Advertising and Differentiated Products (Advances in Applied Microeconomics, Vol. 10)

ADVERTISING AND DIFFERENTIATED PRODUCTS ADVANCES IN APPLIED MICROECONOMICS Series Editor: Michael R. Baye ADVANCES ...

Author: J.P. Nelson | Michael R. Baye

51 downloads 1031 Views 2MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

ADVERTISING AND DIFFERENTIATED PRODUCTS

ADVANCES IN APPLIED MICROECONOMICS Series Editor: Michael R. Baye

ADVANCES IN APPLIED MICROECONOMICS

VOLUME 10

ADVERTISING AND DIFFERENTIATED PRODUCTS EDITED BY

MICHAEL R. BAYE Bert Elwert Professor of Business Economics & Public Policy, Kelly School of Business, Indiana University, USA

JON P. NELSON Professor of Economics, Department of Economics, Pennsylvania State University, USA

2001

JAI An Imprint of Elsevier Science Amsterdam – London – New York – Oxford – Paris – Shannon – Tokyo

ELSEVIER SCIENCE Ltd The Boulevard, Langford Lane Kidlington, Oxford OX5 1GB, UK © 2001 Elsevier Science Ltd. All rights reserved. This work is protected under copyright by Elsevier Science, and the following terms and conditions apply to its use: Photocopying Single photocopies of single chapters may be made for personal use as allowed by national copyright laws. Permission of the Publisher and payment of a fee is required for all other photocopying, including multiple or systematic copying, copying for advertising or promotional purposes, resale, and all forms of document delivery. Special rates are available for educational institutions that wish to make photocopies for non-proﬁt educational classroom use. Permissions may be sought directly from Elsevier Science Global Rights Department, PO Box 800, Oxford OX5 1DX, UK; phone: ( + 44) 1865 843830, fax: ( + 44) 1865 853333, e-mail: [email protected]. You may also contact Global Rights directly through Elsevier’s home page (http://www.elsevier.nl), by selecting ‘Obtaining Permissions’. In the USA, users may clear permissions and make payments through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA; phone: ( + 1) (978) 7508400, fax: ( + 1) (978) 7504744, and in the UK through the Copyright Licensing Agency Rapid Clearance Service (CLARCS), 90 Tottenham Court Road, London W1P 0LP, UK; phone: ( + 44) 207 631 5555; fax: ( + 44) 207 631 5500. Other countries may have a local reprographic rights agency for payments. Derivative Works Tables of contents may be reproduced for internal circulation, but permission of Elsevier Science is required for external resale or distribution of such material. Permission of the Publisher is required for all other derivative works, including compilations and translations. Electronic Storage or Usage Permission of the Publisher is required to store or use electronically any material contained in this work, including any chapter or part of a chapter. Except as outlined above, no part of this work may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without prior written permission of the Publisher. Address permissions requests to: Elsevier Science Global Rights Department, at the mail, fax and e-mail addresses noted above. Notice No responsibility is assumed by the Publisher for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions or ideas contained in the material herein. Because of rapid advances in the medical sciences, in particular, independent veriﬁcation of diagnoses and drug dosages should be made. First edition 2001 Library of Congress Cataloging in Publication Data A catalog record from the Library of Congress has been applied for. British Library Cataloguing in Publication Data A catalogue record from the British Library has been applied for. ISBN: 0-7623-0823-0 ∞ The paper used in this publication meets the requirements of ANSI/NISO Z39.48-1992 (Permanence of 䊊 Paper). Printed in The Netherlands.

CONTENTS LIST OF CONTRIBUTORS

vii

EDITORS’ NOTES

xi

EFFECTS OF ADVERTISING ON U.S. NON-ALCOHOLIC BEVERAGE DEMAND: EVIDENCE FROM A TWO-STAGE ROTTERDAM MODEL Henry W. Kinnucan, Yuliang Miao, Hui Xiao and Harry M. Kaiser

1

THE LONG-RUN DEMAND FOR ALCOHOLIC BEVERAGES AND THE ADVERTISING DEBATE: A COINTEGRATION ANALYSIS N. Edward Coulson, John R. Moran and Jon P. Nelson

31

MANDATED EXCLUSIVE TERRITORIES: EFFICIENCY EFFECTS AND REGULATORY SELECTION BIAS Tim R. Sass and David S. Saurman

55

RACE AND RADIO: PREFERENCE EXTERNALITIES, MINORITY OWNERSHIP, AND THE PROVISION OF PROGRAMMING TO MINORITIES Peter Siegelman and Joel Waldfogel

73

THE VALUE OF ADVERTISING IN A MAGAZINE BUNDLE Craig A. Depken II and Dennis P. Wilson

109

PRICING DYNAMICS OF MULTIPRODUCT RETAILERS Daniel Hosken, David Matsa and David Reiffen

129

PRODUCT INNOVATION IN SERVICES: A FRAMEWORK FOR ANALYSIS Roger Betancourt and David Gautschi

155

v

vi

ASYMPTOTIC EFFICIENCY IN STACKELBERG MARKETS WITH INCOMPLETE INFORMATION Jianbo Zhang and Zhentang Zhang

185

ADVERTISING COOPETITION: WHO PAYS? WHO GAINS? James A. Dearden and Gary L. Lilien

203

A MODEL OF VERTICAL DIFFERENTIATION, BRAND LOYALTY, AND PERSUASIVE ADVERTISING Victor J. Tremblay and Carlos Martins-Filho

221

ALCOHOL ADVERTISING AND ADVERTISING BANS: A SURVEY OF RESEARCH METHODS, RESULTS, AND POLICY IMPLICATIONS Jon P. Nelson

239

LIST OF CONTRIBUTORS Roger Betancourt

Department of Economics, University of Maryland, College Park, MD 20742, USA

N. Edward Coulson

Department of Economics, Pennsylvania State University, University Park, PA 16802, USA

James A. Dearden

Department of Economics, Lehigh University, Bethlehem, PA 18015, USA

Craig A. Depken II

Department of Economics, University of Texas at Arlington, Arlington, TX 76019, USA

David Gautschi

Deloitte & Touche LLP, Seattle, WA 98104, USA

Daniel Hosken

Federal Trade Commission, Washington, D.C. 20580, USA

Harry M. Kaiser

Department of Applied Economics and Management, Cornell University, Ithaca, NY 14853, USA

Henry W. Kinnucan

Department of Agricultural Economics and Rural Sociology, Auburn University, Auburn, AL 36849, USA

Gary L. Lilien

Department of Management Science & Information Systems, Pennsylvania State University, University Park, PA 16802, USA

Carlos Martins-Filho

Department of Economics, Oregon State University, Corvallis, OR 97331, USA vii

viii

David Matsa

Department of Economics, Massachusetts Institute of Technology, Cambridge, MA 02142, USA

Yuliang Miao

International Risk Management, American Express, Phoenix, AZ 85020, USA

John R. Moran

Department of Economics & Center for Policy Research, Syracuse University, Syracuse, NY 13244, USA

Jon P. Nelson

Department of Economics, Pennsylvania State University, University Park, PA 16802, USA

David Reiffen

Federal Trade Commission, Washington, D.C. 20580, USA

Tim R. Sass

Department of Economics, Florida State University, Tallahassee, FL 32306, USA

David S. Saurman

Department of Economics, San Jose State University, San Jose, CA 95192, USA

Peter Siegelman

School of Law, Fordham University, New York, NY 10023, USA

Victor J. Tremblay

Department of Economics, Oregon State University, Corvallis, OR 97331, USA

Joel Waldfogel

The Wharton School, University of Pennsylvania & NBER, Philadelphia, PA 19104, USA

Dennis P. Wilson

Department of Economics, University of Texas at Arlington, Arlington, TX 76019, USA

Hui Xiao

SBS Credit Strategy, American Express, Phoenix, AZ 85027, USA

ix

Jianbo Zhang

Department of Economics, University of Kansas, Lawrence, KS 66045, USA

Zhentang Zhang

Wissenschaftszentrum Berlin (WZB), D-10785 Berlin, Germany

This Page Intentionally Left Blank

EDITORS’ NOTES Guidelines for submitting papers to Advances in Applied Microeconomics and information about planned future volumes may be obtained via the internet on my Homepage: http://php.indiana.edu/ ~ mbaye Michael R. Baye Series Editor Bloomington, IN June 2001 The purpose of this series is to provide a forum for theoretical and empirical research in applied microeconomics. The eleven chapters in this volume provide insight into the strengths and weaknesses of alternative methodological approaches to issues of advertising and differentiated products. It has been a pleasure to be associated with the preparation of the volume. Jon P. Nelson Guest Editor University Park, PA June 2001

xi

This Page Intentionally Left Blank

EFFECTS OF ADVERTISING ON U.S. NON-ALCOHOLIC BEVERAGE DEMAND: EVIDENCE FROM A TWO-STAGE ROTTERDAM MODEL Henry W. Kinnucan, Yuliang Miao, Hui Xiao and Harry M. Kaiser ABSTRACT A two-stage Rotterdam model is estimated to determine the effects of advertising on the demand for non-alcoholic beverages in the United States. Results suggest that advertising has no effect on the demand for non-alcoholic beverages taken as a group. However, the hypothesis that advertising has no effect on the distribution of demand within the nonalcoholic beverage group is ﬁrmly rejected. Coffee and tea are most affected by other beverage advertising, and milk the least. Similarly, juice advertising exerts the largest inﬂuence within the beverage group, and milk advertising the least. Overall, however, the major determinants of the consumption pattern are relative prices and structural change.

Advertising and Differentiated Products, Volume 10, pages 1–29. Copyright © 2001 by Elsevier Science Ltd. All rights of reproduction in any form reserved. ISBN: 0-7623-0823-0

1

2

HENRY W. KINNUCAN ET AL.

I. INTRODUCTION Galbraith’s hypothesis – “If advertising affects the distribution of demand between sellers of a particular product, it must also be supposed that it affects the distribution between products” (Galbraith, 1967, p. 205) – has added signiﬁcance in the context of non-alcoholic beverage advertising. At $1.1 billion in 1994 alone, this product group is one of the most heavily advertised in the U.S. economy. Moreover, two items in the group – milk and fruit juices – are the target of signiﬁcant levels of farm-based generic advertising (over $100 million in 1994). Another $114 million now exists for the “milk moustache” print campaign nominally funded by milk processors (USDA, AMS, 1997). (We say “nominally” because part of the cost is shifted to consumers in the form of higher milk prices.) Generic advertising programs for farm products began in earnest in 1983 when federal legislation authorized a nationwide mandatory “check-off” program for dairy. Funded at 15¢ per hundred pounds of milk marketed, the program generates about $200 million per year for the purpose of promoting U.S. milk and milk products in domestic and export markets. In 1985 Congress authorized similar mandatory programs for beef, pork, and watermelons as part of the farm bill, adding $107 million to the dairy expenditure (Forker & Ward, 1993, pp. 102–103). Today some 50 farm commodities have promotion checkoffs authorized under federal or state legislation, and the annual budget is about $1 billion. These commodities range from minor specialty crops such as cranberries, artichokes, avocados, and raisins to major ﬁeld crops such as wheat, soybeans, and cotton. Industry monies are augmented by federal subsidies for export promotion, as much as $230 million per year over the last decade (Kinnucan & Ackerman, 1995). At $434 million, the dairy and citrus industries are the largest advertisers within the agricultural group. Although substantial research has been done to determine whether generic advertising of ﬂuid milk and fruit juices is proﬁtable (Kaiser, 1997; Lee & Brown, 1992; Ward & Dixon, 1989; Wohlgenant & Clary, 1993), no study has investigated non-alcoholic beverage demand in an integrated framework that takes into account the full array of substitution effects.1 For example, a successful ﬂuid milk advertising campaign might erode the demand and price for citrus products. In addition, the resulting decrease in citrus price could lower the milk price through second-round or feedback effects. These spillover and feedback effects have not been addressed in the milk and citrus advertising literatures, which could cause the estimated returns to be overstated (Kinnucan, 1996).

Effects of Advertising on U.S. Non-alcoholic Beverage Demand

3

In this paper, we determine whether advertising of non-alcoholic beverages has any detectable effect on aggregate demand. Owing to the importance of demand interrelationships, special attention is given to spillover effects, i.e. whether advertising for one beverage affects the demand for related beverages. A secondary objective is to test whether structural change plays a role in the observed consumption pattern, particularly the rise in soft-drink consumption between 1970 and 1994 from 24.3 to 52.2 gallons per person and the decline in milk consumption from 31.3 to 24.7 gallons per person. We begin by discussing the model and data. The hypothesis tests, parameter estimates, and elasticities are then presented and discussed. The paper concludes with a summary of the major ﬁndings.

II. MODEL The Rotterdam model was selected because it is consistent with demand theory (Theil, 1965; Barnett, 1979); it is as ﬂexible as any other local approximating form (Mountain, 1988); it lends itself to advertising applications (Brown & Lee, 1993; Duffy, 1987, 1990, 1991); it appears to be robust to alternative separability assumptions (Moschini, Moro & Green, 1994, pp. 64–69); and prior testing indicated that the estimated advertising effects from the Rotterdam model were similar to those obtained from its major rival, the (linear approximate) Almost Ideal Demand System. Advertising variables are generally incorporated into the Rotterdam model using the Ichimura-Tichner conditions, scaling, or translating (Brown & Lee, 1993; Duffy, 1987, 1995). Testing these speciﬁcations against a simpler form in which advertising enters as a simple shift variable, we found the shift speciﬁcation to be superior. Consequently, we follow inter alia Kinnucan et al. (1997) and adopt the simple-shift speciﬁcation of the absolute-price version of the Rotterdam model.2 Demographics have been shown to be important determinants of milk consumption. For example, Kinnucan (1986) found that age and the non-white population proportion to be signiﬁcant in explaining milk consumption in New York City; similar results were obtained by Ward and Dixon (1989) for the United States. Accordingly, in this study demographics are accounted for by specifying age, dining-out, and trend variables in the model. As with advertising, the age and dining-out variables (to be deﬁned later) were incorporated as intercept shifters, since preliminary testing showed the scaling and translating speciﬁcations to be inferior. Trend is included by specifying an intercept term in the Rotterdam model. The adding-up condition is enforced by

4

HENRY W. KINNUCAN ET AL.

requiring that the coefﬁcients of the demographic variables and the intercepts sum to zero across equations. Because quarterly data were not available on a national basis for any of the beverages except milk, the model is estimated with annual data. An advantage of using annual data is that lag structures need not be speciﬁed. Speciﬁcally, Clarke (1976) ﬁnds that advertising carryover for mature, frequentlypurchased, low-priced items is generally nine months or less (see also Leone, 1995). As for price and income effects, Tomek and Cochrane (1965) posit that long-run demand equations for food items encompass a period of one year or less. Both of these hypotheses are consistent with the beverage demand literature (Brown & Lee, 1993; Kinnucan, 1986; Ward & Dixon, 1989). Following Duffy (1987, 1990, 1991), Goddard and Tielu (1988), and Richards et al. (1997), a two-stage budgeting process is posited. In the ﬁrst stage, the consumer allots her total income to broad commodity groups, one of which is non-alcoholic beverages. In the second stage, the beverage budget is divided among the individual drinks. The basic speciﬁcation is given by:

冘 4

s¯it Dqit = ␮⬘i S¯ Gt DQGt +

␲ijDpjt +

j

and

冘 4

␤ijDAjt + ai + bi Daget + ci Dfafht + vit (1)

j

冋冘册 4

S¯ Gt DQGt = ␮GDQt + ␲G

␮⬘j Dpjt + ␲O Dp + ␤G

j

冋冘册 4

0 t

␮⬘j DAjt + ␤O DA0t

j

+ aG + bG Daget + cG Dfafht + vGt (2) where Eq. (1) corresponds to the second-stage (conditional) demand functions for beverage i (for i = 1, 2, 3, 4 for milk, juices, soft drinks, and coffee and tea, respectively) in year t (for t = 2, 3, . . . , 25 for 1971 to 1994), and (2) corresponds to the ﬁrst-stage (group) demand function. In Eqs (1) and (2), D denotes the logarithmic ﬁrst-difference operator, i.e. Dx = ln xt ⫺ ln xt ⫺ 1; the subscript G denotes the non-alcoholic beverage group; and the subscript O denotes all other (“non-group”) goods. The s¯it term in (1) is the budget share for the i th beverage in year t expressed as an average of the current and preceding year’s budget shares, i.e. s¯it = (sit + sit ⫺ 1)/2, where si = pi qi/ 兺ni pi qi. Similarly, S¯ Gt = 兺4i s¯it is the corresponding group budget share. The term DQGt = 兺4i (¯sit/S¯ Gt)Dqit in (1) and (2) is G’s Divisia volume index, which can be interpreted as a third-order approximation of real expenditure on the nonalcoholic beverage group (Goldberger, 1987, p. 95). The qit term denotes per capita consumption of beverage item i in year t; pjt is the nominal price of beverage item j in year t; Ajt is the real per capita

Effects of Advertising on U.S. Non-alcoholic Beverage Demand

5

advertising expenditure on beverage item j in year t; aget is the proportion of the U.S. population less than age ﬁve in year t; fafht is the ratio of food-awayfrom-home expenditures to food-at-home expenditures in year t; and vit and vGt are random error terms. The intercept terms ai and aG in (1) and (2) are properly interpreted as trend coefﬁcients (Deaton & Muellbauer, 1980, p. 70) and thus serve, along with the coefﬁcients for age and fafh, as tests for structural change. Following Gao and Shonkwiler (1993, p. 316, no. 2), we deﬁne structural change as a shift in the demand function caused by factors other than relative prices, real income, and advertising. The DQt term in (2) is real per capita income in logarithmic ﬁrst-difference form, which may be interpreted as the Divisia volume index corresponding to total consumer expenditure, or “income” for short. The pot term is a price index for non-group goods, and Aot is the total real per capita advertising expenditure on non-group goods. Setting aG = bG = cG = 0 (no group structural change or demographic effects), Eq. (2) reduces to Duffy’s group-demand equation (Duffy, 1987, p. 1060) when price and advertising homogeneity are imposed. In this study, we treat price homogeneity in the group-demand equation (␲G + ␲O = 0) as a maintained hypothesis, but impose advertising homogeneity (␤G + ␤O = 0) only if it is compatible with the data. Following Duffy (1987, p. 1054), this model implicitly assumes that consumer preferences are block independent. Speciﬁcally, non-alcoholic beverages are assumed to form one of the groups in a block-additive utility function. This assumption has three advantages. First, by estimating (1) and (2) simultaneously, the endogeneity problem with respect to group expenditure (LaFrance, 1991) is resolved. Second, by separating the group demand function from the conditional demand functions, more information is provided on how structural change and advertising affect demand. For example, advertising may enlarge total demand (group effect), cause market shares to shift (within-group effect), both, or neither. Equations (1) and (2) permit testing all four cases. Finally, the two-stage approach permits a more parsimonious speciﬁcation, which mitigates multicollinearity problems. The ␮⬘i term in (1) is the conditional marginal share of the i th beverage, and the corresponding ␮G term in (2) is G’s marginal share. As noted by Duffy (1987, p. 1054):

冘 4

␮⬘i = 1

and

␮⬘i = ␮i /␮G

i

where ␮i is the i th’s beverage unconditional marginal share. In this model, the conditional marginal shares play a dual role: they indicate how beverage

6

HENRY W. KINNUCAN ET AL.

expenditure is allocated at the margin (see Eq. (1)); and they serve as weights in the group price and advertising indices (the bracketed terms in Eq. (2)). The model is completed with the following parametric restrictions:

冘 4

␲ij = 0

i = 1, . . . , 4

(price homogeneity)

(3a)

for all i, j

(price symmetry)

(3b)

j = 1, . . . , 4

(Cournot)

(3c)

(adding up)

(3d)

j

␲ij = ␲ji

冘冘冘冘 4

␲ij = 0

i

4

4

ai =

i

4

bi =

i

冘 4

ci = 0, and

i

␮⬘i = 1

i

These conditions, along with advertising adding-up (Basmann, 1956, p. 57),

冘 4

␤ij = 0

j = 1, . . . , 4

(4a)

i

are treated as maintained hypotheses. With prices and expenditure held constant, Selvanathan (1989, pp. 216–218) shows that advertising responses in the absolute-price version of the Rotterdam model are homogeneous of degree zero, i.e.

冘 4

␤ij = 0

i = 1, . . . , 4.

(4b)

j

The symmetry condition, ␤ij = ␤ji

for all i, j

(4c) 3

however, does not necessarily hold (Selvanathan, 1989, pp. 218–219). In this study, restrictions (4b) and (4c) are tested, and imposed only if they are compatible with the data. An implicit assumption underlying (1) is that brand and generic advertising have identical effects on demand. This assumption does not affect soft drinks or coffee and tea, as advertising for these beverages is strictly brand. Nor does it affect milk, since the milk advertising data used in this study are strictly generic. (Brand advertising of milk is minuscule in relation to generic and thus ignored.) For juices, however, the data contain signiﬁcant amounts of both

Effects of Advertising on U.S. Non-alcoholic Beverage Demand

7

types of advertising as there are both strong brands (e.g. SunKist; Citrus Hill), and active support of generic advertising by citrus growers. Thus, to the extent that brand advertising has no effect on market demand, combining the two expenditures may cause the own-advertising coefﬁcient for juices to be biased downward. Conversely, if brand advertising increases market demand, as some research suggests (Brester & Schroeder, 1995; Kaiser & Liu, 1997), the bias should be minimal.

III. DATA The model was estimated with annual time-series data covering the period 1970–1994.4 Data on consumption of ﬂuid milk, fruit juices, soft drinks, and coffee and tea were obtained from Putman and Allshouse (1996, Table 37). Because tea consumption is modest (about seven gallons per person per year) and has changed little (from a low of 6.8 gallons per person in 1970 and 1990 to a high of 7.7 gallons in 1976), data for tea and coffee were combined. Bottled water consumption, which increased from 1.2 gallons per person per year in 1976 (the ﬁrst available ﬁgure) to 9.2 gallons in 1993, is not considered in this study because the series is incomplete. The included beverages accounted for 92.5% of total non-alcoholic beverage consumption in 1993. Price data were obtained primarily from the U.S. Department of Labor’s CPI Detailed Report (1971–1997). To facilitate the computation of budget shares, the CPIs for each beverage were converted to per-gallon prices using standard unit conversions. A composite price series for coffee and tea was obtained by taking the quantity-share weighted average of the tea and coffee prices. As a proxy for the price of juices, the price of frozen orange-juice concentrate was used because orange juice represents the major component of the juice category. As a proxy for the non-group price, we use the Consumer Price Index for all items. The advertising data were obtained from annual issues of AD $ SUMMARY published by Leading National Advertisers, Inc. (LNA, 1970–1995). LNA is a tracking service agency that estimates the advertising expenditures for all brands (including generic advertising organizations such as the National Dairy Board) that spend at least $25,000 per year in a particular medium. The media tracked by LNA include network, spot, syndicated, and cable television; network and national spot radio; magazines (including Sunday supplements); newspapers; and billboards. The advertising data were divided by the CPI for all items (1982–1984 = 100) to remove the effects of inﬂation.5 Sources and complete deﬁnitions for all the variables used in this study, including the CPI,

8

HENRY W. KINNUCAN ET AL.

population, age, and the food-away-from-home variables, are provided in Appendix A.

IV. ESTIMATION PROCEDURE Equations (1) and (2) were estimated jointly using the Iterative Seemingly Unrelated Regression (ITSUR) routine in EViews. However, to determine the robustness of parameters to estimation procedure, we also estimated (1) and (2) as a recursive system with beverage expenditure treated as exogenous. In both estimates, beverage prices are treated as exogenous because Bronsard and Salvas-Bronsard’s (1984) work suggests that price endogeneity is relatively unimportant in demand-system estimation when the commodities in question constitute a small share of consumer income, as is true for non-alcoholic beverages given its budget share of about 2%. By extension, advertising expenditures are treated as (weakly) exogenous, because non-alcoholic beverage advertising over the sample period accounted for less than 2.5% of total U.S. advertising expenditures. Thus, the supply of advertising services to the non-alcoholic beverage sector may be regarded as perfectly elastic, at least for the purposes of estimation. The adding-up conditions imply that only three equations in the conditional system are independent. The usual procedure, followed in this study, is to drop one equation, estimate the remaining system, and calculate the parameters for the omitted equation using the classical restrictions. The advertising homogeneity and symmetry conditions in Eqs (4b) and (4c) were tested using the Wald criterion. Based on these tests, an appropriately restricted model was used to test for structural change and the signiﬁcance of advertising effects. Unless indicated otherwise, a t-ratio of 1.65 is used to indicate parameter signiﬁcance at the 5% level for one-tail tests (own-price and expenditure effects), and 1.96 for two-tail tests. Elasticities (see Appendix B) are evaluated at mean budget shares for 1990–1994, the last ﬁve years in the sample.

V. RESULTS The Wald tests indicated that advertising homogeneity and symmetry in the conditional demand equations are compatible with the data (Table 1). In addition, a t-test indicated that advertising homogeneity could not be rejected in the group demand equation. Accordingly, analysis proceeded with advertising symmetry and homogeneity imposed on both (1) and (2).

Effects of Advertising on U.S. Non-alcoholic Beverage Demand

9

Table 1. Wald Tests of Restrictions on the Conditional Demand Equations. Model

Restriction

Eq. (1)

Price Homogeneity (PH) and Price Symmetry (PS) (Maintained hypotheses) PH, PS and advertising homogeneity (AH) PH, PS, AH and advertising symmetry (AS) PH, PS, AH, AS and ai = 0, all i PH, PS, AH, AS and bi = 0, all i PH, PS, AH, AS and ci = 0, all i PH, PS, AH, AS and bi = ci = 0, all i PH, PS, AH, AS and ai = bi = ci = 0, all i PH, PS and ␤ij = 0, all i, j

1a 1b 1c 1d 1e 1f 1g 1h

Computed ␹2

p-value

–

–

4.556 7.4091 19.944 5.262 5.472 11.867 66.094 85.802

0.2073 0.2847 0.0002 0.1536 0.1403 0.0650 0.0000 0.0000

We next conducted system-wide tests on trend and the advertising variables to determine whether either could be deleted from the second-stage model. Tests in both cases were decisive. In particular, the hypothesis that trend, either singularly, or in combination with demographics, can be deleted from the model is rejected at the p < 0.0002 level (Table 1, models (1c) and (1g)). Similarly, the hypothesis that the advertising variables’ coefﬁcients are jointly zero is rejected with almost no probability of a Type I error (model (1h)). Thus, there is strong evidence from these data that the consumption pattern is inﬂuenced by both advertising and structural change. Parameter estimates from the EViews econometric package are presented in Tables 2 and 3. Tests for serial correlation were borderline and seemed to be more conclusive in the simultaneous model. The simultaneous model, therefore, was estimated with correction for serial correlation under the assumption that the equations share a common auto-regressive parameter. The recursive and simultaneous estimates are similar except that one crossprice coefﬁcient and one trend coefﬁcient gain signiﬁcance in the simultaneous estimates, and one marginal budget share loses signiﬁcance (Table 2). The age effect in the milk equation is insigniﬁcant in the recursive estimate, but signiﬁcant (one-tail test) in the simultaneous estimate; a similar result applies to the “dining-out effect” in the juice equation. Estimation procedure has no signiﬁcant effect on the group demand equation (Table 3). Unless indicated otherwise, the remaining discussion will focus on the simultaneous estimates. Focusing on the second-stage equations (Table 2), all the own-price and expenditure coefﬁcients have the expected sign and most are signiﬁcant. The cross-price coefﬁcients are all positive, which suggests that the beverages are

10

Table 2. Coefﬁcient Estimates of Conditional Demand Equations for U.S. Non-Alcoholic Beverages, 1971–1994. Price Coefﬁcients

age

fafh

␮⬘i

ai

bi

ci

Recursive Estimatesa Milk –0.0453 0.0009 (–4.87)b (1.05) Juices 0.0310 –0.0670 0.0092 0.0219 (3.78) (–2.80) (4.60) (2.36) Soft Drinks 0.0080 0.0287 –0.0551 –0.0047 0.0068 –0.0377 (1.01) (1.63) (–2.76) (–1.76) (0.66) (–2.02) Coffee & Tea 0.0063 0.0073 0.0185 –0.0321 –0.0054 –0.0380 0.0355 0.0078 (1.39) (0.66) (1.44) (–2.42) (–2.16) (–4.56) (2.67) (0.59)

0.0850 (2.73) 0.1909 (2.52) 0.4608 (4.78) 0.2633 (2.78)

–0.0028 0.0560 –0.0632 (–3.46) (1.23) (–2.56) –0.0016 0.1197 0.0772 (–0.78) (1.04) (1.30) 0.0091 –0.0688 –0.0535 (3.42) (–0.50) (–0.70) –0.0047 –0.1077 0.0395 (–1.88) (–0.81) (0.56)

Simultaneous Estimatesc 0.0008 Milk –0.0474 (0.88) (–4.52) Juices 0.0258 –0.0567 0.0078 0.0213 (2.93) (–2.39) (4.04) (2.09) Soft Drinks 0.0114 0.0288 –0.0596 –0.0037 0.0065 –0.0434 (1.30) (1.70) (–3.05) (–1.34) (0.58) (–2.39) Coffee & Tea 0.0102 0.0020 0.0194 –0.0316 –0.0050 –0.0356 0.0404 0.0002 (2.04) (0.19) (1.52) (–2.55) (–2.03) (–4.61) (3.29) (0.02)

0.1142 (4.07) 0.1096 (1.54) 0.5379 (7.50) 0.2382 (4.07)

–0.0028 0.0772 –0.0718 (–3.58) (1.81) (–2.91) –0.0012 0.0753 0.0900 (–0.70) (0.76) (1.68) 0.0085 –0.0561 –0.0471 (3.45) (–0.47) (–0.64) –0.0045 –0.0964 0.0291 (–2.12) (–0.90) (0.46)

␲i1

␲i2

␲i3

␲i4

␤i1

␤i2

␤i3

␤i4

R2

D·W·

0.47 2.05 0.71 2.58 0.56 2.11 0.48 2.53

0.41 1.92 0.74 2.54 0.56 2.03 0.54 2.37

Serial correlation parameter (common to all equations): ␳˙ = –0.2618, t-ratio = –2.08 a Model (1b) of Table 1 estimated by ITSUR regression without correction for serial correlation. All coefﬁcients except expenditure are divided by group budget share. b Numbers in parentheses are asymptotic t-ratios. c Model (1b) of Table 1 estimated jointly with the group demand Eq. (see Table 3) by ITSUR with correction for serial correlation.

HENRY W. KINNUCAN ET AL.

Expend · Intercept

Equation

Advertising Coefﬁcients

Price Coefﬁcients Estimation Procedure Recursivea Simultaneousb

a

Advertising Coefﬁcients Income

Intercept

age

fafh

␲G

␲O

␤G

␤O

␮G

aG

bG

cG

R2

D·W·

–0.00498 (–3.47) –0.00589 (–3.39) [–0.5122]

0.00498 (3.47) 0.00589 (3.39) [0.5122]

0.00009 (0.34) –0.00026 (–0.93) [–0.0225]

–0.00009 (–0.34) 0.00026 (0.93) [0.0225]

0.00165 (0.66) 0.00240 (1.11) [0.2087]

–0.000036 (–0.46) –0.000054 (–0.87) [–0.4696]

–0.00949 (–2.13) –0.01207 (–3.01) [–1.0496]

0.00137 (0.69) 0.00091 (0.56) [0.07913]

0.48

2.10

0.48

2.12

Text Eq. (2) with price and advertising homogeneity imposed estimated by OLS. Text Eq. (2) with price and advertising homogeneity imposed and estimated jointly with text Eq. (1) by ITSUR. Note: Number in parentheses is t-ratio; number in brackets is elasticity. b

Effects of Advertising on U.S. Non-alcoholic Beverage Demand

Table 3. Coefﬁcient Estimates of Group Demand Equation for U.S. Non-Alcoholic Beverages, 1971–1994.

11

12

HENRY W. KINNUCAN ET AL.

conditional net substitutes. Most of the advertising and trend coefﬁcients are signiﬁcant and most of the demographic coefﬁcients are not, as expected based on the system-wide tests (see Table 1). In particular, age and fafh are signiﬁcant only in the milk demand equation. The trend terms (intercepts) are signiﬁcant in all equations except juices. Turning to the ﬁrst-stage equation (Table 3), the advertising variables are not signiﬁcant. Thus, the hypothesis that advertising affects the total demand for non-alcoholic beverages is rejected. Combining this result with the secondstage results, it appears that advertising affects the distribution of expenditures within the non-alcoholic beverage group, but has no effect on total expenditures. This ﬁnding is in line with Duffy’s work (1987, 1990, 1991) on alcoholic beverage advertising in the United Kingdom. The price effects in the group demand equation are highly signiﬁcant (t-ratio of 3.39 in absolute value) whereas the income effect is only marginally signiﬁcant based on a one-tail test (t-ratio of 1.11). Among the demographics, age is signiﬁcant and negative, which suggests that Americans allocate a smaller portion of their income to non-alcoholic beverages as the less-than-ﬁve population segment increases. The fafh variable is not signiﬁcant, nor is the intercept term. Thus, the effects of structural change and dining out appear to be conﬁned to the individual commodity level. The remaining discussion focuses on the elasticities. To evaluate Edgerton’s (1997) critique about the potentially misleading nature of conditional elasticities, we present for comparison purposes both the conditional and the unconditional elasticities. Price Effects The own-price elasticities are all less than one in absolute value, which suggests that non-alcoholic beverage demands are price inelastic (Table 4). For milk, the unconditional own-price elasticity is –0.19, which compares well with Ward and Dixon’s (1989, p. 735) estimate of –0.15 for the U.S. market, and with Goddard and Tielu’s (1988, p. 272) estimate of –0.22 for the Ontario market. Similarly, the unconditional own-price elasticity for juices (–0.40) compares well with Goddard and Tielu’s (GT) estimates of –0.29, –0.27, and –0.28 for, respectively, orange, tomato, and apple juice. For soft drinks and coffee and tea, our estimates of the unconditional own-price elasticities are identical at –0.47, which may be compared to GT’s estimate of –1.0 for soft drinks. (GT’s model excluded coffee and tea.) The unconditional own-price elasticities are absolutely larger than their conditional counterparts. The greatest differences are for soft drinks (–0.48

Price of: Quantity of:

Milk Juices Soft Drinks Coffee & Tea Milk Juices Soft Drinks Coffee & Tea

Milk

–0.1685** 0.1642** 0.0262 0.0803** –0.1922** 0.1236* –0.0459* –0.0290**

Juices

0.0917** –0.3609** 0.0663 0.0157 0.0690* –0.3999* –0.0030* –0.0892*

Soft Drinks

Coffee & Tea

Other Goods

Expenditure

Budget Share

0.0405 0.1833 –0.1372** 0.1528

Conditionala 0.0363** 0.0127 0.0447 –0.2488**

– – – –

0.4060** 0.6976 1.2383** 1.8756**

0.2813 0.1571 0.4344 0.1270

–0.0709* –0.0082* –0.4771** –0.3621*

Unconditionala –0.0131** –0.0721* –0.1059* –0.4768**

0.2072* 0.3560* 0.6319* 0.9571*

0.0844* 0.1450 0.2574* 0.3899*

0.0032 0.0018 0.0050 0.0015

Effects of Advertising on U.S. Non-alcoholic Beverage Demand

Table 4. Hicksian Price and Expenditure Elasticities for U.S. Non-Alcoholic Beverages, Evaluated at 1990–1994 Mean Data Points.

a Based on the simultaneous estimates presented in Tables 2 and 3. Double asterisk indicates that all of the parameters in the applicable elasticity formula (see Appendix B) are individually signiﬁcant; single asterisk indicates that at least one is signiﬁcant; no asterisk indicates none is signiﬁcant.

13

14

HENRY W. KINNUCAN ET AL.

versus –0.14) and the least for juices (–0.40 versus –0.36). As to cross-price relationships, juices and milk are net substitutes regardless of whether conditional or unconditional responses are considered. Coffee and tea and milk, however, switch from net substitutes (conditional response) to net complements. Similarly, soft drinks and milk are independent in the conditional estimates, but net complements in the unconditional estimates. These sign switches, coupled with the numerical differences in the own-elasticities, tend to conﬁrm Edgerton’s concern that conditional elasticities may be inappropriate for policy analysis. Cross-price elasticities for the non-group price variable are positive, which suggests that non-alcoholic beverages and non-group goods are net substitutes. In general, the within-group cross-price elasticities are modest (less than 0.13 in absolute value). This suggests that cross-price effects within the group are weak. For example, the price of coffee and tea is not important in the demand for milk, since the cross elasticity (–0.01), while statistically signiﬁcant, is minute. Expenditure Effects All unconditional income elasticities are between zero and one, which suggests the beverages are normal goods (Table 4). Coffee and tea is the most income responsive (0.39), and milk the least (0.08). The milk income elasticity may be compared to Ward and Dixon’s estimate of 0.29 and GT’s estimate of 0.41. Although one should not make too much of these differences as they may not be statistically signiﬁcant, it is perhaps worth noting that GT’s estimate is identical to our conditional estimate of 0.41. Also, GT found soft drinks to be more expenditure elastic than either milk or juices, which is consistent with our ﬁndings. As expected, the conditional income elasticities for milk, juices, soft drinks, and coffee and tea (0.41, 0.70 (insigniﬁcant), 1.24, and 1.88) are substantially larger than the corresponding unconditional elasticities (0.08, 0.14 (insigniﬁcant), 0.26, and 0.39). The tiny income elasticity for milk (0.08) is consistent with the fact that, despite cyclical changes in the U.S. economy, per-capita milk consumption continues its steady decline (see Fig. 1). Advertising Effects The estimated advertising elasticities afﬁrm the importance of spillover (Table 5). Whereas only half of the own-advertising elasticities are signiﬁcant, fully two-thirds of the cross elasticities are statistically signiﬁcant. Moreover, many of the cross-advertising elasticities are larger in absolute value than the

Effects of Advertising on U.S. Non-alcoholic Beverage Demand

15

Fig. 1. Per Capita Consumption of U.S. Non-Alcoholic Beverages, 1970–1994.

own-advertising elasticities, and in some cases exceed price and income elasticities. Overall, coffee and tea appear to be the most affected by advertising of other commodities, and milk the least (row). Similarly, juice advertising appears to exert the largest inﬂuence within the beverage market, and milk advertising the least (column). Consistent with GT, the estimated own-advertising elasticity for juices is positive and the own-advertising elasticity for soft drinks is negative.6 Also consistent with GT, some of the cross-advertising elasticities have signs that appear to be incompatible with price elasticities. For example, because milk and juice are net substitutes, one might expect that an increase in milk advertising would cause juice demand to decrease. The relevant crossadvertising elasticity (0.048), however, is positive. The estimated own-advertising elasticity for milk is positive, but small (0.002) and not statistically signiﬁcant at usual probability levels (t-ratios of 0.88 and 1.05, respectively, in the simultaneous and recursive models). Thus, we cannot reject the null hypothesis that milk advertising has no effect on milk demand. Given the importance of this parameter from the National Dairy Board’s perspective, and evidence from single-equation demand models that show a positive effect (Tomek & Kaiser, 1999), we estimated the model in double log form and as a (linear approximate) Almost Ideal Demand System

16

Table 5. Advertising, Trend, and Demographic Elasticities for U.S. Non-Alcoholic Beverages, Evaluated at 1990–1994 Mean Data Points. Advertising of: Quantity of:

Milk Juices Soft Drinks Coffee & Tea

0.0028 0.0497** –0.0085 –0.0394** 0.0018 0.0479* –0.0117 –0.0442*

Juices

0.0277** 0.1356** 0.0150 –0.2803** 0.0267* 0.1339* 0.0119 –0.2849*

Soft Drinks

Coffee & Tea

Other Goods

–0.0132 0.0414 –0.0999** 0.3181**

Conditionala –0.0178** –0.2266** 0.0930** 0.0016

–0.0181 0.0329 –0.1149* 0.2954*

Unconditionala –0.0200* 0.0091 –0.2304* 0.0157 0.0279 0.0864* –0.0085 0.0423

– – – –

Trend

age

fafh

Advertising Intensityb

–0.9954** –0.7638 1.9567** –3.5433**

0.2744* 0.4793 –0.1291 –0.7591

–0.2552** 0.5729 –0.1084 0.2283

0.0032 0.0297 0.0207 0.0421

–1.1853* –1.0903 1.3773* –4.4210*

–0.1502** –0.2504 –1.4243 –2.7208

–0.2232* 0.6279 –0.0108 0.3761

0.0032 0.0297 0.0207 0.0421

a Based on the simultaneous estimates presented in Tables 2 and 3. Double asterisk indicates that all of the parameters in the applicable elasticity formula (see Appendix B) are individually signiﬁcant; single asterisk indicates that at least one is signiﬁcant; no asterisk indicates none is signiﬁcant. b Advertising expenditure divided by retail revenue. Advertising intensity for the group as a whole is 0.0199.

HENRY W. KINNUCAN ET AL.

Milk Juices Soft Drinks Coffee & Tea

Milk

Effects of Advertising on U.S. Non-alcoholic Beverage Demand

17

(ALIDS). The milk advertising effect continued to be weak and non-robust in these alternative models, which tends to rule out functional form as the culprit. A potential problem with allocation models such as the Rotterdam or ALIDS is that speciﬁcation or measurement error in one equation can carry over to other equations in the system, contaminating all the estimated coefﬁcients. However, this possibility was rejected, since elasticities from the (singleequation) double-log model were similar to the system estimates. A ﬁnal reason for the insigniﬁcant own-advertising effect for milk may be related to data. Speciﬁcally, the model uses annual data, which could lead to bias since the data interval exceeds the purchase interval (Clarke, 1976). Quoting from Duffy (1990, p. 247), “annual data . . . may be the least suitable frequency at which to detect advertising effects.” However, this critique would imply that none of the advertising effects is signiﬁcant, which clearly is not the case (Table 1, model (1h)). Moreover, Duffy (1990, p. 255) summarizes results from an annual versus quarterly model as follows: “we consistently found that the effects of advertising on total drink demand are always minute, and in the quarterly model when we allow for possible simultaneity in our estimation methods, the responsiveness of (product or group) demand with respect to advertising becomes insigniﬁcantly different from zero.” Thus, although we are reluctant to state categorically that milk advertising has no effect on milk demand (other weaknesses in the model or data may be at work), there is reason to be cautious of claims to the contrary. Demographic and Trend Effects The elasticities for age and fafh, which are signiﬁcant only for milk, are less than 0.26 in absolute value. This suggests that changes in the age structure and dining out are relatively unimportant in explaining the observed consumption pattern. The trend “elasticities,” by contrast, are signiﬁcant in all equations except juices and are relatively large. Speciﬁcally, the elasticities for milk, soft drinks, and coffee and tea are –1.00, 1.96, and –3.54, respectively. These elasticities, as noted by Deaton and Muellbauer (1980, p. 70), are properly interpreted as the annual percentage change in per capita consumption in the absence of changes in other variables in the model.7 Accordingly, there appears to have been a trend increase in the demand for soft drinks, largely offset by trend decreases in the demands for coffee and tea and for milk. These changes are perhaps the most important and obvious shifts in U.S. non-alcoholic beverage consumption over the 25-year sample period (Fig. 1). That economic variables appear to be relatively unimportant in explaining these consumption

18

HENRY W. KINNUCAN ET AL.

shifts suggests that future modeling efforts should stress non-economic variables, perhaps along the lines suggested by Gao and Shonkwiler (1993). Returning to the age effect, note that the conditional and unconditional elasticities in Table 5 have opposite signs, which highlights the importance of induced expenditure effects (a variation on Edgerton’s critique). Speciﬁcally, in the ﬁrst-stage equation, age has a negative sign (Table 3), which indicates that as the less-than-ﬁve population segment increases, Americans allocate a smaller portion of income to non-alcoholic beverages (income effect). In the second-stage equation, age has a positive sign (Table 2), which indicates that milk demand increases as the less-than-age-ﬁve segment increases, holding beverage expenditure constant (allocation effect). That the unconditional age elasticity is negative simply means that the income effect dominates the allocation effect. That is, milk gets a larger share of a smaller beverage budget as the toddler segment increases, but the overall demand for milk decreases due to the smaller budget allocation to beverages. To gauge further the relative importance of the demographic and trend effects, we multiplied the elasticities from the milk equation by the percentage changes in the variables for the period 1990–1994 as indicated in Table 6. Results suggest that the three most important factors affecting milk consumption are trend (–4.89%), price structure (3.91%), and fafh (–2.83%). Thus, for example, the 12.7% increase in fafh between 1990 and 1994 would be expected to decrease per capita milk consumption by 2.83%, ceteris paribus. The fafh variable, therefore, has a relatively important effect on milk consumption, despite its small elasticity (–0.22). This underscores Duffy’s (1989) point that when assessing a variable’s impact, its change as well its elasticity must be considered. The least important factors affecting milk consumption are advertising (0.97%), income (0.29%), and age (–0.06%). Overall, the economic variables had a positive net impact on milk demand, but this sympathetic effect was outweighed by the antagonistic effects of non-economic variables, resulting in a net reduction in per capita consumption between 1990 and 1994 (see Fig. 1). Whether milk advertising can reverse this consumption decline is doubtful, since even with a 135% increase in milk advertising expenditure between 1990–1994 period, the demand curve shifted by a mere 0.24% (if one accepts that the advertising coefﬁcient is in fact signiﬁcantly different from zero).

VI. CONCLUDING COMMENTS The results in this study give only partial support to Galbraith’s hypothesis. In particular, while we can state with conﬁdence that advertising shifts the market

Effects of Advertising on U.S. Non-alcoholic Beverage Demand

19

Table 6. Effects of Prices, Income, Advertising, and Demographics on U.S. Milk Demand, 1990–1994. Variable

Row

Income Prices: Milk Juices Soft drinks Coffee and Tea All other goods

1 2 3 4 5

Net Price Effect: (Sum of rows 1–5) Advertising: Milk Juices Soft drinks Coffee and Tea All other goods

6 7 8 9 10

Net Advertising Effect: (Sum of rows 6–10) Demographics: age fafh trend Net Demographic Effect: (Sum of rows 11–13)

11 12 13

Percent Change

Elasticity

Effect (%)

3.44

0.084

0.29

–8.69 –17.00 –8.90 –1.12 13.39

–0.192 0.069 –0.071 –0.013 0.207

1.67 –1.17 0.63 0.01 2.77

–

–

3.91

135.29 –14.16 –21.55 –30.40 11.68

0.0018 0.0267 –0.0181 –0.0200 0.0091

0.24 –0.38 0.39 0.61 0.11

–

–

0.97

0.38 12.69 –

–0.150 –0.223 –0.995

–0.06 –2.83 –4.98

–

–

–7.87

shares of individual beverages within the non-alcoholic beverage group, there is no evidence of wider substitution effects. The basis for this statement is that the advertising variables in the group demand equation are insigniﬁcant. This means that non-alcoholic beverage advertising in toto has no effect on the beverage market’s size. Similarly, the beverage market is unaffected by advertising outside the non-alcoholic group. Thus, to the extent that Galbraith’s hypothesis is correct, it appears to be conﬁned to within-group effects. The dominant pattern in U.S. non-alcoholic beverage consumption over the past 25 years has been a steady increase in per-capita soft-drink consumption, largely at the expense of coffee consumption and, to a lesser extent, milk consumption. Although changes in the price structure, real income, and advertising have inﬂuenced this pattern, our results suggest that structural change is at work. The basis for this claim is that the intercepts in the second-

20

HENRY W. KINNUCAN ET AL.

stage Rotterdam model are jointly signiﬁcant at almost no probability of a Type I error. Moreover, the trend “elasticities” based on these intercepts are much larger than the price, income, and advertising elasticities, which suggests that important consumption changes can be expected even in the absence of changes in economic variables. Demographic variables, namely Americans’ penchant for dining out and the less-than-age-ﬁve population segment, appear to be conﬁned in their effects to milk, and to be relatively modest in their impacts. As to the spillover issue, our results afﬁrm Green, Carman, and McManus’s statement (1991, p. 70) that “. . . non-brand advertising can have important cross-commodity impacts.” Speciﬁcally, milk advertising (which is strictly generic) was found to have a statistically signiﬁcant effect on the demand for juices and for coffee and tea. Whether these cross-effects are economically signiﬁcant depends crucially on supply response in the separate industries (Piggott, Piggott & Wright, 1995; Kinnucan, Xiao & Hsia, 1996), and thus is an open issue. Still, the existence of spillover effects seems no longer in doubt, which raises questions about the role of government in supporting programs that may beneﬁt one industry at the expense of another. From a research perspective, generic advertising clearly has distributional consequences, an issue that deserves greater attention in the beneﬁt-cost literature.

NOTES 1. An exception to this statement is Goddard and Tielu’s (1988) study of milk advertising in the Ontario market. 2. For a clear discussion of the distinction between the absolute- and relative-price versions of the Rotterdam model, see Selvanathan (1989). 3. Technically, Selvanathan’s (1989, p. 218) symmetry condition is ␤ij ≤ ␤ji for all i, j. The more stringent restriction (4c) is consistent with Theil’s (1980) model, which is based on a stronger assumption about how advertising affects marginal utilities than Selvanathan’s. 4. The sample covers a period of substantial changes in the level of soft drink and milk advertising. For example, milk advertising in the early 1980s (prior to the implementation of federal legislation authorizing the nationwide mandatory checkoff) was about $23 million per year; by 1994 it had more than tripled to $79 million. Soft drink advertising, over the same period, increased from $250 million per year to $462 million. No attempt was made in this study to determine whether the large increases in expenditures affect the response coefﬁcients. 5. Technically, deﬂating advertising expenditure is redundant in the Rotterdam model unless advertising homogeneity is rejected (see Selvanathan, 1989). 6. Negative own-advertising elasticities, although counterintuitive, are surprisingly prevalent in the literature. For example, in Green, Carman, and McManus’s (1991) dried fruit study, negative own-advertising elasticities were obtained for one of three

Effects of Advertising on U.S. Non-alcoholic Beverage Demand

21

commodities in their system-wide estimates, and for two commodities in their singleequation (double log) estimates. In Baye, Jansen, and Lee’s (1992) study, four of six own-advertising elasticities are negative. The theoretical explanation offered by Baye, Jansen, and Lee (1992, pp. 1088–1089) is that the observed advertising effect is composed of two separate effects (akin to the income and substitution effects of a price change), the signs of which are ambiguous. 7. If the trend coefﬁcients are not divided by budget shares, they are interpreted as the annual percentage changes in budget shares rather than (per capita) quantities. To see this, deﬁne the Rotterdam model in the absence of changes in other variables as wi d log qi = ai. Expanding the left-hand-side term gives (pi qi/y)(dqi/qi) = d(piqi)/y = dwi. Thus, the original equation is equivalent to dwi = ai. Note that the trend “coefﬁcients” ai must sum to zero to satisfy the budget constraint, but the trend “elasticities” ai/wi = d log qi need not.

ACKNOWLEDGMENTS This paper is an updated and expanded version of H. Xiao, H. W. Kinnucan, and H. M. Kaiser. “Advertising, Structural Change, and U.S. Non-Alcoholic Drink Demand,” NICPRE Research Bulletin 98-02, Cornell University, March 1998. Appreciation is expressed to Noel Blisard, Jong-Ying Lee, and Phil Vande Kamp for comments on an earlier draft, and to Leen Boon for assistance with data collection. Funds supporting this research were provided in part by the National Institute for Commodity Promotion Research and Evaluation. Responsibility for ﬁnal content, however, rests with the authors.

REFERENCES AD $ Summary, various issues for 1970–1995. Leading National Advertisers, Inc., New York. Barnett, W. A. (1979). Theoretical Foundations for the Rotterdam Model. Review of Economic Studies, 46, 109–130. Basmann, R. L. (1956). A Theory of Demand with Variable Consumer Preferences. Econometrica, 24, 47–58. Baye, M. R., Jansen, D. W., & Lee, J-W. (1992). Advertising Effects in Complete Demand Systems. Applied Economics, 24, 1087–1096. Brester, G. W., & Schroeder, T. C. (1995). The Impacts of Brand and Generic Advertising on Meat Demand. American Journal of Agricultural Economics, 77, 969–979. Bronsard, C., & Salvas-Bronsard, L. (1984). On Price Exogeneity in Complete Demand Systems. Journal of Econometrics, 24, 235–247. Brown, M. G., & Lee, J-Y. (1993). Alternative Speciﬁcations of Advertising in the Rotterdam Model. European Review of Agricultural Economics, 20, 419–436. Clarke, D. G. (1976). Econometric Measurement of the Duration of Advertising Effect on Sales. Journal of Marketing Research, 13, 345–357. Deaton, A., & Muellbauer, J. (1980). Economics and Consumer Behavior. Cambridge University Press, Cambridge. Duffy, M. (1987). Advertising and the Inter-Product Distribution of Demand: A Rotterdam Model Approach. European Economic Review, 31, 1051–1070.

22

HENRY W. KINNUCAN ET AL.

Duffy, M. (1989). Measuring the Contribution of Advertising to Growth in Demand: An Econometric Accounting Framework. International Journal of Advertising, 8, 95–110. Duffy, M. (1990). Advertising and Alcoholic Drink Demand in the U.K.: Some Further Rotterdam Model Estimates. International Journal of Advertising, 9, 247–257. Duffy, M. (1991). Advertising in Demand Systems: Testing a Galbraithian Hypothesis. Applied Economics, 23, 485–496. Duffy, M. (1995). Advertising in Demand Systems for Alcoholic Drinks and Tobacco: A Comparative Study. Journal of Policy Modeling, 17, 557–577. Edgerton, D. L. (1997). Weak Separability and the Estimation of Elasticities in Multistage Demand Systems. American Journal of Agricultural Economics, 79, 62–79. EViews User Guide Version 2.0 (1994). Quantitative Micro Software, Irvine, CA. Forker, O. D., & Ward, R. W. (1993). Commodity Advertising: The Economics and Measurement of Generic Programs. Lexington Books, New York. Galbraith, J. K. (1967). The New Industrial State. Houghton Mifﬂin, Boston. Gao, X. M., & Shonkwiler, J. S. (1993). Characterizing Taste Change in a Model of Meat Demand: Correcting for Spurious Regression and Measurement Errors. Review of Agricultural Economics, 15, 313–324. Goddard, E. W., & Tielu, A. (1988). Assessing the Effectiveness of Fluid Milk Advertising in Ontario. Canadian Journal of Agricultural Economics, 36, 261–278. Goldberger, A. S. (1987). Functional Form and Utility: A Review of Consumer Demand Theory. Westview Press, Boulder, CO. Green, R. D., Carman, H. F., & McManus, K. (1991). Some Empirical Methods of Estimating Advertising Effects in Demand Systems: An Application to Dried Fruits. Western Journal of Agricultural Economics, 16, 63–71. Kaiser, H. M. (1997). Impact of National Generic Advertising on Dairy Markets, 1984–1995. Journal of Agricultural and Applied Economics, 29, 303–313. Kaiser, H. M., & Liu, D. J. (1997). The Effectiveness of Generic and Branded Advertising: The Case of U.S. Dairy Promotion. In: J. L. Ferrero & C. Clary (Eds), Economic Evaluation of Commodity Promotion Programs in the Current Legal and Political Environment. National Institute of Commodity Promotion Research and Evaluation, Cornell University. Ithaca, NY. Kinnucan, H. W. (1986). Demographic Versus Media Advertising Effects on Milk Demand: The Case of the New York City Market. Northeastern Journal of Agricultural and Resource Economics, 15, 66–74. Kinnucan, H. W. (1996). A Note on Measuring Returns to Generic Advertising in Interrelated Markets. Journal of Agricultural Economics, 47, 261–267. Kinnucan, H. W., & Ackerman, K. Z. (1995). Public Funding of Foreign Market Development Programs. In: D. Padberg (Ed.), Public Policy in Foreign and Domestic Market Development. Texas A&M University, College Station. Kinnucan, H. W., Xiao, H., & Hsia, C-J. (1996). Welfare Implications of Increased U.S. Beef Promotion. Applied Economics, 28, 1235–1243. Kinnucan, H. W., Xiao, H., Hsia, C-J. & Jackson, J. D. (1997). Effects of Health Information and Generic Advertising on U.S. Meat Demand. American. Journal of Agricultural Economics, 79, 13–23. LaFrance, J. (1991). When is Expenditure ‘Exogenous’ in Separable Demand Models? Western Journal of Agricultural Economics, 16, 49–62.

Effects of Advertising on U.S. Non-alcoholic Beverage Demand

23

Lee, J-Y., & Brown, M. G. (1992). Commodity Versus Brand Advertising: A Case Study of the Florida Orange Juice Industry. In: H. W. Kinnucan, S. R. Thompson & H-S. Chang (Eds), Commodity Advertising and Promotion (pp. 79–100). Iowa State University Press, Ames. Leone, R. P. (1995). Generalizing What is Known About Temporal Aggregation and Advertising Carryover. Marketing Science, 14, G141-G150. Moschini, G., Moro, D., & Green, R. D. (1994). Maintaining and Testing for Separability in Demand Systems. American Journal of Agricultural Economics, 76, 61–73. Mountain, D. C. (1988). The Rotterdam Model: An Approximation in Variable Space. Econometrica, 56, 477–484. Piggott, R. R., Piggott, N. E., & Wright, V. E. (1995). Approximating Farm-Level Returns to Incremental Advertising Expenditure: Methods and Application to the Australian Meat Industry. American Journal of Agricultural Economics, 77, 497–511. Putman, J. J., & Allshouse, J. E. (1996). Food Consumption, Prices, and Expenditures. Washington, D.C.: Statistical Bulletin Number 928. USDA, ERS. Richards, T. J., Van Ispelen, P., & Kagan, A. (1997). A Two-Stage Analysis of the Effectiveness of Promotion Programs for U.S. Apples. American Journal of Agricultural Economics, 79, 825–837. Selvanathan, E. A. (1989). Advertising and Consumer Demand: A Differential Approach. Economics Letters, 31, 215–219. Theil, H. (1965). The Information Approach to Demand Analysis. Econometrica, 33, 67–87. Tomek, W. G., & Cochrane, W. W. (1965). Long-Run Demand: A Concept and Elasticity Estimates for Meats. Journal of Farm Economics, 47, 717–730. Tomek, W. G., & Kaiser, H. M. (1999). On Improving Econometric Analysis of Generic Advertising Effects. Agribusiness, 15, 485–500. USDA, AMS (1997). Report to Congress on the National Dairy Promotion and Research Program and the National Fluid Milk Processor Program. Washington, D.C.: U.S. Department of Agriculture, Agriculture Marketing Service. USDA, ERS (1996). Food Consumption, Prices, and Expenditures. SB-928. Washington, D.C.: U.S. Department of Agriculture, Economic Research Service. USDL, BLS, various issues (1991–1997). CPI Detailed Report. Washington, D.C.: U.S. Department of Labor, Bureau of Labor Statistics. Ward, R. W., & Dixon, B. L. (1989). Effectiveness of Fluid Milk Advertising Since the Dairy and Tobacco Adjustment Act of 1983. American Journal of Agricultural Economics, 71, 730–740. Wohlgenant. M. K., & Clary, C. R. (1993). Development and Measurement of Farm-to-Retail Price Linkage for Evaluating Dairy Advertising Effectiveness. Agricultural Economics Research, 44, 18–27.

24

HENRY W. KINNUCAN ET AL.

APPENDIX A Data, Sources, and Notes Retail price series were developed in a three step-procedure. First, the U.S. city average price of each beverage in December 1995 was obtained from the CPI Detailed Report. These prices were then divided by each beverage’s CPI for December 1995 (1982–1984 = 100) to obtain an average price for the 1982–1984 base period. The base-period prices were then multiplied by each beverage’s annual CPI (1982–1984 = 100) to form the price series. Because the CPI Detailed Report does not list a price for tea, a modiﬁed version of the foregoing procedure had to be used for tea. In addition, unit conversions and other complications arose with the other beverages. Details are provided below. Fluid milk price – The ﬂuid milk price was proxied as a simple average of whole and low-fat milk prices. The December 1995 U.S. city average price for fresh, whole, fortiﬁed milk is $2.518 per gallon; the corresponding price for fresh, low-fat milk is $2.310 per gallon. Applying the foregoing procedure category to the simple average of these two prices yields a base-period price of $1.806 per gallon. Fruit juice price – The price of frozen orange-juice concentrate was taken as a proxy for fruit-juice price. The December 1995 U.S. city average price of frozen orange-juice concentrate was $1.573 per 16 oz. (473 ml). Since one gallon equals 3,800 ml, a gallon of concentrate cost $12.637. Assuming that concentrate is mixed with water in a 3:1 ratio (one part concentrate to 3 parts water), this implies a December 1995 price of $4.212 per gallon drinking juice. Applying the foregoing procedure to this price yields a base-period price of $3.080 per gallon. Soft drink price – The price of regular cola in two liter containers was taken as a proxy for the price of soft drinks. The December 1995 U.S. city average price of regular cola was $0.996. Using the conversion 3.8 liters equals one gallon, this translates into a December 1995 cola price of $1.892 per gallon. Applying the foregoing procedure to this price yields a base-period price for soft drinks of $1.597 per gallon. Coffee price – The price of coffee was measured as the simple average of instant and ground roast coffee price. The December 1995 U.S. city average price of instant coffee is $10.299 per pound. Each pound of instant makes approximately 186.8 cups of 6.0 oz. ﬂuid coffee or 8.759 gallons. Thus, the December 1995 price of instant coffee is $1.176 per gallon drinking coffee. The December 1995 U.S. city average price of ground roast coffee is $3.507

Effects of Advertising on U.S. Non-alcoholic Beverage Demand

25

per pound. Ground roast coffee makes approximately 59.8 cups of 6.0 oz. ﬂuid coffee, or 2.803 gallons. So the December 1995 price of ground roast coffee is $1.251 per gallon drinking coffee. Applying the foregoing procedure to the simple average of the instant and ground-roast prices yields a base period price for coffee of $0.777 per gallon. Tea price – The tea price series was complicated by the fact that the CPI Detail Report does not list a price for tea andceased publishing a price index for tea in 1977. The latter problem was solved by constructing a price index (1982 = 100) for the period 1975–1995 based on data in Tropical Products: World Markets and Trade (pp. 36–37) provided by the International Tea Committee (ITC). This index was spliced to the USDL’s tea index to obtain an index for the entiresample period 1970–1994. To convert the index to actual prices, the price of tea in 1978 was obtained from ITC data published in Estimated United States Average Retail Price of Food, which lists an average price for tea in 1978 of $1.235 for tea bags, 40-bag package. Assuming that each tea bag produces approximately 7.2 oz. of tea, this translates into 2.242 gallons of liquid tea per package, or a 1978 price of $0.551 per gallon. Dividing this price by the CPI for tea in 1978 (1982 = 100) provides an estimate of the base-period price. Multiplying the base-period price by the annual CPI for tea (1982 = 100) provided the tea price series. The composite price series for coffee and tea was constructed as a quantity-share weighted average of the foregoing tea and coffee prices. The price and quantity series used in this study are given in Table A1. The advertising series and related data are given in Table A2. Basic data sources for the nonprice series and special notes are as follows. (1) q1 to q4 – The source for the quantity data is Putman and Allshouse (1996, Table 37). (2) q1 to p4 – The basic data source for the price series was the U.S. Department of Labor, CPI Detailed Report (various years), which reports average retail food prices for U.S. cities and four regions. This source, however, does not list a price for tea. The sources and methods used to obtain a tea price series are provided in the appendix narrative. (3) a1 to a4 – The basic source for the advertising data is AD $ SUMMARY published by Leading National Advertisers, Inc. (LNA). The relevant LNA categories are as follows: F131 (milk, butter, eggs), F171 (coffee, tea, and cocoa), F172 (fruit drinks), F221 (regular soft drinks), F222 (diet soft drinks), and F223 (non-carbonated soft drinks). Because of deﬁnitional changes and aggregation, several adjustments had to be made before these data could be used for analysis. First, in 1984 LNA broadened the Juice category (F172) to include powdered drinks, which was formerly in the F223 category. At the same time, LNA added a new category (F224), bottled water, which was formerly in F223. Since it was not possible

26

HENRY W. KINNUCAN ET AL.

to isolate the proportion of F172 expenditures that is strictly juice advertising in the redeﬁned series, it was decided that the best approach was simply to add the three categories. That is, in our study, fruit-juice advertising is measured as F172 + F223 + F224. The second adjustment has to do with the F131 category. This category includes expenditures for butter and eggs as well as ﬂuid milk. To isolate the milk expenditures, we collected data for F131 “brands” as follows: National Dairy Board, California Milk Advisory Board, American Dairy Association, United Dairy Industry Association, Mid-Atlantic Farmers’ Milk, Dairymans’ Dairy Products, and Cow Dairyman Association. Thus, the data for ﬂuid milk advertising used in this study refer strictly to generic advertising expenditures as reported by LNA. (The series excludes expenditures by the newly-formed Fluid Milk Processors’ Board as that campaign commenced in 1995, a year later than our sample period.) The third adjustment has to do with missing values. Data prior to 1974 for juices, soft drinks and coffee and tea were unavailable. For milk, no data were available for 1974 and 1975. The latter two data points were obtained by interpolation. For the other beverages, the missing values were “backcast” from the regression equation ADit = ␣ + ␤t + ␥t2 + ␧t, where ADit is the total advertising expenditure for good i in period t as reported by LNA, and t is atrend variable that assumes the values 5, 6, . . . , 10 for 1974–1983. The regressions were run on the combined juice series F172 + F223 + F224, the combined soft-drink series F221 + F222, and the single series F171 for coffee and tea. The missing values for 1970–1973 were computed from the estimated regressions by setting t = 1, 2, 3, and 4, respectively, and computing ADit, when the residuals are zero. Other variables in the study are: AGE – The proportion of the U.S. population less than age ﬁve was obtained from Table B-30 in Economic Report to the President, p. 315. FAFH – Expenditures on food-away-fromhome divided by expenditures on food-at-home, and the data source is Putman and Allshouse (1996, Table 98, p. 36). POP – Resident U.S. population on July 1. Source: Putman and Allshouse (1996, Table 115). CPI – Consumer Price Index for all items for all urban consumers. Source: Putman and Allshouse (1996, Table 101).

Effects of Advertising on U.S. Non-alcoholic Beverage Demand

27

Table A1. Quantity and Retail Price Data for U.S. Non-Alcoholic Beverages, 1970–1994. YEAR 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994

q1 q2 q3 q4 ( – – – – – Gallons/person – – – – –) 31.3 31.3 31.0 30.5 29.5 29.5 29.3 29.0 28.6 28.2 27.6 27.1 26.4 26.3 26.4 26.7 26.5 26.3 25.8 26.0 25.7 25.7 25.4 24.9 24.7

5.7 5.7 6.2 6.0 6.0 6.6 6.9 7.0 6.4 6.8 7.2 7.4 6.8 8.4 7.3 7.7 7.9 8.2 8.2 7.7 6.9 7.9 7.3 8.4 8.6

24.3 25.5 26.2 27.6 27.6 28.2 30.8 33.0 34.2 34.7 35.1 35.4 35.3 35.2 35.9 35.7 35.8 41.9 44.7 45.4 46.3 47.9 48.5 50.2 52.2

40.2 40.4 40.9 40.7 40.7 38.9 40.2 32.0 34.5 36.2 34.0 33.2 32.8 33.3 33.9 34.5 34.6 33.6 32.6 33.0 33.6 33.6 32.9 30.5 28.1

p1 p2 p3 p4 ( – – – – – Dollars/gallon – – – – –) 0.904 0.928 0.942 1.031 1.235 1.236 1.301 1.314 1.390 1.551 1.688 1.783 1.793 1.805 1.819 1.847 1.836 1.871 1.914 2.064 2.288 2.210 2.282 2.309 2.369

1.291 1.341 1.434 1.445 1.497 1.616 1.655 1.993 2.114 2.311 2.473 2.833 2.965 2.973 3.248 3.412 3.239 3.377 3.788 3.905 4.313 4.080 4.270 4.040 4.059

0.609 0.644 0.656 0.674 0.834 1.026 0.994 1.041 1.131 1.234 1.383 1.522 1.562 1.602 1.626 1.642 1.654 1.688 1.688 1.731 1.790 1.805 1.835 1.851 1.848

0.250 0.258 0.255 0.279 0.327 0.358 0.475 0.830 0.774 0.744 0.801 0.696 0.707 0.729 0.813 0.754 0.919 0.806 0.799 0.847 0.833 0.809 0.785 0.765 0.934

Note: The subscripts are deﬁned as follows: 1 = ﬂuid milk, 2 = juices, 3 = soft drinks, and 4 = coffee and tea. See appendix narrativefor sources and explanatory notes.

28

HENRY W. KINNUCAN ET AL.

Table A2. Beverage Advertising and Remaining Data Series, 1970–1994. YEAR 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994

A1 A2 A3 A4 (– – – – – Million dollars – – – – –) 1.903 4.246 11.346 12.101 12.853 13.691 14.529 16.239 15.948 19.144 22.256 22.747 25.643 27.302 4.956 23.056 55.795 54.969 54.844 59.867 28.369 31.653 28.882 72.954 78.969

9.308 21.117 32.924 44.732 31.332 49.711 77.690 83.426 110.466 122.060 120.370 139.172 115.962 148.523 195.280 187.830 186.050 211.660 229.300 246.159 262.460 239.874 228.528 224.990 266.681

24.173 48.346 72.519 96.692 97.004 108.654 135.078 134.631 179.964 237.990 252.695 238.063 257.707 321.234 362.288 384.472 392.375 389.182 457.548 428.224 497.875 477.846 470.847 434.422 462.122

19.711 39.422 59.133 78.844 73.857 89.378 116.005 109.883 167.732 211.525 226.367 226.148 237.892 214.831 244.390 239.620 231.370 215.830 282.690 317.255 340.450 264.986 253.787 260.885 280.454

AGE (%)

FAFH (Ratio)

POP (Thous.)

CPI

8.372 8.304 8.147 7.952 7.709 7.464 7.163 7.067 7.069 7.137 7.224 7.346 7.420 7.489 7.487 7.482 7.464 7.435 7.426 7.483 7.542 7.599 7.637 7.627 7.571

0.356 0.360 0.371 0.375 0.365 0.398 0.427 0.444 0.465 0.474 0.476 0.502 0.527 0.546 0.555 0.561 0.578 0.595 0.608 0.599 0.591 0.589 0.615 0.649 0.666

203984 206827 209284 211357 213342 215465 217563 219760 222095 224567 227225 229466 231664 233792 235825 237924 240133 242289 244499 246819 249402 252131 255028 257783 260341

0.388 0.405 0.418 0.444 0.493 0.538 0.569 0.606 0.652 0.726 0.824 0.909 0.965 0.996 1.039 1.076 1.096 1.136 1.183 1.240 1.307 1.362 1.403 1.445 1.482

Note: The subscripts are deﬁned as follows: 1 = ﬂuid milk, 2 = juices, 3 = soft drinks, and 4 = coffee and tea. See appendix narrative for sources and explanatory notes.

Effects of Advertising on U.S. Non-alcoholic Beverage Demand

29

APPENDIX B Elasticity Formula Conditional Elasticities The conditional income elasticity, which is also the ratio of the individual beverage income elasticity to the group income elasticity, is computed from the second-stage “income” parameter and conditional and unconditional budget shares as follows: MitC = ␮⬘iSGt/sit = [(␮i/␮G)SGt]/sit = (␮i/sit)/(␮G/SGt) The remaining conditional elasticities are computed strictly from second-stage parameters and conditional budgets shares as follows: NijtC = ␲ij/sit

(Conditional Hicksian price elasticities)

B = ␤ij/sit

(Conditional advertising elasticities)

TitC = (ai/sit) ⫻ 100

(Conditional trend “elasticities”)

C ijt

C it

V = (bi/sit)

(Conditional age elasticities)

HitC = (ci/sit)

(Conditional food-away-from-home elasticities)

Unconditional Elasticities The corresponding unconditional elasticities are calculated from ﬁrst- and second-stage parameters and conditional budget share as follows (for complete derivations, see Duffy, 1987): MitU = ␮i/sit = ␮⬘i ␮G/sit NijtU = (␲ij + ␮⬘i ␮⬘j␲G)/sit

i, j = 1, . . . , 4

BijtU = (␤ij + ␮⬘i ␮⬘j␤G)/sit

i, j = 1, . . . , 4

T = 100(ai + ␮⬘i aG)/sit U it

VitU = (bi + ␮⬘i bG)/sit HitU = (ci + ␮⬘i cG)/sit The foregoing elasticities correspond to the within-group effects. Two additional cross-elasticities are available that tell the effect of non-group price and non-group advertising on within-group demands. These elasticities are computed as: U = ␮⬘i ␲O/sit NiOt U BiOt = ␮⬘i␤O/sit.

This Page Intentionally Left Blank

THE LONG-RUN DEMAND FOR ALCOHOLIC BEVERAGES AND THE ADVERTISING DEBATE: A COINTEGRATION ANALYSIS N. Edward Coulson, John R. Moran and Jon P. Nelson ABSTRACT Time-series models of the demand for alcoholic beverages have been criticized for use of annual data; omitted variables; mis-measurement of advertising; simultaneous equations bias; and inadequate attention to nonstationarity and dynamics. This paper reappraises the relationship between alcohol advertising, price, and consumption in a manner which speaks to these issues. Using quarterly data from 1970 : 1 to 1990 : 4 on three beverages (beer, wine, and distilled spirits), we ﬁnd evidence of cointegration between beverage consumption, prices, advertising, and real income. Elasticities obtained from the estimated cointegrating vectors indicate that long-run beverage demands are both price and income inelastic. Moreover, after correcting for each of the problems described above, advertising has virtually no inﬂuence on the steady-state level of alcoholic beverage consumption.

Advertising and Differentiated Products, Volume 10, pages 31–54. Copyright © 2001 by Elsevier Science Ltd. All rights of reproduction in any form reserved. ISBN: 0-7623-0823-0

31

32

N. EDWARD COULSON, JOHN R. MORAN AND JON P. NELSON

I. INTRODUCTION Alcohol consumption and intoxication are associated with a number of undesirable phenomena and behaviors, including drunk driving, crime, alcoholism, suicide, and a variety of alcohol-related health problems. In many instances, the adverse consequences of alcohol use extend beyond the individual consumer, creating social costs which exceed those internalized by the user. For this reason, much attention has been focused on the determinants of alcohol consumption, with particular attention being paid to policy-relevant variables such as price, advertising, and the legal drinking age.1 Unfortunately, the empirical literature on alcohol demand has provided weak and often contradictory evidence on the sensitivity of consumption to these variables (Leung & Phelps, 1993; Saffer, 1993; Cook & Moore, 2000). In part, this lack of consensus reﬂects the inadequacy of available data. Many studies, for example, are based on a small number of annual observations.2 These studies may be prejudiced toward a ﬁnding of modest price and advertising effects, both because of the small sample sizes themselves and because of the inability of yearly observations to pick up higher frequency variation in the data. This view was articulated by Berndt (1991, p. 392), who observed that, “. . . since it is now widely believed that the ‘true’ 90% duration interval of advertising for most established products is a matter of months and certainly is less than a year, one simply should not use annual data to estimate the parameters of the sales-advertising relationship . . . . Further, use of annual data can yield spurious results.” Moreover, in the majority of cases, data on advertising is either not available, or suffers from considerable measurement error.3 The omission of advertising variables from econometric models of alcohol demand not only fails to address the possibility of curbing consumption through advertising restrictions, but also potentially biases the estimated price elasticities. A different, but potentially more fundamental, problem is methodological in nature. Despite the time-series orientation of much of the alcohol demand literature, few researchers have explored the possibility that the variables of interest share a common trend or, in the terminology of Engle and Granger (1987), form a cointegrating relationship. The cointegration/error-correction mechanism approach offers several practical advantages for the estimation of demand elasticities. First, it provides a uniﬁed framework for estimating both short- and long-run elasticities. This distinction is important because demand theory predicts that long-run effects can differ substantially from short-run (impact) responses. Second, in the presence of cointegration, the errorcorrection representation is the only dynamic model of the data-generating

The Long-run Demand for Alcoholic Beverages and the Advertising Debate

33

process that is not misspeciﬁed; other commonly-used speciﬁcations, including models in ﬁrst differences, are potentially biased by the omission of the errorcorrection term. Finally, the presence of cointegration allows one to estimate long-run coefﬁcients in a manner that is free of simultaneous-equations bias (Engle & Granger, 1987; Stock & Watson, 1988).4 This result is of particular importance for alcohol demand studies, which have traditionally employed either single-equation methods or the differential demand system models of Theil (1975/1976) and others. These approaches are problematic in that both methods treat endogenously determined variables (price and advertising) as exogenous regressors.5 Addressing this issue through simultaneous equations methods requires correct speciﬁcation of a structural model, a task which has often proven difﬁcult in practice (Berndt, 1991). The objective of the present study is to reappraise the relationship between alcohol advertising, price, and consumption in a manner which speaks to the issues raised above. Data and measurement issues are addressed through the use of a data set containing quarterly observations on advertising, price, and consumption for three alcoholic beverages (beer, wine, distilled spirits) over the period 1970 : 1 to 1990 : 4. The data set has the additional advantage of having both advertising expenditures and the price of advertising disaggregated by media. This allows for media-speciﬁc deﬂation (which provides an adjustment for the effectiveness of each media), and for the measurement of advertising on a message impressions, as opposed to expenditure, basis. Based on earlier ﬁndings documenting the importance of demographic changes in determining alcohol consumption, we also include two variables designed to capture agerelated differences in the propensity to consume alcohol: (1) the proportion of the population aged 18–29; and (2) the proportion of the population aged 65 and over. In applying a cointegration/error-correction mechanism approach, our paper is complementary to existing work by Johnson et al. (1992), Blake and Nied (1997), and Salisu and Balasubramanyam (1997), who use similar techniques to study alcohol demand in Canada and the United Kingdom. In contrast to these studies, the present study focuses on the United States and generalizes upon previous research through its inclusion of quarterly data on advertising and use of demographic variables. The remainder of the paper is organized as follows. Section II provides a brief summary of the alcohol demand literature, with a focus on advertising. Section III describes the data and documents its relevant properties, presenting evidence that beverage-level consumption is cointegrated with beverage prices, advertising levels, and real income. Section IV estimates long-run demand functions for each beverage, and provides estimates of the associated long-run price, income, and advertising elasticities. Our ﬁndings indicate that long-run

34

N. EDWARD COULSON, JOHN R. MORAN AND JON P. NELSON

beverage demand is both price and income inelastic. The long-run inﬂuence of advertising is notably small, particularly for beer and spirits. We also report estimates of the associated error-correction models, ﬁnding little in the way of short-run dynamics, but signiﬁcant contemporaneous effects associated with changes in the age composition of the population. Section VI contains concluding remarks on the alcohol advertising debate.

II. ALCOHOL ADVERTISING AND DEMAND: A BRIEF SURVEY Empirical research on alcohol use and abuse has grown substantially in the past decade (for surveys, see Leung & Phelps, 1993; Cook & Moore, 2000), but research on alcohol advertising has lagged behind. Incorporating and measuring the quantitative effects of advertising has proven difﬁcult for three reasons. First, readily available data on advertising tends to be limited to aggregate annual or quarterly time-series for speciﬁc brands or product categories. These data are also costly to obtain, except at the aggregate annual level. The aggregate nature of the data has ruled out incorporating advertising into microdata studies of alcohol use and abuse. Recent data innovations include the use of brand-speciﬁc data (Gius, 1996); the use of quarterly data by metropolitan area (Saffer, 1997); and the use of quarterly data by media (Nelson, 1999). The present study adds to these data developments by examining the time-series properties of quarterly data. Second, the response model underlying the measurement of advertising is controversial. Many past studies ﬁnd little or no effect of advertising on the total consumption of alcohol, and modest effects at the beverage level (see footnote 2 above). Saffer (1993) argues that these null results reﬂect measurements taken along the ﬂat portion of the advertising-response function, although he cites only brand-speciﬁc studies to support his general view of a sigmoid-shaped response function. In order to capture higher-frequency variations in alcohol use and advertising, Saffer proposes the use of crosssectional data on advertising bans (Saffer, 1991; Young, 1993; Nelson & Young, 2001); pooled cross-section time-series data (Ornstein & Hanssens, 1985; Goel & Morey, 1995); and quarterly or monthly data (Nelson, 1999; Lariviere et al., 2000). The shape of the product level advertising-response function is an open question, and has been the subject of considerable scrutiny in the past (Assmus et al., 1984; Berndt, 1991; Schmalensee, 1972). The present study avoids the use of annual data that has plagued many past studies. Third, given the difﬁculties of data and modeling, most studies have resorted to relatively simple model speciﬁcations involving alcohol prices, income,

The Long-run Demand for Alcoholic Beverages and the Advertising Debate

35

demographic variables and, depending on the data, selected legal variables such as the minimum legal drinking age or restrictive advertising laws at the state and local level (billboard bans, bans on price advertising). While each individual study makes some progress on resolving the issues of data and econometric methods, there are still several unresolved issues. For example, Goel and Morey (1995) use pooled time-series data for distilled spirits consumption by state, and attempt to account for the lag structure on advertising. However, they use aggregate annual data for advertising, which is the same across all cross-sectional units, and they do not report the method of deﬂation for advertising expenditures. Saffer (1997) measures advertising at the metropolitan level, but studies its effect on trafﬁc fatalities rather than consumption, and does not attempt to account for lag structures. Finally, neither Goel and Morey (1995) nor Saffer (1997) analyze the time-series properties of their data, which leaves open the question of whether key variables like consumption and advertising are nonstationary and/or cointegrated. Saffer (1991) argues that past studies in this area are subject to a host of econometric problems, including use of annual data; omitted variables; mismeasurement of advertising; and simultaneous equations bias. We agree, and would add to this list the potential problems associated with nonstationary data and an inability to adequately capture the dynamics of the consumption/ advertising relationship. Our data set and empirical methodology can be used to rectify many of these problems.

III. DATA DESCRIPTION AND COINTEGRATION TESTS Data Description For each beverage, the basic data consists of quarterly observations for the period 1970 : 1 to 1990 : 4 on: (1) nominal price; (2) nominal advertising expenditures (by media); and (3) physical consumption.6 The basic data are seasonally unadjusted. The beverage-level data is supplemented with quarterly data on disposable personal income and annual data on the U.S. population aged 16 + , 18–29, and 65 + . The population series were converted to quarterly observations through linear interpolation. Nominal disposable income was deﬂated by the CPI, as were each of the beverage price indices. Annual measures of the price of advertising in each media (cost per thousand message impressions) were used to convert each media expenditure series to a message impressions basis. This method of deﬂation follows Schmalensee (1972). The media-level advertising series associated with each beverage were then added

36

N. EDWARD COULSON, JOHN R. MORAN AND JON P. NELSON

together to obtain beverage-level measures of real advertising. As discussed in the introduction, this procedure allows for consistent aggregation by providing an adjustment for the effectiveness of each media. The consumption, real advertising, and real disposable income series were converted to per capita magnitudes using the population aged 16 + . Summary statistics for each variable are displayed in Table 1. Note that per capita consumption grew during the sample period, but the mean growth rates for prices and advertising were negative. To facilitate estimation of various elasticities, all non-population variables were converted to natural logarithms. The two demographic variables, the number of persons aged 18–29 and the number aged 65 and over, were divided by the population aged 16 + to obtain measures of the relative sizes of two segments of the drinking-age population whose propensity to consume alcoholic beverages differs substantially from that of the general population. Additional information on data sources and variable construction methods is provided in Appendix A. Unit Root Tests Preliminary plots of the data revealed a strong seasonal component in the advertising and consumption variables, though not in the real income or price Table 1. Descriptive Statistics. Variable Spirits Consumption Beer Consumption Wine Consumption Spirits Price Beer Price Wine Price Spirits Advertising Beer Advertising Wine Advertising Real Income Population 18–29 Population 65 +

Mean

Std. Dev.

Minimum

Maximum

0.0004 0.0027 0.0076 –0.0076 –0.0036 –0.0045 –0.0062 –0.0001 –0.0006 0.0027 0.2779 0.1504

0.2141 0.1438 0.2452 0.0125 0.0099 0.0106 0.3645 0.3091 0.9268 0.0117 0.0120 0.0066

–0.3536 –0.2413 –0.5118 –0.0280 –0.0263 –0.0278 –1.0064 –0.6561 –1.6362 –0.0339 0.2485 0.1416

0.2999 0.2697 0.5469 0.0801 0.0231 0.0272 0.6238 0.5409 1.6064 0.0421 0.2914 0.1630

Notes: For illustrative purposes, all non-population variables are expressed as quarterly growth rates. Consumption variables are measured in ﬂuid gallons per capita; prices are beverage price indices deﬂated by the CPI; advertising is measured in per capita messages; and real income is per capita real disposable personal income.

The Long-run Demand for Alcoholic Beverages and the Advertising Debate

37

series. In general, a series with a seasonal component may be seasonally integrated, have a deterministic seasonal, a stationary seasonal, or any combination thereof. Moreover, the possibility of unit roots at seasonal frequencies invalidates the standard tests for unit roots at the zero frequency. Fortunately, a uniﬁed test for unit roots at all frequencies (zero, semiannual, and annual) has been developed by Hylleberg et al. (1990). This test, labeled HEGY, can be implemented in the regression form ␸(B)y4,t = ␲1 y1,t ⫺ 1 + ␲2 y2,t ⫺ 1 + ␲3 y3,t ⫺ 2 + ␲4 y3,t ⫺ 1 + ␧t

(1)

where B is the backshift operator, ␸(B) is a polynomial in the backshift operator, xt is the original series, and ␧t is an error term. Using xt, the remaining variables are deﬁned as y1,t = (1 + B + B 2 + B 3)xt y2,t = ⫺ (1 ⫺ B + B 2 ⫺ B 3)xt y3,t = ⫺ (1 ⫺ B 2)xt y4,t = (1 ⫺ B 4)xt Given (1), the presence of unit roots at the zero, semiannual, and annual frequencies can be deduced based on tests of ␲1 = 0, ␲2 = 0, and ␲3 = ␲4 = 0, respectively. A rejection of each restriction indicates the absence of a unit root at the corresponding frequency. Alternative forms of the test allow for the presence of an intercept, a deterministic trend, and deterministic seasonals. Note that, as with the more familiar Dickey-Fuller and augmented DickeyFuller tests, the HEGY test requires that the error term in (1) be white noise. To ensure this, lags of y4,t were added sequentially (beginning with zero) until all Ljung-Box Q-statistics (at lags 1–8) were insigniﬁcant at the 0.10 level. To choose among the various forms of the test, we began with the most general model (i.e. one containing an intercept, trend, and seasonal dummies), and then “tested down” until the appropriate speciﬁcation was determined.7 Results of the HEGY test, which are reported in Table 2, show only weak evidence of unit roots at nonzero frequencies. For three of the series (spirits consumption, beer consumption, wine advertising), the test rejects the presence of both annual and semiannual unit roots at the 0.05 level or better, while for two of the series (spirits advertising, beer advertising), the presence of either the annual or semiannual root is rejected at the 0.05 level or better, with some evidence against the other root being present as well.8 In contrast, the null hypothesis of a unit root at the zero frequency is never rejected. We therefore proceed on the assumption that the variables in question are integrated, if at all, only at the zero frequency. To

38

N. EDWARD COULSON, JOHN R. MORAN AND JON P. NELSON

Table 2. HEGY Unit Root Tests.

Spirits Consumption Spirits Advertising Beer Consumption Beer Advertising Wine Consumption Wine Advertising

Test Version

Lags

T-stat· ␲1 = 0

T-stat· ␲2 = 0

F-stat· ␲3 = ␲4 = 0

C, S, T C, S, T C, S C, S C C, S, T

0 0 0 2 2 0

–1.87 –0.86 –2.80 –1.46 –2.41 0.13

–3.27** –2.27 –4.86*** –4.33*** –0.81 –4.24***

15.23*** 17.30*** 9.17** 4.47 0.85 7.55**

Notes: C, T, and S denote the presence of a constant term, a trend term, and seasonal dummies, respectively. ** and *** indicate signiﬁcance at the 0.05 and 0.01 levels, respectively.

control for seasonal inﬂuences, the consumption and advertising variables were ﬁltered by regressing each series on four seasonal dummies, taking the resulting residuals as the new series. For these variables, all subsequent analysis is based on the ﬁltered data.9 Having rejected the presence of unit roots at nonzero frequencies, tests for unit roots at the zero frequency were conducted by running standard augmented Dickey-Fuller (ADF) tests on each series; ﬁrst in levels and then in ﬁrst differences. The appropriate test version was selected on the basis of the signiﬁcance of the intercept and trend terms (using a 0.10 signiﬁcance level). Lagged values of the dependent variable were added until the residuals were approximately white noise (as indicated by the absence of a signiﬁcant LjungBox Q-statistic at lags 1–8). The results of these tests, which are shown in Table 3, indicate that all of the non-population variables are I(1).10 Cointegration Given our data and the issues to be addressed, a demand model is estimated that includes explanatory variables for own- and cross-advertising, own- and crossprices, and real per capita income. In order to obtain valid estimates for each beverage (i = S, B, W ), we ﬁrst tested for cointegrating, or long-run equilibrium, relationships of the form Qi = ␮ + ␤1AS + ␤2AB + ␤3AW + ␤4PS + ␤5PB + ␤6PW + ␤7R

(2)

where Qi is per capita consumption of beverage i, Ai is the per capita number of advertising messages generated for beverage i, Pi is the relative price of the i th beverage, and R is per capita real income.11 If a group of variables

The Long-run Demand for Alcoholic Beverages and the Advertising Debate

39

Table 3. ADF Unit Root Tests.

Spirits Consumption Spirits Advertising Spirits Price Beer Consumption Beer Advertising Beer Price Wine Consumption Wine Advertising Wine Price Real Income

Test Version

Lags

t-stat

C, T C, T C, T C, T C C, T C C, T – – C C C C, T C, T C, T C, T C – C

3 2 3 2 1 0 3 2 5 4 0 0 5 2 3 2 0 0 0 0

–1.88 –10.05*** –0.86 –8.92*** –2.32 –7.15*** –2.92 –11.62*** –1.50 –2.79*** –1.45 –7.72*** –2.41 –24.14*** 0.12 –10.31*** –2.86 –8.38*** –2.13** –8.18***

Levels 1st Diffs Levels 1st Diffs Levels 1st Diffs Levels 1st Diffs Levels 1st Diffs Levels 1st Diffs Levels 1st Diffs Levels 1st Diffs Levels 1st Diffs Levels 1st Diffs

Notes: C and T denote the presence of a constant and a trend term, respectively, while “ – ” indicates that the test was run with neither a constant nor a trend. ** and *** indicate signiﬁcance at the 0.05 and 0.01 levels, respectively.

(y1, y2, . . . , yn) is cointegrated, then the data generating process takes the form of a vector error-correction (VEC) model written as

冘 p

⌬Yt = ␪0 + ⌸Yt ⫺ 1 +

␪j ⌬Yt ⫺ j + ␧t

(3)

j=1

where Yt ⬅ (y1t, y2t, . . . , ynt)⬘ is an (n ⫻ 1) vector, ⌬ is a ﬁrst-difference operator, ␪0 is an (n ⫻ 1) vector of constants, ␪j ( j = 1, 2, . . , p) and ⌸ are (n ⫻ n) matrices, and ␧t is an (n ⫻ 1) vector of white noise errors (Engle & Granger, 1987). Note that the ⌸ matrix shown in (3) can be expressed as ⌸ = ␣␤⬘, where ␤⬘ is a collection of stacked cointegrating vectors representing the long-run relationships among the variables, and ␣ is a vector of short-run “speed of adjustment” coefﬁcients measuring the speed with which the system returns to equilibrium.

40

N. EDWARD COULSON, JOHN R. MORAN AND JON P. NELSON

Given (3), two tests for cointegration can be conducted based on the eigenvalues of ⌸ (Johansen, 1988, 1991). The ﬁrst, the trace test, can be used for testing the null hypothesis that the number of cointegrating relationships is less than or equal to r against a general alternative. The test statistic is given by

冘 n

␭trace(r) = –T

ln(1 ⫺ ␭i)

(4)

i=r+1

where T is the number of usable observations and ␭1 > ␭2 > . . . > ␭n are the eigenvalues from the estimated ⌸ matrix. Alternatively, one might wish to test the null hypothesis of r cointegrating relationships against the speciﬁc alternative of r + 1. This can be done using the ␭max statistic ␭max(r, r + 1) = –T ln(1 ⫺ ␭r + 1)

(5)

Because the maximum eigenvalue test has a sharper alternative hypothesis, it is generally preferred to the trace test and will be used in what follows. To select the appropriate number of lags for (3), we ran a test-VAR in levels for each of the three beverages, and computed the multivariate Schwarz Information Criterion (SIC) for all lag lengths from 1 to 6. In each case, the SIC was minimized at a lag length of 1, implying that no lagged difference terms should be included in the VEC. However, because this ﬁnding may be due to the large number of variables considered, we also estimated a VEC containing a lagged difference of each explanatory variable. Results from the maximum eigenvalue test are presented in Table 4. Evidence in support of a long-run relationship between beverage-level consumption, prices, advertising, and real income is found for all three beverages in both of the test speciﬁcations considered. As discussed in the introduction, this raises the possibility that the estimated price and advertising

Table 4. Johansen Maximum Eigenvalue Test (# of Cointegrating Vectors). # of Lags in VEC

Spirits

Beer

Wine

0 1

3 1

4 3

1 3

Notes: We take r to be the number of cointegrating vectors if the ␭max test rejects the alternative hypothesis of r + 1 cointegrating vectors at the 0.05 level or better.

The Long-run Demand for Alcoholic Beverages and the Advertising Debate

41

elasticities from previous alcohol demand studies, which have generally estimated models in ﬁrst differences, may be biased by the omission of the error correction terms. In the next section, we will jointly estimate the longand short-run demand relationships for each beverage using a vector error-correction model. This will allow us to separately estimate both long-run and short-run demand elasticities, and to circumvent biases associated with simultaneity among the variables.

IV. LONG-RUN BEVERAGE DEMAND In this section, we estimate the long-run demand for each beverage using the full-information maximum likelihood estimator derived by Johansen (1988, 1991). Estimates of the corresponding long-run demand elasticities are obtained, as are estimates of the associated short-run responses. In the interest of space, only the main results from the VEC models will be discussed. Long-Run Demand Elasticities For each beverage, two VEC models were estimated: one without lagged difference terms, and another containing a single lagged difference of each explanatory variable. The associated cointegrating vectors are reported in Table 5.12 Because all non-population variables have been converted to natural logarithms, the estimated coefﬁcients also are the long-run elasticities of demand. In discussing the elasticities in Table 5, we focus on the “zero-lag’’ speciﬁcation favored by the SIC although, in general, the estimates from the single-lag model are quite similar, especially for the own-price, ownadvertising, and income elasticities emphasized in the discussion below. The cross-elasticities for price or advertising will be discussed only if they are particularly noteworthy. Spirits demand: The demand for spirits is own-price inelastic (elasticity = –0.33) and spirits is a substitute for wine (0.50). The demand for spirits is also income inelastic (0.41). The own-advertising elasticity is only 0.09, and spirits demand does not appear to be appreciably affected by either beer or wine advertising (elasticities = 0.03, –0.03). Beer demand: Beer is own-price inelastic (–0.27). Both of the estimated income elasticities for beer have negative signs. Our observations on this result are given below. The ownadvertising elasticity for beer is only 0.03. Wine demand: The own-price elasticity is –0.59 and the income elasticity is 0.76. The own-advertising

42

N. EDWARD COULSON, JOHN R. MORAN AND JON P. NELSON

Table 5. Long-Run Demand Equations. Spirits

Beer

Wine

Explanatory Variables

0 lag

1 lag

0 lag

1 lag

0 lag

1 lag

Intercept Spirits Price

–0.790 –0.329

–0.448 –0.344*** (0.118) 0.141 (0.123) 0.571*** (0.132) 0.224*** (0.101) 0.045* (0.024) 0.043*** (0.013) –0.039*** (0.007)

0.540 0.023

0.334 –0.073

–1.036 –0.500

–0.274

–0.284

0.162

0.305

–0.270

–0.232

–0.011

–0.024

0.027

0.036

0.001

0.001

–1.491 0.008 (0.291) 0.175 (0.334) –0.593* (0.322) 0.760*** (0.271) 0.236*** (0.060) –0.080** (0.033) 0.189*** (0.017)

Beer Price

0.038

Wine Price

0.501

Real Income

0.406

Spirits Advertising

0.085

Beer Advertising

0.028

Wine Advertising

–0.026

0.120 –0.414 0.512 0.006 –0.085 0.173

Notes: Asymptotic standard errors are shown in parentheses for the two cases where a single cointegrating vector is present. When multiple cointegrating vectors are present, there is no readily-available method for calculating the correct standard errors for the chosen vector and normalization, so standard errors are omitted in these cases. When using Johansen’s maximum likelihood procedure, it is not possible to construct ﬁnite sample conﬁdence intervals using the t-distribution. Instead, the standard normal distribution must be used for determining statistical signiﬁcance. The labels “0 lag” and “1 lag” denote the number of lagged differences of each explanatory variable that were included in the associated error correction model. *, **, and *** indicate signiﬁcance at the 0.10, 0.05, and 0.01 levels, respectively.

elasticity for wine is about 0.19, which is slightly larger than the elasticities for beer and spirits. Table 6 presents a summary of alcohol demand elasticities obtained in the present paper and eight previous empirical studies. As shown in the table, our estimates of the long-run price and income elasticities are generally smaller (in absolute value) compared to those reported by other studies covering Canada, the U.K., and the U.S. This difference may be attributable to a variety of factors, including the time period studied,13 the use of quarterly as opposed to annual data (Bass & Leone, 1983; Leone, 1995), or the inclusion of a more complete set of explanatory variables than has typically been used in previous time series studies. Overall, the demand for alcoholic beverages is price inelastic, with wine having the largest elasticity,

Own-Price Study

Income

Own-Advertising

Country

Time Time Period

Beer

Wine

Spirits

Beer

Wine

Spirits

Beer

Wine

Spirits

Canada U.K. U.K. U.S. U.S. –

1956–83 1963–87 1963–92 1954–85 1964–90 –

–0.14 –0.39 –0.95 –0.71 –0.37 –0.51

–1.17 –1.23 –0.93 –1.11 –0.18 –0.92

0.37 –0.75 –1.32 –1.18 –0.62 –0.97

0.27 0.70 0.89 0.50 0.69 0.61

2.19 2.12 1.61 0.66 0.98 1.51

1.02 1.87 0.98 1.72 1.37 1.39

– 0.06 0.10 – 0.01 0.06

– 0.18 0.18 – 0.10 0.15

– 0.11 0.38 – 0.02 0.17

U.K. U.K. U.K. U.S. U.S. –

1963–88 1963–87 1963–92 1974–94 1970–90 –

–0.24 –0.21 –0.32 –0.20 –0.27 –0.25

–1.10 –1.23 –1.09 –0.69 –0.59 –0.94

–0.76 –0.83 –1.28 –0.11 –0.33 –0.66

1.07 0.64 0.76 0.77 –0.27 0.59

1.36 2.11 1.55 1.84 0.76 1.52

3.36 2.06 0.88 1.07 0.41 1.56

–0.31 0.01 – 0.01 0.03 –0.07

–0.01 0.05 – 0.03 0.19 0.07

0.03 0.03 – –0.02 0.09 0.03

Annual Data Johnson (1992) Duffy (1990) Blake and Nied (1997) Tengene (1990) Nelson and Moran (1995) Average Quarterly Data Duffy (1995) Duffy (1990) Salisu et al. (1997) Nelson (1999) This Study Average

The Long-run Demand for Alcoholic Beverages and the Advertising Debate

Table 6. Comparison of Elasticity Estimates.

43

44

N. EDWARD COULSON, JOHN R. MORAN AND JON P. NELSON

followed by spirits and beer. With respect to income, wine is the most elastic, followed by spirits and beer.14 Our long-run advertising elasticities are all positive, but very small in magnitude.15 This is the same result reported in almost all previous studies. This ﬁnding suggests that alcohol advertising primarily affects brand shares, and has very little impact on beverage demands or total alcohol consumption. The policy implication is that narrow restrictions on alcohol advertising, such as billboard bans, will have little or no impact on alcohol consumption. Reﬂecting an information-search view of the role of advertising, such restrictions might instead tend to increase market concentration or prices. Short-Run Dynamics – Summary Results This section summarizes estimates from a series of ﬁve VEC speciﬁcations for each beverage (displayed in Tables B.1, B.2, and B.3 in Appendix B), incorporating from 0 to 4 lagged differences of each explanatory variable.16 Although the SIC favors the speciﬁcation containing only an error correction term, more general lag structures were estimated to allow for possible carryover effects from price and advertising changes in earlier periods. Although we ﬁnd little in the way of short-run dynamics, two features of the models are worth noting. First, the vast majority of the estimated error-correction terms are statistically signiﬁcant, thus providing additional evidence in favor of the cointegration hypothesis. And second, the demand for alcoholic beverages is signiﬁcantly inﬂuenced by the age composition of the population. The relationship between the youth variable and alcohol consumption is positive in all of the estimated error correction models, except the zero-lag model for wine. The magnitude of the youth coefﬁcient is largest for beer, followed by spirits and then wine. Similarly, the variable for older consumers is negative in most cases, larger than the youth coefﬁcient, and the magnitude is especially important for spirits. These ﬁndings are consistent with Nelson (1997, 1999), and suggest that a continued decline in alcohol consumption in the U.S. will occur as the population ages. Their ﬁndings also provide strong evidence in favor of a demand model that allows for demographic effects (Blake & Nied, 1997; Heien & Pompelli, 1989). The lag structure for advertising is of particular interest, since it is often argued that advertising has a lingering effect on product sales due to cumulative carry-over effects associated with brand loyalty (Berndt, 1991, p. 385). A comprehensive literature review by Assmus et al. (1984) reported a short-term advertising elasticity of about 0.22 and a carry-over elasticity of 0.47 for many “mature” products. Short-term elasticities obtained in models without lagged

The Long-run Demand for Alcoholic Beverages and the Advertising Debate

45

terms are generally higher, and the data interval also inﬂuences these results (Assmus et al., 1984, p. 71). In our case, short-run dynamics are relatively unimportant for any of the explanatory variables, including advertising.17 This result may reﬂect the inclusion of the error-correction terms, which indicate that beverage demands in the U.S. respond quickly to short-term shocks. In the zero-lag speciﬁcation, the spirits, beer, and wine coefﬁcients on the ﬁrst EC term are –1.24, –0.87, and –1.04, respectively. Thus, it appears that consumers quickly adjust their behavior in response to quarter-to-quarter changes in prices and advertising. Perhaps shortening the data interval to monthly data (Lariviere et al., 2000) is required to pick up more of the short-run dynamics and this would aid in the identiﬁcation of advertising carryover effects. A second explanation lies in the nature of advertising for familiar products. Almost all advertising is carried out at the brand level. The absence of short-run dynamics, coupled with the small long-run advertising elasticities, lends support to the view that the primary effect of advertising is to redistribute brand shares, with little in the way of inter-beverage effects.

IV. CONCLUSIONS This paper has estimated the long-run demand for three alcoholic beverages (beer, wine, and spirits) using quarterly data for the period 1970 : 1 to 1990 : 4. Preliminary testing rejected the presence of unit roots at the annual and semiannual frequencies, but did not allow rejection of the hypothesis of unit roots at the zero frequency. Tests conducted using the Johansen maximum eigenvalue test support the hypothesis of a long-run relationship between beverage-level consumption, prices, advertising, and real income for each of the three beverages. In itself, this is reassuring since the inﬂuence of advertising is sometimes cast in social, rather than economic, terms. Elasticities obtained from the estimated cointegrating vectors indicate that long-run beverage demands are both price and income inelastic, with estimated income elasticities that are smaller than those reported in most previous studies. This result reﬂects a number of factors, including the use of quarterly data, a more complete set of explanatory variables, and a focus on a more recent period of time, during which per capita alcohol consumption has grown very little. The estimates for long-run advertising elasticities are extremely small in magnitude, a result which is shared by virtually all previous studies of alcoholic beverages. Much of the existing research on alcohol advertising has been viewed as deﬁcient due to its use of annual data; omitted variables; mis-measurement of advertising; simultaneous equations bias; and inadequate attention to advertising dynamics. Critics argue that these problems are responsible for the inelastic

46

N. EDWARD COULSON, JOHN R. MORAN AND JON P. NELSON

response of alcohol demand to advertising. To this list, we would add that few prior advertising studies have adequately addressed possible nonstationarity of the data. After correcting for almost all of these problems, advertising still plays virtually no role in determining alcohol consumption. Thus, short of total bans, restrictions on alcohol advertising are unlikely to have more than a negligible impact on the use of alcoholic beverages. The implication of these ﬁndings is that alcohol advertising primarily affects brand shares. In contrast, beverage prices are found to have a much larger effect on alcoholic beverage demands than does advertising. These ﬁndings suggest that policies directed at mitigating the external costs of alcohol consumption should concentrate on non-advertising restrictions and controls.

NOTES 1. For contrasting views on the advertising policy debate see Center for Science in the Public Interest (1995) and Waterson (1989). 2. Examples include Duffy (1990), Selvanathan (1989), Tengene (1990), Johnson et al. (1992), Lee and Tremblay (1992), Goel and Morey (1995), and Nelson and Moran (1995). 3. Ideally, advertising should be measured in terms of the total number of message impressions received by consumers. Such a measure should include all relevant media in which the product is advertised, and allow for consistent aggregation by adjusting for the relative effectiveness of each media type. If one is willing to assume that effectiveness is proportional to price, nominal advertising expenditures can be converted to message impressions through the use of media-speciﬁc price deﬂators. Unfortunately, many existing studies have either used undeﬂated expenditures or have failed to deﬂate by media-speciﬁc price indices, while others simply omit consideration of relevant media altogether. Measurement error is therefore a concern when evaluating the existing evidence on alcohol consumption and advertising. 4. As pointed out by Engle and Granger (1987) and Stock and Watson (1988), the OLS estimator in a cointegrating regression is missing the usual “simultaneous equations bias” that occurs in models with jointly endogenous variables (although inference using the usual t-statistic remains invalid). The reason is easy to see. The bias term referred to is, roughly speaking, the ratio of the cross-product of the error term and the endogenous right-hand side variable, and the sum of squares of that variable. When the variables are cointegrated, the numerator is the product of an I(1) and an I(0) variable, while the denominator is the product of two I(1) variables, hence this bias term will vanish asymptotically. 5. The likely endogeneity of advertising was ﬁrst noted by Schmalensee (1972), who demonstrated that under very general conditions proﬁt-maximizing ﬁrms will set advertising expenditures as a constant fraction of sales. Schmalensee also introduced the use of message impressions as the relevant measure of real advertising outlays. 6. We use quarterly price indices rather than an implicit per-unit price, and beverage consumption is measured in ﬂuid gallons. Nelson (1999) shows that demand estimates obtained using ﬂuid gallons and ethanol gallons are very similar. The beginning date of

The Long-run Demand for Alcoholic Beverages and the Advertising Debate

47

1970 was predicated by the availability of quarterly advertising data. Several past studies also suggest a structural shift in alcohol demand that begins around 1969. 7. Beginning with the most general model, the trend terms (or the seasonal dummies) were eliminated if either was insigniﬁcant at the 0.10 level. We then re-ran the model including only the previously signiﬁcant regressor; if it remained signiﬁcant, that speciﬁcation was chosen as being the appropriate form of the test. If it became insigniﬁcant, the model was estimated with only an intercept, and the process was repeated based on the signiﬁcance of the intercept term. In cases where neither the trend term nor any of the seasonal dummies were signiﬁcant in the initial speciﬁcation, the model was re-estimated omitting each in turn to see if the absence of one induced signiﬁcance in the other. Note that all three seasonal dummies were included if any one of them was signiﬁcant at the 0.10 level. 8. In the case of spirits advertising, a t-statistic of –2.27 was obtained for the semiannual root, compared to 0.10 critical values of –2.73 and –2.63 for 48 and 100 observations, respectively. For beer advertising, the F-statistic for testing for an annual unit root was 4.47 relative to 0.10 critical values of 5.50 and 5.56 for 48 and 100 observations, respectively. 9. We seasonally ﬁltered only those variables (consumption and advertising) that displayed a clear seasonal pattern. All other variables were left in their original form (see Hylleberg, 1986). As a sensitivity check, each cointegrating vector was also estimated with unﬁltered data, using quarterly dummies to control for seasonal inﬂuences. The resulting long-run demand elasticities were virtually identical to those obtained using the ﬁltered data. 10. Although, strictly speaking, the ADF test indicates that per capita real income is I(0), this could simply be an artifact of having chosen an insufﬁcient number of lags for the test. When an additional lag is added, the I(1) null hypothesis is no longer rejected. In what follows, we defer to the consensus view that per capita real income should be treated as an I(1) variable. 11. Because they are integrated of a different order, the two population series cannot be cointegrated with the other variables. Instead, these series enter as exogenous regressors in the error-correction models. 12. When multiple cointegrating relationships exist, we follow the convention of selecting the vector with the largest eigenvalue. Unfortunately, in such cases, there is no readily-available method for computing correct standard errors for the chosen vector. 13. Most earlier studies use data sets that start in the mid-1950s or early 1960s. These studies reﬂect a time period characterized by rising per capita alcohol consumption and a trend toward increased spirits and beer consumption. However, by the early 1980s, these trends had largely run their course (see Nelson, 1997, 1999 for discussion). The data employed in the present paper reﬂect a more recent time period during which alcohol consumption has grown very little and spirits consumption has declined sharply over time. 14. The anomalous ﬁnding of a negative income elasticity for beer could be due to the absence of an adding-up constraint on the income terms, which points to the advantages of system-wide estimation. However, several other single-equation studies also report zero or negative income elasticities for beer (see Ornstein & Hanssens, 1985). 15. In the two cases where standard errors can be calculated, the advertising elasticities are estimated quite precisely. For example, the 95% conﬁdence intervals for

48

N. EDWARD COULSON, JOHN R. MORAN AND JON P. NELSON

the own-advertising elasticities for spirits and wine are (–0.002, 0.092) and (0.156, 0.222), respectively. Thus, in these cases, it is possible to reject any sizeable impact of advertising on consumption. Similarly tight conﬁdence intervals can also be constructed for the cross-advertising elasticities. 16. Note that the number of error correction terms included in each model is equal to the number of cointegrating vectors indicated by the maximum eigenvalue test. 17. To further explore the dynamics of the advertising-sales relationship, we calculated impulse response functions (IRFs), which trace out the dynamic response of the endogenous variables to exogenous shocks to the system. The IRFs (which are available upon request from the authors) indicate that convergence to the values implied by the long-run elasticities takes place within just two or three quarters. This quick adjustment is a product of the small number of lags in the system. Note that this also illustrates an advantage of quarterly over annual data, in that the latter cannot model these short-lived dynamic responses. This result renders inappropriate several earlier studies that attempt to model the dynamics of alcohol advertising using annual data.

ACKNOWLEDGMENTS A portion of this research was conducted while the second author was a Fellow in the Robert Wood Johnson Foundation’s Scholars in Health Policy Research Program. We thank Michele Gambera, Hiroyuki Kawakatsu, Ataman Ozyildirim, Randy Sherrod, Norm Swanson, Mark Watson, Chris Wilkins, Jeffrey Wooldridge, and an anonymous reviewer for helpful comments.

REFERENCES Assmus, G., Farley, J. U., & Lehman, D. R. (1984). How advertising affects sales: Meta-analysis of econometric results. Journal of Marketing Research, 21, 65–74. Bass, F. M., & Leone, R. P. (1983). Temporal aggregation, the data interval bias, and empirical estimation of bimonthly relations from annual data. Management Science, 29, 1–11. Berndt, E. (1991). The Practice of Econometrics: Classic and Contemporary. Reading, MA: Addison-Wesley. Blake, D., & Nied, A. (1997). The demand for alcohol in the United Kingdom. Applied Economics, 29, 1655–1672. Center for Science in the Public Interest (1995). Double Dip: The Simultaneous Decline of Alcohol Advertising and Alcohol Problems in the United States. CSPI, Washington, D.C. Cook, P., & Moore, M. J. (2000). Alcohol. In: A. Culyer & J. P. Newhouse (Eds), Handbook of Health Economics (pp. 1629–1673). North-Holland, Amsterdam. Duffy, M. H. (1990). Advertising and alcoholic drink demand in the U.K.: Some further Rotterdam model estimates. International Journal of Advertising, 9, 247–257. Duffy, M. H. (1995). Advertising in demand systems for alcoholic drinks and tobacco: A comparative study. Journal of Policy Modeling, 17, 557–577. Engle, R., & Granger, C. W. J. (1987). Cointegration and error-correction: Representation, estimation, and testing. Econometrica, 55, 251–276. Gius, M. P. (1996). Using panel data to determine the effect of advertising on brand-level distilled spirits sales. Journal of Studies of Alcohol, 57, 73–76.

The Long-run Demand for Alcoholic Beverages and the Advertising Debate

49

Goel, R. K., & Morey, M. J. (1995). The interdependence of cigarette and liquor demand. Southern Economic Journal, 62, 451–459. Heien, D., & Pompelli, G. (1989). The demand for alcoholic beverages: Economic and demographic effects. Southern Economic Journal, 56, 759–770. Hilton, M. E., & Bloss, G. (1993). Economics and the Prevention of Alcohol-Related Problems. NIAAA Research Monograph No. 25. National Institutes of Health, Rockville, MD. Hylleberg, S. (1986). Seasonality in Regression. San Diego: Academic Press. Hylleberg, S., Engle, R. F., Granger, C. W. J., & Yoo, B. S. (1990). Seasonal integration and cointegration. Journal of Econometrics, 44, 215–238. Johansen, S. (1988). Statistical analysis of cointegration vectors. Journal of Economic Dynamics and Control, 12, 231–254. Johansen, S. (1991). Estimation and hypothesis testing of cointegration vectors in Gaussian vector autoregressive models. Econometrica, 59, 1551–1580. Johansen, S., & Juselius, K. (1990). Maximum likelihood estimation and inference on cointegration – With applications to the demand for money. Oxford Bulletin of Economics and Statistics, 52, 169–210. Johnson, J., Oksanen, E. H., Veall, M. R., & Fretz, D. (1992). Short-run and long-run elasticities for Canadian consumption of alcoholic beverages: An error-correction mechanism/ cointegration approach. Review of Economics and Statistics, 74, 64–74. Leone, R. P. (1995). Generalizing what is known about temporal aggregation and advertising carryover. Marketing Science, 14, 141–150. Lariviere, E., Larve, B., & Chalfant, J. (2000). Modeling the demand for alcoholic beverages and advertising speciﬁcations. Agricultural Economics, 22, 147–162. Lee, B., & Tremblay, V. J. (1992). Advertising and U.S. market demand for beer. Applied Economics, 24, 69–76. Leung, S., & Phelps, C. (1993). My kingdom for a drink . . .? A review of estimates of the price sensitivity of demand for alcoholic beverages. In: M. E. Hilton & G. Bloss (Eds), Economics and the Prevention of Alcohol-Related Problems (pp. 1–31). NIAAA Research Monograph No. 25, National Institutes of Health, Rockville, MD. Nelson, J. P. (1997). Economic and demographic factors in U.S. alcohol demand: A growthaccounting analysis. Empirical Economics, 22, 83–102. Nelson, J. P. (1999). Broadcast advertising and U. S. demand for alcoholic beverages. Southern Economic Journal, 65, 774–790. Nelson, J. P., & Moran, J. R. (1995). Advertising and U.S. alcoholic beverage demand: Systemwide estimates. Applied Economics, 27, 1225–1236. Nelson, J. P., & Young, D. J. (2001). Do advertising bans work? An international comparison. International Journal of Advertising, forthcoming. Ornstein, S. O., & Hanssens, D. M. (1985). Alcohol control laws and consumption of distilled spirits and beer. Journal of Consumer Research, 12, 200–213. Osterwald-Lenum, M. (1992). A note with quantiles of the asymptotic distribution of the maximum likelihood cointegration rank test statistics. Oxford Bulletin of Economics and Statistics, 54, 461–471. Saffer, H. (1991). Alcohol advertising bans and alcohol abuse: An international perspective. Journal of Health Economics, 10, 65–79. Saffer, H. (1993). Advertising under the inﬂuence. In: M. E. Hilton & G. Bloss (Eds), Economics and the Prevention of Alcohol-Related Problems (pp. 125–140). NIAAA Research Monograph No. 25, National Institutes of Health, Rockville, MD.

50

N. EDWARD COULSON, JOHN R. MORAN AND JON P. NELSON

Saffer, H. (1997). Alcohol advertising and motor vehicle fatalities. Review of Economics and Statistics, 79, 431–442. Salisu, M., & Balasubramanyam, V. (1997). Income and price elasticities of demand for alcoholic drinks. Applied Economics Letters, 4, 247–251. Schmalensee, R. (1972). The Economics of Advertising. North-Holland, Amsterdam. Selvanathan, E. A. (1989). Advertising and alcohol demand in the U.K.: Further results. International Journal of Advertising, 8, 181–188. Seo, B. (1998). Statistical inference on cointegration rank in error correction models with stationary covariates. Journal of Econometrics, 85, 339–385. Stock, J., & Watson, M. (1988). Variable trends in economic time series. Journal of Economic Perspectives, 2, 147–174. Tengene, A. (1990). The Kalman ﬁlter approach for testing for structural change in the demand for alcoholic beverages in the U.S. Applied Economics, 22, 1407–1416. Theil, H. (1975). Theory and Measurement of Consumer Demand. Vol. 1. North-Holland, Amsterdam. Theil, H. (1976). Theory and Measurement of Consumer Demand. Vol. 2. North-Holland, Amsterdam. Waterson, M. J. (1989). Advertising and alcohol: An analysis of the evidence relating to two major aspects of the debate. International Journal of Advertising, 8, 111–131. Young, D. J. (1993). Alcohol advertising bans and alcohol abuse: Comment. Journal of Health Economics, 12, 213–228.

The Long-run Demand for Alcoholic Beverages and the Advertising Debate

51

APPENDIX A Beverage Consumption: Beer – monthly data on shipments of malt beverages (in barrels) from the Brewer’s Almanac (Washington, D.C.: Beer Institute). Wine – monthly data on wine entering trade channels (in cases) from the Wine Marketing Handbook (NewYork: Jobson). Spirits – monthly data on sales of distilled spirits (in cases) from Liquor Handbook (New York: Jobson). Each monthly series was aggregated by quarter and converted to ﬂuid gallons following standard industry conventions. Beverage Prices: The quarterly CPI series from CPI Detailed Report (Washington, D.C.: Bureau of Labor Statistics) for February, May, August, and November of each year. Income and Population: Quarterly values for personal consumption expenditures from the National Income and Product Accounts, Survey of Current Business (Washington, D.C.: Department of Commerce). All per capita series are based on the resident population aged 16 and over. Total and age group population series from the latest Current Population Reports, Series P-25 (Washington, D.C.: Bureau of the Census), with quarterly data linearly interpolated. Advertising and Advertising Prices: Nominal quarterly expenditures on national advertising of beer, wine, and spirits from LNA Class/Brand QTR $ (New York: Leading National Advertisers). Quarterly data since 1970 are available on six media: magazines, newspaper supplements, outdoor, network TV, spot TV, and network radio. In order to ensure consistent aggregation, each nominal series was ﬁrst deﬂated by its McCann-Erickson media price index (CPM version) from various issues of Advertising Age and Media Decisions. The real series by media (message impressions) were added together and divided by population to obtain per capita series for real advertising by beverage.

52

N. EDWARD COULSON, JOHN R. MORAN AND JON P. NELSON

APPENDIX B Table B.1. Error Correction Models for Spirits. Models (# of Lags) Explanatory Variables

0

1

2

3

4

EC Term # 1 EC Term # 2 EC Term # 3 EC Term # 4

–1.24*** 0.08** 0.04*** –

–1.32*** – – –

–1.65*** – – –

–1.21*** 0.03 0.03** –

–2.95*** 0.21*** 0.03 –0.09***

Intercept % Youngt % Oldt

2.60*** 0.02** –0.21***

1.61*** 0.03*** –0.16***

1.48*** 0.05*** –0.20***

–1.68 0.06** –0.00

2.85 0.08** –0.34***

–0.046 – – – –0.053 – – – –0.007 – – – 0.054*** – – – 0.245 – – – –0.433 – – – –0.500 – – – –0.075 – – –

0.136 –0.180 – – –0.096** –0.022 – – –0.020 –0.019 – – 0.069*** 0.043*** – – 0.340 0.043 – – –0.056 –1.289** – – –0.348 0.653 – – –0.097 0.131 – –

–0.244 –0.312 –0.022 – –0.057 –0.004 –0.020 – –0.070 –0.055 –0.080 – 0.079*** 0.037 0.019 – 0.703 0.200 –0.304 – –0.631 –1.340 0.333 – –0.572 0.366 0.132 – 0.160 0.191 –0.174 –

1.233 0.853 0.673 0.408 –0.211*** –0.151** –0.148* –0.071 –0.063 –0.050 –0.074 0.023 0.125*** 0.080** 0.056 0.024 0.233 –0.529 –0.933** –0.749 –0.162 –0.826 0.236 0.413 –1.203 –0.076 0.151 –0.758 0.301 0.212 0.021 0.664

⌬QS,t ⫺ 1 ⌬QS,t ⫺ 2 ⌬QS,t ⫺ 3 ⌬QS,t ⫺ 4 ΔAS,t ⫺ 1 ΔAS,t ⫺ 2 ΔAS,t ⫺ 3 ΔAS,t ⫺ 4 ΔAB,t ⫺ 1 ΔAB,t ⫺ 2 ΔAB,t ⫺ 3 ΔAB,t ⫺ 4 ΔAW,t ⫺ 1 ΔAW,t ⫺ 2 ΔAW,t ⫺ 3 ΔAW,t ⫺ 4 ΔRt ⫺ 1 ΔRt ⫺ 2 ΔRt ⫺ 3 ΔRt ⫺ 4 ΔPS,t ⫺ 1 ΔPS,t ⫺ 2 ΔPS,t ⫺ 3 ΔPS,t ⫺ 4 ΔPB,t ⫺ 1 ΔPB,t ⫺ 2 ΔPB,t ⫺ 3 ΔPB,t ⫺ 4 ΔPW,t ⫺ 1 ΔPW,t ⫺ 2 ΔPW,t ⫺ 3 ΔPW,t ⫺ 4 R-squared Adj. R-squared Log Likelihood SIC

– – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – 0.60 0.57 1375.56 –52.02

0.69 0.64 1384.83 –50.89

0.75 0.67 1443.06 –49.22

0.75 0.60 1546.84 –46.91

Notes: ** and *** indicate signiﬁcance at the 0.05 and 0.01 levels, respectively.

0.83 0.66 1611.23 –44.47

The Long-run Demand for Alcoholic Beverages and the Advertising Debate

53

Table B.2. Error Correction Models for Beer. Models (# of Lags) Explanatory Variables

0

1

2

3

EC Term # 1 EC Term # 2 EC Term # 3 EC Term # 4

–0.87*** 0.02 0.04*** –0.01**

–1.33*** –0.04** 0.05*** –

–0.06 – – –

Intercept % Youngt % Oldt

–2.87*** 0.03*** 0.14***

–2.55*** 0.04*** 0.10***

–0.02 0.01 –0.01

0.38*** 0.00 –0.03**

–0.36 0.01 0.01

0.335*** – – – 0.031 – – – –0.040** – – – –0.007 – – – –0.255 – – – 0.156 – – – 0.320 – – – –0.238 – – –

–0.534*** –0.562*** – – 0.029 0.012 – – –0.008 0.041** – – –0.013 –0.008 – – –0.165 0.122 – – 0.071 –0.005 – – 0.376 –0.188 – – –0.082 0.034 – –

–0.865*** –0.838*** –0.293** – –0.005 0.011 0.054 – 0.009 0.048 0.032 – –0.001 0.006 0.005 – –0.209 0.106 –0.145 – 0.015 –0.102 –0.128 – 0.266 –0.092 0.088 – 0.069 –0.015 0.126 –

–0.333 –0.446 –0.076 0.155 0.003 –0.021 –0.004 –0.038 –0.020 0.041 0.041 0.025 0.032 0.041** 0.038 0.028** –0.020 0.124 –0.225 –0.089 0.283 0.064 0.187 0.177 –0.002 –0.367 0.089 –0.206 –0.390 –0.420 –0.203 0.091

⌬QB,t ⫺ 1 ⌬QB,t ⫺ 2 ⌬QB,t ⫺ 3 ⌬QB,t ⫺ 4 ⌬AS,t ⫺ 1 ⌬AS,t ⫺ 2 ⌬AS,t ⫺ 3 ⌬AS,t ⫺ 4 ⌬AB,t ⫺ 1 ⌬AB,t ⫺ 2 ⌬AB,t ⫺ 3 ⌬AB,t ⫺ 4 ⌬AW,t ⫺ 1 ⌬AW,t ⫺ 2 ⌬AW,t ⫺ 3 ⌬AW,t ⫺ 4 ⌬Rt ⫺ 1 ⌬Rt ⫺ 2 ⌬Rt ⫺ 3 ⌬Rt ⫺ 4 ⌬PS,t ⫺ 1 ⌬PS,t ⫺ 2 ⌬PS,t ⫺ 3 ⌬PS,t ⫺ 4 ⌬PB,t ⫺ 1 ⌬PB,t ⫺ 2 ⌬PB,t ⫺ 3 ⌬PB,t ⫺ 4 ⌬PW,t ⫺ 1 ⌬PW,t ⫺ 2 ⌬PW,t ⫺ 3 ⌬PW,t ⫺ 4 R-squared Adj. R-squared Log Likelihood SIC

– – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – 0.51 0.47 1407.44 –51.93

0.61 0.53 1437.43 –50.45

0.63 0.52 1459.48 –49.63

0.06 – – –

4

0.73 0.59 1512.00 –47.80

Notes: ** and *** indicate signiﬁcance at the 0.05 and 0.01 levels, respectively.

–0.67 –0.07*** 0.03 –0.02

0.80 0.62 1615.24 –44.57

54

N. EDWARD COULSON, JOHN R. MORAN AND JON P. NELSON

Table B.3. Error Correction Models for Wine. Models (# of Lags) Explanatory Variables

0

1

2

3

4

EC Term # 1 EC Term # 2 EC Term # 3 EC Term # 4

–1.04*** – – –

–1.54*** –0.24*** 0.01 –

–1.34*** – – –

–0.41 0.07** 0.03 –

–0.89*** 0.21*** – –

Intercept % Youngt % Oldt

–2.40*** –0.01 0.18***

5.31*** 0.03 –0.40***

1.19** 0.01 –0.09***

3.26 0.02 –0.26***

0.60 0.02 –0.07

0.204 – – – 0.189 – – – –0.074 – – – 0.005 – – – 1.041 – – – 0.839 – – – 0.859 – – – –0.058 – – –

0.368 0.343** – – –0.130 –0.334*** – – –0.077 –0.104 – – –0.163*** –0.164*** – – 0.927 –0.271 – – –0.165 0.675 – – 0.262 1.515 – – 0.122 –0.774 – –

–0.569** –0.534*** –0.573*** – 0.062 0.053 0.098 – –0.133 –0.112 –0.124 – –0.035 –0.093** –0.058 – –0.322 –0.909 –0.650 – 0.166 –0.161 1.272 – 0.971 1.341 0.438 – 0.985 0.215 1.005 –

–0.278 –0.384 –0.467** –0.118 –0.084 –0.140 –0.119 –0.169 –0.145 –0.087 –0.173 0.084 –0.038 –0.100** –0.027 0.050 –0.181 –0.364 –0.364 0.053 –0.267 –0.546 0.994 –1.234 0.928 0.697 0.418 0.407 0.997 0.722 1.078 0.283

⌬QW,t ⫺ 1 ⌬QW,t ⫺ 2 ⌬QW,t ⫺ 3 ⌬QW,t ⫺ 4 ⌬AS,t ⫺ 1 ⌬AS,t ⫺ 2 ⌬AS,t ⫺ 3 ⌬AS,t ⫺ 4 ⌬AB,t ⫺ 1 ⌬AB,t ⫺ 2 ⌬AB,t ⫺ 3 ⌬AB,t ⫺ 4 ⌬AW,t ⫺ 1 ⌬AW,t ⫺ 2 ⌬AW,t ⫺ 3 ⌬AW,t ⫺ 4 ⌬Rt ⫺ 1 ⌬Rt ⫺ 2 ⌬Rt ⫺ 3 ⌬Rt ⫺ 4 ⌬PS,t ⫺ 1 ⌬PS,t ⫺ 2 ⌬PS,t ⫺ 3 ⌬PS,t ⫺ 4 ⌬PB,t ⫺ 1 ⌬PB,t ⫺ 2 ⌬PB,t ⫺ 3 ⌬PB,t ⫺ 4 ⌬PW,t ⫺ 1 ⌬PW,t ⫺ 2 ⌬PW,t ⫺ 3 ⌬PW,t ⫺ 4 R-squared Adj. R-squared Log Likelihood SIC

– – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – 0.40 0.38 1252.68 –50.76

0.61 0.53 1337.35 –48.01

0.73 0.65 1387.70 –47.63

0.90 0.85 1490.30 –45.50

Notes: ** and *** indicate signiﬁcance at the 0.05 and 0.01 levels, respectively.

0.92 0.86 1515.00 –43.80

MANDATED EXCLUSIVE TERRITORIES: EFFICIENCY EFFECTS AND REGULATORY SELECTION BIAS Tim R. Sass and David S. Saurman ABSTRACT In this paper we analyze the efﬁciency effects of state-mandated exclusive distribution territories in the U.S. beer industry. Using panel data for 48 states over a 10-year period we estimate both ﬁxed-effects and instrumental-variable models of the impact of mandated exclusive territories on beer consumption. We ﬁnd that standard OLS regressions of beer consumption suffer from selection bias, due to the endogeneity of state statutes. Correcting for this bias we estimate exclusive territory mandates increase consumption by between three and eleven%. Our results therefore indicate that exclusive territories in the beer industry increase social welfare and enhance the well-being of consumers.

I. INTRODUCTION Vertically imposed exclusive territories, whereby a manufacturer grants its dealers intrabrand monopoly rights within speciﬁed territories, have been the subject of theoretical debate for many years. Some have argued that exclusive territories promote economic efﬁciency by encouraging the provision of services or other dealer actions that enhance the value of the product to consumers (Telser, 1960; Posner, 1977; Klein & Murphy, 1988). Others have Advertising and Differentiated Products, Volume 10, pages 55–72. Copyright © 2001 by Elsevier Science Ltd. All rights of reproduction in any form reserved. ISBN: 0-7623-0823-0

55

56

TIM R. SASS AND DAVID S. SAURMAN

claimed that exclusive territories reduce social welfare by restricting competition among distributors or facilitating manufacturer price discrimination (Carstensen & Dahlson, 1986; Rey & Stiglitz, 1995). Only recently has systematic empirical evidence been presented to address the theoretical debate. Studies by Culbertson (1989) and Culbertson and Bradford (1991) compare average beer prices across states over the period 1985–1987. They show that states which mandate exclusive territories for beer distributors have higher retail beer prices than non-mandating states. Unfortunately, these price differences reveal little about the welfare effects of exclusive territories. Price could be higher due to decreased competition among manufacturers or dealers or higher prices could be the result of increased demand from dealer promotional efforts. Sass and Saurman (1993) estimate structural supply and demand equations for beer as well as reduced-form price and quantity equations. Using a panel of states over the period 1982–1987, they ﬁnd that state mandated exclusive territories both increase demand and reduce supply. The net effect is a higher equilibrium retail price but no signiﬁcant change in equilibrium quantity. The combination of higher prices and constant output suggests that consumer welfare is unaffected by mandates of exclusive territories while manufacturers and dealers gain. A problem with the aforementioned studies is that they are essentially crosssectional in nature. The observed differences across states could be due to exclusive territories or to other unmeasured factors that vary across states and are correlated with the adoption of laws mandating territories. Put differently, a selection bias may be present in these estimates because state mandates of exclusivity are not randomly assigned. Rather states “self-select” by choosing whether to require the assignment of exclusive territories. One way to deal with the selection-bias problem is to employ time-series analysis, focusing on changes in exclusive territories and market outcomes over time within a given jurisdiction. This is the approach taken by Sass and Saurman (1996), who ﬁnd that Indiana’s ban on exclusive territories in 1979 has produced a 6% decrease in per-capita beer consumption. Unfortunately, Indiana is the only state to ban exclusive territories. Therefore the analysis is limited to a single state and the robustness of the results is uncertain. However, the ﬁnding that exclusive territories increase consumption at least suggests that previous cross-sectional estimates of the impact of exclusive territories may have been biased. We confront the issue of selection bias in the present paper in two ways. First, we extend our interstate data set to cover a longer time period (10 years rather than 6) and estimate a ﬁxed effects model that accounts for the impact

Mandated Exclusive Territories: Efﬁciency Effects and Regulatory Selection Bias

57

of any unmeasured time-invariant variables. Second, we estimate the determinants of state mandates using a probit model. The ﬁtted values of the mandate variable are then used as instruments to provide unbiased estimates of the impact mandated exclusive territories on beer consumption. Our analysis indicates that standard OLS regressions of beer consumption suffer from selection bias. Correcting for this bias with ﬁxed effects or instrumental variables estimators yields results that are consistent with the time-series estimates in Sass and Saurman (1996); exclusive territory mandates increase consumption by between three and 11%. Our results therefore indicate that exclusive territories in the beer industry not only increase social welfare, but enhance the well-being of consumers as well.

II. INSTITUTIONAL AND LEGAL ENVIRONMENT OF BEER DISTRIBUTION Before beginning our empirical analysis it is important to explain the legal and institutional environment in which our tests are conducted. Over 90% of all beer consumed in the United States is produced by the top four domestic brewers: Anheuser-Busch (A-B), Miller Brewing (Miller), Adolph Coors (Coors), and Pabst Brewing Co. (Pabst). Most states prohibit brewers from distributing their beer to wholesalers. Thus except for a handful of companyowned distributors, brewers sell their beer to independent wholesalers who in turn sell the beer to retailers. The use of exclusive territories in the beer industry dates back at least to the repeal of prohibition. As early as the 1940s, both Anheuser-Busch and Miller had established distributor territories and strongly opposed extra-territorial sales by distributors.1 By the time of the 1967 Schwinn ruling,2 the leading brewers had all established exclusive distributor territories.3 After exclusive territories were declared illegal per se in Schwinn, the major brewers rewrote their wholesaler contracts to cover “areas of primary responsibility” rather than explicitly designating exclusive territories.4 After the Schwinn decision was overruled by the Court’s decision in Sylvania,5 brewers eventually shifted back to contractually explicit exclusive territories. In December 1982 A-B adopted new contracts specifying exclusive territories for each of its distributors.6 Other major brewers followed suit in late 1982 and early 1983, establishing exclusive territories in all states except Indiana,7 where exclusive territories were banned in 1979. Although exclusive territories were no longer per se illegal after Sylvania, they were still subject to a rule-of-reason analysis by the courts and could be found illegal if they stiﬂed competition. The possibility that exclusive contracts

58

TIM R. SASS AND DAVID S. SAURMAN

with dealers could be declared illegal and thus void clearly raised the cost to brewers of enforcing territorial exclusivity. Distributors selling outside their territories could potentially avoid termination or other private enforcement mechanisms by seeking relief in the courts. To counter this possibility and thereby strengthen the exclusivity arrangements, brewers lobbied congress for antitrust immunity in 1982, 1983, 1985 and 1987.8 All of these attempts failed however. Brewers and wholesalers were more successful in obtaining antitrust protection from state legislatures. Under the state-action immunity doctrine, actions that are required by state laws are exempt from federal antitrust prosecution. Thus by inducing states to mandate the use of exclusive distribution territories, brewers could shield themselves from potential antitrust actions. In the 1970s ten states adopted laws requiring brewers to establish exclusive distributor territories. From 1980 through 1985, thirteen more states passed laws mandating exclusive territories. The last state to legally require exclusive territories for beer distributors, Florida, did so in 1988.9

III. EMPIRICAL ANALYSIS Panel data consisting of annual observations taken from forty-eight states over the 1983–1992 period are employed to investigate empirically the effects of the state mandates on quantity.10 We proceed along the line of our previous work (Sass & Saurman, 1993) by initially offering the following reduced-form beer quantity speciﬁcation: BEERi,t = f(MANDATEi,t, TAXi,t, NOPRICEADSi,t, ANTIPDLAWi,t, CASHLAWi,t, DEPOSITi,t, MINAGEi,t, BANi,t, POPULATIONi,t, INCOMEi,t, (1) DENSITYi,t, TOURISMi,t, MILEAGEi,t, ␧i,t), where ␧t,i is a random error term. Deﬁnitions and sample means of the variables comprising this speciﬁcation are provided in Table 1. The expected signs of several variables in Eq. (1) need no elaboration (TAXi,t, CASHLAWi,t, DEPOSITi,t, MINAGEi,t, MILEAGEi,t, and NOPRICEADSi,t all negative; POPULATIONi,t, INCOMEi,t and TOURISMi,t, all positive). We make no prediction over the sign of DENSITYi,t. If greater population densities are indicative of lower local transportation and distribution costs, we expect greater consumption all else equal. On the other hand, consumers in relatively dense, urban locales may face a wider range of beverage substitutes as well as leisure-time activities. We also make no a priori statement over the sign of ANTIPDLAWi,t. If brewers would ordinarily ﬁnd it proﬁtable to price discriminate within a state, then ANTIPDLAWi,t would expand output given positive discrimination

Mandated Exclusive Territories: Efﬁciency Effects and Regulatory Selection Bias

Table 1. Variable Names, Sample Means, Descriptions, and Sources. Variable Name

BEER MANDATE

TAX

NOPRICEADS

ANTIPDLAW

CASHLAW

DEPOSIT

MINAGE

BAN POPULATION INCOME

Sample Mean

Description and Source

74897.815 Shipments/apparent consumption of beer, including imports, in thousands of gallons. Source: Brewers almanac. 0.421 Dummy variable = 1 if state requires brewers to establish exclusive distribution territories for at least one-half a year, zero otherwise. Sources: Modern Brewery Age Bluebook, CCH Liquor Control Law Reporter, and state annotated statutes. 0.495 Real state plus federal excise tax on beer, in 1981 dollars per gallon. Sources: Brewers Almanac and Geographical Cost of Living Differences: Interstate and Intrastate, Update 1991. 0.367 Dummy variable = 1 if exterior sign/billboard price advertising is prohibited for at least one-half a year, zero otherwise. Sources: Modern Brewery Age Bluebook, CCH Liquor Control Law Reporter, and state annotated statutes. 0.360 Dummy variable = 1 if anti-price discrimination statute or administrative rule pertaining to brewers in effect for at least one-half a year, zero otherwise. Sources: State annotated statutes and administrative rules. 0.104 Dummy variable = 1 if brewers are not permitted to extend credit to wholesalers for at least one-half a year, zero otherwise. Sources: Modern Brewery Age Bluebook, CCH Liquor Control Law Reporter, and state annotated statutes. 0.196 Dummy variable = 1 if state had a forced deposit law (a law that requires all containers to be returnable and carry a deposit) in effect for at least one-half a year, zero otherwise. Source: Beverage World’s Databank. 0.083 Own-state minimum legal drinking for 3.2% beer age minus border states’ minimum legal drinking age, weighted by the state population and the fraction of the population living within twenty miles of the relevant border. MINAGE is constrained to be zero or greater and reﬂects an incentive for residents of the own-state to purchase alcohol in the bordering state. Sources: Melanie Williams and Frank Chaloupka, 1994 Statistical Abstract of the United States. 0.021 Dummy variable = 1 for Indiana, 0 otherwise. Source: 905 IAC 1–28–1. 2340.219 Total resident adult (age 18 and above) population, in thousands. Source: Resident Population of States. 21311.711 Per-adult disposable income, in thousands of 1981 dollars. Sources: State Annual Summary Tables, 1929–1992 and Geographical Cost of Living Differences: Interstate and Intrastate, Update 1991.

59

60

TIM R. SASS AND DAVID S. SAURMAN

Table 1. Continued. Variable Name

DENSITY

TOURISM

MILEAGE

CVINCOME

CVPOPULATION JUSTCAUSE

STATETAX

EMPLOYMENT

RETAILERS

BLGRATIO

Sample Mean

Description and Source

0.119 Adult population density, thousands per square land mile. Sources: Resident Population of States and Statistical Abstract of the United States. 4.293 Ratio of hotel, motel, motor hotel, and tourist court (SIC 701) receipts to retail sales, multiplied by 100. Data for 1981, 1983–1986, and 1988–1991 estimated by linear interpolation from 1977, 1982, 1987, and 1992 actual data. Sources: Census of Service Industries and Census of Retail Trade. 293.624 Average of distances along the “quickest routes” from most populous city in a state (in 1980) to the nearest AnheuserBusch and Miller breweries in operation for a majority of the year. Sources: Automap, V 2, Modern Brewery Age Bluebook, Beverage Marketing Directory, various newspaper articles, corporate reports, discussions with company ofﬁcials. 0.174 Coefﬁcient of variation of average county per capita personal incomes within a state. Actual data for 1980–1990. Figures for 1991–1992 equal 1990 data. Source: USA Counties 1994 CD-ROM. 1.726 Coefﬁcient of variation of county population within a state. Source: USA Counties 1994 CD-ROM 0.683 Dummy variable = 1 if brewers are not permitted to terminate wholesalers without “just,” “good,” or “reasonable”cause, or cannot terminate wholesalers “unfairly, without due regard to the equities” for at least one-half a year, zero otherwise. Sources: Modern Brewery Age Bluebook, CCH Liquor Control Law Reporter, and state annotated statutes. 0.205 Real state excise tax on beer, in 1981 dollars per gallon. Source: Brewers Almanac and Geographical Cost of Living Differences: Interstate and Intrastate, Update 1991. 0.555 Annual average state employment in wholesale beer and ale distribution (SIC 5181) per thousand adult population. Missing observations for Delaware, Maine, Utah and Wyoming were imputed by linear regression. Source: Employment and Wages: Annual Averages and unpublished Bureau of Labor Statistics data. 4753.500 Number of bars, liquor stores, and groceries in the state. Data for 1981, 1983–1986, and 1988–1991 estimated by linear interpolation from 1977, 1982, 1987, and 1992 actual data. Source: Census of Retail Trade. 0.790 Ratio of bars plus liquor stores to grocery stores in the state. Data for 1981, 1983–1986, and 1988–1991 estimated by linear interpolation from 1977, 1982, 1987, and 1992 actual data. Source: Census of Retail Trade.

Mandated Exclusive Territories: Efﬁciency Effects and Regulatory Selection Bias

61

Table 1. Continued. Variable Name

Sample Mean

DRYPCT

4.565 Percent of state’s population living in counties dry for beer. Source: Brewers Almanac and Resident Population of States [table downloaded from Bureau of the Census FTP site] 28.553 Average of Senate and House members’ constituents age 18 to 44, in thousands. Source: Book of the States and Resident Population of States. 10.641 Percent of state’s population who are members of either the Southern Baptist Convention or the Church of Jesus Christ of Latter Day Saints, 1980. Data for 1981–1989 are based on linear interpolation of 1980 and 1990 data. Data for 1991 and 1992 based on extrapolation of 1980 and 1990 data. Source: Churches and Church Membership in the United States 1980, and 1990.

LEGCON1844

PCTBAPMORM

Description and Source

costs. But with brewers already employing exclusive dealer contracts, such regulations may retard brewers and dealers from adopting the optimal serviceproviding input combinations. If the types and amounts of optimal services varies within a state, producing these optimal levels will yield cost-based price differentials that may lead to prosecution under the anti-price discrimination statute. As such, a ceteris paribus reduction in output would result. Our prime focus is on the output effect of the vertical restraints represented by MANDATEi,t. The anticompetitive perspective holds exclusive territories – and presumably mandates of such – serve to either restrict intrabrand competition or facilitate price discrimination, or both, with an output restriction the likely market outcome. The efﬁciency hypothesis predicts that exclusive territories, by creating a stream of quasi-rents accruing to dealers, serve to induce wholesalers to provide brewers’ optimal level of services. These services may include point-of-sale promotions, advertising, stock rotation to maintain beer freshness, product presentation on store shelves and timely delivery to retailers. Even when granted territorial exclusivity by brewers, wholesalers may still provide sub-optimal service levels by failing to acquire optimal inputs when the future legal status of the contracts is probabilistic, as it is under the Supreme Court’s rule of reason legal posture since Sylvania. State action immunity solidiﬁes the permanency of all exclusivity contracts by securing their future legal status, raising the present value of distributor ﬁrms and inducing wholesalers to provide optimal service levels.11 As a result of the

62

TIM R. SASS AND DAVID S. SAURMAN

federal antitrust shield inherent in the state mandate, any effect on demand, and thus output, of exclusive contracts will be magniﬁed by the state mandate. The anticompetitive and efﬁciency hypotheses are then empirically distinguishable. The anticompetitive camp predicts an unambiguous decline in output associated with MANDATEi,t as only a retail market supply restriction occurs. Support of the efﬁciency hypothesis will be had if an output effect consistent with a supply restriction and demand increase is observed with the state mandates. Additionally, the likelihood of an overall welfare gain as well as a gain in consumer welfare arising from the state mandates of these vertical restraints is greater should output expand. BANi,t captures Indiana’s sole proscription of exclusive territories, and the arguments made above concerning state mandates carry over to the possible output effects of BANi,t. We caution, though, that BANi,t is simply an Indiana dummy variable and therefore has the potential to reﬂect any other Indianaspeciﬁc effects not captured by the remaining control variables. Under the standard assumptions, particularly ␧i,t uncorrelated with the explanatory variables, Ordinary Least Squares (OLS) will yield consistent parameter estimates. These estimates, presented in Table 2, indicate that most of the variables included in Eq. (1) appear to inﬂuence BEER in a fashion consistent with standard economic theory. Our interest here centers on the effect of MANDATE on the quantity of beer exchanged. The OLS estimates indicate that state mandates of exclusive territories have no statistically signiﬁcant effect on output, and offer some support of the efﬁciency perspective of such laws. Given the existing evidence over the effects of these mandates on price,12 this result is suggestive of a demand expansion consistent with consumers valuing additional services provided by wholesalers. Whether this vertical restraint causing a supply restriction, demand expansion, price increase, and zero quantity effect improves or retards overall economic welfare remains to be seen.13 These OLS estimates are potentially inconsistent if the regulatory structure is a choice variable. Since Peltzman (1976), economists have recognized the likelihood of an endogenous regulatory structure.14 That is, the values of a set of latent variables induce states to select themselves into one form of regulatory structure or another. With these latent or unobserved variables determining both BEER and any variable capturing the regulatory structure, then ␧i,t and the regulatory explanatory variable(s) in Eq. (1) will be correlated rendering the OLS estimates inconsistent. One method of accounting for this possibility and obtaining consistent estimates is estimating Eq. (1) by employing a two-way (states and years) ﬁxed effects estimator. The effects of some latent variables on consumption

Mandated Exclusive Territories: Efﬁciency Effects and Regulatory Selection Bias

63

Table 2. Reduced-Form Quantity Estimates, 1983–1992 Dependent Variable = lnBEERi,t (000s of Gallons) (Corrected for Heteroskedasticity – 480 Observations). Estimation Method Explanatory Variables MANDATEi,t ln TAXi,t NOPRICEADSi,t ANTIPDLAWi,t CASHLAWi,t DEPOSITi,t MINAGEi,t BANi,t ln POPULATIONI,t ln INCOMEi,t DENSITYi,t TOURISMi,t ln MILEAGEi,t Adjusted R2 F-Statistic Hausman ␹2 (23 df)

OLS 0.012 (0.706) –0.078*** (3.548) –0.03* (1.644) –0.029** (2.058) 0.009 (0.386) –0.061*** (4.57) –0.038* (1.685) –0.042 (1.496) 0.947*** (140.272) 0.558*** (4.702) –0.266*** (5.73) 0.004*** (9.624) –0.071*** (5.324) 0.981 1133.584***

Fixed Effects 0.033*** (2.916) –0.122*** (5.441)

0.010 (0.732)

0.020 (1.164) –0.012 (1.238)

0.961*** (14.818) 0.406*** (6.580) –2.341*** (5.901) 0.012*** (3.333)

0.999 8453.379***

Instrumental Variables 0.106*** (2.705) –0.076*** (3.115) –0.060** (2.569) –0.026* (1.687) –0.044 (1.386) –0.066*** (4.444) –0.068*** (2.759) 0.055 (1.168) 0.944*** (136.681) 0.572*** (4.949) –0.201*** (3.705) 0.004*** (9.991) –0.076*** (5.908) 0.979 1034.476*** 33.166*

Note: Absolute values of t-ratios are in parentheses. A * indicates signiﬁcance at the 10% level in a two-tailed test, ** at the 5% level, and *** at the 1% level. Estimates of constant terms and year dummy variables not reported in all three equations. Estimates of state and year dummy variables not reported in Fixed Effects equation.

64

TIM R. SASS AND DAVID S. SAURMAN

differences across states may be relatively stable through time and can be captured by a set of state dummy variables. Those relevant but unobserved variables which may change through time (social attitudes towards alcohol, for example) may not vary signiﬁcantly across states. As such, the two-way ﬁxed effects estimator may shed some light on the consistency of the OLS estimator. Should the estimates differ considerably, serious consideration should be extended to the issue of selection bias in the OLS estimates. Several sample characteristics necessitate estimating an attenuated version of Eq. (1). There is little or no variation in several regulatory variables over the sample years yielding high degrees of collinearity among the group and period regressors. Similar problems exist with MILEAGE. As such, we offer the abbreviated version of Eq. (1) presented in Table 2. In general, these ﬁxed effects estimates are similar to those obtained with the OLS method, with the exception of MANDATE. The ﬁxed effects estimates indicate a statistically signiﬁcant output expansion of a little more that 3% per year as a result of brewers and dealers acquiring federal antitrust immunity for their exclusivity contracts. This result offers even stronger support for the efﬁciency hypothesis of such vertical restraints and offers more forceful evidence in support of the notion that these restraints enhance overall economic welfare. While the ﬁxed effects results are suggestive of the presence of selection bias in the OLS estimators, a formal test can be had by specifying the factors that determine the pattern of mandates of exclusive territories across states. To this end, the following is offered to explain the probability of states mandating the adoption of exclusivity contracts: MANDATEi,t = g(ANTIPDLAWi,t, CVINCOMEi,t, CVPOPULATIONi,t, JUSTCAUSEi,t ⫺ 1, STATETAXi,t ⫺ 1, EMPLOYMENTi,t ⫺ 1, RETAILERSi,t ⫺ 1, BLGRATIOI,t ⫺ 1, DRYPCTi,t, POPULATIONi,t, INCOMEi,t, LEGCON1844i,t, PCTBAPMORMi,t, ␮i,t),

(2)

where ␮i,t is a random error term. Deﬁnitions and sample means of the variables in Eq. (2) are also provided in Table 1. Several variables enter Eq. (2) with a one-year lag to avoid potential simultaneity problems associated with employing their contemporaneous values. Several variables in Eq. (2) explain the presence of mandates from the perspective of the demand for such regulation. Others have argued that exclusive territories are one mechanism brewers employ to price discriminate. If so, then the presence of an anti-price discrimination statute and relatively uniform incomes within a state would reduce the incentives for brewers to seek mandates. Under this view, ANTIPDLAWi,t can be expected to enter negatively

Mandated Exclusive Territories: Efﬁciency Effects and Regulatory Selection Bias

65

in Eq. (2), while CVINCOMEi,t would carry a positive coefﬁcient. Under the efﬁciency hypothesis, ANTIPDLAWi,t would not inﬂuence brewers’ calculus concerning the seeking of mandates. Additionally, if geographic heterogeneity of population implies that brewers will have varying levels and mixtures of optimal services across territories, then CVPOPULATIONi,t can be expected to carry a positive sign. Brewers will seek mandates as a substitute for more intensive and costlier monitoring of service levels across territories. Given the control of CVPOPULATIONi,t, a negative coefﬁcient on CVINCOMEi,t would further suggest that brewer’s motive for mandates is one of optimal service provision as opposed to price discrimination. If mandates are motivated by efﬁciency concerns, we anticipate JUST CAUSEi,t ⫺ 1 to inﬂuence positively the probability of observing a mandate. Once exclusive territories are employed, as they have been since 1983, brewers face the potential problem of wholesalers under-providing dealer services. Brewers will attempt to retard these activities, in part, with the threat of termination. But where “just” or “good” cause must be demonstrated, termination becomes more expensive and the odds that some dealers may shirk on service provision increases. Under the efﬁciency hypothesis, mandates raise the potential cost to any maverick dealer of shirking by raising the present value of the distributor ﬁrm. That is, showing cause reduces the probability of a dealer being terminated, but a mandate will increase the penalty imposed on the dealer should a termination occur. As such, brewers will tend to seek mandates to strengthen their exclusive contracts in states requiring the showing of cause in franchise termination. We expect brewers to seek mandates more intensively the more populous the state. Larger populations imply larger markets, and larger gains arising from mandates. Similarly, given state size and that beer appears to be an incomenormal resource, we expect the probability of observing a mandate to be greater the larger is per adult real income. Additionally, if the services or enhanced product quality forthcoming from mandates are income-normal, then greater incomes should raise the probability of observing mandates. As beer wholesalers and their employees can generally expect to beneﬁt from mandates, the larger is this group the greater will be their inﬂuence on the regulatory body supplying mandate regulation. POPULATIONi,t, INCOMEi,t, and EMPLOYMENTi,t ⫺ 1 will all enter Eq. (2) with positive coefﬁcients. We make no prediction over the expected sign of RETAILERSI,t ⫺ 1. While this group may generally beneﬁt from additional services, they will suffer an increase in the price of an input. Additionally, some retailers may see locational rents or rents derived from other sources extracted by wholesalers as a result of state exclusivity mandates.

66

TIM R. SASS AND DAVID S. SAURMAN

The remaining variables attempt to capture effects that other interest groups have on the supply of this regulation. Several states possess “local-option” laws resulting in localities that are “dry” for beer. By requiring all brewers to establish exclusive territories, the diversion of beer from wet to dry areas can be traced more readily by state and local authorities, yielding fewer violations of local-option laws and lower enforcement costs. DRYPCTi,t is expected to be positive. If mandates expand output, then the revenue from the taxation of beer available for legislators to distribute to constituent groups also expands. The alternative to increasing these revenues is to raise the tax rate, assuming higher rates yield higher tax revenues. It would seem that the higher the tax rate, the greater are the political costs to legislators of raising the tax rate and the more attractive becomes expanding tax revenue through an expansion of the tax base. Thus, STATETAXi,t ⫺ 1 is expected to exert a positive inﬂuence on the probability of observing mandates of exclusive dealer contracts. Other groups may oppose mandates on various grounds, thereby raising the opportunity cost to legislators and regulators of requiring exclusive dealer contracts for beer. Some groups stand opposed to alcohol consumption on ideological grounds and would tend to disfavor regulation that leads to increased consumption. We proxy this anti-alcohol sentiment in a state with PCTBAPMORMi,t. If the efﬁciency perspective is correct, this variable should inﬂuence the dependent variable in a negative fashion. Alternatively, in state mandates are indeed anticompetitive, then PCTBAPMORMi,t should carry a negative coefﬁcient. Others (Ornstein & Hannsens, 1985; Lee & Tremblay, 1992) have argued that the proportion of the population that is relatively young (ages 18–44) constitutes a cohort of relatively heavy beer drinkers. If this group represents inframarginal consumers who value the additional services forthcoming from mandates at something less than the price increase, then legislators will face relatively high political costs of permitting state mandates to exist and LEGCON1844i,t will possess a negative coefﬁcient. On the other hand, if members of this group are not inframarginal consumers, and do obtain net gains from additional services, a positive inﬂuence on the probability of observing a mandate would be expected. Hence, we let the data speak as to the inﬂuence of this group. Finally, though we can make no prediction concerning the effect of mandates on retail outlets in general, we can recognize that ﬁnal market sellers differ from others. Bars and liquor stores derive a large fraction of their revenue from the sale of beer in comparison to groceries and would seem to be economically damaged relatively more heavily by the wholesale price increase associated with mandates. If so, then the larger the value of BLGRATIOi,t ⫺ 1, the smaller the probability of observing a mandate in the state. Alternatively, such specialized retailers may well be the prime beneﬁciaries, in

Mandated Exclusive Territories: Efﬁciency Effects and Regulatory Selection Bias

67

a relative sense, of greater service levels and expanded output. Again, we make no a priori statement as to the sign of this variable. Table 3 presents the results of applying a Probit estimator to Eq. (2). The general impression from these results is that the data tend to support the efﬁciency, as opposed to the anticompetitive, hypothesis of mandated exclusive Table 3. Probit Estimates of State Mandates of Exclusive Territories (1983–1992 – 480 Observations). Explanatory Variables ANTIPDLAWi,t ⫺ 1 CVINCOMEi,t CVPOPULATIONi,t JUSTCAUSEi,t ⫺ 1 STATETAXi,t ⫺ 1 EMPLOYMENTi,t ⫺ 1 RETAILERSi,t ⫺ 1 BLGRATIOi,t ⫺ 1 DRYPCTi,t POPULATIONi,t INCOMEi,t LEGCON1844i,t PCTBAPMORMi,t Prediction Percent Pseudo-R2 ␹2 (22 df)

Dependent Variable: MANDATEi,t –0.2374 (1.416) –1.2350 (0.068) 0.7253*** (5.552) 0.4562*** (2.742) 1.2731*** (2.794) 1.5129*** (2.726) –0.0001 (0.598) –1.2633*** (5.042) 0.0244*** (2.985) 0.0003 (1.595) –0.0001 (1.606) –0.0439*** (4.098) –0.0679*** (5.799) 0.741 0.260 174.974***

Note: Absolute values of t-ratios are in parentheses. A * indicates signiﬁcance at the 10% level in a two-tailed test, ** at the 5% level, and *** at the 1% level. Estimates of constant term and year dummy variables have been omitted.

68

TIM R. SASS AND DAVID S. SAURMAN

territories. Existing anti-price discrimination laws appear to have no signiﬁcant effect on the presence of mandates. Given population variation, larger variations in average incomes within states – an ideal precondition for locational price differences which are not cost-based – also has no signiﬁcant inﬂuence on the probability of observing mandates of exclusive dealer contracts. Additionally, the estimated coefﬁcients of PCTBAPMORMi,t and STATETAXi,t ⫺ 1 are both consistent with the idea that the associated groups perceive state mandates to have output-expanding effects. In light of the ﬁxed-effects results of Table 2, we turn to re-estimating Eq. (1) by explicitly recognizing that unmeasured factors appear to affect both consumption/output of beer as well as the likelihood of a state mandating exclusive dealer contracts. That is, when OLS is applied to estimate Eq. (1), MANDATEi,t and ␧i,t are correlated and the OLS estimates are inconsistent. Following Barnow, Cain and Goldberger (1980), we employ the ﬁtted values of MANDATEi,t obtained from the estimates of Eq. (2) as an instrument in obtaining the instrumental variables estimates of Eq. (1) that are presented in Table 2.15 We test for selection bias with the method suggested by Hausman (1978). As the ␹2 statistic indicates, we can reject the null hypothesis of consistent OLS estimates at the 10% level and instead rely on the instrumental variables estimates for drawing inferences over the effects of state mandates on output. These estimates are also seen as generally consistent in both sign and signiﬁcance level with both the ﬁxed effects and the OLS estimates, again save MANDATEi,t. Where mandates of exclusive dealer contracts are in force, the data indicate a slightly larger than 10% output expansion statistically signiﬁcant at the 1% level.

IV. CONCLUDING REMARKS Since the pathbreaking work of Stigler (1971) and Peltzman (1976), economists have viewed the notion of an exogenous regulatory structure with a skeptical eye. With this in mind, we have extended our previous assessments of the output effects of manufacturer-designated exclusive dealer contracts. Our ﬁndings indicate that when this segment of the regulatory structure is treated exogenously, the estimated effects that state mandates have on output are understated in both size and signiﬁcance due to selection bias in the OLS estimators. When such bias is accounted for, the data seem to offer even stronger support for the efﬁciency hypothesis of exclusive territories. We do conclude on a note of caution, however. These results should be viewed as preliminary, pointing the way to further research, as several issues

Mandated Exclusive Territories: Efﬁciency Effects and Regulatory Selection Bias

69

have yet to be investigated. Among these are the appropriate method of accounting for the determinants of the regulatory structure. The beer market since the early 1980s, like other markets, has experienced a particular type of diffusion of several regulatory variables. That is, certain regulations, like mandates, are adopted and then are never repealed. Consequently there is greater variation among states in the age of their regulations than in there existence. Potentially an approach along the lines of Shughart and Tollison (1985), where focus is on predicting the timing of passage rather than the existence of a particular regulatory structure may provide a fruitful method for examining the endogeneity of such regulations across states.

NOTES 1. See The Package Shop, Inc. et al. v. Anheuser-Busch et al., CCH 1987-2 Trade Cases ¶ 67,673 at 59,079. 2. United States v. Arnold Schwinn & Co., 388 U.S. 365 (1967). 3. See The Package Shop, Inc. et al. v. Anheuser-Busch et al., CCH 1987-2 Trade Cases ¶ 67,673 at 59,079; Mendelovitz v. Adolph Coors Co. 693 F.2d 570, 573 (1982); and “Cleary Reviews Current Beer Industry Climate,” Modern Brewery Age, October 18, 1982, at 1. 4. See The Package Shop, Inc. et al. v. Anheuser-Busch et al., CCH 1987-2 Trade Cases ¶ 67,673 at 59,077, 59,107; Adolph Coors Co. v. F.T.C. 497 F2d. 1178 (1974). 5. Continental T. V., Inc. et al. v. GTE Sylvania 433 U.S. 36 (1977). 6. See Orbison (1983). 7. See The Package Shop, Inc. et al. v. Anheuser-Busch et al., CCH 1987-2 Trade Cases ¶ 67,673 at 59,078; Assam Drug Co., Inc. v. Miller Brewing Co. 798 F2d. 311 (8th Cir. 1986) at 313; and State of New York et al. v. Anheuser-Busch et al., cited in 1987-2 CCH Trade Cases ¶ 67,777 at 59,200 and 59,202. 8. See U.S. Senate (1988). 9. For a chronology of exclusive territory mandates (see Sass & Saurman, 1993, p. 175). 10. Data limitations, for the most part, govern the sample space. We exclude data from 1982 as 1983 is the ﬁrst whole year that A-B operated with exclusive wholesaler contracts (except in Indiana) and the year that Miller followed A-B’s lead. Alaska and Hawaii are deleted from our sample due to difﬁculties in measuring transportation distances in a fashion consistent with the other forty-eight states. 11. Sass and Saurman (1993) develop this argument in greater detail. 12. Since brewers and wholesalers have expended resources to obtain the federal antitrust immunity provided by the state mandates, we ignore the possibility that these vertical restraints have no structural effects. 13. Blair and Fesmire (1994) consider inframarginal consumers and examine the conditions under which the imposition of vertical restraints that expand demand will, and will not, improve welfare. Boudreaux and Ekelund (1994) offer a critique of Blair and Fesmire by arguing that given interbrand competition among manufacturers (brewers), the existence of such inframarginal consumers becomes a non-issue and the

70

TIM R. SASS AND DAVID S. SAURMAN

effects observed in our OLS estimates signal an improvement in overall welfare similar to that obtained through a general improvement in product quality. 14. A number of empirical studies, particularly in alcoholic beverage markets, have attempted to account for the endogeneity of legislation in various ways. For some recent examples see Dee (1999), Mast, Benson and Rasmussen (1999) and Sass and Nichols (1996). 15. The instrumental variables estimates in Table 2 assume the presence of selection bias in the OLS estimators is due solely to correlation of the error term with MANDATE. We have also tested the consistency of this estimator versus instrumental variables estimators that assume the real tax, price advertising restrictions, and deposit laws are endogenous. In each of these cases, we could not reject the null hypothesis of consistency of the instrumental variables estimator presented in Table 2.

ACKNOWLEDGEMENTS We are grateful to Frank Chaloupka for supplying some of the data and to Marc Ulrich for providing research assistance. Any errors are solely our responsibility.

REFERENCES Barnow, B. S., Cain, G. C., & Goldberger, A. S. (1980). Issues in the Analysis of Selectivity Bias. Evaluation Studies Review Annual, 5, 43–59. Breusch, T. S., & Pagan, A. R. (1979). A Simple Test for Heteroscedasticity and Random Coefﬁcient Variation. Econometrica, 47, 1287–1294. Beer Institute (various years). Brewers Almanac. Washington, D.C.: The Beer Institute. Beverage Marketing Corporation (various years). Beverage Marketing Directory. Mingo Junction, OH: Beverage Marketing Corporation. Blair, R. D., & Fesmire, J. M. (1994). The Price Maintenance Policy Dilemma. Southern Economic Journal, 60, 1043–1047. Boudreaux, D., & Ekelund, R. B. Jr. (1988). Inframarginal Consumers and the Per Se Legality of Vertical Restraints. Hofstra Law Review, 17, 137–158. Boudreaux, D., & Ekelund, R. B. Jr. (1994). A Note on Vertical Restraints and Inframarginal Consumers: Anatomy of the Analysis. Unpublished manuscript. Bradley, M. et. al. (1992). Churches and Church Membership in the United States 1990. Atlanta, GA: Glenmary Research Center. Business Journals, Inc. (various years). Modern Brewery Age Bluebook. Norwalk, CT: Business Journals, Inc. Carstensen, P. C., & Dahlson, R. F. (1986). Vertical Restraints in Beer Distribution: A Study of the Business Justiﬁcations for and Legal Analysis of Restricting Competition. Wisconsin Law Review, 1–81. Commerce Clearing House, Inc. (various updates). Liquor Control Law Reporter. Chicago, IL: Commerce Clearing House, Inc. Culbertson, W. P. (1989). Beer-Cash Laws: Their Economic Impact and Antitrust Implications. Antitrust Bulletin, 34, 209–229. Culbertson, W. P., & Bradford, D. (1991). The Price of Beer: Some Evidence from Interstate Comparisons. International Journal of Industrial Organization, 9, 275–289.

Mandated Exclusive Territories: Efﬁciency Effects and Regulatory Selection Bias

71

Dee, T. S. (1999). State Alcohol Policies, Teen Drinking and Trafﬁc Fatalities. Journal of Public Economics, 72, 289–315. Hausman, J. A. (1978). Speciﬁcation Tests in Econometrics. Econometrica, 46, 1251–1271. Jordan, W. J., & Jaffee, B. L. (1987). The Use of Exclusive Territories in the Distribution of Beer: Theoretical and Empirical Observations. Antitrust Bulletin, 32, 137–164. Keller Publishing Corp. (various years). Beverage World’s Databank. Shepherdsville, KY: Keller Publishing Corp. Klein, B., & Murphy, K. M. (1988). Vertical Restraints As Contract Enforcement Mechanisms. Journal of Law and Economics, 31, 265–297. Lee, B., & Tremblay, V. J. (1992). Advertising and the U.S. Market Demand for Beer. Applied Economics, 24, 69–76. Mast, B. D., Benson, B. L., & Rasmussen, D. W. (1999). Beer Taxation and Alcohol-Related Trafﬁc Fatalities. Southern Economic Journal, 66, 214–249. McMahon, W. W., & Chang, S. (1991). Geographical Cost of Living Differences: Interstate and Intrastate, Update 1991. Center for the Study of Educational Finance, Illinois State University, Normal, IL. NextBase Ltd. (1991). Automap, Version 2 [computer program]. Phoenix, AZ: Automap, Inc. Orbison, M. H. (1983). Vertical Restraints in the Brewing Industry: Is the Malt Beverage Interbrand Competition Act the Answer? Brooklyn Law Review, 50, 143–189. Ornstein, S. I., & Hanssens, D. M. (1985). Alcohol Control Laws and the Consumption of Distilled Spirits and Beer. Journal of Consumer Research, 12, 200–213. Peltzman, S. (1976). Toward a More General Theory of Regulation. Journal of Law and Economics, 19, 211–240. Posner, R. A. (1977). The Rule of Reason and the Economic Approach: Reﬂections on the Sylvania Decision. University of Chicago Law Review, 44, 1–20. Quinn, B. et. al. (1982). Churches and Church Membership in the United States 1980. Atlanta, GA: Glenmary Research Center. Rey, P., & Stiglitz, J. E. (1995). The Role of Exclusive Territories in Producers’ Competition. Rand Journal of Economics, 26, 431–451. Sass, T. R., & Nichols, M. W. (1996). Scope-of-Practice Regulation: Physician Control and the Wages of Non-Physician Health-Care Professionals. Journal of Regulatory Economics, 9, 61–81. Sass, T. R., & Saurman, D. S. (1993). Mandated Exclusive Territories and Economic Efﬁciency: An Empirical Analysis of the Malt Beverage Industry. Journal of Law and Economics, 36, 153–177. Sass, T. R., & Saurman, D. S. (1996). Efﬁciency Effects of Exclusive Territories: Evidence from the Indiana Beer Market. Economic Inquiry, 34, 597–615. Stigler, G. J. (1971). The Theory of Economic Regulation. Bell Journal of Economics, 2, 3–21. Shughart, W. F. II, & Tollison, R. D. (1985). Corporate Chartering: An Exploration in the Economics of Legal Change. Economic Inquiry, 23, 585–599. Telser, L. G. (1961). Why Should Manufacturers Want Fair Trade? Journal of Law and Economics, 3, 86–105. U.S. Department of Commerce, Bureau of the Census (various years). Census of Retail Trade.Washington, D.C.: U.S. Department of Commerce, Bureau of the Census. U.S. Department of Commerce, Bureau of the Census (various years). Census of Service Industries. Washington, D.C.: U.S. Department of Commerce, Bureau of the Census.

72

TIM R. SASS AND DAVID S. SAURMAN

U.S. Department of Commerce, Bureau of the Census (1995). Resident Population of States [table downloaded from Census Bureau FTP site]. Washington, D.C.: U.S. Department of Commerce, Bureau of the Census. U.S. Department of Commerce, Bureau of the Census (various years). Statistical Abstract of the United States. Washington, D.C.: U.S. Department of Commerce, Bureau of the Census. U.S. Department of Commerce, Bureau of the Census (1994). USA Counties 1994 [CD-ROM]. Washington, D.C.: U.S. Department of Commerce, Bureau of the Census. U.S. Department of Commerce, Bureau of Economic Analysis (1993). State Annual Summary Tables, 1929–1992 [computer diskette]. Washington, D.C.: U.S. Department of Commerce, Bureau of Economic Analysis. U.S. Department of Labor, Bureau of Labor Statistics (various years). Employment and Earnings, Annual Averages. Washington, D.C.: U.S. Department of Labor, Bureau of Labor Statistics. U.S. Senate (1988). The Malt Beverage Interbrand Competition Act: Hearings Before the Committee on the Judiciary, United States Senate, 100th Congress, 1st. Session. Washington, D.C.: U.S. Governtment Printing Ofﬁce. White, H. (1978). A Heteroscedasticity Consistent Covariance Matrix and a Direct Test for Heteroscedasticity. Econometrica, 48, 817–838.

RACE AND RADIO: PREFERENCE EXTERNALITIES, MINORITY OWNERSHIP, AND THE PROVISION OF PROGRAMMING TO MINORITIES Peter Siegelman and Joel Waldfogel ABSTRACT Market provision of radio programming is beset by possible inefﬁcient underprovision of formats that appeal to small audiences, for which the social beneﬁts of programming – but not advertising revenue – exceed their costs. Larger markets have more programming, so their listeners derive beneﬁts from being in the same market as others with similar preferences, a mechanism we term “preference externalities.” Yet, because white and minority content preferences are substantially different, preference externalities are positive only within group. We expect problems of inefﬁcient underprovision to be more likely for small minority populations. We ﬁnd evidence that policies promoting minority ownership increase the amount of minority-targeted programming.

Advertising and Differentiated Products, Volume 10, pages 73–107. Copyright © 2001 by Elsevier Science Ltd. All rights of reproduction in any form reserved. ISBN: 0-7623-0823-0

73

74

PETER SIEGELMAN AND JOEL WALDFOGEL

I. INTRODUCTION Economic theory predicts that markets will do a poor job of allocating public goods. It is no surprise, therefore, that we see relatively little market provision of national defense. Curiously, however, the U.S. relies almost exclusively on the market to provide one of the few other textbook examples of public goods, radio broadcast signals.1 Market provision of radio programming is beset by an important potential problem: inefﬁcient underprovision of unpopular formats whose social beneﬁt – but not advertising revenue – exceeds their costs.2 Because commercial radio programming is ﬁnanced through advertising, the market will provide only those stations whose advertising revenues are sufﬁcient to cover their costs. Advertising revenues, in turn, are largely a function of the number of listeners a station can garner. The more people there are in a given geographic market, therefore, the greater is the number of stations that market can support, and hence the greater the variety of stations that are available to listeners. Listeners thus derive beneﬁts from being in the same market as others with similar preferences – nearby individuals with similar tastes help defray the ﬁxed costs of providing the programming they all prefer. We term these beneﬁts “preference externalities.” See Waldfogel (1999) for an extensive discussion of preference externalities in radio broadcasting. George and Waldfogel (2000) presents evidence on preference externalities in daily newspaper markets. These studies outline how preference externalities can be positive, negative, or zero across groups. If preference externalities are important, it follows that small groups with distinct preferences are especially likely to be inefﬁciently underserved by the radio market. The focus of this paper is on one particular kind of group – racial and ethnic minorities.3 We show that Black and white (and Hispanic/Anglo) preferences in radio programming are substantially different. Hence, minorities and whites in effect constitute separate radio markets. This in turn means that the problem of inefﬁcient underprovision will be more acute for minorities, simply because their smaller populations confer smaller preference externalities on each other. It is important to state at the outset the difﬁculty of identifying absolute instances of inefﬁcient underprovision. Because radio signals are unpriced, we never observe their value to listeners; hence, we cannot conclusively demonstrate circumstances in which the social beneﬁt of potential programming, but not the associated ad revenue alone, exceeds its cost. Instead, we advance a theoretical argument about underprovision of programming to small

Race and Radio

75

audiences, and then show empirically that the argument plausibly applies to minority listeners in the U.S. today. Although not explicitly couched in terms of preference externalities, there appears to be widespread concern about the level of programming offered to racial and ethnic minorities. For example, the FCC has employed a series of policies to promote minority ownership of stations, and at least one rationale for these ownership preferences is to increase the amount of minority-targeted programming on the air.4 But empirical support for the importance of preference externalities as the mechanism responsible for underprovision of broadcasting to minorities has thus far been lacking. The simple facts tell a somewhat contradictory story. On the one hand, the number of minority-owned radio stations remains low (3.4%), and the fraction of black-owned stations has declined from 3.0% in 1993 to 2.2% in 1997. At the same time, however, the number of black-targeted stations has increased. The remainder of this paper attempts to assess the empirical evidence for the existence of preference externalities that result in underprovision of broadcasting favored by racial and ethnic minorities. We proceed in four steps. First, we offer a theoretical sketch that compares the actual and efﬁcient provision of radio broadcasting. We suggest that the market can inefﬁciently underprovide programming appealing to small audiences at the same time as it inefﬁciently overprovides formats popular among large audiences. Second, we brieﬂy describe the data used in the study. In the third section, we empirically examine the inﬂuence of local radio markets’ racial/ethnic composition on the types of programming that are provided. We begin with striking evidence of preference externalities, which operate only within ethnic groups. In a given geographic market, additional blacks, Hispanics, or whites confer a beneﬁt only on members of their own group, not on other ethnicities. We then turn to explain the mechanism by which these preference externalities operate. We show that preferences for programming among minorities and whites are very different, that the amount of local minority-targeted programming depends on the size of the minority (but not the white) population, and that minority-targeted programming attracts minority audiences to radio listening. In the ﬁnal section of the paper, we discuss the use of preferences for minority ownership as a possible solution to the problem of inefﬁcient underprovision. At a minimum, such preferences only make sense if two conditions are met: (1) minority-owned stations do actually broadcast minoritytargeted programming; and (2) the additional minority-targeted programming they provide must not simply crowd-out existing programming. We ﬁnd strong empirical support for both conditions.

76

PETER SIEGELMAN AND JOEL WALDFOGEL

Two strands of literature are relevant to the present inquiry. First, there are a few studies concerned speciﬁcally with justiﬁcations for minority ownership preferences in U.S. broadcasting. Spitzer (1991) presents theoretical arguments and evidence justifying minority preferences in broadcasting, as well as an excellent overview of the relevant legal doctrines and policies (many of which have, however, changed since 1991). Dubin and Spitzer (1993) examine the station-level relationship between minority ownership and radio station programming formats as of 1987. Both papers document that minority-owned stations are more likely than white-owned stations to broadcast in minoritytargeted formats, leading the authors to conclude that increased minority ownership increases the amount of minority programming. However, because both minority- and white-owned stations may compete for minority audiences – indeed, we show that most minority-targeted stations are white-owned – the Dubin and Spitzer evidence does not indicate whether minority ownership affects the total available amount of minority-targeted programming or simply replaces an equivalent amount of white-owned, minority-targeted programming. We build on existing work in a variety ways. We employ more recent, longitudinal data, and examine the relationship between the amount of minority ownership and the amount of minority programming at the market level, which allows for an analysis of “crowding out” that is precluded by a ﬁrm-level analysis. The other relevant strand of the theoretical literature asks whether markets for differentiated products will generally provide the right products. For example, see Spence (1976a, b), Dixit and Stiglitz (1977), or Mankiw and Whinston (1984). In the context of broadcast markets in particular, these questions have been addressed by Steiner (1952), Spence and Owen (1977), Beebe (1977), and, more recently, Anderson and Coate (2000). A small empirical literature focuses on the adequacy of entry into radio broadcasting. Papers include Berry and Waldfogel (1999b), which examines the possible inefﬁciency of excessive entry; Berry and Waldfogel (1999a), which asks whether public radio corrects problems of inefﬁcient underprovision; and Rogers and Woodbury (1996), who document that audience size increases in programming variety and that programming variety increases with audience diversity.

II. EFFICIENT AND ACTUAL PROVISION OF RADIO BROADCASTING Radio signals are pure public goods. A station should be provided to a local market if its social beneﬁt – including beneﬁts to listeners and advertisers –

Race and Radio

77

exceeds its cost.5 Given the presence of existing stations, an additional station should be provided only if the sum of its incremental beneﬁts across listeners (the total beneﬁt with the new station, less total beneﬁt of pre-existing stations) plus net new ad revenue (revenue of the new station less business diverted from existing stations) exceeds its cost. Because the only part of their social beneﬁt that commercial radio stations can capture is advertising revenue, we expect some inefﬁcient underprovision of relatively unpopular programming.6 This places attention on the sort of station allocation brought forth by the market. This section discusses the determinants of radio broadcast entry with an eye toward the circumstances that mitigate (or worsen) the underprovision problem; that is, the failure to provide programming with social beneﬁt – but not advertising revenue – in excess of its cost. Because commercial radio programming is ﬁnanced through advertising, the market provides only those stations that can cover their costs with advertising revenue. Under free entry – and if station ﬁxed costs do not vary with the size of the market – we expect the number of stations to be roughly proportional to population. We expect that listener welfare – as reﬂected by the number of listeners – will increase in the number of local stations which, in turn, depends on the size of the local market. Thus, listeners derive a beneﬁt from the size of the local market. In a larger market, more varieties of programming can garner sufﬁcient audiences to cover operating costs with advertising revenues, so that listeners derive external beneﬁts from being in the same market as others with similar preferences. We term these beneﬁts preference externalities. Market size is important because a larger potential audience can undo the inefﬁcient underprovision problem. Imagine a proposed station that would attract 5% of the local population as its audience, each of them valuing the station at $100 per year.7 Suppose that advertising revenue is $100 per listener/ year, and that the station costs $750 per year to operate. If the local population is 100, the station is not provided, even though its social beneﬁt exceeds its cost. If the local population is instead 200, then the station can proﬁtably enter, since it can cover its costs with ad revenue alone. This example illustrates another important and unusual feature of the radio market: listener valuations are irrelevant in determining whether a station will operate, because listeners do not pay for radio. In the example above, the station would not be provided in a 100-person market, even if its potential listeners valued it at $1000 each. Conversely, in the 200-person market, the station would be provided even if its listeners valued it at only $5 each (or even less). Throughout this paper, we assume that advertisers value all listeners alike, so that only the total number of listeners matters in determining station format and

78

PETER SIEGELMAN AND JOEL WALDFOGEL

revenue. There is considerable anecdotal evidence that the race of a station’s listeners does matter to advertisers, however.8 Advertisers apparently value white listeners more highly than Hispanic or black listeners, presumably because the former have larger incomes and spend more on consumer goods on average.9 The analysis in the rest of this paper focuses exclusively on how their smaller numbers affect the programming variety offered to minorities. But to the extent that racial or ethnic minorities are less desirable advertising targets than whites, the market penalizes them with an additional handicap that might strengthen the case for some kind of intervention. Overall market size can undo inefﬁcient underprovision only to the extent that all listeners have the same preferences and comprise a single homogenous market. A natural question to ask, then, is whether all potential listeners confer similar preference externalities on one another. For example, do whites beneﬁt from the presence of blacks in the same geographic market, and vice-versa? Or do individuals only beneﬁt only from being in the same geographic market as others in the same “preference group?” Large populations attract more entry and can therefore go a greater distance toward undoing inefﬁcient underprovision than small populations, and black and Hispanic populations are small. Across 244 markets in 1993, blacks make up an average of only 9.9%, and Hispanics only 6.4% of total population. Hence, if listening preferences do differ across groups – so that each group only confers positive beneﬁts on its own members – we expect more severe underprovision of black- and Hispanic-targeted programming than whitetargeted programming. All of this leads us to question the adequacy of programming for minority listeners. Classical music fans are one sort of minority. While they are not a legally protected class, they – along with jazz and news fans – do have state and federal subsidies designed to promote their preferred programming.10 As we demonstrate below, racial and ethnic minorities have distinct preferences, and in most markets are small in numbers – precisely the conditions under which theory suggests the market will underprovide their preferred programming.

III. DATA We have station-level information for commercial stations in 244 markets in 1993 and 1997. We observe total listening, call letters, AM/FM status, programming format, owner identity, owner race (whether white, black, or Hispanic), and whether the station broadcasts from inside or outside of the metropolitan area where it is received for 5219 underlying stations in 1993 and

Race and Radio

79

for 5990 underlying stations in 1997. We observe station-level black and Hispanic listening for a subset of the metropolitan areas.11 The data are derived from a variety of sources. Programming format, owner identity, and AM-FM status information are from Duncan’s American Radio, Spring 1993 and 1997. Duncan classiﬁes stations into 43 formats, which we report in Table 1. Listening data from are from Arbitron’s Radio USA, Spring 1993 and 1997. We use Arbitron’s average quarter hour (AQH) as our listening measure. AQH listening share is the percentage of persons in a given group listening to radio for at least ﬁve minutes during an average quarter hour period between 6AM and midnight. We obtained owner race information from the National Telecommunications Information Agency, which maintains lists of radio and television broadcast facilities owned by blacks, Hispanics, and other minorities.12 The organization of the station-level data merits some discussion. Each observation is not a station but rather a city-station pair. A station such as WCBS-AM in New York also has substantial numbers of listeners in other nearby markets, such as Stamford, Bridgeport, New Haven, etc. The entry for WCBS-AM in New York shows its New York listening and that it broadcasts from inside that metropolitan area. Its entry in Bridgeport, by contrast, shows its Bridgeport listening and that it broadcasts from outside of the Bridgeport metro area. We treat simulcasting stations (multiple transmitters simultaneously broadcasting the same programming on different frequencies) as single stations in each market where they are received. Our basic sample of 244 markets (for which we have both 1993 and 1997 data, excluding minority listening data) covers areas that included 167 million persons in 1993. Table 2 reports basic summary statistics. Black radio station ownership declined substantially between 1993 and 1997: the average number of black-owned stations received in the markets in our dataset fell by 15.4%, from 0.65 to 0.55. By contrast, Hispanic station ownership increased over the same period, from an average of 0.18 to 0.28 stations per market. The total number of stations and programming formats received in these markets both grew: average stations per market rose from 21.4 in 1993 to 24.5 in 1997, while average available formats per market grew from 11.5 to 14.9. The growth in overall variety (total number of formats) was mirrored by the rise in programming aimed at minority audiences. Despite the decline in black ownership, the average number of stations broadcasting programming targeted at black audiences increased by 27%, from 1.5 in 1993 to 1.9 in 1997, while Hispanic-targeted stations grew 57%, from 0.68 to 1.07 per market.13 Despite the growth in both available stations and programming variety, overall AQH listening declined from 16.80% in 1993 (meaning that 16.8% of the population

80

PETER SIEGELMAN AND JOEL WALDFOGEL

Table 1. Stations and Listening, by Race and Format, 1997. 101 Markets with Black Listening

54 Markets with Hispanic Listening

Percent of Listening: Format Adult Contemporary (AC) AC/Contemp. Hit Radio Adult Contemporary/New Rock AC/Soft Adult Contemporary Album Oriented Rock (AOR) AOR/Adult Contemporary AOR/Classic Rock Album Oriented Rock/New Rock Album Oriented Rock/Progressive Black Black/Adult Contemporary Black/Gospel Black/Oldies Black/Talk Big Band/Nostalgia Big Band/Nostalgia/Religious Country Country/Full Service Contemporary Hit Radio (CHR) CHR/Adult Contemporary CHR/New Rock Contemporary Hit Radio/Urban Classical Classic Album Oriented Rock Classic Hits Ethnic Easy Listening Full Service/Variety Full Service/Variety/Talk Gospel Jazz News News/Talk Oldies Religious Soft Adult Contemporary Spanish Sports Talk Talk/Classic AOR Talk/Full Service Talk/Jazz Unknown

Percent of Listening:

Percent of Non- Black Percent of NonHispanic Stations Black Stations Hispanic 5.9 2.1 0.6 0.1 5.3 0.1 0.2 3.0 1.0 7.5 3.2 1.2 1.0 0.2 4.7 0.0 13.0 0.1 5.6 0.8 0.2 1.1 1.2 3.3 1.5 0.2 0.2 0.9 1.2 2.0 2.0 1.4 2.5 5.6 5.4 3.4 2.9 2.9 6.0 0.2 0.0 0.1 0.1

6.7 2.9 1.1 0.1 6.0 0.1 0.1 3.7 1.3 1.7 0.8 0.0 0.1 0.0 4.2 0.0 11.9 0.2 6.7 0.9 0.3 2.7 2.4 4.0 1.5 0.1 0.0 0.7 2.4 0.1 2.3 3.0 3.2 6.7 1.2 5.2 7.4 2.1 6.1 0.4 0.0 0.2 0.0

2.0 0.8 0.3 0.1 0.7 0.0 0.0 0.5 0.1 32.5 18.3 1.8 2.4 1.4 0.5 0.0 1.5 0.0 2.5 0.3 0.1 7.7 0.5 0.5 0.2 0.6 0.0 0.2 0.8 3.8 6.5 2.9 1.0 1.7 2.5 2.3 0.2 1.0 1.7 0.1 0.0 0.1 0.0

5.0 2.3 1.1 0.1 4.5 0.1 0.1 3.1 1.8 1.6 1.4 0.1 0.3 0.3 4.2 0.1 10.6 5.2 0.4 0.1 0.1 1.5 1.8 4.1 1.1 0.4 0.2 0.5 0.6 0.4 3.0 1.9 3.3 5.2 3.5 3.8 16.1 3.0 7.0 0.1 0.1 0.1 0.0

5.4 2.9 1.5 0.1 4.9 0.1 0.0 3.3 1.6 4.0 4.6 0.0 0.5 0.4 4.1 0.0 8.7 6.0 0.8 0.0 0.2 4.2 3.0 3.8 1.1 0.3 0.0 0.3 1.7 0.3 4.2 4.6 3.4 6.1 1.2 4.7 0.5 2.7 8.0 0.5 0.0 0.2 0.0

3.4 1.5 0.6 0.0 2.4 0.0 0.0 1.9 0.6 1.6 1.7 0.0 0.0 0.0 1.2 0.0 4.0 8.6 0.4 0.0 0.0 6.3 0.9 2.3 0.3 0.2 0.0 0.1 0.2 0.0 2.0 1.1 1.1 4.4 0.9 3.2 45.7 1.0 2.2 0.0 0.0 0.0 0.0

Race and Radio

81

Table 2. Per-Market Averages. 1. All 244 Markets 1993 Stations Black-Targeted Black-Owned Hispanic-Targeted Hispanic-Owned AQH Listening Formats Owners HHI Population (000) Black Population Hispanic Population

21.39 1.52 0.65 0.68 0.19 16.80 11.48 18.65 1269.83 685.18 84.32 69.54 2. 73 Markets with Black Listening Data in both Years 1993

Stations Black-Targeted Black-Owned Non-Black AQH Listening Black AQH Listening Formats Owners HHI Non-Black Population Black Population

25.33 3.53 1.60 16.71 18.51 13.48 21.60 1040.87 1235.61 236.33

3. 31 Markets with Hispanic Listening Data in both Years 1993 Stations Black-Targeted Black-Owned Non-Hispanic AQH Listening Hispanic AQH Listening Formats Owners HHI Non-Hispanic Population Hispanic Population

31.65 4.39 0.97 16.83 18.18 14.94 26.65 806.38 1824.05 443.26

1997 24.55 1.93 0.55 1.07 0.28 15.78 14.88 14.92 2091.66 707.74 – –

1997 28.01 4.26 1.36 15.67 18.14 17.25 16.45 1995.44 1262.79 245.34

1997 33.90 6.16 1.39 15.96 17.71 18.65 20.52 1578.67 1827.58 503.75

82

PETER SIEGELMAN AND JOEL WALDFOGEL

listened to radio for 5 minutes during an average quarter hour) to 15.78% in 1997. Ownership concentration increased sharply, with the HHI rising from 1297 in 1993 to 2092 in 1997. Panel 2 of Table 2 reports the same variables for the 73 markets with black listening data for both 1993 and 1997. The patterns in the full dataset are also observed in these markets. Here, we report black and non-black listening separately.14 Black AQH listening is roughly 11% (1.9 percentage points) higher than non-black listening. While 16.7% of non-blacks listen to radio for ﬁve minutes during an average quarter hour in 1993, 18.5% of blacks listen. AQH listening declines over the period for both blacks and non-blacks, however. Panel 3 reports the same variables for the 31 markets with Hispanic listening data in both 1993 and 1997. Patterns are similar.

IV. DETERMINANTS OF MINORITY-TRAGETED PROGRAMMING 1. Direct Evidence of Preference Externalities We begin our characterization of minority-targeted programming with this paper’s central fact, which is demonstrated in Table 3: a group’s AQH listening share increases in its own population size, but is invariant with respect to the size of other groups. Column 1 demonstrates that overall AQH listening (for all groups together) increases with local population, our measure of the size of the market. This is powerful and direct evidence of a macro-level preference externality: a larger audience brings forth more stations and greater variety, thereby attracting listeners who would otherwise not have tuned in to radio. The remaining columns in the ﬁrst part of Table 3 show how particular groups’ AQH listening vary with own-group and other-group population. The results are striking. For each group, AQH listening grows in own-group size and is completely invariant with the size of the remaining population. That is, blacks listen more to the radio in markets where there are more blacks, but additional whites have no effect on black listening. Preference externalities in radio programming thus operate only within racial or ethnic groups. Results are similar when regressions include region dummies. The size of the preference externality varies across groups. An additional million whites in a market increases the market’s white AQH listening by 0.4 percentage points. An additional million blacks or Hispanics raises their respective AQH listening by 3 and 1 percentage points.15 This evidence makes it clear that a group’s population has an important effect on its radio listening. Below we detail the mechanism underlying this effect. The latter half of Table

All Listening

NonBlack Listening

Black Listening

Non-Hisp· Listening

Hispanic Listening

All Listening

Dep· Var· = AQH*100 Constant Population

15.569* (0.082) 0.3030* (0.053)

Non-Black Population Black Population Non-Hisp. Population Hispanic Population R-sq N

0.1184 244

15.150* (0.145)

17.401* (0.229)

0.403* (0.162) –0.436 (0.846)

–0.090 (0.256) 3.002* (1.335)

0.1735 99

0.1591 99

15.467* (0.103)

0.390* (0.103) –0.489 (0.345) 0.2887 51

Non-Black Listening

Black Listening

Non-Hisp· Listening

Hispanic Listening

Race and Radio

Table 3. Direct Evidence of Preference Externalities, 1997.

Dep· Var· = ln(AQH/(1-AQH)) 17.161* (0.309)

–0.011 (0.193) 1.067** (0.649) 0.1121 51

–1.693* (0.006) 0.022* (0.004)

0.1134 244

–1.726* (0.011)

–1.564* (0.016)

0.031* (0.013) –0.034 (0.065)

0.006 (0.018) 0.202* (0.096)

0.1674 99

0.1438 99

–1.700* (0.012)

–1.580* (0.022)

0.029* (0.008) –0.035 (0.026) 0.2776 51

0.0002 (0.014) 0.071 (0.046) 0.1077 51

Notes: All population ﬁgures are in millions. The left side of the table reports OLS regressions using the listening share*100 as the dependent variable, allowing easy interpretation of coefﬁcients. The right side of the table reports regressions using the log-odds ratio of the share as the dependent variable. Standard Errors in parentheses. Asterisk indicates 95% signiﬁcance level. Double asterisk indicates 90% signiﬁcance level.

83

84

PETER SIEGELMAN AND JOEL WALDFOGEL

3 performs the exercise substituting the log odds ratio of AQH listening for AQH listening itself as the dependent variable, giving rise to a dependent variable that varies continuously and without bound, above and below zero. Substantive results are similar.16 2a. Do Preferences Differ By Race? Even given the direct evidence of preference externalities above, minority groups can only experience inadequate programming if they prefer different programming from whites. For example, it would be odd to say that the market provides insufﬁcient programming for left-handed persons. The reason is that despite their relatively small numbers, left-handed listeners’ preferences in radio are presumably identical to those of right-handers. If preferences do not differ by race/ethnicity, then there is no meaningful distinction between the adequacy of programming variety available to minorities and the adequacy of programming variety generally. This section therefore compares white, black, and Hispanic choices in radio programming. We ﬁnd very little overlap in listenership – by and large, blacks listen to black format stations, whites listen to white format stations, and Hispanics to Hispanic format stations.17 Columns 2 and 3 of Table 1 report 1997 listening data, by format and race (black and non-black), for the 101 markets reporting black listening data. It is obvious that blacks and non-blacks listen to very different programming. Just over half of black listening is concentrated in only two formats, Black, and Black/Adult Contemporary, which account for less than 2.5% of non-black listening. Blacks make up the majority of listeners to stations in seven formats: Black, Black/Adult Contemporary, Black/Gospel, Black/Oldies, Black/Talk, Gospel, and Ethnic. (We classify these formats as “black-targeted.”) Other formats attracting substantial amounts of black listening include Contemporary Hit Radio/Urban and Jazz. Altogether, black-targeted formats attract 61% of all black listeners, but only about 3% of white listeners. The Duncan index is commonly used to measure segregation – that is, the degree to which the allocation of blacks and whites to neighborhoods or formats differs from shares that are proportional to each group’s population share. The index gives the proportion of all blacks and whites who would have to move (change format) in order to achieve completely integrated listening. For radio, the average 1997 black/white Duncan index is 72.2, which is comparable to levels of black/white residential segregation.18 The last two columns of Table 1 report 1997 listening data, by format and Hispanic status, for the 54 markets with 1997 Hispanic listening data. Like

Race and Radio

85

blacks, Hispanics listen to different programming than non-Hispanics. Hispanic listeners make up the majority of listeners to the broad Spanishlanguage format, “Spanish,” which attracts 45.7% of Hispanic listening. We classify Spanish stations as Hispanic-targeted. Other formats substantial numbers of Hispanic listeners include Contemporary Hit Radio (attracting 8.6% of Hispanic listeners) and Contemporary Hit Radio/Urban (6.3%). Hispanic listeners are somewhat less segregated than blacks, with an average Duncan index of 46.9. 2b. Do Minority Listeners Value Minority-Targeted Programming? The previous section demonstrated that whites, blacks, and Hispanics each listen to different programming. A slightly stronger test for whether groups value programming targeted towards them is to ask whether targeted programming actually attracts listeners from non-listening. If additional targeted programming reduces the number of non-listeners, the new listeners reveal that they prefer the programming to whatever outside option they forego by turning on the radio. Table 4 reports results of regressions of black and white AQH listening percentages on the numbers of white-targeted and black-targeted radio stations for 1997. The ﬁrst ﬁve columns show OLS regressions of the AQH share on numbers of stations. The remainder of the table shows OLS regressions in which the dependent variable is the log odds of AQH listening – that is, ln(share listening/share not listening). The results are clear. Listening increases in the number of own-race stations and is far less sensitive to the number of other stations. The ﬁrst column shows that overall listening increases in the total number of stations. We then examine the relationship between a group’s listening and the number of stations targeted at it, along with the number of stations targeted at other listeners. Each group’s listening depends strongly on the stations targeted at it, and to a lesser extent – or not at all – on those targeted at other groups.19 The OLS results are likely to suffer from the endogeneity problem that entry will tend to occur in markets where there is a high (but unobservable) tendency to listen. This would bias the coefﬁcients in Table 4 upward. To correct for this problem, we require instrumental variables that determine entry of black, Hispanic, and white-targeted stations without directly affecting AQH listening. Measures of market size, such as population, are natural candidates. We explore these instruments next.

86

Table 4. 1997 Listening and Stations, by Type of Stations and Listening. All Listening

NonBlack Listening

Black Listening

Non-Hisp· Listening

Hispanic Listening

All Listening

Dep· Var· = AQH*100 Constant

Non-BlackTargeted Stations Black-Targeted Stations Non-HispanicTargeted Stations HispanicTargeted Stations R-sq N

0.2120 244

12.586* (0.436)

15.067* (0.813)

0.120* (0.015) 0.038 (0.051)

0.068* (0.028) 0.313* (0.096)

0.3921 99

0.1347 99

14.829* (0.522)

0.044* (0.019) –0.023 (0.042) 0.1049 51

Black Listening

Non-Hisp· Listening

Hispanic Listening

Dep· Var· = ln(AQH/(1-AQH)) 17.474* (0.882)

–0.027 (0.032) 0.152* (0.072) 0.0958 51

–1.801* (0.016) 0.005* (0.0006)

0.2106 244

–1.923* (0.034)

–1.728* (0.058)

0.009* (0.001)

0.005* (0.002)

0.3912 99

0.1304 99

–1.748* (0.039)

–1.566* (0.062)

0.003 (0.004) 0.003* (0.001) –0.001 (0.003) 0.0997 51

0.022* (0.007) –0.002 (0.002) 0.011* (0.005) 0.1005 51

Notes: Standard errors in parentheses. Listening is measured as the log-odds ratio of AQH listening. Asterisk indicates 95% signiﬁcance level. Double asterisk indicates 90% signiﬁcance level.

PETER SIEGELMAN AND JOEL WALDFOGEL

All Stations

14.142* (0.215) 0.067* (0.008)

Non-Black Listening

Race and Radio

87

2c. Station Entry and Market Size Given that blacks, Hispanics, and whites have different listening preferences, we expect the number of stations targeted to each group to vary with the size of the group in a given geographic market. Here we distinguish between stations broadcasting from inside and outside the metro area. Table 5 reports regressions of the number of inside stations targeted towards different groups on ﬁrst- and second-order terms in the groups’ populations, as well as the number of outside stations targeting the group.20 The ﬁrst ﬁve columns of the table include all markets for which we have data. As expected, the number of group-targeted stations increases as the group increases in size; but curiously, the number of group-targeted stations decreases in the size of the remaining population. We suspect that this was due to technological limits on the number of stations in large markets, where the broadcast spectrum is likely to be so crowded as to forestall additional entry. In the last ﬁve columns, we report the same regressions for markets with under 2.5 million persons in 1997. The curious result is diminished but does not go away. In all but the largest markets, the number of group-targeted stations increases as the group’s population rises, and is less sensitive to the population of the other group. Waldfogel (1999) documents that negative cross-group effects in entry reﬂect minority listeners switching from white-targeted to minority-targeted stations as the latter become available. These regressions indicate that the size of a market’s minority population determines the number of minority-targeted stations, while the population of whites in the market is largely irrelevant. In Table 6 we revisit the relationship between station entry (by target group) and group AQH listening share, using IV estimates with terms in population as instruments.21 The IV results in Table 6 are virtually identical to the OLS results in Table 4: each group’s listening depends only on the number of stations targeted at it; the number of stations targeted at the other group has no effect on its listening. These results provide strong evidence that groups value programming that targets them, and are far less sensitive to non-targeted programming.22 3. Problems with Market Provision of Minority Programming Preference externalities are not just a theoretical curiosity: it is clear that blacks confer beneﬁts on other blacks, Hispanics on other Hispanics, and whites on other whites. The market therefore provides fewer stations appealing to racial and ethnic minorities, relative to whites, simply because these groups are less numerous. Without knowing something about the value that listeners place on

88

Table 5. 1997 Inside Station Entry and Population, by Race and Hispanic Status. All Stations

NonBlack

Black

NonHispanic

Hispanic

All Stations

All Markets Constant Population (millions) Pop. Squared

15.220* (0.450) 6.614* (0.436) –0.347* (0.040)

Non-Black Pop

Black Pop. Black Pop. Sq.

2.556* (0.285)

7.316* (1.009) –0.396* (0.108) –9.936* (4.798) 2.537 (2.347)

–2.280* (0.454) 0.178* (0.049) 15.699* (2.104) –5.587* (1.039)

Non-Hisp. Pop. Non-Hisp. Pop. Sq. Hisp. Pop Hisp. Pop. Sq. Outside Stations Targeting this Group R-sq N

–0.338* (0.031) 0.6902 244

–0.409* (0.048) 0.8069 99

–0.119 (0.139) 0.4385 99

Black

NonHispanic

Hispanic

Markets with Population < 2·5 million 14.490* (0.944)

1.881* (0.392)

5.501* (0.702) –0.281* (0.062) –1.648 (3.331) –0.652 (0.827)

–1.297* (0.353) –0.018 (0.030) 14.571* (1.645) –2.597* (0.406)

–0.352* (0.057) 0.8382 51

–0.018 (0.167) 0.7289 51

11.750* (0.496) 21.519* (1.515) –7.255* (0.767)

–0.365* (0.025) 0.7121 231

Notes: Standard errors in parentheses. Asterisk indicates 95% signiﬁcance level. Double asterisk indicates 90% signiﬁcance level.

10.746* (0.893)

1.429* (0.403)

19.427* (2.786) –7.024* (1.323) –2.490 (13.097) –5.147 (26.008)

–5.442* (1.314) 1.285* (0.628) 47.000* (5.637) –62.311* (11.442)

–0.371* (0.043) 0.7668 87

–0.035 (0.121) 0.5924 87

12.475* (1.154)

0.820* (0.402)

14.162* (4.312) –4.308* (2.089) 11.503 (14.848) –25.945 (24.982)

–6.024* (1.638) 1.134 (0.787) 53.143* (5.854) – 64.660* (9.507) –0.243 (0.136) 0.8203 41

–0.390* (0.054) 0.7304 41

PETER SIEGELMAN AND JOEL WALDFOGEL

Non-Black Pop. Sq.

14.562* (0.643)

NonBlack

All Listening

Non-Black Listening

Black Listening

Non-Hisp· Listening

Hispanic Listening

All Listening

Dep· Var· = AQH*100 Constant All Stations Non-BlackTargeted Stations BlackTargeted Stations Non-HispanicTargeted Stations HispanicTargeted Stations R-sq N

13.100* (0.425) 0.109* (0.017)

0.1484 244

11.871* (0.805)

11.222* (1.683)

0.142* (0.025) 0.088 (0.103)

0.2250 99

Black Listening

Non-Hisp· Listening

Hispanic Listening

Dep· Var· = ln(AQH/(1-AQH)) –1.977* (0.062)

–1.992* (0.119)

0.160* (0.051)

0.011* (0.002)

0.011* (0.004)

0.747* (0.214)

0.007 (0.008)

0.051* (0.015)

0.1928 99

12.817* (0.956)

Non-Black Listening

Race and Radio

Table 6. 1997 Group Listening and Targeted Stations (IV).

15.031* (1.502)

–1.878* (0.032) 0.008* (0.001)

–1.900* (0.072)

–1.736* (0.106)

0.122* (0.035)

0.063 (0.055)

0.009* (0.003)

0.005 (0.004)

–0.006 (0.061)

0.197* (0.096)

–0.002 (0.005)

0.014* (0.007)

0.3128 51

0.1137 51

0.3060 51

0.1138 51

0.1453 244

0.2197 99

0.1793 99

Notes: Standard errors in parentheses. Asterisks indicate 95% signiﬁcance level. Instruments include relevant populations and their square. We obtain similar results with other speciﬁcations. See text for details.

89

90

PETER SIEGELMAN AND JOEL WALDFOGEL

programming, it is impossible to be certain that minorities are being underserved in an absolute sense. But it is clear that minorities are being underserved relative to whites. All listeners face the problem of inefﬁcient underprovision, but minority populations generate smaller preference externalities for their distinct group of listeners. Consequently, less-numerous black and Hispanic populations do less to mitigate the generic inefﬁcient underprovision problem. We note, however, that despite the relative paucity of black-targeted programming, blacks listen to radio more than whites (recall Table 2). Given this fact, it is by no means obvious that underprovision of service to blacks is severe. On the other hand, one can interpret greater black listening, even when facing fewer group-targeted varieties of programming, as evidence that blacks place greater value on radio than whites, in which case their greater listening might nevertheless be consistent with underprovision.

V. DOES MINORITY OWNERSHIP PROMOTE MINORITY PROGRAMMING Our evidence suggesting that minority listeners face more severe underprovision problems than whites raises the possibility that policy can affect the situation. The FCC has pursued various policies designed to promote minority ownership of radio stations (see Spitzer, 1991 and Appendix A). Ultimately, we would like to know whether there is an economic rationale for these policies. A threshold question is simply whether minority ownership preferences actually promote minority programming. Efﬁcacy is a necessary, but not sufﬁcient, condition for justifying such programs. In a 1987 cross-section of radio stations, Dubin and Spitzer (1993) ﬁnd that minority-owned stations are more likely to broadcast minority-targeted programming than are white-owned stations. In this section we build on Dubin and Spitzer’s ﬁndings in four signiﬁcant ways. First, we use more recent data, for both 1993 and 1997. Second, in addition to cross-sectional variation, we are also able to make use of time-series variation, examining the relationship between changes in minority programming and changes in minority ownership. Third, we have not only changes in minority ownership between 1993 and 1997 but also a policy shift generating a plausible source of exogenous variation in minority ownership. This is important in overcoming the possible endogeneity problems in a simple cross-sectional regression of format on ownership. Fourth, and most important, we use market-level, rather than station-level data, allowing us to measure the impact of minority-owned stations on minoritytargeted programming, net of any crowding out.

Race and Radio

91

1. Exogenous Policy Changes, 1993–1997 There were at least two important changes in the regulatory regime governing minority broadcasters between 1993 and 1997. First, in January of 1995 Congress repealed the FCC’s Tax Certiﬁcate policy, which granted favorable treatment of capital gains on sales of broadcast licenses to minority owners.23 This had provided a substantial tax subsidy for license holders who sold their licenses to minorities. A second important change came with the passage of the 1996 Telecommunications Act, under which the rules limiting station ownership were substantially modiﬁed. As of 1992, FCC rules speciﬁed that an individual or entity could not own more than 18 AM and 18 FM stations nationally (with up to 3 more if they were controlled by minorities or small businesses). The limits were raise to 20 of each type of station in Sept. 1994. The Telecommunications Act of 1996 completely eliminated national limits, and relaxed local limits as summarized in Table 7. Relaxed ownership limits may have raised the value of stations to incumbent owners who are able to hold multiple stations. To the extent that non-minority owners are probably more able than black owners to hold multiple stations, we would expect that relaxing ownership limits would prompt black owners to sell stations to whites. These policy shifts unleashed a torrent of radio station merger activity. The FCC approved transfers of almost 4,000 stations in 1996.24 Between 1993 and 1997 the market level HHI’s for the 244 markets with valid data in both years

Table 7. Ownership Restrictions Under the 1996 Telecommunications Act. Market Size (Number of Stations)

Maximum Number of Stations That Can be Owned by a Single Entity

Maximum Number in a Single Service (AM or FM)

45 or more 30–44 15–29 14 or Fewer

8 7 6 Min(5, N/2) where N is total stations in the market

5 4 4 3

Source: 47 C. F. R. §73.3555H(a)(1)(i)-(iv) (1998). Note: The Telecommunications Act abolished national caps, which had previously been set at a maximum of 20 stations in each service. Local limits had been set at 4 stations in a single market.

92

PETER SIEGELMAN AND JOEL WALDFOGEL

nearly doubled, from an average of 1270 in 1993 to an average of 2092 in 1997, as Table 2 documents.25 Many in the minority broadcasting community watched with concern as minority station owners sold their stations to white owners of large numbers of stations (Irving et al., 1998). We view the 1996 Telecommunications Act and the elimination of the favorable capital gains treatment as exogenous increases in the demand for minority-owned stations by white owners because they occurred for reasons unrelated to the underlying demand for minority programming. The drop in minority ownership that followed on these measures is therefore plausibly exogenous, since it did not depend on changes in either minority listening or advertising revenues from minority-targeted programming. This in turn means that changes in minority-targeted programming over this period are a measure of the efﬁcacy of minority ownership preferences: if minority programming changes as the result of an exogenous shock to minority ownership, we can be conﬁdent that the causation runs from ownership to programming, and hence that preferences which increase ownership thereby promote minority-targeted programming. Moreover, since ownership restrictions under the Telecommunications Act were relaxed differentially according to market size, we can use market size measures as instruments for changes in station ownership by race. In this section we examine the relationship between minority ownership and programming using three separate approaches. First, we present data on the distribution of stations by programming format and owner race. We then present cross section evidence for 1993 and 1997 and longitudinal evidence, with and without instrumenting for changes in ownership. 2. Who Broadcasts Minority-Targeted Programming? Table 8 shows the distribution of stations by format and owner race for 1997. A surprising fact emerging from this table is that, while almost all minorityowned stations broadcast minority-targeted content, most stations broadcasting minority-targeted programming are actually white-owned. Of 139 black-owned city-stations in 1997, all but 23 (16%) were in the six black-targeted formats. Of these 23 stations, moreover, 8 were in formats that attract substantial numbers of black listeners – Jazz and Contemporary Hit Radio/Urban – meaning that nearly 90% of black-owned stations broadcast to a substantially black target audience. Yet, most black-targeted stations are white-owned. For example, whites own 169 (72%) of the 236 stations broadcasting in the Black format. Whites own 90 of 107 stations broadcasting in the Black/Adult Contemporary format, 22 of 32

Race and Radio

93

Table 8. Station Ownership by Format and Owner Race/Ethnicity. Format Adult Contemporary (AC) AC/Album Oriented Rock AC/Contemp. Hit Radio AC/New Rock AC/Soft Adult Contemporary Album Oriented Rock (AOR) AOR/Adult Contemporary AOR/Classical AOR/New Rock AOR/Progressive Black Black/Adult Contemporary Black/Gospel Black/Oldies Black/Talk Big Band Big Band/Easy Listening Big Band/Religious Big Band/Talk Country Country/Full Service Contemporary Hit Radio (CHR) CHR/Adult Contemporary CHR/Black CHR/New Rock CHR/Urban Classical Classic Album Oriented Rock Classic Hits Ethnic Easy Listening Easy ListeningSoft Adult Contemporary Full Service Full Service/Talk Gospel Jazz Kids News News/Talk Oldies Religious Soft Adult Contemporary Spanish Sports Talk Talk/Classic AOR Talk/Full Service Talk/Jazz Unknown

White

Black

Hispanic

418 4 128 42 8 374 3 19 198 80 169 90 22 20 1 315 3 1 1 904 14 409 51 2 9 49 70 265 83 12 13 1 96 78 54 115 1 75 170 393 266 253 223 169 407 7 3 5 6

0 0 0 0 0 1 0 0 2 0 67 17 10 6 5 0 0 0 0 0 0 0 0 0 0 5 0 0 0 2 0 0 0 0 8 3 0 0 0 1 9 0 2 1 0 0 0 0 0

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 71 2 1 0 0 0 0

94

PETER SIEGELMAN AND JOEL WALDFOGEL

Black/Gospel stations, 20 of 26 Black/Oldies stations, 54 of 62 Gospel stations, and 12 of 14 Ethnic stations. (An interesting exception with potential signiﬁcance for an analysis of viewpoint diversity is that whites own only 1 of 6 Black/Talk stations.) These results clearly indicate that black ownership is not necessary for the provision of black programming. A question that they do not answer is whether black ownership increases the number of stations broadcasting black-targeted programming. Given that white owners frequently provide black-targeted programming, it is entirely possible that additional black-owned and targeted stations simply reduce the number of black-targeted stations provided by white owners, a “crowding out” effect. We turn to this question next. 3. Cross Section Evidence on Owner Race and Targeted Programming To measure the impact of black and Hispanic ownership on the volume of programming targeted at these two groups, Table 9 presents regressions of the number of group-targeted stations in a market on the number of group-owned stations in the market, ﬁrst without, then with additional controls. Controls include group population and its square, as well as the total number of stations and the total number of formats. Using either the 1993 or 1997 cross section, the coefﬁcients on the number of minority-owned stations tend to be quite large (between 0.69 and 1.31 for blacks, between 1.12 and 2.56 for Hispanics). These results imply that each additional minority-owned station begets roughly one additional net source of minority-targeted programming, suggesting that minority-owned stations do not simply replace white-owned, minority targeted stations. Results are virtually identical when we use inside stations as the dependent variable and treat outside stations as an additional explanatory variable. While interesting, these estimates are vulnerable to a concern that the positive estimated coefﬁcient may arise because both the number of minorityowned stations and the number of minority-oriented formats depend on some third unobserved factor. To address this concern, we make use of the panel feature of the data. 4. Longitudinal Evidence Table 10 reports regressions of the change in the number of black- and Hispanic-targeted stations on the change in the number black- and Hispanicowned stations. The OLS coefﬁcient estimate is 0.248 for blacks (with a standard error of 0.092). The OLS Hispanic coefﬁcient is 0.784 (0.165). These

Race and Radio

Table 9. Cross Sectional Evidence on Minority Ownership and Programming. Black-Targeted Stations 1993 Constant Number of BlackOwned Stations Number Hisp.-Owned

0.770* (0.106) 1.152* (0.083)

Number of Stations

1997 0.225 (0.317) 0.685* (0.101)

1.215* (0.136) 1.312* (0.119)

–0.018 (0.019) 0.073 (0.047) 6.063* (1.061) –2.058* (0.478)

Number of Formats Black Pop. 1993 Black Pop. Sq. 1993

Hispanic-Targeted Stations 1993 0.709 (0.457) 0.900* (0.130)

0.278* (0.099)

0.622* (0.257)

0.367* (0.124)

0.730* (0.369)

2.136* (0.163)

1.144* (0.150) 0.049* (0.016) –0.144* (0.038)

2.537* (0.166)

1.518* (0.167) 0.092* (0.022) –0.191* (0.044) 9.823* (1.117) –2.500* (0.380) 9.520 (1.115) –2.440 (0.038) 0.6753 244

0.038 (0.026) –0.046 (0.056) 7.383* (1.329) –3.032* (0.628)

Hisp. Pop. 1993 Hisp. Pop. Sq. 1993 R-sq. N

0.4406 244

0.5393 244

0.3358 244

0.4340 244

1997

0.4136 244

8.975* (0.891) –2.246* (0.299) 0.6834 244

0.4908 244

Note: Standard errors in parentheses. Asterisk indicates 95% signiﬁcance level.

95

96

PETER SIEGELMAN AND JOEL WALDFOGEL

Table 10. Longitudinal Evidence on Minority Ownership and Programming.

Constant ⌬Black-Owned Stations ⌬HispanicOwned Stations 1993 Population 1993 Population Squared R-sq N

⌬BlackTargeted Stations OLS

⌬BlackOwned Stations OLS

⌬BlackTargeted Stations IV

⌬HispanicTargeted Stations OLS

⌬HispanicOwned Stations OLS

⌬HispanicTargeted Stations IV

0.437* (0.070) 0.249* (0.093)

–0.003 (0.060)

0.515* (0.091) 0.987* (0.442)

0.322* (0.065)

0.051 (0.031)

0.291* (0.089)

0.787* (0.165)

0.0288 244

–0.163* (0.077) 0.003 (0.007) 0.0556 244

0.0252 244

0.0864 244

1.136** (0.688) 0.050 (0.040) 0.002 (0.004) 0.0583 244

0.0105 244

Notes: Standard errors in parentheses. Asterisk and double asterisks indicate signiﬁcance at the 95 and 90% levels, respectively. Instruments for IV regressions include 1993 population and its square. First-stage regressions are reported in columns 2 and 5.

results suggest substantial but not complete crowding out: they imply that the net effect of an additional minority owned station is to increase minority targeted formats by only one-quarter to three-quarters of a station. Although running regressions in changes eliminates the problem of ﬁxed unobservable factors affecting both ownership and format, other potential problems remain. First, changes in a group’s ownership may be endogenous. Second, changes in ownership may be measured with error. One important source of measurement error is the fact that minority ownership is self-reported on a survey sent to all radio stations, not all of which are completed or are legible.26 The structure of changes in regulatory ownership limits described above suggests that market size measures related to the number of stations can serve as instruments for the change in minority ownership.27 Fortunately, instrumental variables (IV) addresses measurement error as well as endogeneity concerns. Columns 2 and 5 of Table 10 show the associated ﬁrst-stage regressions. While the instrument does not work especially well for either group, population has higher signiﬁcance for blacks than for Hispanics. The estimated IV coefﬁcients on the change in group ownership are 0.987 (0.442)

Race and Radio

97

for blacks and 1.136 (0.688) for Hispanics. In other words, after controlling for the possible endogeneity of changes in minority ownership, adding a minority owned station increases the number of minority-format stations by roughly 1.0, implying no crowding-out of white-owned, minority-targeted stations. We obtain virtually identical results when we disaggregate inside and outside stations, treating the change in inside stations as the dependent variable and the change in outside stations as an additional exogenous variable.

VII. EXPLANATIONS The absence of crowding-out documented above implies that, even though white owners commonly provide black-targeted programming, black owners enter in situations that white owners avoid. Otherwise, exogenously adding new black-owned stations to a market should have no effect on the total number of black-targeted stations. Two possible explanations come to mind, both discussed in Spitzer (1991). First, black owners may have informational advantages that allow them to proﬁtably enter markets that whites cannot. This seems highly implausible, given that the majority of black-targeted stations are white-owned. If true, however, this hypothesis would imply that black ownership would have an effect on listening, over and above the effect of black programming. We can examine this hypothesis by comparing listening to black- and white-owned black-targeted stations located in the same markets. Under a strong version of the informational advantage hypothesis, blackowned black-targeted stations should attract more listeners per station than white-owned black-targeted stations. But this is not true. In markets with both white and black-owned black-targeted stations, the average black-owned blacktargeted station had an average of 4,970 listeners while the average white-owned black-targeted station had 6,840, nearly 38% more. This difference also arises – and is signiﬁcant – in a regression that includes format and market ﬁxed effects.28 We can do a second test for black informational advantage by regressing black AQH listening for a metro area on the total number of stations, the number of black-targeted stations, and the number of black-owned stations (virtually all of which are black-targeted). If black owners have an advantage over white owners, we would expect to ﬁnd that black listening is a positive function of the number of black-owned stations, over and above any effect arising from the number of black-targeted stations. When we run this test, however, the coefﬁcient on the number of black-owned stations is small and insigniﬁcant. The evidence does not support the informational advantage hypothesis for black owners.

98

PETER SIEGELMAN AND JOEL WALDFOGEL

For Hispanics the evidence is slightly different. In markets with both Hispanic and non-Hispanic-owned Hispanic-targeted stations, Hispanic-owned stations have an average of 5,850 listeners while non-Hispanic-owned stations have an average of 6,030, only 3% more. In a regression of Hispanic listeners on market dummies and a Hispanic ownership dummy (there is only one Hispanic format, so there are no format dummies), the coefﬁcient on Hispanic ownership is 1,790. This indicates that for Hispanics, there is a positive ownerrace effect on listening (controlling for other factors), although the effect is not statistically signiﬁcant (t-stat = 1.12). In regressions of Hispanic AQH listening on the total number of number of stations, the number of Hispanic-targeted stations, and the number of Hispanic-owned stations, the coefﬁcient on the number of Hispanic-owned stations is small and never signiﬁcant. On balance, then, there is no convincing evidence of an informational advantage for black or Hispanic owners. This seems economically plausible, since as Spitzer notes, even if minority programmers have special insights into the minority market, there is little reason to think that white owners could not simply buy this expertise, obviating the need for minority ownership. A second possibility is that black owners enter for “ideological” reasons, which means that they are willing to forego some proﬁts in order to a provide a particular sort of programming.29 This hypothesis would rationalize the observation that black-owned and targeted stations have fewer listeners, on average, that their white-owned counterparts (in markets with both white and black-owned black-targeted stations). Black owners’ willingness to accept smaller returns could explain why greater black ownership increases blacktargeted programming: additional black owners are willing to enter low-proﬁtability market niches (programming to small black audiences) that whites would not enter. The “ideological” theory predicts lower returns for black-owned stations. Unfortunately, we lack proﬁt data that would support an adequate test of this prediction.

VIII. CONCLUSION This paper has demonstrated three important facts about the market for commercial radio. First, we document the existence of preference externalities: individuals are better off when they are located in markets with others who share their preferences in radio programming. Second, we ﬁnd that these externalities operate only within-group, which should not be surprising, given the disparate tastes of whites, blacks, and Hispanics. Our third empirical ﬁnding is that minority ownership increases the net amount of minoritytargeted programming. Even though most minority-targeted stations are

Race and Radio

99

white-owned, markets with more minority-owned stations also have more minority-targeted stations, which means that minority-owned stations add to the total programming available to minority listeners. While these facts are clear, their normative implications are not as obvious. Given that preferences externalities can overcome inefﬁcient underprovision for large audiences, the small size of local black and Hispanic audiences in most markets means that the preference externality mechanism is not available to correct this problem. An efﬁciency-minded regulator might thus want to consider other ways of promoting minority-targeted programming.30 In this vein, policies promoting minority ownership seem like a plausible method for increasing welfare by correcting inefﬁcient underprovision. An important caveat is in order, however. Although we can make a theoretical case for the possibility that markets will underprovide programming of interest to small audiences, we cannot isolate particular instances of inefﬁcient underprovision. Because listeners get radio programming without paying, we cannot determine its value to listeners, and hence we cannot ascertain whether the market fails to provide programming whose social value is greater than its cost. Given that it is impossible to measure listener valuation directly, there are two indirect methods that might be used to shed some light on the efﬁciency questions involved. One alternative is to deduce the listener valuation implicit in current regulatory policies. The argument works like this. While the marginal entrant may attract many listeners, in a market with many stations, she attracts a small number of new listeners to radio, even though her station has substantial operating costs. In general, it must be the case that the net increase in total (market) ad revenue associated with the marginal station falls short of its cost. For example, suppose the marginal station costs $1 million to operate and generates $1.2 million in advertising revenues. Suppose that only $0.3 million of this $1.2 million constitutes a net addition to total market advertising revenues; the remaining $0.9 million represents revenues (listeners) diverted from pre-existing stations. If the marginal station costs $1 million and generates only $0.3 million in net new revenue, then in order for its entry to be optimal there must be at least $0.7 million in beneﬁts going to someone else – in this case, listeners. Berry and Waldfogel (1999b) calculate empirically that in order for the existing pattern of entry to be optimal, regulators must believe that the value of the marginal station to listeners is roughly three times its value to advertisers.31 The larger the listener valuation of programming, the greater the problem of inefﬁcient underprovision, since listener valuation is the portion of social beneﬁt that program providers cannot capture as revenue. The Berry and

100

PETER SIEGELMAN AND JOEL WALDFOGEL

Waldfogel result thus suggests that if we assume current entry patterns are optimal, inefﬁcient underprovision is likely to be a serious problem, since implicit listener valuations are large relative advertising revenues. A second normative benchmark comes from other areas of radio regulation. While the unpriced nature of radio makes it difﬁcult, if not impossible, to identify inefﬁcient underprovision, U.S. broadcasting policy does directly subsidize programming in some formats, namely classical music, jazz, and news. Other research (Berry & Waldfogel, 1999a) suggests that roughly a third of government support for “public” classical stations supports stations that either have local commercial competition in the format or would, in the absence of the public station. Efﬁciency – beneﬁts in excess of costs – is an ideal benchmark against which to judge the market for minority-targeted programming. But given that existing policy already subsidizes some formats that would otherwise attract relatively few listeners, the economic case for intervention to increase the amount minority-targeted programming is strengthened.32 Much work remains. First, it would be useful to characterize the sort of minority-targeted programming available in more detail. Existing format designations are very coarse. Most black-targeted stations are uninformatively labeled simply “Black.” Are these talk stations, urban top 40 stations, religious stations? The Hispanic-targeted designations are even less helpful. Virtually all are simply “Hispanic.” It would be useful, therefore, simply to characterize the availability of various sorts of minority-targeted stations. Some markets have multiple black-targeted stations. With ﬁner format information, it would be possible to know whether these stations duplicate one another, as opposed to offering substantially distinct programming. Second, the forces operating in the provision of radio broadcasting to blacks and Hispanics also operate, to varying extents, in the provision of local television, as well as other media (such as newspapers). Relatively few markets have black or Hispanic news outlets. If minority preferences in news are different, it is possible that the current local TV news conﬁguration – three virtually identical news shows – is inferior to, say, two virtually identical whiteoriented shows along with, say, one minority oriented program. In the absence of viewership data, this is no more than speculation. It would be interesting, however, to study viewing data, by race and Hispanic status, by market. Do minorities watch in greater proportions in markets where they are more numerous (and where, one can assume, there is programming closer to their tastes and interests)? Third, to what extent can emerging technologies correct the possible inefﬁcient underprovision that we describe? The FCC has licensed two

Race and Radio

101

companies to broadcast satellite radio in the U.S.. Do these companies plan to include minority-targeted programming? Because they are charging for their service, they avoid the theoretical problem of underprovision because they can capture listener valuation as revenue. However, will their pricing make the service appealing for minority listeners to adopt? Should satellite providers with capacity for, say, 100 channels be encouraged to “sublet” some of their channels to minority providers? Emerging technologies that effectively increase the size of the relevant market may make it possible to harness preference externalities to solve this problem through the market. Such solutions may require regulatory foresight.

NOTES 1. See Head (1985). With the widespread adoption of cable, television signals are no longer textbook examples of a public good, and technological change threatens radio’s status as well. In October 1997, the Federal Communications Commission (FCC) granted the CD Radio company a license to broadcast 100 channels of commercial-free radio to listeners in cars. The company expects to charge $10 per month. (See www.cdradio.com). 2. The problem of potentially inefﬁcient market underprovision of classical music, jazz, and news in the U.S. is discussed in Berry and Waldfogel (1999a). A second potential problem is inefﬁcient overprovision. In a sufﬁciently large market, entrants may divert listeners from incumbent stations, causing the private beneﬁt of entry to exceed the social beneﬁt. This leads to excess entry. From a social planner’s perspective, excess entry can be a problem because excessive resources are devoted to station operation. But excessive entry does not pose a problem for listeners. Our approach in this paper is to examine the adequacy of programming for minority listeners. Hence we are concerned only with correction of potential underprovision, not with possible overprovision. Berry and Waldfogel (1999b) measure the social inefﬁciency of free entry into radio broadcasting, viewed from the standpoint of the market participants (buyers and sellers of advertising). 3. One obvious source of distinct programming preferences is language. Another way to identify small groups with distinct preferences is by format: Classical music and jazz, both subsidized in the U.S., come to mind as possible candidates (see Berry & Waldfogel, 1999a). One might also examine the adequacy of provision for listeners by age or gender. 4. A summary of the FCC’s main racial preference policies is provided in Appendix A. Spitzer (1991) suggests that minority ownership preferences are rational responses to inadequate provision of minority-targeted programming, and hence do not violate the 14th Amendment’s Equal Protection clause. (Note, however, that the terms of the debate have changed considerably since 1991.) Recent reductions in minority ownership have prompted the FCC to review the impact of changes in its ownership rules on diversity. See, for example, the statement of FCC Commissioner Ness (1998). 5. See Samuelson (1954) for a discussion of efﬁcient public goods provision.

102

PETER SIEGELMAN AND JOEL WALDFOGEL

6. Inefﬁcient underprovision will occur whenever ad revenue < cost of provision < total social beneﬁt (ad revenue and listener value). The ﬁrst condition guarantees that the station will not be provided; the second that it should optimally be provided. This is a standard problem in differentiated products. Unless perfect price discrimination is possible, some goods with total beneﬁt in excess of costs will not be provided (see Spence, 1976a, b). 7. This is their willingness to pay to listen to the station in the presence of commercials. This example describes the provision of the ﬁrst station in the market. Entry in the presence of existing stations, while it makes the analysis more complicated because we must now consider net revenue, does not alter the basic intuition. This example also assumes free entry, so that a station should enter if its social beneﬁt exceeds its operating cost. If entry is not free, a station’s social beneﬁt would need to exceed the social beneﬁt of the marginal incumbent for its entry to be efﬁcient. 8. Ofori (1998) presents evidence that minority-targeted stations attract fewer advertising dollars per listener than white-targeted stations. 9. A controversial memo written in 1998 by one advertising consultant encouraged advertisers “to minimize or eliminate advertising with black- and Hispanic-targeted radio stations, saying, ‘When it comes to delivering prospects, not suspects, the urbans [Hispanic and especially black radio stations] deliver the largest amount of listeners who turn out to be the least likely to purchase.’ Buying advertising on ethnic stations would mean ‘losing the more important white segment of the population,’ the memo said.” (Billboard, 1998). 10. See Berry and Waldfogel (1999a). 11. We observe 1993 black listening for 75 metropolitan areas and Hispanic listening for 31 markets. We observe 1997 black listening in 99 markets and Hispanic listening in 51 markets. 12. See, NTIA (1993–1997). The owner race data were compiled from self-reports to a survey mailed by the NTIA to all radio stations. The data may fail to include some black-owned stations, especially in the earlier years. 13. Black-targeted formats include “Black,” “Black/Gospel,” “Black/Adult Contemporary,” “Black/Oldies,” “Black/Talk,” “Gospel,” and “Ethnic.” We classify only “Spanish” as Hispanic-targeted. 14. Non-black includes all persons who are not black. We use this category because Arbitron reports only total and black listening. Non-black listening is the difference between total and black listening. 15. The second panel of Table 3 again uses OLS, but examines an alternative speciﬁcation for the dependent variable, using the log odds of AQH listening – that is, ln(share listening/share not listening). This speciﬁcation arises more naturally from the behavior of utility-maximizing listeners (see Berry & Waldfogel, 1996), but is more difﬁcult to interpret by eyeball. Since the results are qualitatively identical to those in the ﬁrst speciﬁcation, we concentrate on the former. 16. One might weight these regressions by the number of Arbitron diaries in each metro area’s sample. While we do not have the numbers of Arbitron diaries, the samples are roughly proportional to metro area population. We veriﬁed that weighting by population does not change any of the substantive results in Tables 3–6. 17. Black and white viewers also have substantially different preferences in television shows. Between September 21 and November 29, 1998, the top 5 network

Race and Radio

103

television shows among whites ranked 118th, 124th, 7th, 118th, and 10th, respectively, among black viewers. See Sterngold (1998). 18. Let sij be the share of all listeners of type i (i = b,w) listening to format j in a given market. Then the Duncan index for that market is D = 100 *⌺j | sbj ⫺ swj | /2. By comparison, the Duncan index for residential segregation in the 15 Northern cities with the largest black populations in 1980 was 80.1 (Massey and Denton, 1993, p. 64). Alternatively, let ␦ = Min(sbj, swj)/Max(sbj, swj), where sbj is the percentage of all black listeners listening to format j and similarly for whites. There are only two formats, News and Soft Adult Contemporary, for which ␦ is greater than 0.5. 19. The measure of stations in table 4 includes all group-targeted stations received in the metro area, both those broadcasting from inside and outside the metro area. When we include the inside and outside stations of each relevant group separately, we obtain sensible patterns. Inside stations have larger coefﬁcients. Results in Table 4 are also robust to the inclusion of region dummies. 20. We treat outside stations as an exogenous explanatory variable here. The rationale for doing so is that outside entry occurs for reasons unrelated to the local market. This argument is more valid, the more local is radio advertising. According to Duncan (1994), roughly three quarters of radio ads are local, although this fraction varies by market. 21. We performed these estimates in a variety of ways: (1) disaggregating inside and outside stations and treating the outside stations as exogenous, and (2) excluding markets with 2.5 million or more people. All results are substantively similar to those reported, which include the full sample and do not distinguish between inside and outside stations. 22. Note that listeners could still value variety even if we observed no effect of the number of stations on listening. Even if total listening is invariant with station entry, listeners are at least weakly better off with entry, as they get weakly more preferred choices. 23. See 26 U.S.C §1071 (repealed in 1995). 24. See Ness (1997). 25. See Berry and Waldfogel (2001) for an analysis of the effect of increasing concentration on programming variety following the increases in concentration after the 1996 Telecommunications Act. 26. The Commerce Department admits that it misclassiﬁed as many as 20 minority owned stations in 1996. 27. The ownership limits established by the Telecommunications Act vary with the number of stations in a local market. But we cannot use “the number of stations” to measure market size because the relevant number of stations for regulatory purposes is an engineering concept that depends on the signal contours of the stations involved (Aronowitz, 1998). This means that there are many possible measures of “the number of stations in the market.” For tractability, we use population as a simple measure of market size. 28. While inconsistent with a strong informational advantage for black owners, these results are consistent with a black advantage in reaching small market segments otherwise unserved by white owners. 29. Morton and Podolny (1998) examine similar issues in the California wine industry.

104

PETER SIEGELMAN AND JOEL WALDFOGEL

30. One solution to the underprovision problem, in principle, is to increase audience sizes. If all blacks lived in Atlanta, for example, then they would enjoy preference externalities large enough to undo some inefﬁcient underprovision. While communication policy cannot, of course, induce mass-migration to create large local audiences, emerging technologies (satellite radio, Internet radio) can accomplish the same thing by making the entire country (or world) into a single market. 31. We stress that this valuation does not come from listeners themselves. Rather, it comes from policy makers: it is the valuation that renders regulatory decisions optimal. If the true listener valuation is smaller, free entry may result in an excessive number of stations. 32. Of course, public broadcasting is controlled by an entirely different regulatory institution, with different goals and constituencies from the FCC. And there may also be important legal differences between race-neutral subsidization of classical music and race-conscious efforts to increase the amount of minority-targeted programming. Nevertheless, the comparison is still illuminating, at least at the theoretical level. 33. The exact mechanism by which the FCC granted licenses, and the role minority preferences played in it, is extremely complicated. For a description of the rules before 1993, see, e.g. Miracle Strip Communications, 4 FCC Rcd. 5064 (1989). In essence, the FCC ranked applicants for licenses according to two public interest objectives: “Best Practicable Service to the Public,” and “Diversiﬁcation of Control of the Mass Media.” Diversiﬁcation meant whether or not the applicant had any other media holdings. For the period relevant to our data (after Miracle Strip was decided in 1989), race was not a factor in the diversiﬁcation analysis. Racial preferences were taken into account in deciding which applicant offered the Best Practicable Service, as follows. First, applicants were ranked on the basis of the Intensity of Owner Participation or “Integration” of management and ownership (I), where I = (100 ⫻ (hours per week/ 40))2 ⫻ (% ownership interest). For example, if a half-owner participates half time, I = (100 ⫻ 20/40)2 ⫻ 0.5 = 1250. The owner’s racial background was then considered, along with other factors such as local residence, previous broadcast experience, etc., as a “plus factor” that could make up for a lower Intensity of Participation score. It is unclear how much weight the combined other factors have: but there is precedent for the qualitative factors being unable to overcome a difference of 12.5% in the quantitative measures. See New Continental Broadcasting Co., 88 FCC 2d 830, 850 para. 35 (Rev. Bd. 1983) (holding that a “clear” quantitative differential of 12.5% (1250 using Index) cannot be overcome by the qualitative attributes of the competing applicant’s integrated owners). A divided Supreme Court upheld the constitutionality of the “plus factor” and the distress sale provisions (see below) in Metro Broadcasting v. FCC, 497 U.S. 547 (1990). However, the Commission’s entire integration policy was invalidated as arbitrary and capricious in Bechtel v. FCC, 10 F.3d 875 (D.C. Cir. 1993), so credit for minority ownership could not be linked to integration after 1993. Contested applications for licenses were frozen between 1994 and 1998, when the FCC adopted a competitive bidding process for contested applications, with a special bidding credit for ‘new entrants’ that appears to be speciﬁcally designed to increase minority participation. See, In the Matter of Implementation of Section 309(j) of the Communications Act,Competitive Bidding for Commercial Broadcast . . . Licenses, 1998 FCC LEXIS 4290 at 1. 34. Apparently this was used largely for allocating low-power TV licenses and was not a factor in the radio market.

Race and Radio

105

35. Codiﬁed in 26 U.S.C §1071. The tax certiﬁcate policy was used in “281 sales of AM, FM and TV stations” between 1978 and its elimination in 1995. (See Notice of Proposed Rule Making, 10 FCC Rcd 2788, 2789 (1995). 36. Statement of Policy on Minority Ownership of Broadcasting Facilities, 68 FCC 2d 979 (1978). The distress sale provision was used 42 times between 1978 and 1995. See, Notice of Proposed Rulemaking, 10 FCC Rcd. 2788, 2789 (1995). 37. The limits were raised to 20 AM and 20 FM stations in 1994, then essentially eliminated by the 1996 Telecommunications Act. See 47 C.F.R. §73.3555H(a)(1)(i)-(iv) (1998). 38. See, Statement of Policy on Minority Ownership of Broadcasting Facilities, 68 FCC 2d 979 (1978). ComTrain is a management training program run by the NTIA’s MTDP (Minority Telecommunications Development Program). Other policies not on their face designed to assist minority broadcasters may nevertheless disproportionately beneﬁt minority applicants for station licenses. For example, the Telecommunications Development Fund was established under the 1996 Telecommunications Act to provide capital to small businesses. 39. See 47 C.F.R. §73.2080. These requirements were overturned by the D.C. Circuit in Lutheran Church-Missouri Synod v. FCC, 1998 U.S. App. LEXIS 7387 (D. C. Cir.).

ACKNOWLEDGMENTS We thank Yun-Sug Baik and Yu Li for able research assistance, and Shelly Cagner of Arbitron for providing access to their data. Siegelman’s work on this paper was completed while he was visiting at the University of Connecticut Law School.

REFERENCES Anderson, S. P., & Coate, S. (2000). Market Provision of Public Goods: The Case of Broadcasting. NBER Working Paper No. 7513, January. Aronowitz, A. (1998). Email communication. February 24. The Arbitron Company (1994). Radio Metro Market Guide, 1993–1994. The Arbitron Company, New York. The Arbitron Company (1993, 1997). Radio USA, Spring 1993, Spring 1997. New York: The Arbitron Company. Beebe, J. H. (1977). Institutional Structure and Program Choices in Television Markets. Quarterly Journal of Economics, 91, 15–37. Berry, S. T., & Waldfogel, J. (1999a). Public Radio in the United States: Does it Correct Market Failure or Cannibalize Commercial Stations? Journal of Public Economics, 71, 189–211. Berry, S. T., & Waldfogel, J. (1999b). Free Entry and Social Inefﬁciency in Radio Broadcasting. RAND Journal of Economics, 30, 397–420. Berry, S. T., & Waldfogel, J. (2001). Do Mergers Increase Product Variety? Evidence from Radio Broadcasting. Quarterly Journal of Economics, 116, 1009–1025. Billboard Magazine (1988). Programming Newsline. May 30. Dixit, A., & Stiglitz, J. E. (1977). Monopolistic Competition and Optimum Product Diversity. American Economic Review, 67, 297–308.

106

PETER SIEGELMAN AND JOEL WALDFOGEL

Dubin, J. A., & Spitzer, M. L. (1993). Testing Minority Preferences in Broadcasting. California Institute of Technology Working Paper 856. Duncan, J. H. (1993, 1997). Duncan’s American Radio. Indianapolis: Duncan’s American Radio. Duncan, J. H. (1994). Duncan’s Radio Market Guide. Indianapolis: Duncan’s American Radio. George, L., & Waldfogel, J. (2000). Who Beneﬁts Whom in Daily Newspaper Markets? NBER working paper 7944. Head, S. W. (1985). World Broadcasting Systems : A Comparative Analysis. Belmont, CA: Wadsworth. Irving, L., Brown, K., Williams, T. L., & James, L. D. (1992–1998). Minority Commercial Broadcast Ownership in the United States. Washington, D.C.: The Minority Telecommunications Development Program, National Telecommunications and Information Administration, U.S. Department of Commerce. Mankiw, N. G., & Whinston, M. D. (1986). Free Entry and Social Inefﬁciency. RAND Journal of Economics, 17, 48–58. Massey, D. S., & Denton, N. A. (1993). American Apartheid : Segregation and the Making of the Underclass. Cambridge, MA: Harvard University Press. Ness, S. (1998). Separate Statement of Commissioner Susan Ness (Re: Review of the Commission’s Broadcast Ownership Rules and Other Rules Adopted Pursuant to Section 202 of the Telecommunications Act of 1996). Washington, D.C.: FCC. Ness, S. (1997). Remarks of Commissioner Susan Ness before the Michigan Association of Broadcasters. Washington, D.C.: FCC. Ofori, K. A. (1998). When Being No. 1 Isn’t Enough: the Impact of Advertising Practices on Minority-Formatted and Minority-Owned Broadcasters. Washington, D.C.: Civil Rights Forum on Communications Policy. Rogers, R. P., & Woodbury, J. R. (1996). Market Structure, Program Diversity, and Radio Audience Size. Contemporary Economic Policy, 14, 81–91. Samuelson, P. A. (1954). The Pure Theory of Public Expenditures. Review of Economics and Statistics, 36, 387–389. Scott Morton, F. M., & Podolny, J. M. (1998). Love or Money? The Effects of Owner Motivation in the California Wine Industry. Unpublished Manuscript, Stanford University. Spence, A. M. (1976a). Product Selection, Fixed Costs, and Monpolistic Competition. Review of Economic Studies, 43, 217–236. Spence, A. M. (1976b). Product Differentiation and Welfare. American Economic Review, 66, 407–414. Spence, A. M., & Owen, B. (1977). Television Programming, Monopolistic Competition, and Welfare. Quarterly Journal of Economics, 91, 103–126. Spitzer, M. L. (1991). Justifying Minority Preferences in Broadcasting. Southern California Law Review, 64, 293–418. Steiner, P. O. (1952). Program Patterns and the Workability of Competition in Radio Broadcasting. Quarterly Journal of Economics, 66, 194–223. Sterngold, J. (1998). A Racial Divide Widens on Network TV. New York Times, (December 29), p. A1. Waldfogel, J. (1999). Preference Externalities: An Empirical Study of Who Beneﬁts Whom in Differentiated Product Markets. NBER working paper 7391.

Race and Radio

107

APPENDIX A A Summary of the FCC’s Minority Preferences Policy

How Applied

Effective Dates

Grants of Licenses by the FCC “Plus-Factor”

If > 1 applicant for single license, minority ownership considered as a plus factor.33

1973–1993

Lottery Preference

Special credit for minority bidders on stations awarded by lottery.34

1983–Current?

Sales of Licenses Between Private Parties Tax Certiﬁcates

Favorable tax treatment for capital gains realized on sale of station license to minority buyer.35

1978–Jan. 1995

“Distress Sale” Provision

Relaxed procedural requirements when owner whose qualiﬁcations to hold a license have been called into question sells to minority enterprise.36

1978–Current?

Station Ownership Limits

Allowed a single owner to take a non-controlling interest in up to 3 stations per service (AM or FM) nationwide beyond the national caps on total station ownership, if the additional stations were controlled by minorities; relaxed the national cap by 2 for minority owners.37

1978–1996

Operational Requirements or Regulations Training Programs

Special programs to train minority broadcasters.38

1978–Current

Employee Afﬁrmative Action

Reporting requirements for employment and recruitment of minorities, with possible consequences for license renewal if requirements were not met.39

1987–1998

This Page Intentionally Left Blank

THE VALUE OF ADVERTISING IN A MAGAZINE BUNDLE Craig A. Depken II and Dennis P. Wilson ABSTRACT Existing analyses of circulation industries have often postulated that consumers consider advertising to be a “bad,” thereby requiring publishers to subsidize cover price (and thus circulation revenue) with advertising revenues. This paper uses data describing 117 U.S. magazines for the years 1996–1998 to determine the value of advertising in a magazine bundle. For four of ﬁve magazine genres, advertising is found to increase average newsstand circulation and newsstand price. These results indicate that advertising is a shift parameter in the newsstand demand for magazines, as opposed to the alternative hypothesis that advertising is a bad that subsidizes cover price.

I. INTRODUCTION A circulation industry is typically deﬁned as a market in which publishers disseminate information on a regular basis to subscribers, either in a futures market, e.g. subscriptions, or a spot market, e.g. newsstand sales. The economic analysis of circulation industries has primarily focused on the publisher’s perspective. Traditionally, the tradeoff between circulation revenue, generated by sales to subscribers, and revenue from advertising has been investigated (for example, Reddaway, 1963; Blair & Romano, 1993; Chaudhri, 1998). Advertising and Differentiated Products, Volume 10, pages 109–128. Copyright © 2001 by Elsevier Science Ltd. All rights of reproduction in any form reserved. ISBN: 0-7623-0823-0

109

110

CRAIG A. DEPKEN II AND DENNIS P. WILSON

From the point of view of a publisher, the implications from bundling advertising and the information product can be profound. Bundling theory suggest that bundles may be more valuable to consumers if two complementary goods are sold together, e.g. shoes and shoelaces, but may also be a means of extracting surplus from consumers. Stigler’s (1968) seminal example of distributors bundling popular movies with unpopular movies is an example of surplus extraction. Adams and Yellen (1976) suggest the alternative. Combining bundling theory with the economics of circulation industries leads to some interesting questions. Does advertising increase or decrease the value of the bundle of advertising and editorial content? Does advertising act as a pure enhancement to the publisher’s revenue or does advertising have any (negative) feedback on the ultimate quantity of bundles sold? Public policy has frowned on bundling, especially when designed to extract surplus from consumers. Speciﬁcally, the Clayton Act of 1914 prohibits commodity bundling when its impact is anti-competitive. However, circulation industries have avoided antitrust investigation for bundling even though they have bundled advertising and editorial content over a long period. One explanation is that, while consumers are forced to purchase advertising, they are not necessarily required to consume the advertising (as claimed by Blair & Romano, 1993, and modeled generally by Grossman & Shapiro, 1984). This situation, while perhaps not entirely acceptable to advertisers, may be tolerated by consumers (and public policy enforcers) if advertising reduces the value of the bundle, in essence reducing the price that publishers can charge. This would correspond with a decrease in the (potential) real cost of editorial content. An alternative explanation is that bundling advertising with editorial content does not extract from, but rather adds to, the surplus enjoyed by consumers. This alternative is the focus of this paper. Existing analyses of circulation industries have often postulated that consumers consider advertising to be a “bad,” thereby requiring publishers to subsidize cover price (and thus circulation revenue) with advertising revenues. While this is a common conjecture, it has been questioned on intuitive grounds, although little empirical evidence is proffered to support either side of the argument. Instead of focusing on the advertising question from the point of view of the publisher/advertiser, Becker and Murphy (1993) focus on a representative consumer for which advertising is a choice variable in consumption. This approach differs from traditional models of advertising (for example, Dorfman & Steiner, 1954; Dixit & Norman, 1978; Stahl, 1989; Stegeman, 1991) in which advertising acts as a pure shift parameter in the demand for an advertising ﬁrm’s product. Implicit in the Becker and Murphy approach is that

The Value of Advertising in a Magazine Bundle

111

advertising is not an omnipresent phenomenon; rather, advertising can be actively avoided or pursued by consumers depending upon their preferences. Unfortunately, Becker and Murphy do not provide empirical evidence to support their theoretical model, although they do provide rather convincing anecdotal evidence. This paper attempts to ﬁll this gap in the literature focusing on circulation industries by investigating the net effect of advertising levels on U.S. magazine sales and cover prices. The U.S. magazine industry is a convenient vehicle to test the BeckerMurphy hypothesis that advertising may be considered as a good, i.e. people are willing to pay to consume advertising. Other industries, such as cable and satellite television subscriptions and pay-per-view sports events, might also provide data compatible with an empirical test of a corollary to the Becker and Murphy hypothesis that advertising is a consumer choice variable. Speciﬁcally, such industries would facilitate a test of whether advertising is a “bad,” i.e. something individuals would pay to avoid. However, the data describing these other industries have proven difﬁcult to collect. Therefore, our attempts to empirically test Becker and Murphy focus on the magazine industry. The characteristics of magazines make the industry conducive to empirical implementation of the Becker-Murphy hypothesis. Speciﬁcally, newsstand sales are necessarily voluntary purchases of a bundle, comprised of advertising and editorial content. We attempt to extricate the value placed on the advertising portion of this bundle, which allows for a direct test of Becker and Murphy. If advertising reduces the value of the bundle, as reﬂected in newsstand price, then it can be deduced that advertising is a bad. However, if advertising raises the value of the bundle, consumers are found to treat advertising as a good. Magazines are published in fairly speciﬁc genres, with particular editorial contents and readership characteristics that make it plausible for advertising to be a desirable component of the bundle. The relative stability of circulation levels and the titles available in the U.S. magazine market makes the analysis of advertising’s impact on the demand for various genres of magazines relatively straightforward. This study employs data describing 117 U.S. magazines from 1996 through 1998, separated into ﬁve genres. Advertising is found to increase both newsstand circulation and newsstand price for four of the ﬁve genres. The empirical evidence suggests that for the majority of magazines available, advertising is a valued component of the bundle that comprises a magazine. Thus, the U.S. magazine industry is found to support Becker and Murphy’s hypothesis that consumers actively choose the level of advertising they consume.

112

CRAIG A. DEPKEN II AND DENNIS P. WILSON

The remainder of the paper is structured as follows. The next section reviews the popular treatment of magazines (and newspapers) from the point of view of the publisher, and reviews the Becker and Murphy alternative. This section also highlights the differences between the existing publisher-focused literature and Becker and Murphy’s consumer-focused approach. Section III presents an empirical model, describes the data, and presents the results. The ﬁnal section offers concluding remarks and suggestions for future research.

II. ADVERTISING IN CIRCULATION INDUSTRIES Magazine and newspaper markets can be grouped together under the term “circulation industry” (Chaudhri, 1998). While new media outlets, such as the Internet and cable television, have been investigated in recent years (Bakos & Brynjolfsson, 2000; Emmons & Prager, 1997, respectively), the existing theory dealing with circulation industries has remained consistent over the past thirty years. While focus of this study is the U.S. magazine industry, the implications could be extended to other circulation industries. For a market of its size, the magazine industry has received relatively little attention in the literature. However, the newspaper industry has been the focus of a number of studies. Because of the similarity of the two industries, we discuss the previous approaches in both markets. The investigation of newspapers and magazines started in the traditional Structure-ConductPerformance paradigm in the 1930s (Cover, Thompson & Cohenour, 1931). Early in the investigation of circulation industries it was recognized that the industry provides two distinct but not mutually exclusive sources of revenue: circulation sales and advertising revenue.1 This was less a revelation of economists than recognition of how the magazine industry was evolving from predominantly local magazines to large nationally distributed magazines, such as Time, Newsweek, and Fortune (Abrahamson, 1997). Early analyses of newspaper monopolies viewed the end product as a bundle of advertising and editorial content (Cover, Thompson & Cohenour, 1931). Blair and Romano (1993) present a model of a newspaper monopolist in which they analyze the positive feedback of advertising on circulation. Their main focus is the price elasticity of demand for newspapers, which they claim is underestimated using traditional econometric models that do not accurately account for the feedback of advertising on circulation that, in turn, motivates more advertising. Their empirical support is borrowed from Rosse (1980). Chaudhri (1998) analyzes a circulation-industry monopolist that weighs the costs and beneﬁts of increasing advertising levels while simultaneously determining optimal circulation levels. He ﬁnds that price-taking behavior in

The Value of Advertising in a Magazine Bundle

113

both circulation and advertising leads to marginal cost pricing in both submarkets. However, market power in the advertising market may motivate the information provider to sell subscriptions at below marginal cost, a ﬁnding supported in other models of multi-product monopolists (e.g. Tirole, 1988). The intuition is that the more subscribers a publisher has, the more the publisher can charge for the advertising included in the bundle (Abrahamson, 1997). Round and Bentick (1997) investigate the subscription discounts offered on Australian magazines. While our focus is on newsstand sales, some of their ﬁndings are of interest in the current context. They ﬁnd that the subscription discount increases with the number of issues a magazine has in a given year, the newsstand price and a measure of economies of scope. We employ similar variables in our empirical analysis of newsstand sales. Abrahamson (1997) contends that circulation levels increase the value of a given stock of advertising, thereby increasing its value to advertising ﬁrms. Lowering cover price below publication costs, using advertising revenue to subsidize readership, increases circulation. However, Thompson (1989) argues that increased circulation may actually reduce the value of a given stock of advertising because of a diluted readership. Small readership may have speciﬁc characteristics that are of value to advertising ﬁrms, such as a speciﬁc disposable income level or revealed preferences for certain goods. The dilution of these desirable characteristics could in turn reduce advertising revenue. Thompson considers advertising to be an enhancement to circulation but does not investigate the impact of advertising on cover price. An alternative interpretation of Thompson’s ﬁndings follows Steiner’s (1952) network externality approach to U.S. radio broadcasts. In Steiner’s theory, there is a positive externality to other individuals experiencing a broadcast, in essence creating a club good. The fact that one is a member of the “club,” whether it be a listener base or a readership base, may enhance the value of consumption. The idea being that one can share the experiences of listening, watching or reading, with others.2 These investigations approach the problem from the publisher’s or the advertiser’s point of view. However, claims such as those by Abrahamson that advertising revenues heavily subsidize the editorial content for readers are questionable. Reddaway (1963), using data from a Royal Commission, estimated that at most only one-third of advertising revenue would be available to subsidize readers. Further, casual observation of U.S. magazines shows that those magazines with an overwhelming amount of advertising tend to have little editorial content.

114

CRAIG A. DEPKEN II AND DENNIS P. WILSON

The effect of advertising on circulation is clearly a point to consider. It is typically assumed that the direct impact of advertising on circulation is positive. Thus, the more advertising there is in a given issue, the greater is the quantity demanded because of the informational value of the advertising. In a uniform-pricing environment, the increase in circulation, ceteris paribus, requires a decrease in cover price. From the point of view of the publisher, any loss in circulation revenue is recovered, on the margin, by gains in advertising revenues (for example, Corden, 1953; Reddaway, 1963; Dertouzos & Trautman, 1990). While this feedback hypothesis has potential merit, empirical support is relatively sparse. Thompson (1989) provides evidence that the English newspaper industry seems consistent with this feedback mechanism. Rosse (1980) conﬁrms this effect. To the authors’ knowledge there are no existing studies that test the value of advertising to consumers. Rather, all previous studies have focused on the value of advertising to the publisher or the advertiser. The economic theory of commodity bundling suggests that consumers may be forced to purchase, if not directly consume, certain components of the bundle even if they do not value the component.3 Such claims have been indirectly made in the newspaper and magazine markets (Abrahamson, 1997). Stigler (1968) proposes that bundles exist because one or both components of the bundle would command less value if separately marketed. Thus, Stigler views commodity bundling as a method of extracting consumer surplus. This is likely the case for the cable television industry (the authors dislike having to purchase ﬁve shopping networks in order to obtain ESPN Classic). Alternatively, Adams and Yellen (1976) propose that complementary goods may be worth more in a bundle than separately because of reduced search and transaction costs. Despite also being a media outlet, it is not immediately clear that the magazine industry is directly analogous to cable television. The magazine market provides for differentiation in at least two dimensions: editorial and advertising content. Some magazines, such as Modern Bride and Computer Shopper, are comprised almost entirely of advertising content. On the other hand, magazines such as Liberty and The Harvard Business Review consist almost entirely of editorial content. The differentiation between levels of advertising and editorial content is augmented by competition in and across genres of editorial/advertising content. While some magazines are very general in nature, e.g. Time or Newsweek, others are very specialized, e.g. UFO Magazine or Potato Grower. The generalization or specialization of information allows magazine publishers to differentiate their bundle of editorial and advertising content and target different audiences. These efforts provide added

The Value of Advertising in a Magazine Bundle

115

value to potential advertisers. Therefore, magazines can be viewed as a differentiated-good market with imperfect-competition, akin to the ready-to-eat cereal or automobile industries. The claim that magazines bundle advertising and editorial content to extract consumer surplus is belied by magazines such as Shoes and Swimsuit Magazine, which are comprised solely of shoe and swimsuit advertisements, respectively. Further refutations of the “advertising is a bad” hypothesis are the several women’s (and recently men’s) fashion magazines that provide, on average, hundreds of pages of advertising with relatively little editorial content. The advertising of these magazines is clearly aimed at a particular target audience. Casual observation at any hair-salon provides anecdotal evidence that the purpose of these magazines is primarily to deliver advertising content, with editorial content being of secondary concern. Oftentimes, the table-of-contents for a magazine such as Cosmopolitan is not reached until forty pages of advertising have been viewed. Such a situation does not lend credence to the editorial value of Cosmopolitan.4 If advertising is viewed as an asset to the magazine bundle, advertising may subsidize, rather than be subsidized by, editorial content. This has not been empirically investigated primarily because of a lack of supporting theory. For example, Blair and Romano (1993) claim that advertising should enhance circulation but do not investigate the impact of advertising on cover price, which would conﬁrm whether advertising is truly a good or a bad. However, Becker and Murphy (1993) provide a convenient framework with which to test the possibility that advertising may have a net positive impact on the magazine bundle. Becker and Murphy theorize that advertising is a choice variable for consumers, with a shadow price that is positive or negative depending on whether advertising is a good or bad. This allows either for advertising to be a burden to consumers for which they must be compensated or to be a component for which consumers are willing to pay for. Unlike other models in which individuals do not ignore unsolicited advertisements (e.g. Grossman & Shapiro, 1984), the actual value of advertising is revealed through the behavior of consumers. The aggregation of individual decisions regarding the impact of advertising on utility is a direct extension of Becker and Murphy’s model. Although the theoretical model provided by Becker-Murphy is unique and compelling, they do not provide any empirical support for their theory, beyond anecdotal suggestions.5 If it is possible to identify that advertising enhances the value of a magazine bundle, having controlled for any possible feedback effects, then advertising in this context can be considered a net good. If

116

CRAIG A. DEPKEN II AND DENNIS P. WILSON

Fig. 1. The Value of Advertising in a Magazine Bundle.

advertising reduces the value of the bundle, then advertising can be considered a bad. The hypothesized effects of changes in advertising on circulation and newsstand price are illustrated in Fig. 1. If circulation and newsstand price respond to changes in advertising in the same direction, then Becker and Murphy’s hypothesis that advertising acts as a demand shifter for the magazine bundle is applicable. Thus, if advertising leads to lower circulation and lower newsstand prices, advertising can be concluded to be an unambiguous bad in the magazine industry. An increase in both circulation and price indicates that advertising acts as an unambiguous good in the bundle of a magazine. If circulation responds positively to changes in advertising and newsstand price responds negatively, advertising is an ambiguous bad that subsidies the price of the magazine bundle as hypothesized by Abrahamson. If, however, advertising leads to a negative effect on circulation and a positive effect on price, advertising is an ambiguous good with value to a speciﬁc market niche. The following empirical analysis attempts to determine the value of advertising to the bundle of editorial content and advertising in the U.S. magazine industry. Since magazine newsstand sales are voluntary and may rely as much on advertising content as editorial sophistication, the magazine industry is a compatible ﬁt for testing the value of advertising to the bundle.

III. THE VALUE OF ADVERTISING IN A MAGAZINE BUNDLE To implement an empirical test, data describing 117 U.S. magazines for 1996 through 1998 are employed. The data allow for a two-equation system with

The Value of Advertising in a Magazine Bundle

117

which to test whether advertising is a net good or bad for consumers of magazines. Three-stage least squares are applied to a fully identiﬁed, twoequation system describing the demand for magazines. The system treats newsstand sales and newsstand price as endogenous. Explanatory variables follow previous empirical models of circulation industries and provide for interesting discussion on the economics of the U.S. magazine industry (e.g. Cechetti, 1986; Thompson, 1989). This study focuses on the impact of advertising on both magazine circulation and newsstand price. The signs of the estimated parameters are indicative of the value placed on advertising in the magazine bundle. Magazines were grouped into genres based upon their editorial (and advertising) content.6 The following genres are employed: Business and News (BN); Entertainment, General, and Travel (EGT); Health, Home, Food, and Lifestyle (HHFL); Sports, Cars, and Men’s (SCM); and Women’s, Fashion, and Parenting (WFP). While the choice of which genre a magazine belongs is often commonsensical, e.g. U.S. News and World Report and Cosmopolitan, www.magazines.com was consulted for genre clariﬁcations of lesser known magazines. While not an internationally recognized authority, this discount magazine retailer provides a representative, consistent delineation of genre. The variables used in this study and their sources are listed in Table 1. The descriptive statistics for the entire sample are reported in Table 2. We list the magazines used in the study in Appendix 1 and the genre-speciﬁc descriptive statistics of selected variables are reported in Appendix 2. Circulation is measured as newsstand sales in thousands of copies. Cover price is measured in nominal dollars. Recognizing that magazines are targeted to different markets, the inﬂuence readership characteristics have on the demand for a particular magazine is controlled by the inclusion of several variables (see Reddaway, 1963; Thompson, 1989). Each magazine’s readership characteristics are captured with the average income and age of readers, the ratio of male to female readers, the average number of readers per copy sold, and the percentage of circulation sales that occurs at newsstand.7 To roughly control for economies of scale on the part of publishers, those magazines that have a strictly local distribution are identiﬁed (see Reddaway, 1963; Dertouzos & Trautman, 1990). To proxy for the advertising to editorial content ratio, the number of advertising pages is interacted with each genre. While it would be desirable to directly measure the advertising-to-editorial content ratio, a direct measure is problematic. It is difﬁcult to develop an objective measure of what constitutes an advertising page for different magazines. Some magazines include

118

CRAIG A. DEPKEN II AND DENNIS P. WILSON

Table 1. Variables Used in the Study. Variable Name

Description

Source

NSCIRC NSPRICE INCOME AGE NUMREAD MFRATIO ISSUES LOCAL NSBASE ADSVOL ADSISSUE BN EGT HHFL SCM WFP YR96 YR97 YR98

Newsstand Circulation (1,000s) Newsstand Price ($) Ave. Reader Income ($) Ave. Reader Age (Years) Ave. Number of Readers per copy Male-Female Reader Ratio Number of Issues Local Distribution % Newsstand Sales Adv. Pages per Volume Adv. Pages per Issue Business and News Entertainment, General and Travel Health, Home, Food and Leisure Sports, Cars, Men’s Women’s, Fashion and Parenting Year 1996 Year 1997 Year 1998

Audit Bureau of Circulation Audit Bureau of Circulation Media Mark Media Mark Media Mark Media Mark Audit Bureau of Circulation Self Generated Audit Bureau of Circulation Advertising Age Advertising Age Self Generated Self Generated Self Generated Self Generated Self Generated Self Generated Self Generated Self Generated

advertising and editorial content on the same page, thereby making it difﬁcult to determine exactly the number of pages of editorial content and advertising content. Publishers self-reveal the number of pages of advertising space but not the number of pages of editorial content. Simply measuring the total number of pages published would suffer from another measurement problem. Some magazines include advertising pages in the total page count, others include only editorial content in the total page count, and still others include a page in its count if it contains “sufﬁcient” editorial content. Short of manually counting each page of each issue of each magazine in the sample, a direct measure of advertising-to-editorial content ratio would be impossible. Using the interactive proxy is expected to capture the inter-genre differentiation in advertising-tocontent ratios. Casual observation indicates that magazines in the same genre tend to be of similar size and frequency of both advertising and editorial content (both in quantity and quality). The data are used to estimate a fully identiﬁed, two-equation system that describes magazine demand and captures the feedback effects of advertising on circulation and newsstand price. The circulation equation is deﬁned as

The Value of Advertising in a Magazine Bundle

119

Table 2. Descriptive Statistics for Entire Sample. Variable

Mean

NSCIRC NSPRICE INCOME AGE READERS MFRATIO ISSUES LOCAL (1 = Yes) NSBASE ADSVOL ADSISSUE Business and News (1 = Yes) Entertainment, General and Travel (1 = Yes) Health, Home, Food and Leisure (1 = Yes) Sports, Cars, Men’s (1 = Yes) Women’s, Fashion and Parenting (1 = Yes) Yr96 (1 = Yes)a Yr97 (1 = Yes) Yr98 (1 = Yes) Observations a

338.19 3.30 49615.88 38.56 5.51 1.72 16.61 0.03 21.84 1294.52 94.27 0.11 0.19 0.19 0.24 0.24 0.30 0.33 0.33 343

Std· Deviation Minimum Maximum 510.01 0.902 11214.145 6.219 2.569 2.238 13.39 0.185 20.064 855.533 74.59 0.323 0.398 0.397 0.428 0.432 0.463 0.472 0.472

0.647 4058.847 1.190 5.990 25937.00 86115.000 23.60 51.90 1.860 17.480 0.021 12.021 6.00 55.00 0.00 1.00 0.368 89.477 138.16 5063.299 11.89 658.66 0.00 1.00 0.00 1.00 0.00 1.00 0.00 1.00 0.00 1.00 0.00 1.00 0.00 1.00 0.00 1.00

Year dummy variables do not sum to 1.00 because of rounding.

NSCIRC = ␤0 + ␤1NSPRICE + ␤2INCOME + ␤3AGE + ␤4AdvBN + ␤5AdvEGT + ␤6AdvHHFL + ␤7AdvSCM + ␤8AdvWFP + ␤9ISSUES + ␤10ISSUESSQ + ␤11YR96 + ␤12YR97 + ␤13YR98 + ␧

(1)

where ␤’s are parameters to be estimated and ␧ is a zero-mean error structure. The variables are as described in Table 1. It is hypothesized that INCOME, ISSUES, and perhaps AGE enhance magazine circulation whereas PRICE and ISSUESSQ should reduce magazine circulation. The remaining variables are ambiguous in their effects on newsstand circulation. In the context of the Becker-Murphy hypothesis, if the advertising-genre coefﬁcients are positive, advertising is found to increase circulation. This may indicate that advertising is a good from the point of view of magazine consumers. However, if the coefﬁcients are negative, advertising is found to decrease circulation and may be viewed as a consumer bad. To help identify the true value placed on advertising by consumers, the equation for the newsstand price of magazines is deﬁned as

120

CRAIG A. DEPKEN II AND DENNIS P. WILSON

NSPRICE = ␥0 + ␥1NSCIRC + ␥2MFRATIO + ␥3LOCAL + ␥4AdvBN + ␥5AdvEGT + ␥6AdvHHFL + ␥7AdvSCM + ␥8AdvWFP + ␥9READERS␥10NSBASE + ␥11YR96 + ␥12YR97 + ␥13YR98 + ␻, (2) where the ␥’s are parameters to be estimated and ␻ is a zero-mean error structure. Again, the variables are as described in Table 1. It is hypothesized that newsstand circulation should reduce the price of magazines because, an increase in the quantity of magazines on the market should reduce the uniform newsstand price publishers can charge. On the other hand, LOCAL, READERS, and NSBASE should increase the price of magazines. Local distribution may preclude a publisher internalizing returns to scale available to a nationally distributed magazine publisher (the average circulation of local magazines is approximately 150,000 copies less than nationally distributed magazines). Therefore, one may expect local distribution to cause an increase in cover price (see Dertouzos & Trautman, 1990). The average number of readers-per-copy would be expected to increase the cover price, as the publisher has only one chance to transfer consumer surplus from readers; it is impossible for the publisher to charge those who free ride on the purchase of another. The percentage of sales that occur at newsstands introduces an element of risk to the publisher, as subscription sales are guaranteed. The introduction of risk into circulation revenue may cause an increase in cover price. The remaining variables are ambiguous in their expected effects.8 The estimated parameters of the advertising proxies, when combined with the results from the circulation equation, allow for a test of whether advertising enhances the value of the bundle offered by magazine publishers. If advertising reduces circulation but also increases newsstand price, publishers may be targeting a speciﬁc audience that is willing and able to purchase the advertising information contained within a given magazine.9 This would provide support for the Becker-Murphy hypothesis that advertising is a good, but would indicate that treating advertising as a good may only hold for a small proportion of the population. On the other hand, if advertising increases both circulation and newsstand price, it would indicate that advertising is treated as a good by a relatively large proportion of the magazine audience. This result would offer strong support of the Becker-Murphy hypothesis. The estimation results for Eqs (1) and (2) are presented in Tables 3 and 4, respectively. The results conform to economic intuition and allow for a discussion of the demand for U.S. magazines, especially in the context of advertising. The majority of the estimated parameters have the expected sign

The Value of Advertising in a Magazine Bundle

121

and are statistically signiﬁcant. The standard errors are calculated using the asymptotic covariance matrix taking advantage of the correlation between the two equations. Analyzing the equations in turn, the estimates for the circulation equation are reported in Table 3. Price is found to have a negative and statistically signiﬁcant impact on circulation, indicating a downward-sloping demand curve for magazines. A one-dollar increase in cover price leads the average circulation to decline by approximately 177,000 copies. As readership income increases, however, the average circulation declines, perhaps indicating magazines are treated as an inferior good. As would be expected, an increase in the number of issues per volume increases circulation. However, the marginal increase in circulation is tempered by the negative parameter estimate on issues squared. This indicates that, from the point of view of a publisher, there is an interior solution for the optimal number of issues in a given volume. Of particular interest in the circulation equation are the estimated parameters on the interaction of genre dummy variables with advertising levels. These interactions proxy for the advertising to editorial content ratio, assumed (and conﬁrmed by casual observation) to be relatively consistent within a given genre. All advertising proxies are statistically signiﬁcant except for the Business and News genre.

Table 3. 3SLS Estimation Results for Circulation Equation Dependent Variable is Newsstand Circulation. Variable NSPRICE INCOME AGE ISSUES ISSUESSQ AdvBN AdvEGT AdvHHFL AdvSCM AdvWFP Yr96 Yr97 Yr98 N R2

Coefﬁcient

Std· Error

t-Statistic

p-Value

–177.439 –0.012 –0.344 65.774 –0.833 0.911 2.470 4.390 2.380 1.904 582.571 610.384 630.294 343 0.290

54.280 0.002 3.538 11.375 0.179 0.792 0.946 0.835 0.907 0.396 124.792 125.581 127.308

–3.26 –4.45 –0.09 5.78 –4.65 1.14 2.60 5.25 2.62 4.80 4.66 4.89 5.02

0.001 0.000 0.922 0.000 0.000 0.250 0.009 0.000 0.009 0.000 0.000 0.000 0.000

122

CRAIG A. DEPKEN II AND DENNIS P. WILSON

These estimates indicate that, for the average magazine in the remaining genres, advertising tends to increase circulation levels. Taken on their own, these parameter estimates can be obtained under two possible scenarios. First, advertising may be subsidizing cover price, therefore advertising simply moves the market for a particular magazine down its own demand curve. On the other hand, advertising may act as a demand shifter for magazines, as suggested by Becker and Murphy, indicating that magazine purchasers treat advertising as a net good. Estimating simultaneously the newsstand price equation solves this identiﬁcation problem. Table 4 reports the estimation results for the newsstand price equation. The inverse relationship between price and quantity of magazines demanded is conﬁrmed by the negative parameter estimate on circulation. A magazine restricted to local distribution has a cover price on average $0.79 greater than nationally distributed magazines, indicating the inability of local magazines to internalize returns of scale. These ﬁndings support Dertouzos and Trautman (1990), who ﬁnd that publishing has large returns to scale. A 1% increase in the percentage of sales through newsstands causes an average increase of approximately $0.01 per copy, supporting the intuition that an increased risk in circulation revenue would cause a minor increase in sales price. The number of readers-per-copy also has a positive impact on cover prices. For every readerper-copy the average cover price increases by approximately $0.08. The Table 4. 3SLS Estimation Results for Newsstand Price Equation Dependent Variable is Newsstand Price. Variable NSCIRC MFRATIO LOCAL READERS NSBASE AdvBN AdvEGT AdvHHFL AdvSCM AdvWFP Yr96 Yr97 Yr98 N R2

Coefﬁcient

Std· Error

t-Statistic

p-Value

–0.001 0.100 0.793 0.080 0.014 0.008 0.007 0.009 0.003 0.002 2.263 2.288 2.363 343 0.338

1.89 ⫻ 10–3 0.025 0.219 0.020 0.004 0.001 0.001 0.001 0.001 0.000 0.145 0.147 0.143

–7.56 3.96 3.60 3.91 3.34 6.91 3.90 6.56 2.03 3.58 15.54 15.54 16.44

0.000 0.000 0.000 0.000 0.001 0.000 0.000 0.000 0.042 0.000 0.000 0.000 0.000

The Value of Advertising in a Magazine Bundle

123

estimated parameter of the male-to-female readership ratio is positive and statistically signiﬁcant, indicating that magazines with a readership of relatively more males charge a higher cover price, perhaps revealing a stronger preference for magazines amongst men. Each of the estimated parameters on the advertising genre variables is positive and statistically signiﬁcant. While the absolute magnitudes seem small, we calculate the contribution of advertising to the average cover price. In each genre this contribution is calculated as approximately $0.82 in Business and News; $0.45 in Entertainment, General and Travel; $0.79 in Health, Home, Food and Leisure; $0.24 in Sports, Cars and Men’s; and $0.28 in Women’s, Fashion and Parenting. The estimates indicate that readers treat the advertising portion of the average magazine bundle as having a positive value, i.e. advertising is a good. If advertising had carried a negative estimated parameter, indicating that advertising reduced cover price, it would have been possible to interpret magazine advertising as a bad. Combining the estimated results, advertising can be inferred to be a shift parameter in the demand for magazines. Thus, advertising increases both circulation and the value of the bundle provided by the publisher. These results provide strong support for the model presented by Becker and Murphy. Magazine purchases are purely voluntary and often there are several close substitutes within a given genre. If advertising were a bad, one would expect to see individuals being compensated to consume advertising or willing to pay to avoid advertising. However, the empirical results indicate that individuals value the advertising provided in the magazine bundle and suggests that future research should carefully consider the alternative assumption that advertising is necessarily welfare reducing.

IV. CONCLUSIONS Research on the market for magazines and newspapers often assume the publisher to be a monopolist in both advertising and circulation levels. Further, it has often been assumed, without corroborating empirical evidence, that advertising subsidizes consumers through a reduction in cover price. This is equivalent to assuming advertising is inherently a “bad” that reduces the overall value of a magazine’s bundle of advertising and editorial content. Thus, consumers must be compensated for the consumption of advertising by a corresponding increase in editorial quality, editorial quantity, or a reduction in cover price. Several authors assume advertising increases circulation but fail to discuss the value of advertising in the bundle as reﬂected in both circulation and newsstand price. An alternative theory offered by Becker and Murphy

124

CRAIG A. DEPKEN II AND DENNIS P. WILSON

(1993) treats advertising as a choice variable on the part of consumers in which advertising can be a good or a bad. In the context of a magazine bundle, the Becker-Murphy hypothesis provides the motivation for this study: to extricate the value of advertising to consumers of U.S. magazines. While seemingly benign, the differences between the publisher-oriented approach used in traditional models of circulation industries and the consumeroriented approach of Becker and Murphy have important implications for publishers. If advertising is treated as a valued component of the magazine bundle, then the optimal level of advertising in a given magazine may be greater than predicted by traditional circulation models. Indeed, many theories of circulation markets predicated on the negative marginal utility of advertising might well need reinvestigation if the results in this study withstand scrutiny. This paper takes advantage of a unique data set describing 117 U.S. magazines for the years 1996–1998. A fully identiﬁed, two-equation system, in which newsstand sales (quantity demanded) and newsstand cover price are endogenous, is estimated. Using three-stage least squares, the impact of advertising on magazine demand is found to conﬁrm the theory postulated by many: for the majority of magazine genres, advertising increases average circulation. However, such a ﬁnding on its own is not enough to discern the value placed on the advertising portion of the magazine bundle. The results from the magazine-price equation provide evidence that advertising enhances newsstand price. These results, combined with those obtained in the circulation equation, indicate that advertising is a shift parameter in the demand for magazines, as opposed to the alternative hypothesis that advertising is a bad and subsidizes cover price. Our data also include several variables that, to date, have not been used to investigate the mechanics of circulation industries. We ﬁnd that local distribution tends to increase cover price, perhaps because of a lack of economies of scale. Greater dependence on newsstand sales likewise increases newsstand price, perhaps because of increased risk borne by the publisher. Further, the more readers-per-copy a magazine enjoys the greater the newsstand price. Finally, we ﬁnd that the greater the male-to-female readership ratio, the greater the average newsstand price. This ﬁnal ﬁnding has no existing theoretical support, but may be an avenue of future research.

NOTES 1. Buchan and Siegfried (1978) provide a detailed analysis of the vertical relationships in the U.S. magazine industry. We rely upon their explanation as it accords with the way the industry operates today.

The Value of Advertising in a Magazine Bundle

125

2. This seems very much the case with Home Box Ofﬁce’s show “The Sopranos.” While the show is one of the most popular on television, it is relatively exclusive because of the reduced number of households with access to HBO. 3. Emmons and Prager (1997) relate commodity bundling to the cable television industry. 4. Indeed, today’s reader of Cosmopolitan would likely not recognize a 1930s issue of the magazine, which started out as a public policy vehicle. 5. Tremblay and Tremblay (1995) and Farr, Tremblay, and Tremblay (2001) have empirically applied Becker and Murphy’s model to the beer industry and the cigarette industry, respectively. By considering the social welfare effect, the beneﬁts of advertising, such as reduced search costs, are outweighed by the cost of higher prices and negative social externalities in both industries. However, the ban of cigarette advertising acts to lower social welfare by decreasing market competition and increasing ﬁrms market power. Moreover, these studies use industry level data, and not ﬁrm speciﬁc data. 6. The motivation for grouping magazines by genres follows from Reddaway’s study of English newspapers, Round and Bentrick’s (1997) analysis of Australian magazines, and Steiner’s (1952) approach to U.S. radio broadcasts. 7. It should be noted that the readership characteristics reported by Media Mark reﬂect the readership of both subscription and newsstand purchasers. We implicitly assume that the characteristics of newsstand purchasers closely resemble the characteristics of a magazine’s total readership. 8. The ratio of male-to-female readers is correlated, although not perfectly, with the genre breakdowns (see Appendix 2). Those magazines in the women’s, fashion, and parenting genre tend to have relatively low male-to-female ratios, whereas magazines in the sports, cars, and men’s genre tend to have relatively high male-to- female ratios. Magazines in the other genres have less of a gender correlation with the genre breakdowns. The expected parameter sign on this variable is ambiguous. It is not clear that having relatively more male than female readers would cause an increase in price, there being no speciﬁc theory to predict such a result. 9. In support of this intuition, consider a magazine such as The Robb Report. This magazine specializes in advertising very expensive consumer products and services.

ACKNOWLEDGMENTS The authors wish to thank Melissa Lind for her assistance in gathering the data used in this study.

REFERENCES Abrahamson, D. (1997). Magazines in the Twentieth Century. In: M. Blanchard (Ed.), History of Mass Media in the United States: An Encyclopedia. New York: Garland. Adams, W., & Yellen, J. (1976). Commodity Bundling and the Burden of Monopoly. Quarterly Journal of Economics, 90, 475–498.

126

CRAIG A. DEPKEN II AND DENNIS P. WILSON

Bakos, Y., & Brynjolfsson, E. (2000). Bundling and Competition on the Internet. Marketing Science, 19, 63–82. Becker, G., & Murphy, K. (1993). A Simple Theory of Advertising as a Good or a Bad. Quarterly Journal of Economics, 108, 941–964. Blair, R., & Romano, R. (1993). Pricing Decisions of the Newspaper Monopolist. Southern Economic Journal, 59, 721–732. Buchan, R., & Siegfried, J. (1978). An Economic Evaluation of the Magazine Distribution Industry. Antitrust Bulletin, 23, 19–50. Cecchetti, S. (1986). The Frequency of Price Adjustment: A Study of the Newsstand Prices of Magazines. Journal of Econometrics, 31, 255–274. Chaudhri, V. (1998). Pricing and Efﬁciency of a Circulation Industry: The Case of Newspapers. Information Economics and Policy, 10, 59–76. Corden, W. M. (1953). The Maximisation of Proﬁt by a Newspaper. Review of Economic Studies, 20, 181–190. Cover, J., Thompson, P., & Cohenour, V. (1931). Newspaper Advertising Rates and Circulation. Journal of Business, 4, 115–126. Dertouzos, J., & Trautman, W. (1990). Economic Effects of Media Concentration: Estimates from a Model of the Newspaper Firm. Journal of Industrial Economics, 39, 1–14. Dixit, A., & Norman, V. (1978). Advertising and Welfare. Bell Journal of Economics, 9, 1–17. Dorfman, R., & Steiner, P. (1954). Optimal Advertising and Optimal Quality. American Economic Review, 44, 826–836. Emmons, W. III, & Prager, R. (1997). The Effects of Market Structure and Ownership on Prices and Service Offerings in the U. S. Cable Television Industry. RAND Journal of Economics, 28, 732–50. Farr, D., Tremblay, C., & Tremblay, V. (2001). The Welfare Effect of Advertising Restrictions in the U.S. Cigarette Industry. Review of Industrial Organization, 18, 147–160. Grossman, G., & Shapiro, C. (1984). Informative Advertising with Differentiated Products. Review of Economic Studies, 51, 63–81. Reddaway, W. (1963). The Economics of Newspapers. Economic Journal, 73, 201–218. Rosse, J. (1980). Vertical Price Fixing in Newspaper Distribution: A Per Se Rule That Makes Everyone Worse Off. Studies in Industry Economics 112, Stanford University. Round, D., & Bentick, T. (1997). Magazine Subscription Discounts in Australia. Review of Industrial Organization, 12, 555–577. Stahl, D. (1989). Oligopolistic Pricing and Sequential Consumer Search. American Economic Review, 79, 700–712. Stegeman, M. (1991). Advertising in Competitive Markets. American Economic Review, 81, 210–223. Steiner, P. (1952). Program Patterns and Workability of Competition in Radio Broadcasting. Quarterly Journal of Economics, 66, 194–223. Stigler, G. (1968). A Note on Block Booking. In: The Organization of Industry. Homewood, IL: Irwin. Thompson, R. (1989). Circulation Versus Advertiser Appeal in the Newspaper Industry: An Empirical Investigation. Journal of Industrial Economics, 37, 259–271. Tirole, J. (1988). The Theory of Industrial Organization. Cambridge, MA: MIT Press. Tremblay, C., & Tremblay, V. (1995). Advertising, Price and Welfare: Evidence from the U.S. Brewing Industry. Southern Economic Journal, 62, 367–381.

The Value of Advertising in a Magazine Bundle

127

APPENDIX 1 The following table lists the magazines used in the study. Bold print indicates local distribution. Business and News

Entertainment, General and Travel

Health, Home, Food, Lifestyle

Sports, Cars, Men’s

Women’s, Fashion and Parenting

Barron’s Black Enterprise

CondeNastTraveler Discover

Automobile Car & Driver

Allure Bridal Guide

Business Week Entrepreneur Forbes Fortune Home Ofﬁce Computing Inc.

Ebony Jet Life National Enquirer National Geographic

Architectural Digest Better Homes and Gardens Bon Appetit Consumer’s Digest Cooking Light Country Home Country Living

Easyriders Esquire Family Handyman Field and Stream Four Wheeler

Child Cosmopolitan Elle Essence First for Women

Family Circle

Golf Digest

Glamour

Kiplinger’s

National Geographic Traveler New York

Food and Wine

Golf Magazine

Money

People

Good Housekeeping

Golf World

Newsweek PC Computing Time U.S. News and World Report

Popular Science Premier Reader’s Digest Rolling Stone

Gourmet Health Home House Beautiful

GQ Guns & Ammo Hot Rod Motor Trend

Harper’s Bazaar Ladies Home Journal Mademoiselle McCall’s Modern Bride New Woman

Scientiﬁc American Smithsonian Soap Opera Digest Spin The New Yorker Town and Country TV Guide Texas Monthly Yankee

Martha Stewart Living Men’s Health Metropolitan Home Midwest Living Muscle & Fitness Organic Gardening Prevention Shape Southern Living Sunset Traditional Home Weight Watchers

Outdoor Life Penthouse Playboy Popular Mechanics Road and Track Runner’s World Ski Skiing Sport Sporting News Sports Aﬁeld Sports Illustrated

Parenting Parents Redbook Self Seventeen Teen U.S. Vanity Fair Victoria Vogue Woman’s Day Working Woman Ym

128

CRAIG A. DEPKEN II AND DENNIS P. WILSON

APPENDIX 2 The following table reports descriptive statistics for selected variables by genre. Genre

Variable

Mean

Std. Dev.

Minimum

Maximum

BN

Newsstand Price Newsstand Sales (1000s) Male-Female Ratio %Newsstand Sales Average Age Average Income Number of Observations

3.60 94.97 1.74 10.12 41.75 62949.69 41

0.78 68.50 0.53 10.16 2.68 12995.50

2.95 19.60 0.744 2.12 36.90 32611.00

4.99 260.53 3.02 44.37 46.70 86115.00

EGT

Newsstand Price Newsstand Sales (1000s) Male-Female Ratio %Newsstand Sales Average Age Average Income Number of Observations

3.10 460.06 1.09 18.11 40.02 47169.55 65

0.97 805.93 0.833 18.92 6.39 12024.71

1.19 7.51 0.08 0.36 24.60 25937.00

4.95 4058.84 5.36 84.86 51.91 74654.79

HHFL

Newsstand Price Newsstand Sales (1000s) Male-Female Ratio %Newsstand Sales Average Age Average Income Number of Observations

3.35 342.33 0.36 17.39 43.67 52058.47 69

0.94 471.11 0.25 10.86 3.92 9161.04

1.29 19.67 0.07 2.44 30.70 34558.00

5.00 2275.50 1.42 50.21 51.60 77972.00

SCM

Newsstand Price Newsstand Sales (1000s) Male-Female Ratio %Newsstand Sales Average Age Average Income Number of Observations

3.61 165.59 4.91 19.68 35.87 48868.32 82

0.79 163.511 2.34 18.95 4.45 9200.75

2.50 0.64 1.28 0.42 28.8 32903.00

5.99 713.79 12.02 69.63 47.90 69805.00

WFP

Newsstand Price Newsstand Sales (1000s) Male-Female Ratio %Newsstand Sales Average Age Average Income Number of Observations

2.97 532.32 0.16 36.56 34.19 43580.33 82

0.82 496.63 0.11 23.78 6.08 5889.64

1.29 8.51 0.02 1.34 23.60 30104.00

4.99 2171.75 0.59 89.47 48.40 60606.80

PRICING DYNAMICS OF MULTIPRODUCT RETAILERS Daniel Hosken, David Matsa and David Reiffen ABSTRACT This paper attempts to broaden our understanding of retail pricing dynamics by providing some systematic evidence about U.S. grocery prices. Using a large data set containing prices on twenty categories of goods from thirty U.S. metro areas for the period 1988–1997, we ﬁnd a number of empirical regularities. Sales are common phenomenon in that retailers seem to have a “regular” price, and most deviations from that price are downward. There is also considerable heterogeneity in sale behavior across goods within a category, such as cereal. Within each category of goods, retailers regularly put some items on sale, while other items are rarely, if ever, put on sale. Finally, the probability of a sale on an item appears to be greater when demand for that item is higher. These results suggest that retailers use complicated strategies in pricing the items they sell that differ across items and over time. Studies that use retail prices and do not account for the process determining retail prices are likely to yield misleading results.

Advertising and Differentiated Products, Volume 10, pages 129–153. Copyright © 2001 by Elsevier Science Ltd. All rights of reproduction in any form reserved. ISBN: 0-7623-0823-0

129

130

DANIEL HOSKEN, DAVID MATSA AND DAVID REIFFEN

I. INTRODUCTION As consumers, we all have some familiarity with the complex pricing strategies employed by supermarkets. For instance, among the more than 20,000 items they carry, supermarkets choose to offer only a small fraction of the items they sell at a low “sale” price each week. Increasingly, data are available that track price and quantity information, and economists have begun using the data to learn more about retailer and consumer behavior. Two principle types of empirical studies have employed these data. First, economists have estimated demand systems in consumer product industries to determine if products are close or distant substitutes. Second, in some retail environments, economists have attempted to directly measure the extent to which competing retailers constrain each other’s pricing (see, for example, Ashenfelter et al., 1998). In principle, the increased availability of highly disaggregated retail price data should allow economists to better determine the substitutability/complementarity between products and measure the effects of retail competition on price. Despite the increase in the quantity of data available, there is relatively little research on how to correctly use these data to estimate economically meaningful measures of closeness between products or outlets. This lack of research creates a potentially serious problem for policy makers who would like to use the results from empirical studies to inform policy decisions. For example, often when competing consumer products companies merge, they will hire economists to conduct a statistical study showing the degree of substitutability across a group of products, including those of the merging ﬁrms. These studies typically use retail scanner data collected at grocery stores, drug stores, or mass-merchandisers. Whether these substitution measures correspond to the measures of interest to a policy-maker depends critically upon the source of price variation. As shown below, most of the variation in consumer prices used to estimate substitution patterns comes from items being placed on “sale” by the grocery chain, not from general changes in wholesale prices. Previous theoretical and empirical research suggests that if retailers place items on sale to, in part, intertemporally price discriminate, then the substitution measures estimated using contemporaneous price and quantity data in a demand model will not correspond to the substitution measures relevant to policy-makers. Price measurement is also an important issue in estimating the relationship between market structure and prices. For example, suppose a researcher would like to exploit the wealth of data available from grocery store scanners in order to measure how the “price” charged by a retailer is affected by competition from other retailers. Given that a typical grocery stores sells thousands of

Pricing Dynamics of Multiproduct Retailers

131

distinct items, the question of how to construct the grocery store’s “price” becomes important. One approach to measuring price would be to construct a bundle of products corresponding to a typical consumer’s purchases. However, even this task is difﬁcult because retailers pursue different pricing strategies for different types of goods. For example, because consumers are more aware of the prices of popular products (e.g. Tide or Cheerios), grocery stores have strong incentives to charge low prices on these items to maintain a low-price image. In contrast, consumers know relatively less about the prices of most unbranded products, giving retailers less of an incentive to charge low prices for these products. In addition, some retail products often move between relatively high everyday prices and low sale prices while other products are rarely offered on sale. The goals of this paper are to provide systematic evidence about grocery pricing behavior, and offer guidance to researchers on how to use retail prices to both correctly estimate demand systems and measure the competitive signiﬁcance of competing retailers. For our empirical work, we use a large nonpublic data set obtained from the Bureau of Labor Statistics (BLS) which contains more than 350,000 monthly quotes on 20 categories of consumer products from 30 U.S. cities. We use this data set to establish a number of pricing regularities. For instance, we ﬁnd that most products have a regular retail price, and that most deviations from that price are downward, likely the result of sales. In order to develop empirical studies which appropriately use retail pricing data, one must developing an understanding of why retailers pricing strategies are so complex. Towards this end, we present and test a theory of sales. Speciﬁcally, in an earlier paper (Hosken, Matsa, & Reiffen, 2000), we derived a model which shows that, other things equal, retailers will place products that are more popular on sale more often. Here, we ﬁnd strong empirical results consistent with this prediction. For example, we observe there is substantial heterogeneity in which grocery items are placed on sale. Within a product category, e.g. cereal, we ﬁnd that some brands are quite likely to go on sale, while others almost never go on sale. Additionally, we ﬁnd that products that have known seasonal increases in demand are more likely to go on sale in periods of high demand than low demand.

II. LITERATURE REVIEW Two key features of the supermarket industry are that each ﬁrm sells a large number of individual products and that the typical consumer purchases many individual products in each visit. Casual observation suggests that the pricing

132

DANIEL HOSKEN, DAVID MATSA AND DAVID REIFFEN

policies adopted by supermarkets differ across goods and vary over time for each good. Speciﬁcally, a typical pattern is for a supermarket to put a group of products on sale each week and to advertise those prices, with the advertised products changing from week to week. The question of why there are temporary price reductions – that is, why retailers’ proﬁt-maximizing prices change over time – inherently must be analyzed in a dynamic setting. In contrast, the question of why the prices of a subset of goods are advertised can be analyzed in a static setting. Section A presents theories of dynamic pricing, while Section B focuses on the static question of why multiple products are put on sale each period. A. Theories of Pricing Dynamics Previous research provides two types of explanations for the sales phenomenon. First, Conlisk, Gerstner and Sobel (1984) have suggested that sales can be used to price discriminate between consumers based on differences in both their demand elasticity and their willingness to wait (which is analytically similar to differences in costs of inventorying). In their model, sales arise because periodic price reductions lead to a large volume of purchases by highelasticity customers. Hence, this strategy allows the monopolist to charge a low price to high-elasticity customers, while most of the purchases by low-elasticity customers are at a high price.1 Second, Varian (1980) has suggested that sales arise because consumers differ in their willingness to shop, and retailers compete for those consumers who will only buy at the store with the lowest prices. The only symmetric equilibria in Varian’s model feature mixed strategies, where all retailers choose their price from an identical continuous distribution. Sobel (1984) combines elements of the two preceding models to explain sales. In his model, there are multiple retailers, and high-value (low-elasticity) consumers are not only willing to pay more for the good and less willing to wait (as in Conlisk, Gerstner & Sobel), but they also are loyal to one retailer (as in Varian).2 The basic characteristics of the equilibrium in Sobel’s model resemble the equilibrium in the Conlisk, Gerstner, and Sobel model. Retailers charge a high price when the number of non-loyal customers is small, but as the number grows, it eventually becomes proﬁtable to reduce price to attract nonloyal customers. The key difference between the monopoly and multiple retailer equilibria is that competing retailers will consider having a sale sooner than a monopolist.3 Another difference is that “sale” prices are lower in the Sobel model. Finally, one can extend the model to show that the difference between the monopoly and multiple retailer cases is a general one. That is, a

Pricing Dynamics of Multiproduct Retailers

133

reduction in the number of competing retailers reduces the frequency and depth of sales, but does not affect the non-sale price of any good. Hosken and Reiffen (2001) extend the Sobel analysis by considering competition between multiproduct retailers. An implication of their model is that competition between retailers always leads to some goods being on sale in each period, while others will be at their “regular” price. Because any individual good will only be on sale infrequently, the identity of the goods sold at low prices changes from period to period. As such, the model implies that sales may even occur for products for which intertemporal price discrimination is not feasible, such as products like lettuce or bananas that cannot be inventoried (the reason is that one or more goods must be on sale each period). B. Costly Information and Multiproduct Retailers The literature on the price promotion of multiproduct retailers tends to focus on the information value of the advertising. A contribution that is particularly relevant in the supermarket context is the work of Lal and Matutes (1989, 1994), who model competition between multiproduct retailers. The main question that Lal and Matutes focus on is whether retailers charge the same markups on all items they sell. They show that (under certain circumstances, particularly when the retailers’ advertising costs and the consumers’ transportation costs are not too trivial or too great) retailers will charge relatively low markups on advertised items and large markups on unadvertised items.4 Thus, their paper provides an economic explanation for the “loss leader” strategy used by many retailers. Hosken, Matsa, and Reiffen extend the Lal and Matutes model to consider how retailers decide which products to advertise. They show that, other things equal, products that are relatively more popular (that is, are consumed by a higher fraction of the population) are more likely to go on sale. The intuition behind the result is that when advertising is costly, a retailer will try to reach the largest number of consumers at the lowest cost. Thus, by advertising a low price on a more popular item, the retailer attracts a large number of consumers who will also buy unadvertised items with relatively high margins. While the theories we have summarized are highly stylized, they do yield a number of predictions we can examine in the data. For instance, the intertemporal price discrimination models of Sobel and Conlisk, Gerstner, and Sobel, predict that retail prices of certain kinds of products will normally be at a high everyday price and periodically be discounted to a low sale price. Hosken and Reiffen’s multiproduct retailer model predicts differences in sale

134

DANIEL HOSKEN, DAVID MATSA AND DAVID REIFFEN

behavior between categories of goods (e.g. perishable vs. non-perishable goods). Finally, the Hosken, Matsa and Reiffen model predicts that competing ﬁrms will be more likely to place popular products on sale than unpopular products. The empirical analysis in Section IV of the paper shows that these predictions appear to hold in the empirical distribution of prices.

III. DATA Many researchers estimating demand functions for speciﬁc consumer products use retail scanner data available from ﬁrms such as A. C. Nielsen. The information these ﬁrms can provide is quite detailed. For example, Nielsen sells price and quantity data as disaggregated as the weekly price and quantities of a particular stock keeping unit (or sku), e.g. an 18 ounce container of Skippy Creamy peanut butter, in a speciﬁc city. However, as a practical matter, gaining access to a broad set of products for a long time period is prohibitively expensive for most researchers. In this study, we make use of a data set provided to us by the Bureau of Labor Statistics (BLS). While this data set does not contain the speciﬁc brand information that data from a ﬁrm such as A. C. Nielsen provides, this data set contains a wide range of product prices for from many cities for a ten year time period. In collecting the data used to calculate the Consumer Price Index (CPI), the BLS collects prices from a large number of retailers in 88 geographic areas, collecting prices of speciﬁc items in up to 94 categories of goods.5 The goal of the BLS is to accurately measure changes in the prices consumers face. For this reason the BLS uses a sampling scheme that collects product prices that approximate consumer expenditure patterns. In addition, because these data are used to measure price movements over time, the BLS takes great pains to accurately measure the price of speciﬁc products over time. Thus, the underlying price quote data used in the construction of the CPI will allow us to observe how prices vary over time. Below, we describe in detail how the BLS collects its price data. Within each product category, the BLS samples the price of a speciﬁc item at the same store monthly for up to 5 years. That is, if in the ﬁrst month the BLS uses a 2-liter bottle of Pepsi as its cola product in a speciﬁc store, then it will continue collecting pricing data on 2-liter bottles of Pepsi as its cola item as long as the store remains in the sample and 2-liters bottles of Pepsi remain on its shelf. The number of retailers sampled in each area increases with the area’s population. In each geographic area the BLS changes all of the stores in its sample every ﬁve years. Hence, the largest potential number of observations in any individual price series is 60. The choice of which speciﬁc item(s) in a

Pricing Dynamics of Multiproduct Retailers

135

category to sample from each supermarket is based on a revenue weightedaverage randomization. For example, if 2-liter bottles of Pepsi represent 10% of a store’s cola revenue, then the BLS randomization results in a 10% chance that 2-liter Pepsi will be the sampled cola product. The data we use in this study consist of individual price series for speciﬁc products. For example, each price series in the cola category in Chicago contains monthly observations on the price of a speciﬁc brand and container size of cola at a retail outlet in the Chicago area, for up to 60 consecutive months.6 Most product categories have multiple price series in each geographic area. Unfortunately, the price series provided to us do not contain information that identiﬁes the speciﬁc product and package size sampled within each category.7 We only know that all of the prices within a price series correspond to prices for a speciﬁc product at a speciﬁc store within a category. We do not know, for example, whether that speciﬁc cola product is a 12-pack of Coke or a 2-liter bottle of Pepsi. We also cannot identify the store or chain associated with each price series. Hence, we cannot determine when two series are taken from the same store or chain.8 The data set we received from the BLS contains all of the price series the BLS collected on 20 categories of goods (see Table 2 for a list of the speciﬁc categories) from 30 geographic areas (see Table 3 for a list of the speciﬁc areas) for the period 1988–1997.9 Tables 1–4 provide descriptive information about the data set.10 Table 1 shows that the observations are fairly evenly distributed throughout the sample period, although some years have more observations than others. Table 2 presents both the number of unique price series and the number of observations for each product category. Our data contain far more

Table 1. Description of Data Set by Year. Year

Percentage of Observations

1988 1989 1990 1991 1992 1993 1994 1995 1996 1997

11.4% 10.0% 9.6% 9.9% 10.1% 9.2% 9.3% 10.3% 9.8% 10.4%

136

DANIEL HOSKEN, DAVID MATSA AND DAVID REIFFEN

information on some grocery products (e.g. ground beef and white bread) than others (e.g. baby food and paper products). This reﬂects a policy on the part of the BLS to collect more data on products that are viewed as more important in measuring the CPI. Table 3 shows the number of price series and items by geographic area. The sample contains much more information from larger population areas than smaller areas. Table 4 presents a frequency distribution of the length of the individual price series by category. As discussed earlier, under the BLS sampling scheme, an individual price series can be as long as 5 years. However, as seen in Table 4, only a small fraction of price series in our sample are 5 years long. In fact, most of the price series are less than 2 years in length for all product categories except ground beef, eggs, orange juice, and lettuce. According to the BLS, there are two reasons why many of our price series have relatively short lengths. The ﬁrst reason is that we obtained the same ten calendar years (1988–1997) of data for all cities. Because the BLS changes its sample of stores for 20% of its cities each year, 80% of the observations in the ﬁrst year Table 2. Description of Data Set By Product. Product Baby Food Bananas Canned Soup Cereal Cheese Snacks Cola Drinks Cookies Crackers Eggs Frozen Dinners Frozen Orange Juice Ground Beef Hotdogs Lettuce Margarine Paper Products Peanut Butter Soap and Detergents White Bread Total

Number of Price Series

Number of Observations

299 1142 1310 1631 1233 1288 1116 750 311 905 561 491 909 471 672 477 620 342 820 1043

6579 26284 26480 26603 27183 21654 19343 14125 6982 27915 7561 13703 27551 9594 25687 11826 7018 9188 10158 24663

16391

350097

Pricing Dynamics of Multiproduct Retailers

137

of our data are part of a series that began in a previous year. Hence, 80% of the observations for 1988 will be part of a price series that began outside of our sample period. Similarly, 80% of the observation for 1997 will be part of a price series that will conclude outside of our sample period. This means that for the 80% of 1988 observations that are parts of price series that began before 1988, the maximum series length will be 48 months, and for 60% of the observations the maximum series length will be 36 months, etc.

Table 3. Descriptive of Data Set by Region. Region Atlanta Boston Buffalo Chicago Cleveland Dallas Dayton Denver Detroit El Paso Greater Los Angeles Jacksonville Kansas City Los Angeles Miami Minneapolis New Orleans Suburbs of New York City Philadelphia Portland Richmond St. Louis San Diego San Francisco Scranton Seattle Syracuse Tampa Tucson Washington, D.C. Total

Number of Price Series

Number of Observations

361 570 317 1765 492 536 289 341 1069 323 557 297 374 1694 387 337 375 685 830 289 385 654 331 947 335 355 311 280 369 536

6547 11022 5866 40019 9730 10657 6733 6231 21404 7312 15682 7118 6033 35487 7116 6379 6812 17816 17270 5565 8102 13530 5556 25186 6752 6566 8577 5515 7658 11856

16391

350097

138

DANIEL HOSKEN, DAVID MATSA AND DAVID REIFFEN

The second reason is that if the BLS surveyor arrives at the store and cannot ﬁnd the exact product and package size of a particular item, she selects a new product in that category and creates a new price series. In the data set, it appears this is the primary reason why most of the price series are so short. For some of the product categories, e.g. canned soup or frozen dinners, this explanation seems plausible. These product categories have many different individual brands and package sizes, and it seems reasonable to believe that the life span of a randomly selected product is short. However, for more stable categories, e.g. cola drinks, we ﬁnd this explanation less credible. The two leading brands of cola (Coke and Pepsi) come in four different varieties (the permutations of with, and without, sugar and caffeine) that were on the market with a commanding market share throughout the sample period. It seems unlikely that changes in the product mix would result in 40% of the price series for cola drinks being less than one year in length. The unexpectedly short

Table 4. Sample Description: Frequency Distribution of Length of Time Series. Product

All Products Baby Food Bananas Canned Soup Cereal Cheese Snacks Cola Drinks Cookies Crackers Eggs Frozen Dinners Frozen Orange Juice Ground Beef Hotdogs Lettuce Margarine Paper Products Peanut Butter Soap and Detergents White Bread

Less than 1 year

1 to 2 years

2 to 3 years

3 to 4 years

4 to 5 years

5 years or more

37.8% 44.1% 23.6% 37.3% 51.5% 37.0% 45.3% 40.9% 43.9% 31.2% 19.0% 56.7% 26.5% 19.0% 40.3% 6.8% 32.1% 64.4% 28.4% 61.0% 34.6%

24.4% 17.4% 28.4% 30.5% 24.5% 23.1% 28.3% 25.7% 24.2% 28.6% 23.2% 24.4% 20.3% 23.4% 22.5% 17.9% 24.3% 22.2% 16.0% 23.6% 21.8%

15.7% 16.1% 26.4% 12.7% 10.1% 16.4% 12.8% 21.1% 15.1% 18.0% 16.3% 11.8% 16.7% 17.8% 18.1% 19.1% 14.2% 9.4% 22.6% 9.4% 17.4%

10.1% 11.0% 21.5% 9.1% 7.2% 8.7% 8.4% 10.8% 6.5% 9.3% 13.2% 4.8% 14.5% 13.5% 8.7% 15.4% 9.3% 2.0% 13.1% 2.2% 10.6%

8.8% 7.4% 0.1% 7.9% 5.2% 11.3% 4.7% 1.5% 7.8% 10.6% 19.5% 2.1% 15.1% 18.3% 8.9% 27.7% 15.9% 0.9% 13.2% 3.1% 11.8%

3.2% 4.0% 0% 2.5% 1.5% 3.5% 0.5% 0% 2.5% 2.3% 8.8% 0.2% 6.9% 8.0% 1.5% 13.1% 4.2% 0.6% 6.7% 0.6% 3.8%

Pricing Dynamics of Multiproduct Retailers

139

duration of many of the individual price series appears to be the major shortcoming of the BLS data set. However, while the short length of some of our price series weakens our ability to detect price changes, it should not induce any bias into our analysis. One ﬁnal weakness with the data that might affect our ability to detect sales is that prices are sampled monthly, whereas previous research suggests that sales last either one or two weeks and the ideal frequency of observation is weekly (see Hosken & Reiffen, or Pesendorfer, 2002). In a large sample, this should not affect the proportion of our observations that are sales, but it will reduce our ability to detect sales. The reason that sales are more difﬁcult to observe is only partially due to the reduced number of observations. A more fundamental problem arising from having less-frequent observations is that the retailer’s costs are more likely to change between observations than if the data were weekly. Thus, some of the price movements we detect may reﬂect wholesale price changes rather than sales.

V. EMPIRICAL METHODS AND RESULTS The purpose of this study is to identify some pricing regularities and relate them to the theories described in Section II. In particular, we wish to demonstrate the importance of sales, describe some of features of sale behavior, and examine the implications of this source of retail price variation on empirical studies that use retail prices. First, we document the extent to which products in our data have a “regular” price. We do this by calculating how often an individual product’s price is at its “typical” level. Speciﬁcally, we conduct the following calculation: we ﬁrst divide the data set into individual price series for each calender year (e.g. the tenth price series for crackers in Chicago for 1996). Then, for each annual price series, we calculate the modal price. Finally, for each category we determine the frequency with which prices in each individual price series are equal to their modal values. Table 5 presents summary statistics to characterize the extent to which products have a regular price. Speciﬁcally, it presents frequency distributions for each product category describing how often the prices in each individual time series are equal to their modal values for that year. With the exception of eggs and lettuce, the products’ prices are equal to their modal value at least 50% of the time. Furthermore, with the exception of eggs, lettuce, and bananas, more than 25% of products are at their modal prices at least 75% of the time. This evidence shows that over the course of a year, in spite of both sales and

140

DANIEL HOSKEN, DAVID MATSA AND DAVID REIFFEN

wholesale price changes, most products have a regular price and are priced at that level most of the time. The second aspect of the sale phenomenon is evaluating what happens when price is not at its regular level. If sales are important, then we would expect that when prices are not at their regular level, they are signiﬁcantly more likely to be below the regular price than above it. Hence, we test for sales by comparing the percentage of deviations from the modal price that are above versus below the mode for each type of product in our sample. This comparison demonstrates the relative importance of retail margin changes versus wholesale price changes in affecting retail prices. If retail prices only change as the result of permanent changes in wholesale prices, then we would expect the percentage of retail prices above the mode to be as large as the percentage of deviations below the mode.11 Conversely, a ﬁnding that when the price is not at Table 5. Summary of Frequency Distributions of How Often Price Quotes are at Their Modal Value.

Product Baby Food Bananas Canned Soup Cereal Cheese Snacks Cola Drinks Cookies Crackers Eggs Frozen Dinners Frozen Orange Juice Ground Beef Hot Dogs Lettuce Margarine Paper Products Peanut Butter Soap and Detergent White Bread

Percentage of Products at Modal Price less than 25% of the Time

Percentage of Products at Modal Price less than 50% of the Time

Percentage of Products at Modal Price more than 75% of the Time

Annual Price Series

0.3% 1.1% 0.1% 0.4% 0.7% 0.1% 1.4% 0.4% 0.4% 16.9% 0.1% 1.2% 0.8% 0.8% 73.2% 0.9% 0.5% 0.3% 0.6% 0.4%

10.1% 31.4% 15.0% 16.6% 21.4% 9.9% 25.7% 14.6% 20.3% 63.4% 13.4% 27.7% 28.7% 24.6% 84.6% 24.6% 12.7% 21.1% 11.2% 16.5%

50.2% 23.0% 43.5% 44.0% 42.7% 54.1% 42.3% 53.4% 37.7% 15.6% 51.7% 29.4% 32.1% 41.8% 2.3% 39.2% 49.1% 38.1% 52.2% 47.4%

745 3055 3231 3366 3138 2832 2444 1744 799 2877 1109 1463 2848 1116 2563 1300 1312 990 1889 2787

Pricing Dynamics of Multiproduct Retailers

141

its mode, it is generally below the mode suggests that retail price changes are primarily driven by retailer behavior. As seen in Table 6, for every category, prices below the mode are much more likely to occur than prices above the mode. In each product category, the difference between the number of downward deviations from the mode is higher than the number of upward deviations by a statistically signiﬁcant amount. Thus, the data suggest that sales are an important cause of retail price variation for a wide variety of goods sold by retailers. As discussed above, it is likely that this result would only be enhanced if we had data at weekly, rather than monthly, intervals. In this paper we focus on differences in sale behavior across products. To examine these differences, we must ﬁrst operationalize the idea of a sale as a signiﬁcant temporary reduction in the price of a retail item. We do this by saying that a sale occurs if a product’s price falls by some ﬁxed amount in a given month and then rises by a similar amount in the next observed month.12 Tables 7 and 8 present some general facts about the prevalence of sales in our sample. In Table 7, we see that the likelihood that an item is placed on sale varies widely across categories of products. For instance, the baby food products sampled almost never go on sale, while colas and crackers go on sale fairly frequently. We note that, consistent with Hosken & Reiffen, the two most perishable products, bananas and lettuce, appear to be the most likely to go on sale using our deﬁnition. However, this conclusion requires some caution, as some of the “sale” behavior we detect is likely due to the greater volatility of the underlying wholesale prices for these items resulting from seasonal variation in output. There also appear to be variations in sale behavior across the U.S. To create our measure of the probability of observing a sale in each geographic region, we restricted our attention to the ten items sampled in each of the thirty regions in our data set.13 To calculate the averages presented in Table 8, we ﬁrst calculate the average probability of a sale for each of the ten product categories. We then take the simple average of those category probabilities to construct the probability of observing a sale in a region. The differences in sale behavior vary signiﬁcantly across the U.S. For example, in the Miami area the probability of seeing a sale of at least 10% was about 0.05 versus 0.09 in the Chicago area. Further, these cross-sectional differences in sale behavior seem to be robust to the exact sale deﬁnition used. While the evidence presented thus far suggests that sales are an important cause of variation in retail prices, it is not clear that sales are an important cause of price variation for all retail products. Of particular interest is the prediction of the Hosken, Matsa and Reiffen model that popular products (deﬁned as

142

DANIEL HOSKEN, DAVID MATSA AND DAVID REIFFEN

Table 6. Percentage of Prices Above and Below the Annual Modal Price By Product. Product Baby Food Bananas Canned Soup Cereal Cheese Snacks Cola Drinks Cookies Crackers Eggs Frozen Dinners Frozen Orange Juice Ground Beef Hotdogs Lettuce Margarine Paper Products Peanut Butter Soap and Detergents White Bread a b

Percentage Above Modea

Percentage Below Modea

Z-Statisticb

9.5% (592) 14.0% (3371) 10.5% (2615) 11.6% (2885) 12.8% (3238) 7.0% (1453) 10.5% (1872) 7.8% (1049) 7.8% (516) 25.6% (5795) 7.8% (552) 12.3% (1560) 11.8% (2996) 10.2% (908) 18.2% (4206) 11.1% (1222) 9.2% (602) 11.5% (984) 8.7% (832) 10.6% (2462)

16.6% (1032) 28.2% (6791) 20.3% (5043) 20.3% (5038) 19.7% (4986) 17.2% (3581) 23.5% (4184) 18.6% (2491) 25.7% (1699) 32.4% (7346) 21.6% (1531) 27.5% (3479) 25.6% (6480) 24.3% (2170) 65.0% (15007) 23.4% (2576) 22.3% (1454) 22.2% (1904) 20.8% (1996) 18.0% (4183)

3.95 (0.0000) 15.88 (0.0000) 10.81 (0.0000) 9.85 (0.0000) 8.15 (0.0000) 9.40 (0.0000) 11.80 (0.0000) 8.09 (0.0000) 8.66 (0.0000) 8.55 (0.0000) 7.24 (0.0000) 11.86 (0.0000) 15.22 (0.0000) 8.92 (0.0000) 53.84 (0.0000) 8.95 (0.0000) 6.94 (0.0000) 7.03 (0.0000) 7.79 (0.0000) 8.11 (0.0000)

Number of observations in parentheses. P-values in parentheses.

Pricing Dynamics of Multiproduct Retailers

143

products consumed by a large proportion of consumers) are placed on sale most frequently. Because it is difﬁcult to operationalize this concept of popularity, we instead use several indirect tests of their hypothesis. The ﬁrst test turns on the observation that within each category, if popularity is stable over time, then the same goods will be on sale repeatedly. Hence, if one observes product A on sale in period t, the theory implies that product A is (in expectation) more popular than product B (where both goods are in the same category). Therefore, we would expect product A to be on sale more often than product B in future periods. Conversely, if products are randomly chosen to be put on sale (say due to unexpectedly high inventories for particular goods), sales would be equally likely on all products. Thus, we wish to compare the theory’s implication that the probability of observing a sale in subsequent periods is higher for products that had a sale in the base periods to the null hypothesis that the probability of observing a sale on a particular product in subsequent periods is independent of whether that product was on sale in a base period.

Table 7. Probability of Sale by Product Category. Product Baby Food Cereal Canned Soup Peanut Butter Cheese White Bread Soap and Detergent Paper Products Margarine Cookies Eggs Snacks Frozen Orange Juice Ground Beef Frozen Dinner Hot Dogs Cola Crackers Bananas Lettuce For All Products

Observations

5% Sale

10% Sale

15% Sale

20% Sale

25% Sale

5670 22193 22655 8197 23227 21247 4180 2936 10415 11844 25009 17596 12175 24946 5834 8053 16581 5989 23306 23101 295154

0.0219 0.0463 0.0516 0.0661 0.0619 0.0585 0.0730 0.0811 0.0825 0.0881 0.1111 0.0802 0.0881 0.1080 0.0921 0.0929 0.0977 0.1186 0.1455 0.2321 0.0959

0.0144 0.0303 0.0347 0.0378 0.0449 0.0511 0.0522 0.0586 0.0587 0.0628 0.0648 0.0684 0.0702 0.0711 0.0754 0.0777 0.0794 0.0945 0.1378 0.1896 0.0736

0.0083 0.0232 0.0235 0.0221 0.0215 0.0392 0.0349 0.0351 0.0396 0.0382 0.0366 0.0551 0.0536 0.0476 0.0614 0.0596 0.0589 0.0725 0.1160 0.1523 0.0547

0.0048 0.0179 0.0155 0.0138 0.0223 0.0291 0.0239 0.0191 0.0293 0.0237 0.0263 0.0451 0.0447 0.0322 0.0425 0.0468 0.0431 0.0533 0.0983 0.1224 0.418

0.0025 0.0134 0.0104 0.0083 0.0133 0.0204 0.0144 0.0116 0.0204 0.0141 0.0197 0.0299 0.0313 0.0208 0.0255 0.0340 0.0286 0.0359 0.0826 0.0967 0.0306

144

DANIEL HOSKEN, DAVID MATSA AND DAVID REIFFEN

We test this hypothesis as follows. For every price series longer than 2 years, we record whether that price series experienced a sale during the ﬁrst twelve months for which we have data. We then divide the sample into two parts: the ﬁrst contains price series that have a sale in the ﬁrst twelve months and the second contains those price series that do not have a sale. Within each product category we then calculate two conditional probabilities: the probability that a price series would experience a sale during the second year of the sample, conditional on the product being in the ﬁrst group (i.e. having a sale within the ﬁrst 12 months), and the probability of a sale in the second year conditional on Table 8. Probability of Sale by Region. Region Atlanta Boston Buffalo Chicago Cleveland Dallas Dayton Denver Detroit El Paso Greater Los Angeles Jacksonville Kansas City Los Angeles Miami Minneapolis New Orleans Suburbs – New York City Philadelphia Portland Richmond St. Louis San Diego San Francisco Scranton Seattle Syracuse Tampa Tucson Washington, D.C.

Observations 5% Sale 10% Sale 15% Sale 20% Sale 25% Sale 3759 7475 3547 25088 6263 6715 4758 3435 13406 5307 9788 4513 3705 22456 4726 3419 4053 11716 10342 38531 5127 3408 15281 4245 3717 7934 5771 3376 5007 7671

0.0922 0.0942 0.0905 0.1134 0.0976 0.0992 0.0817 0.0979 0.0901 0.0878 0.1067 0.0945 0.1069 0.1040 0.0683 0.0994 0.0768 0.0953 0.0960 0.1145 0.1049 0.0879 0.1083 0.1117 0.0879 0.1068 0.0979 0.0888 0.1185 0.0877

0.0697 0.0750 0.0680 0.0937 0.0766 0.0754 0.0658 0.0725 0.0710 0.0654 0.0791 0.0700 0.0842 0.0795 0.0513 0.0715 0.0559 0.0749 0.0805 0.0888 0.0859 0.0672 0.0781 0.0845 0.0687 0.0844 0.0810 0.0625 0.0891 0.0639

0.0505 0.0596 0.0491 0.0725 0.0583 0.0543 0.0510 0.0559 0.0552 0.0490 0.0581 0.0512 0.0632 0.0598 0.0372 0.0509 0.0424 0.0564 0.0649 0.0717 0.0637 0.0524 0.0579 0.0620 0.0492 0.0677 0.0670 0.0453 0.0603 0.0439

0.0368 0.0468 0.0401 0.0587 0.0459 0.0429 0.0391 0.0410 0.0438 0.0343 0.0422 0.0438 0.0478 0.0471 0.0294 0.0359 0.0292 0.0427 0.0528 0.0581 0.0482 0.0423 0.0451 0.0449 0.0335 0.0495 0.0544 0.0337 0.0434 0.0333

0.0257 0.0357 0.0304 0.0451 0.0357 0.0320 0.0288 0.0340 0.0301 0.0245 0.0323 0.0322 0.0337 0.0364 0.0198 0.0270 0.0209 0.0314 0.0383 0.0465 0.0344 0.0312 0.0328 0.0319 0.0238 0.0381 0.0412 0.0240 0.0330 0.0236

Pricing Dynamics of Multiproduct Retailers

145

being in the second group. We then test the null hypothesis that the conditional probability of observing a sale in the second 12-month period is the same for both groups. The results appear in Table 9.14 For every product category in our sample the conditional probability of observing a sale is larger, often substantially larger, if the price series experienced a sale within the ﬁrst 12 months. In fact, in 19 of the 20 hypotheses tests listed, we reject the null hypothesis with a z-statistic greater than 2.5.15 For example, of the 62 cereal price series that experienced a sale of at least 15% within their ﬁrst 12 months in the sample, 50.0% experienced at least one additional 15% sale in the second 12 months of the sample period, while only 18.6% of the 274 price series that did not experience a sale within the ﬁrst 12 months experienced at least one 15% sale in the second 12 months. The difference in these conditional probabilities is different from zero at any conventional level of statistical signiﬁcance (z = 5.20). We interpret this as strong evidence that there is substantial heterogeneity across products in the likelihood of having a sale. Within categories, retailers appear to systematically place some products on sale more often than others. This result is robust across 20 large categories of goods, over time, across the U.S. and for ﬁve different deﬁnitions of sales (–5%, 10%, 20%, and 25%, as well as the 15% reported here). Unfortunately, the BLS data do not allow us to relate product characteristics (e.g. a product’s market share) to the probability of going on sale. However, the data suggest that products differ widely in the frequency with which they are put on sale. This result is consistent with the prediction that more popular products (deﬁned as those being consumed by a larger proportion of consumers) should go on sale more often. However, because we cannot know individual product identities in the BLS data, we do not know the relative popularity of products, and this result is a rather indirect test of this hypothesis. A somewhat more direct test takes advantage of the fact that some goods become more popular at certain times of the year. The theory predicts that as a product becomes more popular, it becomes more likely to be put on sale. The hypothesis we test is that in each of these categories, the frequency of sales rises during the high-demand period. Of the twenty products in our sample, we identify ﬁve which have predictable seasonal increases in demand. The demand for soup increases in the fall and winter (October thru March), peanut butter demand increases as part of back to school planning in August and September, egg demand increases around Easter, and ground beef and hot dog demand increases during the summer barbeque season (June, July and August). Further, because the costs of producing these items are not seasonal, we are reasonably conﬁdent that any change in sale behavior is a result of retailers’ reactions to changes in demand

146

DANIEL HOSKEN, DAVID MATSA AND DAVID REIFFEN

Table 9. Percent of Price Series Experiencing at Least One Sale of 15% or more in the Second Year, Conditional on Whether there is a Sale within the First Year. Product

Baby Food Bananas Canned Soup Cereal Cheese Snacks Cola Drinks Cookies Crackers Eggs Frozen Dinners Frozen Orange Juice Ground Beef Hot Dogs Lettuce Margarine Paper Products Peanut Butter Soap and Detergent White Bread

Conditional On at Least One Sale within the First Year (no. of price series)

Conditional on No Sale within the First Year (no. of price series)

Z-statistic (p-value)

80.0% (5) 78.2% (367) 43.4% (76) 50.0% (62) 51.9% (106) 61.2% (103) 59.7% (124) 69.8% (43) 72.9% (48) 49.7% (157) 59.5% (42) 62.8% (94) 60.6% (175) 63.1% (65) 90.6% (383) 64.3% (56) 66.7% (9) 32.3% (31) 54.5% (22) 54.2% (131)

2.2% (92) 49.6% (121) 13.7% (299) 18.6% (274) 14.8% (290) 25.6% (172) 23.9% (155) 16.3% (135) 23.2% (56) 23.3% (305) 26.2% (42) 28.5% (137) 31.4% (287) 31.5% (74) 75.7% (74) 25.2% (127) 35.9% (39) 7.8% (129) 16.7% (42) 15.0% (253)

7.04 (0) 6.02 (0) 5.81 (0) 5.20 (0) 7.56 (0) 5.86 (0) 6.07 (0) 6.71 (0) 5.08 (0) 5.75 (0) 3.09 (.0022) 5.18 (0) 6.16 (0) 3.29 (.0012) 3.64 (.0002) 5.04 (0) 1.69 (.0910) 3.70 (.0002) 3.15 (.0016) 8.07 (0)

Pricing Dynamics of Multiproduct Retailers

147

Table 10. Probability of Sale for Various Products in Relatively High and Low Periods of Demand: Sale = 15% reduction.

Product Ground Beef Hot Dogs Eggs Canned Soup Peanut Butter

Probability of Sale in High Demand Period

Probability of Sale in Low Demand Period

Z-Statistic for difference in Probability

0.06422 0.07245 0.06926 0.02677 0.03228

0.04515 0.05701 0.03529 0.01876 0.01897

4.10 1.75 2.78 4.25 3.27

rather than supply. The results of these tests are presented in Table 10. The results strongly support the theoretical analysis. We see for both of the sale deﬁnitions considered, retailers are more likely to put these items on sale in periods of high demand, and that these differences are statistically signiﬁcant in virtually all cases, at any conventional signiﬁcance level. Thus, our data suggest that retailers systematically lower the prices of items that experience increases in demand. While these results are not surprising to anyone who shops in a grocery store, the Hosken, Matsa and Reiffen analysis provides an explanation for this phenomenon: A retailer attracts a consumer by offering more consumer surplus than its rival. Because it is costly for a retailer to inform consumers of the price of any individual item, other things equal, the least costly way for retailers to assure a given level of surplus to the largest number of consumers is to put items on sale that are attractive to the widest audience possible. Hence, when products have known upward spikes in demand, we would expect retailers to ﬁnd it more attractive to put these items on sale. Using the BLS data we have seen that products appear to have a regular price and that most deviations from that price appear to be sales. There is also substantial heterogeneity across products in the likelihood a retailer puts the product on sale. Within each product category, some products are far more likely to go on sale than others. Finally, we have seen some evidence that suggests products are more likely to be put on sale when they are more popular, e.g. eggs at Easter.

V. DISCUSSION As noted above, the increased availability of store and market-level retail pricing data has led to an increase in research based on that data. Two important

148

DANIEL HOSKEN, DAVID MATSA AND DAVID REIFFEN

strands of this research are product-speciﬁc demand elasticity studies, and studies of the relationship between retail market structure and pricing. Our analysis of the BLS price data yields some important insights into retail pricing behavior that affect the interpretation of the results of such research. Products appear to have a regular price level that is maintained through relatively long periods of time. Relatedly, a subset of products are periodically temporarily discounted from this everyday price. These temporary discounts; i.e. sales, are an important component of retail price variation, and are typically retailerspeciﬁc.16 The fact that retailers are not setting prices by simply charging a ﬁxed markup on their wholesale costs has implications for studies that wish to use retail prices to estimate demand elasticities for consumer products for at least two reasons. The ﬁrst reason relates to correctly measuring the prices and quantities used in studies that measure consumer demand elasticities. Most studies appear to use highly aggregated price and quantity data (either for a region, such as the northeastern U.S., or a metropolitan area). However, because retailers within a market will often be charging different prices for the same item in a market at a point in time (e.g. if one retailer is offering the item on sale), the average price charged in a market for an item will be a poor measure of the actual price consumers face. Thus, to avoid this type of measurement error, a researcher should collect the price and quantity data for consumer products at the level of the speciﬁc retailers in a city.17 Unfortunately, there are also problems with using retailer-speciﬁc price and quantity data. For example, if many consumers choose to buy a speciﬁc item at the lowest price retailer in the city, then a retailer having a sale on that item (e.g. gallons of milk) may see its sales of that product surge, even if sales for the item in that city are barely affected. In this case, using retailer speciﬁc data can result in elasticities being overestimated.18 While both types of data have measurement problems, the prudent approach by researchers would be to use both types of data in a study to check the robustness of their results. The second, and more problematic reason results from retailers using sales to both compete with one another and to intertemporally price discriminate. As discussed earlier, Sobel, Pesendorfer and Hosken and Reiffen each develop models that show that retailers who are selling goods that can be inventoried by consumers (e.g. non-perishable goods) have incentives to charge different prices over time to price discriminate against consumers who have high inventory costs. In these models, in each period in which the retailer does not have a sale, the demand curve for the next period shifts further to the right. Thus, the strategic behavior by the retailer in period t affects the level of the demand curve in period t + 1. Existing empirical evidence suggests that retailers

Pricing Dynamics of Multiproduct Retailers

149

are currently engaging in such price discrimination, validating our concern. For example, in Pesendorfer’s study of the demand for ketchup he ﬁnds that the current period demand for ketchup is a function of lagged prices. In addition, it is well known that retailers sell very large quantities during sales that would imply unrealistic demand elasticities in the absence of these consumer inventory effects.19 When intertemporal price discrimination by retailers is an important source of retail price variation, the elasticities calculated using contemporaneous price and quantity data measure how much observed purchasing increases when a store has a sale, not what the level of purchases in a time period would be if the retail price of the item permanently lowered its price to a given level. In order to determine if a merger will induce the merged ﬁrm to increase price following a merger, a researcher needs to know how consumers would respond to a permanent change in the price distribution (regular price, sale price, and frequency of sale) of a good. It is for this reason that the demand elasticities estimated using contemporaneous price and quantity data provide limited information to policy makers. There are similar problems in using retail prices to measure the effect of competition between retailers. Retailers use different strategies to price the different items they sell. For example, grocery chains report that consumers are very sensitive to the prices charged for the most popular products they sell (e.g. leading brands like Tide or Cheerios, or frequently purchased non-branded products like milk and ground beef). Because these products are frequently purchased by consumers and sold by a large number of retailers, consumers have a good sense of what the rival retailers typically charge for these items. For example, grocery store operators try to very closely peg their prices on these “price sensitive” items to their rivals in order to maintain their price image in a market. If a grocery chain’s milk prices deviate from their typical level, the chain stands to lose signiﬁcant business. In contrast, consumers are less aware of the prices of items they purchases less frequently (e.g. pancake mix or canned corn). Consequently, retailers face less of a penalty if their prices on these items are relatively high. These strategies appear to carry over to a retailer’s decision of which items to put on sale. As discussed earlier, by offering consumers a sale, retailers commit to offering consumers low prices on a set of items. Because advertising a sale is costly, retailers want to advertise the items that are most likely to attract consumers to the store; that is, retailers will advertise relatively popular items. Because retailers pursue different pricing strategies for different kinds of goods, we believe that the (quantity-weighted) average price of a bundle containing all goods sold by a supermarket is unlikely to be the most useful tool

150

DANIEL HOSKEN, DAVID MATSA AND DAVID REIFFEN

for evaluating competition. Retailers face the strongest incentives to lower prices on the products that will bring them the most customers. Thus, we expect that the price of popular, frequently purchased, items would be most affected by retail competition. Similarly, competition may also manifest itself through the type of sales retailers offer consumers (how many items, depth of discounts) in addition to the regular price level. Recall, in Sobel’s model of retailer competition, the number of rivals in a market does not affect the “regular” price, it only affects the frequency of sale and depth of sale. Further, because consumers appear to anticipate sales, a disproportionate number of consumer purchases take place during sales. Thus, the consumers who may be harmed by a merger (the removal of a rival from a market) may not be the consumers who purchase at the regular price, but the price sensitive consumers who time their purchases to coincide with sales. For these reasons, we recommend that researchers analyze the frequency and depth of sales to measure the effects of competition (or changes in competition). Further, when examining price levels, researchers are more likely to observe the effects of retail price competition on the most popular items retailers sell.

VI. CONCLUSIONS Food retailers adjust their prices in ways that are often bewildering to economists. This paper attempts to broaden our understanding of retail pricing dynamics by providing some systematic evidence about U.S. grocery prices. Using a large data set containing information on twenty categories of goods from thirty U.S. metro areas for the period 1988–1997, we ﬁnd a number of empirical regularities. First, for each of twenty categories of goods in our BLS sample, stores seem to have a “regular” price, and most deviations from that price are downward. Second, we ﬁnd there is considerable heterogeneity in sale behavior across goods in each category; within each category of goods, the same items are regularly put on sale, while other items are rarely, if ever, put on sale. Third, the probability of a sale on an item appears to be greatest when demand for that item is highest. While many aspects of retailer behavior are beyond the scope of this paper, our data do reveal some patterns in their decision-making. For example, retailers appear to pursue different pricing strategies for different grocery items. In particular, it appears that more popular products are more likely to go on sale. In addition, the evidence we have found is consistent with models that predict retailers will periodically offer items at a reduced price to intertemporally price discriminate against consumers with high inventory costs.

Pricing Dynamics of Multiproduct Retailers

151

These results suggest that retailers do not play a passive role in bringing ﬁnal goods to market (e.g. simply marking up the price of all goods they sell a ﬁxed amount or percentage). This observation has important implications for interpreting results based on retail pricing data. For instance, most of the variation in retail prices that is used to identify elasticities in studies of demand for speciﬁc consumer products will be the result of retail sale behavior, not changes in wholesale costs. If retailers are using sales to intertemporally price discriminate, then these estimated demand elasticities will not measure the true relationship between consumer prices and consumption, but will instead measure how consumer purchases change in response to predictable short term decreases in retail prices. The elasticity that is of interest to decision makers (how much consumer consumption will fall if all prices increase) requires the researcher to observe a shift in the entire distribution of consumer prices, sale and non-sale, (such as would occur following a changes in wholesale prices). Similarly, studies that use retail prices to measure the relationship between consumer prices and competition should be careful in choosing which retail prices to analyze. For example, in Sobel’s model changes in the number of market participants only affect product’s sale prices, not their regular prices. Thus, if only popular items go on sale, it is conceivable that only popular items will be affected by changes in the level of competition.

NOTES 1. Lal and Matutes (1989) use a similar explanation for competing multiproduct retailers using different (static) pricing strategies for their array of goods. In their model, each retailer has a low price on a different good, which causes low transportation cost consumers to buy at more than one store each period, but allows the retailers to charge high prices on some items to high transportation cost/high reservation value consumers. Banks and Moorthy (1999), show that coupons can be another way of offering low prices to low reservation price/low search cost customers, while maintaining high prices to high reservation price/high search cost consumers. 2. Pesendorfer (2002) generalizes Sobel’s analysis by introducing a third type of consumer; store-loyal, but low-value. 3. In contrast to the monopoly retailer case, with competing retailers the probability that a sale may occur becomes positive as soon as the expected proﬁt from selling to the accumulated low-value consumers at a low price equals the proﬁt from selling to the loyal consumers at their higher reservation value. 4. In the Lal and Matutes (1994) model consumers realize that when they go shopping they will purchase a bundle of products (some advertised and some unadvertised) and choose the retailer (or retailers) that will sell them that bundle at the lowest cost. They understand that the unadvertised products will be sold at relatively high prices, and incorporate this information into their decision making process. In their

152

DANIEL HOSKEN, DAVID MATSA AND DAVID REIFFEN

model advertising is, in essence, a commitment device that keeps the retailer from charging the consumer high prices on all of its products once she has sunk the transportation costs of visiting the store. 5. A category is a fairly narrow classiﬁcation of consumer goods, e.g. cola drinks, eggs, and white bread are BLS categories. 6. Some of the price series have lengths longer than 5 years because the BLS collected an additional year of data for the regions that were rotated out in 1997 for the update of the CPI. 7. The BLS does maintain the information on the speciﬁc retailer and product surveyed, however, for conﬁdentiality reasons they cannot release this data. 8. For this reason, we cannot use the BLS data to examine any implications regarding the relationship of prices movements on products within a store. 9. The BLS occasionally updates the sampling scheme it uses to collects consumer prices. We choose to collect data from the time period 1988–1997 because the BLS used the same sampling scheme throughout the time period. 10. The BLS imputes prices for missing values. However, because the goal of our study was to study the way actual consumer prices changed over time, we deleted all of the imputed prices from our data set, roughly 5% of the observations. 11. Given there is upward trend to pricing due to inﬂation, other things equal we would expect most wholesale price changes to be increases in price. Implicitly in this analysis, we assume there is no systematic pattern in wholesale price changes, e.g. manufacturers changing prices every March. 12. We consider ﬁve different levels of price reductions in our deﬁnition of a sale, discounts of at least – 5%, 10%, 15%, 20%, and 25%. 13. The products included are bananas, canned soup, cereal, cheese, chips, cola drinks, eggs, ground beef, lettuce, and white bread. 14. We consider the ﬁve different minimum price decreases in our sale deﬁnition (–5%, 10%, 15%, 20%, and 25%). In the interest of brevity, only the results for the 15% deﬁnitions are presented here. 15. The corresponding number of z-statistics over 2.5 using all 5 sale deﬁnitions was 91 out of 100. Note that for some of the comparisons of conditional probabilities, the number of price series is very small. In these cases it is incorrect to assume that the difference in proportions is approximately normal, and instead we simply interpret the computed z-statistics as measures of the size of the difference between conditional probabilities. 16. While it is true that retailers often receive promotional allowances from manufacturers to subsidize a sale, our understanding is that these promotions are offered simultaneously to all retailers in a geographic area, and that the individual retailer decides how much of thepromotional discount will be passed thru to consumers. In addition, Hosken and Reiffen show that price changes across retailers are not correlated. The lack of correlation in changes in retailer prices suggests that individual retailers play an important role in both choosing the timing and depth of sales. 17. If demand curves are linear, this measurement problem should not be an issue, however, in general it would be. 18. This problem is probably not insurmountable. For example, by controlling for which ﬁrm has the lowest price on a particular item at a point in time (e.g. with an intercept shifter), it may be possible to control for this store switching effect.

Pricing Dynamics of Multiproduct Retailers

153

19. For example, a study described by Progressive Grocer (December 1992) found that the quantity of soda sold increased between 600% and 800% and that the quantity of ﬂour sold increased between 940% and 1800% during a sale.

ACKNOWLEDGMENTS The opinions expressed in this paper are those of the authors, and not necessarily those of the Federal Trade Commission or any of its individual Commissioners. We would like to thank Steve Scutt for his assistance in putting together the data set, and Sara Harkavy and Morgan Long for providing excellent research assistance. We would also like to thank Cindy Alexander, Jim Ferguson, Charles Thomas and Aileen Thompson for their helpful comments on previous drafts.

REFERENCES Ashenfelter, O., Ashmore, D., Baker, J. B., & McKernan, S-M. (1998). Identifying the ﬁrmspeciﬁc cost pass-through rate. Washington, D.C.: Federal Trade Commission, Bureau of Economics Working Paper 220. FTC. Banks, J., & Moorthy, S. (1999). A model of price promotions with consumer search. International Journal of Industrial Organization, 17, 371–398. Conlisk, J., Gerstner, E., & Sobel, J. (1984). Cyclic pricing by a durable goods monopolist. Quarterly Journal of Economics, 99, 489–505. Hosken, D., & Reiffen, D. (2001). Multiproduct retailers and the sale phenomenon. Agribusiness, 17, 115–137. Hosken, D., Matsa, D., & Reiffen, D. (2000). How do retailers adjust prices? Evidence from store level data. Washington, D.C.: Federal Trade Commission, Bureau of Economics Working Paper 230. FTC. Lal, R., & Matutes, C. (1994). Retail pricing and advertising strategies. Journal of Business, 67, 345–370. Lal, R., & Matutes, C. (1989). Price competition in multimarket duopolies. Rand Journal of Economics, 20, 516–537. Pesendorfer, M. (2002). Retail sales: A study of pricing behavior in super markets. Journal of Business, forthcoming. Sansolo, M. (1992). The real power of promotion. Progressive Grocer, (December), 36–41. Sobel, J. (1984). The timing of sales. Review of Economic Studies, 51, 353–368. Varian, H. R. (1980). A model of sales. American Economic Review, 70, 651–659.

This Page Intentionally Left Blank

PRODUCT INNOVATION IN SERVICES: A FRAMEWORK FOR ANALYSIS Roger Betancourt and David Gautschi ABSTRACT Services now play a remarkably prominent role in modern economies. Not surprisingly, economists and marketing researchers have begun to turn their attention to the analysis of activities in the so-called tertiary sector. In this chapter we attempt to contribute to the effort of systematizing the analysis of service institutions by integrating perspectives from the economic analysis of institutions and property rights, on the one hand, and the economic analysis of retailing and distribution services, on the other hand. In so doing, we propose a set of evaluative criteria to be applied to the assessment of the evolution of service institutions, as well as a tableau for analyzing the emergence of various institutional forms. The tableau organizes the three primitive economic activities of production, distribution, and consumption on temporal and spatial dimensions. As such, the tableau applies the notion of relational constraints that have the property of reducing uncertainty and transaction costs, thus being welfare enhancing. The tableau and the evaluative criteria enable us to explore a range of issues, such as joint-ness of production and consumption, divided ownership of property rights, and the effects of technological progress that are inherent in the process of product innovation in services.

Advertising and Differentiated Products, Volume 10, pages 155–183. Copyright © 2001 by Elsevier Science Ltd. All rights of reproduction in any form reserved. ISBN: 0-7623-0823-0

155

156

ROGER BETANCOURT AND DAVID GAUTSCHI

I. INTRODUCTION a relevant analytical process cannot be divorced from purpose and, consequently, is itself a primary notion – that is, a notion that may be clariﬁed by discussion and examples but never reduced to other notions by a formal deﬁnition (N. Georgescu-Roegen, The Entropy Law and the Economic Process, 1971, p. 213).

We are living through a moment in history characterized by rapid technical progress and institutional change on a grand scale. For instance, the term information revolution is applied to the current period and comparisons are usually made with major innovations in history such as the printing press, e.g. Drucker (1998). Major institutional changes in terms of economic and political integration (the European Union) and disintegration (the Soviet Union) have taken place in the last two decades. During the past one hundred years, the transformation of the economic activities deﬁning the American economy has been profound. An important aspect of this transformation has been the increasing importance of services in economic activity. For example, in 1960 personal consumption expenditures on services were 25.8% of U.S. GDP and by 1999 this percentage had increased to 39.5%, U.S. Statistical Abstract (2000, Table 715). This pattern is not uniquely American, as it is widely acknowledged that the growth of modern developed economies appears to a large extent to be founded on the expansion of the so-called service or tertiary sector, for example Inman (1986) or Griliches (1992). While the attention of economists and marketing scientists has been progressively directed to the analysis of research questions germane to this sector, a dominant paradigm or framework has yet to emerge in order to discipline or systematize inquiry, thereby to establish convention. Some contributions exist, however, that point to components of such a framework. Nordhaus (1997) proposes and implements a version of the hedonic method for evaluating the services of light throughout history. Oi (1997) discusses the evaluation of major product innovations such as the air conditioner, but does not commit himself to the use of a particular procedure. Wernerfelt (1994) proposes an efﬁciency criterion for evaluating marketing designs that requires taking account of any impacts on the pay-off functions of the agents involved, including agents adjacent to the channel where the design is introduced. The property rights literature, for example Barzel (1997), the transaction costs literature, for example Williamson (1985), and more generally the new institutional economics literature, for example Eggertsson (1990), provide a broad basis for additional components. In this paper we integrate these components in an attempt to move toward a framework for the analysis of service institutions.

Product Innovation in Services: A Framework for Analysis

157

From the perspective of a ‘consumer’1, a service can be viewed as the dual of work that the ‘consumer’ might otherwise conduct. This means that if a market service reduces work for the ‘consumer’, then the institution providing the market service is likely to be sustained so long as it is not unproﬁtable for it to do so. Whether or not the institution is efﬁcient is more difﬁcult to determine.2 If one limits the analysis to the set of alternative institutional arrangements between a provider and a ‘consumer’, then Wernerfelt’s efﬁciency criterion can be applied. More generally, accounting for the system of economic activities that contribute to the ultimate fulﬁlment of some consumption aim, this criterion may be difﬁcult to apply in practice. Viewing a market service – as distinct from a market good – as a commodity supplied purposively to limit or restrict consumption activities, we direct our attention throughout the paper to the level of the institution, rather than to the level of the product or the ﬁrm. We subscribe to the deﬁnition that an institution is a set of constraints imposed on human interaction, for example, Nabli and Nugent (1989) or North (1990). Thus, a service institution is a set of constraints imposed on human interactions for the purpose of reducing work that ‘consumers’ would otherwise conduct. Since some industries are characterized as service industries, whereas others are not, it would be useful to provide some guide to their classiﬁcation for purposes of the analysis of any specialized service institution. In most instances, a producer has the option of providing either a good or a service to a consumer. This is true of personal services, as a barber or a doctor, for example, could choose to sell the instruments of haircuts [scissors, razor blades, hair dryers, and hair preparations] and medical care [stethoscopes, tongue depressors, thermometers, and medicine] to their customers rather than the haircut or the medical treatment, respectively. This is also true of capital intensive or extremely technical contexts. For example, Boeing could decide to operate airports and air transportation services connecting airports rather than to sell airplanes. In all of these cases, what is consumed can ultimately be simpliﬁed to some kind of service so that the consumption activity can be relatively clearly identiﬁed. Care must be exercised, however, in determining what constitutes the production activity. Undoubtedly, the determination will depend on the context of the analysis. If Boeing is included in the economic context, then the production activity is bounded by culmination of the creation of an airplane not a ﬂight. If Moe the barber or Dr. Welby, M.D. is included in the analysis, then the relevant production activity is the service rendered in either instance and not the creation of the instruments used in the performance of the service. To be sure, economic agents may participate in both kinds of production activities,

158

ROGER BETANCOURT AND DAVID GAUTSCHI

to wit, General Motors produces cars to which it transfers title to consumers and General Motors rents or leases cars to consumers who engage in limited consumption activities. Thus, the analysis of service institutions in terms of productive activities will be context-speciﬁc and, thereby, difﬁcult to use for classiﬁcation purposes. Our analysis of service institutions relies on the concept of property rights: namely, the ability to consume, to earn income from, and to exchange assets. Property rights in an economic sense are conceptually distinct from the legal right of ownership or the legal right to use, both typically enforced through legal contracts. For instance, any service not fully charged for on the margin, such as cool air in an air-conditioned store, is a residual that is at least partly relinquished to the public domain. Because measurement and enforcement costs combine to make it difﬁcult to delineate property rights completely, wealth maximizing individuals will devote resources to the capture of such residuals. Recognizing the existence of measurement and enforcement costs, service providers will seek novel formats for maximizing proﬁts that restrict consumers in some respects while allowing consumers opportunities for capture in other respects. We proceed below, in Section II, by drawing upon the notions of household production, e.g. Becker (1965), distribution services, e.g. Betancourt and Gautschi (1988), and the economic analysis of property rights, e.g. Barzel (1997), to generate a framework for the evaluation of product innovation in services through the introduction or elimination of service institutions. The household production model provides optimizing mechanics and distribution services provide a link to the analysis of property rights and to the outputs of service institutions. The integration of these concepts leads to a mechanism similar to that proposed by Nordhaus (1997) for evaluating the beneﬁts of a particular service institution. Furthermore, the integration suggests the identiﬁcation of three primitive economic activities: production, distribution, and consumption. In Section III we propose a novel tableau, separating these primitive economic activities across space and time, that helps systematize our analysis. Among other things the tableau brings out an ignored characteristic of the constraints deﬁning service institutions: namely, they can enhance welfare by reducing uncertainty. Incidentally this characteristic of constraints is not new in economics, since it also appears in the industrial organization literature on vertical restraints, the game theory literature on commitment and the macro literature on coordination failures. What is different in this context is the pervasiveness with which it applies in the tableau. This pervasiveness is explained in terms of the relational constraints of information theory.

Product Innovation in Services: A Framework for Analysis

159

Employing the tableau to investigate the nature of different service institutions impels us to confront a range of issues commonly associated with services and that lend a peculiar status to services as distinct from goods. To wit, in Section IV we demonstrate that the often assumed property of joint-ness of production and consumption in services is not a deﬁning characteristic of services. In Section V, we extend the argument to demonstrate that the principal economic function of some service institutions is to accomplish the separation of primitive economic activities across time or space. In Section VI we explore the role of specialized service institutions in fostering or permitting gains from exchange resulting from the division of ownership rights in market transactions. In Section VII we address the inﬂuence of technical progress on institutional change, and we grapple with the issue of determination of competitive boundaries in the context of service institutions. Finally, in Section VIII we conclude by bringing together the arguments of the six substantive sections.

II. DISTRIBUTION SERVICES AND PROPERTY RIGHTS: AN INTEGRATION Our aim in this section is to integrate important insights of two separate bodies of literature: recent literature on the analysis of retail institutions and another literature on the analysis of property rights. We begin by noting that all organizations are institutions, i.e. they impose constraints on human interactions, but all institutions are not organizations, e.g. zoning laws are retail institutions but they are not retail organizations.3 Since retail institutions are intrinsically service institutions, they provide a useful initial reference for the analysis of other service institutions. The economic function of a retail organization, in general, is to provide explicit market goods or services to ﬁnal consumers bundled with a set of distribution services that can be classiﬁed into ﬁve broad categories: accessibility of location, product assortment [depth and breadth], assurance of product delivery [at the desired time or in the desired form], information, and ambiance.4 These distribution services play a fundamental role in all subsequent discussion. For, they play a dual role in linking retailers and consumers. First they operate as ﬁxed inputs in the consumption or purchase activities of consumers, thus as any retail organization increases the levels of distribution services consumers incur lower [distribution] costs in order to attain given levels of satisfaction.5 Second these distribution services operate as outputs of retail organizations, hence when any retail organization increases the

160

ROGER BETANCOURT AND DAVID GAUTSCHI

levels of distribution services its total costs must rise (or at least not decrease). From the point of view of the property rights literature, each of these distribution services can be viewed as representing valued attributes of commodities exchanged. These distribution services lower for sellers or buyers distribution or transaction costs associated with exchange.6 Our analysis proceeds by treating transaction costs as mapping into elements of the set of distribution services mentioned above. While these distribution services were identiﬁed in the context of retail institutions, they are associated with any exchange (Betancourt, 1993). One characteristic of these distribution services is that, typically, they are not explicitly priced, and as a consequence can present opportunities for wealth capture by third parties. The integration of the analysis of property rights with the role of distribution services in facilitating exchange will reveal how institutions lower transaction costs and how shifting of these costs through the provision of different levels of distribution services affect the appropriation of the gains from exchange. Let us consider the convenience store as a retail organization to illustrate the integration of distribution services and property rights. From the perspective of the literature on distribution services the convenience store provides characteristically: low levels of assortment (i.e. narrow and shallow assortment), high levels of assurance of product delivery at the desired time (24 hours a day, for example), and high levels of accessibility of location. From the perspective of the property rights literature, the narrow and shallow assortment of the convenience store represents a restriction that allows divided ownership enhancing gains from exchange. That is, the limit on assortment enhances the gains from exchange for the store by lowering its costs of remaining open long hours; the limit enhances the gains for consumers by lowering their costs of planning and scheduling their purchase activities for certain types of frequently purchased goods and services. Indeed, it is this means of restricting or dividing ownership of the right of patronage that permits the convenience store to lower transaction costs. In order to evaluate the welfare effects stemming from the existence of an institution, it would be helpful to have a formalization. One way of analyzing this service institution is to look at the exchange gains resulting from its existence. We cast the formalization ﬁrst from the perspective of the consumer in terms of the cost difference in attaining a given level of utility assuming the institution exists against its not existing. Relying on a household production framework, this can be measured as an expenditure function differential as follows:

Product Innovation in Services: A Framework for Analysis

E0 = E(p*I , p, DI, Z0I ) ⫺ E(p*NI, p, DNI, Z0NI)

161

(1)

where p* represents a vector of retail prices, p represents a vector of other prices facing the consumer, D represents the levels of distribution services and Z0 the levels of consumption activities that yield utility.7 E( . ) is a classical expenditure function that is nondecreasing, linear homogeneous and concave in all prices and nonincreasing in distribution services. The ﬁrst function on the right hand side of (1) represents the costs of attaining a given level of utility conditioned on the existence of the institution (I) and the implied levels of prices, distribution services, and consumption activities. The second function represents the costs of attaining the same level of utility given that the institution does not exist and that the consumer faces the corresponding implied levels of retail prices and distribution services while being able to adjust consumption activities. For an efﬁcient institution to arise or to persist, E0, should be nonpositive since the second function represents optimization subject to an additional constraint (namely, that the institution cannot exist) relative to the ﬁrst one. In terms of the analysis of dissipation in the property rights literature, E0, represents the beneﬁts lost by eliminating the institution while allowing producers and consumers to adjust on remaining margins (p*, D, and Z0).8 It is worth clarifying why the expression in (1) may be positive. The emergence and disappearance of institutions need not involve marginal adjustments around an equilibrium. Their evolution may entail the rise and fall of constraints, and the movement from one equilibrium to another. This means that one cannot rule out the possibility of welfare loss in this evolutionary process even if one expects the normal or typical case to be one of welfare gains. An interesting illustration is the development of hypermarkets outside urban areas in France. They arise in the 1950s as a result of restrictions on store size in urban areas (Loi Royer) and they expand rapidly. With the lifting of some of these size restrictions to spur competition, supermarkets expand more rapidly than hypermarkets in France during the 1980s. Interestingly, legal restrictions on store size are reintroduced in France in 1996. It is unlikely that expansion of hypermarkets was welfare improving, nonpositive according to (1), in each of these three very different sets of circumstances. On the production side the gains from the existence of the institution can be measured as a proﬁt differential associated with the aggregate proﬁts of all agents consuming resources to permit the ultimate fulﬁlment of some consumption aim, that is, ␲0 = [p*I XI ⫺ C(v, XI, DI)] ⫺ [p*NIXNI ⫺ C(v, XNI, DNI)]

(2)

162

ROGER BETANCOURT AND DAVID GAUTSCHI

where C is a cost function for the retailer, X is a quantity vector of market goods or services produced or provided and v are the input prices faced by the retailer. All other terms have been deﬁned already.9 For an efﬁcient institution to arise or persist, ␲0 should be non-negative since the second function entails optimization subject to an additional constraint (namely, that the institution can not exist) relative to the ﬁrst one.10 If one assumes the process of entry and exit to lead to zero proﬁts given that the institution exists or not, then (1) also measures the (negative of the) beneﬁts to society from the existence of the institution. If we allow for nonzero proﬁts, then the beneﬁts to society will be measured by ␲0 ⫺ E0. The following proposition summarizes the argument: Proposition 1: The beneﬁts to society from the existence of an institution can be measured as the sum of the change in costs to consumers and the change in proﬁts to producers which result from its elimination. The critical role played by distribution services in the analysis of retail institutions underscores the signiﬁcance of distribution activities as primitive activities distinct from production and consumption activities. The emphasis in the property rights literature on the gains from exchange, many of which accrue due to restrictions that enhance individual rights by allowing divided ownership, suggests that there may be beneﬁts from the separation of these economic activities. The integration of these two literatures has helped to identify the economic function of a very speciﬁc service institution, namely, the convenience store. By separating primitive economic activities of any service institution we would seek to identify additional economic functions. We proceed now to introduce such separation systematically.

III. A TABLEAU OF PRIMITIVE ECONOMIC ACTIVITIES: A GUIDE TO ALTERNATIVE SERVICE INSTITUTIONS In this section we present a simple tool that will enable us to analyze the nature of alternative service institutions. We endeavor to construct a tableau not so much for classifying service institutions as for analyzing their speciﬁc economic functions. Features that are often attributed to services, such as difﬁculties of storage (Lovelock, 1989), quality measurement (Holmstrom, 1985), joint-ness (e.g. Bawa & Hale, 1995), and non-standardization (e.g. Zeithaml et al, 1985) may be illustrated in speciﬁc cells of the tableau and demonstrated not to hold at all in others. We begin by formally designating the

Product Innovation in Services: A Framework for Analysis

163

three conceptually distinct activities of production, distribution, and consumption as the primitive economic activities. Distribution encompasses all activities associated with conveying a product to market for ultimate consumption, and in conventional economic analysis such activities are often subsumed under production. As our conceptual framework is grounded on the notion of household production, we acknowledge that all economic agents [producers, distributors, and consumers] have production functions. The boundary between production and distribution, especially, is determined in any context by the consumption aim. That is, the output of a production activity is intended to fulﬁll a consumption aim; the output of a distribution activity is intended to permit such fulﬁlment.11 The conceptual distinction of the primitives and the ordered connections between them imply the imposition of certain constraints. But these constraints differ from the conventional constraints used in the machinery of economic optimization. It is necessary to emphasize this point because of the ingrained views of constraints in economics, as limiting possibilities on choices, thereby lowering welfare. Indeed, we have applied this convention in the development of our conceptual framework in Section II. That analysis, however, considers two situations under certainty, and the constraint in that section is the elimination of the service institution. In the analysis of restrictions that permit divided ownership in the property rights literature, the imposition of the restriction lowers overall uncertainty and the associated transaction costs. Constraints that apply in the tableau, permitting us to distinguish one cell from another, are relational constraints in the sense of information theory (Shannon, 1948; Shannon & Weaver, 1949). Such constraints play a dual role (Gatlin, 1972). These relational constraints, for example, restrict how a commodity can be consumed and, consequently, have the welfare enhancing effect of reducing uncertainty with respect to the feasibility of alternative consumption procedures. With respect to the primitive economic activities, a relational constraint is limiting by imposing a particular conﬁguration of activity. Nonetheless, within a given cell and conditioned on a speciﬁc consumption activity, the constraint allows the emergence of variety and novelty in organizational forms that can serve that activity. Constraints that have this dual role have been characterized as context sensitive constraints that make complexity possible, for example Juarrero (1999, Ch. 9). In the same vein the limit on assortment that allows the convenience store to exist can be thought of as a relational constraint. In fact, the reason the emergence of an institution such as the convenience store is welfare enhancing springs from the deﬁning characteristic of a relational constraint, namely, that it lowers uncertainty. In viewing the convenience store as a service institution

164

ROGER BETANCOURT AND DAVID GAUTSCHI

that imposes a constraint in the form of a restricted assortment permitting divided ownership of the right of patronage, the constraint lowers overall uncertainty and such transaction costs as storage and waiting. Thus, Proposition 1 can also be viewed ex post as a mechanism for measuring the beneﬁts of imposing a limit on assortments that allows this service institution to emerge relative to a situation where the absence of limits on assortment prevents the service institution from emerging.12 Our approach raises an aggregation issue, and we acknowledge that assignment of an institution to a cell in the tableau would be sensitive to the unit of analysis. For example, the institutional structure characterizing the transactions involving a consumer, travel agent, and an airline would depend on whether the relevant economic activity entails production and consumption of a claim to a trip or, simply, the production and consumption of a trip. In the former case, we can think of the consumer and distributor of the claim in the same space (the travel agent’s ofﬁce) and of both separate spatially from the producer (airline) while production, distribution and consumption (acquisition) of the claim are joint in time; in the latter case we can think of the distribution services provided by the travel agent in securing the claim to a trip as separate from production and consumption of the trip in space and time.13 Since such characteristics are sensitive to the purpose of the analysis, the tableau provides a systematic means for conducting the economic analysis. But, as any given service institution could be assigned to different cells depending on the economic context, the tableau is not intended as a convention for classifying institutions. In brief, the tableau is constructed on two dimensions: time and space. On each dimension, we identify ﬁve different combinations of these three primitive economic activities, depending on whether they are carried out jointly or separately with one another in this dimension. The tableau is, thus, bounded by two cases: joint-ness in time and space of production, distribution, and consumption, at one extreme, and complete separation in time and space of production, distribution, and consumption at the other extreme. All other cases are intermediate cases involving joint-ness or separability to some extent, temporally or spatially. In all, we identify twenty-ﬁve distinct conﬁgurations of the three primitive economic activities in the context of the tableau. The tableau is presented in Fig. 1. In the remaining sections of the paper we address a number of speciﬁc issues that acknowledgment of the tableau compels us to identify and resolve. We express the fundamental implications of the tableau of primitive economic activities in the following proposition:

Product Innovation in Services: A Framework for Analysis Time Space {P, D, C} D | {P, C} C | {P, D} P | {C, D} P|D|C

165

{P, D, C}

D | {P, C}

C | {P, D}

P | {C, D}

P|D|C

1 6 11 16 21

2 7 12 17 22

3 8 13 18 23

4 9 14 19 24

5 10 15 20 25

Key: | denotes separation { } denotes joint activities

Fig. 1. Table of Primitive Economic Activities.

Proposition 2: Assignment of a specialized service institution to a cell in the tableau implies the imposition of a relational constraint on a given consumption activity, at a given level of aggregation, served by a speciﬁc conﬁguration of production and distribution activities.

IV. JOINT-NESS The difﬁculty in drawing generalizations from the analysis of market services as distinct from market goods has impelled us to adopt the tableau in order to systematize our analysis. In this section we apply the tableau to explore the generality of one purported service characteristic: Namely, a conventional view that joint-ness is characteristic of market services, as distinguished from market goods, imposes the restriction that production and consumption are nonseparable activities. At the extreme this would involve distribution activities, as well. Thus, cell no. 1 of the tableau is the extreme case of joint-ness in service production. Examples of such extremes abound: a meal consumed at a local restaurant without having reserved a table in advance or attending a live event at a local arena, having paid for admission at the time and site of the event, to name just two. The contextual activities determining the joint-ness are, respectively, consuming a meal out of the home and consuming live entertainment out of the home. The specialized service institutions that produce and distribute their products jointly with consumption in these contexts are the local restaurant and the local arena. Instances of joint-ness are possible on either the spatial or the temporal dimension at the exclusion of joint-ness on the other, and joint-ness is possible with any pairing of the three primitive activities. Returning to the examples in

166

ROGER BETANCOURT AND DAVID GAUTSCHI

the preceding paragraph, one could reserve a table in advance of one’s arrival at the restaurant, thus separating, in time only, an aspect of distribution from production and consumption of the meal and dining experience that are to be accomplished jointly in time and space [cell no. 2]. A live prize ﬁght that may be viewed in a speciﬁc geographic region only in studios connected to the ﬁght arena by closed circuit TV joins production, consumption, and distribution of the event in time, and joins consumption and distribution in space while keeping production separate in this dimension [cell no. 16]. Interesting asymmetries among combinations of the three primitive economic activities are possible, as well. Consider the situation of those tourists who ply the streams and slopes of Maine’s Baxter State Park, or who ride on the tour buses around Yosemite Valley, or who explore the garden paths and atelier at Monet’s Giverny. In each case, signiﬁcant outputs of the relevant production that now attract the tourists occurred years or even centuries in advance of the tourists’ present-day consumption. Moreover, a signiﬁcant aspect of distribution occurred jointly with production in time: Monet’s water lilies still grow in the same manner from the same spot that ﬁrst inspired him to record his impressions on canvas, some of which still hang in the atelier; El Capitan rises from the ﬂoor of Yosemite Valley where it was ﬁrst placed eons ago; and the waters of Chimney Pond still bathe the moose in the shadow of Mt. Katahdin as they did even before Percival Baxter gifted the place to the State of Maine. As a consequence of this, tourists are attracted to the precise space where these magniﬁcent outputs were placed or distributed for succeeding generations to consume14 [cell no. 3]. In a sense one could view these natural wonders as durable goods. Such a view would imply that what is consumed is a ﬂow of services, thus joining production and consumption in time and space. Our view is that production – just as distribution – has many facets, and one important facet is the manufacture or creation of the durable good as a store of services that could ﬂow from it. Here we invoke Georgescu-Roegen’s (1971) thesis that reality is a seamless web and, consequently, such analytical categories as the primitive economic activities naturally introduce artiﬁcial boundaries. The service institutions that can be assigned to cell no. 1 do stand in stark contrast to those that can be assigned to any other cell in the tableau, as all other cells depict situations of some degree of separation of the three primitive economic activities. At the opposite extreme, production, distribution, and consumption can be completely separate in time and space [cell no. 25], and one need look no further than the quest for a bag of charcoal from the convenience store as one prepares a barbecue at home. In this instance, production is accomplished by a manufacturer remotely located in time and

Product Innovation in Services: A Framework for Analysis

167

space from the store; distribution is produced by the store; and consumption can only begin to take place upon return home with a bag of charcoal in hand. These examples are sufﬁcient to establish the following: Proposition 3: Joint-ness in production, distribution, and consumption is widespread in services markets, but it is neither a necessary nor a sufﬁcient condition to distinguish a service product from any other market product.

V. AN ECONOMIC FUNCTION OF SERVICE INSTITUTIONS Proposition 3 merely eliminates joint-ness in the primitive activities as a deﬁning characteristic of services. In this section we seek to establish that the removal of joint-ness in the primitive activities is a critical function of many service institutions. We shall attempt to establish this by considering some carefully chosen examples. Credit cards, such as Visa and MasterCard, present an interesting example of a specialized service institution that serves as a mechanism for separating an aspect of distribution from production and consumption in time and space. A credit card provides a mechanism whereby certain distribution services are explicitly sold to interacting customers. The explicit price the consumer pays for the services of this service institution is a fee for the card; the explicit price the merchant pays for the service institution’s services is a percentage of those sales transacted with the card. An important economic function of this service institution is the unbundling of certain distribution services from the explicit goods and services transacted. From the perspective of the consumer this service institution’s economic function is to provide high levels of assurance of product delivery at the desired time in a variety of locations together with information on the availability of this service across space and time. The institution produces these services by evaluating individual consumer’s credit worthiness, on the one hand, and by negotiating with merchants to honor the card on the other hand. The institution distributes these services by issuing cards to consumers and signs or decals to merchants. Consumption of these services occurs when the credit card is employed in a transaction. From this vantage point, production, distribution, and consumption of these services are typically separate across space and time [cell no. 25]. The service institution’s economic function from the perspective of the merchant is to provide high levels of assurance of product delivery in the desired form, namely, payment in legal tender and, possibly to a lesser extent, assurance of product delivery at the desired time. Hence from the point of view

168

ROGER BETANCOURT AND DAVID GAUTSCHI

of the merchant, consumption and distribution of the service occur jointly in time and space [cell no. 19] when the consumer uses the credit card.15 As a specialized service institution that separates at least one distribution service from all of the distribution services that are invoked in an economic transaction, the credit card presents a clear illustration of how various kinds of costs may be shifted among the three interacting agents: the consumer, the merchant, and the credit card agency. To enjoy the beneﬁt of greater assurance of product delivery at disparate merchant venues, the consumer willingly incurs the explicit cost of entitlement to the use of the credit card. To produce sufﬁciently high levels of assurance of product delivery for the subscribers of its services, the credit card agency incurs direct costs of negotiating with many merchants on behalf of subscribing consumers who would desire credit privileges. In doing so the credit card agency enjoys scale and network economies that any individual consumer would be unable to achieve for herself. Similarly, any given merchant who participates in the credit purchase programs of the credit card agency willingly incurs the opportunity cost of a discounted invoice on all credit transactions in order to receive the beneﬁt in the form of high levels of assurance of product delivery. The credit card agency incurs the direct costs of producing this beneﬁt for the participating merchant by handling all billing of subscribing consumers. In doing so, the credit card agency enjoys scale economies that any given merchant would ﬁnd difﬁcult to realize were he to do this for himself. Such instances of the shifting of costs across the market establishes the grounds for the existence of the service institution as expressed in the expenditure differential of Section II. Specialized service institutions do not emerge only to accomplish separation of distribution activities from production and consumption activities. For example, the spatial separation of the primitive activities that is characteristic of most of the services produced and consumed in the communications and information services industries are illustrations of the separating of consumption from production. Video-conferencing, a service institution that is gaining prominence in certain settings, has been demonstrated to be a beneﬁt to society in the new area of “telemedicine” by separating activities in space while joining them in time (Gautier, 1995). Such an institution permits physicians located in the emergency room of Massachusetts General Hospital to provide physicians’ services to clients entering the aid station at Logan Airport, using the intermediation of para-medics. In this case, production, distribution, and consumption of the physicians’ services take place simultaneously; consumption and distribution occur in the same space, but production is separated from both spatially [cell no. 16]. It is the separation of primitive economic activities across space that is the main economic function of this specialized service

Product Innovation in Services: A Framework for Analysis

169

institution. One should note that the technology of video-conferencing does not exclusively enable institutions corresponding to a particular cell. For example, video-conferencing makes possible the production and distribution of a lecture in one locale and its simultaneous transmission to consumers in another locale [cell no. 11]. Other institutions in the communications domain, such as voice mail and electronic mail, have emerged to allow separation of production from consumption especially in time. In the parlance of property rights, voice and electronic mail services confer on consumers the right to receive [i.e. consume] messages produced and distributed to them by others whenever they want. For instance, use of a voice mail service at the local telephone company separates production distribution and consumption of phone messages in space and allows joint-ness in time for production and distribution while keeping consumption separate in time [cell no. 23]. If the message is stored in a recorder on the consumer’s phone, however, production is separate from distribution and consumption in space, which in turn are joint, while consumption remains separate from production and distribution in time [cell no. 18]. Finally, a cell phone allows for space separation and time joint-ness in the production, distribution and consumption of phone messages [cell no. 21]. These examples are sufﬁcient to establish the following: Proposition 4: One of the primary economic functions of service institutions is to separate at least some aspects of production, distribution, and consumption across space and time.

VI. DIVIDED OWNERSHIP AND SERVICE INSTITUTIONS In his analysis of property rights, Barzel (1997, p. 114) reconciles the enhancement of individual wealth with the imposition of restrictions that promote divided ownership of commodities. Such restrictions help delineate rights properly, thus preventing wealth capture on the part of non-owners of the commodities or of their attributes. These restrictions resulting in divided ownership are relational constraints, as deﬁned in Section III, in that by limiting rights in at least one dimension they lower uncertainty in capturing the full beneﬁts of the exchange in other dimensions. To illustrate this brieﬂy, we consider again the example of the convenience store. A consumer who is seeking a bag of charcoal for an imminent barbecue dinner could rush to a supermarket. After searching through the extensive assortment, the consumer would then queue at a checkout lane. In the checkout

170

ROGER BETANCOURT AND DAVID GAUTSCHI

lane, other consumers ahead in line could likely have more items to purchase, more questions requiring immediate answers, and more complex transactions possibly requiring credit approval. In a sense, these consumers, perhaps unwittingly but genuinely, engage in capture of a scarce commodity: the charcoal bag-toting consumer’s time. Nevertheless, those consumers who might engage in slow, complex transactions in the supermarket would be unable to exercise such behavior in the convenience store because of the restrictions the convenience store imposes on them. That is, the limited assortment, restricted layout, refusal to honor purchases by check, and so forth all combine to eliminate the opportunities of other consumers to capture the time of the charcoal bag seeking consumer. To put this another way: the restrictions embodied in the convenience store as an institution lower a consumer’s uncertainty, below what it would be at the supermarket, with respect to owning her own time while exercising her right of patronage over a limited assortment. Borrowing an example from Barzel, commercial use of a refrigerator by a consumer is usually prohibited under the conditions of the manufacturer’s warranty. This restriction imposed by the warranty allows the manufacturer to lower the consumer’s uncertainty of obtaining the full beneﬁts from use of the refrigerator. As Barzel notes, such restrictions delineating rights also facilitate the emergence of institutions that enhance the gains from exchange, such as second hand markets in the case of refrigerators. Reconciliation of welfare gains and the imposition of restrictions is an intriguing and unconventional proposition in economics, and it naturally raises some questions. How might ownership be divided? Why might divided ownership result in gains from exchange? Let us consider the example of an automatic teller machine (ATM) network. An economic function of this institution is to provide assurance of product delivery at the desired time, an important aspect of distribution, by separating the acquisition of the means of payment for any relevant economic activity from other aspects of distribution, production, and consumption across space and time. That is, the ATM network divides spatially and temporally for each of its users the right to engage in exchange. Moreover, without ATM networks some exchanges by absent-minded or myopic consumers who neglect to carry sufﬁcient cash would not take place at all. Ownership of the right to use the ATM is also commonly divided among consumers by means of a personal identiﬁcation number (PIN), limiting the right of access to speciﬁc agents. Any speciﬁc consumption context that can be assigned to either row 2 or column 2 of the tableau in Fig. 1 can be accomplished using an ATM. In principle we can measure the gains from the spatial separation in cell no. 6, for

Product Innovation in Services: A Framework for Analysis

171

example, made possible by the ATM network with the framework of Section II. If one assumes that the ATM market is competitive, then proﬁts are zero and the beneﬁts to society facilitated by this institution conforming to the organization of the primitive activities in this row are measured by the reduction in the costs to the representative consumer of attaining a given level of utility from speciﬁc consumption activities with the ATM network relative to the situation without it. This could be aggregated over the number of ATM users and over the set of speciﬁc consumption contexts for which cell no. 6 can serve as a description. There are many examples, such as attending plays or ﬁlms, enjoying restaurant meals, etc. What makes these gains from exchange both feasible and important is that these networks allow divided ownership of the right to use any ATM on the network for all consumers participating on the network. It is also clear from this discussion that the separation of primitive economic activities in cell no. 6 made feasible by the ATM network increases society’s welfare in part by increasing the number of transactions that can take place.16 The ATM example illustrates the dividing of ownership over space of each user’s right to use the network, as well as the dividing of ownership across consumers. Moreover, there is also dividing of ownership rights over time in the separation of the right to access cash at any time from the right of access to other bank services, which are available during ofﬁce hours. Consider again the example of voice mail. Its economic function is to provide assurance of product delivery at the desired time. This specialized service institution enhances the gains from exchange by allowing consumers to receive messages whenever they want. These gains can be measured in terms of the reduced costs to the representative consumer of having these services relative to the situation without them, assuming proﬁts to be zero for the sake of simplicity. From society’s point of view these beneﬁts can be aggregated over all users of the service. Once again without this service institution some exchanges would not take place, so the number and variety of messages transmitted is enhanced by the existence of voice mail. What’s most interesting in this case, however, is that there are two different institutional arrangements for providing voice mail. From the point of view of the consumption activity this service can be provided through an institutional arrangement where the consumer buys a phone with a tape device to store messages [cell no. 18] or through another where the consumer buys a service from a digital network [cell no. 23], which is a service institution supported by a telephone company for its subscribers. This is an interesting example where a service provider competes with a manufacturer in satisfying a consumption aim by shifting the costs of the storage function from the consumer to the

172

ROGER BETANCOURT AND DAVID GAUTSCHI

distributor.17 In both cases, however, the producer of the tape and the provider of the storage service have access to the consumer through a third party’s willingness to restrict its ownership of access mechanisms to the ﬁnal customer by sharing either the telephone apparatus or the telephone service.18 Because divided ownership, if feasible, reduces uncertainty for both parties to an exchange it alleviates the problems of adverse selection (hidden information) and moral hazard (hidden action). Consider the instance of commercial software, such as Lotus 123. From the perspective of the producer, a buyer could use the product in ways that could limit the proﬁt to be gained from its sale. For example, a buyer could embed Lotus 123 into software tools that she produces and sells to another buyer. Unless there were some mechanism for metering the use of the embedded Lotus 123 program, Lotus Development Corporation would not receive any rents from the exchanges that the software developer would engage in with her clients. Consequently, Lotus Development Corporation licences the use of the Lotus 123 program to each user, and prohibits transfer of the use rights of the program to any other potential user. This means that those users who buy the software developer’s tools must have engaged in a transaction to acquire the licence to use Lotus 123. This restriction has value to Lotus as it reduces moral hazard, and it has value – although not necessarily the same – to all users.19 In transferring the right to use Lotus 123 to an individual user, Lotus Development Corporation also imposes another set of restrictions that reduce another aspect of uncertainty from the perspective of the user. Software markets are widely known to be subject to inexorable technical improvement, as new versions or upgrades of existing software packages are announced frequently. The consumer, recognizing that this is the case, may have difﬁculty knowing when it is best to enter the market. Software producers such as Lotus Development Corporation, can warrant to the adopters of their software that upgrades will be sold to them at discounted prices whenever they are made available to the market. In so doing, the software producer is alleviating an adverse selection problem for the buyer by giving existing consumers privileged access to the rights to use improved forms of the product that will be probably available at some future point in time. Those who have purchased earlier versions of the software, therefore, must be able to distinguish themselves from ﬁrst-time adopters. They are able to do so, by displaying to the software retailer the registration number of their initial licence. Hence, in the case of software, ownership is divided in rather complex ways across users and across time. In both instances the restrictions dividing ownership (limitations on resale and commitments to discount) enhance the gains from exchange by reducing uncertainty.

Product Innovation in Services: A Framework for Analysis

173

These illustrations demonstrate that the phenomenon of increases in individual wealth through restrictions that enhance divided ownership is much more extensive than Barzel may have argued originally. Thus, we have: Proposition 5: Service institutions enhance the gains from exchange by reducing uncertainty through the promotion of divided ownership by economic agents over assets and services (especially distribution services) across space and time.

VII. TECHNOLOGY, COMPETITIVE BOUNDARIES AND SERVICE INSTITUTIONS In this section we consider the inﬂuence that technology may have on service institutions, and are especially concerned with the issue of how technological progress may induce change in service institutions as economic agents engage in various marginal adjustments in response to a modiﬁed technological environment. We adopt an expansive view of technology as a combination of methods, procedures, instruments, knowledge, and skills that may be exploited to accomplish some productive aim.20 At any transaction point, technologies, so deﬁned, exist on both sides of the market and condition the nature of exchange. What is valued by one party to an exchange may depend on that which that party would ﬁnd difﬁcult to accomplish for himself or herself; hence technological progress for that party may alter the nature of ensuing exchanges. In Section V we argued that an economic function of some service institutions is to separate aspects of primitive economic activities across space and time. This ability to separate activities that previously could not be separated illustrates an important role of technology in product innovation in services: it provides feasibility. Indeed, this role of technology is a well established one and is not limited to modern services. A clear example of its foundation pertains to strategies and conventions of human subsistence. In agrarian societies that are subject to severe winters, the consumption of speciﬁc kinds of fresh fruits and vegetables is likely to take place only during the growing seasons that are of limited duration. Thus, in the summer, a household may consume exceptionally high quantities of strawberries, tomatoes, leafy lettuce, corn, peaches, and so forth. As autumn unfolds, the household’s consumption switches to apples, squash, beans, and potatoes. In the depths of winter and in the early spring, the household may consume fruits and vegetables only if it has applied a storage technology of one kind or another that involves altering the chemistry of the food: canning. In any event,

174

ROGER BETANCOURT AND DAVID GAUTSCHI

in the most primitive societies, production, distribution, and consumption of foodstuffs are joint in time and space. The application of storage technology based on some preservation method permits production to be separated from consumption in time. The invention of modern refrigeration methods presents an instance of technological progress that has complex consequences on the evolution of service institutions. A great variety of foodstuffs may be frozen without appreciable deterioration or alteration of the chemical properties of the fresh version. Of course there are exceptions, for example bananas and lettuce. This means that a supplier with access to refrigerated storage technology could bring to market “fresh” oranges or corn in the middle of the summer and in the middle of the winter. The beneﬁt to the consumer is that the variety of meals is enhanced at any point in the year. Incremental technological advances such as the acquisition of food preservation methods ﬁrst open new cells of the tableau for organizing the primitive activities temporally. For instance, cells 3 and 5 of row no. 1 of the tableau represent possible alternative institutional conﬁgurations exploiting the advances presented by food preservation technologies that can serve the basic consumption activity of subsisting by separating consumption and production in time. Secondly, within any given cell, the attributes that may be valued in exchange may vary. For example, an agent who purchases large lots of fresh produce from farmers in the summer months and freezes the produce, may specialize in providing assurance of product delivery in the desired form (only tomatoes, for example) throughout the year or assortment depth and breadth (ten different varieties of tomatoes, peaches, oranges, etc.) at any point in the year. In either of these cells, however, a food preservation technology such as cold storage may induce variety in the satisfaction of the basic consumption activity of subsisting. It is interesting to note that prior to refrigeration, the technology applied to extend the nutritional life of a perishable food involved processing it in some way to preserve at least some of its nutritional qualities over time. Milk was processed to make cheese and yoghurt; ﬁsh and meats were cured; fruits and vegetables were canned. These food preservation technologies effectively added a step to the production activity that could be construed as a distribution service (assurance of product delivery in the desired form). Hence, the technological progress associated with early forms of food preservation constituted an alteration of a service institution, as the relational constraints imposed on the primitive economic activities joined production and distribution in order to unbundle consumption temporally. This technology also abetted the expansion of the agricultural producing unit, enlarging the expanse of the

Product Innovation in Services: A Framework for Analysis

175

market, permitting realization of scale economies. Using the framework of Section II, not only does the consumer receive a beneﬁt as a result of the existence of an institution of food preservation, the supplier is encouraged to expand to enjoy scale economies that present proﬁt potential in conjunction with the exploitation of the technology.21 In the present environment, the year long availability of the kiwi fruit in retail markets as far removed as New Zealand and California has resulted from the entrepreneurial activity of someone who has realized the advantages of combining air transport and refrigeration with two different harvesting seasons.22 A similar story applies to the world-wide distribution of fresh cut ﬂowers from Holland. Thus, in short, the potential consequences of technological progress are to expand the limits of the market in time and in space. This also means, in terms of the tableau, that technological progress presents the opportunity for new institutional conﬁgurations of the primitive economic activities to emerge for highly disaggregated consumption activities as the set of choice alternatives is expanded. In the case of refrigeration alone, a perishable commodity is rendered more durable, enabling the temporal separation of consumption and production. In the case of high speed transportation, it is the perishability of a commodity that is preserved or rendered more durable, thus reducing the separation of consumption and production in time. In either case, however, there is technological progress that enlarges the boundaries of the market across space and time. Technological progress has the potential for redrawing competitive boundaries in a variety of ways, and some of these may be difﬁcult to anticipate. For instance, in bygone times a consumer in Paris may have developed some familiarity and appreciation for the kiwi fruit by consuming concoctions made from kiwi extract that complemented other inputs of meals. Today, the successors of these former consumers can consume the fresh fruit at almost any time of the year, and use it to substitute for other fresh fruits that its extract may have complemented in previous times. Indeed, progress in transportation technology has brought remarkable changes to the retail markets of grocery products, as the assortment of fresh foods has continuously expanded. Thus with respect to the offerings of some fresh foods, a restaurant in the U.S. can compete with a restaurant in Europe regardless of the continental origins of the fresh food. Of course high speed ground transport modes or air transport modes are not only amenable to the transport of fresh food products. Progress in air transportation and communications technologies has created ever more complex forms of substitution, exacerbating the problem of deﬁning the boundaries of markets. More than one hundred years ago, as railroads were expanding over the globe, communications in the form of postal service were

176

ROGER BETANCOURT AND DAVID GAUTSCHI

enhanced as a complementary service of this new transportation service. Letters could contain far more content than a telegram, and as transportation technology continued to improve throughout this century letters could be conveyed to their addressees with progressively less delay. Communications technology has progressed to the point now, however, where more information can be conveyed in real time to an addressee by means of the public switched digital network than in a lengthy document conveyed through the postal service or by some other courier service. In fact, the “Reach out and touch someone” advertising campaign of ATT is an attempt to suggest that the institutions of modern telecommunications can substitute in some instances for the institutions of modern transportation. The recognition of such novel sources of competition has prompted Boeing, for example, to buy full-page advertisements in the Financial Times depicting situations that are concluded with the tagline “. . . sometimes there is no substitute for being there.” Because of the geographic expansion of the market for kiwi fruit afforded by technological progress principally in air transport, and the temporal expansion of the market afforded by refrigeration, the supplier of kiwi fruit to the market is able to divide ownership of the right of market access among more than simply those consumers who reside in close proximity to the points of production. Service institutions made possible by these changes enhance the welfare of consumers and the proﬁts of suppliers as compared to the condition where these institutions do not exist. These institutions, however, emerge, rise, and decline over decades not years. For instance, the introduction of the kiwifruit at the retail level in California took place in 1961 but the California KiwiFruit Commission was not create until 1980. Similarly, it is well known that the provision of information on new products through advertising, for example, is more extensive than for old products, Brozen (1982, p. 173). Most of the characteristics of technological innovation in other sectors also affect or characterize innovation in the service sector but they may be more difﬁcult to identify in a service context. For instance, creative destruction, the distinction between cost-reducing and product innovation, and interdependence of innovations all arise in the context of service institutions. We conclude by illustrating these features. In the last ﬁfty years the corner grocery store has effectively disappeared in the United States. Instead, the retail landscape is now well represented by convenience stores, supermarkets and even vending machine operations. There is another process of creative destruction in the U.S. retail landscape, however, that is less well known: conventional supermarkets are in the process of being replaced by formats with broader, and in some cases deeper assortments such as superstores. For instance, between 1990 and 1998 the share of sales of

Product Innovation in Services: A Framework for Analysis

177

supermarkets made by superstores has gone from 34% to 42% while the share of sales by conventional supermarkets has declined from 35% to 19%.23 This process also took place during the 1980s. One of the best examples of a cost-reducing innovation in services is the introduction of the Universal Product Code, which initially allowed the widespread diffusion of scanners in the U.S. grocery trade, spread to other areas of domestic retailing and is by now adopted far beyond the conﬁnes of retailing and the U.S. border. From zero registrations of manufacturers in 1970, the Uniform Code Council acknowledged cumulative registrations of 177,000 manufacturer speciﬁc numbers by January of 1997. The number of actual products is unknown but undoubtedly larger, since each manufacturer can assign digits to identify each product after its registration number.24 This innovation is also suggestive of the interdependence of innovations in service institutions, since the rapid expansion of formats with broad and deep assortments in grocery retailing would have been far more costly without the economies at the checkout counter and in inventory management made feasible by the U.P.C. Ironically, service institutions characterized by these newer formats with broader and deeper assortments can and in many cases should be viewed as forms of product innovation in services! One can also ask if our familiar convenience store is an example of a novel service institution that is a product innovation or a cost reducing innovation in services. The answer may be that it illustrates both features. Its characterization in terms of narrow and shallow assortments, extended hours and convenient location suggests that one should view it as a product innovation. On the other hand, since most convenience stores are operated as business format franchises, elements of cost reducing innovations associated with this institutional form are relevant to the economic feasibility of this service institution. Finally, we consider vending machine operations. They are a service institution that permits the joining of distribution – and in some cases, even aspects of production – and consumption in space and time. Still, the vending machine offers a more restricted assortment than the convenience store, and partly as a result of this, vending machines can be placed in more accessible locations than convenience stores. Moreover, as in the case of the convenience store, the limit on assortment provided by the vending machine is beneﬁcial only when one can divide the right of patronage over many consumers. The more accessible are vending machines, the more pronounced is the separation between production and distribution at least in terms of keeping the machines stocked.25 The introduction of frozen products in their assortment provides a simple and clear illustration of technological change in the form of a product

178

ROGER BETANCOURT AND DAVID GAUTSCHI

innovation by a service institution, perhaps in response to competition from convenience stores. We summarize the arguments of this section in the following proposition: Proposition 6: Technological change allows the emergence of service institutions that: (1) separate primitive economic activities across space and time; (2) provide variety and novelty in satisfying given consumption aims; (3) redraw the competitive boundaries of various markets.

VIII. CONCLUDING REMARKS Our integration of the property rights literature with the analysis of distribution services – by treating these services as valued attributes of any exchange – generates two important results. First, it provides a conceptual framework for evaluating the beneﬁts of service institutions comparable to recently developed approaches for products with new service characteristics. Secondly, it leads to the separation of economic activities into three primitive forms: production, distribution, and consumption. A systematic look at all possible conﬁgurations of these three primitive forms across space and time yields two immediate results for the analysis of service institutions. It shows that the existence of these institutions implies the imposition of a relational constraint which reduces uncertainty and the associated transaction costs. Moreover, it also shows that the conceptual framework of Section II continues to be applicable for the evaluation of the beneﬁts to society. Several additional results derive from looking at service institutions from this perspective. First, it becomes clear that joint-ness is not a deﬁning characteristic of services. Second, it also becomes clear that an important economic function of many service institutions is the separation of these primitive economic activities across space or time. Third, the relational constraints implied by the existence of service institutions enhance the gains from exchange by allowing divided ownership of commodities and distribution services by economic agents across space or time. This result extends the analyses in the property rights literature over a much wider domain, and it should facilitate a more systematic application of this idea. Finally, our analysis provides a somewhat different view of the consequences of technical change with respect to the existence of service institutions, novel forms of organization by these institutions, and the nature of market boundaries. The topic under analysis is a vast one, and there are many issues that we have either ignored or treated in an incomplete fashion. Nonetheless, we hope to have moved the development of a framework for the analysis of service

Product Innovation in Services: A Framework for Analysis

179

institutions sufﬁciently forward so that a rigorous application of these ideas to more narrowly deﬁned topics is, thereby, feasible.

NOTES 1. That is, an economic agent occupying the demand side of a market. Thus, our ‘consumer’ can be a business ﬁrm or a household. 2. In the case of households, it is possible that the work reduced may have yielded satisfaction. For example, restaurant meals reduce work for the household but may reduce the joy of cooking for some of its members. 3. According to North (1989) organizations are differentiated from other institutions, explicitly or implicitly, by their membership. 4. The concepts and how they map into the distribution costs faced by consumers are discussed in more detail in Betancourt and Gautschi (1986, 1988). Empirical measurements at the industry level are developed in Betancourt and Gautschi (1993) and at the store level in Betancourt and Malanoski (1999). A similar classiﬁcation is provided by Oi (1992). 5. The implications of this characteristic for demand analysis are explored in Betancourt and Gautschi (1990, 1992). 6. For instance, Barzel (1982) deﬁnes measurement cost as mapping directly into the costs of providing product information and analyzes its implications. We are merely saying that similar mappings exist for other transaction costs and distribution services. For example, the use of warranties to provide quality assurance is a means of lowering transaction costs due to potential enforcement problems by increasing the distribution service assurance of product delivery in the desired form. 7. This formulation is analogous to the one proposed by Nordhaus (1997) for evaluating a new product with new service characteristics. The narrower one he uses in the case of light is appropriate for evaluating a new product generating existing service characteristics. 8. The adjustment in the vector Z0 has to be such as to leave the level of utility unchanged. 9. Implicit in our analysis of both consumer and producer beneﬁts is that the ﬁrst term always pertains to the existence of an institution, even if the institution is infeasible in actual practice, which is what happens in general with market failure. 10. This argument assumes no change in market structure. For instance, if the proﬁts were to arise as a result of a change in market structure from competition to monopoly due to the introduction of the institution, they should be subtracted from, rather than added to, the consumer gains. For an evaluation of welfare gains under imperfect competition in a simpler setting, see Hausmann (1997). 11. One should note that the identiﬁcation of the three primitive economic activities applies at any level of the marketing system, supply chain or channel conﬁguration. 12. The concepts of vertical restraints in industrial organization, commitment in game theory and coordination mechanisms in macroeconomics generate similar situations. They imply the imposition of constraints, and these constraints enhance welfare by lowering uncertainty and the associated transaction costs. 13. In the tableau, the former case belongs in cell no. 16 and the latter case belongs in cell no. 7.

180

ROGER BETANCOURT AND DAVID GAUTSCHI

14. Because tourists are attracted to the same places in such large numbers, public agencies with the responsibility of preserving the outputs of the production could restrict the spatial joint-ness primarily to production and consumption by limiting the number of tourists who could visit these sites. As such restrictions would likely be implemented at reservation agencies situated closer to the points where tourists reside, a signiﬁcant aspect of distribution would be separate from production and consumption in space [cell no. 8]. 15. The discussion reveals one network feature of credit cards as a service institution. Since the services provided are distinct for consumers and merchants, this institution functions as a two-way network in the terminology of Economides (1996). Thus, our analysis is useful in applying his classiﬁcation scheme for networks to service institutions. There are other aspects of the network that we ignore here, for example the role of banks in providing short-term credit (Ausubel, 1991). 16. Recently it has been argued that one of the main costs of a lack of impartial third party enforcement is the failure to exist, or operation at very low transaction levels, of markets where transactions are not self-enforcing. Moreover, the deﬁciencies in these markets have been identiﬁed as one of the main reasons for underdevelopment, e.g. Clague, Keefer, Knack and Olson (1999). 17. In practice the institutional arrangement relying on the service provider dominates in ofﬁce contexts and the one relying on the tape manufacturer dominates in the residential market. 18. This feature also suggests an underlying institutional background in terms of a legal system and contract enforcement mechanisms that can support or retard the development of service institutions which we are not addressing explicitly. 19. If, in the absence of the restriction, the program were not available or were available only at a higher cost, users would be disadvantaged. The beneﬁts of the restriction can be measured with the criterion of Section II. The beneﬁts would differ across users in that those who would want to transfer the program to other potential users would face lower costs without the restriction. 20. This view encompasses the standard economic view of technology, e.g. Salter (1966), as well as the more recent view of technologies as networks, e.g. Scazzieri (1993). 21. The expansion of scale is especially illustrated by the displacement of traditional geographic market boundaries that may have been previously limited by inefﬁcient transportation technologies. For instance during the Renaissance Marco Polo successfully introduced spices from the Orient into the European market, exploiting a combination of improvements in preservation and sea transport technologies. The consequence of this was to extend the limit of the Oriental spice market by spawning an industry, the spice trade, over which wars were fought. 22. For a brief history of the kiwi fruit see the site of the Californian KiwiFruit Commission at < www.kiwifruit.org > . 23. These ﬁgures are calculated from Table 1280 of the U.S. Statistical Abstract (2000). 24. Our discussion of the U.P.C. is based on Brown (1997). 25. There are limits on accessibility, though, as it is not efﬁcient to locate vending machines in consumers’ homes.

Product Innovation in Services: A Framework for Analysis

181

ACKNOWLEDGMENTS An earlier version of this paper, titled “ The Nature of Service Institutions: Welfare Enhancing Constraints and Primitive Economic Activities,” was presented at the Marketing Science Conference in Sydney, Australia, at the Marketing Camp, University of Washington, at the Research Seminar in Marketing, Washington University, and at the Industrial Organization Workshop, University of Maryland. The ﬁrst-mentioned author gratefully acknowledges ﬁnancial support provided by the U.S. Agency for International Development under Cooperative Agreement No. DHR-0015-A00-0031-00 to the Center on Institutional Reform and the Informal Sector (IRIS) and administered by the Ofﬁce of Economic and Institutional Reform, Center for Economic Growth, Bureau for Global Programs, Field Support and Research. The second-mentioned author gratefully acknowledges a summer research grant from the Research and Travel Committee, School of Business, University of Washington. The views expressed here are his own, and not necessarily those of Deloitte & Touche LLP. We appreciate the thoughtful comments and suggestions of participants in the aforementioned seminars and want to acknowledge speciﬁcally those of L. Ausubel, M. Cao, C. Clague, P. Cramton, M. Cremer, J. Nelson, M. Oyler, A. Sen, J. Ruth, B. Tietje, C. Weinberg, and B. Wernerfelt on earlier drafts. We are solely responsible for any errors remaining.

REFERENCES Ausubel, L. (1991). The Failure of Competition in the Credit Card Market. American Economic Review, 81, 50–81. Barzel, Y. (1997). Economic Analysis of Property Rights (2nd ed.). New York: Cambridge University Press. Barzel, Y. (1982). Measurement Cost and the Organization of Markets. Journal of Law and Economics, 17, 73–96. Bawa, K., & Hale, A. (1995). Service Characteristics and Switching Behavior: An Empirical Investigation, paper presented at Informs Marketing Science Conference, Sydney. Becker, G. (1965). A Theory of the Allocation of Time. Economic Journal, 75, 493–517. Betancourt, R. (1993). An Economic Analysis of the U.S. Distribution System. OECD, Economics Department Discussion Paper No. 135. Betancourt, R., & Gautschi, D. (1986). The Evolution of Retailing: A Suggested Economic Interpretation. International Journal of Research in Marketing, 3, 217–232. Betancourt, R., & Gautschi, D. (1988). The Economics of Retail Firms. Managerial and Decision Economics, 9, 133–144. Betancourt, R., & Gautschi, D. (1990). Demand Complementarities, Household Production, and Retail Assortments. Marketing Science, 9, 146–161.

182

ROGER BETANCOURT AND DAVID GAUTSCHI

Betancourt, R., & Gautschi, D. (1992). The Demand for Retail Products and the Household Production Model: New Views on Complementarity and Substitutability. Journal of Economic Behavior and Organization, 17, 257–275. Betancourt, R., & Gautschi, D. (1993). The Outputs of Retail Activities: Concepts, Measurement, and Evidence from U.S. Census Data. Review of Economics and Statistics, 75, 294–301. Betancourt, R., & Malanoski, M. (1999). An Estimable Model of Supermarket Behavior: Prices, Distribution Services and some Effects of Competition. Empirica, 26, 55–73. Brown, S. (1997). Revolution at the Check-Out Counter: The Explosion of the Bar Code. Cambridge: Harvard University Press. Brozen, Y. (1982). Concentration, Mergers, and Public Policy. Macmillan, New York. Clague, C., Keefer, P., Knack, S., & Olson, M. (1999). Contract Intensive Money: Contract Enforcement, Property Rights, and Economic Performance. Journal of Economic Growth, 4, 185–211. Drucker, P. (1998). The Next Information Revolution. Forbes, (August). Economides, N. (1994). The Economics of Networks. International Journal of Industrial Organization, 14, 673–699. Eggertsson, T. (1990). Economic Behavior and Institutions. London: Cambridge University Press. Gatlin, L. (1972). Information Theory and The Living System. New York: Columbia University Press. Gautier, A. (1995). An Analysis of the Market Opportunities in Cybermedecine, mimeo. University of Washington, School of Business Administration. Georgescu-Roegen, N. (1971). The Entropy Law and the Economic Process. Cambridge: Harvard University Press. Griliches, Z. (1992). Output Measurement in the Service Sector. Chicago: University of Chicago Press. Haussman, J. (1997). Valuation of New Goods Under Perfect and Imperfect Competition. In: T. Bresnahan & R. Gordon (Eds), The Economics of New Goods. Chicago: University of Chicago Press. Holmstrom, B. (1985). The Provision of Services in a Market Economy. In: R. Inman (Ed.), Managing the Service Economy: Problems and Prospects. London: Cambridge University Press. Inman, R. (Ed.) (1985). Managing the Service Economy: Problems and Prospects. London: Cambridge University Press. Juarrero, A. (1999). Dynamics in Action. Cambridge: MIT Press. Lovelock, C. (1988). Managing Services: Marketing, Operations, and Human Resources. Englewood Cliffs, NJ: Prentice-Hall. Nabli, M., & Nugent, J. (1989). The New Institutional Economics and Development. Amsterdam: North-Holland Press. Nordhaus, W., 1997. Do Real Output and Real Wage Measures Capture Reality? The History of Light Suggests Not. In: T. Bresnahan & R. Gordon (Eds), The Economics of New Goods. Chicago: The University of Chicago Press. North, D. (1990). Institutions, Institutional Change, and Economic Performance. London: Cambridge University Press. Oi, W. (1992). Productivity in the Distributive Trades: The Shopper and the Economics of Massed Reserves. In: Z. Griliches (Ed.), Output Measurement in the Service Sector. Chicago: University of Chicago Press.

Product Innovation in Services: A Framework for Analysis

183

Oi, W. (1997). Welfare Implications of a New Product: The Air Conditioner. In: T. Bresnahan & R. Gordon (Eds), The Economics of New Goods. Chicago: University of Chicago Press. Salter, W. E. G. (1969). Productivity and Technical Change. London: Cambridge University Press. Scazzieri, R. (1993). A Theory of Production: Tasks, Processes, and Technical Practices. Oxford: Oxford University Press. Shannon, C. (1948). The Mathematical Theory of Communication. Bell System Technical Journal, 27, 623–656. Shannon, C., & Weaver, W. (1949). The Mathematical Theory of Communication. Urbana, IL: University of Illinois Press. U.S. Bureau of the Census (2000). Statistical Abstract of the United States. Washington, D.C.: Department of Commerce. Wernerfelt, B. (1994). An Efﬁciency Criterion for Marketing Design. Journal of Marketing Research, 31, 462–470. Williamson, O. (1985). The Economic Institutions of Capitalism: Firms, Markets, Relational Contracting. New York: Free Press. Zeithaml, V., Pasuraman, A., & Berry, L. (1985). Problems and Strategies in Services Marketing. Journal of Marketing, 49, 33–46.

This Page Intentionally Left Blank

ASYMPTOTIC EFFICIENCY IN STACKELBERG MARKETS WITH INCOMPLETE INFORMATION Jianbo Zhang and Zhentang Zhang ABSTRACT This paper examines the asymptotic (in)efﬁciency of Stackelberg markets with incomplete information. Firms who are early in the queue make their quantity choices based on limited information and their output choices are likely to deviate from those optimal under complete information. Due to the presence of both payoff externality and information externality, the output deviations of early ﬁrms have a lasting effect on all subsequent output decisions. Consequently, the total market output diverges from the competitive equilibrium output even as the number of ﬁrms goes to inﬁnity. That is, Stackelberg markets with incomplete information are asymptotically inefﬁcient with probability one.

I. INTRODUCTION It is well known that markets may fail because of limited competition or the existence of incomplete information.1 The inefﬁciency due to these two sources, however, can be respectively eliminated in a large market as the number of ﬁrms tends to inﬁnity. Wilson (1977) demonstrates that in a sealed bid tender auction where each bidder has private information, the winning bid

Advertising and Differentiated Products, Volume 10, pages 185–202. 2001 by Elsevier Science Ltd. ISBN: 0-7623-0823-0

185

186

JIANBO ZHANG AND ZHENTANG ZHAN

will converge in probability to the true value of the object as the number of bidders grows toward inﬁnity. This result is extended and generalized by Milgrom (1979), who obtains necessary and sufﬁcient conditions for convergence.2 Regarding the inefﬁciency from market power, Novshek (1980) and Robson (1990) investigate Cournot and Stackelberg markets respectively, and show that under general demand and U-shaped average cost curves, the inefﬁciency arising from market power will disappear asymptotically as the minimum efﬁcient scale tends to zero. That is, both Cournot and Stackelberg equilibria under complete information converge to the competitive equilibrium as the number of ﬁrms tends to inﬁnity. A natural question to raise is whether inefﬁciency in markets with both market power and incomplete information is also eliminated as the number of ﬁrms converges to inﬁnity? The asymptotic property of the Cournot model with incomplete information has been investigated in the literature. Palfey (1985) shows that, under certain assumptions on the information structure, a Cournot market with unknown demand becomes efﬁcient as the number of ﬁrms goes to inﬁnity. Li (1985) obtains the same result by endogenizing ﬁrms’ decision to share information.3 Vives (1988) demonstrates that Palfey and Li’s result depends on the production technology exhibiting constant returns to scale. However, so far in the literature, no studies have looked at whether the inefﬁciency in Stackelberg markets with incomplete information could be eliminated as the number of ﬁrms tends to inﬁnity. The question is whether large Stackelberg markets aggregate information efﬁciently? This paper attempts to ﬁll this gap. The information structure arising in Stackelberg competition is similar to the one in information cascade literature introduced by Banerjee (1992), and Bikchandani, Hirshleifer, and Welch (1992). Under such an information structure, agents take actions sequentially after observing the action history and a private signal. An information externality occurs since each agent’s private information is revealed, perfectly or imperfectly, through its action to the following agents and may thereby alter their beliefs about the underlying uncertainty. This information externality may give rise to information cascades.4 Although the literature on information cascade illustrates how inefﬁciencies can be generated in a sequential action model through information externality, it is not well suited to analyze the information aggregation problem in large Stackelberg markets. The reason for this is that information cascade models assume that there is no strategic value for a player early in the queue to manipulate its action in order to inﬂuence the actions of the following players. That is, they assume that there are no strategic interactions between players and thereby no payoff externality.5 By assuming

Asymptotic Efﬁciency in Stackelberg Markets with Incomplete Information

187

away the strategic interactions, the information cascade literature captures the inefﬁciencies resulting only from information externalities but not from payoff externalities.6 In a Stackelberg market with incomplete information, the effects of every ﬁrm’s action on the payoffs of its successive ﬁrms are two-fold. First, the strategic interactions between leaders and followers create payoff externality. Second, every ﬁrm’s action conveys its private information and thereby affects its following ﬁrms’ belief about the unknown state, which creates information externality. It is this payoff externality interwoven with the information externality, as shown below, that drives the efﬁciency loss even as the number of ﬁrms goes to inﬁnity. The main result of this paper can be illustrated by considering the following scenario: a number of ﬁrms engage in Stackelberg competition making their production choices sequentially. The nature of demand is unknown to the ﬁrms. In addition to receiving a private signal, each ﬁrm observes all the actions of the preceding ﬁrms and tries to infer their private information through these actions. Based on the private signal and the inferred public information, every ﬁrm makes its quantity choice. In general, this game is an extended signaling game with many players where the quantity choice of each leader is a signal about its private information to all its followers.7 Remember that dynamic games with incomplete information tend to have multiple equilibria and there is no exception in this game.8 The reﬁnement of perfect Bayesian equilibrium adopted in this paper is called the extended intuitive criterion, which is an extended version of the intuitive criterion of Cho and Kreps (1987). More precisely, the intuitive criterion is applied to every continuation game of the extended signaling game. Since every continuation game satisﬁes the single crossing property, the extended intuitive criterion leaves us with a unique separating equilibrium. This implies that every ﬁrm’s quantity choice fully reveals its private information to all its followers. Therefore, according to the strong law of large numbers, the true state of demand is eventually revealed to the ﬁrms who are further back in the queue. The basic point is as follows: in the information cascade models with no payoff externality, the revelation of the truth necessarily forces the later players to take actions that are optimal under complete information. This cannot happen in the current model precisely due to the payoff externality that is present. Intuitively, the ﬁrst ﬁrm can get either a high or a low signal with a positive probability regardless of what the true state of demand is. As a consequence, its quantity choice may be different from the quantity choice under complete information. The quantity choice of the ﬁrst ﬁrm will then affect the quantity choices of all the ﬁrms later in the queue, due to the payoff externality. As a result, output deviations of early ﬁrms have a lasting effect on all subsequent output decisions and the total market output

188

JIANBO ZHANG AND ZHENTANG ZHAN

does not converge to the competitive equilibrium. This is true despite the fact that ﬁrms further back in the queue have almost complete information about demand.9 The rest of the paper is organized as follows. In Section II we set up the model. In Section III we investigate the asymptotic (in)efﬁciency of the Stackelberg market. In Section IV we make some additional comments and conclude the paper.

II. MODEL Consider a Stackelberg market with N ﬁrms making production choices sequentially. Firms are assumed to have constant marginal cost; i.e. c(qn) = cqn, n = 1, 2, . . . , N. There are no ﬁxed costs. Following Novshek and Sonnenschein (1982) and Vives (1988), ﬁrms’ private information is about the demand intercept: p = a + S ⫺ bQ; where Q is total market output and it is assumed that a > c and b > 0. State S is a random variable which is distributed over a ﬁnite state space ⍀ 傺 R 1. The unknown state could be interpreted as a parameter affecting consumers’ taste, so that a higher state would result in a higher market demand and vice versa. Firms do not know the realization s of S, but have common initial prior distribution ␮0(s) = prob(S = s), which is assumed to be non-degenerate. The information structure of the game is as follows: at the beginning of the game, a state s is drawn randomly from the ﬁnite state space ⍀ and remains ﬁxed throughout. Each ﬁrm makes output choice based on its private signal and the public information in order to maximize its proﬁts. Following Lee (1993), the private signal of ﬁrm n, xn, is drawn randomly from a Bernoulli distribution xn苸{0, 1}. The draw of a signal is conditionally i.i.d. given the state. ¯ 10 The fact that AN Firm n chooses qn from its action set AN, where AN = [0, Q]. is compact and convex guarantees the existence of equilibrium. Given the state of nature s, the information set of ﬁrm n is given by ⍀n = {hn, xn}, where hn = (q1, q2, . . . , qn ⫺ 1) (h1 = ␾) is the history of actions and xn is the private signal of ﬁrm n. The behavior strategy of ﬁrm n, ␴n(qn | (hn, xn)) is a mapping from the information set to the action set. Each ﬁrm, before making its output choice, has an identical initial prior belief ␮0(s) over the states. After observing the output choices of its preceding ﬁrms as well as receiving a private signal, the ﬁrm updates its prior belief according to Bayes’ rule. The output choice is optimal with respect to the posterior belief. Let ␮n = prob(s | (hn, xn)) denote the posterior belief of ﬁrm n. Firm n’s expected proﬁt function is given by E␮n␲n = E{(a + S ⫺ c ⫺ bQ)qn | ⍀n}, where Q is the total market output. We solve for perfect Bayesian equilibrium.

Asymptotic Efﬁciency in Stackelberg Markets with Incomplete Information

189

Given the number of ﬁrms N and the realized signal vector xជ = (x1, x2, . . . , xN), a perfect Bayesian equilibrium (hereafter, PBE) is deﬁned as follows: Deﬁnition: A PBE is a set of strategies (␴*1, . . . , ␴*N) and posterior beliefs (␮1, . . . , ␮N) such that for any ␮0, hn and n = 1, 2, . . . , N, (P1) ␴*n(qn | (hn, xn))苸arg max E␮n␲n(hn, qn, q*n + 1(qn), . . . , q*N(qn)); qn

(P2) ␴*n + 1(qn)苸arg max E␮n+1␲n + 1(hn + 1, qn + 1, q*n + i + 1(qn + i), . . . , q*N(qn + i)), for any qn+1

i = 1, 2, . . . , N ⫺ n; (B) ␮n(s | (hn, xn)) is derived from the prior ␮0(s), ␴*n ⫺ 1(qn ⫺ 1 | (hn ⫺ 1, xn ⫺ 1)) and ␴*n ⫺ 1 + i(qn ⫺ 1) according to Bayes’ rule, when applicable.11 That is, a PBE of an extended signaling game with N players requires that the strategies yield a PBE for every continuation game. To simplify the analysis, we make the following assumptions. Assumption 1: The signals are unbiased estimators of the state. That is, E(xn | s) = s. From Assumption 1, we have that s = prob(xn = 1 | s) and 1 ⫺ s = prob(xn = 0 | s). That is, it is more likely to get a good signal (xn = 1) when the state of nature is good. Assumption 2: The conditional probability of signal, xn = 1, is strictly between 0 and 1. That is, 0 < prob(xn = 1 | s) < 1. Assumption 2 implies that for any given state, there is a positive probability that each ﬁrm gets either signal xn = 1 or signal xn = 0. Assumption 2 rules out degenerate cases. To extend the spirit of subgame perfection to this game, we would like to require that the strategies yield PBE for every continuation game starting from every possible history hn. We make the following assumptions on players’ beliefs at the start of each continuation game. Assumption 3: For any history hn, player n + 1 to player N have the same interim beliefs about the state of nature given hn. Recall that dynamic games with incomplete information tend to have multiple equilibria because Bayes’ rule has no bite in out-of-equilibrium events and any posterior beliefs are admissible at out-of-equilibrium information sets. Consequently, the game above will potentially have multiple equilibria, unless we make use of a reﬁnement. Using the intuitive criterion introduced by Cho

190

JIANBO ZHANG AND ZHENTANG ZHAN

and Kreps (1987), we propose an “extended intuitive criterion” which means that we apply the intuitive criterion of Cho and Kreps (1987) to every continuation game. In other words, we extend the intuitive criterion to our extended signaling game with many players.

III. STACKELBERG COMPETITION WITH INCOMPLETE INFORMATION We can now investigate whether a Stackelberg market with incomplete information is asymptotically efﬁcient in the presence of both information and payoff externalities. There are N ﬁrms making their output choice sequentially in a hierarchical Stackelberg framework. Without loss of generality, the order of ﬁrms’ actions is assumed to be exogenously given by 1, 2, . . . , N.12 Let us examine pooling equilibrium ﬁrst. At a pooling equilibrium, ﬁrm n produces the same output no matter what signal he gets (0 or 1). The followers of ﬁrm n, ﬁrm n + 1 to ﬁrm N, therefore update their posterior believes only based on their private signals. This implies that for any n, the information set of ﬁrm n is simply ⍀n = {xn}. Hence, ﬁrm n’s expected payoff function can be simpliﬁed to E␮n␲n = E{(a + S ⫺ c ⫺ bQ)qn | xn}. The existence and reﬁnement of the pooling equilibria are given in the following proposition: Proposition 1: Given the number of ﬁrms N and the realized signal vector xជ = (x1, x2, . . . , xN), there exists a continuum of pooling PBE. Furthermore, the extended intuitive criterion eliminates all the pooling equilibria. Proof: At a pooling equilibrium, each ﬁrm updates its posterior belief only based on its own private signal, i.e. ⍀n = {xn} for n = 1, . . . , N. For any given N and history hn, a pooling equilibrium is said to survive the extended intuitive criterion if it survives intuitive criterion of Cho and Kreps (1987) in every continuation game. We solve this game backward. Step 1. Continuation Game N For any given hN ⫺ 1, this continuation game consists of only the Nth ﬁrm whose equilibrium output is given by q*N苸arg max E [qN(a ⫺ c + S ⫺ bQ) | xN] qN

(A1)

Asymptotic Efﬁciency in Stackelberg Markets with Incomplete Information

191

Step 2. Continuation Game N ⫺ 1 For any given hN ⫺ 2, this continuation game consists of ﬁrm (N ⫺ 1) and ﬁrm N. Let qN ⫺ 1 denote a pooling equilibrium for ﬁrm N ⫺ 1. Firm (N ⫺ 1)’s expected payoff is

冘

E[qN ⫺ 1(a ⫺ c + S ⫺ bQN ⫺ 1 ⫺ bq*N) | xN ⫺ 1],

N⫺1

where QN ⫺ 1 =

qi and q*N(qN ⫺ 1) is given by (A1).

i=1

Thus, the best way to sustain qN ⫺ 1 as a pooling equilibrium is to assume that ﬁrm N believes that ﬁrm N ⫺ 1 gets signal xN ⫺ 1 = 1 when it observes q⬘N ⫺ 1 ≠ qN ⫺ 1. So qN ⫺ 1 will indeed be a pooling equilibrium if and only if the following conditions are satisﬁed (M1): E[␲(qN ⫺ 1, q*N) | xN ⫺ 1 = 1] ≥ max E[␲(q⬘N ⫺ 1, q⬘* N ) | xN ⫺ 1 = 1], q⬘N⫺1

(M2): E[␲(qN ⫺ 1, q*N) | xN ⫺ 1 = 0] ≥ max E[␲(q⬘N ⫺ 1, q⬘* N ) | xN ⫺ 1 = 0], q⬘N⫺1

where q*N(qN ⫺ 1) is given by (A1) and q⬘* N (q⬘N ⫺ 1) is given as follows: q⬘* N 苸arg max E[q⬘N(a ⫺ c + S ⫺ bQ) | (xN ⫺ 1 = 1, xN)]. q⬘N

Therefore, there exists a continuum pooling equilibria qN ⫺ 1苸[qN ⫺ 1, qN ⫺ 1], where qN ⫺ 1 and qN ⫺ 1 are the lower and upper bound of qN ⫺ 1 which satisﬁes (M1) and (M2). In order to eliminate this continuum of pooling equilibria, we apply the intuitive criterion of Cho and Kreps (1987) to this continuation game. Deﬁne q⬘N ⫺ 1 < qN ⫺ 1 by the smallest root of E[␲(q⬘N ⫺ 1, r*N) | xN ⫺ 1 = 1] = E[␲(qN ⫺ 1, q*N) | xN ⫺ 1 = 1], where r*N苸arg max E[rN(a ⫺ c + S ⫺ bQN ⫺ 2 ⫺ bq⬘N ⫺ 1 ⫺ brN) | (xN ⫺ 1 = 0, xN)]. rN

Now, playing q⬘N ⫺ 1 ⫺ ␧ (for ␧ > 0) is equilibrium dominated for ﬁrm N ⫺ 1 with signal xN ⫺ 1 = 1, but not for ﬁrm N ⫺ 1 with signal xN ⫺ 1 since E[␲(q⬘N ⫺ 1, q*N) | xN ⫺ 1 = 0] ⫺ E[␲(qN ⫺ 1, q*N) | xN ⫺ 1 = 0] =

E(S | (xN ⫺ 1 = 1, xN)) ⫺ E(S | (xN ⫺ 1 = 0, xN)) > 0. 2b

192

JIANBO ZHANG AND ZHENTANG ZHAN

Therefore, ﬁrm N’s posterior belief should put all the weight on xN ⫺ 1 = 0 following output q⬘N ⫺ 1 ⫺ ␧. However, ﬁrm N ⫺ 1 who gets xN ⫺ 1 = 0 prefers to play q⬘N ⫺ 1 ⫺ ␧ to qN ⫺ 1. Thus, qN ⫺ 1 is not pooling output anymore. Step 3. Continuation Game N ⫺ 2 For any given hn ⫺ 3, this game consists of ﬁrm N ⫺ 2, ﬁrm N ⫺ 1 and ﬁrm N. From assumption 3, ﬁrm N ⫺ 1 and ﬁrm N have the same beliefs after observing qN ⫺ 2. Applying the similar argument and technique used in the last continuation game to this continuation game, we can eliminate pooling equilibria in this continuation game. Continuing this process for every continuation game, we will then eliminate all the pooling equilibria. Q.E.D. Next, we study separating equilibrium. For a separating equilibrium, every ﬁrm’s quantity choice perfectly reveals its private information. Therefore, ﬁrm n’s information set ⍀n can be reduced to ⍀n = {x1, x2, . . . , xn ⫺ 1, xn}. Consequently, the expected payoff function of ﬁrm n is given by E␮n␲n = E{(a + S ⫺ c ⫺ bQ)qn | (x1, . . . , xn)}. The existence and the reﬁnement of the separating equilibria are given by the following proposition: Proposition 2: Given the number of ﬁrms N and the realized signal vector xជ = (x1, x2, . . . , xN), there exists a continuum of separating PBE. Furthermore, the extensive intuitive criterion eliminates all but one separating equilibrium which is given as follows:

冉冊

1 Q(N, xជ ) = 1 ⫺ N 2

a⫺c 1 + b b

冘 N

n=1

E(S | ⍀n) , 2n

p(N, xជ ) = a ⫺ bQ(N, xជ ) + S.

(1) (2)

Proof: See Appendix. Proposition 1 and 2 imply that the extended intuitive criterion equilibrium reﬁnement leaves us with a unique PBE. Before proceeding to investigate the asymptotic (in)efﬁciency of Stackelberg markets in the presence of both information and payoff externalities, we present the following useful lemmas. Lemma 1: Along the unique PBE path, the best responses of ﬁrm n’s followers, ﬁrm n + 1 to ﬁrm N, satisfy respectively dqn + 1 1 dqn + 2 1 dqN 1 =⫺ , =⫺ , . . . , = ⫺ N ⫺ n; dqn 2 dqn 4 dqn 2

Asymptotic Efﬁciency in Stackelberg Markets with Incomplete Information

193

where n = 1, 2, . . . , N ⫺ 1, and ⍀n = {x1, x2, . . . , xn ⫺ 1, xn}. Proof: See Appendix. From Lemma 1, it is trivial to show that

冉冊

dQ 1 = dqn 2

N⫺n

. This is an important

result as it implies that the impact of a ﬁrm’s output choice on total output is smaller the further back in the game the ﬁrm is. Conversely, the output choices of early ﬁrms have a lasting effect on all subsequent output decisions. In order to have a base for comparison, we use the perfect competitive equilibrium outcome with complete information as a benchmark.13 The following lemma is trivial to obtain. Lemma 2: Let (Q0(s), p0(s)) be the competitive equilibrium outcome with a⫺c+s and p0(s) = c. complete information. Then Q0(s) = b We are now ready to state our main result regarding the asymptotic (in)efﬁciency of Stackelberg markets with incomplete information. Theorem: For any realization s of S. Let (Q(N | s), p(N | s)) be the vector of random variables which represents the unique (stochastic) PBE given by (1) and (2). Then E(Q(N | s), p(N | s)) converges to some random vector (Q*(s), p*(s)) as N goes to inﬁnity. For almost all realizations s, (Q*(s), p*(s)) ≠ (Q0(s), p0(s)).14 Proof: From Proposition 2, for any N and xជ, there exists a unique PBE given by Eqs (1) and (2). For any realization s of S, it is trivial that

冉冊

1 Q(N | s) = 1 ⫺ N 2

a⫺c 1 + b b

冘 N

n=1

[E(S | ⍀n) | s] , 2n

and p(N | s) = a ⫺ bQ(N | s) + s; where ⍀n = {x1, x2, . . . , xn ⫺ 1, xn}. As N → ⬁ , according to the Martingale Convergence Theorem a⫺c 1 EQ(N | s) → + b b

冘 ⬁

n=1

E[E(S | ⍀n) | s] ⬅ Q*(s), 2n

194

JIANBO ZHANG AND ZHENTANG ZHAN

and

冘 ⬁

Ep(N | s) → c + s ⫺

n=1

E[E(S | ⍀n) | s] ⬅ p*(s). 2n

Let s* be the unique solution to the following equation

冘 ⬁

s=

n=1

E[E(S | ⍀n) | s] . 2n

It is then trivial that (Q*(s), p*(s)) ≠ (Q0(s), p0(s)) unless s = s*.

Q.E.D.

The Theorem shows that the Stackelberg output is inefﬁciently low (Q*(s) < Q0(s)) when the true state of demand is favorable (s > s*), while it is inefﬁciently high (Q*(s) > Q0(s)) when the true state of demand is unfavorable (s < s*). In sum, the Stackelberg output is asymptotically inefﬁcient with probability one.15 The intuition behind the above result is as follows. Firms make their production decisions sequentially based on their private information as well as the inferred public information of the preceding ﬁrms. Since every continuation game of the Stackelberg game satisﬁes the single crossing property, the extended intuitive criterion selects a separating equilibrium, which implies that the leaders’ quantity choices fully reveal their private information. Consequently, the ﬁrms who are sufﬁciently back in the queue have almost complete information, according to the Martingale Convergence Theorem. However, the ﬁrms who are early in the queue have very limited information about the unknown demand and their quantity choices tend to be different from the choices under complete information. In addition, these early ﬁrms’ production choices affect the output choices of the later ﬁrms due to the payoff externality existing in the game. Therefore, the deviations of the early ﬁrms’ output choice have a lasting effect on all subsequent output decisions and causes the total market output to be diverging from the competitive equilibrium output even as the number of ﬁrms goes to inﬁnity. The above theorem can be further illustrated in the following example. Example: Suppose the initial prior distribution ␮0(s) is the uniform distribution over (0, 1). The expected value of posterior distribution in the unique separating PBE can be simpliﬁed as

冋冘册

1 1+ E(S | ⍀n) = n+2

N

xn ,

n=1

Asymptotic Efﬁciency in Stackelberg Markets with Incomplete Information

where ⍀n = {x1, x2, . . . , xn ⫺ 1, xn}.16

冘 ⬁

Therefore,

n=1

E[E(S | ⍀n) | s] = 2n

a⫺c 1 + Hence, Q(N | s) → b b

冘 ⬁

n=1

冘 ⬁

n=1

195

ns + 1 by Assumption 1. 2n(n + 2)

ns + 1 ; p(N | s) → c + s ⫺ 2n(n + 2)

冉

From Taylor expansion ln(1 ⫺ r) = ⫺ r +

冘冊 ⬁

n=1

r2 rn +. . .+ +. . . 2 n

ns + 1 . 2n(n + 2)

for ⫺ 1 < r < 1,

we have that

冘 ⬁

n=1

1 ns + 1 = (6 ⫺ 8 ln 2)s + (4 ln 2 ⫺ 2.5) ≈ 0.46s + 0.27. 2n n + 2

Therefore, Q(N | s) →

a⫺c 1 + (0.46s + 0.27) ⬅ Q*(s), and b b

p(N | s) → c + (0.54s ⫺ 0.27) ⬅ p*(s). 1 1 It is trivial that Q*(s) < Q0(s) if s > , Q*(s) > Q0(s) if s < , and otherwise 2 2 Q*(s) = Q0(s).

IV. CONCLUDING REMARKS In this paper, we have demonstrated that large Stackelberg markets do not aggregate information efﬁciently, even if the technology exhibits constant return to scale. That is, in the presence of incomplete information, Stackelberg markets are asymptotically inefﬁcient with probability one. This is because the early ﬁrms make their production choices based on very limited information and consequently tend to over- or under-produce. In addition, the payoff externality ensures that the quantity choices of the early ﬁrms have a lasting effect on the output decisions of all subsequent ﬁrms. As a result, the over- or under-production of the early ﬁrms gets carried over and drives the efﬁciency loss. The extended intuitive criterion selects a unique separating equilibrium, which ensures that each ﬁrm’s private information is fully revealed to the successive ﬁrms and accordingly the underlying uncertainty is gradually

196

JIANBO ZHANG AND ZHENTANG ZHAN

resolved along the queue as the number of ﬁrms becomes larger. Therefore, ﬁrms who are sufﬁciently far back in the queue have almost complete information about the demand. In this sense, there is no efﬁciency loss from the information externality per se, and there is no possibility for a non-fully revealing information cascade to occur. It would be interesting to investigate a class of games where agents’ actions do not fully reveal their private information.17 In this case, a non-fully revealing information cascade may arise as discussed in Banerjee (1992), Bikchandani, Hirshleifer, and Welch (1992), and Zhang and Zhang (1995). The efﬁciency loss in these games is therefore expected to be larger due to the additional inefﬁciency from information externalities.

NOTES 1. See, for example, Akerlof (1980) for a classical discussion of the problem. 2. Swinkels (1996) shows that discriminatory private value auctions for multiple objects are asymptotically efﬁcient as the number of players grows. 3. The incentives for Cournot oligopolists to share information have been studied extensively in the literature, for example, by Li (1985), Shapiro (1986), and Vives (1984). They conclude that when the uncertainty is about a ﬁrm-speciﬁc parameter, perfect revelation is the unique equilibrium. On the other hand, when the uncertainty is about a common parameter, no information sharing is the unique equilibrium. 4. An information cascade is deﬁned as the convergence of the sequence of actions (Lee, 1993). A fully revealing information cascade is said to occur if the limit is optimal under the true state. Otherwise, a non-fully revealing information cascade occurs. 5. Payoff externality refers to that situation where a player’s action not only affects his own payoff but also other agents’ payoffs. 6. Similarly, Vives (1993) studies the speed of convergence to the rational expectations equilibrium in a simple dynamic model of rational learning between agents. However, he assumes that the action of one player does not affect the proﬁts of other periods. That is, he assumes that there are no strategic interactions between players across periods. 7. In this extended signalling game, every player has private information while in the standard signalling games, only the ﬁrst player has private information. 8. To obtain a unique equilibrium a reﬁnement is necessary. 9. In our set-up, we do not assume that there is a designed mechanism which solicit information before implementing the transactions as in Gul and Postlewaite (1992). Neither do we assume there is a market auctioneer who pools information and sets market clearing price as in Rutichini, Satterthwaite and Williams (1994). Instead, the private information is revealed through the quantity choices of the ﬁrms.

冋

冘 N

册

1 1 a⫺c+ [E(S | ⍀n)] . b N n=1 This is because any choice of quantity greater than Q results in a strict expected loss, regardless of other ﬁrms’ choices.

10. The upper bound of ﬁrm n’s output is given by Q =

Asymptotic Efﬁciency in Stackelberg Markets with Incomplete Information

197

11. Note that if qn ⫺ 1 is not part of ﬁrm (n–1)’s optimal strategy for some private signal, observing qn ⫺ 1 is a probability 0 event, and Bayes’ rule does not pin down posterior beliefs. Any posterior beliefs are then admissible. 12. It is an interesting problem to endogenize the order of ﬁrms’ actions. Mailath (1993) studies endogenous sequencing of ﬁrm decisions in a duopoly setup with asymmetric information between ﬁrms. Chamley and Gale (1994) and Zhang (1997) endogenize the sequential choice of agents in an information cascade context. 13. Our main results would not change if we use the competitive equilibrium outcome under incomplete information as the benchmark since the stochastic competitive equilibrium outcome converges in probability to a degenerate distribution, as the number of ﬁrms goes to inﬁnity (see Vives, 1988). 14. We are grateful to an anonymous referee for the way this theorem is now stated. 15. The fact that the stochastic Stackelberg equilibrium outcome does not converge to the competitive equilibrium outcome is equivalent to stating that there is no convergence to a degenerate distribution since the degenerate Stackelberg equilibrium outcome converges to the competitive equilibrium outcome (see Robson, 1990). 16. See Welch (1992). 17. One example of this class of games is a Stackelberg game where each ﬁrm can only observe some but not all its preceding ﬁrms’ actions. 18. We have implicitly assumed that ﬁrms have rational expectations, i.e. E(qi(⍀i) | ⍀n) = qi; i > n. 19. We therefore assume that a⫺c>2

冋

N

冘 N

min(E(S | ⍀1), . . . , E(S | ⍀N)) ⫺

n=1

E(S | ⍀n) 2n

册

in order to guarantee that q*n > 0.

ACKNOWLEDGMENTS We would like to thank Larry Samuelson for his insightful discussions on the topic of this paper. We would also like to thank Andreas Blume, Marc Dudey, Dan Kovenock for comments on earlier drafts, and seminar participants at the Southern Economic Association Meetings, E.A.R.I.E. in Leuven, and WZB. Finally, we would like to thank the editor and an anonymous referee for their helpful comments.

REFERENCES Akerlof, G. A. (1970). The market for lemons: Quality uncertainty and the market mechanism. Quarterly Journal of Economics, 84, 488–500. Banerjee, A. (1992). A simple model of herd behavior. Quarterly Journal of Economics, 107, 797–817. Bikchandani, S., Hirshleifer, D., & Welch, I. (1992). A theory of fads, fashion, custom, and cultural changes as information cascades. Journal Political Economy, 100, 992–1026.

198

JIANBO ZHANG AND ZHENTANG ZHAN

Chamley, C., Gale, D., 1994. Information revelation and strategic delay in a model of investment. Econometrica, 62, 1065–1085. Cho, I. K., & Kreps, D. (1987). Signaling games and stable equilibria. Quarterly Journal of Economics, 102, 179–221. Gul, F., & Postlewaite, A. (1992). Asymptotic efﬁciency in large exchange economies with asymmetric information. Econometrica, 60, 1273–1292. Lee, I. H. (1993). On the convergence of informational cascades. Journal of Economic Theory, 61, 395–411. Li, L. (1985). Cournot oligopoly with information sharing. Rand Journal of Economics, 16, 521–536. Mailath, G. (1993). Endogenous sequencing of ﬁrm decisions. Journal of Economic Theory, 59, 169–182. Milgrom, P. R. (1979). A convergence theorem for competitive bidding with differential information. Econometrica, 47, 679–688. Novshek, W. (1980). Cournot equilibrium with free entry. Review of Economic Studies, 47, 473–486. Novshek, W., & Sonnenschein, H. (1982). Fulﬁlled expectations Cournot duopoly with information acquisition and release. Bell Journal of Economics, 13, 214–218. Palfrey, T. R. (1985). Uncertainty resolution, private information aggregation and the Cournot competitive limit. Review of Economic Studies, 52, 69–83. Robson, A. J. (1990). Stackelberg and Marshall. American Economic Review, 80, 69–82. Shapiro, C. (1986). Exchange of cost information in oligopoly. Review of Economic Studies, 53, 433–446. Swinkels, J. (1996). Asymptotic efﬁciency for discriminatory private value auctions. Northwestern University Discussion Paper No. 1173. Vives, X. (1988). Aggregation of information in large Cournot markets. Econometrica, 56, 851–876. Vives, X. (1993). How fast do rational agents learn? Review of Economic Studies, 60, 329–347. Welch, I. (1992). Sequential sales, learning and cascades. Journal of Finance, 47, 695–733. Wilson, R. (1977). A bidding model of perfect competition. Econometrica, 44, 511–518. Zhang, J. (1997). Strategic delay and the onset of investment cascades. Rand Journal of Economics, 28, 188–205. Zhang, J., & Zhang, Z. (1995). Information externality, limited observations and the emergence of truth in sequential decisions. WZB Working Papers, FS IV, pp. 95–25.

APPENDIX Proof of Proposition 2: For a separating equilibrium, every ﬁrm’s quantity choice perfectly reveals its private information. Hence, the information set of ﬁrm n can be simpliﬁed to ⍀n = {x1, x2, . . . , xn ⫺ 1, xn}. In addition, a separating equilibrium is said to survive the extended intuitive criterion if it survives intuitive criterion of Cho and Kreps (1987) in every continuation game. We solve this game backward.

Asymptotic Efﬁciency in Stackelberg Markets with Incomplete Information

199

1. Continuation game N For any given hN ⫺ 1, this game consists of only ﬁrm N whose equilibrium output is given by q*N苸arg max E[qN(a ⫺ c + S ⫺ bQ) | ⍀N], where ⍀N = {x1, x2, . . . , xN}. qN

(A2)

2. Continuation game N ⫺ 1 For any given hN ⫺ 2, this game consists of ﬁrm N ⫺ 1 and ﬁrm N. At a separating equilibrium, the private information of ﬁrm N ⫺ 1 is fully revealed to ﬁrm N through its quantity choice. Therefore, for ﬁrm N ⫺ 1 with signal xN ⫺ 1 = 1, it chooses the following optimal quantity q*NH⫺ 1苸arg max E[qN ⫺ 1(a ⫺ c + S ⫺ bQN ⫺ 1 ⫺ bq*NH(qn ⫺ 1)) | (⍀N ⫺ 2, xN ⫺ 1 = 1)], qN⫺1

冘

(A3)

N⫺1

where QN ⫺ 1 =

qi and q*NH(qN ⫺ 1) is given by (A2) with xN ⫺ 1 = 1. That is

i=1

q*NH(qN ⫺ 1)苸arg max E[qN(a ⫺ c + S ⫺ bQ) | (⍀N ⫺ 2, xN ⫺ 1 = 1, xN)]. qN

(A4)

On the other hand, for ﬁrm N ⫺ 1 with signal xN ⫺ 1 = 0, a separating equilibrium qN ⫺ 1 is such that the following conditions are satisﬁed jointly (S1): max E[␲(qN ⫺ 1, q*NH) | (⍀N ⫺ 2, xN ⫺ 1 = 1)] ≥ E[␲(qN ⫺ 1, r*N) | (⍀N ⫺ 2, xN ⫺ 1 = 1)], qN⫺1

(S2): E[␲(qN ⫺ 1, r*N) | (⍀N ⫺ 2, xN ⫺ 1 = 0)] ≥ max E[␲(qN ⫺ 1, q*NH) | (⍀N ⫺ 2, xN ⫺ 1 = 0)], qN⫺1

where q*NH is given by (A4) and r*N苸arg max E[qN(a ⫺ c + S ⫺ bQ) | (⍀N ⫺ 2, xN ⫺ 1 = 0, xN)]. qN

(S1) says that when ﬁrm N ⫺ 1 gets signal xN ⫺ 1 = 1, it does not want to produce the output which corresponds to signal xN ⫺ 1 = 0. (S2) says that when ﬁrm N ⫺ 1 gets signal xN ⫺ 1 = 0, it does not want to produce output which conveys signal xN ⫺ 1 = 1. Therefore, there exists a continuum separating equilibria qLN ⫺ 1苸[qLN ⫺ 1, qLN ⫺ 1], where qLN ⫺ 1 and qLN ⫺ 1 are the lower and upper bound of qLN ⫺ 1 which satisﬁes (S1) and (S2). Therefore, there exists a continuum separating equilibria. The ﬁrm with signal xN ⫺ 1 = 1 prefers playing q*HN ⫺ 1 while the ﬁrm with signal xN ⫺ 1 = 0 prefers playing qLN ⫺ 1苸[qLN ⫺ 1, qLN ⫺ 1]. From (S1), it is clear that playing qLN ⫺ 1 is equilibrium dominated for the ﬁrm with signal xN ⫺ 1 = 1, but not for the ﬁrm

200

JIANBO ZHANG AND ZHENTANG ZHAN

with signal xN ⫺ 1 = 0. So ﬁrm N’s posterior belief should put all the weight on signal xN ⫺ 1 = 0 following qLN ⫺ 1. Let q*NL⫺ 1 denote q*NL⫺ 1苸arg max E[␲(qN ⫺ 1, r*N) || (⍀N ⫺ 2, xN ⫺ 1 = 0)] qN⫺1

L N⫺1

Then q* is the unique separating equilibrium surviving the elimination of weakly dominated strategies for ﬁrm with signal xN ⫺ 1 = 0. Hence, for any hN ⫺ 2, the equilibrium reﬁnement of this continuation game leaves us with a set of unique PBE strategy (q*N ⫺ 1, q*N), where q*N is given by (A2) and q*N ⫺ 1 is given by q*N ⫺ 1苸arg max E[qN ⫺ 1(a ⫺ c + S ⫺ bQN ⫺ 1 ⫺ bq*N) | ⍀N ⫺ 1]. qN⫺1

(A5)

3. Continuation game N ⫺ 2 For any given hn ⫺ 3, this game consists of ﬁrm N ⫺ 2, ﬁrm N ⫺ 1 and ﬁrm N. From assumption 3, ﬁrm N ⫺ 1 and ﬁrm N have the same beliefs after observing qN ⫺ 2. By the similar reasoning as in last continuation game, the elimination of weakly dominated strategies of this continuation game leave us with a set of unique separating PBE proﬁle (q*N ⫺ 2, q*N ⫺ 1, q*N), where q*N is given by (A2), q*N ⫺ 1 is given by (A5) and q*N ⫺ 2 is given by q*N ⫺ 2苸arg max E[qN ⫺ 2(a ⫺ c + S ⫺ bQN ⫺ 2 ⫺ bq*N ⫺ 1 ⫺ bq*N) | ⍀N ⫺ 2]. (A6) qN⫺2

Continuing this process for every continuation game, the extended intuitive criterion leaves us with a unique separating PBE satisfying q*n苸arg max E[qn(a ⫺ c + S ⫺ bQ) | ⍀n], for n = 1, 2, . . . , N. qn

Therefore, along the unique separating PBE path, we have

冘 N

a ⫺ c ⫺ bqn ⫺ bQN ⫺ bqn

冘

dqi + E(S | ⍀n) = 0; dqn i=n+1

N

where QN =

qj and n = 1, 2, . . . , N.18

j=1

Applying Lemma 1 (which is proved below) and rearranging the above equation, we have

冉

q*n = 2N ⫺ n

冊

a ⫺ c + E(S | ⍀n) ⫺ QN ; n = 1, 2, . . . , N.19 b

Asymptotic Efﬁciency in Stackelberg Markets with Incomplete Information

201

Summing over n = 1, 2, . . . , N, we have that for every N and the realized signal vector xជ = (x1, . . . , xN), the unique stochastic PBE which survives the extended intuitive criterion is given by

冉冊

1 Q(N, xជ ) = 1 ⫺ N 2

a⫺c 1 + b b

冘 N

n=1

E(S | ⍀n) . 2n

Thus, p(N, X) = a ⫺ bQ(N, X) + S.

Q.E.D.

Proof of Lemma 1: We prove this lemma by mathematical induction. Along the unique separating PBE path, we have that a ⫺ c ⫺ 2bqN ⫺ bQN ⫺ 1 + E(s | ⍀N) = 0; where

冘 N⫺1

QN ⫺ 1 =

qj .

j=1

Thus, the best response of ﬁrm N to ﬁrm (N ⫺ 1)’s output is satisﬁes that dqN 1 =⫺ . dqN ⫺ 1 2

Now suppose that the lemma holds for ﬁrm n + 1. That is, along the unique separating PBE path, the best responses of ﬁrm (n + 1)’s followers, ﬁrm n + 2 to ﬁrm N, satisfy dqn + 2 1 dqn + 3 1 =⫺ , =⫺ , dqn + 1 2 dqn + 1 4

From the above, it is trivial that

dqN 1 = ⫺ N ⫺ (n + 1) dqn + 1 2

⭸qj 1 = ⫺ for any j ≥ n + 2. ⭸qn + 1 2

(L1) (L2)

Now we want to show the lemma also holds true for ﬁrm n. From the proof of Proposition 2, ﬁrm (n + 1)’s best response along the unique separating PBE path can be derived as follows

冘 N

a ⫺ c ⫺ 2bqn + 1 ⫺ bQn ⫺ b

冘 N

qj ⫺ bqn + 1

j=n+2

where

冘 n

Qn =

j=1

qj .

dqJ + E(S | ⍀n + 1) = 0; dqn + 1 j=n+2

202

JIANBO ZHANG AND ZHENTANG ZHAN

Rearrange it and applying (L1), we have

冋冉冊册 1 1+ 2

冘 N

N⫺n⫺1

bqn + 1 + b

qj = a ⫺ c ⫺ bQn + E(S | ⍀n + 1)

j=n+2

Taking the derivative with respect to qn, we have

冋冉冊册 1+

where

1 2

冘 N

N⫺n⫺1

b

dqn + 1 dqj +b =⫺b dqn dq n j=n+2

(L3)

dqn + 2 1 dqn + 1 1 =⫺ ⫺ from (L2). dqn 2 dqn 2

Similarly, dqn + 3 1 dqn + 2 1 dqn + 1 1 1 dqn + 1 1 =⫺ ⫺ ⫺ =⫺ ⫺ , . . . , and dqn 2 dqn 2 dqn 2 4 dqn 4 dqN 1 dqn + 1 1 = ⫺ N⫺n⫺1 ⫺ N ⫺ n ⫺ 1. dqn 2 dqn 2

冘 N

Therefore,

冋冉冊册冉

dqj 1 =⫺ 1⫺ dqn 2 j=n+2

N⫺n⫺1

1+

冊

dqn + 1 . dqn

Substituting the above back into (L3) and rearrange it, we have that dqn + 1 1 =⫺ . dqn 2

Hence, dqn + 2 1 dqn + 1 1 1 =⫺ ⫺ =⫺ , dqn 2 dqn 2 4 dqn + 3 1 1 dqn + 1 1 =⫺ ⫺ = ⫺ , . . . , and dqn 4 4 dqn 8 dqN 1 1 dqn + 1 1 = ⫺ N⫺n⫺1 ⫺ N⫺n⫺1 = ⫺ N⫺n . dqn 2 2 dqn 2

Q.E.D.

ADVERTISING COOPETITION: WHO PAYS? WHO GAINS? James A. Dearden and Gary L. Lilien ABSTRACT We develop a competitive advertising model, where a ﬁrm’s advertising spending can be divided into two parts. One part, which we call generic advertising, affects only total market demand. The second component of that spending, brand advertising, affects market share directly, but may also have an effect on total market demand. We investigate how the proﬁt margins of the ﬁrms, the advertising elasticities, the base market shares of the ﬁrms, and the market demand effect of brand advertising interact to determine the total amount of advertising spending in the market, who pays and how they pay (the ratio of generic to brand advertising). We also show that, in general, a market where generic advertising expenditures are set cooperatively will see higher expenditures of generic advertising than will a purely competitive market.

I. INTRODUCTION Tolstoy summed up his views on business as War and Peace. Brandenburger and Nalebuff (1996), noting the term War and Peace, renamed that view

Advertising and Differentiated Products, Volume 10, pages 203–219. Copyright © 2001 by Elsevier Science Ltd. All rights of reproduction in any form reserved. ISBN: 0-7623-0823-0

203

204

JAMES A. DEARDEN AND GARY L. LILIEN

Coopetition. The coopetiton idea is that, in many markets, there are forces that drive ﬁrms to increase industry proﬁts – to cooperate – and forces that drive them simultaneously to increase their own respective proﬁts at the expense of their competitors’ proﬁts – to compete. We focus on advertising here, where ﬁrms advertise to grow the total market (driving primary demand) – the cooperation – as well as to capture their share of that demand – the competition. When considering the sort of advertising that beneﬁts an entire industry, the ﬁrst types of product markets that come to mind are commodities, especially agricultural products. Agricultural economists have focused considerable attention on the generation of primary demand through what is called generic advertising (Forker & Ward, 1993). They argue that the historic development of commodity advertising (primarily in the agricultural products arena) grew out of two forces: the need for large communications spending to generate primary product demand and to “pull” products through distribution channels, and the inability of individual agents (mostly farmers) to capture the returns associated with that spending. That is, in agricultural products markets, the industry is better off with increased generic advertising, but no single agent (farmer) wants to pay for the increased advertising. As a result, government programs and industry associations have developed to help address this perceived underspending in advertising and solve the free rider problem. But pressures to cooperate as well as to compete in the advertising that generates overall market demand go well beyond commodity product producers. For example, the Video Store Distribution Association has put pressure on the ﬁlm industry to contribute to a generic advertising campaign for video rentals (Arnold, 2000) and similar arguments are being made for digital television (Snoddy, 1998). While brand-speciﬁc advertising exceeds generic advertising by a ten to one ratio in the food industry, generic advertising is growing at a faster rate than brand advertising (Sun et al., 1993). But perhaps even this rapid growth understates the generic advertising effect. When Peter Lehmann advertises its wine as “Barossa Valley born and bred: the Art of Australian Winemaking” (Wine Spectator, June 30, 2001) that ad message seems designed both to increase demand of Australian wines at the expense of other regions as well as to draw sales directly for Peter Lehmann’s wines. Also, in the area of alcoholic beverages, Woodside (1999) shows that brand-speciﬁc advertising may have signiﬁcant category effects. That is, even if an ad message aims to increase the demand for a particular product, and does not directly aim to increase category demand, other products in the category may beneﬁt. However, Nelson (2001)

Advertising Coopetition: Who Pays? Who Gains?

205

ﬁnds that studies of brand advertising fail to establish a spillover effect of successful brand advertising on the marketwide demand for alcoholic beverages. Leeﬂang et al. (2000) propose that a product category’s sales is simply the aggregation of sales of brands in that category: if all brands are advertising to increase their own sales, an increase in category sales will quite likely occur. This observation sets up one of the central assumptions of our model: both generic and brand advertising generate overall market demand. That is, brand advertising affects not only market shares; it also affects overall market demand. Whether a ﬁrm’s brand advertising affects only its own demand or whether there is a spillover effect on overall market demand is open to question and depends on the industry in question. Roberts and Samuelson (1988) in their study of the cigarette industry ﬁnd that cigarette brand advertising increases overall market demand, but has no effect on market shares. Alternatively, Gasmi, Laffont, and Vuong (1992) in their study of the soft drink market ﬁnd the opposite result. Soft drink brand advertising affects market shares without affecting overall market demand. Our model permits the entire range of brand advertising effects, from affecting only market shares to affecting only overall market demand. In this paper, we segment a ﬁrm’s advertising spending into two parts. One part only affects total market demand, which we call generic advertising. The second component of that spending, brand advertising, affects market shares directly, and may also have an effect on total market demand (as the Peter Lehmann ad would seem to do). We investigate how the proﬁt margins of the ﬁrms, the advertising elasticities, the base market shares of the ﬁrms and both the market demand effect and market share effect of brand advertising interact to determine the types of advertising spending in the market, who pays and how they pay (the ratio of generic to brand advertising). In our game-theoretic analysis of advertising competition, we investigate the relationship between the effect of brand advertising on overall market demand and generic and brand advertising expenditures. We explore this relationship both by expressing the equilibrium advertising expenditures in terms of advertising elasticities and by analyzing numerical cases. In our model, we ﬁnd two reasons for reduced generic advertising. The ﬁrst reason is that when the generic advertising is noncooperative – the government or industry associations do not enforce payments to industry-wide generic advertising campaigns – the free-rider problem results. We demonstrate by means of an example that the movement from noncooperative generic advertising campaigns to cooperative industry-wide campaigns results in

206

JAMES A. DEARDEN AND GARY L. LILIEN

greater generic advertising. The second reason is due the market demand effect of the brand advertising. This market demand effect means that brand advertising crowds out and reduces the need for generic advertising. Note that our paper is related to Krishnamurthy (2000), which also considers brand and generic advertising. In Krishnamurthy, however, brand advertising affects market shares but not overall market demand. Hence, Krishnamurthy investigates only the ﬁrst reason for reduced generic advertising. We examine three additional issues. The ﬁrst is the free-rider problem. In Theorem 1 and Corollary 1, for the case in which generic advertising is noncooperative, we characterize the types of ﬁrms that contribute to the advertising and the types of ﬁrms that free ride. Second, by means of a numerical case, we demonstrate that the movement from noncooperative to cooperative (i.e. the ﬁrms set generic advertising expenditures to maximize industry-wide proﬁt) results in greater expenditures on generic advertising. Third, and a minor point of the paper, in Theorem 1 and by means of a numerical case, we demonstrate that noncooperative generic advertising expenditures by the ﬁrms in an industry are strategic substitutes. That is, if one ﬁrm reduces its generic advertising expenditures, then a competitor will strategically respond by increasing its expenditures on generic advertising. Alternatively, if one ﬁrm increases its advertising in response to an increase by its competitor, then the ﬁrm’s brand advertising is a strategic complement of its competitor’s. (See Bulow, Geanakoplos & Klemperer, 1985, for formal deﬁnitions of strategic substitutes and strategic complements.) In a numerical case, within the context of our model, we also characterize whether brand advertising expenditures are strategic substitutes or complements. Before proceeding, we note that unintended consequences of generic advertising can arise. For example, the advertising of orange juice in the U.S. has lead to free riding by imports. That increase in imports has eroded the expected impact of advertising on supporting price by an estimated 66% (Brown, Lee & Spreen, 1996) which, in turn, may lead to a decrease in total market supply. In addition, Kinnucan and Miao (2000) develop a model that shows that juice producers beneﬁt from generic ﬂuid milk advertising, while soft drinks, coffee and tea producers lose. In the context of our model, the advertising by one market segment – for example U.S. orange growers – has an effect on the overall market demand for oranges in the U.S. We also note that other concerns have been raised about U.S. government mandated agricultural advertising programs. For example, Chung and Kaiser (2000) show, using data from the New York milk industry, that the distribution of beneﬁts of generic advertising programs varies by ﬁrm size, with larger ﬁrms

Advertising Coopetition: Who Pays? Who Gains?

207

seeing greater beneﬁts but smaller beneﬁts per unit sold. We demonstrate that with noncooperative generic advertising, as opposed to the cooperative advertising of the generic case, the ﬁrms that do the generic advertising is determined by the product of each ﬁrm’s proﬁt margin and market share. As a result, even if smaller ﬁrms receive greater beneﬁt per dollar spent on generic advertising, they may be the ones that free ride in the noncooperative equilibrium. Simply, these smaller milk producers could be better off free riding than they are contributing to the industry-wide campaign. This result could explain why small, independent hog farmers oppose federally mandated contributions to the national pork advertising campaign (New York Times, June 11, 2001), and the third largest mushroom grower, United Foods, and not the largest two growers, opposes federally mandated contributions to the national mushroom advertising campaign (New York Times, April 18, 2001; case decided June 25, 2001 in favor of United Foods). We proceed as follows. In the next section we specify our model. Then in Section III we analyze that model, proving the theorem that identiﬁes those ﬁrms that free ride by spending no money on generic advertising. In Section IV, we investigate four two-ﬁrm special cases that highlight the implications of our model for the interplay of generic and brand advertising. In Section V, we compare one of those cases to a ﬁfth special case – cooperative equilibrium – where the expenditure on generic advertising is determined cooperatively. Finally, in Section VI we sum up, and discuss model extensions and implications for future research.

II. THE MODEL We consider a market in which n ﬁrms choose expenditures on brand and generic advertising. We denote ﬁrm j’s expenditures on generic advertising by gj and brand advertising by bj. Both generic advertising and brand advertising inﬂuence overall market demand, d, but only brand advertising affects market shares, (s1, . . , sn). In the market demand function, we let w苸[0, 1] measure the relative inﬂuence on overall market demand of brand advertising relative to generic advertising. We use a to denote the effective total advertising devoted to overall market demand

冘 n

a=

i=1

(gi + wbi)

208

JAMES A. DEARDEN AND GARY L. LILIEN

The overall market demand function is

冉冘

冊

n

d = M(a) = M

(gi + wbi) .

i=1

Firm j ’s market share is a function of brand advertising sj =

Aj(bj)

冘 n

,

Ai(bi)

i=1

where Aj(bj) is ﬁrm j ’s attraction function. This us/(us + them) formulation has a long history in the marketing science literature and has good empirical validity (see for example Bell, Keeney & Little, 1975; Cooper, 1993; Krishnamurthy, 2000). Let pj denote ﬁrm j ’s unit proﬁt margin, before considering advertising costs. Firm j ’s proﬁt, ␲j, as a function of advertising expenditures, (g, b) = (g1, . . . . , gn, b1, . . . , bn), is then

冉冘冘冊 n

␲j(g, b) = pjM

n

gj + w

j=1

bj

j=1

Aj(bj)

冘 n

⫺ gj ⫺ bj .

Ai(bi)

i=1

While this model is superﬁcially similar to other models, such as that of Krishnmurthy (2000), its main distinguishing characteristic is its explicit consideration of brand speciﬁc advertising effects on total demand. That characteristic (along with the technical assumptions below) is key to deriving our main model insights. Technical Assumptions (1) M(a) is continuous, twice differentiable, strictly increasing, and strictly concave. ⭸M(a) = 0. (2) lim a→ ⬁ ⭸a (3) Aj(bj) is continuous, nondecreasing, and concave. (This assumption implies that the market share function, sj(b), is either independent of bj or increasing and strictly concave in bj .) ⭸Aj(bj) = 0. (4) lim bj → ⬁ ⭸bj

Advertising Coopetition: Who Pays? Who Gains?

209

These technical assumptions ensure that equilibria exist both in the noncooperative and cooperative advertising cases.

III. ANALYSIS In our model, all ﬁrms simultaneously set generic and brand advertising expenditures. In this section the ﬁrms set those expenditures noncooperatively. Speciﬁcally, given the advertising expenditures by the other ﬁrms, which we denote as (g ⫺ j, b ⫺ j), ﬁrm j sets (gj, bj) to maximize its proﬁt, ␲j, resulting in Nash equilibria to the advertising game. (In Section V, we consider the environment in which the ﬁrms noncooperatively set brand advertising expenditures, while they cooperatively set the generic advertising expenditure, 兺ni= 1 gi, to maximize joint total proﬁt, 兺ni= 1 ␲i .) A Nash equilibrium is an expenditure proﬁle, (g*, b*), such that for each ﬁrm j ␲j((g*j , b*j ), (g*⫺ j, b*⫺ j)) ≥ ␲j((gj, bj), (g*⫺ j, b*⫺ j)) for any (gj, bj). In the Nash equilibrium, Firm j chooses (gj, bj) to maximize ␲j. Our technical assumptions imply that a Nash equilibrium exists and can be characterized by the ﬁrst-order conditions: ⭸␲j(g*, b*) ⭸M(g*, b*) = pj sj(b*) ⫺ 1 ≤ 0 ⭸gj ⭸a

(1)

with strict equality if g*j > 0; and ⭸␲j(g*, b*) ⭸M(g*, b*) ⭸sj(b*) = wpj sj(b*) + pj M(g*, b*) ⫺1≤0 ⭸bj ⭸a ⭸bj

(2)

with strict equality if b*j > 0. (For more on the existence and characterization of the ﬁrst-order conditions for the Nash equilibrium, see Mas-Colell, Whinston & Green, 1995.) In expression ⭸M(g*, b*) sj(b*) is the marginal effect of generic advertising (1), the term pj ⭸a transmitted through overall market demand; while in expression (2), the term ⭸M(g*, b*) sj(b*) is the marginal effect of brand advertising transmitted wpj ⭸a through overall market demand. Note that the marginal effect of brand advertising is w times the effect of generic advertising on overall market ⭸sj(b*) is the marginal effect demand. In expression (2), the term pj M(g*, b*) ⭸bj of brand advertising transmitted through the ﬁrm’s market share.

210

JAMES A. DEARDEN AND GARY L. LILIEN

Theorem 1 below makes two points. First, this theorem identiﬁes those ﬁrms that free ride in any Nash equilibrium by spending no money on generic advertising, and receiving the beneﬁts of the expenditures by other ﬁrms. We let ˜ denote the set of ﬁrms that free ride in any Nash equilibrium. This ﬁrst point N is a classic result on Nash equilibria in games of simultaneous contributions to fund public goods, or generic advertising in our case. The theorem shows that a ﬁrm spends no money on generic advertising if, given the expenditures on generic advertising, the marginal revenue to the ﬁrm of increasing industry expenditures on generic advertising beyond what competitors are spending is less than its marginal cost of advertising. Those ﬁrms with lower proﬁt margins or lower market shares are most likely to spend no money on generic advertising. The second point is that if more than two ﬁrms are in N\N˜ – the set of ﬁrms that contribute to the generic advertising – then there is a continuum of Nash equilibria to the advertising game. The Nash equilibria can be characterized by the sum of expenditures on generic advertising. Theorem 1. Consider a Nash equilibrium (g*, b*). If g*j > 0, then ⭸␲j((gj, g*⫺ j), b*) ⭸M((gj, g*⫺ j), b*) = pj sj(b*) ⫺ 1 = 0 for some gj > 0. ⭸gj ⭸a

If ⭸␲j((gj, g*⫺ j), b*) ⭸M((gj, g*⫺ j), b*) = pj sj(b*) ⫺ 1 < 0 for each gj > 0, ⭸gj ⭸a

then g*j = 0. Moreover, if (g*, b*) is a Nash equilibrium, then any proﬁle of advertising expenditures (g**, b*) that satisﬁes

冘冘 g*j =

˜ j苸N\N

g** j

˜ j苸N\N

is also a Nash equilibrium. Proof. The proof follows from Expression (1) and that

冉冘 n

⭸M

(gi + wbi)

i=1

⭸a

冊

冉冘 n

⭸a = ⭸gj

⭸M

(gi + wbi)

i=1

⭸a

for each gj, gk. This relationship also implies that j ≠ k.

冊

⭸a ⭸M(g, b) = ⭸gk ⭸a

⭸g*j ˜ = ⫺ 1 for each j苸N\N, ⭸gk Q.E.D.

Advertising Coopetition: Who Pays? Who Gains?

211

Corollary 1. Consider a Nash equilibrium (g*, b*). If g*j > 0, then pjsj(b*) = max {pksk(b*)}. Moreover, if pjsj(b*) < max { pksk(b*)}, k≠j

k≠j

then g*j = 0. Proof. From Theorem 1, if ﬁrms j, k pay for the generic advertising, then ⭸M(g*, b*) ⭸M(g*, b*) sj(b*) = pk sk(b*). pj ⭸a ⭸a ⭸M(g*, b*) Since is the same for each ﬁrm in the market, the above condition ⭸a Q.E.D. implies that pjsj(b*) = pksk(b*). Remark. Corollary 1 tells us that the product of the proﬁt margin and the market share, pjsj(b*), must be the same for each ﬁrm that pays for generic advertising. The equilibrium in elasticity form. Now that we have characterized which ﬁrms spend for generic advertising, we can characterize the Nash equilibrium advertising expenditures in terms of advertising elasticities. Consider ﬁrm j, and suppose this ﬁrm spends on both generic and brand advertising, i.e., g*j > 0 and b*j > 0. Manipulation of ﬁrm j ’s Nash equilibrium conditions, expressions (1) and (2), yields ␧sj,bk b*j b*j 1 = = , a* n (1 ⫺ w) ␧M,a (g*i + wb*i )

冘 i=1

where

a* ⭸M(g*, b*) ⭸a M(g*, b*) is the elasticity of market demand with respect to effective generic advertising, a; and ⭸s (b*) bj ␧sj,bj = j ⭸bj sj(b*) is the elasticity of ﬁrm j ’s market share with respect to its brand advertising, bj. Considering ﬁrm j’s advertising decision, the ratio of brand advertising to the effective total advertising devoted to overall market demand b*j b*j = a* n (g*i + wb*i )

␧M,a =

冘 i=1

212

JAMES A. DEARDEN AND GARY L. LILIEN

for constant elasticity of advertising demand functions, is increasing in both ␧sj,bj /␧M,a and w. That is, in elasticity terms, the greater is the relative effectiveness of ﬁrm j ’s brand advertising to the industry advertising devoted to overall market demand (i.e. the greater is ␧sj,bj /␧M,a), the greater is the ratio of ﬁrm j ‘s brand advertising to the industry advertising devoted to overall market demand. Moreover, the larger the effect of ﬁrm j ’s brand advertising on market demand (i.e. the greater is w), the greater is ﬁrm j ’s brand advertising to the industry advertising devoted to overall market demand. We can also assess the ratio of the total advertising expenditures devoted to overall market demand to the sum of brand advertising expenditures as

冘 n

冘 n

b*i

i=1

a*

=

冘

冘 n

b*i

i=1

n

1 = (1 ⫺ w)

␧s i,bi

i=1

␧M,a

(g*i + wb*i )

i=1

Thus, the ratio of industry brand advertising to industry advertising devoted to

冘 n

overall market demand is increasing in both

␧si,bi /␧M,a and w.

i=1

These elasticity conditions are reminiscent of the Dorfman-Steiner (1954) condition on the ratio of a monopolist’s optimal advertising expenditures to optimal revenues in terms of price and advertising elasticities. We expect that our elasticity conditions will prove useful for future empirical work.

IV. SEVERAL ILLUSTRATIVE SPECIAL CASES We examine four two-ﬁrm special cases to highlight four important points about the interplay of generic and brand advertising. • In Case 1, we investigate how the effect of brand advertising on overall market demand (i.e. w) affects equilibrium generic and brand advertising expenditures. • In Case 2, we examine the environments in which ﬁrms spend money on only generic advertising, speciﬁcally those equilibria in which both ﬁrms pay for the generic advertising. This case permits us to investigate the response by one ﬁrm to changes in generic advertising expenditures by the other ﬁrm. • In Case 3, we highlight environments with no brand advertising and in which only one of the ﬁrms pays for the generic advertising. • In Case 4, we focus on brand advertising, investigating whether brand advertising expenditures are strategic substitutes (i.e. if one ﬁrm increases its

Advertising Coopetition: Who Pays? Who Gains?

213

expenditures on brand advertising, the other ﬁrm responds by reducing its expenditures) or strategic complements (i.e. if one ﬁrm increases its expenditures on brand advertising, the other ﬁrm responds by increasing its expenditures). In all four cases, we specify the attraction function Aj(bj) = ␣j + ␤jbej j, where ␣j > 0, ␤j > 0, and ej苸[0, 1], and the overall market demand function M(g, b) = ␥(g1 + g2 + w(b1 + b2))␦ where ␥ > 0 and ␦苸[0, 1]. In these attraction functions (Hanssens et al., 2001), ␣j represents the inherent attractiveness of brand j independent of its advertising (brand equity) and ␤j represents the relative effect of brand advertising on brand attraction. For simplicity here, our total market demand model yields zero sales with no advertising. Case 1. How w, the effect of brand advertising on overall market demand, affects generic and brand advertising. In this case, we let n = 2, ␥ = 1, ␦ = 1/2, p0 = p1 = p, e1 = e2 = e, ␣1 = ␣2 = 0, and ␤1 = ␤2 = 1. Here, the two ﬁrms have the same proﬁt margins, and have the symmetric attraction functions. Solving expressions (1) and (2) for the Nash equilibrium yields the following: If w ≤

1 : 2e + 1

g*0 + g*1 =

p2(1 ⫺ 2we ⫺ w) , and 16(1 ⫺ w)

b*0 = b*1 =

p2e . 16(1 ⫺ w)

If w >

1 : 2e + 1

g*0 = g*1 = 0, and b*0 = b*1 =

1 2 p w(1 + 2e)2. 32

Implications of Case 1. This case shows how equilibrium advertising expenditures depend on w – the direct effect of brand advertising on overall market demand. Speciﬁcally, brand advertising is increasing in w and generic 1 , then the ﬁrms do not spend advertising is decreasing in w. If fact, if w > 2e + 1 money on generic advertising. We obtain this result because a ﬁrm’s brand advertising has two effects on revenue – it increases overall market demand and

214

JAMES A. DEARDEN AND GARY L. LILIEN

it increases the ﬁrm’s market share of that demand – while generic advertising has only one revenue effect – it increases overall market demand. Hence, the greater is w, the greater is the effect of brand advertising on overall market demand, and as a result, the greater are the expenditures on brand advertising. With increased brand advertising due to the increase in w, then the greater the total advertising devoted to overall market demand, a = (g1 + g2 + w(b1 + b2)). With an increase in a, the marginal revenue of generic advertising is smaller, 1 , in this and as a result, the ﬁrms spend less on generic advertising. If w > 2e + 1 case, brand advertising expenditures are sufﬁciently large (due both to the market demand and the market share effects) so that, given the equilibrium expenditures of brand advertising, the marginal revenue of generic advertising is always less than the marginal cost of the generic advertising (which is 1), resulting in no generic advertising. Note also that brand advertising expenditures are increasing and generic advertising expenditures are decreasing in e – a measure of the effect of brand advertising on market shares. If brand advertising has a greater effect on market shares, then ﬁrms increase brand advertising expenditures. Then, the greater brand advertising, which also affects overall market demand, crowds out the generic advertising, and ﬁrms accordingly reduce generic advertising expenditures. Case 2. Generic advertising and multiple equilibria. In this case, n = 2, p1␣1 = p2␣2 = p␣, w = 0 and ␤1 = ␤2 = 0. In this case, brand advertising has no effect on either market demand or market advertising. Hence, b*1 = b*2 = 0. The advertising in the market is entirely generic advertising, and g*1 + g*2 = (p␣␥␦)1/(1 ⫺ ␦). Implications of Case 2. In this case, the two ﬁrms’ generic advertising expenditures are strategic substitutes – an increase in one ﬁrm’s advertising expenditure results in a decrease in advertising expenditure by the other ﬁrm. From expression (1), we can derive ﬁrm j’s optimal generic advertising expenditure in terms of ﬁrm i’s generic advertising expenditure. This function gj(gk) is ﬁrm j’s best response function. Here, and in general for any Nash equilibrium where both ﬁrms j and k spend money on generic advertising, the slope of this best-response function is ⭸gj(gk) = ⫺ 1. ⭸gk

Advertising Coopetition: Who Pays? Who Gains?

215

Thus, for each unit increase in generic advertising expenditure by ﬁrm k, ﬁrm j will respond by reducing its generic advertising expenditure by one unit. Case 3. Generic advertising and only one ﬁrm pays. In this case, n = 2, ␣1 p1 ≠ ␣2 p2, w = 0, and ␤1 = ␤2 = 0. Case 3 is like case 2 in that brand advertising has no effect on either market demand or market shares. Hence, b*1 = b*2 = 0. Unlike Case 2, however, in this case, ␣1 p1 ≠ ␣2 p2. If ␣j pj > ␣k pk, by Theorem 1, only ﬁrm j spends for generic advertising. Speciﬁcally, g*j = ( pj␣j␥␦)1/(1 ⫺ ␦) and g*k = 0. In this case, because ﬁrm j earns the greater marginal revenue of generic advertising than does ﬁrm k, i.e. ( pj␣j)1/(1 ⫺ ␦)(␥␦)1/(1 ⫺ ␦) > ( pk␣k)1/(1 ⫺ ␦)(␥␦)1/(1 ⫺ ␦) ﬁrm j does all of the generic advertising. Implications of Case 3. An interesting point of this example is that the larger ﬁrm in this market – the ﬁrm with the greater market share – is not necessarily the ﬁrm that pays for the generic advertising; which ﬁrm pays depends on the product of market shares and proﬁt margins. Hence, a large ﬁrm with a small proﬁt margin might still be able to free ride on a smaller, but more proﬁtable ﬁrm. Case 4. Brand advertising: strategic substitutes or complements? In this case, n = 2, ␥ = 1, ␦ = 1/2, w = 1, ␣1 = ␣2 = 0, and e1 = e2 = 1. If w = 1, then brand and generic advertising have the same effect on overall market demand. Expressions (1) and (2) then imply g*1 = g*2 = 0, i.e. the ﬁrms do no generic advertising. Firm j’s equilibrium brand advertising expenditure is b*j =

p2j pk␤j␤k ␣j ⫺ . ( pj␤j + pk␤k)2 ␤j

One interesting point about this case is the slope of the brand advertising best response functions: ⭸bj(bk) ( pj␤j + pk␤k)( pj␤j ⫺ pk␤k) = . bk ␤j␤k p2j pk

It follows directly that ⭸bj(bk) ⭸bj(bk) > 0 if ( pj␤j ⫺ pk␤k) > 0, and < 0 if ( pj␤j ⫺ pk␤k) < 0. bk bk

Implications of Case 4. The result above means that ﬁrm j’s brand advertising is a strategic complement of ﬁrm k’s brand advertising if the

216

JAMES A. DEARDEN AND GARY L. LILIEN

product of ﬁrm j’s proﬁt margin and the marginal effect of its brand advertising on its own attractiveness is greater than that of ﬁrm k. Otherwise, ﬁrm j’s brand advertising is a strategic substitute for ﬁrm k’s brand advertising. As in case 3, this result does not mean that the ﬁrm with the greater market share is the ﬁrm that strategically responds by matching increases in its competitor’s brand advertising with increases in its own brand advertising. Market shares are partly determined by the constant or brand equity terms in the attraction functions (the ␣ terms) which do not appear in the equations above. So once again, a ﬁrm with higher proﬁt margin or greater marginal attractiveness response to advertising, rather than the largest ﬁrm, might be the one that responds.

V. COOPERATIVE GENERIC ADVERTISING As Forker and Ward (1993) point out, in a number of markets, either a government agency or an industry association sets a budget for generic advertising, and then allocates that budget amongst the ﬁrms that are expected to beneﬁt. We model that situation as follows. Let gN denote the generic advertising expenditure that the ﬁrms collectively choose. Suppose the objective of the ﬁrms in the industry by choosing gN is to maximize the sum of industry proﬁts. The ﬁrms individually and noncooperatively choose brand advertising expenditures, b. By the technical assumptions above, an equilibrium, denoted by (goN, b*), exists and can be characterized by the ﬁrst-order conditions:

冘 n

⭸

␲j(goN, b*)

j=1

⭸gN

冘 n

=

j=1

pj

⭸M(goN, b*) sj(b*) ⫺ 1 ≤ 0 ⭸a

(3)

with a strict equality if g*N > 0. ⭸␲j(goN, b*) ⭸M(goN, b*) ⭸sj(b*) = wpj sj(b*) + pjM(goN, b*) ⫺1≤0 ⭸bj ⭸a ⭸bj

(4)

with a strict equality if b*j > 0. To compare this situation with those in the preceding section, we construct case 5 in which we examine the expenditures on cooperative generic advertising and noncooperative brand advertising. Case 5. How w, the effect of brand advertising on overall market demand, affects cooperative generic advertising and noncooperative brand advertising. In this case, as in Case 1, n = 2, ␥ = 1, ␦ = 1/2, p0 = p1 = p, e1 = e2 = e,

Advertising Coopetition: Who Pays? Who Gains?

217

␣1 = ␣2 = 0, and ␤1 = ␤2 = 1. And Case 5 has the same market demand function and market share functions as those of Case 1, which we will compare it to. Solving (3) and (4) for the equilibrium advertising expenditures yields the following: 2 If w ≤ : e+1 p2(2 ⫺ we ⫺ w) goN = , and 4(2 ⫺ w) p2e . b*0 = b*1 = 4(2 ⫺ w) 2 If w > : e+1 go0 + go1 = 0, and 1 2 p w(1 + 2e)2. b*0 = b*1 = 32 Implications of Case 5 (comparison with Case 1). We compare the noncooperative and cooperative equilibrium generic advertising expenditures for cases 1 and 5. 1 If w ≤ , then 2e + 1 p2(2 ⫺ we ⫺ w) p2(1 ⫺ 2we ⫺ w) 6 ⫺ 9w + 2w2e + 3w2 goN ⫺ (g*0 + g*1) = ⫺ = >0 4(2 ⫺ w) 8(1 ⫺ w) (2 ⫺ w)(1 ⫺ w) and the cooperative generic advertising promotes increased generic advertising. 1 2 <w≤ , then If 2e + 1 e+1 p2(2 ⫺ we ⫺ w) goN ⫺ (g*0 + g*1) = ⫺ 0 > 0, 4(2 ⫺ w) and again cooperative generic advertising promotes an increased expenditure on generic advertising. 2 : If w > e+1 goN ⫺ (g*0 + g*1) = 0 ⫺ 0 = 0, then, with either cooperative or noncooperative generic advertising, the ﬁrms spend no money on generic advertising.

218

JAMES A. DEARDEN AND GARY L. LILIEN

Thus the expenditure on generic advertising that maximizes the sum of the proﬁts of the ﬁrms in the cooperative situation is greater than the sum of generic advertising expenditures in the noncooperative situation. By collectively determining the generic advertising expenditure, the ﬁrms solve the free-rider problem created by the positive externality of the generic advertising. With noncooperative advertising, ﬁrm j for example chooses its generic advertising expenditure that maximizes its own proﬁt and ignores the positive effect of its advertising on ﬁrm k. With cooperative generic advertising, ﬁrms j and k internalize this positive externality and increase the expenditure on generic advertising.

VI. DISCUSSION AND IMPLICATIONS Our model and analysis show that the interplay between generic and brand advertising is subtle and complex. Although our model is simple and we were able to obtain clean analytical results for certain special cases, our analysis shows how a few market factors like brand equity, proﬁt margin, brand advertising attractiveness response, and the spillover effect of brand advertising on overall market sales can inﬂuence total advertising spending, who pays and what they pay for (brand or generic advertising). Our results are of the contingent variety as our model admits qualitatively different results in different market situations. While we feel that while our results have intuitive appeal, the model that generated those results is quite simple and limited. In particular, our model is static while markets are dynamic; good advertising response models admit delayed or carryover effects as well as instantaneous ones (Hanssens et al., 2001). And our model admits only one marketing mix element, ignoring the interplay of advertising, price, promotional efforts, and the like. Our model appears to be amenable to empirical validation, predicting both the split and the total expenditure on adverting in different market situations, and when we would expect to see free riding. We hope that the model has helped clarify a few issues associated with the coopetition phenomenon in advertising and that it has also highlighted a number of interesting issues that could be the subject for future research.

ACKNOWLEDGMENTS The authors thank the participants of the Lehigh University Economics Seminar Series.

Advertising Coopetition: Who Pays? Who Gains?

219

REFERENCES Arnold, T. K. (2000). VSDA wants studios to pay for awareness campaign. Video Store, 22, (October 8–14), 1. Bell, D. P., Keeney, R. L., & Little, J. D. C. (1975). A market share theorem. Journal of Marketing Research, 12, 136–141. Brandenburger, A. M., & Nalebuff, B. J. (1996). Coopetition. New York: Doubleday. Brown, M. G., Lee, J-Y., & Spreen, T. H. (1996). The impact of generic advertising and the free rider problem: A look at the U.S. orange juice market and imports. Agribusiness, 12, 309–317. Bulow, J., Geanakoplos, J., & Klemperer, P. (1985). Multimarket oligopoly: Strategic substitutes and complements. Journal of Political Economy, 93, 488–511. Chung, C., & Kaiser, H. M. (2000). Distribution of generic advertising beneﬁts across participating ﬁrms. American Journal of Agricultural Economics, 82, 659–664. Cooper, L. (1993). Market share models. In: J. Eliashberg & G. L. Lilien (Eds), Handbook in Operations Research and Management Science, Volume 5 Marketing (pp. 259–314). New York: Elsevier. Dorfman, R., & Steiner, P. (1954). Optimal advertising and optimal quality. American Economic Review, 44, 826–836. Gasmi, F., Laffont, J.-J. , & Vuong, Q. (1992). An Econometric Analysis of Collusive Behavior in a Soft-Drink Market. Journal of Economics and Management Strategy, 1, 277–311. Hanssens, D., Parsons, L., & Schultze, R. (2001). Market Response Models: Econometric and Time Series Analysis (2nd ed.). Boston: Kluwer. Forker, O. D., & Ward, R. W. (1993). Commodity Advertising: The Economics and Measurement of Generic Programs. New York: Lexington Books. Kinnucan, H., & Miao, Y. (2000). Distributional impacts of generic advertising on related commodity markets. American Journal of Agricultural Economics, 82, 672–678. Krishnamurthy, S. (2000). Enlarging the pie v. increasing one’s own slice: An analysis of the relationship between generic and brand advertising. Marketing Letters, 11, 37–48. Leeﬂang, P. S. H., Wittink, D. R., Wedel, M., & Naert, P. (2000). Building Models for Marketing Decisions. Boston: Kluwer. Mas-Colell, A., Whinston, M., & Green, J. (1995). A Course in Microeconomic Theory. New York: Oxford University Press. Nelson, J. P. (2001). Alcohol advertising and advertising bans: A survey of research methods, results, and policy implications. In: Advances in Applied Microeconomics (this volume). Amsterdam: JAI Press. New York Times (2001). Supreme Court roundup; Justices return to commercial speech quandary. April 18, A18. New York Times (2001). Unpopular fee makes activists of hog farmers. June 11, A1. Roberts, M., & Samuelson, L. (1988). An Empirical Analysis of Dynamic, Nonprice Competition in an Oligopolistic Industry. Rand Journal of Economics, 19, 988, 200–220. Snoddy, R. (1998). Digital needs an ad campaign to turn on viewers. Marketing, January 8, 8. Sun, T. Y., Blaylock, J. R., & Allshouse, J. E. (1993). Dramatic growth in mass media food advertising in the 1980s. Food Review, 16, 36–38. Woodside, A. (1999). Advertising and consumption of alcoholic beverages. Journal of Consumer Psychology, 8, 167–186.

This Page Intentionally Left Blank

A MODEL OF VERTICAL DIFFERENTIATION, BRAND LOYALTY, AND PERSUASIVE ADVERTISING Victor J. Tremblay and Carlos Martins-Filho ABSTRACT In this paper, we analyze the impact of advertising and quality decisions on price competition in a duopoly setting. Firms are able to differentiate their products vertically and use persuasive advertising to increase consumer brand loyalty. The model predicts that the high quality ﬁrm will advertise more intensively than the low quality ﬁrm in both covered and uncovered markets. Because consumers are assumed to be informed about product characteristics, advertising neither signals high quality nor discourages ﬁrms from lowering product quality unexpectedly. Instead, advertising is persuasive and is used to dampen price competition, enabling ﬁrms to avoid the Bertrand Paradox. This model provides one explanation for the coexistence of name (heavily advertised) and generic (sparsely advertised) brands.

I. INTRODUCTION An analysis of advertising and product differentiation in an oligopoly setting is complicated because they have multiple dimensions. For example, Kaldor (1949–1950) and Bain (1950) argue that advertising is persuasive in nature.

Advertising and Differentiated Products, Volume 10, pages 221–238. 2001 by Elsevier Science Ltd. ISBN: 0-7623-0823-0

221

222

VICTOR J. TREMBLAY AND CARLOS MARTINS-FILHO

That is, it changes peoples’ tastes by creating subjective differentiation and inducing people to purchase commodities they would not buy otherwise. Thus, this form of advertising increases market power and is socially wasteful. At the other extreme, Stigler (1961), Telser (1964), and Nelson (1974), emphasize the informative role of advertising. By providing consumers with useful information, advertising leads consumers to lower priced commodities with more preferred characteristics. Thus, informative advertising reduces market power and improves market performance. Product differentiation can be described in several ways. First, products can be classiﬁed as horizontally differentiated, a case where consumers have different preference orderings over product characteristics. For example, consider a market with two competing banks that differ only in terms of their geographic location within a city. In general, consumers will not unanimously prefer one bank over another, but instead, will prefer the bank with a more convenient location. Thus, this is sometimes called spatial differentiation (Hotelling, 1929; d’Aspremont et al., 1979). Alternatively, when consumers agree over the preference ordering of products, the market is said to be vertically differentiated. The best example of vertical differentiation occurs when products differ in terms of quality. With this form of differentiation, all consumers most prefer the product of highest quality, ceteris paribus. Finally, in Lancaster’s (1966) more general framework, consumers have preferences over one or more horizontal and/or vertical characteristics, not the commodities themselves. Thus, goods are deﬁned as bundles of characteristics, and those with a similar mix of characteristics are classiﬁed as product substitutes. Previous studies analyze particular types of advertising and product differentiation in an imperfectly competitive setting. For instance, Grossman and Shapiro (1984) show how informative advertising affects price competition in an oligopoly market with horizontal differentiation. Milgrom and Roberts (1986) demonstrate that high quality producers may use advertising to signal quality, while Klein and Lefﬂer (1981) show that advertising ﬁrms have a greater incentive to maintain high product quality. Dixit and Norman (1978) demonstrate that both persuasive and informative advertising can be socially excessive in imperfectly competitive markets when advertising changes consumer tastes.1 Hallagan and Joerding (1983) analyze the impact of persuasive advertising in a monopolistically competitive market. More recent research has examined the impact of persuasive adverting on socialization (Lee, 1997) and price competition in markets with horizontal differentiation (Von der Fehr & Stevik, 1998; Block & Manceau, 1999) and with homogeneous goods (Tremblay & Polasky, 2000). To date, however, there has

A Model of Vertical Differentiation, Brand Loyalty, and Persuasive Advertising

223

been no research on the impact of vertical differentiation and persuasive advertising on price competition. The purpose of this research is to develop a duopoly model in which ﬁrms compete in price, quality, and persuasive advertising. Consumers accurately observe product quality but differ in their willingness to pay for brands of different quality. Although the mechanism is a matter of conjecture, we assume that persuasive advertising operates by increasing consumers’ willingness to pay for the advertised brand.2 This structure enables ﬁrms to distance themselves vertically and use persuasive advertising to enhance their market positions. Firms are assumed to play a multistage game of perfect and complete information. That is, each ﬁrm knows the history of play and the payoffs of both ﬁrms (Gibbons, 1992). Because advertising is purely persuasive in nature, ﬁrms choose advertising in the initial stage. Then, ﬁrms compete in quality and price. In the spirit of Caves and Porter (1977), our model provides one explanation for the existence of name (heavily advertised) and generic (sparsely advertised) brands. The ability to advertise and adjust product quality allows ﬁrms to create greater real and perceived distance between products, which dampens price competition and enhances market power. These features are common to many consumer goods markets. For example, the work by Wiggins and Raboy (1996) indicates that Chiquita Banana competes with suppliers of generic bananas by using a persuasive form of advertising to sell bananas of superior quality at a premium price. Similarly, Budweiser beer, Gateway computers, and Kodak ﬁlm follow a similar strategy when competing with their generic counterparts.3 Finally, both Wills and Mueller (1989) and Putsis and Cotterill (1999) ﬁnd that advertising-induced product differentiation plays a key role in explaining the price premium paid for name brand products in many food product markets.

II. MARKET CONDITIONS Two ﬁrms (indexed by 1 and 2) compete in a market, and each ﬁrm produces a single product, deﬁned as brands 1 and 2. Strategic variables are price, advertising, and product quality. Quality is indexed by the parameter s, where a higher value of s indicates a higher level of quality, s苸[sL, sH], sH , sL苸⺢, and sH > sL > 0. For simplicity, we assume that variable and ﬁxed costs of production are zero.4 Following Mussa and Rosen (1978), Gabszewicz and Thisse (1979), and Wauthy (1996), consumers have heterogeneous preferences that are represented by a taste parameter, ␸. Tastes are assumed to be distributed uniformly

224

VICTOR J. TREMBLAY AND CARLOS MARTINS-FILHO

over the interval [␸L, ␸H], where ␸H > ␸L ≥ 0. Each ␸-type consumer maximizes the following utility function for brand i: Ui(␸) =

再

␸si ⫺ pi , for ␸ > pi /si 0, for ␸ ≤ pi /si .

(1)

where pi is the price and si is the quality level associated with brand i. All consumers prefer a higher quality product, but a consumer with a higher ␸ will have a stronger preference for quality and will be more loyal to the higher quality brand, ceteris paribus. Thus, this speciﬁcation accounts for elements of vertical product differentiation and consumer brand loyalty. We assume there exists a marginal consumer with taste parameter ␸m, who is indifferent between buying from ﬁrm 1 or 2. That is, ␸ms1 ⫺ p1 ⬅ ␸ms2 ⫺ p2, which implies that ␸m ⬅ ( p1 ⫺ p2)/(s1 ⫺ s2). Thus, the value of ␸m can be thought of as the quality-adjusted premium that the marginal consumer is willing to pay for product 1. Firm demand depends upon ␸L, ␸m, ␸H, p1, p2, s1, and s2. Consumers have unit demands, and the market is either covered or uncovered. A duopoly market is said to be covered if every consumer makes a purchase of one brand or the other. In an uncovered market, some consumers are priced out of the market and do not make a purchase.5 When the market is covered and assuming ﬁrm 1 is the high quality producer, the demand for ﬁrm 1 ’s product, D1, is deﬁned as the set of all consumers with preferences greater than ␸m.6 That is: D1 = ␸H ⫺ ␸m p1 ⫺ p2 = ␸H ⫺ . (2) s1 ⫺ s2 Likewise, the demand for product, 2, D2, is: D2 = ␸m ⫺ ␸L p1 ⫺ p 2 = ⫺ ␸L. (3) s1 ⫺ s2 Figure 1 describes the connection between consumer preferences and ﬁrm demand. In this case, the market is covered and both ﬁrms have non-negative demands since ␸H > ␸m > ␸L > p2 /s2. In addition, the ﬁgure shows that an increase in s1 relative to s2 (or a decrease in p1 relative to p2) causes ␸m to shift right, D1 to fall, and D2 to increase. Once ␸m exceeds ␸H, however, ﬁrm 2 becomes a monopolist and the market is said to be preempted (Wauthy, 1996). When the market is uncovered, some consumers do not make a purchase in the market. Thus, to derive ﬁrm demands requires one to identify ␸m and the

A Model of Vertical Differentiation, Brand Loyalty, and Persuasive Advertising

225

Fig. 1. Covered Market.

consumer who is indifferent between purchasing from ﬁrm 2 and not making a purchase at all. This latter consumer is characterized as having a taste parameter ␸ = p2/s2, such that U2(␸) = ␸s2 ⫺ p2 = 0. In an uncovered market, the demand for product 1 is the same as in the covered market, but the demand for product 2 includes only those consumers with taste parameters between ␸m and p2/s2. That is, D2 =

p1 ⫺ p2 p2 ⫺ . s1 ⫺ s2 s2

(4)

Figure 2 describes consumer preferences and ﬁrm demand functions for brands 1 and 2. In this setting, both ﬁrms have non-negative demands and the market

226

VICTOR J. TREMBLAY AND CARLOS MARTINS-FILHO

Fig. 2. Uncovered Market.

is uncovered because ␸H > ␸m > p2/s2 > ␸L. This is evident from the ﬁgure, as those consumers with taste parameters between p2/s2 and ␸L receive negative utility from the purchase of either brand and refrain from making a purchase. Given that ﬁrm demand is a function of consumer taste parameters, ﬁrms may invest in persuasive advertising in an effort to shift preferences in proﬁtable directions. For example, assume that ﬁrm 1 can use advertising to increase the number of customers loyal to the high quality brand by increasing ␸H and ﬁrm 2 can use advertising to increase the number of customers loyal to the low quality brand by decreasing ␸L (when the market is covered). This form of advertising is purely persuasive in nature because it expands the size of the market by converting non-consumers into new consumers who make purchases only because advertising persuades them to do so.

A Model of Vertical Differentiation, Brand Loyalty, and Persuasive Advertising

227

In this framework, a consumer is deﬁned as someone who views these commodities as goods, while a non-consumer views them as bads. We assume there exists a pool of non-consumers who have malleable preferences. Persuasive advertising expands the size of the market by convincing some nonconsumers that bads are in fact goods. This is consistent with the deﬁnition of persuasive advertising, as it changes peoples’ tastes and causes them to purchase commodities they would not normally buy. If consumers did not have unit demands, as deﬁned in our model, then persuasive advertising could further increase market demand by causing existing consumers to purchase more output than they would buy before being exposed to advertising. In our model, however, persuasive advertising increases demand by simply inducing new consumers to enter the market. We assume the existence of twice differentiable functions A1(␸H; z): ⺢ + → ⺢ + and A2(␸L; z): ⺢ + → ⺢ + , which deﬁne the level of advertising expenditures (A) necessary to produce an upper bound taste parameter ␸H and a lower bound taste parameter ␸L, respectively, for every z > 0. Parameter z is common to both functions and is associated with advertising technology. A higher z implies that advertising is more expensive or is less effective at increasing demand. Since the demand for ﬁrm 1’s product increases with ␸H and assuming it becomes increasingly more expensive to expand the size of the market, A⬘1(␸H ; z) > 0 and A⬙1(␸H ; z) > 0. However, since ﬁrm 2’s advertising increases demand by reducing ␸L when the market is covered, A⬘2(␸L; z) < 0 and A⬙2(␸L; z) > 0.7 We assume that A1 and A2 are symmetrical with respect to the center of the taste parameter support.

III. QUALITY, PRICE, AND ADVERTISING IN AN UNCOVERED MARKET Firms are initially assumed to compete in an uncovered market and play a three stage game of perfect and complete information. In the ﬁrst stage, ﬁrms simultaneously choose advertising. In the second stage, they choose quality simultaneously, and in the third stage they simultaneously choose prices. In each sub-game, ﬁrms have observed the outcome of the previous stage. Because advertising is persuasive and a marketing campaign can take a considerable amount of time to develop and implement, ﬁrms are assumed to establish a marketing strategy before playing the quality and pricing stage games. In an uncovered market with the demand and cost conditions described above, ﬁrm proﬁts are ␲1 = p1(␸H ⫺ p1/x + p2/x) ⫺ A1(␸H ; z) and ␲2 = p2( p1/ x ⫺ p2/x ⫺ p2/s2), where x ⬅ (s1 ⫺ s2). Because ﬁrm 2’s demand is not a function

228

VICTOR J. TREMBLAY AND CARLOS MARTINS-FILHO

of taste parameters, only ﬁrm 1 has an incentive to use persuasive advertising in this setting. Solving recursively to obtain the sub-game perfect equilibrium, the best reply functions of the ﬁnal-stage pricing game are obtained from each ﬁrm’s ﬁrst order conditions. p1 =

2␸H s1(s1 ⫺ s2) ; 4s1 ⫺ s2

p2 =

␸H s2(s1 ⫺ s2) . 4s1 ⫺ s2

(5)

These results indicate that as s1 approaches s2, vertical differentiation diminishes and the Nash equilibrium approaches the simple Bertrand equilibrium where price equals marginal cost. Thus, vertical differentiation helps dampen price competition. In stage two of the game, ﬁrms simultaneously choose quality. Anticipating the pricing game in the previous stage, ﬁrm proﬁt equations can be found by substituting the best reply functions into the original proﬁt equations. ␲1 =

4␸2H s21(s1 ⫺ s2) ⫺ A1(␸H ; z); (4s1 ⫺ s2)2

␲2 =

␸2H s1s2(s1 ⫺ s2) . (4s1 ⫺ s2)2

(6)

The best reply for ﬁrm 1 is easily determined because proﬁt is increasing in s1. Thus, ﬁrm 1 will choose s1 = sH. Given this, the best reply for ﬁrm 2 is s2 = (4/7)sH. This “4/7 rule” is originally due to Choi and Shin (1992). Because only ﬁrm 1 has an incentive to advertise, it will choose advertising levels to affect ␸H in an optimal way. Anticipating information from previous stages, ﬁrm 1’s proﬁt equation can be expressed in terms of ␸H and sH by substituting the best reply functions from the quality game into ﬁrm 1’s proﬁt equation in (6). The resulting proﬁt equation for ﬁrm 1 becomes: ␲1 =

7␸2H sH ⫺ A1(␸H ; z). 48

(7)

At this stage, the ﬁrm’s goal is to choose the level of advertising that produces the proﬁt maximizing value of ␸H . The following proposition describes the characteristics of the resulting subgame perfect equilibrium when the market is uncovered. Proposition 1. Consider an uncovered market where ␸H > p2/s2 > ␸L ≥ 0 and ﬁrms play the dynamic game described above. Then, (a) the proﬁt functions in the ﬁrst stage of the game are: ␲1 =

7␸2H sH ⫺ A1(␸H ; z); 48

␲2 =

␸2H sH ; 48

(8)

A Model of Vertical Differentiation, Brand Loyalty, and Persuasive Advertising

229

(b) the sub-game perfect equilibrium will have the following properties: (i) (ii) (iii) (iv) (v)

s*1 > s*2 A*1 > A*2 = 0 p*1 > p*2 D*1 > D*2 ␲*1 ≠ ␲*2

Proof. See the Appendix. In this setting, ﬁrm behavior and performance are asymmetric, and only the high quality ﬁrm has an incentive to advertise. Firms differentiate vertically, with the high quality ﬁrm producing more output and charging a higher price than its lower quality competitor. Comparative static results are obtained by applying the Implicit Function Theorem to ﬁrm 1’s ﬁrst order condition at the ﬁnal stage of the game.8 These results indicate that a decrease in the price of advertising (z) and an increase in the upper limit of product quality (sH ) cause ﬁrm 1’s optimal level of advertising to increase, which leads to higher prices and greater market power for both ﬁrms. Thus, both vertical differentiation and persuasive advertising help uphold the principle of differentiation by softening price competition. To derive a simple closed-form solution to the game, we assume that A1(␸H; z) = z(␸H ⫺ kH)2 for ␸H > kH > 0. The parameter kH identiﬁes the upper bound of the parameter taste interval when ﬁrm 1 does not advertise and represents consumer preferences that are unadulterated by advertising. In addition, this speciﬁcation assures that A⬘1(␸H ; z) > 0 and A⬙1(␸H ; z) > 0. In this case, ﬁrm 1’s best reply in the ﬁnal stage of the game is to advertise so that: ␸H =

48kH z . 48z ⫺ 7sH

(9)

The sub-game perfect equilibrium to this game is described below.9 s*1 = sH > s*2 = (4/7)sH A*1 = p*1 =

12kH sH z 24kH sH z > p*2 = 48z ⫺ 7sH 7(48z ⫺ 7sH)

D*1 = ␲*1 =

49k2H s2H z > A*2 = 0 (48z ⫺ 7sH)2

28kH z 14kH z > D*2 = 48z ⫺ 7sH 48z ⫺ 7sH

7k2H sH z 48k2H sH z2 ≠ ␲*2 = 48z ⫺ 7sH (48z ⫺ 7sH)2

(10)

230

VICTOR J. TREMBLAY AND CARLOS MARTINS-FILHO

This illustrates that ﬁrm 1 advertises more intensively, and that the market power of both ﬁrms increases as z decreases and as sH increases.10 It also shows that ﬁrm 1’s advertising and that the price, output, and proﬁt of both ﬁrms increase in kH. This implies that the market power of ﬁrms increases with a wider distribution of pre-advertising preferences for quality. As an example, we let sH = 2, sL = 1, z = 2, and kH = 2, conditions that ensure ﬁrm participation in an uncovered market. The resulting sub-game perfect equilibrium is: s*1 = 2 > s*2 = 1.1429 A*1 = 0.2322 > A*2 = 0 (11) p*1 = 1.1707 > p*2 = 0.3345 D*1 = 1.3659 > D*2 = 0.6829 ␲*1 = 1.3659 > ␲*2 = 0.2284 This illustrates that the degree of asymmetry can be rather extreme, with the high quality ﬁrm being much larger in size, advertising much more intensively, charging a higher price, and earning greater proﬁt than its low quality competitor. These results indicate that heavily advertised name brand products may coexist with unadvertised generic brands if ␸L is low enough to produce an uncovered market.11 This will occur if this extreme consumer has little taste for quality. Marketing research indicates that this is very likely in real markets since some consumers always purchase the cheapest brand when making a purchase (Economist, 1992 and Foxall & Goldsmith, 1994, Chapter 2). In such a market, only the high quality producer will have an incentive to use a persuasive form of advertising. Thus, this model provides one rationale for markets where only high quality producers advertise.

IV. QUALITY, PRICE, AND ADVERTISING IN A COVERED MARKET Next, we analyze a covered market where ﬁrm proﬁt functions are ␲1 = p1(␸H ⫺ p1 /x + p2 /x) ⫺ A1(␸H ; z) and ␲2 = p2( ⫺ ␸L + p1/x ⫺ p2 /x) ⫺ A2(␸L; z). In this case, ﬁrm 1 will use advertising to increase ␸H , and ﬁrm 2 will use advertising to decrease ␸L. Assuming that the game is played with the same timing as in the previous section, ﬁrms will ﬁrst solve the pricing game, which produces the following best reply functions. (2␸H ⫺ ␸L)(s1 ⫺ s2) (␸H ⫺ 2␸L)(s1 ⫺ s2) ; p2 = . (12) p1 = 3 3

A Model of Vertical Differentiation, Brand Loyalty, and Persuasive Advertising

231

As in the uncovered market case, the Nash equilibrium approaches the simple Bertrand equilibrium as s1 approaches s2. When ␸H < 2␸L, however, ﬁrm 2 will exit the market, creating a monopoly for ﬁrm 1. Thus, to assure a duopoly structure, we assume that ␸H > 2␸L when ﬁrm 2 chooses to advertise and the market is covered. In stage two of the game, ﬁrms simultaneously choose quality. Anticipating the pricing game in the previous stage, ﬁrm proﬁts are given by: ( ⫺ 2␸H + ␸L)2(s1 ⫺ s2) ⫺ A1(␸H; z); 9 ( ⫺ ␸H + 2␸L)2(s1 ⫺ s2) ⫺ A2(␸L; z). ␲2 = 9

␲1 =

(13)

Because 1’s proﬁts increase in s1 and 2’s proﬁts decrease in s2, their respective best replies are s1 = sH and s2 = sL. That is, ﬁrms choose maximal real differentiation. Finally, ﬁrms choose the levels of advertising that produce the proﬁt maximizing taste parameters of ␸H and ␸L. The characteristics of the resulting sub-game perfect equilibrium are described in the following proposition. Proposition 2. Consider a covered market where ␸H > 2␸L, ␸L > p2/s2 ≥ 0, and ﬁrms play the dynamic game described above. Then, (a) the proﬁt functions in the ﬁrst stage of the game are: ( ⫺ 2␸H + ␸L)2(s1 ⫺ s2) ⫺ A1(␸H ; z); 9 ( ⫺ ␸H + 2␸L)2(s1 ⫺ s2) ⫺ A2(␸L; z); ␲2 = 9

␲1 =

(14)

(b) the sub-game perfect equilibrium to this game is asymmetric and has the following properties: (i) (ii) (iii) (iv) (v)

s*1 > s*2 A*1 > A*2 p*1 > p*2 D*1 > D*2 ␲*1 ≠ ␲*2

Proof. See the Appendix. When the market is covered and both ﬁrms advertise, ﬁrm behavior and performance remain asymmetric. As with the uncovered market case, the high quality ﬁrm advertises more, charges a higher price, and sells more output than the low quality ﬁrm.

232

VICTOR J. TREMBLAY AND CARLOS MARTINS-FILHO

Applying the Implicit Function Theorem and Cramer’s rule to the system of ﬁrst order conditions in the ﬁnal stage game yields comparative static results. As in an uncovered market, the equilibrium level of advertising for each ﬁrm increases as the cost of advertising (z) decreases and as the potential vertical distance between products (sH ⫺ sL) increases. These changes also cause the equilibrium price, output, and proﬁts of both ﬁrms to increase. To derive a simple closed-form solution that is comparable to the previous case, we again assume that A1(␸H ; z) = z(␸H ⫺ kH)2, ␸H > kH > 0. To ensure symmetry, we assume that A2(␸L; z) = z(␸L ⫺ kL)2 when the market is covered, with kH > kL ≥ ␸L > 0. In this case, kL represents the value of the lower bound on the parameter taste interval when ﬁrm 2 does not advertise. Thus, ﬁrm 1 has an incentive to use advertising to increase ␸H , and ﬁrm 2 has an incentive to use advertising to decrease ␸L. Both of these actions increase the size of the market by causing new consumers to purchase products only because they are persuaded to do so by advertising. The sub-game perfect equilibrium to this game is described below. s*1 = sH > s*2 = sL A*1 =

4y2z(2kH y ⫺ 6kH z + 3kLz)2 4y2z(2kLy + 3kH z ⫺ 6kLz)2 > A*2 = 2 w w2

3 y z( ⫺ 2kH y + 6kH z ⫺ 3kLz) 3 y z(2kLy + 3kH z ⫺ 6kLz) > p*2 = w w 3 z( ⫺ 2kH y + 6kH z ⫺ 3kLz) 3 z(2kLy + 3kH z ⫺ 6kLz) > q*2 = q*1 = w w

p*1 =

␲*1 =

(15)

yz(9z ⫺ 4y)(2kH y ⫺ 6kH z ⫺ 3kLz)2 yz(9z ⫺ 4y)(2kLy + 3kH z ⫺ 6kLz)2 > ␲* = 2 w2 w2

where w ⬅ (4y2 ⫺ 24yz + 27z2) and y ⬅ sH ⫺ sL.12 This illustrates that ﬁrm behavior and performance are asymmetric. Firms differentiate their products vertically, and the high quality ﬁrm advertises more intensively, is larger in size, charges a higher price, and earns a higher proﬁt margin than the low quality producer. In this case, however, both ﬁrms advertise to attract new customers and expand the size of the market. To illustrate the properties of the model, we select parameter values that ensure a sub-game perfect equilibrium and a covered market when both ﬁrms participate in the game. These conditions are met, for example, when sH = 2, sL = 1 (y = 1), z = 2, kH = 2, and kL = 1. This generates the following equilibrium values:13

A Model of Vertical Differentiation, Brand Loyalty, and Persuasive Advertising

233

s*1 = 2 > s*2 = 1 A*1 = 0.3828 > A*2 = 0.0078 p*1 = 1.3125 > p*2 = 0.1875

(16)

q*1 = 1.3123 > q*2 = 0.1875 ␲*1 = 1.3398 > ␲*2 = 0.0273 This example illustrates that the degree of ﬁrm asymmetry can be large, as the high quality ﬁrm is much larger in size and advertises more intensively than the low quality ﬁrm. From this equilibrium point, the comparative statics are quite plausible.14 The advertising, price, output, and proﬁts of both ﬁrms decrease with the cost of advertising (z), increase with the potential vertical distance between products (y), and increase as diversity in preferences expands (that is, as kH increases and/or kL decreases), ceteris paribus. As in an uncovered market, market power is enhanced by the ability of ﬁrms to distance themselves vertically and effectively use persuasive advertising to change consumer tastes in proﬁtable directions. For both covered and uncovered markets, the results are consistent with Klein and Lefﬂer (1981) and Milgrom and Roberts (1986) who ﬁnd that heavily advertised, name brand products will develop in markets where product quality is an issue. Unlike in their models, however, advertising neither encourages ﬁrms to maintain quality nor serves as a signal of quality. Instead, our model shows that both real vertical differentiation and purely persuasive advertising enhance market power and induces the high quality ﬁrm to advertise more intensively than the low quality ﬁrm.

V. CONCLUDING REMARKS A duopoly model is developed to show how vertical product differentiation and persuasive advertising affect price competition in covered and uncovered markets. In both cases, ﬁrms distance themselves vertically and use persuasive advertising to create subjective differentiation in an effort to reduce price competition and avoid the Bertrand paradox. The model predicts that ﬁrms are more likely to use persuasive advertising to market high quality brands. When the market is uncovered, only the high quality ﬁrm has an incentive to advertise. Although both ﬁrms advertise when the market is covered, ﬁrm behavior is still asymmetric, with the high quality ﬁrm advertising more intensively than the low quality ﬁrm. This is consistent with the empirical work of Caves and Green (1996), who ﬁnd that advertising increases with quality for experience goods. Because consumers know the

234

VICTOR J. TREMBLAY AND CARLOS MARTINS-FILHO

quality level of each brand, however, advertising does not serve as a signal of quality and does not induce the high quality ﬁrm to maintain quality. Instead, persuasive advertising is used to enhance consumer brand loyalty and the market power of ﬁrms. Thus, our game-theoretic model provides an alternative explanation for the positive correlation between advertising and product quality.

NOTES 1. For more discussion of their provocative results, see Fisher and McGowan (1979), Shapiro (1980), and Becker and Murphy (1993). See Tremblay and Tremblay (1995) and Farr et al. (2001) for empirical applications of these models to the brewing and cigarette industries. 2. Empirical evidence indicates that advertising does increase aggregate demand (Seldon & Jung, 1995), and psychologists ﬁnd that mere exposure to advertising has a positive effect on consumers’ attitudes toward advertised brands (Zajonc, 1968; Witt, 1991; Kokkinaki & Lunt, 1999). Akerlof and Kranton (2000) argue that advertising may change consumer tastes by manipulating social norms and notions of a consumer’s identity. Alternatively, since consumers’ willingness to pay for quality is thought to be a positive function of income (Mussa & Rosen, 1978; Tirole, 1990, 1996–1997), persuasive advertising may change preferences by inducing people to behave as if they are in a higher income class. 3. James Taylor, Gateway’s Vice President of Marketing, describes Gateway’s persuasive form of advertising as exuding a “gentle humor . . . a built-in politeness beﬁtting a company based far off in South Dakota with a Holstein cow as its mascot” (Johnson, 1997). For a more complete discussion of the beer, computer, and ﬁlm industries, see Consumer Reports (July 1983), Tremblay (1985), Scherer (1994), Elzinga (2001), Kadiyali (1996), and Greer (1998). 4. Our main conclusions are unchanged as long as the rate of increase in costs due to a quality improvement is sufﬁciently low. 5. To preserve our duopoly structure, we assume that either ␸H > ␸m > ␸L > p2/s2, in which case the market is covered, or ␸H > ␸m > p2/s2 > ␸L, in which case the market is uncovered. 6. As each ﬁrm has an incentive to be the high quality ﬁrm in our model, this rank could be due to chance or a ﬁrst mover advantage (implying a dynamic structure to the quality stage game). 7. This follows Grossman and Shapiro’s (1984) speciﬁcation of advertising technology, except that advertising is purely persuasive in our case. 8. We assume that the equilibrium is locally stable (Bulow et al., 1985). That is, the Hessian matrix of the second order conditions of proﬁt maximization for both ﬁrms is negative deﬁnite. 9. This is a unique Nash equilibrium for the following reasons. First, the ﬁrst and second order conditions of proﬁt maximization are satisﬁed in each stage game. Second, the equilibrium price exceeds marginal cost for each ﬁrm [since second order conditions require that sH < (48/7) z and kH, sH, and z are ﬁnite and greater than zero]. Third, the solution is interior for an uncovered market, a condition that is met when p2/s2 = (6kH z)/

A Model of Vertical Differentiation, Brand Loyalty, and Persuasive Advertising

235

(48z ⫺ 7sH ) > ␸L ≥ 0. Finally, since the demand functions are continuous and piecewise linear, it is not proﬁtable for either ﬁrm to undercut the other. 10. With zero ﬁxed costs, ␲1 > ␲2 when 0 < sH < 5.88z and ␲1 < ␲2 when 5.88 < sH < 6.86z. Second order conditions of proﬁt maximization require that sH < 6.86z. 11. Another possible cause of this result is that it may be easier for the high quality producer to use advertising to create loyalty to its brand than for the low quality ﬁrm to negate it. The fact that rivals rarely use comparative advertising to attack the reputation of other ﬁrms suggests that this is a possibility. Self imposed guidelines for comparison advertising of the American Association of Advertising Agencies, which state that “The competition should be fairly and properly identiﬁed but never in a tone of voice that degrades the competitive product or service,” may also make it difﬁcult for the low quality ﬁrm to use advertising to negate consumer loyalty to the high quality brand (Mandell, 1984, p. 459). 12. Assuming ﬁrm participation, which requires that z > (2/3)y, this is a unique Nash equilibrium for the following reasons. First, the ﬁrst and second order conditions of proﬁt maximization are satisﬁed in each stage game. Second, the equilibrium price exceeds marginal cost for each ﬁrm [since z > (2/3)y and y, kL, kH , and z are ﬁnite and greater than zero]. Third, conditions will be discussed shortly under which an interior solution does exist when the market is uncovered (␸H > ␸m > p2/s2 > ␸L). Finally, since the demand functions are continuous and piecewise linear, it is not proﬁtable for either ﬁrm to undercut the other. 13. All of the conditions of a Nash equilibrium are met and the market is covered under these conditions because ␸H = 2.4375 > ␸m = 1.125 > ␸L = 0.9375 > p2/s2 = 0.1875 in equilibrium. 14. To assure a covered market and ﬁrm participation in this case, the following conditions must hold: z > 1.1111, 0.6667 < kL < 1.2, and 1.6667 < kH < 3.0.

ACKNOWLEDGMENTS We would like to thank Laura Connolly, Art O’Sullivan, Steve Polasky, and Carol Tremblay for helpful comments. Dong Woon Noh and Okmyung Bin provided valuable research assistance. The usual caveat applies.

REFERENCES Akerlof, G. A., & Kranton, R. E. (2000). Economics and Identity. Quarterly Journal of Economics, 115, 715–753. Bain, J. S. (1956). Barriers to New Competition. Oxford: Basil Blackwell. Becker, G. S., & Murphy, K. M. (1993). A Simple Theory of Advertising as a Good or Bad. Quarterly Journal of Economics, 104, 941–964. Block, F., & Manceau, D. (1999). Persuasive Advertising in Hotelling’s Model of Product Differentiation. International Journal of Industrial Organization, 17, 557–574. Bulow, J. I., Geanokoplos, J. D., & Klemperer, P. D. (1985). Multimarket Oligopoly: Strategic Substitutes and Complements. Journal of Political Economy, 93, 488–511.

236

VICTOR J. TREMBLAY AND CARLOS MARTINS-FILHO

Caves, R. E., & Greene, D. P. (1996). Brands’ Quality Levels, Prices, and Advertising Outlays: Empirical Evidence on Signals and Information Costs. International Journal of Industrial Organization, 14, 29–52. Caves, R. E., & Porter, M. (1977). From Entry Barriers to Mobility Barriers: Conjectural Decisions and Contrived Deterrence to New Competition. Quarterly Journal of Economics, 91, 241–261. Choi, C. J., & Shin, H. S. (1992). A Comment on a Model of Vertical Product Differentiation. Journal of Industrial Economics, 40, 229–231. Consumer Reports (1983). Beer. July, 342–378. Cremer, H. C., & Thisse, J.-F. (1991). Location Models of Horizontal Differentiation: A Special Case of Vertical Differentiation Models. Journal of Industrial Economics, 39, 383–390. d’Aspremont, C., Gabszewicz, J. J., & Thisse, J.-F. (1979). On Hotelling’s Stability in Competition. Econometrica, 47, 1145–1151. Dixit, A., & Norman, V. (1978). Advertising and Welfare. Bell Journal of Economics, 9, 1–17. Economist (1992). Strategic Shopping. September 26, 82–87. Elzinga, K. G. (2001). Beer. In: W. Adams & J. W. Brock (Eds), The Structure of American Industry (10th ed., pp. 85–113). Prentice Hall, Englewood Cliffs. Farr, S., Tremblay, C. H., & Tremblay, V. J. (2001). The Welfare Effect of Advertising Restrictions in the U.S. Cigarette Industry. Review of Industrial Organization, 18, 147–160. Fisher, F. M., & McGowan, J. J. (1979). Advertising and Welfare: Comment. Bell Journal of Economics, 10, 726–727. Foxall, G. R., & Goldsmith, R. E. (1994). Consumer Psychology for Marketing. New York: Routledge. Gabszewicz, J. J., & Thisse, J.-F. (1979). Price Competition, Quality, and Income Disparities. Journal of Economic Theory, 20, 327–338. Gabszewicz, J. J., & Thisse, J.-F. (1992). Location. In: R. J. Aumann & S. Hart (Eds), Handbook of Game Theory (Vol. 1, pp. 281–304). Amsterdam: North Holland. Gibbons, R. (1992). Game Theory for Applied Economists. Princeton: Princeton University Press. Greer, D. R. (1998). Beer: Causes of Structural Change. In: L. L. Duetsch (Ed.), Industry Studies (2nd ed., pp. 28–64). London: M. E. Sharpe. Grossman, G. M., & Shapiro, C. (1984). Informative Advertising with Differentiated Products. Review of Economic Studies, 51, 63–81. Hallagan, W., & Joerding, W. (1983). Polymorphic Equilibrium in Advertising. Bell Journal of Economics, 14, 191–201. Hotelling, H. (1929). Stability in Competition. Economic Journal, 39, 41–57. Johnson, B. (1997). Gateway Exudes Politeness Before NFL’s Big Game. Advertising Age, (January 20), 12. Kadiyali, V. (1996). Entry, its Deterrence, and its Accommodation: A Study of the A Study of the U.S. Photographic Film Industry. Rand Journal of Economics, 27, 452–478. Kaldor, N. (1949–1950). The Economics of Advertising. Review of Economic Studies, 17, 1–21. Klein, B., & Lefﬂer, K. B. (1981). The Role of Market Forces in Assuring Contractual Performance. Journal of Political Economy, 89, 615–641. Kokkinaki, F., & Lunt, P. (1999). The Effect of Advertising Message Involvement on Brand Attitude Accessibility. Journal of Economic Psychology, 20, 41–51. Lancaster, K. (1966). A New Approach to Consumer Theory. Journal of Political Economy, 74, 132–157.

A Model of Vertical Differentiation, Brand Loyalty, and Persuasive Advertising

237

Lee, L. W. (1997). Persuasive Advertising and Socialization. International Journal of the Economics of Business, 4, 203–214. Lefﬂer, K. G. (1981). Persuasion or Information? The Economics of Prescription Drug Advertising. Journal of Law and Economics, 34, 45–74. Mandell, M. I. (1984). Advertising. Englewood Cliffs, NJ: Prentice Hall. Milgrom, P., & Roberts, J. (1986). Price and Advertising Signals of Product Quality. Journal of Political Economy, 94, 796–821. Mussa, M., & Rosen, S. (1978). Monopoly and Product Quality. Journal of Economic Theory, 18, 301–317. Nelson, P. (1974). Advertising as Information. Journal of Political Economy, 82, 729–754. Putsis, W. P. Jr., & Cotterill, R. W. (1999). Share, Price, and Category Expenditure – Geographic Market Effects and Private Labels. Managerial and Decision Economics, 20, 175–187. Seldon, B. J., & Jung, C. (1995). The Length of the Effect of Aggregate Advertising on Aggregate Consumption. Economics Letters, 48, 207–211. Shapiro, C. (1980). Advertising and Welfare: Comment. Bell Journal of Economics, 11, 749–752. Scherer, F. M. (1994). Industry Structure, Strategy, and Public Policy. New York: Harper Collins. Stigler, G. J. (1961). The Economics of Information. Journal of Political Economy, 69, 213–225. Telser, L. (1964). Advertising and Competition. Journal of Political Economy, 72, 551–567. Tirole, J. (1988). The Theory of Industrial Organization. Cambridge, MA: MIT Press. Tremblay, C. H., & Tremblay, V., J. (1995). Advertising, Price, and Welfare: Evidence from the U.S. Brewing Industry. Southern Economic Journal, 62, 367–381. Tremblay, V. J. (1985). Strategic Groups and the Demand for Beer. Journal of Industrial Economics, 34, 183–198. Tremblay, V. J., & Polasky, S. (2000). Advertising with Subjective Horizontal and Vertical Product Differentiation. Mimeo, Oregon State University. Von der Fehr, N.-H., & Stevik K. (1998). Persuasive Advertising and Product Differentiation. Southern Economic Journal, 65, 113–126. Wauthy, X. (1996). Quality Choice in Models of Vertical Differentiation. Journal of Industrial Economics, 44, 345–353. Wiggins, S. N., & Raboy, D. G. (1996). Prices Premia to Name Brands: An Empirical Analysis. Journal of Industrial Economics, 44, 377–389. Wills, R. L., & Mueller, W. F. (1989). Brand Pricing and Advertising. Southern Economic Journal, 56, 383–395. Witt, U. (1991). Economics, Sociobiology, and Behavioral Psychology on Preferences. Journal of Economic Psychology, 12, 557–573. Zajonc, R. (1968). The Attitude Effects of Mere Exposure. Journal of Personality and Social Psychology, Monograph Supplement, 9, 1–27.

238

VICTOR J. TREMBLAY AND CARLOS MARTINS-FILHO

APPENDIX Proof of Proposition 1 (i) (ii)

(iii) (iv) (v)

From the quality stage of the game, it was shown that, s*1 = sH > s*2 = (4/7)sH . Because ␸L does not enter ﬁrm 2’s demand function, only ﬁrm 1 has an incentive to use persuasive advertising. As long as it is proﬁtable for ﬁrm 1 to advertise, then A*1 > A*2 = 0. Because s*1 > s*2 and given the price equations in (5), p*1 > p*2. Because s*1 > s*2 and given the demand functions in (2) and (4) and the bestreply price equations in (5), D*1 > D*2. Given the ﬁnal stage proﬁt functions in (8), ␲*1 > ( < )␲*2 if A*1 is sufﬁciently small (large). Q.E.D. Proof of Proposition 2

(i) From the quality stage of the game, it was shown that s*1 = sH > s*2 = sL. (ii) Regarding advertising, the ﬁnal step proﬁt functions are used to derive the ﬁrst order conditions of proﬁt maximization for each ﬁrm (second order conditions are assumed to hold): ⭸␲1 = ␣*1 ⫺ A⬘1(␸*H ) = 0; ⭸␸H

⭸␲2 = ␣*2 ⫺ A⬙2(␸*L) = 0. ⭸␸L

4(2␸*H ⫺ ␸*L)(sH ⫺ sL) ; 9

␣*2 ⬅

(A1)

where: ␣*1 ⬅

4(-␸*H + 2␸*L)(sH ⫺ sL) . (A2) 9

Because | ␣*1 | > | ␣*2 |,| A⬘1(␸*H) | > | A⬘2(␸*L) |. By symmetry and strict convexity of the advertising functions, A*1 > A*2. (iii) Because ␸H > 2␸L and s*1 > s*2 and given the best-reply price equations in (12), p*1 > p*2. (iv) Because ␸H > 2␸L and s*1 > s*2 and given the demand functions in (2) and (3) and the best-reply price equations in (12), D*1 > D*2. (v) Given the final stage profit functions, ␲*1 > ( < )␲*2 if A*1 ⫺ A*2 is sufficiently small (large). Q.E.D.

ALCOHOL ADVERTISING AND ADVERTISING BANS: A SURVEY OF RESEARCH METHODS, RESULTS, AND POLICY IMPLICATIONS Jon P. Nelson ABSTRACT This chapter surveys the literatures on advertising bans and alcohol consumption or abuse, and advertising expenditures and alcohol consumption. Studies of state-level bans of billboards are examined as well as studies of international bans that cover broadcasting media. For expenditures, the survey concentrates on econometric methods and the existence of an industry advertising-sales response function. Selected results from survey-research studies of advertising and youth alcohol behaviors also are discussed. The chapter concludes that advertising bans do not reduce alcohol consumption or abuse; advertising expenditures do not have a marketwide expansion effect; and survey-research studies of youth behaviors are seriously incomplete as a basis for public policy. Results of the survey are applied to the Supreme Court’s Central Hudson test for constitutionality of restrictions on commercial speech.

I. INTRODUCTION In recent years there has been a great deal of controversy over the role of advertising as a possible stimulus to alcohol consumption and as a contributor Advertising and Differentiated Products, Volume 10, pages 239–295. Copyright © 2001 by Elsevier Science Ltd. All rights of reproduction in any form reserved. ISBN: 0-7623-0823-0

239

240

JON P. NELSON

to alcohol abuse, including underage purchase and consumption of alcohol. Although the general issue is longstanding, the current controversy stems in part from four recent actions: ﬁrst, a 1991 report on Youth and Alcohol – Controlling Alcohol Advertising, by the U.S. Department of Health and Human Services (DHHS, 1991) and the accompanying public statement by Surgeon General Novello. Second, the decision in 1996 by the Distilled Spirits Council to lift the industry’s 48-year-old voluntary ban on liquor advertising on television and radio. Third, a federal appeals court decision in 1996 that upheld a constitutional challenge to a Baltimore City ordinance prohibiting outdoor advertising of alcoholic beverages in most areas of the City. Fourth, successful efforts at banning tobacco advertising, including the April 1999 ban of billboard advertising as part of the Master Settlement Agreement. The last two actions in particular sparked a number of local ordinances that severely restrict billboards and other publicly visible displays of alcohol advertising, including ordinances passed in Chicago, Cleveland, Detroit, Los Angeles, and Oakland. The purpose of this chapter is to review critically the available studies that attempt to establish an econometric relationship between: (1) advertising bans and alcohol consumption or abuse; and (2) advertising expenditures and alcohol consumption. Because the literature on expenditures has been previously reviewed, the survey concentrates on advertising bans as a possibly effective public policy.1 The portion of the survey that deals with expenditure studies is focused on recent studies and advances in econometric techniques. Among previous surveys of alcohol advertising, the work by Saffer (1993, 1995, 1996, 1998) stands out for both the number of econometric issues raised and the number of surveys published in various outlets. Further, a recent encyclopedia entry by Saffer, Grossman and Chaloupka (2000, p. 47) concluded that advertising bans can reduce alcohol consumption, which is a stronger statement than was contained in Saffer’s earlier surveys. In sharp contrast, a selective survey by Cook and Moore (2000, p. 1645) concluded that available evidence “. . . precluded a conﬁdent conclusion about whether the regulation of commercial advertising is a potentially important policy instrument.” Further, a survey of recent econometric and related studies by the National Institute on Alcohol Abuse and Alcoholism (NIAAA, 2000, p. 422) found that “. . . the results of research on the effects of alcohol advertising are mixed and not conclusive.” Hence, an objective of the present survey is to determine which of these conclusions is warranted by a thorough examination of the evidence on advertising bans and expenditures. I also discuss several survey-research studies of youth alcohol behaviors and television advertising. Section II provides a summary of commercial speech law as established by the Supreme Court. The Central Hudson four-part test of constitutionality is the

Alcohol Advertising and Advertising Bans

241

legal framework necessary for policy evaluation of many advertising regulations. Section III examines econometric studies of advertising bans, alcohol consumption, and alcohol abuse. Both local bans of publicly visible displays (billboards and window displays) and broadcast advertising bans are discussed. Section IV examines recent studies of advertising expenditures, with an emphasis on advances in econometric methods. A number of econometric approaches are discussed, which involve issues of speciﬁcation and measurement of demand functions containing advertising as an explanatory variable. A review of several survey-research studies of television advertising and youth alcohol behaviors also is presented in this section, and I discuss the relationship between “adult” alcohol consumption and measures of youth alcohol behaviors. Section V addresses the issue of a market- or industry-wide advertising response function. New empirical evidence is presented for beer, wine, and spirits that raises important questions regarding advertising spillovers and the existence of a relationship between brand or category advertising and the marketwide demand for alcohol. Only a few previous alcohol studies have addressed this issue in a comprehensive manner, and the empirical work by Gius (1996) on liquor brand advertising is almost singular. The new evidence presented in Section V further demonstrates the difﬁculty of inferring an industry-wide (or beverage-wide) relationship from data on brand (or category) advertising and market shares. Section VI contains conclusions from the survey.

II. COMMERCIAL SPEECH LAW: THE FIRST AMENDMENT AND THE CENTRAL HUDSON “BALANCING” TEST Advertising of alcoholic beverages is a potential public health issue if it can be shown that advertising has a direct and material effect on alcohol consumption or that advertising has a direct and material effect on alcohol abuse outcomes (generally or for speciﬁc populations). With this in mind, the law on commercial speech provides the framework necessary for evaluation of public policies involving advertising bans or regulation of media and content. This section summarizes the law on commercial speech and, in particular, develops the Supreme Court’s four-part “balancing” test in its 1980 landmark Central Hudson decision. I ﬁrst trace the history of the Central Hudson test, and then examine two recent court cases that take up the constitutionality of restrictions on advertising of alcoholic beverages. These cases provide the impetus for the remainder of the survey.

242

JON P. NELSON

Among the Federal agencies, the Federal Trade Commission (FTC) regulates commercial speech that is deceptive or unfair, including ads that target youth for adult products (Starek, 1997). “Deceptive” advertising is deﬁned by the FTC as advertising that would likely affect an individual’s conduct in a material way when the individual is acting reasonably under the circumstances (Beales & Muris, 1993, p. 25). “Unfair” advertising involves damages that are both likely and substantial, and which are not offset by countervailing beneﬁts to consumers or competition. Although restrictive petitions on alcohol advertising have been received by the FTC (1985), the agency does not attempt to regulate advertising based solely on media or content, such as television advertisements based on emotional appeals or lifestyle themes. The Bureau of Alcohol, Tobacco and Firearms also regulates advertising claims that are deceptive, misleading, disparaging of a competitor’s product, obscene, or indecent. Lastly, the Federal Communications Commission (FCC) has the statutory authority to insure that broadcasters serve “the public interest” (O’Neil, 1997). FCC rules and regulations do not forbid advertising of alcoholic beverages in broadcasting, although the three major networks have self-imposed guidelines. At the state level, a number of individual states have restricted alcohol advertising, ranging from outright bans of print and outdoor advertising to prohibitions on off-premise advertising of alcohol prices. The Supreme Court’s decision in 44 Liquormart in 1996 ended a number of these controls, especially bans of price advertising. Industry self-regulation also has been an important part of the debate on alcohol advertising. While self-regulation is an interesting topic (Calfee, 1997; FTC, 1999), this section is only concerned with the limits on advertising permissible under the First Amendment to the Constitution, and provides a review of leading Supreme Court cases. Readers seeking more indepth analysis should consult specialized treatises on commercial speech and advertising (Moore et al., 1998; Rome & Roberts, 1985). A. Early Constitutional Cases, 1942–1979 In 1942, the Supreme Court decided the case of Valentine v. Chrestensen, 316 U.S. 52 (1942). Mr. Chrestensen owned a former Navy submarine, which he exhibited for proﬁt at a pier in New York City. He had prepared a handbill advertising the boat, which solicited customers for a stated admission fee. A City sanitary ordinance forbade the distribution of commercial and business advertising in the streets, except for bills and newspapers devoted to information or public protest. Advised that he was violating the ordinance, Chrestensen produced a double-sided bill that included a protest against the City and that removed only the commercial information on the admission fee.

Alcohol Advertising and Advertising Bans

243

The City advised that distribution of the revised bill would violate the law, and it arrested the respondent when he proceeded with its distribution. Asserting a violation of the First Amendment, Chrestensen was granted injunctive relief by the lower courts. Upon appeal, the Supreme Court considered whether the City’s ordinance was an unconstitutional abridgement of the freedom of the press and speech. The Court disregarded the political protest, and held unanimously that “the Constitution imposes no . . . restraint on government as respects purely commercial advertising” (316 U.S. 54). The ruling left to the states all questions of promotion of “gainful occupation in the streets” (316 U.S. 52). For the next 30 years the three-page Valentine decision, which was unsupported by any citations or historical reasoning, summarized the law on commercial speech. However, in the 1970s the value of constitutional protection for commercial speech was reconsidered. In Bigelow v. Virginia, 421 U.S. 809 (1975), the Court struck down a Virginia statute that prohibited advertising that encouraged or promoted abortions. A New York City abortion referral service had placed ads in an “underground weekly newspaper” that served the University of Virginia. In a 7–2 decision, the Court held that Virginia’s legitimate interest in the quality of medical care did not outweigh its citizens’ First Amendment right to receive information about lawful medical services. Valentine was repudiated, although the majority declined to decide the precise limits to which a state might legally regulate or prohibit advertising in order to advance a legitimate public interest (421 U.S. 825). A year later in 1976, the value of commercial speech was revisited in Virginia State Board of Pharmacy v. Virginia Citizens Consumer Council, Inc., 425 U.S. 748 (1976). The issue was a Virginia statute that made it unprofessional conduct for a pharmacist to advertise prices of prescription drugs. The case was brought by a Virginia consumer group asserting a right to receive the competitive beneﬁts of price advertising. In a 7–1 decision, the Court held that: (1) First Amendment protection for information on drug prices was a protection enjoyed by both retailers (speakers) and consumers (hearers); (2) since speech was protected under the First Amendment, the advertising of prices was protected speech notwithstanding its commercial character; and (3) the pharmacists’ justiﬁcation for the ban as maintaining a high degree of professionalism was insufﬁcient. With regard to the last issue, the pharmacists claimed that advertising might raise prices due to added expenses, destroy stable pharmacist-customer relationships, and reduce the status of a pharmacist to a mere retailer (425 U.S. 768). Faced with these age-old claims of ruinous and destructive competition, Justice Blackman held that

244

JON P. NELSON

People will perceive their own best interests if only they are well enough informed, and . . . the best means to that end is to open the channels of communication rather than close them. If they are truly open, nothing prevents the “professional” pharmacist from marketing his own assertedly superior product . . . . It is precisely this kind of choice, between the dangers of suppressing information, and the dangers of its misuse if it is freely available that the First Amendment makes for us. Virginia is free to require whatever professional standards it wishes of its pharmacists . . . but it may not do so by keeping the public in ignorance of the entirely lawful terms that competing pharmacists are offering (425 U.S. 770).

The ruling in Virginia Pharmacy made clear that some speech regulations might survive constitutional challenges, such as “mere time, place, and manner” restrictions that could be justiﬁed without reference to content. Having accorded a limited degree of protection to commercial speech, the Court during the period 1976–1979 struck down restrictive regulations for contraceptive advertising, lawyer advertising, and posting of “For Sale” or “Sold” yard signs by real estate agents. In the latter case, Linmark Associates, Inc. v. Township of Willingboro, 431 U.S. 85 (1977), the ruling held that an ordinance designed to prevent “panic selling” in a racially-integrated community near Fort Dix could not be sustained. The Court held that the ordinance was not necessary to achieve the government’s objective and in any event, impermissibly impaired the ﬂow of truthful information. Justice Marshall reasoned that Willingboro has proscribed particular types of signs based on their content because it fears their ‘primary’ effect . . . . That the proscription applies to only one mode of communication . . . does not transform it into a ‘time, place or manner’ case . . . . If the ordinance is to be sustained, it must be on the basis of the township’s interest in regulating the content of communication, and not on any interest in regulating the form (431 U.S. 95–96).

The decision also noted that respondents had failed to prove that “For Sale” signs were in fact a major cause of panic selling or that the ban was effective in reducing sales (431 U.S. 95). B. Central Hudson: A Four-Prong Test of Constitutionality In 1980, the Court distilled earlier cases and worked out a four-prong test to be used in balancing governmental interests in regulation of advertising with the interests protected by the First Amendment. Central Hudson Gas & Electric Corp. v. Public Service Commission of New York, 447 U.S. 557 (1980), involved a ban of promotional advertising of electricity originally enacted during the energy crisis of the early 1970s, which was later extended in the interest of conservation. Purely institutional and informative advertising was allowed, but the Commission argued that promotional ads would stimulate

Alcohol Advertising and Advertising Bans

245

electricity consumption, provide the public with misleading signals regarding energy conservation, and unfairly disrupt electricity rate structures. The utility challenged the ban in New York state courts, claiming violation of the First and Fourteenth Amendments. However, the prohibition was upheld by the Court of Appeals of New York on the grounds that the governmental interests outweighed the value of the commercial speech. This outcome was reversed by the Supreme Court in an 8–1 decision, including three concurring judgments. The opinion by Justice Powell reasoned that We have rejected the “highly paternalistic” view that the government has complete power to suppress or regulate commercial speech . . . . Our decisions have recognized the “commonsense” distinction between speech proposing a commercial transaction, which occurs in an area traditionally subject to government regulation, and other varieties of speech . . . . The Constitution . . . accords a lesser protection to commercial speech than to other constitutionally guaranteed expression . . . . The protection available for particular commercial expression turns on the nature both of the expression and of the governmental interests served by its regulation (447 U.S. 562).

Given this difference, Justice Powell discerned a four-prong test for protection of commercial speech on constitutional grounds (447 U.S. 566): (1) The speech must concern lawful activity and not be misleading; (2) The government’s interest asserted to justify the regulation must be substantial; (3) The regulation must directly advance the governmental interest asserted; and, (4) The regulation must not be more extensive than is necessary to serve that interest. The test can be divided into two parts. The ﬁrst prong articulates a “strict scrutiny” standard or a sufﬁcient condition for government regulation, while the remaining prongs provide the necessary protection from regulation. The third prong suggests an area where econometric and other statistical evidence is particularly of value, since this criterion can be framed in statistical terms as a null hypothesis. From the government censor’s viewpoint, the null hypothesis is that the regulation in question will not directly advance the asserted government’s interest. The positive evidence marshaled in support of an advertising ban or other regulation must be sufﬁciently strong that the null can be rejected with a high degree of conﬁdence. Further, the restriction must have a material effect on the interest, or in the words of Justice Powell “. . . the regulation may not be sustained if it provides only ineffective or remote support for the government’s purpose” (447 U.S. 564). From the plaintiff’s viewpoint,

246

JON P. NELSON

the null hypothesis is that the proposed regulation is direct and material, and the negative evidence must be sufﬁciently strong that the null can be rejected with a high degree of conﬁdence. The fourth prong also is capable of statistical testing and comparative policy analysis. Justice Powell noted that this criterion “. . . recognizes that the First Amendment mandates that speech restrictions be ‘narrowly drawn’ . . . [and] we review with special care regulations that entirely suppress commercial speech in order to pursue a nonspeech-related policy” (447 U.S. 565–566). C. Further Constitutional Developments, 1981–1995 During the 1980s, a number of cases tested the limits of the Central Hudson formula. In Bolger v. Youngs Drug Products Corp., 463 U.S. 60 (1983), the Court considered a Postal Service regulation that prohibited the unsolicited mailing of advertisements for contraceptives. The mailings in question also included informational materials. The government asserted an interest in protecting children from offensive materials. A unanimous ruling held that the party seeking to uphold a restriction on commercial speech carries the burden of justiﬁcation. The Court noted that the regulation would provide only a marginal degree of protection because many magazines contained advertisements for contraceptives, and it held that “. . . the government may not reduce the adult population . . . to reading only what is ﬁt for children” (463 U.S. 73). In Zauderer v. Ofﬁce of Disciplinary Counsel of the Supreme Court of Ohio, 471 U.S. 626 (1985), the ruling again stressed the state’s evidentiary burden in a case involving a challenge to a lawyer advertising regulation. The Court reiterated that the state must show a substantial government interest justifying the restriction and must demonstrate that the restriction vindicated the interest through the least restrictive means available (471 U.S. 647). However, in Posadas de Puerto Rico Associates v. Tourism Company of Puerto Rico, 478 U.S. 328 (1986), the Court considered a law that legalized casino gambling and permitted advertising aimed at tourists, but banned casino advertising within Puerto Rico. The commonwealth asserted an interest in protecting the welfare of its citizens, despite (or because of) the availability of a commonwealth lottery and other local gambling events (478 U.S. 354; see also McChesney, 1997, p. 105). In a 5–4 decision, Justice Rehnquist held that “. . . it is permissible for the government to take the step of allowing the conduct, but reducing the demand through restrictions on advertising” (478 U.S. 346). Although Justice Rehnquist appeared to apply the Central Hudson test, he did

Alcohol Advertising and Advertising Bans

247

not require the government to uphold the fourth prong. Moreover, the third prong, which might appear to justify the majority decision, was never seriously analyzed. The government was only required to show a “reasonable belief” that its restriction on speech would directly advance the asserted interest. In a sharply worded dissent, Justice Brennan citing Virginia Pharmacy Board argued that “. . . the First Amendment presupposes that people will perceive their own best interest if only they are well enough informed, [and] . . . the best means to that end is to open the channels of communication, rather than to close them” (478 U.S. 358). As the Supreme Court has since recognized, the Posadas ruling made little sense. The confusion that followed led to a series of decisions during the 1990s demonstrating that a government censor must satisfy the third and fourth prongs. In 1993, the Court considered the case of Edenﬁeld v. Fane, 507 U.S. 761 (1993), involving a Florida statute forbidding advertising by public accountants. Applying the third prong, the ruling in this case required that the government “. . . must demonstrate that the harms it recites are real and that its restriction will in fact alleviate them to a material degree” (507 U.S. 771). In City of Cincinnati v. Discovery Network, Inc., 507 U.S. 410 (1993), the fourth prong was invoked. The issue was a City ordinance that revoked petitioners’ permits to use 62 newsracks placed on public property (out of 1,500–2,000 citywide racks used by other publications). The racks were used for the distribution of free magazines containing advertisements and other information on current events. The City alleged that the magazines were “commercial handbills,” and that the ordinance was designed to advance its interest in the safety and appearance of streets and sidewalks. However, the Court’s decision held that “. . . it was the city’s burden to establish a ‘reasonable ﬁt’ between its legitimate interests in safety and esthetics and its choice of a limited and selective prohibition of newsracks . . . it has not ‘carefully calculated’ the costs and beneﬁts associated with the burden on speech imposed by its prohibition” (507 U.S. 416). Finally, in Rubin v. Coors, 514 U.S. 476 (1995), the Court unanimously struck down a Federal law passed in 1935 that prohibited labels on beer containers from displaying alcohol content, including words that suggested high content. The opinion by Justice Thomas reasoned that the government had provided only “anecdotal evidence and educated guesses” that the labeling ban inhibited “strength wars” among brewers, and he argued that the government’s interest could be advanced in a less intrusive manner. The decision also noted that wine and spirits producers were required to disclose alcohol content and that a brewer’s wish to disclose alcohol content did not prove a desire to compete on content.

248

JON P. NELSON

D. The 44 Liquormart Decision and Its Application in Anheuser-Busch v. Schmoke In 1996, the Supreme Court again took steps to reduce the confusion created by earlier decisions by invalidating a Rhode Island ban on price advertising of alcoholic beverages in 44 Liquormart, Inc. v. Rhode Island, 517 U.S. 484 (1996). The state of Rhode Island had enacted statutory provisions that: (1) prohibited vendors licensed in Rhode Island – as well as out-of-state manufacturers, wholesalers, and shippers – from advertising the price of any alcoholic beverage offered for sale in Rhode Island; and (2) prohibited Rhode Island media from publishing or broadcasting any advertisements, even referring to sales in other states, that made reference to the price of any alcoholic beverage. In addition, a regulation of the state Liquor Control Administration (LCA) provided that no placard or sign visible from the exterior of a package store could make any reference to prices. The state asserted that the regulations advanced its interest in temperance among its citizens. The Rhode Island laws were enacted in 1956, but similar restrictions had existed for many years in a number of states. Many of these laws dated to the Great Depression era and the repeal of Prohibition, and were blatantly anticompetitive in origin (McGahan, 1995). Indeed, the Rhode Island Liquor Store Association joined with the state as a petitioner in this case, and a lower court decision noted that the laws were in fact designed to protect smaller retailers from price competition. In 1991, complaints from competitors were received by the LCA regarding an advertisement placed by Liquormart in a Rhode Island newspaper.2 The advertisement did not mention liquor prices directly, and it also noted that state laws prohibited the package store from price advertising. However, the advertisement stated low prices for non-liquor items, and it included the word “WOW” in large letters next to pictures of vodka and rum bottles. The LCA assessed a ﬁne of $400. The operators of several liquor stores in Massachusetts also attempted to place price ads in Rhode Island newspapers, but these attempts were refused. The store operators joined together and ﬁled a First Amendment action against the ban. The district court found that: (1) the state’s off-premise price ban had no signiﬁcant impact on levels of alcohol consumption in Rhode Island; and (2) the price ban was unconstitutional because it did not directly advance the state’s asserted interest in reducing alcohol consumption and was more extensive than necessary to serve that interest (829 F. Supp 543, 1993). The court of appeals reversed, reasoning that: (1) there was merit in Rhode Island’s claim that competitive price advertising would lower prices and lower prices would ultimately produce more alcohol sales; and (2) the Twenty-First Amendment that repealed

Alcohol Advertising and Advertising Bans

249

Prohibition gave the statutes an added presumption of validity (39 F.3d 5, 1994). The Supreme Court unanimously reversed the appeals court, holding that the ban did not directly advance the state’s substantial interest in promoting temperance, and the ban was more extensive than necessary to serve that interest. With respect to the third prong, Justice Stevens noted that “. . . the State has presented no evidence to suggest that its speech prohibition will signiﬁcantly reduce marketwide consumption” (517 U.S. 506; emphasis in the original). With regard to the fourth prong, the decision determined that “. . . it is perfectly obvious that alternative forms of regulation that would not involve any restriction on speech would be likely to achieve the State’s goal of promoting temperance. As the State’s own expert conceded, higher prices can be maintained by direct regulation or by increased taxation” (517 U.S. 507). Finally, Justice Stevens clariﬁed the reason for protecting commercial speech: When a State entirely prohibits the dissemination of truthful, nonmisleading commercial messages for reasons unrelated to the preservation of a fair bargaining process, there is far less reason to depart from the rigorous review that the First Amendment generally demands (517 U.S. 501; emphasis added).

The Court’s application in 44 Liquormart of a rigorous Central Hudson analysis suggests the minimum level of scrutiny for evaluating restrictions on the ability of speakers to provide, and their audiences to receive, messages about lawful products and services. The government censor must prove – and not merely assert or assume as part of a regulatory process – that a ban of commercial speech directly and materially advances a substantial government interest. The government also must demonstrate that the restriction is no more extensive than is necessary to promote that interest.3 As noted above, these criteria can be formulated as statistical hypotheses and applied to relevant data for the market in question. Hence, a thorough application of the Central Hudson test necessarily involves an economic analysis of the market, especially examination of the empirical effects of advertising or advertising bans on marketwide demand and related outcomes.4 Further, the balancing of interests required by Central Hudson is a familiar procedure to economists; indeed, it is nothing more than the principle of marginal evaluation of beneﬁts and costs (and cost effectiveness) that underlies all of economic analysis. Moreover, beneﬁts or costs may be weighted to advance certain objectives. For example, placing a heavy burden of proof on the government censor adds the necessary element of protection for the “free trade in ideas.” As noted by Justice Stevens, “The Constitution presumes that attempts to regulate speech are more

250

JON P. NELSON

dangerous than attempts to regulate conduct . . . speech restrictions cannot be treated as simply another means that the government may use to achieve its ends” (517 U.S. 512). However, following the ruling in 44 Liquormart, the Supreme Court vacated and remanded to the appeals court for further consideration the case of Anheuser-Busch, Inc. v. Schmoke, Mayor of Baltimore City, 101 F.3d 325 (1996). Schmoke involved a City ordinance banning most outdoor advertising of alcoholic beverages based on a concern for the “welfare and temperance of minors.” Upon reconsideration, the appeals court upheld the ordinance, reafﬁrmed the district court’s judgment, and concluded that 44 Liquormart did not require it to alter these decisions. The appeals court agreed with the City’s assertion that: (1) there was a direct and material link between advertising and youth alcohol consumption; and (2) the ordinance expressly targeted persons who cannot be legal users of alcoholic beverages. The appeals court rejected plaintiffs’ argument that less restrictive means were available. A dissenting opinion by Judge Butzner noted the lack of an evidentiary hearing by the courts, and he reasoned that “. . . each party should be given the opportunity to present evidence on this issue and to test the strength of the opposing party’s evidence . . . the First Amendment requires careful evaluation . . . performed by a judge – not by a city council” (101 F.3d 332). The Central Hudson formula and its application in recent decisions are background necessary for economic analysis of advertising bans and other government regulations that affect alcohol advertising. Using the third and fourth prongs, it is possible to correctly pose the analytical questions that courts seek to answer in deciding on the constitutionality of a restriction on commercial speech. As demonstrated in 44 Liquormart, the Supreme Court requires something more that educated guesses for constitutional scrutiny of an advertising ban. Hence, Judge Butzner’s statement and the decisions in the two alcohol cases provide the impetus for the remainder of this survey.5 I now turn to the task of reviewing the empirical evidence with regard to: (1) advertising bans and alcohol consumption; and (2) advertising expenditures and alcohol consumption. I also discuss the relationship between research ﬁndings in these two areas and youth alcohol consumption.

III. STUDIES OF ALCOHOL ADVERTISING BANS Communication with consumers is a difﬁcult and costly process; hence, it is also an incomplete activity. Many advertisements are ineffective or reach

Alcohol Advertising and Advertising Bans

251

individuals who have little or no interest in the message. Indeed, many consumers are highly skeptical of advertising claims (Beales & Muris, 1993; Calfee, 1997). However, it cannot be inferred from this outcome that the messages are without social value. As the Supreme Court recognized in Virginia Pharmacy, “. . . advertising, however tasteless and excessive it sometimes may seem, is nonetheless dissemination of information” (425 U.S. 765). Further, advertising is clearly an important part of the competitive process, and this is especially true in markets dominated by branded products that are familiar and well known. Each year the alcoholic beverage industry spends upwards of $1 billion on advertising using conventional media and it is believed to spend an equal amount on other forms of promotion. Does this substantial expenditure affect only brand shares or does it also affect marketwide demand? Do bans of advertising reverse the marketwide effects of advertising and result in materially lower levels of alcohol consumption and abuse? This section examines the second of these two questions. Expenditure studies are discussed in the next section. Two types of econometric studies are reviewed: ﬁrst, studies that use state data containing cross-sectional or timeseries cross-section (panel) variation; and second, studies that use panel data for a sample of countries. The advertising bans in the cross-national studies are wider and cover more media and beverages. Some countries, such as Sweden, restrict virtually all forms of alcohol advertising, including bans on broadcast and print advertising of all beverages. The bans in the state-level studies, however, occur at a lower level of aggregation and the evidence from these studies is applicable to local policy issues, such as bans of billboard advertising. In both cases, the bans are often longstanding, and behavior of individuals should reﬂect the duration of censorship. Hence, I argue below that past or lagged youth alcohol behaviors are captured by studies of advertising bans. Saffer (1991, 1993, 1995) argued that advertising bans are “ideally suited” for examination of the effects of advertising on alcohol consumption and abuse. His view is that many data sets contain insufﬁcient variation and measure the advertising-response function in the region of small marginal returns to sales. In order to increase the variation in the data, Saffer (1993, pp. 134–138) proposed that studies examine state-level and international bans, data for selected media such as billboards and broadcasting, and higher-frequency time series, such as monthly and quarterly data. This section examines fourteen studies of state-level and cross-national advertising bans. The studies are presented in chronological order by category.

252

JON P. NELSON

A. Studies of State-Level Advertising Bans (1) Schweitzer et al. (1983) studied advertising bans using a cross-section of 35 states for the year 1975 (excluded states reﬂect missing data or geographic size). The authors examined state-level per-capita consumption of beer and spirits as well as alcoholism and alcohol-related mortality. Both reduced form and structural models were estimated. The most important variables were income, beer prices, tourism, unemployment, and the minimum legal drinking age. The explanatory variable for advertising was BAN, which was a binary variable equal to one if advertising was prohibited, and zero if any was allowed. In a reduced-form model, BAN was statistically insigniﬁcant for spirits consumption; weakly signiﬁcant and negative for beer consumption; and insigniﬁcant for the abuse outcomes. When the authors estimated the structural model, BAN was insigniﬁcant for both beer and spirits. Using the point estimates, their policy simulations found that “. . . a prohibition on advertising for alcoholic beverages leads not to a general reduction in alcoholic beverage consumption but rather a shift from beer consumption to spirits consumption” (Schweitzer et al., 1983, p. 123; emphasis in the original). (2) Hoadley et al. (1984) examined per-capita consumption of distilled spirits using a cross-section of 48 states for the years 1955, 1960, 1965, 1970, 1975, and 1980. The authors ran separate regressions for each of the six years and pooled models, and analyzed the effects of a host of state regulations and controls. The dummy variables for advertising controls included ADBILS (whether or not billboard advertising was prohibited); ADSIGNS (restrictions on exterior advertising); and ADPRICES (restrictions of price advertising of any kind). The pooled model also included dummy variables for regions and years. In the pooled model, ADBILS had the wrong sign, and ADSIGNS and ADPRICES were insigniﬁcant. The signiﬁcant variables included prices, income, tourism, religion, and state monopoly control. The authors concluded that advertising bans “. . . have been almost totally ineffective as a deterrent on alcohol consumption . . . [and] the most anomalous result comes with restrictions on billboard advertising, where results showed a consistent and fairly large effect in the wrong direction” (Hoadley et al., 1984, p. 396). (3) Ornstein and Hanssens (1985) used a cross-section of 50 states and the District of Columbia for the period 1974–1978. The years and states were pooled together to form a panel of 255 observations. They estimated separate regressions for beer and distilled spirits. Wine was deleted because of the lack of price data. In addition to other alcohol control variables, three

Alcohol Advertising and Advertising Bans

253

advertising dummy variables were included: BILLB (bans of billboard advertising); PPRINT (bans of print price advertising); and PBILLB (bans of billboard price advertising). Mixed results were obtained for these variables. For spirits demand, they found that allowing billboards decreased spirits consumption, while allowing price advertising increased consumption by a small amount. The latter ﬁnding is consistent with the view that price advertising leads to lower prices and higher consumption. The variables with the largest elasticities were price, income, tourism, and state monopoly control. For beer demand, BILLB and PPRINT were insigniﬁcant, and the coefﬁcient for PBILLB indicated that allowing beer price advertising on billboards had a small positive effect on consumption. The authors suggested that beer advertising primarily affects brand shares and not market demand (Ornstein & Hanssens, 1985, p. 210). The main determinates of beer consumption were the youth demographic variable and the minimum legal age, while the price, income, and tourism elasticities were smaller for beer compared to spirits. Further, beer consumption was not signiﬁcantly different between monopoly and license states. The authors concluded that “. . . control laws affecting price have the greatest impact on consumption . . . [but] the inﬂuence of control measures is small relative to that of sociodemographic and economic variables that affect consumers’ overall attitudes toward drinking” (Ornstein & Hanssens, 1985, pp. 210–211). (4) Wilkinson (1985) studied the relationships between total alcohol consumption, advertising, and highway fatalities using a sample of 48 states for 1976–1979. He included two measures of advertising restrictions: ﬁrst, whether states prohibited all advertising in periodicals or on billboards (ADPRHBT); and, second, whether states prohibited price advertising in periodicals and on billboards (ADPRICE). He noted that advertising bans can increase or decrease consumption depending on the net effects of advertising on entry barriers and information ﬂows (Wilkinson, 1985, p. 65). Using a recursive speciﬁcation, the panel model was estimated using two-stage least squares with a random-effects error term. The regression results for alcohol consumption indicated that ADPRICE had a signiﬁcantly negative effect on the per-capita quantity of ethanol consumed, while ADPRHBT was insigniﬁcant (Wilkinson, 1985, p. 113). In Wilkinson (1987), these results are extended to account for possible simultaneity among alcohol consumption, prices, and alcohol outlets. The sample was 45 states for the years 1976–1980. The signiﬁcant determinates of total alcohol consumption were price, income, outlets, religion, state monopoly control, and the minimum legal age. For advertising regulations,

254

JON P. NELSON

he found that the price advertising coefﬁcient was negative and signiﬁcant. However, the additional restraint of forbidding all advertising had an insigniﬁcant effect on demand. The researcher concluded that “. . . the direct effects of regulation on alcohol sales are relatively small. Regulatory restrictions have their largest effect via their impact on the number of outlets” (Wilkinson, 1987, p. 17). (5) In Nelson (1990a, b), I examined the effects of economic and regulatory variables on per-capita consumption of beer, wine, and spirits. The sample was a cross-section of 48 continental states and the District of Columbia for the year 1982. An important feature of these studies was the use of an improved set of price variables for alcoholic beverages, including wine prices. The advertising variable ADBAN was equal to one if the state banned price advertising in print media, including billboards. States that banned all advertising in print or on billboards also were included in this category. Hence, ADBAN combined the two categories used by Wilkinson (1985, 1987) and others. The results indicated that ADBAN was not a signiﬁcant determinate of alcoholic beverage consumption. The signiﬁcant variables were prices, income, tourism, number of outlets, and the minimum legal age (Nelson, 1990a, p. 94). Hence, I concluded that “. . . there is no effect on consumption of advertising bans” (Nelson, 1990b, p. 232). (6) In Nelson (2001), I examined a sample consisting of 45 states for the period 1982–1997. The sample size was 720, and the study improved on previous state-panel studies by examining per-capita total ethanol consumption as well as beverage demands. As a result, I was able to demonstrate substitution among beverages as a response to restrictive alcohol laws and regulations. None of the previous studies had examined both the pattern of demand and the net effects of advertising bans. The study examined the effects of billboard bans; bans of price advertising in print and on billboards; state monopoly control of retail outlets; and minimum legal drinking ages for each of the beverages. The model speciﬁcation also included variables for the own-price, cross-price, income, tourism, age distribution, unemployment rate, regional dummies, and state-speciﬁc exponential time trends. The results indicated that bans of billboards increased consumption of spirits and wine, and reduced consumption of beer. The results for spirits replicated earlier ﬁndings discussed above. More important, billboard bans did not have a negative effect on total alcohol consumption, and this result held across several different samples of states and time periods. When the data were divided into two time periods, billboard bans increased total alcohol demand

Alcohol Advertising and Advertising Bans

255

during 1982–1988, but were insigniﬁcant thereafter. During both time periods, price advertising bans were associated with lower consumption of spirits, higher consumption of beer, and no signiﬁcant effects on total alcohol consumption. This is consistent with price bans raising prices, and with positive spillovers for substitute beverages. The signiﬁcant and material determinates of alcohol demand were prices, income, tourism, youth population, minimum legal age, and state monopoly control. The study concluded that “. . . billboards account for only 8% of total alcohol advertising. Hence, the elimination of this medium would not be expected to substantially or materially affect alcohol consumption” (Nelson, 2001, p. 17). (7) Markowitz and Grossman (1998, 2000) examined the relationship between alcohol consumption and domestic violence toward children. In their ﬁrst study, a sample of 1,147 married or cohabiting individuals was drawn from the 1976 Physical Violence in American Families survey. The authors estimated a reduced-form model in which violent outcomes were affected by state excise tax rates on beer (a proxy for state-level beer price differences), illegal drug prices, and state-level regulatory variables, including outlet density, prohibitions of grocery sales, and laws that restrict alcohol advertising (billboard bans, window display bans, price advertising bans). The outcome variables were measures of the probability of violent acts against children in a family, regressed on the above variables and family characteristics, including income, age, race, etc. The authors found that increased beer taxes and restrictive availability laws would be effective in reducing domestic violence. However, they concluded that “. . . the advertising variables . . . are never statistically signiﬁcant individually or as a set . . . [and] laws restricting advertising of beer are shown to be ineffective in reducing violence” (Markowitz & Grossman, 1998, pp. 318– 320). In an extension of this work, Markowitz & Grossman (2000) separated the data by gender and increased the sample size by employing data from a 1985 survey. When they pooled the data, there were no measurable effects of advertising on violence, including billboard bans and window display bans. Hence, they concluded that “. . . increasing the sample size does not result in measurable effects of the advertising . . . on violence” (Markowitz & Grossman, 2000, p. 280). In summary, seven studies have examined the effects of state-level advertising bans on alcohol consumption and abuse, including billboard bans and bans of other visible displays. These studies are summarized in Table 1. In a few instances a signiﬁcant effect was found, but these results are small in

256

Table 1. Summary of State-Level Studies of Advertising Bans: Print and Outdoor. Coefﬁcient Estimates (st. error)

R-Sq

Sample & Years

Dependent Variables

Advertising Variables

Schweitzer et al. (1983) Hoadley et al. (1984) Ornstein and Hanssens (1984) Wilkinson (1985, 1987)

35 states: 1975

log per-capita beer; log per-capita spirits per-capita spirits

binary variable: if ads banned = 1 binary variable: if billboards allowed binary variable: if billboards allowed binary variable: if ban of billboards & print ads binary variable: if price ad bans in print or billboards (incl. total bans) binary variable: if ban of liquor billboards

–0.110 (0.06) 0.125 (0.10) –0.099 (na)**

0.764 0.790 0.890

–0.105 (0.04)* 0.004 (0.02) –0.008 (0.03)

0.794 0.748 0.653

0.089 (0.06) 0.061 (0.05) –0.043 (0.09) –0.206 (0.14)

n.a. 0.629 0.567 0.564

–0.057 (0.01)* 0.102 (0.03)* 0.129 (0.02)* 0.001 (0.01)

0.799 0.890 0.767 0.803

binary variables: if beer price ads ban; billboard ban; window display ban

0.122 (0.10) –0.097 (0.16) –0.059 (0.11)

n.a.

Nelson (1990a)

48 states: pooled sample for six years 51 states: pooled sample for ﬁve years 51 states: pooled sample for four years and ﬁve years 49 states: 1982

Nelson (2001)

45 states: pooled sample for 16 years (1989–1997 results are reported)

Markowitz and Grossman (1998, 2000)

1998: 1,147 families 2000: 2,675 families (female results are reported)

log per-capita spirits log per-capita beer log per-capita total ethanol, 1976–1979 & 1976–1980 log per-capita beer; log per-capita wine; log per-capita spirits log per-capita beer; log per-capita wine; log per-capita spirits; log per-capita total ethanol severe child violence any child violence

Notes: * indicates signiﬁcant at 1% level or better; and ** for 5% or better. Page numbers for coefﬁcient estimates are Schweitzer et al. (1983, p. 118); Hoadley et al. (1984, p. 118); Ornstein and Hanssens (1984, pp. 207, 210); Wilkinson (1985, p. 113; 1987, p. 12); Nelson (1990a, p. 92); Nelson (2001, p. 23); and Markowitz and Grossman (2000, p. 279). n.a. = not available.

JON P. NELSON

Study

Alcohol Advertising and Advertising Bans

257

magnitude or statistically fragile (e.g. Markowitz & Grossman, 1998, p. 318). The policy issue is whether or not there is a robust negative relationship between bans and drinking, but the evidence clearly speaks against this outcome. None of these studies produced results that support the null hypothesis that advertising bans will materially reduce alcohol consumption or alcohol abuse. Similar results have been found for bans on cigarette advertising on college campuses (Czart et al., 2001). Additional comments on the policy implications of these ﬁndings are contained in the concluding portion of this section. Cross-national studies of broadcast advertising bans are examined next. B. Cross-National Studies of Broadcast Advertising Bans (1) Smart and Cutler (1976) studied a 14-month ban of all alcohol advertising, which was instituted in British Columbia (B.C.) in 1971 by the Social Credit government. The law prohibited advertising in newspapers, radio, television, and outdoor advertising on billboards, notice-boards, and the like. Advertisements that originated outside of the province were not affected by the ban. The ban was not popular and it was ended in late 1972 when the NDP party won the provincial election. Smart and Cutler (1976, p. 16) used data on per-capita beverage consumption for the yearly period 1962–1972 and monthly from January 1970 to December 1973. For the annual data, they compared per-capita consumption by beverage for B.C. and Ontario. After detrending the monthly data, they examined before-andafter consumption levels in B.C. They concluded that “. . . both the yearly and monthly analysis of beer, wine or liquor consumption show no substantial effect of the ban” (Smart & Cutler, 1976, p. 20). Hence, there was no effect of a ban that included broadcast advertising of alcoholic beverages. (2) Ogborne and Smart (1980) examined a Manitoba law passed in 1974 that banned all beer advertising from the province’s electronic and print media. Wine and spirits producers continued to advertise in print media, and some wine advertising occurred on television. Using time-series methods applied to monthly beer sales for January 1970 to January 1978, the authors were unable to detect any effects of the ban. A comparison of Manitoba with Alberta over the same time period revealed that beer sales in the two provinces did not differ signiﬁcantly during the four years covered by the advertising ban (Ogborne & Smart, 1980, p. 295). Hence, there was no effect on the beverage speciﬁcally targeted by the broadcast ban.

258

JON P. NELSON

(3) Makowsky and Whitehead (1991) studied the termination in 1983 of a 58-year-old advertising ban of alcoholic beverages in Saskatchewan. The termination affected provincial advertising on radio, television, newspapers, and magazines for beer and wine. Spirits advertising on radio and TV continued to be banned under regulations of the Canadian Radio and Television Commission. A similar comprehensive advertising ban existed in the province of New Brunswick, but it was not terminated at this time. Monthly sales of beer, wine, and spirits were examined for the years 1981–1987 using Box-Jenkins interrupted time-series analyses for Saskatchewan (treatment area) and New Brunswick (control area). Differential or beverage-speciﬁc effects were found. The termination of the ban increased beer sales in Saskatchewan, reduced sales of spirits, and left unaffected the consumption of wine and total ethanol. The authors concluded that there was “. . . evidence of an impact of the change in legislation regarding alcohol advertising in terms of a substitution effect of beer sales for spirits sales. Advertising does not, however, affect total consumption” (Makowsky & Whitehead, 1991, p. 566). (4) Saffer (1991) examined the cross-national effects of laws that ban broadcast advertising of alcoholic beverages. He used a panel of 17 countries over the period 1970–1983. The countries were members of the Organization of Economic Cooperation and Development (OECD), including the U.S. and Canada. Saffer regressed per-capita ethanol consumption on several explanatory variables, including advertising bans for all beverages; bans of broadcast advertising of spirits; real price; real income; alcohol sentiment (a binary variable for percentage alcohol consumed as beer and wine); and tourism. Separate regressions also were estimated for motor vehicle fatality rates and liver cirrhosis mortality rates as alcohol abuse outcomes. Among the difﬁculties in this type of study are the substantial differences that exist in a cross-national data set. For example, cultural differences (“drinking sentiment”) are difﬁcult to measure as are many other determinates of international drinking levels and patterns. This is especially the case for the Scandinavian countries, which operate state monopolies for production and retailing and which also have very strict laws on drunk driving. The author recognizes this problem; he states that “. . . the omission of these variables could result in biased estimates of the effects of advertising bans” (Saffer, 1991, p. 71). Using a yearly ﬁxed-effects model, the signiﬁcant variables for consumption were price, income, drinking sentiment, tourism, and advertising bans. The author concluded that “. . . advertising bans have a signiﬁcant effect in reducing all three measures” (Saffer, 1991, p. 78). For the consumption

Alcohol Advertising and Advertising Bans

259

measure, he reported that “. . . countries with bans on spirits advertising have about 16% lower alcohol consumption than countries with no bans and that countries with bans on beer and wine advertising have about 11% lower alcohol consumption than countries with bans only on spirits advertising” (Saffer, 1991, p. 77). These are large estimates, which is an indication of possible omitted variable bias. Further, three other crossnational studies have been unable to replicate these ﬁndings. (5) Young (1993) studied the same OECD data used by Saffer. However, Young (1993, p. 225) corrected for serial correlation in the data, and he also changed the model speciﬁcation by using binary variables for each country and by directly entering binary variables for the type of advertising restrictions in place. His ﬁrst set of estimates showed that “. . . the mix of signs and general lack of statistical signiﬁcance implies that there is no convincing evidence to support the hypothesis that advertising bans are negatively associated with consumption or death rates” (Young, 1993, p. 222). Next, Young disaggregated the consumption data and estimated the demand for each beverage. He found that a ban of spirits advertising resulted in higher beer consumption and a ban of all advertising resulted in higher spirits consumption. Wine consumption was unaffected by a total ban and negatively affected by a ban of spirits advertising (Young, 1993, p. 224). This is an important ﬁnding since it showed that bans can lead to substitution among beverages. Young concluded that the “. . . the relationships between advertising bans and consumption of speciﬁc types of alcoholic beverages are largely inconsistent with the notion that bans reduce consumption” (Young, 1993, p. 227). (6) Calfee and Scheraga (1994) examined the alcoholic beverage markets in France, Germany, Netherlands, Sweden, and the U.K. The time period varied by country and was approximately 1970–1990. Sweden banned all alcohol advertising after 1979. Advertising expenditures were used as the explanatory variable for four countries, and Sweden was treated as a control case. This is the only study available that uses cross-national data on annual advertising expenditures. Further, advertising outlays varied substantially, so statistical inferences were possible. As noted by the authors, in France, “. . . advertising more than tripled during the years under examination, and in two others [Netherlands and the U.K.], advertising approximately doubled” (Calfee & Scheraga, 1994, p. 303). The regression model included price, income, and a time trend. Advertising expenditures were insigniﬁcant for Germany, Netherlands, and the U.K., and signiﬁcantly negative for France. Regression results for the four countries without bans were not appreciably different from Sweden, and

260

JON P. NELSON

alcohol consumption declined in all ﬁve countries during the study period. The study concluded that “. . . social forces other than prices and income were bringing about a strong reduction in demand for alcoholic beverages, and advertising did nothing to ward off this trend towards reduced consumption” (Calfee & Scheraga, 1994, p. 305). (7) Nelson and Young (2001) updated and expanded the earlier analyses of seventeen OECD countries. We argued that broadcast bans can lead to price reductions due to reduced product differentiation and lower costs, and might therefore increase price competition among producers of more homogeneous goods. Hence, banning advertising could have the perverse effect of increasing alcohol consumption. We also discussed how other alcohol control policies could affect the regression results, and how consumption levels in Scandinavian countries are underreported due to cross-border purchases, duty-free imports, and legal and illegal home production. The expanded consumption model included explanatory variables for real price, real income, drinking sentiment (percent consumption as wine), tourism, age distribution, and unemployment. Several of these variables were omitted by Saffer (1991). Using data for 1977–1995, we found that bans of broadcast advertising of spirits led to greater alcohol consumption and higher rates of motor vehicle fatalities. Moreover, wider bans of additional media and beverages were not consistently related to alcohol consumption or abuse. Real prices were shown to be an important determinant of alcohol consumption, and the youth variable was statistically signiﬁcant as a determinate of highway fatalities. This study conﬁrmed the earlier ﬁndings of Young (1993) and Calfee and Scheraga (1994), and all three studies refuted the ﬁndings of Saffer (1991) for advertising bans in OECD countries. Table 2 summarizes the results in the seven international studies of alcohol advertising bans. Only the study by Saffer (1991) suggests that a ban of broadcast advertising will have a signiﬁcant and material effect on alcohol consumption and abuse. The other six studies do not support this result, including three studies that used similar OECD data. Hence, it must be concluded that the available evidence does not convincingly support the hypothesis that broadcast (and wider) advertising bans reduce alcohol consumption or alcohol abuse. In a related study, Saffer and Chaloupka (2000) examined the effects of tobacco advertising bans by using an international panel that covered 22 OECD countries for the period 1970–1992. An innovation in this study was the measurement of the number of media banned in each country, which ranged

Dependent Variables

Advertising Variables

per-capita beer

comparison with Ontario and beforeafter ban time-series analysis

Study

Sample & Years

Smart and Cutler (1976)

British Columbia, 1962–1972 & 1970– 1973; before-after Manitoba, 1970– 1978; before-after Saskatchewan, 1981–1987; beforeafter end of ban

per-capita gallons beer, wine, spirits

monthly sales of beer, wine, spirits and total ethanol

Box-Jenkins model for Saskatchewan and New Brunswick

17 OECD nations: 1970–1983; bans of broadcast ads 17 OECD nations: 1970–1983; bans of broadcast ads 5 OECD nations: 1970–1990; expend. on advertising; Sweden as control 17 OECD nations: 1977–1995; bans of broadcast ads

log ethanol per capita

binary var. bans of spirit ads; bans of all beverages binary var. bans of broadcast ads; bans of all beverages advertising expend. for France, Germany, Netherlands, and U.K. Sweden bans all ads binary var. bans of spirit ads; bans of all beverages

Ogborne and Smart (1980) Makowsky and Whitehead (1991)

Saffer (1991)

Young (1993)

Calfee and Scheraga (1994)

Nelson and Young (2001)

log ethanol per capita log ethanol per capita

log ethanol per capita

Coefﬁcient Estimates (std. error) Mann-Whitney U-test; T-test (not signiﬁcant) no signiﬁcant decrease beer increase; spirits decrease no impact for wine or total ethanol –0.169 (0.04)* –0.313 (0.09)*

R-Sq n.a.

n.a. n.a.

Alcohol Advertising and Advertising Bans

Table 2. Summary of Cross-National Studies of Broadcast Advertising Bans.

0.590

0.013 (0.03) 0.035 (0.03)

0.990

–0.090 (0.04)** 0.018 (0.09) 0.079 (0.11) –0.024 (0.06) 0.139 (0.02)* 0.049 (0.03)

0.980 0.880 0.970 0.840 0.658

261

Notes: * indicates signiﬁcant at 1% level or better; and ** for 5% or better. Page numbers for coefﬁcient estimates are Smart and Cutler (1976, pp. 17, 20); Ogborne and Smart (1980, p. 294); Makowsky and Whitehead (1991, p. 562); Saffer (1991, p. 75); Young (1993, p. 226); Calfee and Scheraga (1994, p. 300); and Nelson and Young (2001, p. 286).

262

JON P. NELSON

from zero-to-two media (“weak” bans); three-to-four media (“limited” bans); and ﬁve-to-seven media (“comprehensive” bans). The authors treated the “weak ban” category as the control group, which implies that bans of one or two media did not substantially affect tobacco use compared to zero bans. In this data set, the number of bans in many countries increased over time, so the authors were able to study both cross-national and time-series variation of advertising restrictions. They report that “. . . the Limited Ban variables are generally not signiﬁcant, while the Comprehensive Ban coefﬁcients are almost all signiﬁcant” (Saffer & Chaloupka, 2000, p. 1130). Further, they ﬁnd that “. . . a limited set of advertising bans will not reduce the total level of advertising expenditures, but will simply result in substitution to the remaining non-banned media” (Saffer & Chaloupka, 2000, p. 1134). Finally, the authors conclude that “. . . the analysis in this paper suggests that the new ban on outdoor advertising, required by the 1999 U.S. tobacco settlement, will have little effect on consumption . . . [and] will result in substitution to the remaining three forms of [non-banned] advertising and to increased use of tobacco promotion” (Saffer & Chaloupka, 2000, p. 1135). C. Why Don’t Alcohol Advertising Bans Work? The fourteen studies reviewed above do not support a statistically signiﬁcant or material effect of alcohol advertising bans, including selective bans of outdoor media and comprehensive bans of broadcast media and other print media. Except for Saffer’s (1991) study, the evidence is neither mixed nor inconclusive. There is evidence that some bans result in substitution between beverages or media, but there is no convincing evidence that selective or more comprehensive bans have signiﬁcant effects on marketwide demand for all alcohol beverages. Additionally, some bans are designed to affect competitive conditions in the market, and favor one beverage over others as a form of rentseeking behavior (Calfee, 1997; McGahan, 1995; Peltzman, 1971; Yandle, 1983). At the beverage level, the results are mixed and suggest that substitution among beverages is a possible effect of a ban. However, the null hypothesis that advertising bans reduce alcohol consumption (or abuse) must be rejected. Conceivably, a total ban of all alcohol advertising and promotion might have some effect on behavior, although anecdotal evidence for the Soviet Union and the U.S. Prohibition-era argue against even this extreme result. However, two additional questions remain. First, if advertising bans do not affect alcohol demand, then what economic and social processes are possibly affected by such restrictions? I offer four responses to this question. Second, can the results in this section be applied to youth alcohol behaviors? None of

Alcohol Advertising and Advertising Bans

263

the studies examined above is speciﬁcally concerned with youth alcohol consumption, although a large econometric literature exists on this general issue (Chaloupka et al., 2001; Cook & Moore, 2000; Grossman et al., 1994; Kenkel, 1993). I also discuss why the empirical studies in Tables 1 and 2 are relevant for policy analysis of youth alcohol consumption problems. There are four reasons why advertising bans might not work in the ways imagined by policymakers. First, given a selective ban, substitution toward non-banned media is always possible. This issue has not been investigated thoroughly, although some evidence is available from expenditure studies. Also, a non-alcohol investigation by Seldon and Jung (1993) obtains empirical estimates of substitution elasticities among messages in broadcast, print, direct mail, and other media. Broadcast advertising is the most effective of the four media that they examine, yet bans of this media have not been shown to reduce alcohol consumption. Second, a ban of one media can stimulate innovations within the set of non-banned media and other means of promotion. Harrison and Godfrey (1989) discuss the difﬁculty of alcohol advertising controls in Europe given new media such as broadband cable, videotex, and satellite television. Third, advertising can have both a price effect and an output effect. Advertising that increases product differentiation can reduce the price elasticity of demand or shift consumer expenditures toward higher-priced beverages (Ambler, 1996; Kaul & Wittick, 1995). Some evidence regarding this effect is offered in Section V. More generally, Motta (1997) develops a theoretical model of an advertising ban under conditions of monopoly and oligopoly. His basic point is that a ban of persuasive advertising could either increase or decrease consumption, since advertising affects both the level of demand at given prices and the level of prices that sellers ﬁnd optimal. Hence, a ban of persuasive advertising could reduce prices enough that alcohol consumption increases, rather than decreases. This result is counterintuitive, but consistent with ﬁndings in some studies. Similar possibilities can be found in theoretical models due to Dixit and Norman (1978) and Milgrom and Roberts (1986). Fourth, as recognized by the Supreme Court, advertising in mature markets can affect only brand (or beverage) shares, and there is no important spillover to the market as a whole. Numerous studies of advertising expenditures support this conclusion. For example, a meta-analysis of 128 marketing studies of frequently-purchased consumer products found that “. . . advertising has a relatively minor impact on product class sales” (Assmus et al., 1984, p. 72). The second important question is the possible effect of an advertising ban on youth alcohol behaviors. Does advertising cause or predispose youth to drink alcohol? No empirical studies have addressed this issue directly for billboard bans or broadcast bans. However, there are at least four reasons why the results

264

JON P. NELSON

in Tables 1 and 2 are applicable to youth. First, “adult” alcohol consumption and measures of youth alcohol consumption are highly correlated. For example, both adult consumption and youth bingeing have declined by more than 20% from the peak levels that occurred in 1981. Using time-series data for the period 1975–1997, I regressed per-capita ethanol consumption on data for binge drinking by 12th-graders (measured as a percentage of responses) from the Monitoring the Future surveys. The R2 for the regression was 0.884, and the residual never exceeded about 2% in absolute value (compared to a sample mean of 35%).6 The close relationship between these measures also has been employed in a comprehensive model of youth drinking by Cook and Moore (2001), which is discussed along with survey-research studies in Section IV. Second, studies by social psychologists using survey-research data and “path analysis” show a signiﬁcant and direct effect of parents’ drinking (and parents’ approval) on youth alcohol behaviors. In contrast, measures of alcohol advertising in these studies always have small indirect effects, and the results are not always statistically signiﬁcant. This is consistent with the ﬁndings from studies of “adult” consumption and advertising bans and expenditures. Third, alcohol advertising bans are often longstanding as many of these restrictions date to the repeal of Prohibition. If advertising has a substantial inﬂuence on youth behaviors (as frequently claimed), then it is reasonable to expect that current “adult” behavior will reveal this inﬂuence through a negative effect of a ban on consumption. However, this result is not found across a wide variety of bans, time periods, and countries. Fourth, numerous econometric studies include variables for youth demographics and the minimum legal drinking age. These variables are almost always signiﬁcant as determinates of per-capita alcohol demand, especially beer consumption. However, advertising bans have not been found to materially decrease per-capita beer consumption. This difference is striking, and cannot be explained away by the level of data aggregation.

IV. STUDIES OF ALCOHOL ADVERTISING EXPENDITURES Virtually all econometric studies of alcohol advertising expenditures come to the conclusion that advertising has little or no effect on marketwide (or beverage) alcohol demand. Using coefﬁcient point estimates, some studies using annual data obtain small advertising elasticities of about 0.1 or less. An advertising elasticity of 0.1 implies that a 50% increase (decrease) in advertising would increase (decrease) alcohol consumption by only 5%. It is doubtful that an effect of this magnitude would pass the third prong of the

Alcohol Advertising and Advertising Bans

265

Central Hudson test. In 1998, the shares by media for alcohol advertising were broadcast advertising, 66%; print advertising, 26%; and outdoor advertising, 8%. The implication is that virtually a total ban of broadcast advertising would be required to reduce alcohol consumption by 5%, but international studies of broadcast bans fail to demonstrate that even this large change would have an effect on consumption. None of the studies reviewed in this section provides support for the null hypothesis that advertising expenditures increase marketwide alcohol consumption in a material way. The literature surveys by Saffer (1993, 1995, 1996, 1998) are highly critical of this result. He argued that “. . . studies that use national data on annual alcohol advertising expenditures measure advertising at a high level [of aggregation] with little yearly change and are likely to ﬁnd no effect on consumption” (Saffer, 1996, p. 266; emphasis added). This argument is based on the existence of a market- or industry-wide sales response function, which is ﬂat in the region of observed levels of annual advertising (Saffer, 1995, p. 85; 1996, p. 268; 1997, p. 431; 1998, p. 783). I discuss the existence of an industryresponse function in Section V. In addition to studies that examine advertising bans, Saffer argued that data sets with greater variation are required, such as higher-frequency monthly and quarterly data. He also discussed a number of potential econometric problems with time-series studies, including simultaneous equation bias; cumulative effects of advertising; temporal aggregation bias; and measurement of real advertising expenditures (Saffer, 1993). Recent studies of expenditures have dealt with all of these issues, but advertising has not been shown in these studies to signiﬁcantly affect marketwide alcohol consumption. This section summarizes recent studies, with a focus on methodological issues and econometric procedures used in time-series expenditure studies.7 A. Econometric Studies of Advertising Expenditures (1) Studies Using High-Frequency Quarterly and Monthly Data. A number of recent studies use monthly or quarterly data on advertising expenditures and alcohol consumption. These data capture seasonal or other data variations (such as “ﬂighting” or “pulsing” strategies by advertisers), which might conceivably inﬂuence consumption. U.S. studies using quarterly data are Coulson et al. (2001), Franke and Wilcox (1987), and Nelson (1997, 1999); U.K. studies using quarterly data include Duffy (1990, 1991a, b, 1995, 2001); and Canadian and Australian studies using monthly data are Lariviere et al. (2000) and Smith (1990). The Canadian study is a thorough treatment of advertising dynamics within a demand

266

JON P. NELSON

system for alcoholic beverages and soft drinks. The researchers found only weak advertising effects on aggregate alcohol expenditures, and they concluded that “. . . advertising is not effective in enlarging markets” (Lariviere et al., 2000, p. 147). (2) Studies Using Simultaneous-Equation Methods. A number of studies of alcohol advertising use single-equation methods for one or more beverages, including studies by Goel and Morey (1995), Lee and Tremblay (1992), and Tegene (1990). However, other recent studies use multiequation methods, and either provide coefﬁcient estimates that account for simultaneity or include tests for endogeneity. U.S. studies that estimate structural models using system-wide methods are Nelson (1997, 1999) and Nelson and Moran (1995); U.K. studies include Duffy (1990, 1991a, b, 1995, 2001); and a Canadian study is Lariviere et al. (2000). Cointegration analysis is another econometric method that accounts for simultaneousequation bias (Stock & Watson, 1988, p. 166). Studies that apply cointegration methods include Blake and Nied (1997) and Duffy (1991a) for the U.K., and Coulson et al. (2001) for the U.S. Additional studies that test for endogeneity include Duffy (1991a, b, 1995, 2001), Lee and Tremblay (1992), Nelson (1999), and Nelson and Moran (1995). A study by Duffy (2001) uses quarterly data and conducts a thorough investigation of functional forms for demand systems. The Hausman-Wu test is applied to the hypothesis that advertising is orthogonal to the equation disturbances, and the study found strong support for exogeneity of beer, wine, and spirits advertising (Duffy, 2001, p. 445). The researcher concluded that “. . . advertising is found to have no signiﬁcant effect upon the ‘product composition’ or ‘level’ of total alcoholic drink composition in the U.K.” (Duffy, 2001, p. 437). (3) Studies that Account for Advertising Dynamics. A frequent claim is that advertising has a cumulative or “lingering” effect on demand. Hence, lagged values of advertising expenditures are possible determinates of consumption. However, economic and marketing studies of the depreciation rate on past advertising usually ﬁnd that expenditures are almost fully depreciated in less than a year’s time. Clarke (1976) surveyed the literature on this issue for a broad sample of consumer products. He reported that “. . . 90% of the cumulative effect of advertising on sales of mature, frequently purchased, low-priced products occurs within 3 to 9 months of the advertisement. The conclusion that advertising’s effect on sales lasts for months rather than years is strongly supported” (Clarke, 1976, p. 355). Additional surveys by Assmus et al. (1984) and Leone (1995) reach the same general conclusion (see also Boyd & Seldon, 1990). Given a high rate

Alcohol Advertising and Advertising Bans

267

of depreciation, it would be surprising if lagged advertising has a substantial effect on alcohol consumption. Studies that investigate advertising dynamics using quarterly and monthly data include Coulson et al. (2001), Duffy (1991a), and Lariviere et al. (2000). None of these studies found substantial effects of lagged (or current) advertising on alcohol demand. (4) Temporal Aggregation Bias. Theoretical work by Bass and Leone (1983) showed that the magnitude of the advertising coefﬁcient should increase with the length of the data interval employed in the analysis. In order to examine this result, they estimated a distributed lag model using 16 years of bimonthly advertising data for ﬁve brands of a frequently purchased and heavily-advertised product category. The model was estimated using data aggregates for two months, four months, six months, and twelve months. The general result was that the estimate of the advertising coefﬁcient increased gradually in size as the data interval was lengthened, and the overall increase was substantial (Bass & Leone, 1983, p. 6). Comparing quarterly with annual estimates for the ﬁve brands, the average increase in the ﬁve coefﬁcients was about 83% (from 1.41 to 2.58). Leone (1995) surveyed other theoretical models of temporal aggregation and concluded that regardless of the model “. . . it is clear that . . . the advertising coefﬁcient should increase as the level of aggregation increases” (Leone, 1995, p. 149). This point was overlooked in the survey of alcohol advertising by Fisher and Cook (1995, p. 146), who concluded that the advertising elasticity was about 0.10. In fact, studies using quarterly data obtain considerably smaller point estimates of this elasticity. Duffy (1990, p. 9) used quarterly U.K. data for 1963–1983, and obtained advertising elasticities for total alcohol demand in the range from 0.020 to 0.039. Nelson (1999, p. 786) used quarterly U.S. data for 1977–1994, and obtained point estimates for total alcohol demand of about 0.001 or less. In other words, a 50% reduction in U.S. advertising would reduce per-capita alcohol consumption by only 0.05%. This is much smaller than the 5% reduction obtained from estimates based on annual data. Moreover, the advertising coefﬁcient estimates in these studies are not statistically different from zero. It seems quite clear that elasticities of this size would not pass the Central Hudson test. (5) Advertising Measurement Issues. The issue of advertising measurement was ﬁrst raised by Backman (1967). In path-breaking work, Schmalensee (1972, p. 245) showed that consistent aggregation of expenditures could be obtained by deﬂating by media-speciﬁc price indexes, where each price index accounts for a medium’s audience exposure and coverage. This effect

268

JON P. NELSON

is referred to as the medium’s “reach,” which is the percentage of the potential audience reached by an advertising campaign. The price indexes are expressed in cost per thousand (CPM), which measures the cost of reaching 1000 potential customers. Hence, deﬂated expenditures are measures of the number of real message “impressions” due to each medium, and the advertising coefﬁcient in a demand function is an estimate of the marginal “response” or marginal productivity of messages on quantity of sales. Deﬂating advertising expenditures by broader price indexes (such as the implicit deﬂator for GDP) is generally satisfactory for ease of replication, but a better analytical procedure is to use the McCannErickson CPM indexes or comparable price indexes. Studies for the U.S. that use the McCann-Erickson indexes include Coulson et al. (2001), Franke and Wilcox (1987), Gius (1991, 1996), Lee and Tremblay (1992), Nelson (1999), and Nelson and Moran (1995). In Nelson (1999), real advertising expenditures also are disaggregated by media in order to explore possible differential effects on alcohol demand of broadcast, print, and outdoor media. The results for the three beverages and total alcohol indicate that, regardless of the medium, advertising has little or no effect on demand. (6) Studies that Adjust for Serial Correlation and Data Stationarity. Serial correlation and stationarity are relevant concerns for studies that use timeseries data. Studies that include explicit adjustments for serial correlation include Calfee and Scheraga (1994), Duffy (1991a), Goel and Morey (1995), Lee and Tremblay (1992), and Nelson and Young (2001). Studies that adjust for serial correlation using ﬁrst-order (or higher) differences of the data include Duffy (1990, 1991a, b, 1995, 2001), and Nelson and Moran (1995). In Nelson (1997, 1999) the data are fourth-differenced to remove pure seasonal variation, and the intercept term captures any remaining exogenous trend in the transformed data. Using more recent techniques, studies that investigate stationarity of the data include Blake and Nied (1997), Coulson et al. (2001), Duffy (1991a), and Lariviere et al. (2000). None of these studies ﬁnds that advertising has a signiﬁcant or material effect on alcohol consumption. The study by Lee and Tremblay (1992) is a thorough investigation of the demand for beer. The model included the real price of beer, real price of whiskey, real price of coladrinks, real disposable income, real advertising, real stock of advertising, lagged consumption, and youth demographics. The advertising elasticities were never statistically signiﬁcant, and the point estimates were very small (0.006 to 0.019). The empirically important determinates of beer demand included the price of beer, prices of substitutes, and youth demographics.

Alcohol Advertising and Advertising Bans

269

With respect to beer advertising, the study concluded that “. . . there is no support for the hypothesis that advertising has a signiﬁcant positive effect on market demand” (Lee & Tremblay, 1992, p. 74). In summary, despite numerous advances in data and econometric techniques, there is little or no evidence of a marketwide relationship between alcohol advertising and alcohol consumption. The whole notion of an industry-wide response function must be called into question. Only small effects at the beverage level have been reported, but these effects always disappear when consumption is aggregated across beverages. Further, I have argued above that these studies and their ﬁndings are applicable to issues of youth alcohol behaviors. I next turn to an alternative approach to the issue of youth drinking and advertising. B. Survey-Research Studies: Path Analysis of Youth Alcohol Behaviors Survey research is designed to gather self-reports of drinking behaviors and of exposure and responses to advertising. Path analysis is a technique for estimating structural (causal) models that include hypothetical latent (unobservable) variables, and which encompasses and extends regression, econometric, and factor analysis procedures (Bollen, 1989). The observable determinates of a latent variable are referred to as indicator variables, which contain errors of measurement. A path diagram is a pictorial representation of a system of simultaneous equations. A particular feature of path analysis is the statistical isolation of direct effects and indirect (mediated) effects. For example, advertising has a possible direct effect on beer consumption, and possible direct and indirect effects on drunk driving.8 The indirect effects of advertising are mediated through latent variables (for example, risk perception), and the total effect is the sum of direct and indirect effects (Bollen, 1989, pp. 4, 36, 376). Single-equation regression models estimate the direct effects of observable variables, and might understate overall effects if indirect effects are substantial (Bollen, 1989, p. 38). Structural econometric models and reducedform models typically deal with only observable (manifest) variables, although direct, indirect, and total effects are clearly possible within these models (see Kinnucan et al., 2001; Nelson, 1997; Nelson & Young, 2001). Modern path analysis models are estimated using maximum likelihood procedures or generalized least-squares, and are therefore subject to the full range of econometric problems, including speciﬁcation bias, measurement errors, sample selection bias, missing data, outliers, multicollinearity, lack of replication, and the like (Bollen, 1989, pp. 55–61).9

270

JON P. NELSON

There are numerous studies of youth alcohol behaviors that apply path analysis or related techniques such as conﬁrmatory factor analysis and hierarchical regression analysis. No attempt is made to survey all of this large and growing ﬁeld of inquiry.10 Rather, I examine only three representative studies in this area, and critically discuss the authors’ methods, results, and policy implications. These studies attempt to measure the indirect effect of television advertising on youth alcohol behaviors. None of the studies establishes a statistically signiﬁcant direct effect of advertising on youth drinking or, more important, has identiﬁed the magnitude of the indirect effect of advertising. Methodologically, this omission is surprising, since the measurement of indirect effects and their standard errors is well-known in the literature on structural modeling.11 As a consequence, the application of path analysis to the debate on alcohol advertising is seriously incomplete and presently of limited usefulness for public policy evaluation. (1) Grube and Wallack (1994) studied the effects of television exposure in a cross-section sample of 468 ﬁfth- and sixth-graders (10–13 years old). The children were asked to state their intention to drink as an adult for each of four beverages – beer, wine, wine coolers, and liquor – on a 4-point scale from “not at all” to “at least once per week.” However, only 5% (or 23) of the children had ever had more than a taste of an alcoholic beverage (Grube & Wallack, 1994, p. 255). The statistical implications of this sample feature are not discussed. The endogenous variable – “drinking intention as an adult” – is a latent variable, which is constructed by combining the four responses by beverage (as factor loading). The authors fail to report the mean value for the survey responses for each beverage, the mean of the latent variable (or its distribution), or other summary information about their sample and variables. In order to measure awareness of advertising, the students were shown ﬁve still photographs from currently televised beer commercials (with brand names removed), and they were asked to name the brand being advertised. They also were asked to recall beer brands and to match brands and slogans. This approach is used despite the fact that the intentions variable covers all forms of alcohol, and is not a brand-speciﬁc measure. Awareness of beer advertising, beliefs about positive aspects of drinking (sociability), and beliefs about the negative aspects of drinking (health problems) also are latent variables, which are inﬂuenced by multiple indicators in the authors’ measurement model (Grube & Wallack, 1994, p. 256). For example, “awareness of advertisements” depends signiﬁcantly on ﬁve indicators: knowledge of slogans; knowledge of brands; respondents’ exposure to television sports; weekend

Alcohol Advertising and Advertising Bans

271

PM television watching; and parents’ drinking (Grube & Wallack, 1994, p. 257). The number of television and other advertising messages viewed is not measured. Other statistically signiﬁcant exogenous variables included in the structural model are gender, peer approval, peer drinking, parents’ education, parents’ approval, and parents’ drinking. The statistical results yield coefﬁcient estimates that can be used to obtain both direct and indirect effects for relevant variables, but the authors fail to provide the latter estimates or their standard errors. For example, the standardized path coefﬁcients for the direct effects on “intentions” for gender, peer approval, parents’ education, and parents’ drinking are 0.14, 0.26, 0.11, and 0.28, respectively (Grube & Wallack, 1994, p. 257). A surprising result is that peer drinking only indirectly affects “intentions” through the knowledge of brands. Parents’ approval and parents’ drinking also have indirect effects on “intentions” through positive beliefs and advertising awareness, respectively. “Positive beliefs” is a latent variable with a structural coefﬁcient of 0.32, which can be decomposed into effects due to parents’ approval and advertising awareness. The coefﬁcients are 0.22 and 0.47, respectively. Hence, the indirect path effect of “advertising awareness” (through positive beliefs) on drinking intentions is (0.47)(0.32) = 0.15, which is smaller than the direct effects of peer approval or parents’ drinking by a factor of two. Similarly, the mediated or indirect effect of TV sports on drinking intentions is given by (0.23)(0.47)(0.32) = 0.03, and that for weekend PM TV watching is (0.24)(0.47)(0.32) = 0.03, where 0.23 and 0.24 are the structural coefﬁcients. For brand awareness, the indirect effect is (0.38)(0.47)(0.32) = 0.06, and that for slogans is given by (0.17)(0.47)(0.32) = 0.03. While television exposure would be affected by a broadcast advertising ban, the indirect effects of this medium are smaller by a factor of nine compared to the direct effects of peer approval or parents’ drinking. These results cannot be regarded as evidence of a material effect of advertising on youth drinking intentions. Furthermore, Grube and Wallack (1994) do not measure advertising exposure for media other than television. They do not report diagnostics for data outliers, tests for collinearity, or summary measures of goodness of ﬁt. A large number of relevant variables are omitted, including parents’ income, occupation, religion, and the like. Finally, they fail to report the indirect coefﬁcients and their standard errors. (2) In a longitudinal study, Connolly et al. (1994) investigated alcohol consumption at age 18 years and prior advertising exposure. The sample covered 667 teenagers in New Zealand, who were interviewed at ages 13, 15, and 18. Participants in the study underwent a full day’s assessment,

272

JON P. NELSON

including personal recalls of alcohol-related advertising exposure (print, broadcast, entertainment images) collected in face-to-face interviews at ages 13 and 15. For self-reported alcohol use at age 18, the respondents completed a computer questionnaire (Connolly et al., 1994, p. 1257). Standard regression analysis is used to investigate the relationship between alcohol consumption and earlier recall of alcohol seen in the mass media. Alcohol consumption is measured by average quantity, maximum quantity, and frequency of drinking at different locations, summed over drinking locations (Connolly et al., 1994, p. 1258). The non-media variables include gender, socioeconomic status, living situation, respondents’ occupation, and peer approval of drinking. This small number of explanatory variables is considered despite the fact that the same individuals had been followed in the study since birth (Connolly et al., 1994, p. 1257). No explanatory variables are included for market conditions (e.g. prices), parents’ drinking, peers’ drinking, religion, respondents’ or parents’ income, and the like. Despite the use of standard regression analysis, the authors do not report regression coefﬁcients (only p-values for Student t-tests are reported). Hence, it is impossible to judge the magnitude of the reported effects. No relationship whatsoever is found between broadcast advertising (recalls at ages 13 and 15) for wine and spirits consumption by women or men. For women, only the number of hours of TV watched at age 13 (but not at age 15) has a weakly positive effect on the amount of beer, wine, and spirits consumption. For men, there are no signiﬁcant relationships of any kind for wine and spirits. For men, the number of broadcast messages recalled at age 15 (but not at age 13) has a weakly positive effect on the average and maximum amount of beer consumption. However, for women, the number of alcohol promotions recalled at age 13 (but not at age 15) is negatively associated with the frequency of beer consumption (Connolly et al., 1994, pp. 1260, 1262). This is the wrong coefﬁcient sign. In summary, this study is deﬁcient as a basis for informed public policy, despite the authors’ contrary claim (Connolly et al., 1994, p. 1262). First, the magnitude of advertising effects cannot be judged, which is a common problem in many social learning studies. Second, the regression relationships ﬁt the data very poorly (R2 values are all less than 0.17), and only about 10–17% of the total variance in the dependent variables is actually explained. This suggests that the results are subject to serious omitted variable bias. Third, the authors report 66 regression coefﬁcients for advertising, and only ﬁve coefﬁcient estimates have p-values less than 0.05, and one of these has the wrong sign. In any study with a large number of

Alcohol Advertising and Advertising Bans

273

estimates for the same inﬂuence, random variation (and pre-testing bias) can easily result in three or four signiﬁcant coefﬁcients. (3) Austin and Knaus (2000) surveyed 273 children (third-, sixth-, and ninthgraders) in Washington state. The survey was a pencil-and-paper questionnaire designed to assess media (television) effects on “precursors to drinking” among children for whom alcohol consumption was not yet occurring (Austin & Knaus, 2000, p. 13). The authors hypothesize that children’s attitudinal expectancies (drinking makes you happy) will be positively predicted by attitudinal responses to questions on alcohol desirability (women in beer ads are good looking) and identiﬁcation (I wish I could be like TV people). All three measures are correlated with each other (correlations range from 0.35 to 0.48), which is a possible difﬁculty with this study. As an additional step, they measure “alcohol predrinking behavior” by third- and sixth-graders using children’s preferences for toys and other items with beer brand logos (e.g. beach towels, hats). A positive value was recorded if the respondent chose a beer-logo item and a negative value for a soda pop-logo item. The mean for this variable is –2.85, indicating that most children picked the soda items (Austin & Knaus, 2000, p. 18). Sixth- and ninth-graders also were asked to self-report risky behavior, including frequency of use of tobacco and alcohol (Austin & Knaus, 2000, p. 18). Frequency of use is reported on a six-point scale from never used (1) to use about every day (6), and the mean alcohol value is 1.58. This indicates that most respondents had infrequently used alcohol, and suggests that the risky behavior may include outliers or is a discrete variable that would be better examined in an ordered probit or logit model. Control variables in the analysis include gender, grade level, ethnic background, and parents’ education. There are no observable measures of advertising exposure in this study, such as hours watched of TV or recall of slogans and brands. This hinders replication of results by other researchers and cross-study comparisons. The authors hypothesize that desirability should predict identiﬁcation, which should predict expectancies, which should predict both risky and predrinking behaviors. Estimates of the standardized coefﬁcients are 0.22, 0.18, 0.43, and 0.02, respectively (Austin & Knaus, 2000, p. 20). The last coefﬁcient is not statistically signiﬁcant, and only grade-level predicts predrinking behavior. Desirability also predicts expectancies, with a coefﬁcient of 0.31. Given the path structure of the model, the indirect effect of desirability on risky behavior is given by (0.22)(0.18)(0.43) + (0.31)(0.43) = 0.15. In contrast, the sum of direct and indirect effects of grade-level are 0.39 + (0.27)(0.18)(0.43) + (0.43)(0.43) = 0.60. Hence, exogenous variables with direct effects on

274

JON P. NELSON

behavior are far more important than the indirect effects of advertisingrelated variables, which repeats the results in Grube and Wallack (1994). Given the limited number of control variables in this study, the results also are suspect due to omitted variable bias and other econometric problems that largely go undiscussed by the authors. The authors’ hypothesis regarding predrinking behavior is not supported, although they largely ignore this negative result. More important, since there are no measures of advertising exposure, the study cannot be used to determine the effects of removing speciﬁc types of advertising on youth behaviors. Finally, the authors do not report a coefﬁcient that measures the magnitude of the relationship between risky behavior and predrinking behavior for the sixthgrade students. The basis for their strong policy conclusion is a simple statistical comparison of mean differences by grade-level, and a zero-order correlation of 0.20 (Austin & Knaus, 2000, pp. 20, 22). Path analysis models have not been utilized in a manner that is suitable for the questions that must be addressed by the courts (and must be considered by governments) under the Central Hudson formula. The empirical results in path analysis studies are biased and incomplete. None of the studies establishes the magnitude of the (indirect) effect of advertising messages on a well-deﬁned measure of alcohol consumption. Stated in broad terms, the models and results in path analysis studies can be likened to a coin-ﬂipping experiment. There are four or ﬁve possible steps to the actual youth behavior of interest, including advertising exposure, desirability, advertising awareness or identiﬁcation, positive beliefs, and, lastly, actual drinking behavior or expectancies. The resulting outcome is analogous to the probability of ﬂipping a fair coin, and getting a “heads” four times in a row. The joint probability of this event is (0.5)4 = 0.0625. It is a serious question whether a probability this small can be used to support the third prong of the Central Hudson test. Furthermore, a comparison of the estimates for advertising and parents’ behavior suggests that more cost-effective solutions exist under the fourth prong. In contrast to these studies, a comprehensive econometric study of youthful drinking is found in Cook and Moore (2001). Unfortunately, there are no advertising variables in this study. The authors use data from the National Longitudinal Survey of Youth (NLSY) for 1982–1985 and 1988–1989 for youth aged 14 to 22. The sample size is about 50,000. They note that youthful drinking has declined over time, and that youth drinking in the U.S. is below that found in many other countries (Cook & Moore, 2001, p. 377; see also Smart & Ogborne, 2000b). They examine two binary dependent variables for alcohol consumption, including consumption in the past 30 days and binge

Alcohol Advertising and Advertising Bans

275

drinking (6 + drinks on four or more occasions) in the past 30 days. Probit regressions are estimated for these two variables, including predetermined variables for sex, race, ethnicity, age, cognitive ability, parents’ education, family composition (e.g. living with one parent), location of residence, and religion (Cook & Moore, 2001, p. 387). Many of these variables are not included in survey-research studies. In addition, the authors include a number of variables that might be endogenous, including parents’ drinking, years of school completed, living arrangements, marital status, number of children in household, school and employment status, the respondent’s weight, income, and educational aspiration. The alcohol control variables in Cook and Moore (2001, p. 388) are the beer excise tax and (lagged) minimum legal age. Further, per-capita consumption of ethanol by adults is introduced as a control for all other state-level factors that might inﬂuence drinking, including alcohol availability, outlet licensing, liability rules, social contagion, and the local drinking culture. The results show that the minimum legal age and the beer excise tax are effective deterrents to youthful drinking and bingeing, and “adult” consumption has an important and consistent effect on youth behaviors. The last result is consistent with results in path analysis studies, but Cook and Moore’s results also illustrate the incompleteness of the model speciﬁcations in survey-research studies. Their results also demonstrate that studies of “adult” or aggregate alcohol consumption are applicable to issues of youth alcohol behaviors. The researchers conclude that “. . . youthful drinking decisions are closely linked to overall consumption” (Cook & Moore, 2001, p. 421).

V. BRAND ADVERTISING AND THE CHIMERA OF AN INDUSTRY-RESPONSE FUNCTION Saffer (1995, 1996, 1997, 1998) argued that the industry-wide response function is ﬂat in the region measured by annual data on expenditures. This claim also has appeared in the policy literature (American Medical Association, 1986, p. 1487). As I have demonstrated above, empirical studies using higherfrequency data fail to show that alcohol advertising has an effect on marketwide sales. Furthermore, studies of advertising bans fail to demonstrate that billboard and broadcast bans reduce total alcohol consumption. I now turn to the evidence on the relationship between brand (or category) advertising and beverage consumption. Three examples of this relationship are developed: light beer and total beer demand; wine coolers and total wine demand; and vodka and total spirits demand. Each of these examples demonstrates the difﬁculty of making marketwide inferences based on the marketing success of a particular brand or beverage. Successful brand promotion produces increased sales of the

276

JON P. NELSON

brand, but it does not follow that this success spills over to the market (or category) as a whole. An obvious reason for this failure is that the primary effect of advertising is to redistribute brand shares. Hence, a movement along the advertising-sales function for Brand A causes a shift in the advertising-sales function for other brands, including brands produced by the ﬁrm selling Brand A. Marketwide sales can increase or decrease depending on the net effect. I argue below that examining market segments with successful brand advertising campaigns reveals information about these net effects and the possible shape or existence of an industry advertising-sales response function. Current notions about the shape of the industry-response function depend on studies of brand advertising and market shares of leading brands and ﬁrms. Simon (1970, 1980) provided thorough surveys of the early marketing literature on this topic. He was primarily interested in the possibility that there might be increasing returns to advertising over some range of experience. He concluded that marginal returns to advertising diminish throughout, but all of the data that Simon examined refer to speciﬁc brands. I am aware of only one study of brand advertising for alcoholic beverages that provides some direct evidence on the existence of an industry-response function for alcoholic beverages. Gius (1996) examined brand-level sales and advertising of distilled spirits. In his model, brand advertising has an own-brand effect (due to reduced sales of rival brands) and a spillover or marketwide effect on sales. Gius argued that the rival-brand advertising effect in the own-brand’s demand function will be positive if the spillover effect is greater than the decrease of own-sales caused by rival-brand advertising (Gius, 1996, p. 73). As an empirical test, Gius studied a panel of sixteen leading liquor brands for the period 1976–1989. Advertising was measured by deﬂated (real) national brand-level messages obtained from Leading National Advertisers. Other variables included price, rival price, rival advertising, income, and a time trend. The regression results indicated that own-brand advertising had a signiﬁcant positive effect on ownbrand demand, but rival-brand advertising was insigniﬁcantly different from zero. Hence, he concluded that “. . . brand-level spirits advertising results in only brand switching and does not increase the size of the spirits market” (Gius, 1996, p. 75). As a general model of this process, consider advertising under oligopoly conditions where sellers conform to Cournot-Nash assumptions, expecting rivals’ advertising outlays to remain unchanged in response to their outlays (Scherer & Ross, 1990, p. 594). Brand A’s advertising elasticity can be decomposed into three terms: (1) the elasticity of Brand A’s output with respect to the spillover or market-expansion effect of Brand A’s advertising; (2) the elasticity of Brand A’s output with respect changes in its own market share or

Alcohol Advertising and Advertising Bans

277

the rivalrous effect of Brand A’s advertising; and (3) the negative elasticity of Brand A’s output with respect to rivals’ advertising induced by changes in Brand A’s advertising. Successful brands are those for which the ﬁrst two terms more than offset the negative effect of the third term. These brands should have both increasing market shares and increasing quantities of sales. Hence, studying brands and categories that exhibit rising sales due to successful advertising reveals information about the spillover effect of Brand A’s advertising on marketwide demand. In this section, I examine three examples of successful brand advertising of alcoholic beverages – Miller Lite beer, Bartles & Jaymes wine cooler, and Absolut vodka. There is no convincing evidence of an industry-wide spillover effect associated with these successful brands. For light beer, the product category expands, but the beer market does not. For wine coolers, the category and the market expand, but this success is temporary. For vodka, both the vodka category and the spirits market decline steadily over time. These examples fail to support the notion of a long-run marketwide effect of successful brand-level advertising.12 Brand advertising affects brand shares in the alcoholic beverage industry, but the examples fail to demonstrate a marketwide spillover. This is consistent with studies of advertising expenditures at various levels of aggregation. A. Light Beer and the Market for Beer Light beer (“Lite”) was introduced to the national mass market by Miller Brewing in 1975. Under the direction of new management by its parent Philip Morris, Miller set out to segment the beer market by target marketing to consumer groups (Scherer & Ross, 1990, p. 583). Advertising for Miller Lite stressed its low calorie content, and the message “Taste’s Great, Less Filling” pushed by aging ex-athletes was an attraction for the weight-conscious consumer. Anheuser-Busch responded in 1977 to Lite’s success with its Natural Light brand, followed in 1978 and 1981 by Michelob Light and Bud Light brands (message: “Gimme a Light”). These brands were backed by substantial advertising campaigns (Greer, 1993, p. 107). The battle of the brands had been joined, and advertising budgets exploded to reﬂect this.13 Miller spent about $14.8 million on Lite promotion in 1976, principally on television. By 1984, Miller’s spending had grown to $67 million. AnheuserBusch (A-B) spent $10.7 million introducing Natural Light in 1977, with about 94% occurring on television and radio. A-B’s spending on Bud Light and Michelob Light was $79 million in 1984. This was more than $10 per barrel for each brand, compared to $1.60 for A-B’s Budweiser brand and $3.87 per barrel by Miller Lite. Nominal advertising expenditures for Lite peaked in 1986 at

278

JON P. NELSON

$86 million ($4.53 per barrel). Nominal advertising of Bud Light peaked in 1988 at $70 million ($7.32 per barrel), while spending on Michelob Light was $18 million ($7.05 per barrel). Advertising outlays on light beer began to drop off after 1988 as market shares stabilized and other brands and categories were introduced. In 1999, A-B spent $105 million ($3.66 per barrel) on Bud Light and $23 million on Michelob Light ($8.33 per barrel). For the two A-B brands, 98% of the total budget was spent on television advertising. Miller spent about $83 million on advertising of Lite ($5.20 per barrel), with 92% of the spending accounted for by television and 7% by magazine advertising. In constant 1999 dollars, Miller’s total budget for Lite was about $117 million in 1986. Hence, Miller’s real advertising outlays fell by 29% as its market share shrank. The consequence of all this spending was a marked shift over time in the market shares of brands and beer categories. Light beer sales were about 35.4 million barrels in 1984, or 20% of the beer market. By 1990, sales were 59.1 million barrels and the market share was 31%. In 1999, light beer sales were 83.2 million barrels, or 42% of the beer market. Hence, consumption of lowcalorie beer has grown both in absolute terms and as a percentage of market. Bud Light was the second leading brand in 1999, with 14.6% of the beer market (compared to 5.8% in 1990). Miller Lite was the third leading brand in 1999, with 8.1% of the market (compared to 10.0% in 1990). It seems likely that advertising played an important role in shifts of brand and category sales in the market for beer. However, it does not follow that increased category advertising or brand success expanded the beer market as a whole. Table 3 and Fig. 1a show percapita consumption of light beer and all beer for the period 1980–1999. Although per-capita sales of low-calorie beer grew by 189% during this period, marketwide consumption of beer declined by about 11% from 33.7 gallons per capita in 1980 to 30.1 gallons per capita in 1999. Advertising in the light beer category might have had something to do with slowing the overall rate of decline, but this has not been demonstrated empirically. In any event, marketing successes at the brand and category levels failed to expand the beer market. A noticeable spillover effect is not present in these data, except at the category level. Further, as noted by Elzinga (2001, p. 101), popular-priced beer declined from 60% of the market in 1970 to less that 15% in 1998 as consumers shifted to light beer, imports, craft beers, and other premium-priced categories. Hence, it is equally likely that market segmentation shifted consumer preferences toward higher-priced brands and categories, with resultant lighter wallets and movement up the market demand schedule. Advertising and product differentiation may have reduced the quantity of sales, although not necessarily the dollar value of sales.

Alcohol Advertising and Advertising Bans

279

Table 3. Alcoholic Beverage Consumption, 1980–1999 (gallons per capita; ages 18 + ). Year

Light Beer

All Beer

Percent Light

Wine Coolers

All Wine

Percent Coolers

Vodka

All Spirits

Percent Vodka

1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999

4.37 5.02 6.00 6.23 6.45 7.16 7.71 8.01 8.48 8.96 9.88 10.38 10.58 10.95 10.66 10.70 11.14 11.56 12.03 12.62

33.71 33.99 33.48 33.09 32.63 32.26 32.61 32.42 32.10 31.74 32.22 31.20 30.81 30.58 30.48 29.83 29.84 29.69 29.96 30.10

13.0 14.8 17.9 18.8 19.8 22.2 23.6 24.7 26.4 28.2 30.7 33.3 34.3 35.8 35.0 35.9 37.3 38.9 40.2 41.9

– – – 0.05 0.19 0.45 0.71 0.71 0.62 0.51 0.50 0.35 0.24 0.15 0.12 0.10 0.10 0.07 0.05 0.04

2.93 3.04 3.04 3.08 3.20 3.31 3.37 3.30 3.10 2.88 2.77 2.52 2.53 2.33 2.39 2.42 2.53 2.56 2.57 2.65

– – – 1.6 5.9 13.6 21.1 21.5 20.0 17.7 18.0 13.9 9.5 6.4 5.0 4.1 4.0 2.7 2.0 1.5

0.53 0.53 0.50 0.50 0.48 0.48 0.47 0.45 0.46 0.45 0.45 0.42 0.41 0.40 0.39 0.39 0.40 0.39 0.40 0.41

2.75 2.70 2.59 2.52 2.46 2.37 2.22 2.17 2.09 2.03 2.02 1.85 1.88 1.80 1.74 1.68 1.66 1.66 1.67 1.69

19.3 19.6 19.3 19.8 19.5 20.2 21.2 20.7 22.0 22.2 22.3 22.7 22.8 22.2 22.4 23.2 24.1 23.5 24.0 24.3

Sources: Beer data – Beer Industry Update (various years) and Adams/Jobson Beer Handbook (various years). Wine data – Adams/Jobson Wine Handbook (various issues). Spirits data – Adams/ Jobson Liquor Handbook (various issues). Population – U.S. Bureau of the Census, Resident population estimates of the U.S. by age and sex.

B. Wine Coolers and the Market for Wine Wine coolers were introduced in 1981 by the original California Cooler. The ﬁrst coolers were a mixture of white wine, fruit juices, and carbonated water. Coolers generally contained about 5–6% alcohol by volume, or about the same amount as most beer. Initially, coolers were identiﬁed with the California lifestyle and were regarded as more acceptable to women than beer. This theme was reﬂected in brand names such as “California Splash” and “Golden Wine.” By 1985, there were more than 100 wine cooler brands on the market produced by wineries and other leading beverage producers such as Brown-Forman Distillers, Schenley, and Seagram. Most of these brands would quietly disappear and some would be converted to malt-based products, which are

280

JON P. NELSON

Fig. 1. (a) Light Beer and Beer per capita, 1980–1999; (b) Wine Coolers and Wine per capita, 1980–1999; (c) Vodka and Spirits per capita, 1980–1999.

Alcohol Advertising and Advertising Bans

281

taxed at lower rates than wine. In 1983, only 3.3 million cases of wine coolers were sold, but this amount increased to 14.1 million cases in 1984; 33.3 million in 1985; 52.7 million in 1986; and 53.6 million cases in 1987. It was all downhill after 1987, and only 3.0 million cases of wine coolers were sold in 1999 (Fig. 1b). In the peak year, coolers accounted for more than 21% of all wine sales, but captured only 1.5% of the wine market in 1999. Within the category, the leading marketing success story was E & J Gallo’s brand, sold under the trade-name “Bartles & Jaymes.” Frank Bartles and Ed Jaymes, the plain-looking older spokesmodels, were the creation of the Riney advertising agency. The tagline “We thank you for your support,” was repeated over and over during the seven years that the Frank & Ed campaign was aired on television. Gallo’s name was never mentioned in the ads (and Jaymes never spoke), so many people believed that Frank & Ed actually produced the brand. The connection between the avuncular duo and a California lifestyle is elusive, but the ad campaign is ranked among the Top 100 TV commercials of all time (Kanner, 1999, p. 134) and it won an Advertising Age award for effective use of humor. The ads also pushed Bartles & Jaymes to the top of the wine cooler market in 1986 – just one year after its national launch. Total category spending on wine cooler ads peaked in 1986 at $152.5 million ($2.89 per case), which was a greater amount per case than any other alcoholic beverage, and 94% was spent on broadcast advertising. Since 1990, Bartles & Jaymes has shared the top spot in a shrinking category with Seagram’s Coolers. In the early years, both ﬁrms heavily promoted their products, and together dominated the category after 1989. In 1990, Gallo spent $11.9 million on advertising, almost all on television ads, and Seagram spent $7.3 million. However, faced with declining demand, Seagram in 1998 spent only $952,000 on promotion, while Gallo dropped all advertising of its cooler brand in that year. Consumer support for Frank & Ed had come and gone. A few other lowalcohol refreshers remain on the market, but most products are now categorized as malt coolers or malt liquors. Except for Coor’s Zima brand, promotion of these products is minimal or nonexistent. Table 3 and Fig. 1b show per-capita consumption of wine coolers and all wine for the period 1984–1999. In contrast to light beer, increased sales of coolers played a temporary role in expanding the market for wine. Undoubtedly part of this increase came at the expense of sales of other alcoholic beverages, including beer. However, despite substantial initial success by two ﬁrms with considerable marketing skills, the category has not been a long-term success. Contrary to frequent claims, advertising did not create and sustain consumer demand. Some winemakers initially hoped that consumers would try wine coolers, and then move up to table wines and other premium products. Clearly,

282

JON P. NELSON

this did not happen. The wine cooler story illustrates the increase in advertising that usually accompanies new product introductions, but this heightened activity did not spillover to the broader market on a long-term basis. Indeed, causality appears to run from sales to advertising, rather than visa versa (Berndt, 1991, p. 398). Brand advertising alone is not sufﬁcient to affect the long-run distribution of demand between product categories or sustain a longterm increase in marketwide demand. C. Vodka and the Market for Distilled Spirits In 1980, vodka consumption in the U.S. was 0.53 gallons per capita and total spirits consumption was 2.75 gallons per capita. However, by 1999, these two amounts had declined to 0.41 gallons and 1.69 gallons per capita, respectively (Table 3 and Fig. 1c). Although the vodka category declined in absolute terms, vodka’s share of the spirits market rose from 19% in 1980 to 24% in 1999, making it the largest single category of spirits. In 1980, a new brand of vodka entered the crowded U.S. market for alcoholic beverages. Absolut Vodka is an advertising icon (Hamilton, 2000; Lewis, 1996). A premium-priced brand, it is produced by Vin & Spirits, the Swedish government’s alcohol monopoly (no alcohol advertising of any kind is allowed in Sweden). The brand’s U.S. advertising campaign is known for its creative print ads, which feature the distinctive Absolut bottle and a two-word message that plays on its name, such as “Absolut Perfection” and “Absolut Joy.” The campaign evolved over time to feature artists, cities and locations, objects, holidays, designer-name clothes, and other novel or upscale images. In 1989, Absolut was only the third best brand by sales among vodkas and the 12th leading spirits brand overall. In 1999, Absolut was the second best seller among vodkas (behind Smirnoff) and the third leading brand overall. Case sales rose from about 2.25 million cases in 1989 to 4.05 million in 1999, or an increase of 80%. Case sales of vodka during the same period remained stable at about 35 million cases per year. Success at the brand level did not translate into success for the category (or market) as a whole. Absolut Vodka, and distilled spirits generally, are heavily promoted, but this has not translated into either increased category or market sales. In 1984, Absolut spent about $9.65 per case on advertising, compared to $1.52 per case for category leader Smirnoff. In 1989, Absolut spent $10.6 million ($4.71 per case) on advertising, principally in magazines, making it the third leading vodka brand by total advertising expenditure (behind Dewar’s and Smirnoff). Total spending on spirits was $272.8 million ($1.75 per case). In 1999, Absolut – now distributed by marketing giant Seagram – spent $32.4 million ($8.00 per

Alcohol Advertising and Advertising Bans

283

case), and was the leading brand among all distilled spirits by total advertising expenditure. The second most advertised brand was Bacardi rum, which spent $15.8 million ($2.20 per case). Total spending on spirits in 1999 was about $321.3 million ($2.23 per case), with 79% of the ad dollars spent on magazines and newspapers. Table 3 and Fig. 1c display per-capita consumption of vodka and all distilled spirits for the period 1980–1999. Both vodka sales per capita and total consumption of spirits have declined steadily over time. With absolute clarity, this illustrates that brand-level success has not translated into expansion of either the vodka category or the market for distilled spirits. The remarkable success of Absolut as a brand has not had a noticeable spillover effect, which suggests that an industry-response function does not exist. Successive brand advertising translates into increased brand sales, but any marketwide (or crossbeverage) spillovers must be remarkably small in magnitude. This refutes Saffer’s claims. The three examples presented in this section as well as the results in expenditure studies show that the brand success fails to translate into a statistical relationship between industry-level advertising and marketwide sales. D. Advertising in a Declining Market The alcoholic beverage market has been declining since the early 1980s, or earlier in the case of spirits (Nelson, 1999). Per-capita consumption of spirts peaked in 1975; beer in 1981; and wine in 1986. Total ethanol consumption per capita peaked in 1981, and declined by 21% between 1981 and 1997 (NIAAA, 1999, p. 15). It would not be expected that total advertising would increase in a declining market, although this does not rule out increased advertising at the brand or category level. Table 4 displays nominal and real advertising outlays for the industry, including a percentage breakdown by major advertising media. Real advertising has declined by over 35% since the peak year of 1986. Note that the 1980s reﬂect remarkable levels of advertising by a few brands, and not an industry-wide explosion. Alcohol consumption overall was declining. There is no evidence from the three examples that these changes can be represented as a movement along a well-deﬁned industry-response function. Alcohol advertising affects brand shares, and virtually all studies of advertising expenditures at the beverage level support this conclusion. Alcohol advertising has not been shown to increase total alcohol (or beverage) consumption, regardless of the level of aggregation or time period.

284

JON P. NELSON

Table 4. Long-Term Trends in Alcohol Advertising. Year 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999P % Change, 1986–1999

Total Alcohol Ads Real Alcohol Ads Percent Percent Percent (million $) (1996 million $) Broadcast Print Outdoor 395.6 456.8 534.0 703.0 865.2 906.9 1014.9 1108.7 1182.9 1284.4 1293.0 1400.2 1374.7 1319.4 1200.4 1050.4 1119.5 1074.7 970.7 1000.9 1027.5 1008.8 1087.0 1187.6 1242.2 –

979.9 1067.5 1171.3 1442.1 1641.7 1580.5 1618.7 1667.0 1708.4 1788.9 1746.1 1850.6 1766.1 1639.8 1436.6 1209.7 1247.2 1172.0 1030.9 1041.1 1046.4 1008.8 1069.2 1154.6 1189.5 –35.7%

44.0 46.6 52.5 54.6 54.6 55.1 56.6 58.1 62.0 66.0 68.2 73.5 73.5 74.4 68.2 64.8 66.4 68.5 70.4 69.4 68.2 68.5 66.5 66.3 64.2 –

45.6 43.6 39.3 38.7 38.9 38.3 36.3 33.9 31.2 27.2 25.6 22.0 21.4 21.2 26.9 29.8 28.1 26.2 23.4 22.6 23.2 24.3 27.5 25.8 26.9 –

10.4 9.8 8.2 6.7 6.6 6.6 7.1 8.0 6.9 6.8 6.2 4.5 5.1 4.4 4.8 5.4 5.5 5.3 6.2 8.0 8.6 7.3 6.0 7.9 8.9 –

Notes: 1975–1991 data from Impact (various issues). 1992–1999 data from LNA/Competitive Media (various issues). Preliminary (P) data for 1999. Nominal data deﬂated by the GDP implicit price deﬂator (1996 = 100) from the Economic Report of the President.

VI. CONCLUSIONS My purpose in this chapter has been to provide an answer to a speciﬁc question. Is there a direct and material effect of advertising on the overall level of alcohol consumption or on alcohol abuse? The bulk of the scientiﬁc evidence presented in this survey indicates that the answer to this question is “No.” Studies of statelevel bans of billboards and publicly visible displays fail to demonstrate that selective bans reduce consumption. Studies of international bans that cover

Alcohol Advertising and Advertising Bans

285

more media and beverages reach the same conclusion for more comprehensive bans. Studies of advertising expenditures fail to ﬁnd statistically signiﬁcant effects of advertising on alcohol consumption, despite numerous advances in econometric methods. Studies of brand advertising fail to establish a spillover effect of successful brand advertising on the marketwide demand for alcoholic beverages. Under conditions of brand rivalry in a declining market, an industrywide advertising-response function simply does not exist. Furthermore, survey-research studies have failed to measure the magnitude of the effect of advertising on youth intentions or behavior in a manner that is suitable for policy analysis. These studies do ﬁnd a direct effect of parents’ drinking on such behavior, which can be incorporated into future studies of youth alcohol behaviors (see Cook & Moore, 2001). This direct effect also means that results from aggregate studies of alcohol consumption are applicable to youth behaviors. Policy analysis of regulation of alcohol advertising must begin with the Central Hudson formula. Commercial speech regulations that concern lawful activity must be shown to directly and materially advance a substantial government interest. What I have demonstrated in this chapter is that the third prong of the Central Hudson formula cannot be satisﬁed. Given these negative results, legislative bodies should abandon symbolic polices that restrict speech, and turn instead to remedies that have been demonstrated to be effective for youth drinking problems, such as rigorous enforcement of minimum legal age laws or increased penalties for illegal sale to minors. Cost-effective alcohol policies should be based on something other than widespread mis-beliefs about the nature or effectiveness of commercial speech.

NOTES 1. Previous literature surveys include Calfee (1997), Calfee and Scheraga (1994), Fisher (1993), Fisher and Cook (1995), FTC (1985), NIAAA (2000), and Smart (1988). 2. Calfee (1997) demonstrates how government agencies use controls of commercial speech to achieve political ends, which may include the suppression of market competition. Meeting transcripts of the Liquor Control Administration showed that Rhode Island liquor retailers constantly monitored competitors’ ads for price claims, with the aim of getting the offender’s license revoked. In a remarkable parallel to Valentine, one retailer asked if it could run an ad that simply protested the Administration’s price ban. The Administrator denied the request on grounds that this would signal low prices (Calfee, 1997, p. 111). Hence, political speech was construed as commercial speech, and regulated. 3. Posadas created the possibility of a “vice”or “harmful product” exception to First Amendment protection; see also United States v. Edge Broadcasting Co., 509 U.S. 418

286

JON P. NELSON

(1993), in which a split Court upheld a FCC regulation prohibiting advertising of state lotteries, except within the state where the lottery was authorized. However, in 44 Liquormart, Justice Stevens dismissed the notion of a vice exception for alcohol, citing the Court’s ruling in Rubin v. Coors, 514 U.S. 476 (1995). He argued that a vice exception would be “difﬁcult, if not impossible, to deﬁne” (517 U.S. 514). In Greater New Orleans Broadcasting Assoc., Inc. v. United States, 527 U.S. 173 (1999), the Court unanimously ruled that the FCC could not prohibit the advertising of non-tribal casino gambling that might be heard in neighboring states where private casinos were unlawful. Justice Stevens reasoned that federal statutes presently accommodate both pro-gambling and anti-gambling policies, and that “. . . some form of gambling is legal in nearly every state” (527 U.S. 187). He also recognized that advertising may do little more than promote brand choice, since “. . . it is reasonable to assume that much of that advertising would merely channel gamblers to one casino rather than another” (527 U.S. 189). 4. For an economic analysis of the effects of 44 Liquormart, see Milyo & Waldfogel (1999). Numerous law review articles discuss commercial speech and the decisions examined here; see Blue (1996), Costello (1997), O’Neill (1998), Post (2000), Ritter (1997), Sackett (1983), and Troy (1999). For a legal-economic analysis that takes issue with many of the Court’s recent decisions, see McChesney (1997), arguing that the Court has failed to develop a consistent economic view of the value of commercial speech as originally stated in Virginia Pharmacy and Bates v. State Bar of Arizona, 433 U.S. 350 (1977). For an economic analysis of early cases, see Coase (1974, 1977), comparing protection of political and commercial speech and anticipating the Central Hudson doctrine. See Lefﬂer (1983) for an economic analysis of a case involving a complete ban of billboard advertising in San Diego. 5. During the October 2000 term, the Supreme Court ruled on two cases involving advertising. In United States v. United Foods, Inc., No. 00–276 (decided June 25, 2001), the issue was whether the federal government could compel an independent mushroom grower to pay a marketing-order assessment used primarily to fund generic advertisements, and which beneﬁtted competitors of the respondent. In a 6–3 decision, the Court held that the First Amendment prevents the government from compelling individuals to express certain views. The government did not rely on Central Hudson to challenge the appeals court decision. In Lorillard v. Reilly, No. 00-596 (decided June 28, 2001), the Court considered a Massachusetts law (passed in 1999) that prohibited visible advertising of cigarettes and other tobacco products within a 1,000 feet of schools, parks, and playgrounds, effectively banning all outdoor signs and window displays. Point-of-sale signs also were banned if they were placed lower than ﬁve feet from the ﬂoor in stores accessible to minors. The issues in Lorillard were, ﬁrst, preemption of state actions under the Federal Cigarette Labeling and Advertising Act (FCLAA) and, second, the constitutionality of “location-based” restrictions of commercial speech under the Central Hudson formula. In a 5–4 decision, the majority held that the FCLAA preempted cigarette regulations with respect to outdoor and pointof-sale advertising. In order to avoid preemption, the state would have to uniformly ban outdoor displays for all products. On the second issue, the ruling found that the cigarette restrictions satisﬁed the third prong of the Central Hudson test, but failed to satisfy the fourth prong. Citing Cincinnati v. Discovery Network, the ruling held that the sale and use of tobacco by adults was a legal activity, and the state’s regulations unduly impinged on both the speaker’s ability to propose, and the hearer’s opportunity to consider, a legal

Alcohol Advertising and Advertising Bans

287

commercial transaction. Justice O’Connor concluded that the state had failed to show that its regulations were “not more extensive than necessary to advance the State’s substantial interest in preventing underage tobacco use” (No. 00-596, 35). The in-store regulations for smokeless tobacco and cigars were held to violate both the third and fourth prongs. Other state restrictions affecting seller conduct, and not speech, were upheld by the decision. The dissenting opinion stressed preemption by the FCLAA. 6. Data for the regression are from NIAAA (1985, 1999) and NIH (2000a, b) for total U.S. ethanol consumption per capita (ages 20 + ) and prevalence of binge drinking by 12th-graders, respectively. The bivariate regression was Binge = –9.38 + 15.75 (Total U.S. ethanol); R-sq = 0.884. The t-statistic on the slope coefﬁcient is 13.2. For similar evidence for other countries, see Smart and Ogborne (2000a, b). 7. In addition to the issues discussed here, many other econometric issues have been investigated. Studies that adjust for cross-section heteroscedasticity include Nelson (2001) and Nelson and Young (2001). Studies that account for data outliers include Nelson (1990b) and Nelson and Young (2001). Studies that test for structural change or account for exogenous taste changes include Calfee and Scheraga (1994), Duffy (1991b, 1995), Lee and Tremblay (1992), Nelson (1990a, 2001), Nelson and Moran (1995), and Tegene (1990). 8. A study by Saffer (1997) failed to show that local alcohol advertising expenditures, including billboards, have a direct and material effect on proxies for youth drunk driving fatalities. Alcohol consumption is not measured in this study. Rather, the author substitutes for the possible (indirect) determinates of drinking prior to driving, which includes advertising. The two regressions with insigniﬁcant advertising coefﬁcients were for 18–20-year olds, although coefﬁcients for older drivers were signiﬁcantly positive. Saffer (1997, p. 438) argued that this may be due to the cumulative effects of advertising, but the results in expenditure studies clearly speak against this relationship. 9. The path analysis approach to structural estimation was originally developed in the 1920s and 1930s by the biometrician Sewall Wright, but one of its early applications was the identiﬁcation and estimation via instrumental variables of a structural supplydemand model for butter (Goldberger, 1972, p. 986). Wright’s estimation of causal relationships is a forerunner of other methods of simultaneous equations estimation, although most of his empirical work involved recursive models where the paths do not return on themselves (Goldberger, 1972, p. 988). Blalock (1963) and Duncan (1966) published early expositions of path analysis for sociologists, and Goldberger (1971) published a comparison of structural methods in econometrics with those used in psychology. Two edited collections that represent the multidisciplinary origins of modern path analysis are Blalock (1971) and Goldberger and Duncan (1973). Standard treatments of causal modeling for the practitioner include Asher (1983), Bollen (1989), Loehlin (1998), and Long (1983). 10. The three studies reviewed here are representative of the much larger literature with regard to methods, results, and policy implications. Brief literature reviews are found in Grube (1993, 1995). Grube (1995, p. 117) concluded that “. . . available studies indicate that alcohol advertising may have small but signiﬁcant effects on the beliefs, intentions, and possibly behaviors of young people.” See also Martin (1995, p. v), concluding that the evidence from survey-research studies provides only weak, limited, and inconclusive support for the belief that alcohol advertising affects the initiation and amount of youthful drinking. A more comprehensive review by the NIAAA (2000) also

288

JON P. NELSON

discusses econometric, experimental, and survey-research studies. With regard to survey studies, the NIAAA found that “the evidence . . . is far from conclusive (NIAAA, 2000, p. 423) . . . [and] the survey study designs employed thus far have not been able to establish whether, for example, the advertisements caused the beliefs and behaviors, or whether preexisting beliefs and behaviors led to an increased awareness of the advertisements” (NIAAA, 2000, p. 412). Each of these reviews casts considerable doubt on whether the results from survey-research studies can be applied to the policy issue of youth drinking. 11. Indirect effects are the products of the structural coefﬁcients along the path (or paths) in question. Previous work that lays out the mathematical relationships between indirect and direct effects includes contributions by Alwin and Hauser (1975), Asher (1983, p. 35), Bollen (1989, p. 37), Finney (1972), Fox (1980, 1985), and Short and Hennessy (1994). Standard errors for indirect effects are discussed by MacKinnon and Dwyer (1993), Sobel (1988), and others. 12. Sources of information and data for the three examples include Adams/Jobson Beer Handbook (Adams Business Media, various editions); Adams/Jobson Wine Handbook (Adams Business Media, various editions); Adams/Jobson Liquor Handbook (Adams Business Media, various editions); Beer Industry Update (Beer Marketer’s Insights, various editions); Beverage World (Keller International, various issues); Brewers Almanac (Beer Institute, various editions); Impact:U.S. News and Research for the Wine, Spirits and Beer Executive (Sanken Communications, various issues); LNA/ Mediawatch Multi-Media Service: Competitive Media Reporting (Distilled Spirits Council of the U.S., various issues); Marketing & Media Decisions (Decisions Publications, various issues); and Modern Brewery Age (MBA, Inc., various issues). 13. Television advertising was the primary weapon for this battle. There is a limited body of research that focuses on substitution among advertising media. Bresnahan (1984) and Seldon et al. (2000) estimated the degree of substitutability between television, radio, and print advertising media in the U.S. beer industry, and found a high degree of substitutability between media. See also Färe et al. (2001) and Seldon and Jung (1993) for comparable results for brewing ﬁrms and all product advertising, respectively.

ACKNOWLEDGMENTS This paper reﬂects research begun in 1985 when I testiﬁed before the Pennsylvania Liquor Control Board on its ban of beer price advertising. Gene Waye was instrumental in my involvement at that time. I have beneﬁtted greatly from collaboration, assistance, and discussions with Mike Baye, Jack Calfee, Lee Carpenter, Ed Coulson, Jim Donahue, Mark Gius, Don Kenkel, John Moran, Mark Roberts, Reg Smart, and Doug Young. I also want to acknowledge the excellent research assistance on earlier projects received from Sharon Bee, Mohit Behende, Sheng-Wen Chang, Tsui-Fang Lin, and Janice Stenger.

Alcohol Advertising and Advertising Bans

289

REFERENCES Alwin, D. F., & Hauser, R. M. (1975). The decomposition of effects in path analysis. American Sociological Review, 40, 37–54. Ambler, T. (1996). Can alcohol misuse be reduced by banning advertising? International Journal of Advertising, 15, 167–174. American Medical Association, Board of Trustees (1986). Alcohol: Advertising, counteradvertising, and depiction in the public media. Journal of the American Medical Association, 256, 1485–1488. Asher, H. B. (1983). Causal Modeling (2nd ed.). Newbury Park, CA: Sage. Assmus, G., Farley, J. U., & Lehmann, D. R. (1984). How advertising affects sales: Meta-analysis of econometric results. Journal of Marketing Research, 21, 65–74. Austin, E. W., Knaus, C. (2000). Predicting the potential for risky behavior among those ‘too young’ to drink as a result of appealing advertising. Journal of Health Communications, 5, 13–27. Backman, J. (1967). Advertising and Competition. New York: University of London Press. Bass, F. M., & Leone, R. P. (1983). Temporal aggregation, the data interval bias, and empirical estimation of bimonthly relations from annual data. Management Science, 29, 1–11. Beales, J. H., & Muris, T. J. (1993). State and Federal Regulation of National Advertising. Washington, D.C.: AEI Press. Berndt, E. R. (1991). The Practice of Econometrics: Classic and Contemporary. Reading, MA: Addison-Wesley. Blake, D., & Nied, A. (1997). The demand for alcohol in the United Kingdom. Applied Economics, 29, 1655–1672. Blalock, H. M. (1963). Correlated independent variables: The problem of multicollinearity. American Journal of Sociology, 42, 233–237. Blalock, H. M. (Ed.) (1971). Causal Models in the Social Sciences. Chicago: Aldine-Atherton. Blue, T. D. (1996). Over the edge: The Fourth’s Circuit commercial speech analysis in Penn Advertising & Anheuser-Busch. North Carolina Law Review, 74, 2086–2107. Bollen, K. A. (1989). Structural Equations with Latent Variables. New York: Wiley. Boyd, R., & Seldon, B. J. (1990). The ﬂeeting effect of advertising: Empirical evidence from a case study. Economics Letters, 34, 375–379. Bresnahan, T. F. (1984). The demand for advertising by medium: Implications for the economies of scale in advertising. In: P. M. Ippolito & D. T. Scheffman (Eds), Empirical Approaches to Consumer Protection Economics (pp. 135–163). Washington, D.C.: Federal Trade Commission. Calfee, J. E. (1997). Fear of Persuasion: A New Perspective on Advertising and Regulation. Washington, D.C.: AEI Press. Calfee, J. E., & Scheraga, C. (1994). The inﬂuence of advertising on alcohol consumption: A literature review and an econometric analysis of four European nations. International Journal of Advertising, 13, 287–310. Chaloupka, F. J., Grossman, M., & Saffer, H. (2001). The effects of price on alcohol consumption. Alcohol Research & Health, forthcoming. Clarke, D. G. (1976). Econometric measurement of the duration of advertising effects on sales. Journal of Marketing Research, 13, 345–357. Coase, R. H. (1974). The market for goods and the market for ideas. American Economic Review, 64, 384–391.

290

JON P. NELSON

Coase, R. H. (1977). Advertising and free speech. Journal of Legal Studies, 6, 1–34. Connolly, G. M., Casswell, S., Zhang, J-F., & Silva, P. A. (1994). Alcohol in the mass media and drinking by adolescents: A longitudinal study. Addiction, 89, 1255–1263. Cook, P. J., & Moore, M. J. (2000). Alcohol. In: A. Culyer & J. P. Newhouse (Eds), The Handbook of Health Economics (Vol. 1B, pp. 1629–1673). Amsterdam: North-Holland. Cook, P. J., & Moore, M. J. (2001). Environment and persistence in youthful drinking patterns. In: J. Gruber (Ed.), Risky Behavior Among Youths: An Economic Analysis (pp. 375–437). Chicago: University of Chicago Press. Costello, S. P. (1997). Strange brew: the state of commercial speech jurisprudence before and after 44 Liquormart, Inc. v. Rhode Island. Case Western Law Review, 47, 681–749. Coulson, N. E., Moran, J. R., & Nelson, J. P. (2001). The long-run demand for alcoholic beverages and the advertising debate: A cointegration approach. In: Advances in Applied Microeconomics (this volume). Amsterdam: JAI Press. Czart, C., Pacula, R. L., Chaloupka, F. J., & Wechsler, H. (2001). The impact of prices and control policies on cigarette smoking among college students. Contemporary Economic Policy, 19, 135–149. Dixit, A. K., & Norman, V. D. (1978). Advertising and welfare. Bell Journal of Economics, 9, 1–17. Duffy, M. (1990). Advertising and alcoholic drink demand in the U.K.: Some further Rotterdam model estimates. International Journal of Advertising, 9, 247–257. Duffy, M. (1991a). Advertising in demand systems: Testing a Galbraithian hypothesis. Applied Economics, 23, 485–496. Duffy, M. (1991b). Advertising and the consumption of tobacco and alcoholic drink: A systemwide analysis. Scottish Journal of Political Economy, 38, 369–385. Duffy, M. (1995). Advertising in demand systems for alcoholic drinks and tobacco: A comparative study. Journal of Policy Modeling, 17, 557–577. Duffy, M. (2001). Advertising in consumer allocation models: Choice of functional form. Applied Economics, 33, 437–456. Duncan, O. D. (1966). Path analysis: sociological examples. American Journal of Sociology, 72, 1–16. Elzinga, K. G. (2001). Beer. In: W. Adams & J. Brock (Eds), The Structure of American Industry (10th ed., pp. 85–113). Upper Saddle River, NJ: Prentice-Hall. Färe, R., Grosskopf, S., Seldon, B. J., & Tremblay, V. J. (2001). Advertising efﬁciency and the choice of media mix: A case of beer. Unpublished paper, Department of Economics, Oregon State University. Finney, J. M. (1972). Indirect effects in path analysis. Sociological Methods and Research, 1, 175–186. Fisher, J. C. (1993). Advertising, Alcohol Consumption, and Abuse: A Worldwide Survey. Westport, CT: Greenwood Press. Fisher, J. C., & Cook, P. A. (1995). Advertising, Alcohol Consumption, and Mortality: An Empirical Investigation. Westport, CT: Greenwood Press. Fox, J. (1980). Effect analysis in structural equation models: Extensions and simpliﬁed methods of computation. Sociological Methods and Research, 9, 3–28. Fox, J. (1985). Effect analysis in structural equation models II: Calculation of speciﬁc indirect effects. Sociological Methods and Research, 14, 81–95. Franke, G., & Wilcox, G. (1987). Alcoholic beverage advertising and consumption in the United States, 1964–1984. Journal of Advertising, 16, 22–30.

Alcohol Advertising and Advertising Bans

291

Gius, M. P. (1991). The Extent of the Market in the Liquor Industry: An Empirical Test of Localized Brand Rivalry, 1970–1988. Unpublished Ph.D. dissertation, Department of Economics, Pennsylvania State University. Gius, M. P. (1996). Using panel data to determine the effect of advertising on brand-level distilled spirits sales. Journal of Studies on Alcohol, 57, 73–76. Goel, R., & Morey, M. (1995). The interdependence of cigarette and liquor demand. Southern Economic Journal, 62, 451–459. Goldberger, A. S. (1971). Econometrics and psychometrics: A survey of commonalities. Psychometrica, 36, 83–107. Goldberger A. S. (1972). Structural equations models in the social sciences. Econometrica, 40, 979–1001. Goldberger, A. S., & Duncan, O. D. (Eds) (1973). Structural Equation Models in the Social Sciences. New York: Seminar Press. Greer, D. F. (1993). Beer: Causes of structural change. In: L. L. Duetsch (Ed.), Industry Studies (pp. 85–115). Englewood Cliffs, NJ: Prentice-Hall. Grossman, M., Chaloupka, F. J., Saffer, H., & Laixuthai, A. (1994). Effects of alcohol price policy on youth: A summary of economic research. Journal of Research on Adolescence, 4, 347–364. Grube, J. W. (1993). Alcohol portrayals and alcohol advertising on television: Content and effects on children and adolescents. Alcohol Health & Research World, 17, 61–66. Grube, J. W. (1995). Television alcohol portrayals, alcohol advertising, and alcohol expectancies among children and adolescents. In: S. E. Martin (Ed.), The Effects of the Mass Media on the Use and Abuse of Alcohol (pp. 105–121). Washington, D.C.: NIAAA. Grube, J. W., & Wallack, L. (1994). Television beer advertising and drinking knowledge, beliefs, and intentions among schoolchildren. American Journal of Public Health, 84, 254–259. Hamilton, C. (2000). Absolut: Biography of a Bottle. London: Texere. Harrison, L., & Godfrey, C. (1989). Alcohol advertising controls in the 1990s. International Journal of Advertising, 8, 167–180. Hoadley, J. F., Fuchs, B. C., & Holder, H. D. (1984). The effect of alcohol beverage restrictions on consumption: A 25-year longitudinal analysis. American Journal of Drug and Alcohol Abuse, 10, 375–401. Kanner, B. K. (1999). The 100 Best TV Commercials . . . . And Why They Worked. New York: Times Business. Kaul, A., & Wittick, D. R. (1995). Empirical generalizations about the impact of advertising on price sensitivity and price. Marketing Science, 14, 151–160. Kenkel, D. S. (1993). Prohibition versus taxation: Reconsidering the minimum drinking age. Contemporary Policy Issues, 12, 137–150. Kinnucan, H. W., Kaiser, H. M., Miao, Y., & Xiao, H. (2001). Effects of advertising on U.S. nonalcoholic beverage demand: Evidence from a two-stage Rotterdam model. In: Advances in Applied Microeconomics (this volume). Amsterdam: JAI Press. Lariviere, E., Larue, B., & Chalfant, J. (2000). Modeling the demand for alcoholic beverages and advertising speciﬁcations. Agricultural Economics, 22, 147–162. Lee, B., & Tremblay, V. J. (1992). Advertising and the U.S. market demand for beer. Applied Economics, 24, 69–76. Lefﬂer, K. B. (1983). The prohibition of billboard advertising: An economic analysis of the Metromedia decision. Supreme Court Economic Review, 1, 113–133. Leone, R. P. (1995). Generalizing what is known about temporal aggregation and advertising carryover. Marketing Science, 14, 141–150.

292

JON P. NELSON

Lewis, R. W. (1996). Absolut Book: The Absolut Vodka Advertising Story. Boston: Journey Editions. Loehlin, J. C. (1998). Latent Variable Models: An Introduction to Factor, Path, and Structural Analysis (3rd ed.). Mahwah, NJ: Lawrence Erlbaum. Long, J. S. (1983). Covariance Structure Models: An Introduction to LISREL. Beverly Hills, CA: Sage. MacKinnon, D. P., & Dwyer, J. H. (1993). Estimating mediated effects in prevention studies. Evaluation Review, 17, 144–158. Makowsky, C. R., & Whitehead, P. C. (1991). Advertising and alcohol sales: A legal impact study. Journal of Studies on Alcohol, 52, 555–567. Markowitz, S., & Grossman, M. (1998). Alcohol regulation and domestic violence towards children. Contemporary Economic Policy, 16, 309–320. Markowitz, S., & Grossman, M. (2000). The effects of beer taxes on physical child abuse. Journal of Health Economics, 19, 271–282. Martin, S. E. (Ed.) (1995). The Effects of the Mass Media on the Use and Abuse of Alcohol. Washington, D.C.: NIAAA. McChesney, F. S. (1997). De-Bates and Re-Bates: The Supreme Court’s latest commercial speech cases. Supreme Court Economic Review, 5, 81–139. McGahan, A. M. (1995). Cooperation in prices and advertising: Trade associations in brewing after repeal. Journal of Law and Economics, 38, 521–559. Milgrom, R., & Roberts, J. (1986). Price and advertising signals of product quality. Journal of Political Economy, 94, 796–821. Milyo, J., & Waldfogel, J. (1999). The effect of price advertising on prices: Evidence in the wake of 44 Liquormart. American Economic Review, 89, 1081–1096. Moore, R. L., Farrar, R. T., & Collins, E. L. (1998). Advertising and Public Relations Law. Mahwah, NJ: Lawrence Erlbaum. Motta, M. (1997). Advertising bans. Unpublished paper, Department of Economics, Universitat Pompeu Fabra. Nelson, J. P. (1990a). State monopolies and alcoholic beverage consumption. Journal of Regulatory Economics, 2, 83–98. Nelson, J. P. (1990b). Effect of regulation on alcoholic beverage consumption: Regression diagnostics and inﬂuential data. In: R. R. Watson (Ed.), Drug and Alcohol Abuse Reviews: Prevention (pp. 223–243). Clifton, NJ: Humana Press. Nelson, J. P. (1997). Economic and demographic factors in U.S. alcohol demand: A growthaccounting analysis. Empirical Economics, 22, 83–102. Nelson, J. P. (1999). Broadcast advertising and U.S. demand for alcoholic beverages. Southern Economic Journal, 66, 774–790. Nelson, J. P. (2001). Advertising bans, monopoly, and alcohol demand: Testing for substitution effects using panel data. Unpublished paper, Department of Economics, Pennsylvania State University. Nelson, J. P., & Moran, J. R. (1995). Advertising and U.S. alcoholic beverage demand: Systemwide estimates. Applied Economics, 27, 1225–1236. Nelson, J. P., & Young, D. J. (2001). Do advertising bans work? An international comparison. International Journal of Advertising, 20, 273–296. Ogborne, A. C., & Smart, R. G. (1980). Will restrictions on alcohol advertising reduce alcohol consumption? British Journal of Addiction, 75, 293–296. O’Neil, R. M. (1997). Alcohol Advertising on the Air: Beyond the Reach of Government? Washington, D.C.: The Media Institute.

Alcohol Advertising and Advertising Bans

293

O’Neill, W. D. (1998). Governmental restrictions on beverage alcohol advertising after 44 Liquormart v. Rhode Island. Saint Louis University Law Journal, 42, 267–300. Ornstein, S. O., & Hanssens, D. M. (1985). Alcohol control laws and the consumption of distilled spirits and beer. Journal of Consumer Research, 12, 200–213. Peltzman, S. (1971). Pricing in public and private enterprises. Journal of Law and Economics, 14, 109–147. Post, R. (2000). The constitutional status of commercial speech. UCLA Law Review, 48, 1–57. Ritter, D. (1997). The First Amendment, commercial speech, and the future of tobacco advertising after 44 Liquormart, Inc. v. Rhode Island. Wayne Law Review, 43, 1505–1532. Rome, E. P., & Roberts, W. H. (1985). Corporate and Commercial Speech: First Amendment Protection of Expression in Business. Westport, CT: Quorum. Sackett, J. (1983). Alcoholic beverage advertising and the First Amendment. University of Cincinnati Law Journal, 52, 861–900. Saffer, H. (1991). Alcohol advertising bans and alcohol abuse: An international perspective. Journal of Health Economics, 10, 65–79. Saffer, H. (1993). Advertising under the inﬂuence. In: M. E. Hilton & G. Bloss (Eds), Economics and the Prevention of Alcohol-Related Problems (pp. 125–140). Washington, D.C.: NIAAA. Saffer, H. (1995). Alcohol advertising and alcohol consumption: Econometric studies. In: S. E. Martin (Ed.), The Effects of the Mass Media on the Use and Abuse of Alcohol (pp. 277– 295). Washington, D.C.: NIAAA. Saffer, H. (1996). Studying the effects of alcohol advertising on consumption. Alcohol Health & Research World, 20, 266–272. Saffer, H. (1997). Alcohol advertising and motor vehicle fatalities. Review of Economics and Statistics, 79, 431–442. Saffer, H. (1998). Economic issues in cigarette and alcohol advertising. Journal of Drug Issues, 28, 781–793. Saffer, H., & Chaloupka, F. (2000). The effect of tobacco advertising bans on tobacco consumption. Journal of Health Economics, 19, 1117–1137. Saffer, H., Grossman, M., & Chaloupka, F. (2000). Alcohol control. In: B. Bouckaert & G. De Geest (Eds), Encyclopedia of Law and Economics (pp. 44–48). Northampton, MA: Elgar. Scherer, F. M., & Ross, D. (1990). Industrial Market Structure and Economic Performance (3rd ed.). Boston: Houghton Mifﬂin. Schmalensee, R. (1972). The Economics of Advertising. Amsterdam: North-Holland. Schweitzer, S. O., Intriligator, M. D., & Salehi, H. (1983). Alcoholism: An econometric model of its causes, its effect and its control. In: M. Grant, M. Plant & A. Williams (Eds), Economics and Alcohol: Consumption and Controls (pp. 107–127). New York: Harwood. Seldon, B. J., Jewell, R. T., & O’Brien, D. M. (2000). Media substitution and economies of scale in advertising. International Journal of Industrial Organization, 18, 1153–1180. Seldon, B. J., & Jung, C. (1993). Derived demand for advertising messages and substitutability among the media. Quarterly Review of Economics and Finance, 33, 71–86. Short, L. M., & Hennessy, M. (1994). Using structural equations to estimate effects of behavioral intervention. Structural Equation Modeling, 1, 68–81. Simon, J. L. (1970). Issues in the Economics of Advertising. Urbana, IL: University of Illinois Press. Simon, J. L. (1980). The shape of the advertising response function. Journal of Advertising Research, 20, 11–28.

294

JON P. NELSON

Smart, R. G. (1988). Does alcohol advertising affect overall consumption? A review of empirical studies. Journal of Studies on Alcohol, 49, 314–323. Smart, R. G., & Cutler, R. E. (1976). The alcohol advertising ban in British Columbia: Problems and effects on beverage consumption. British Journal of Addiction, 71, 13–21. Smart, R. G., & Ogborne, A. (2000a). Drinking and heavy drinking by students in 18 countries. Drug and Alcohol Dependence, 60, 315–318. Smart, R. G., & Ogborne, A. (2000b). Drug use and drinking among students in 36 countries. Addictive Behaviors, 25, 455–460. Smith, D. I. (1990). Consumption and advertising of alcoholic beverages in Australia, 1969–86. Drug and Alcohol Review, 9, 33–43. Sobel, M. E. (1998). Direct and indirect effects in linear structural equation models. In: J. S. Long (Ed.), Common Problems/Proper Solutions: Avoiding Errors in Quantitative Research (pp. 46–64). Newbury Park, CA: Sage. Starek, R. B. (Commissioner, FTC) (1997). Alcohol advertising and the First Amendment. Prepared remarks presented before the American Bar Association, August 4, 1997. Washington, D.C.: Federal Trade Commission. Stock, J., & Watson, M. (1988). Variable trends in economic time series. Journal of Economic Perspectives, 2, 147–174. Tegene, A. (1990). The Kalman ﬁlter approach for testing structural change in the demand for alcoholic beverages in the U.S. Applied Economics, 22, 1407–1416. Troy, D. E. (1999). Advertising: Not ‘low value’ speech. Yale Journal on Regulation, 16, 85–144. U.S. Department of Health and Human Services, Ofﬁce of the Inspector General (1991). Youth and Alcohol: Controlling Alcohol Advertising that Appeals to Youth. Washington, D.C.: DHHS. U.S. Federal Trade Commission (1985). Recommendations of the Staff of the Federal Trade Commission: Omnibus Petition for Regulation of Unfair and Deceptive Alcoholic Beverage Advertising and Marketing Practices. Washington, D.C.: FTC. U.S. Federal Trade Commission (1999). Self-Regulation in the Alcohol Industry: A Review of Industry Efforts to Avoid Promoting Alcohol to Underage Consumers. Washington, D.C.: FTC. U.S. National Institute on Alcohol Abuse and Alcoholism (1985). U.S. Apparent Consumption of Alcoholic Beverages based on State Sales, Taxation, or Receipt Data, Alcohol Epidemiologic Data Reference Manual, Vol. 1. Washington, D.C.: NIAAA. U.S. National Institute on Alcohol Abuse and Alcoholism (1999). Apparent Per Capita Alcohol Consumption: National, State, and Regional Trends, 1977–97, Surveillance Report No. 51. Washington, D.C.: NIAAA. U.S. National Institute on Alcohol Abuse and Alcoholism (2000). Alcohol advertising: What are the effects? In: 10th Special Report to the U.S. Congress on Alcohol and Health (pp. 412– 426). Washington, D.C.: NIAAA. U.S. National Institutes of Health (2000a). Monitoring the Future: Overview of Key Findings, 1999. Washington, D.C.: NIH. U.S. National Institutes of Health (2000b). Monitoring the Future: National Survey Results, 1975–1999. Washington, D.C.: NIH. Wilkinson, J. T. (1985). Alcohol and Accidents: An Economic Approach to Drunken Driving. Unpublished Ph.D. dissertation, Department of Economics, Vanderbilt University. Wilkinson, J. T. (1987). The effects of regulation on the demand for alcohol. Unpublished paper, Department of Economics, University of Missouri.

Alcohol Advertising and Advertising Bans

295

Yandle, B. (1983). Bootleggers and Baptists in the market for regulation. In: J. F. Shogren (Ed.), The Political Economy of Government Regulation (pp. 29–53). Norwell, MA: Kluwer. Young, D. J. (1993). Alcohol advertising bans and alcohol abuse: Comment. Journal of Health Economics, 12, 213–228.

This Page Intentionally Left Blank