Quantitative Economic History
The chapters in this book use the analytical tools and theoretical framework of economic...
60 downloads
1651 Views
1MB 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
Quantitative Economic History
The chapters in this book use the analytical tools and theoretical framework of economics to interpret quantitative historical evidence, offering new ways to approach historical issues and suggesting entirely new types of evidence outside conventional archives. Rosenbloom has gathered together seven essays from leading quantitative economic historians, illustrating the breadth of scope and continued importance of quantitative economic history. All of the chapters explore in one way or another the economic and social transformations associated with the emergence of an industrial and post-industrial economy, with most focusing on the transformations of the U.S. economy in the late nineteenth and early twentieth centuries, the technological innovations that factored into this transformation, and the relationship between industrialization and rising wealth inequality. This book will be of great interest to students and researchers engaged with U.S. Economic and British Demographic History, as well as quantitative economists in general. Joshua L. Rosenbloom is Professor of Economics and Associate Vice Provost for Research and Graduate Studies at the University of Kansas.
Routledge explorations in economic history
1 Economic Ideas and Government Policy Contributions to contemporary economic history Sir Alec Cairncross 2 The Organization of Labour Markets Modernity, culture and governance in Germany, Sweden, Britain and Japan Bo Stråth 3 Currency Convertibility The gold standard and beyond Edited by Jorge Braga de Macedo, Barry Eichengreen and Jaime Reis 4 Britain’s Place in the World A historical enquiry into import controls 1945–1960 Alan S. Milward and George Brennan 5 France and the International Economy From Vichy to the Treaty of Rome Frances M. B. Lynch 6 Monetary Standards and Exchange Rates M.C. Marcuzzo, L. Officer and A. Rosselli
7 Production Efficiency in Domesday England, 1086 John McDonald 8 Free Trade and its Reception 1815–1960 Freedom and trade: volume I Edited by Andrew Marrison 9 Conceiving Companies Joint-stock politics in Victorian England Timothy L. Alborn 10 The British Industrial Decline Reconsidered Edited by Jean-Pierre Dormois and Michael Dintenfass 11 The Conservatives and Industrial Efficiency, 1951–1964 Thirteen wasted years? Nick Tiratsoo and Jim Tomlinson 12 Pacific Centuries Pacific and Pacific Rim economic history since the 16th century Edited by Dennis O. Flynn, Lionel Frost and A.J.H. Latham
13 The Premodern Chinese Economy Structural equilibrium and capitalist sterility Gang Deng 14 The Role of Banks in Monitoring Firms The case of the crédit mobilier Elisabeth Paulet 15 Management of the National Debt in the United Kingdom, 1900–1932 Jeremy Wormell 16 An Economic History of Sweden Lars Magnusson 17 Freedom and Growth The rise of states and markets in Europe, 1300–1750 S. R. Epstein 18 The Mediterranean Response to Globalization Before 1950 Sevket Pamuk and Jeffrey G. Williamson 19 Production and Consumption in English Households 1600–1750 Mark Overton, Jane Whittle, Darron Dean and Andrew Hann 20 Governance, the State, Regulation and Industrial Relations Ian Clark 21 Early Modern Capitalism Economic and social change in Europe 1400–1800 Edited by Maarten Prak
22 An Economic History of London, 1800–1914 Michael Ball and David Sunderland 23 The Origins of National Financial Systems Alexander Gerschenkron reconsidered Edited by Douglas J. Forsyth and Daniel Verdier 24 The Russian Revolutionary Economy, 1890–1940 Ideas, debates and alternatives Vincent Barnett 25 Land Rights, Ethno Nationality and Sovereignty in History Edited by Stanley L. Engerman and Jacob Metzer 26 An Economic History of Film Edited by John Sedgwick and Mike Pokorny 27 The Foreign Exchange Market of London Development since 1900 John Atkin 28 Rethinking Economic Change in India Labour and livelihood Tirthankar Roy 29 The Mechanics of Modernity in Europe and East Asia The institutional origins of social change and stagnation Erik Ringmar
30 International Economic Integration in Historical Perspective Dennis M. P. McCarthy
36 Political Competition and Economic Regulation Edited by Peter Bernholz and Roland Vaubel
31 Theories of International Trade Adam Klug Edited by Warren Young and Michael Bordo
37 Industrial Development in Postwar Japan Hirohisa Kohama
32 Classical Trade Protectionism 1815–1914 Edited by Jean Pierre Dormois and Pedro Lains 33 Economy and Economics of Ancient Greece Takeshi Amemiya 34 Social Capital, Trust and the Industrial Revolution: 1780–1880 David Sunderland 35 Pricing Theory, Financing of International Organisations and Monetary History Lawrence H. Officer
38 Reflections on the Cliometrics Revolution Conversations with economic historians Edited by John S. Lyons, Louis P. Cain, and Samuel H. Williamson 39 Agriculture and Economic Development in Europe Since 1870 Edited by Pedro Lains and Vicente Pinilla 40 Quantitative Economic History The good of counting Edited by Joshua L. Rosenbloom
Quantitative Economic History The good of counting
Edited by Joshua L. Rosenbloom
First published 2008 by Routledge 2 Park Square, Milton Park, Abingdon, Oxon OX14 4RN Simultaneously published in the USA and Canada by Routledge 270 Madison Ave, New York, NY 10016 This edition published in the Taylor & Francis e-Library, 2008. “To purchase your own copy of this or any of Taylor & Francis or Routledge’s collection of thousands of eBooks please go to www.eBookstore.tandf.co.uk.” Routledge is an imprint of the Taylor & Francis Group, an informa business © 2008 Selection and editorial matter, Joshua L. Rosenbloom; individual chapters, the contributors All rights reserved. No part of this book may be reprinted or reproduced or utilized in any form or by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying and recording, or in any information storage or retrieval system, without permission in writing from the publishers. British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library Library of Congress Cataloging in Publication Data Rosenbloom, Joshua L. Quantitative economic history: the good of counting/Joshua L. Rosenbloom [editor]. p. cm. – (Routledge explorations in economic history) Includes bibliographical references and index. 1. Economic history–Statistical methods. 2. Economic history–Methodology. 3. United States–Economic conditions–Statistical methods. 4. Social change–United States–History–Statistical methods. I. Title. II. Title: Good of counting. HC51.R657 2008 330.9–dc22 2007043218 ISBN 0-203-92813-X Master e-book ISBN ISBN10: 0-415-77349-0 (hbk) ISBN10: 0-203-92813-X (ebk) ISBN13: 978-0-415-77349-2 (hbk) ISBN13: 978-0-203-92813-4 (ebk)
Contents
List of figures List of tables List of contributors Preface Acknowledgements
1
Editor’s introduction: the good of counting
ix x xii xv xviii
1
JOSHUA L. ROSENBLOOM
2
An economic history of bastardy in England and Wales
8
JOHN ERMISCH
3
Epidemics, demonstration effects, and investment in sanitation capital by U.S. cities in the early twentieth century
34
LOUIS CAIN AND ELYCE ROTELLA
4
Profitability, firm size, and business organization in nineteenth-century U.S. manufacturing
54
JEREMY ATACK AND FRED BATEMAN
5
Railroads and local economic development: the United States in the 1850s
78
MICHAEL R. HAINES AND ROBERT A. MARGO
6
Did refrigeration kill the hog–corn cycle?
100
LEE A. CRAIG AND MATTHEW T. HOLT
7
Measuring the intensity of state labor regulation during the Progressive Era REBECCA HOLMES, PRICE FISHBACK AND SAMUEL ALLEN
119
viii
8
Contents
Reexamining the distribution of wealth in 1870
146
JOSHUA L. ROSENBLOOM AND GREGORY W. STUTES
Publications of Thomas J. Weiss Index
170 173
Figures
2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 4.1 4.2 4.3 4.4 6.1 7.1 7.2 7.3 7.4 7.5 8.1
Births outside marriage (per 1,000 births) and general fertility rate (per 1,000 women aged 15–44) Births outside marriage (per 1,000 births) and marriages of single women (10,000s) Marriages of single women, births outside marriage and the unemployment rate Birth rate outside marriage, per 1,000 unmarried women Birth rate within marriage, per 1,000 married women Proportion of women not married Equilibria in proportions becoming a single mother Percent of births outside marriage, Western Europe Changes in births outside of marriage (BoM) 1979–2004 and changes in average unemployment rate (UR) 1973/79–1980/87 Distribution of unweighted rates of return in manufacturing Distribution of rates of return by invested capital Distributions of corporations in states with size limits Firms by organizational type and capital The hog-corn price ratio, 1870–1940 Plots of raw sum of labor regulations in 1919 and 1899 ESW index, 1899 ESW index, 1919 National labor regulation ESW-indexes, 1899–1924, all labor and manufacturing labor Change in ESW index, 1919–1899 Relationship between the share of wealth owned by the top 1 percent and the Theil index of inequality
9 12 13 15 16 17 21 22 23 58 60 64 65 107 128 134 134 135 135 160
Tables
2.1 2.2 2.3 2.4 2.5 3.1 3.2 3.3 3.4 4.1 4.2 4.3 4.4 4.5 4.6 5.1 5.2 5.3 5.4 5.5 5.6 6.1 6.2
VAR model of births outside marriage (per 1,000 births) and marriages of single women (10,000s), 1849–1913 VAR model of births outside marriage (per 1,000 births) and marriages of single women (10,000s), 1871–1939 First birth rate before marriage: differences by educational group First birth rate before marriage: differences by educational group First union: differences by educational group Waterborne disease death rate regression results Dates of adoption of water treatment techniques Years of high expenditure on water capital Years of high expenditure on sewer capital Average rates of return by region, 1850–1880 T-test for equality between means The growth of general incorporation laws, 1850–1880 Rate of return by implied organizational form and year Capital-labor ratios by firm size and by organizational form, 1850–1880 Probability of bankruptcy and growth among small and large firms, 1850–1880 Railroad access in 1850 and 1860 Treatment effects of gaining rail access: percent urban, output per acre, and improved acres/square mile Factor price regressions: counties with no rail access in 1850: difference-in-difference estimates North–South differences: treatment effects of rail access in 1860, no access in 1850: difference in difference estimates Preliminary IPUMS results for Indiana Preliminary IPUMS results for Illinois Implied percentage change in seasonal values for the hog–corn price ratio Percentage change in U.S. pork production resulting from the adoption of mechanical refrigeration
13 14 24 25 28 35 44–5 47 50 59 60 62 65 68 69 83–4 87–8 89 90 92–3 94–5 112 113
Tables xi 7.1 7.2 7.3
7.4
7.5
8.1 8.2
8.3 8.4 8.5 8.6 8.7
Year of introduction of labor commission, factory inspectors, department of labors, and industrial commissions Number of states with each type of law, 1895, 1908, 1918, and 1924 Cross-sectional correlations of manufacturing real value added per wage earner with measures of labor regulation for years 1919, 1909, and 1899 Cross sectional correlations of changes in manufacturing real value added per wage earner with changes in measures of labor regulation for periods 1919–1909, 1919–1899, and 1909–1899 OLS and fixed effects estimates of coefficients in regressions of the natural log of manufacturing value added in 1967 dollars on the variables listed Summary statistics, 1870 IPUMS and selected sub-samples Average value of property owned, share of property owned by top 1% of wealth holders, and share owning any wealth, by state and region, 1870 Probit estimates of the determinants of property ownership, 1870 OLS estimates of the determinants of the value of property owned, 1870 Within group inequality, selected population groups, 1870 National inequality arising from within and between group inequality, for selected population subgroups, 1870 OLS estimates of determinants of state inequality, 1870
121–2 123–6
137
138
139 149
150–1 154–5 158–9 162 163 166–7
Contributors
Samuel Allen is Assistant Professor at the Virginia Military Institute after completing a two-year post-doctoral position at the University of California at Davis. In 2006, he received the John Heinz award for best dissertation on Social Insurance from the National Academy of Social Insurance for The Economics and Politics of Workers’ Compensation, 1930–2000. Jeremy Atack is Professor of Economics and History at Vanderbilt University and is a Research Associate at the National Bureau of Economic Research. As a graduate student at Indiana University he worked as a research assistant to Fred Bateman on A Deplorable Scarcity, co-authored with Thomas Weiss. Subsequently, he co-authored several papers with Fred Bateman and Weiss on manufacturing profitability and on the diffusion steam power. In addition to numerous articles in journals and book chapters on agricultural development, he has maintained his interest in manufacturing and is currently working on a book on U.S. industrialization with Bateman and Robert Margo. Fred Bateman is the Nicholas A. Beadles Professor of Economics at the University of Georgia. He was previously Professor of Business Economics at Indiana University. With Thomas Weiss he wrote A Deplorable Scarcity and many articles on nineteenth-century manufacturing, collaborating on several NSF grants to collect data samples from the manuscripts of the federal censuses. He co-authored To Their Own Soil with Jeremy Atack along with two chapters on agricultural development in the northern states for the Cambridge Economic History of the United States. They also co-edited the chapter on manufacturing for the Millennial Edition of Historical Statistics of the United States. Louis Cain is Professor of Economics at Loyola University Chicago, Adjunct Professor of Economics at Northwestern University, and Acting Director of Research at the Center for Population Economics at the University of Chicago. His research currently focuses on entrepreneurship in antebellum America, the growth of Chicago before the Great Fire, the urban mortality penalty, and the relationship between urban mortality and expenditures by cities on sanitation.
Contributors
xiii
Lee A. Craig is Alumni Distinguished Professor of Economics at North Carolina State University. He is the author of four books and more than seventy other publications on economic and business history. In addition to his appointment at N.C. State, Dr. Craig has been a research fellow and research economist at the National Bureau of Economic Research in Cambridge, Massachusetts. He is a trustee of both the Economic History Association and the Cliometric Society, and he is a former fellow of the Center for Demographic Studies at Duke University and the Seminar für Wirtschaftsgeschichte, Universität München. John Ermisch is a professor of economics at the Institute for Social and Economic Research, University of Essex and a Fellow of the British Academy. Formerly, he was Bonar-Macfie Professor in the Department of Political Economy at the University of Glasgow (1991–94), and a senior research officer at the National Institute of Economic and Social Research. In addition he has been co-editor of Journal of Population Economics and is a past President of the European Society for Population Economics. He is the author of three books and of numerous articles in economic and demographic journals. Price Fishback is the Frank and Clara Kramer Professor of Economics at the University of Arizona and is a Research Associate at the National Bureau of Economic Research. His book with Shawn Kantor, A Prelude to the Welfare State: The Origins of Workers’ Compensation in 2000 received a Paul Samuelson Certificate of Excellence from the TIAA-CREF Institute and the Richard A. Lester Prize from the Industrial Relations Section at Princeton University. Michael R. Haines is Banfi Vintners Distinguished Professor of Economics at Colgate University (Hamilton, NY) and is a Research Associate at the National Bureau of Economic Research. He currently serves on the Council of the Interuniversity Consortium for Political and Social Research. He is a past Vice-President and President of the Social Science History Association and currently its treasurer. He is the author of three books and several dozen articles and chapters in books and professional journals; and has served on the editorial boards of the Journal of Economic History and Explorations in Economic History. Rebecca Holmes is the Chief Economist at the Salt River Project electric utility in Phoenix, Arizona. She received the Alan Nevins Prize for best dissertation in North American Economic History in 2004 for her work on The Impact of State Labor Regulations on Manufacturing Input Demand During the Progressive Era. Matthew T. Holt is Professor and Wickersham Chair of Excellence in Agricultural Research in the Department of Agricultural Economics at Purdue University. He is currently on the Board of Directors of the American
xiv
Contributors
Agricultural Economics Association, He has also served as Associate Editor for the American Journal of Agricultural Economics and the Journal of Business and Economic Statistics. He has published extensively on topics as far ranging as the role of risk in agricultural production, specification, and estimation of inverse demand systems, and the role of nonlinear dynamics in agricultural and commodity markets. Robert A. Margo is Professor of Economics and African-American Studies at Boston University, and a Research Associate of the National Bureau of Economic Research. He is the editor of Explorations in Economic History, and has served on the editorial boards of many journals, including the Journal of Economic History, the American Economic Review, and the Quarterly Journal of Economics. He is the author or co-author of four books and over one hundred articles, book chapters and book reviews. Elyce Rotella is Professor of Economics at Indiana University. Her research currently focuses on the economic history of women in the American labor force, on borrowing and spending by ordinary people in the past, and on the relationship between urban morality and expenditures by cities on sanitation. Joshua L. Rosenbloom is Professor of Economics and Associate Vice Provost for Research and Graduate Studies at the University of Kansas. He is also a Research Associate of the National Bureau of Economic Research. He is currently collaborating with Thomas Weiss and Peter C. Mancall on a project to construct detailed regional estimates of the pace of economic growth in North America before 1800. He is the author of one book and over 40 articles on different aspects of American economic history. He is a past member of the board of Editors of the Journal of Economic History. Gregory W. Stutes is Associate Professor of economics at Minnesota State University in Moorhead, where he also serves as the director of the Center for Economic Education. He completed his dissertation at the University of Kansas in 2004 on the causes and consequences of the distribution of real estate wealth in nineteenth-century America.
Preface
This volume had its origins in a conference held in April 2006 at the University of Kansas, in Lawrence, Kansas, in honor of Thomas Weiss. The participants in the conference included several of Weiss’s students and many of his research collaborators. Along with Weiss they are among the leading proponents of the value of quantitative approaches to understanding economic and social history. Thomas Weiss, Tom to his friends, developed a calculating eye on the golf courses in and around his native Poughkeepsie, New York. Finding the angles on a golf course not all that different from those in the principles of economics, Tom leveraged his skills on the links into a scholarship to Holy Cross. After college, Tom gave up on his PGA Tour hopes and moved on to the University of North Carolina at Chapel Hill, completing his doctorate in 1967.1 At Chapel Hill, Tom met his friend and long-time collaborator, Fred Bateman, who had come there to work with William Parker. Although Tom missed the opening shots of the Cliometric Revolution, he witnessed and participated in many of its battles.2 The Cliometric approach employs quantitative evidence, informed by economic theory, to better understand the historical circumstances from which that evidence was generated. During his career, Weiss has embodied the twin ideals of careful attention to historical context and the innovative use of quantitative evidence to advance our understanding of the development of the American economy. Following in the footsteps of his mentor, Robert Gallman, Weiss has been a leading contributor to the development of a diverse set of data bases upon which our understanding of American economic development rests. These contributions began with his dissertation, which undertook the measurement of capital, labor, and output in the U.S. service sector between 1840 and 1900. This work remains the benchmark against which other similar histories are measured. Weiss’s starting point was the vast array of data collected by the United States Census. However, the categories used and the questions asked by nineteenth-century government officials rarely fit neatly into the conceptual frameworks developed by twentieth-century economists to classify and measure economic activity, in particular the national income and product accounts developed by Gallman’s mentor, Simon Kuznets. The development of approaches to fill in missing categories of data and to bring consistency across
xvi
Preface
the measurements made at different dates requires attention to detail and a great deal of imagination, qualities that are both abundantly evident in Weiss’s work. Building upon his work on the service sector, Weiss took an interest in measuring labor inputs more broadly. He subsequently made significant contributions to the measurement of the U.S. labor force and its sectoral distribution in the nineteenth century. Building on work begun by Stanley Lebergott, Weiss carefully re-examined census data to more accurately apportion laborers between different sectors. His revision of earlier estimates led to a reinterpretation of nineteenth-century U.S. growth. Specifically, he found that laborers had been over counted in the agricultural sector, especially earlier in the century. Thus, re-allocating these workers from agriculture to other sectors tended to increase estimates of agricultural labor productivity early in the century, with the result that subsequent growth rates were lowered. This finding in turn revised thinking on the pace of U.S. agricultural advancement. While such painstaking research does not always win headlines, getting the details right is important because these basic series underlie so many other accounts. In addition to his contributions to the development of the statistical underpinnings of our understanding of American economic development, Weiss has also contributed to the debate about how to interpret that development. With Bateman, he collected one of the first micro-samples drawn from the manuscripts of nineteenth-century manufacturing censuses.3 The rich cross-sectional evidence on the characteristics of a representative sample of manufacturing establishments allowed Bateman and Weiss to offer an insightful analysis of the differential rates of industrialization in the antebellum North and South and to shed new light on the impacts of slavery on the course of southern economic development. Bateman and Weiss carved a careful theoretical and empirical path between scholars who argued antebellum southerners were irrational in their devotion to an agrarian way of life, and those who saw an efficient allocation of resources via regional comparative advantage. Subsequently, in conjunction with his long-time University of Kansas colleague, Joshua Rosenbloom, and two of Bateman’s students, Jeremy Atack and Lee Craig, Weiss has written on a wide range of topics including, but hardly limited to, agricultural labor productivity, nutrition, mortality, U.S. capital markets, the diffusion of steam power, and colonial economic growth. More recently, Weiss has turned his attention to the economic history of tourism, reflecting his tastes in leisure, which run toward travel, fine wine, and good food. Weiss’s contributions to economic history extend well beyond his own research. As a teacher and colleague he has inspired and taught other scholars in diverse ways. Many of his students at the University of Kansas recall his courses as the most demanding and rewarding part of their studies. No matter whether he was teaching introductory economics or graduate economic history, Weiss managed to find ways to engage the students, to teach them about the application of economics to real world problems, and to illustrate the insights of economic theory and history. His reputation as a classroom teacher was unequaled in his
Preface xvii years at Kansas. Once, when Weiss, Craig, Rosenbloom, and a few other Cliometricians, who shall remain nameless for their own good, were in a bar in Oxford, Ohio, they managed to join the members of a woman’s volleyball team visiting from another campus. Incredibly, one of the young women knew of Weiss’s classroom reputation from a friend of hers who had taken one of his classes. The high standards of his scholarship and the generosity of his spirit have led to many honors and recognitions. Thomas Weiss has served as editor of the Journal of Economic History, as Executive Secretary and President of the Economic History Association, and as a trustee of the Cliometric Society. He has been a long time Research Associate of the National Bureau of Economic Research, and was one of the first recipients of the famed Clio “can.” All those who have had the opportunity to work closely with Tom know how fortunate they have been. His ability to combine exacting standards of scholarship with an irrepressible sense of fun, and an enjoyment of good food and drink, and exotic travel set an example to which all scholars might aspire.
Notes 1 Weiss describes his decision to pursue economic history in “An Interview with Thomas Weiss,” The Newsletter of the Cliometric Society (Summer 2002, vol. 17, no. 2: 3–4) 2 A complete listing of the Weiss’s publications can be found at the end of this volume. 3 James Foust was also involved in the collection of these data, but not in their subsequent analysis. Jeremy Attack, a student of Bateman’s, was also involved in this project, initially as a graduate student, and has subsequently extended this data collection effort to include a number of post-bellum manufacturing censuses as well.
Acknowledgments
I wish to thank Jeremy Atack, Fred Bateman, Lou Cain, Lee Craig, John Ermisch, Price Fishback, Michael Haines, and Bob Margo whose role in producing this volume extends far beyond the chapters they have contributed. Their encouragement, advice, and editorial assistance have been invaluable. I would also like to thank Joseph Sicilian for his help planning the conference held at the University of Kansas, and Leanea Wales for taking care of all the logistical issues surrounding the conference. Finally, I wish to express my appreciation to the Department of Economics at the University of Kansas and Chancellor Robert Hemenway for assistance with conference expenses.
1
Editor’s introduction The good of counting Joshua L. Rosenbloom
BOSWELL: Sir Alexander Dick tells me, that he remembers having a thousand people in a year dine at his house: that is, reckoning each person as one, each time that he dined there. JOHNSON: That, Sir is about three a day. BOSWELL: How your statement lessens the idea. JOHNSON: That, Sir is the good of counting. It brings every thing to a certainty, which before floated in the mind indefinitely. BOSWELL: But Omne ignotum pro magnifico est: one is sorry to have this diminished. JOHNSON: Sir, you should not allow yourself to be delighted with error. (Boswell’s Life of Johnson (Oxford University Press ed.: Oxford 1934): IV, p. 204 (18 April 1783, Aetat. 74))
One of the principal achievements of the cliometric revolution is the approach to quantitative economic history that it spawned (McCloskey 1978). Economic historians have, of course, always relied upon quantitative evidence, but beginning in the 1950s cliometricians began to analyze quantitative data in new ways. Most importantly they explicitly acknowledged the connection between economic theory and quantitative evidence. Theory not only provided guidance about what to measure, but also how to measure it, and in turn, then, suggested entirely new types of evidence that had not previously been subjected to careful analysis. Even now, after more than half a century, the field of quantitative economic history continues to yield new insights about the past. One important contribution of quantitative economic history has been the extension and refinement of measures of national income for periods before the formation of modern government statistical agencies. The reconstruction of these historical time series describing the pace and pattern of economic growth provides an essential foundation and consistency check for narrative accounts and more narrowly focused histories of particular sectors or industries. Not only does theory provide a framework for assembling readily available statistical information about the past, but it has proved essential in working out methods to bridge gaps in the historical record through creative reconstruction.1 In addition to providing a basis for reconstructing the quantitative dimensions
2
J.L. Rosenbloom
of economic growth the Cliometric approach has made accessible new kinds of historical evidence that were simply beyond the scope of traditional historians’ consideration. For example, scholars considering the role President Andrew Jackson’s “War” on the Second Bank of the United States played in the depression of 1837 had long relied on the arguments of various knowledgeable contemporaries to support their competing interpretations. By turning to data on changes in the country’s stock of monetary gold, however, quantitative economic historians were able to cut through these arguments and convincingly resolve the debate by showing that the causes of the depression arose largely from changes in international markets rather than any action of the President (Temin 1969; Rockoff 1971). In much the same, way quantitative economic historians mobilized a vast array of new data to re-examine debates about the economics of slavery. Despite the controversy that some of their work initially elicited, the profitability and economic rationality of the slave system was conclusively established through the use of a wide array of statistical sources that had not previously been examined.2 The point is not that quantitative economic history is in some sense better than other approaches to the past. Rather, by mobilizing the methods of the social sciences, quantitative economic historians have opened up new ways of looking at the past that complement and extend other more traditional approaches. By focusing on collective behavior and developing techniques to describe both central tendencies and variations around them quantitative economic historians are able to pose and answer questions that are not readily accessible to more traditional historians (Fogel 1983: 29). While traditional historical approaches focus on the particular, quantitative economic history provides a way to set individual experience in a broader context, to assess the representativeness of individual experience and to explore systematic patterns of variation in data that cannot be discovered from case studies alone. New ways of looking at the past have also encouraged the collection of new types of quantitative evidence. Since the 1950s quantitative economic historians have begun to systematically collect and analyze a variety of data sources outside the conventional archives of the historian. Beginning with the work of William Parker and Robert Gallman, who collected data from a sample of farms across the Cotton South in the 1860 census of population and linked these to data drawn from the census of agriculture, quantitative economic historians have collected and made available computer readable samples from a vast array of manuscript censuses.3 The Integrated Public Use Microdata Series based at the University of Minnesota now makes available to scholars large, nationally representative samples of every extant U.S. population census since 1850, as well as a growing collection of international data.4 Samples of a number of nineteenth-century manufacturing censuses are also now available. The initial work on these sources was undertaken by Fred Bateman and Thomas Weiss (1981), who collected samples from the manufacturing censuses of 1850, 1860, and 1870. This work has been continued and extended by Jeremy Atack, who has added samples from 1880.5
Editor’s introduction
3
Like the field of quantitative economic history, the essays in this volume are unified by a common set of methods rather than by a particular chronological or topical focus. Many of them take advantage of large data samples to examine aspects of the past. Several describe insights derived from efforts to collect new sources of data, or to combine existing data sources to yield new insights. What all of them illustrate are the ways in which these methods continue to offer new insights about the past. This volume begins with two essays addressing the intersection between economic and demographic history, one of the areas in which quantitative approaches have made great contributions to understanding historical forces that cannot be discerned from conventional documentary sources, as Weiss’s work on U.S. mortality rates indicates (Haines et al. 2003). In Chapter 2, John Ermisch examines how the economic environment and technology affected childbearing outside marriage in England and Wales since 1845. In it he finds that, up to World War I, higher unemployment rates discouraged marriage and increased non-marital birth rates, which is consistent with poorer labor market conditions discouraging marriages among pregnant would-be brides, thereby increasing bastardy. During the inter-war period, higher unemployment continued to produce postponement of marriages, but non-marital births were no longer linked to unemployment. Free access by unmarried women to the contraceptive pill after 1975 and the legalisation of abortion dramatically altered the environment for decisions concerning marriage and childbearing, causing the postponement of marriage and the rise in cohabiting unions. The consequent rise in the unmarried population accounts for most of the explosion in the proportion of births outside marriage, reaching 42 percent of births in 2004. Ermish argues that non-marital childbearing had been discouraged by strong social stigma against it, but a temporary change in the determinants of non-marital childbearing that raised it, like the large rise in unemployment in the early 1980s, produced rapid erosion of the stigma and a self-reinforcing rise in non-marital births. In Chapter 3 Louis Cain and Elyce Rotella focus on deaths rather than births, exploring the factors that influenced the timing of the dramatic improvements in public health that were accomplished through investments in the provision of water and sewage treatment in U.S. cities. During the first 30 years of the twentieth century, American cities experienced dramatic declines in deaths from diseases associated with bad water and poor sewage disposal. Cities bought this reduction by spending large amounts on sanitation systems. Cain and Rotella estimate that a 1 percent increase in sanitation expenditures was associated with a 3 percent decline in the mortality rate from typhoid fever, diarrhea, and dysentery. In the nineteenth century, sanitation was largely supplied privately by the service sector initially explored by Weiss (1975). With continued population and economic growth, these critical services began to move into the public sector. This chapter examines the role that epidemics and demonstration effects played in causing cities to undertake these expenditures. Empirical examination of epidemics reveals that cities that suffered an episode of increased waterborne
4
J.L. Rosenbloom
mortality often responded with higher spending. Demonstration effects took place in a number of ways over time ranging from prominent engineers visiting sanitation works in the middle of the nineteenth century to formal demonstrations projects constructed on a large scale in the years just before the Great Depression. Cain and Rotella argue that epidemics had their biggest impact on the demand for sanitation services, while demonstration effects had their biggest impact on the supply of sanitation services. Together these forces created the conditions whereby American cities allocated the funds that dramatically reduced death rates. The next three chapters focus on the transformation of the American economy that took place in the late nineteenth and early twentieth centuries. In Chapter 4, Weiss’s long-time collaborators Jeremy Atack and Fred Bateman pursue a natural extension of their earlier work with Weiss on the profitability of nineteenth-century manufacturing. Using expanded samples drawn from the manuscripts of the censuses of manufactures for 1850, 1860, 1870, and 1880, they examine the relationship between firm size and profitability. Many nineteenth-century populists appear to have assumed that larger firm size was synonymous with monopoly power and associated with higher profitability, a belief that played some role in the rise of government regulation intended to offset the perceived advantages of the concentration of economic power. Nonetheless empirical evidence from the twentieth century has generally revealed an inverse relationship between firm size and profitability and Atack and Bateman demonstrate that this same relationship also held in the nineteenth century as well. Moreover, they show that firm size and organizational structure were closely intertwined, with larger firms being much more likely to be organized as corporations rather than sole-proprietorships or partnerships. If profits were decreasing with firm size, it is natural to ask why was average firm size rising at this time? Atack and Bateman argue that two factors help to rationalize this pattern; first, larger businesses enjoyed a lower cost of capital, reflected in their higher capital–labor ratios, and attributable to their access to impersonal capital markets and, second, rates of return for larger firms and corporations were far less volatile than those of smaller firms or those with different organizational structures, so they were much more likely to persist over time In Chapter 5 Robert Margo and Michael Haines revisit a topic addressed by Weiss in a 1998 paper with Lee Craig and Raymond Palmquist. Craig, Palmquist and Weiss (1998) collected county level data on access to rail and water transportation for 1850 and 1860, and used these data to study the impact of such access on agricultural land values. Margo and Haines extend this previous work in two ways. First, they shift attention to those counties that gained rail access in the 1850s; that is, rather than relying on the cross-section correlation between the presence of the railroad and economic outcomes, they seek to measure the “treatment effect” of the railroad. Second, they consider a broader range of economic outcomes from gaining rail access that are predicted from standard economic models of transportation improvements, such as the Von Thunen model – not just land values as examined by Craig, Palmquist, and Weiss. The results of their
Editor’s introduction
5
investigation are mixed: although some predictions based on the models they use are confirmed, others are not. Perhaps their strongest conclusion is that antebellum railroads promoted urbanization and the growth of the service sector, which also links this paper with Weiss’s previous research. In Chapter 6 Matthew Holt and Lee Craig look at a topic that has long been of interest to historians and economists, namely the hog cycle. The paper elaborates on issues raised in Weiss’s research with Craig and Michael Haines (Craig and Weiss 1993, 1998, 2000; Haines et al. 2003). One of the themes of this body of work was how the markets for agricultural products allocated nutrients to non-farm households. In terms of both the inputs dedicated to their production and the value of their output, corn and hogs played a large role in the farm economy and non-farm consumption. Furthermore, because of the interrelated nature of the two markets – the majority of the corn crop was fed to hogs – the swings in their respective prices were not independent of one another, and the relationship became known as the hog cycle. Craig and Holt address three questions related to the cycle: How were hog cycles propagated? How were they eventually ameliorated? What was the social savings from their amelioration? The authors argue that cycles were propagated by the combination of weather shocks to the corn market and the inability of hog farmers to smooth production in response to the resulting feed price shocks. The seasonal cycle was ameliorated by the advent of the mechanical refrigeration, which allowed farmers to both smooth and increase production. The increase in production represented a social saving. Craig and Holt estimate this savings was in the neighborhood of 0.17 percent of aggregate output, which in turn corresponded with a one to two percent increase in food consumption measured by calories or protein. In the last two chapters of this book the focus shifts to issues related to work and welfare. In Chapter 7 Rebecca Holmes, Price Fishback and Samuel Allen break new ground in developing techniques to measure the extent of regulation in labor markets during the late nineteenth and early twentieth centuries. The hallmark of Weiss’s research has been his interest in developing the best way to measure economic concepts. He has carefully built on the work of prior scholars to construct measures of the labor force, wages, Gross National Product, and productivity. All of these concepts are based on economic theory, but it turns out that it is often difficult to measure precisely the theoretical construct. Weiss has led the way in trying to improve measurement of these concepts. The Holmes, Fishback, and Allen chapter follows in that same vein, offering multiple ways of aggregating state labor legislation into a general measure of the extent of labor regulation at the state level. After a detailed exploration of various approaches to aggregating multiple dimensions of regulation, they use their newly created measure of labor regulation to try to establish the relationships between regulation and labor productivity in manufacturing. Proponents and opponents of the labor legislation offered different views of how the regulation would influence labor productivity. Holmes, Fishback, and Allen offer a first examination of which of these two groups were correct, and conclude that the regulation did not retard labor productivity growth.
6
J.L. Rosenbloom
In the final chapter of the book Joshua Rosenbloom and Gregory Stutes study the distribution of wealth in the United States in 1870 using data from the 1870 population census available in the Integrated Public Use Microdata Series (IPUMS). Economic historians have had a long-standing interest in the relationship between economic development and inequality, and the recent widening of income and wealth disparities in the United States has heightened this interest among a broader audience. Rosenbloom and Stutes are not the first scholars to examine the 1870 census data. Indeed, Weiss himself employed the data in his research with Bateman (1981) and Craig (1993), but the large size of the IPUMS sample enables Rosenbloom and Stutes to look at these data in considerably more detail than has been possible in previous studies. On the strength of these data they argue that it appears likely that increasing industrialization and urbanization were linked to rising inequality. In addition to documenting a substantial increase in wealth inequality between 1870 and the early twentieth century they find that cross-state variations in inequality in 1870 are in large part explained by differences in the level of urbanization and manufacturing employment. The chapters in this volume do not constitute a comprehensive overview of quantitative economic history, but they do illustrate the breadth of topics and innovative research methods that characterize recent work in this field. Each of these essays reflects the interplay between theory and measurement that is the hallmark of the Cliometric approach to history and the hallmark of Weiss’s extensive contributions to the field. Theory provides the starting point and motivation for each essay and serves to guide decisions about what to measure and how measurements will be made. But theory is carefully aligned with an understanding of the sources that are available and the kinds of measurement that they will support. In this conjunction of sound historical methods and economic theory these essays carry forward the approach that Weiss has exemplified throughout his academic career.
Notes 1 Among the earliest efforts to reconstruct national income is Gallman (1966). See also David (1967) and Weiss (1994). More recently Romer (1986) and Gordon and Balke (1989) have undertaken efforts to refine some of the methods used in these earlier works. 2 There is a vast literature on this topic. See Fogel (1989) for a comprehensive overview of this literature and the many debates it spawned. 3 See Parker (1970) for a discussion of these data. 4 These data have been compiled by the Minnesota Population Center under the direction of Steven Ruggles. See Ruggles et al. (2004) and www.ipums.org for additional information and documentation. 5 These can be downloaded from his website www.vanderbilt.edu/econ/faculty/ Atack/atackj.htm.
References Bateman, Fred and Weiss, Thomas (1981) A Deplorable Scarcity: The Failure of Industrialization in the Slave Economy, Chapel Hill: University of North Carolina Press.
Editor’s introduction
7
Craig, Lee A. and Weiss, Thomas (1993) “Agricultural Productivity Growth During the Decade of the Civil War”, Journal of Economic History 53: 527–48. Craig, Lee A. and Weiss, Thomas (1998) “Nutritional Status and Agricultural Surpluses in the Antebellum United States”, in Studies on the Biological Standard of Living in Comparative Perspectives, John Komlos and Joerg Baten (eds), Stuttgart: Franz Steiner Verlag. Craig, Lee A. and Weiss, Thomas (2000) “Hours at Work and Total Factor Productivity Growth in 19th-Century U.S. Agriculture”, Advances in Agricultural Economic History 1: 1–30. Craig, Lee A, Palmquist, Raymond and Weiss, Thomas (1998) “Transportation Improvements and Land Values in the Antebellum United States: Hedonic Approach”, The Journal of Real Estate Finance and Economics 16: 173–90. David, Paul A. (1967) “The Growth of Real Product in the United States Before 1840: New Evidence, Controlled Conjectures”, Journal of Economic History 27: 151–97. Fogel, Robert William (1983) “ ‘Scientific’ History and Traditional History”, in Robert William Fogel and G.R. Elton, Which Road to the Past? Two Views of History, New Haven and London: Yale University Press. Fogel, Robert William (1989) Without Consent or Contract: The Rise and Fall of American Negro Slavery, New York and London: W. W. Norton. Gallman, Robert E. (1966) “Gross National Product in the United States, 1834–1906”, in Dorothy Brady (ed.), Output, Employment and Productivity in the Unites States after 1800, Studies in Income and Wealth, vol. 30, New York: National Bureau of Economic Research. Gordon, Robert J. and Balke, Nathan S. (1989) “The Estimation of Prewar Gross National Product: Methodology and New Evidence”, Journal of Political Economy 97: 38–92. Haines, Michael R., Craig, Lee A. and Weiss, Thomas (2003) “The Short and the Dead: Nutrition, Mortality, and the ‘Antebellum Puzzle’ in the United States”, Journal of Economic History 63: 382–413. McCloskey, Donald. N. (1978) “The Achievements of the Cliometric School”, Journal of Economic History 38: 13–28. Parker, William N. (1970) “Introduction: The Cotton Economy of the Antebellum South”, Agricultural History 44: 1–4. Rockoff, Hugh (1971) “Money, Prices and Banks in the Jacksonian Era,” in The Reinterpretation of American Economic History, Robert Fogel and Stanley L. Engerman (eds), New York: Harper & Row, 1971: 448–58. Romer, Christina (1986) “New Estimates of Prewar Gross National Product and Unemploment”, Journal of Economic History 46: 341–52. Ruggles, Steven, Sobek, Matthew, Alexander, Trent, Fitch, Catherine A., Goeken, Ronald, Hall, Patricia Kelly, King, Miriam, and Ronnander, Chad (2004) Integrated Public Use Microdata Series: Version 3.0 [Machine-readable database], Minneapolis, MN: Minnesota Population Center [producer and distributor]. Temin, Peter (1969) The Jacksonian Economy, New York and London: W. W. Norton. Weiss, Thomas (1975) The Service Sector in the United States, 1839 through 1899, New York: Arno Press. Weiss, Thomas (1994) “Economic Growth Before 1860: Revised Conjectures”, in Thomas Weiss and Donald Schaeffer (eds), American Economic Development in Historical Perspective, Stanford: Stanford University Press.
2
An economic history of bastardy in England and Wales John Ermisch*
Four centuries of bastardy in England A remarkable feature of English demographic history is the explosion in childbearing outside marriage during the last quarter of the twentieth century, after 400 years of relative stability. Figure 2.1 plots the number of live births outside marriage per 1,000 live births for England and Wales between 1845 and 2003. With the exception of a ‘spike’ during the World War II, the percentage of births outside marriage stayed within the range of 3.9 to 7.0% between 1845 and 1960. In the late sixteenth century it was about 3%, fell to 1% in the mid-seventeenth century, and then rose to about 6% at the beginning of the nineteenth century (Laslett 1977: Fig. 3.2). Thus, over 400 years, the percentage moved within a relatively narrow range. This chapter focuses on the last 160 years, particularly the period since 1870. It aims to explain fluctuations in non-marital childbearing in England and Wales during the period of relative stability and the explosion in the last quarter of the twentieth century. Figure 2.1 also shows the comparable non-marital birth ratio for the USA since 1940. It is a little lower than the English/Welsh ratio from 1940–55, but shows a steady increase from the early 1960s to the early 1990s. The increase in the English/Welsh ratio is much more rapid, is concentrated in the period 1980–2003, and eventually overtakes the American ratio in the 1990s. Background: fertility, birth control and social stigma, 1845–1940 Figure 2.1 also shows the evolution of fertility since 1845, as measured by the General Fertility Rate (GFR). Fertility began its steep long-term fall in the second half of the 1870s, reaching a low point just before the World War II. As in many other European countries, non-marital birth rates (non-marital births relative to unmarried women, standardised by age) fell in parallel with marital birth rates until the late 1930s (Shorter et al. 1971). It is important for the analysis of non-marital childbearing to consider how fertility was controlled to produce the long fertility decline. Szreter (1996: 398–9) argues convincingly that ‘attempted abstinence within marriage was the single most widespread and frequently used method of birth
450
180
400
160
350
140
300 250
GFR-E&W Right scale
120 100
200
80
150
60
100 50
40 bom-USA Left scale
20 0
1845 1852 1859 1866 1873 1880 1887 1894 1901 1908 1915 1922 1929 1936 1943 1950 1957 1964 1971 1978 1985 1992 1999
0
bom-E&W Left scale
9
Per 1,000 women
Per 1,000 births
An economic history of bastardy
Figure 2.1 Births outside marriage (per 1,000 births) and general fertility rate (per 1,000 women aged 15–44).
control’ during the long decline, particularly before the World War I. He identifies (p. 420): conscious, attempted abstinence to restrict births as the main cause of reduced coital frequency in the late Victorian and Edwardian period . . . a public discourse explicitly promoting the virtues of sexual continence, primarily on moral and health grounds was, in fact, consciously developed and elaborated during this period. Birth control during the inter-war years was similar to the preceding 50 years. Low marital birth rates in 1930s were primarily attained through intended low coital frequencies – the reported levels of birth control use (even if underreported) and the low effectiveness of these methods cannot account for the low birth rate. A culture of sexual restraint supported abstinence as the main control on marital fertility and kept births outside marriage low. While knowledge about contraception improved during the inter-war years, the unmarried had difficulty obtaining such information. At least as far back as the seventeenth and eighteenth centuries, childbearing outside marriage in England was associated with courtship. The fact that premarital childbearing tended to be higher when marriage ages were lower, and that the age at first ‘illegitimate’ birth was approximately equal to the age at first marriage (Laslett 1980; Oosterveen et al. 1980), suggest that it was part of the courtship process. In times when general marriage opportunities were good there would be more courtship, hence more sexual activity, and more risk of nonmarital births when, for a number of reasons, a particular marriage failed to take place. Wrightson (1980: 190) interprets the proportion of births outside marriage in seventeenth-century England as ‘an index of the degree of disjunction
10
J. Ermisch
between socially acceptable premarital sexual activity and particular marital opportunities’. Premarital sex continued to be a part of courting behavior among working-class people throughout the nineteenth century. As Cook (2004: 17) explains, individual caution backed up by community sanctions stopped most couples from marrying unless they had sufficient savings and income to support a new separate household containing wife and children. Premarital sexual intercourse was regulated by this system. Most unmarried women would not have sexual intercourse except with a partner who had agreed to marriage, and the man would not make this offer until he could afford to do so. Note that this is not the same as delaying intercourse until marriage was imminent. . . . There was probably considerable leeway if the woman did not become pregnant. Cook (2004: 105–6) contends that women’s increasing sexual caution and diminishing opportunity to relaxed premarital sexual activity (enforced by adult control) spread from the middle classes to the working classes, and that this contributed to the decline in the non-marital birth rate. Nevertheless, premarital sex remained an important part of courting behavior among working-class people into the early twentieth century. Having intercourse marked a point in courtship, a staging post en route to marriage and household formation. Despite this, premarital births not quickly followed by marriage to the father brought extreme disapproval from parents and neighbors, and sometimes abandonment by the mother’s parents, often because they were too poor to support her. If her family abandoned her, she had to fall back on one of a number of harsh institutions, the most infamous and dreaded of which was the workhouse. According to Humphries (1988: 87), the workhouse was still dealing with about one-fifth of all non-marital births in the early twentieth century, and more in the Victorian era. Other institutions had a variety of names – reformatories, penitentiaries, refuges, maternity homes and ‘mother and baby’ homes – what they had in common was tough regimes in which young mothers lost all personal rights, freedom and privacy. The fear engendered by these institutions must have provided a strong reinforcement of the social stigma associated with childbearing outside marriage. The next section demonstrates that, up to World War I, fluctuations in the percentage of births outside marriage responded to the unemployment rate. Higher unemployment discouraged marriage and increased non-marital births, which is consistent with poorer labor market conditions discouraging marriages among pregnant would-be brides, thereby increasing bastardy. This is further support for the hypothesis that non-marital childbearing was associated with courtship. During the inter-war period, higher unemployment continued to produce postponement of marriages, but non-marital childbearing was no longer linked to unemployment. The remainder of the chapter considers the period after World War II. An
An economic history of bastardy
11
important technological change dominates this period: the introduction of the contraceptive pill in 1961. The third section shows that the increase in the percentage of births outside marriage from 9% in 1975 to 42% in 2003 is mainly accounted for by a large fall in the proportion of women aged 20–34 who are married. This is in turn associated with a dramatic rise in cohabiting unions. As these unions are short-lived before either dissolving or being converted into marriage, non-marital childbearing appears, therefore, to be associated with modern courtship. The fourth section presents a theory of marriage market search (courtship) in which out-of-wedlock childbearing may be a rational choice, and the fifth section explains how it can become widespread when a woman’s welfare as a single mother is influenced by the prevalence of single mothers in the population. A possible catalyst for the large increase in nonmarital childbearing throughout Europe after 1980 may have been the substantial increase in unemployment in the first half of the 1980s. The sixth section presents evidence that socio-economic differences in the chances of having a child before marriage widened as non-marital childbearing became more common. The seventh section discusses the diffusion of cohabiting unions from the better educated to the rest of the population, which may provide an alternative or complementary explanation for the explosion of non-marital childbearing, and the eighth section presents conclusions.
Unemployment and bastardy before World War II A simple theory of how bastards are born can be based on two premises that are supported by English socio-demographic history, outlined in the first section. The first is that getting married meant establishing a new household, which required sufficient earnings and reasonably good economic prospects. The second is that premarital sex was not uncommon, particularly when both partners had the intention of marrying one another at the time of the sexual act, and contraception was usually not practised and inefficient when it was. For instance, there is evidence that in rural England in the late eighteenth and nineteenth century 30–40% of brides had a birth within eight months of marriage (Hair 1970: Table 1), consistent with pregnancy at marriage for a substantial proportion of brides. For comparison, soon after World War II (1951–55), registration statistics indicate that 18% of first births were born within eight months of marriage. If for some reason the marriage planned by a pregnant woman did not take place, then the child would be born outside marriage. Death of the father-to-be is a clear example, and helps explain the ‘spikes’ in the non-marital birth ratio and the number of marriages of single women during World War I shown in Figure 2.2. Thus, a change in the environment that reduced marriages would also increase the birth rate outside marriage. We can examine the consistency of the historical experience with this simple theory by exploiting the long series on marriages of single women. Data on the stock of single women, which is required to convert the marriages into a marriage rate, is not available for the
12
J. Ermisch 80 70 60 50 40 30 20
0
bom Marriages 1845 1848 1851 1854 1857 1860 1863 1866 1869 1872 1875 1878 1881 1884 1887 1890 1893 1896 1899 1902 1905 1908 1911 1914 1917 1920 1923 1926 1929 1932 1935 1938
10
Figure 2.2 Births outside marriage (per 1,000 births) and marriages of single women (10,000s).
entire period. But as the stock is likely to move slowly, this is not a major problem. Over the period 1930–2002, the number of marriages to single women and the marriage rate of single women move together closely – their correlation coefficient is 0.844, and the correlation coefficient between the first differences of the two variables is 0.963. Focussing on the period before the disruption of the world wars, 1845–1913, neither live births outside marriage per 1,000 live births (bom), nor the marriages of single women are stationary series, both are trended as Figure 2.2 shows.1 Their first differences are, however, stationary. Estimates of a vector auto-regression (VAR) involving these two differenced endogenous variables are shown in Table 2.1. More marriages of single women reduce the proportion of births outside marriage in subsequent years, and an increase in the proportion of births outside marriage subsequently increases marriages of single women. Each variable ‘Granger-causes’ the other.2 The negative impact of marriages on bom one year later is consistent with the simple theory’s prediction that shocks that reduce marriages increase births outside marriage. The positive effect of bom on subsequent marriages of single women may reflect postponed marriages of women who had become single mothers. One such shock may be poorer labor market conditions, as indicated by a higher unemployment rate. Southall and Gilbert (1996) present evidence that during the period 1860–1914 there were fewer marriages when unemployment (reported by various unions operating unemployment insurance schemes) was higher. Whether unemployment also affected births outside marriage is investi-
An economic history of bastardy
13
Table 2.1 VAR model of births outside marriage (per 1,000 births) and marriages of single women (10,000s), 1849–1913*
bomt–1 – bomt–2 bomt–2 – bomt–3 mart–1 – mart–2 mart–2 – mart–3 constant Granger causality: p value R2
bomt – bomt–1
mart – mart–1
0.12 (0.12) 0.24 (0.12) –0.64 (0.24) –0.31 (0.22) –0.01 (0.14) 0.003 0.22
0.17 (0.06) 0.079 (0.064) 0.23 (0.12) –0.11 (0.11) 0.281 (0.07) 0.002 0.24
Note *Standard errors in parentheses.
gated here for the period 1870–1913 using the new estimates of British unemployment compiled by Boyer and Hatton (2002). Their estimate of the unemployment rate (ur), bom and marriages to single women are plotted in Figure 2.3. The Dickey Fuller test indicates that ur is a stationary series during the period 1870–1913, and it will be assumed that ur is strictly exogenous. Estimation of a vector autogression (VAR) with first differences in bom and marriages being endogenous and ur being exogenous indicates that lags of the endogenous variables do not have significant effects. Thus, the model can be simplified to that shown in the first two columns of Table 2.2. The estimated parameters indicate that a higher unemployment rate raised births outside marriage. It also initially reduced marriages to single women, but there was a strong recovery in 18 16
60
14 50 12 40
10
30
8
Percent
BoM per 1,000 births and 10,000s of marriage
70
6 20 4 10
Marriage BoM Unempr
0
18
7 18 1 7 18 4 7 18 7 8 18 0 8 18 3 8 18 6 8 18 9 9 18 2 9 18 5 9 19 8 0 19 1 0 19 4 0 19 7 1 19 0 1 19 3 1 19 6 1 19 9 2 19 2 2 19 5 2 19 8 3 19 1 3 19 4 3 19 7 40
0
2
Figure 2.3 Marriages of single women, births outside marriage and the unemployment rate.
14
J. Ermisch
Table 2.2 VAR model of births outside marriage (per 1,000 births) and marriages of single women (10,000s), 1871–1939 1871–1913*
bomt–1 – bomt–2 mart–1 – mart–2 urt urt–1 constant R2 DW LM (lag 1),p-val. LM (lag 2),p-val.
1921–1939*
bomt – bomt–1
mart – mart–1
bomt – bomt–1
mart – mart–1
– – 0.23 (0.10) 0.20 (0.10) –2.81 (0.44) 0.47 1.92 1.00 0.62
– – –0.44 (0.04) 0.37 (0.04) 0.63 (0.18) 0.74 1.85 0.69 0.82
0.21 (0.10) – 0.02 (0.12) –0.16 (0.11) 1.31 (0.96) 0.31 1.78 0.71 0.64
– 0.34 (0.31) –0.54 (0.20) 0.52 (0.17) 0.83 (1.83) 0.41 1.19 0.28 0.49
Note *Standard errors in parentheses.
marriages in the following year. The prolonged positive effect of unemployment (over two years) on bom probably reflects the gestation lag in births in conjunction with the annual measurement of births. These estimates are consistent with poorer labor market conditions discouraging marriages among pregnant would-be brides, thereby increasing bastardy. As working class people are affected more by fluctuations in labor market conditions, these results are consistent with the long established tendency for non-marital childbearing to be disproportionate among poorer members of society, which goes back to at least the sixteenth century in Britain (e.g. Oosterveen et al. 1980, Smout 1980). Note that the model implies that persistently higher or lower unemployment would alter the trend in bom, but in fact the unemployment rate fluctuated around a constant mean of about 5.8%. During the inter-war period, the relationship between bastardy and unemployment appears to have disappeared. This is suggested by Figure 2.3, and the second set of parameter estimates in Table 2.2 confirms this. Note that marriages of single women continued to be discouraged by higher unemployment, but with nearly complete recovery in the following year. It is likely that the very high level of inter-war unemployment discouraged childbearing generally by sharply diminishing economic prospects, so that abstinence was more likely to be practised in the lead-up to a marriage. As a consequence, fewer women became pregnant before marriage, thereby sharply reducing the impact of unemployment on births outside marriage. It is certainly the case that the general fertility rate was much lower in the inter-war period than in the decade before World War I (70 births per 1,000 women aged 15–44 compared with 104; see Figure 2.1).3
Demographic accounting, 1938–2003 The proportion of births outside marriage depends on age-specific non-marital and marital fertility rates and proportions of women who are not married. For
An economic history of bastardy 70
Unemployment rate, right axis
50
Aged 25–29
40
Aged 30–34
16 14 12 10 8
30 Aged 20–24 20
0
6 4
Aged 15–19
2 0
1938 1941 1944 1947 1950 1953 1956 1959 1962 1965 1968 1971 1974 1977 1980 1983 1986 1989 1992 1995 1998 2001
10
Percent
Per 1,000
60
15
Figure 2.4 Birth rate outside marriage, per 1,000 unmarried women.
the years preceding 1938 it is not possible to obtain annual estimates of the marital status distribution of the population by age, and this explains the reliance on the overall bom indicator in the discussion of bastardy in the pre-World War II years. This section focuses on accounting for changes in bom over the period since 1938, when such information is available. Figure 2.4 shows the age-specific fertility rates outside marriage, a direct measure of the propensity to have a birth outside marriage relative to the ‘population at risk’. After the World War II ‘spike’, these non-marital fertility rates did not return to their 1938 levels, and from the mid-1950s they increased dramatically for all age groups, peaking in 1964 (for those aged 20–34) and then declining until the mid-1970s. As comparison with Figure 2.5 indicates, marital fertility rates showed a similar ‘baby boom and bust’ pattern, but rose and fell proportionately less around their much higher levels. Figure 2.4 suggests no clear relationship between non-marital birth rates and the unemployment rate. The number of births outside marriage among women in the j-th age group in year t (BOMtj) can be written as BOMtj (Popt)(atj)(1 – mtj)(fomtj), where Popt is the female population aged 15–34 in year t, atj is the proportion of the population aged 15–34 in the j-th age group, mtj is the proportion of the female population in the j-th age group who are married, and fomtj is the fertility rate of the unmarried women in the j-th age group.4 The number of births inside marriage for women in the j-th age group is defined analogously: BIMtj (Popt)(atj)(mtj)(fimtj), where fimtj is the fertility rate of married women in the j-th age group. Then the proportion of births to women aged 15–34 outside marriage is given by: bomt (jBOMtj)/(jBIMtj + jBOMtj)
(1)
16
J. Ermisch
350 300 Aged 20–24
250 200
Aged 25–29 150 100
Aged 30–34
2001
1995
1998
1989
1992
1983
1986
1980
1977
1974
1971
1968
1962
1965
1959
1956
1953
1950
1947
1944
1941
0
1938
50
Year
Figure 2.5 Birth rate within marriage, per 1,000 married women.
where j indicates summation over the five-year age groups 15–19, 20–24, 25–29 and 30–34. We can decompose changes in bomt between any two years by holding each of the various components of BOMtj and BIMtj constant at base year values. First consider the period 1938–64. Among women aged 15–34, 4.4% of births were outside marriage in 1938, and at the peak of the baby boom this percentage was 7.4%.5 Figure 2.6 shows the large declines in the proportion of women not married among women aged 20–34. The decomposition indicates that the percentage of births outside marriage to women aged 15–34 would have increased from 4.4% in 1938 to 17.2% in 1964, rather than the actual value of 7.4% in 1964, if the proportions married in each age group had remained at their 1938 values while the age structure and fertility rates of unmarried and married women changed as they actually did. Thus, large rises between 1938 and 1964 in the age-specific proportions of women married is mainly responsible for the moderate increase in bom in the face of the large increase in non-marital fertility rates (fomtj). Changes in the proportion of women married also played a large role in accounting for changes in bom during the period 1975–2003. Figure 2.4 shows that the age-specific fertility rates of unmarried women increased from the mid1970s to the early 1990s and then were relatively constant after that, at levels higher than at the 1964 peak. Age-specific fertility rates of married women aged 20–34 exhibited a moderate upward trend (Figure 2.5). As shown in Figure 2.6, there was a large rise in the proportion of women who are not married. The decomposition for the 1975–2003 period indicates that if the proportions married in each age group had remained at their 1975 values while the age struc-
An economic history of bastardy
17
1 0.9 Aged 15–19 0.8 0.7 0.6 0.5
Aged 20–24
0.4 Aged 25–29
0.3 0.2
Aged 30–34 0.1 2001
1995
1998
1992
1986
1989
1983
1980
1977
1971
1974
1968
1965
1962
1959
1956
1953
1947
1950
1941
1944
1938
0
Year
Figure 2.6 Proportion of women not married.
ture and fertility rates of unmarried and married women changed as they actually did, the proportion of births outside marriage to women aged 15–34 would have increased from 9.1% in 1975 to only 11.6% in 2003, rather than the actual value of 44.2% in 2003.6 Thus, less than one-tenth of the 1975–2003 increase in the proportion of births outside marriage can be accounted for by simultaneous changes in components other than proportions married in each age group. The large role played by changes in the proportions married is also clear if we consider what would have happened if age-specific non-marital and marital fertility rates were held constant at their 1975 level. This decomposition indicates that percentage of births outside marriage among women aged 15–34 would have risen to 40%, very close to the observed 44%. The shift from legal marriage to cohabiting unions as the mode of first partnership mainly accounts for the delay of first marriage in Britain and the rise in the proportion of women not married (Ermisch and Francesconi 2000). Among women born in the 1950s, about one-fourth cohabited in their first livein partnership. This proportion increased to three-fifths among women born in the 1960s and to 85% among women born in the 1970s.7 But there has also been a delay in the age of first partnership: comparing women born in the 1950s, 1960s and 1970s, the median age at first partnership has risen from 22 to 24 to 25, respectively. The seventh section discusses possible reasons for the diffusion of cohabiting unions. These are, of course, accounting exercises. Similar socio-economic factors
18
J. Ermisch
may have produced the rise in cohabitation, the delay in first partnership and an increase in the propensity to have children outside marriage, making the changes in non-marital fertility rates and the changes in proportions married interdependent.
The decision to have a child outside marriage Throughout the 1950s and most of the 1960s, unmarried women continued to have no recognised legal right of access to contraception. The absence of reliable contraception produced many unplanned pregnancies, and the risks of and taboos against abortion made the outcome of premarital pregnancies relatively clear – have the child, with or without a husband. The contraceptive pill changed all that. It was introduced in 1961 to family doctors (General Practitioners – GPs), and in 1962 the Family Planning Association (FPA) started offering it to married women, provided they had permission of their GP. It became available from the National Health Service (NHS) from December 1961, but only to married women whose health would be endangered by further pregnancy. In 1966, the NHS allowed GPs to charge for pill prescriptions not given for medical reasons – pill sales rose sharply from then on, helped by a fall in its price during the 1960s. In 1968, the FPA gave branches permission to provide contraceptives to unmarried women, and from 1970 they were required to make provision. Contraception became free to all women from 1975.8 Abortion became legal in 1969. The free availability of contraception and abortion after 1975 is likely to have played an important role in the postponement of marriage discussed in the previous section and illustrated in Figure 2.6. Before the pill and legal abortion, there were considerable costs from delaying marriage – sexual abstinence or pregnancy risk. By reducing these costs, the pill encouraged all women and men to delay marriage to a time when their tastes, character and economic position were better formed.9 In particular, widespread unmarried cohabitation is inconceivable without the pill (with legal abortion as a backup). After 1975, ‘accidents’ were no longer a convincing reason why women had a child outside marriage. Why did such births not fall when more women remained single and had the means to avoid non-marital childbearing if they wanted to do so? For instance, in Japan the average age at marriage also increased dramatically to levels similar to that in England, but only about 2% of children are born outside marriage.10 A better understanding of the increase in non-marital childbearing requires a behavioral model of the decision to have a birth outside marriage that takes into account the reliability of modern fertility control and the interdependency between marriage and childbearing decisions. It builds on the observation that it takes time to meet potentially suitable members of the opposite sex. Such marriage market frictions affect who marries whom, the gains from each marriage and the distribution of gains between spouses. They also open the possibility of childbearing outside marriage as a rational choice, even when a woman can
An economic history of bastardy
19
control her fertility perfectly.11 When a man and woman are in a relationship, the man can choose whether to marry the woman or not, if she will have him. While a woman faces the same choice, she can also choose to have a child by the man and raise it without the father. If the social welfare system, parental support or the father’s child support are generous enough, a woman’s welfare when raising a child by herself would be greater than what she obtains when single and childless. But there are also costs in terms of future marriage market prospects associated with raising a child alone: a single woman with child may find it more difficult to meet potential husbands while looking after a child. A woman who has a relationship with a man she does not wish to marry, or who will not marry her, would choose to have a child by the man if the short-run welfare gain exceeds the long-term costs in terms of her marriage prospects. An important implication of this model is that couples who find each other to be mutually acceptable marriage partners wait to have children within marriage, while a woman may have a child outside marriage if this is not the case. This suggests that sexual relationships that produce a child outside marriage should be much less likely to lead to marriage than those that do not. In general it is difficult to observe the outcomes of relationships, but we can observe the outcome of cohabiting unions. The evidence indicates that cohabiting unions that produce children are much less likely to be converted into marriage and more likely to break up than childless ones (Ermisch and Francesconi 2000). Births in cohabiting unions make up about 60% of recent non-marital births in England and Wales. Those women who expect to obtain a significant increase in welfare when they marry suffer a greater long-term cost by having a child while single than women whose marriage prospects are such that they expect to gain little from marriage. Thus, women with poorer marriage prospects should be more likely to have children outside marriage. If marriage market prospects are worse for poorer women (e.g. those with low educational attainments), because they can only marry men with low incomes, we would expect that poorer women would be more likely to have a child before marriage, a prediction which is repeatedly confirmed (e.g. Ermisch 2001; Del Bono 2004).12 Another clear implication of the model is that non-marital childbearing is more likely if, all else being equal, state benefits for single mothers are higher.13 Could improvements in benefits explain the rise in non-marital birth rates between the mid-1970s and early 1990s, shown in Figure 2.4? The British benefit system for non-employed single mothers is called Income Support (Supplementary Benefit before 1991); it has an implicit 100% tax rate on earnings and other income, which strongly discourages employment. Increases in women’s wages increase welfare in the single, childless and married states, but have little effect on welfare as a single mother because the vast majority of single mothers are not employed, while benefits only increase welfare in the single mother state. The ratio of benefits for a single mother with one child to women’s average full-time hourly wage fluctuated during the 1970s and then fell from 1980 until the late 1990s, after which it has stabilised. Higher state benefits clearly cannot be a stimulant to the rise in non-marital birth rates since 1980.
20
J. Ermisch
Conditions of higher unemployment tend to reduce men’s incomes, particularly those men whom women with poorer marriage market attributes might have a chance of marrying. Thus, in labor markets in which the unemployment rate is higher, the value of waiting childless for the right man to come along is reduced relative to the welfare from having a child as a single mother. Poor employment opportunities also reduce a woman’s opportunity cost of having a child on her own. Thus, like our simple theory of the second section, this model also leads us to expect that a higher unemployment rate increases childbearing outside marriage, but for different reasons.14 The large but temporary increase in unemployment in the first half of the 1980s shown in Figure 2.4 does not appear to be consistent with sustained higher non-marital birth rates, but the next section suggests it could have a role to play.
Social influence and interaction A woman’s welfare as a single mother is reduced by ‘social stigma’ attached to being one, and stigma may also reduce her marriage prospects further, thereby discouraging non-marital childbearing. The stigma discussed earlier in the chapter continued to exist throughout the 1950s and early 1960s (Szreter 1996: 577). If such stigma declines, then more women would decide to have a child (on her own) with a man whom she rejects as her husband (or who rejects her as a wife), and among women who object to abortion, fewer would marry in response to a pregnancy. Evidence on the outcomes of conceptions outside marriage is consistent with a decline in stigma.15 For example, the percentage of premarital conceptions among women aged under 20 leading to births within marriage fell from 55% in 1969 to 2% in 2003. The corresponding percentages for women aged 20–24 were 45% (1969) and 4% (2003). There were complementary rises in the percentage of conceptions leading to births outside marriage: from 9% to 56% for women aged under 20 and from 16% to 56% for women aged 20–24.16 Thus, the intervention of marriage between premarital conception and birth has become relatively rare. This section presents a model that can produce rapid erosion of such stigma, which would both increase non-marital childbearing directly and increase the single population. Suppose that, because social stigma is less when more women become single mothers, the utility associated with being a single mother relative to the utility associated with remaining childless is larger when more women in her reference group (e.g. defined by neighborhood) become mothers outside marriage. Then the probability that a woman becomes a single mother when she and her partner do not agree to marry increases with the expected proportion of women in a woman’s reference group who have a birth outside marriage in this situation.17 This is what we shall call a social interaction effect. A social equilibrium occurs when people’s expectations are consistent with the average proportion in the reference group who become single mothers when a couple does not marry; that is, when the actual proportion in the reference group is equal to the expected proportion. The non-linear curve in Figure 2.7
An economic history of bastardy
21
1 0.9 0.8 C
Actual proportion
0.7 0.6 0.5 0.4
B
0.3 0.2 0.1
A
0 Expected proportion
Figure 2.7 Equilibria in proportions becoming a single mother.
plots the relationship between actual and expected proportions becoming single mothers for a relatively large social interaction effect, and the 45-degree line represents the condition that the actual and expected proportions are equal. The points at which the curve intersects the line are ‘social equilibria’. Figure 2.7 illustrates the possibility of more than one social equilibrium. If we make some plausible assumptions about dynamics, then we can characterise an equilibrium like B in Figure 2.7 as unstable.18 When women respond sufficiently to what others are doing, as assumed in Figure 2.7, temporary changes in the socio-economic environment that alter non-marital childbearing behavior and/or expectations can produce dramatic changes in the proportion who become single mothers, by causing a move from a type A equilibrium to a type C equilibrium. For example, some temporary development that encouraged women to expect that the proportion becoming single mothers would be above that associated with equilibrium B in Figure 2.7 would produce a permanent dramatic increase in the proportion. In this sense, ‘history matters’ for the selection of the low-level or high-level equilibrium. It is difficult to believe that the fundamental determinants of the decision to become a single mother have changed by an order of magnitude sufficient to explain the large increase in births outside marriage between 1975 and 2004. The increase may, however, reflect temporary changes in these determinants, the effects of which were magnified by social interaction effects. One possible driver of these changes was the steep rise in unemployment in first half of the 1980s, which may have reduced the pool of men that poorer women would find acceptable to marry, thereby discouraging them from waiting to begin childbearing within marriage. Figures 2.4 and 2.6 show that the timing of its increase
22
J. Ermisch
roughly coincides with the large increase in non-marital birth rates and the postponement of marriage. Even though unemployment subsequently fell, the rise may have been sufficient to have moved the actual proportion becoming single mothers above that in an equilibrium like B in Figure 2.7, which in turn altered expectations for subsequent cohorts, ultimately leading to an equilibrium like C. More generally, we are looking for catalysts that initiated the dynamics driven by social interaction effects. Comparison across countries of changes in the non-marital birth percentage may shed some light on what these might be. Figure 2.8 shows acceleration in its increase around 1980 in a number of European countries. The similarity between France and England and Wales is particularly striking, but there were also rapid rises in the Netherlands, Ireland, Spain and Portugal, starting about 1980. Other European countries have also seen large increases in more recent years: Germany since the mid-1990s (reaching 29% in 2004) and Italy since the late 1990s (reaching 17% in 2004). Whatever the catalysts might be, it appears that they are not specific to one country. Unemployment increased substantially during the first half of the 1980s in all of these countries (e.g. see Nickell et al. 2005), and so it is a candidate for such a catalyst. Figure 2.9 provides a ‘scatter plot’ between the change in the percentage of births outside marriage (bom) between 1979 and 2004 and the increase in the average unemployment rate (ur) between 1973–79 and 1980–87 for a number of European countries and the USA. A longer period has been used for the change in bom to allow for social interaction effects. Spain is an outlier on the far right of Figure 2.9, but the broad tendency is for the rise in extra-marital childbearing to be larger for countries experiencing a larger increase in unemployment in the 1980s.19
60 50 40 Ireland France Netherlands E&W Spain Portugal Germany Italy
30 20 10
1960 1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004
0
Figure 2.8 Percent of births outside marriage, Western Europe.
An economic history of bastardy
23
40 y = 0.97117 + 19.723 R2 = 0.201
Change in bom 1979–2004
35 30 25 20 15 10 5 0
0
2
4 6 8 10 Change in UR 1973/79–1980/87
12
14
Figure 2.9 Changes in births outside of marriage (bom) 1979–2004 and changes in average unemployment rate (UR) 1973/79–1980/87.
Educational differences in premarital birth rates – changes by birth cohort The model in fourth section predicts that women with poorer economic prospects are more likely to become single mothers. To be more concrete, it seems likely that women with lower educational attainments have poorer prospects in marriage and labor markets, and their marriage decisions may also be more sensitive to unemployment because the incomes of their potential spouses are affected more by it. Also, assume that a person’s reference group consists of those with a similar level of education. Those with ‘low education’ would be represented by the curve in Figure 2.7, while those with ‘high education’ would be represented by a similar, but lower curve, which only intersects the 45-degree line once. Thus, initially there is only a small difference between the two education groups in the equilibrium proportions of women becoming single mothers and both are low. Suppose that there is a temporary change in the environment (e.g. higher unemployment) that increases the actual proportion of women in the lower education group who become single mothers above the middle (unstable) equilibrium point (B). Then the dynamics of social interaction would drive them to the higher equilibrium point (C). In this new equilibrium, the difference between the education groups in their equilibrium proportion would be very large. Individual data on women’s birth histories are used to explore whether educational differences in non-martial childbearing widened as it increased in prominence.
24
J. Ermisch
The data come from the British Household Panel Survey (BHPS) retrospective histories of cohabiting unions, marriages and births, which have been updated with information during the panel, 1993–2005.20 For simplicity, a dichotomous indicator of educational attainments is used to measure educational differences between women. It is whether or not a woman’s highest qualifications did not exceed ‘O-levels’ (later, ‘General Certificate of Secondary Education’), which are usually obtained by the time a person is 16 years old. Denote these women as having ‘low education’. Three sets of cohorts are distinguished: those born in the 1950s, in the 1960s and the 1970s, who reach the primary ages for having births before marriage about 20 years later. Note that all of these cohorts ‘come of age’ in the post-pill era. The dependent variable of interest is the ‘hazard rate’ of a premarital first birth for woman j at age t, denoted hjt. Women are assumed to be at risk for such a birth from the age of 14, and they drop out of the population at risk when they marry or when they have a first birth. Women who remained childless and never-married from age 14 to the time of the last survey in which they are present remain in the population at risk until the time of their last interview.21 In order to have a simple parameter to compare across cohorts, a ‘proportional hazard’ model is estimated for each cohort; it takes the form hjt = g(t) exp(Ej)
(2)
where Ej = 1 denotes a woman leaving education with ‘low education’ and Ej = 0 denotes a woman with higher qualifications. The function g(t) is an unspecified function of age. This model has the property that, for two women j and k of the same age, the ratio of their hazard rates is equal to hjt/hkt = exp[(Ej –Ek)]. The parameter is estimated by Cox’s partial likelihood method, and Table 2.3 reports the estimate of exp(), which is the ratio of the premarital birth hazard of a woman with ‘low education’ to that of a woman with ‘high education’. The
Table 2.3 First birth rate before marriage: differences by educational group Birth cohort:
1950–60
exp()* 3.04 SE (0.60) p-value of test of b not different from 1950–60 value n/a p-value of joint test of no difference in b Log-rank test, 2(1) 35.43 N-high education 468 N-low education 795 N 1,263 N of premarital births Note *Cox proportional hazard model of a pre-marital first birth.
148
1960–70
1970–80
3.95 4.41 (0.45) (0.44) 0.255 0.096 0.244 171.98 265.54 851 1,375 986 1,062 1,837 2,437 419
516
An economic history of bastardy
25
Table 2.4 First birth rate before marriage: differences by educational group A. Before first live-in partnership Birth cohort: exp(b)* SE p-value of test of b not different from 1950–60 value p-value of joint test of no difference in b Log-rank test, 2(1) N-high education N-low education N N of pre-partnership births
1950–60 2.22 (0.37) 24.22 514 871 1,385 182
1960–70
1970–80
3.98 4.71 (0.49) (0.66) 0.003 0.000 0.0008 145.96 149.88 887 1,330 1,075 1,066 1,962 2,396 361
278
1950–60
1960–70
1970–80
1.74 (0.61)
2.30 (0.42) 0.562 0.697 21.96 414 375 789
2.50 (0.36) 0.406 43.45 464 447 911
131
218
Note *Cox proportional hazard model of first birth before the first partnership.
B. Within first cohabiting union Birth cohort: exp(b)* SE p-value of test of b not different from 1950–60 value p-value of joint test of no difference in b Log-rank test, 2(1) N-high education N-low education N N of within first cohab.union births
2.54 147 155 302 37
Note *Cox proportional hazard model of first birth within the first cohabiting union.
estimates of this hazard ratio increase across cohorts, although only the 1970s hazard ratio is significantly larger than the 1950s ratio (at the 0.10 level). Table 2.4 breaks down non-marital first births between those before the first live-in partnership and those in the first cohabiting union. In each case, the hazard ratio is higher for later birth cohorts, although only significantly so for births outside of a live-in partnership.22 This evidence is weakly consistent with the emergence of a ‘high level equilibrium’ for less educated women.
The diffusion of cohabiting unions The rise in non-marital childbearing in many countries illustrated in Figure 2.8 has been strongly associated with the increase in cohabiting unions. For instance, there is a tendency for countries with a higher proportion of women cohabiting in their twenties to have a larger percentage of births outside
26
J. Ermisch
marriage (Ermisch 2003: Figure 11.6). In 2004, about 64% of births outside marriage in England and Wales occurred in cohabiting unions.23 The percentages are higher in a number of other countries. For example, among women born during 1964–69 in France, 75% of non-marital births were in cohabiting unions (Le Goff 2002); in Portugal and Norway, 80% of non-marital births in 2004 and 2005, respectively, were in such unions, and 90% of non-marital births in Sweden during 1995–97 were in cohabiting unions.24 In the USA, one-half of non-marital births to non-Hispanic white women during 1990–94 were to cohabiting parents (Bumpass and Lu 2000). Thus, an alternative, but also possibly complementary, explanation for the dramatic increase in bastardy in the last quarter century is that the large rise in the proportion of women who cohabit in their first partnership (rather than marry) is responsible. The reason that this may be the case follows from examining the dynamics of cohabiting unions in Britain. Analysis of the BHPS data indicates that the time spent living together in cohabiting unions before either marrying each other or the union dissolving is usually very short – the median duration was about two years for women born in the 1950s and 1960s, rising to three and a half years for those born in the 1970s. There has been an upward trend in the proportion of women’s first cohabiting unions that dissolve rather than turn into marriage: 30%, 37% and 50% for women born in the 1950s, 1960s and 1970s, respectively. Most of those who re-partner after their cohabiting union dissolved also start their next partnership by cohabiting. It takes about two years for one half to have formed a new partnership, which is again subject to the high risk of dissolution. Thus, the time spent cohabiting, the relatively high risk that the union dissolves and the time it takes to cohabit again all contribute to a longer time before any marriage takes place and, therefore, more time at risk to have a birth outside marriage. Adding to this time at risk is the delay in first partnerships for more recent cohorts discussed earlier. At fixed age-specific fertility rates outside of a live-in partnership and in cohabiting unions, the increase in the exposure time to these non-marital birth rates would have increased nonmarital childbearing. This is consistent with the large accounting contribution of the decline in the proportion of women married to the rise in the proportion of births outside marriage discussed in the third section. But why did average non-marital fertility rates not fall when more women cohabited? Women had the means (contraception and legal abortion) to avoid non-marital childbearing if they wanted to do so, and so the substitution of cohabiting unions for marriages need not have raised non-marital fertility. This consideration leads us back to the explanations suggested by the models in the fourth and fifth sections: purposeful decisions to have a child outside marriage and erosion of the stigma against doing so. Also, the demographic accounting begs the question: why did cohabiting unions become the common type of first partnership? If there was social stigma operating against cohabitation, its erosion and the dramatic increase in cohabitation may be due to a social interaction model like that outlined in the fifth section, applied to cohabiting unions. It could also be the case that cohabiting unions and partnership delay were only
An economic history of bastardy
27
attractive when the stigma associated with childbearing outside marriage was less important. In addition to the contraceptive pill and legal abortion, another possible driving force for the delay and change in mode of first partnership may have been important changes in young people’s education. For instance, one-third of 18-year-olds were in full-time education in 1992, compared with 15% in 1979. Bagnoli and Bergstrom (1993) argue that young men who expect to prosper in later life will postpone marriage until their success becomes evident to potential marriage partners. Those who do not expect their economic status to advance much will seek to marry at a relatively young age. The careers of those who obtain university degrees take longer to develop and their earnings peak later in life. Thus, as more young men go on to higher education, the proportion of young men who think it is worth waiting to signal their better economic status is likely to increase. The same may now apply to women, and also women who are university graduates prefer to marry graduate men. Beyond a certain age, young people may nevertheless prefer to have a live-in partner, particularly in an era of reliable contraception, and cohabiting unions cater for this preference while allowing them to postpone making a long-term commitment. Thus, higher educational achievement would both encourage young people to enter live-in partnerships later and to cohabit when they do. The short spells of cohabitation that are observed are consistent with the argument that it is used while waiting to signal economic success and as a learning experience before stronger commitments are made. Table 2.5 presents evidence from the BHPS on the timing of a first live-in partnership and the odds of cohabiting in this partnership. Panel A presents the estimates of a proportional hazard model like (2) for the hazard of a first union. It indicates that there is little change between the 1950s and 1960s birth cohorts in the educational difference in the timing of a first union – the hazard rate for ‘low educated’ (as defined earlier) women is about 30% higher – but the difference in union formation rates widens for the 1970s cohort – i.e. the higher educated wait even longer to start their first partnership. Panel B presents estimates of a logit model for the odds of cohabiting relative to marrying in a woman’s first union. As the arguments above predict, low educated women born in the 1950s and 1960s were much less likely to cohabit in their first partnership, but as cohabiting in one’s first partnership became almost universal (85%), this is no longer true for the 1970s cohort.25 The results are consistent with social interaction effects spreading cohabitation widely from better educated women, who always had an incentive to cohabit before marrying, to a large proportion of the population. Through its link to the widespread substitution of cohabiting unions for direct marriage, the increase in non-marital childbearing may, therefore, be due in part to the expansion of higher education acting in conjunction with social interaction effects. For this argument to be correct it is not necessary to show that university graduates were themselves having children in cohabiting unions to a significant degree (they were not). They were the pioneers of cohabiting unions,
28
J. Ermisch
Table 2.5 First union: differences by educational group A. Age at first union Birth cohort:
1950–60 1.30 (0.08)
1960–70
1970–80
exp(b)* SE p-value of test of b not different from 1950–60 value p-value of joint test of no difference in b N-high education N-low education N
517 877 1,394
1.28 1.39 (0.07) (0.08) 0.148 0.003 0.012 901 1,334 1,085 1,071 1,986 2,405
N of first unions
1,100
1,581
1,291
Note *Cox proportional hazard model of first partnership formation..
B. Odds of cohabiting union relative to marriage Birth cohort: b* SE p-value of test of b not different from 1950–60 value p-value of joint test of no difference in b N-high education N-low education N N of cohab.unions
1950–60 –0.68 (0.14) 395 705 1,100 298
1960–70
1970–80
–0.36 0.18 (0.10) (0.15) 0.023 0.000 0.000 692 622 889 669 1,581 1,291 964
1,071
Note *Logistic model of odds of cohabiting in first partnership.
but social interaction effects spread them widely through society, to people who would have stronger incentives to have children within such unions (see the fourth section).
Conclusions Over the period 1845–1960, the percentage of births outside marriage moved within a small range, averaging about 5%. Up to World War I, higher unemployment discouraged marriage and increased non-marital births, with a recovery in marriages in the subsequent year. This pattern is consistent with poorer labor market conditions discouraging marriages among pregnant would-be brides, thereby increasing bastardy. During the inter-war period, higher unemployment continued to produce postponement of marriages, but non-marital childbearing was no longer linked to unemployment. After 1960, childbearing outside marriage began to climb slowly, and it
An economic history of bastardy
29
exploded after 1980, reaching 42% in 2004. This was partly driven by a steep increase in age-specific non-marital births rates among women aged 20–34 from the mid-1970s to the early 1990s, after which they stabilised at a high level. At fixed average non-marital and marital age-specific birth rates, the increase in the proportion of births outside marriage can be mainly accounted for by a large fall in the proportion of women aged 20–34 who are married, which is in turn associated with a dramatic rise in cohabiting unions. These unions are short-lived before either dissolving or being converted into marriage. But this begs the question: why did average non-marital fertility rates not fall when more women cohabited? Women had the means – abortion became legal in 1969 and the contraceptive pill was freely available to unmarried women from 1975 – to avoid non-marital childbearing if they wanted to do so, and so the substitution of cohabiting unions for marriages need not have raised non-marital fertility. A theory of marriage market search (courtship) in which out-of-wedlock childbearing is an option suggests why it may be a rational choice, even when fertility can be controlled. Because of social stigma against single mothers, a woman’s welfare as a single mother is likely to be influenced by the prevalence of single mothers in the population. When their prevalence is low, non-marital childbearing is discouraged. A temporary change in the determinants of nonmarital childbearing that raises it, like the large rise in unemployment in the first half of the 1980s, could have produced rapid erosion of the stigma and a selfreinforcing rise in childbearing outside marriage. This dynamic was likely to be concentrated among a segment of the population who already had stronger incentives to have a child before marriage, such as women with low educational attainments. There is indeed evidence that differences in the chances of having a child before marriage by women’s educational level widened as childbearing outside marriage became more common. An alternative, or complementary, explanation stresses the role of the rise in cohabiting unions and delay in partnership. These generated an increase in non-marital births by increasing the unmarried population. This view also points to the operation of a social influence model in explaining the dramatic rise in cohabitation, and the chapter provides evidence of a diffusion of cohabiting unions from the better educated to the less educated population.
Appendix: the BHPS In Autumn 1991, the BHPS interviewed a representative sample of 5,500 households, containing about 10,000 persons. The same individuals are re-interviewed each successive year, and if they split off from their original households to form new households, all adult members of these households are also interviewed. Similarly, children in the original households are interviewed when they reach the age of 16. Thus, the sample remains broadly representative of the population of Britain as it changes through the 1990s. The core questionnaire elicits information about income, labor market behavior, housing conditions, household composition, education and health at each yearly interview.
30
J. Ermisch
The second wave of the BHPS collected, during the last quarter of 1992, complete fertility histories and also histories of all spells of marriage and cohabitation from a representative sample of 9459 adults aged 16 and over throughout Great Britain. Information on cohabitation is elicited by the following question: ‘As you know some couples live together without actually getting married. Have you ever lived with someone as a couple for three months or more?’ If the answer is yes, questions then proceed to ask how many of such partnerships he/she had and the months and years at which they started and stopped living together. Booster samples for Scotland and Wales were added in 1999, and demographic histories were collected for these respondents in 2001. In 2001, a Northern Ireland sample was added, and demographic histories were collected from these respondents in 2002. These histories are matched with the information obtained from the annual panel interview during the period 1991–2003. This produces a history for each person through the last year that they participated in the panel, and also a partial history for people joining the panel who did not report a complete history in 1992, 2001 or 2000.
Notes * I am grateful to Tim Hatton for advice and comments on earlier versions of this chapter, to Chiara Pronzato for constructing the demographic history files combining retrospective and prospective panel information and to conference participants for comments on an earlier version of the chapter. 1 In contrast to yt being covariance-stationary, it has a ‘unit root’ if it takes the form yt = + yt–1 + ut, where ut follows a stationary and invertible autoregressive movingaverage (ARMA) process. The null hypothesis for the Augmented Dickey-Fuller (DF) test is that = 1 in the model yt = + yt–1 + ut. Rejection of the null is consistent with stationarity. The DF statistics in the level of the variables never come close to rejecting the null. 2 A variable x is said to Granger-cause variable y if, given past values of y, past values of x are useful for predicting y. Operationally, rejection of the hypothesis that the coefficients of the past values of x are jointly zero in a regression with these values and past values of y is evidence of Granger causality. 3 The absence of an impact of unemployment on births outside marriage may also reflect the pervasive culture of sexual restraint that had evolved by the inter-war years. 4 Over the period 1938–64, women aged 15–34 produced about 85% of all births and 90% of all births outside marriage. During 1964–2003, women aged 15–34 produced about 90% of all births and 93% of all births outside marriage. 5 The corresponding percentages for births to women of all ages were 4.3% and 7.2%. 6 The corresponding percentage for births to women of all ages were 9.1% and 41.4%. 7 These estimates come from the partnership histories in the BHPS, described in the Appendix. 8 Goldin and Katz (2002) demonstrate a similar diffusion of the pill among single women in the USA, but for different reasons. 9 See Goldin and Katz (2002) for a similar argument for the role of the pill in delaying marriage for a large fraction of all American young people in the 1970s. 10 This is not because of low levels of premarital sex: the proportion of Japanese female junior college and university students who reported having sexual intercourse increased from 11 to 51% between 1974 and 1999. The pill was not legalised in Japan
An economic history of bastardy
11 12
13 14
15
16 17
18 19
20 21 22
23 24 25
31
until 1999. The main contraception is condoms, backed up by abortion, which is both legal and socially acceptable. Premarital pregnancies that are not quickly followed by marriage are almost always aborted; see Retherford and Ogawa (2006: 19–20). What follows is a brief description of the matching model presented in Burdett and Ermisch (2002) and Ermisch (2003: Chapter 7). Other models integrating non-marital childbearing and marriage decisions, which assume transferable utility and no marriage market frictions (Willis 1999; Neal 2004), also predict that if there is childbearing outside marriage in equilibrium, it is women with lower ‘endowments’ (and, therefore, valued less in the marriage market) who are responsible for it. The same prediction comes out of most economic models of non-marital fertility; see Rosenzweig (1999), Nechyba (2001) or Neal (2004). Analysis of a large cohort of women born in 1970 (Del Bono 2004), who were making childbearing and partnership decisions in the late 1980s and 1990s, indicates that women living in counties with higher male unemployment were more likely to become a mother outside a live-in partnership and less likely to enter a partnership. This is consistent with the model’s prediction. In the official statistics, conceptions are defined as pregnancies resulting in live births, still births or legal terminations under the 1967 Abortion Act; miscarriages and spontaneous and illegal abortions are excluded. Since abortions are available from the NHS, it appears likely the statistics include virtually all induced abortions. The data come from Birth Statistics (various years, Tables 12.5 and 12.6). The residual in each case is conceptions terminated by abortion. Until the mid-1970s, a growing proportion of premarital conceptions in each five-year age group were terminated, after which time these proportions have been relatively constant. This type of model was discussed by Schelling (1978), and more recently by Manski (1993) and Durlauf and Young (2001). Formally, the model can be written as P = H[ + Pe + z], where P is the probability that a women becomes a single mother, Pe is the expected proportion becoming a single mother in that person’s ‘reference group’, z denotes a vector of individual attributes, H[·] is a specified continuous, strictly increasing distribution function, such as the logistic distribution, and , and are parameters. Social stigma/social influence/social approval effects would imply > 0. Nechyba (2001) analyses a similar model to investigate the effects of state benefits to single mothers on non-marital childbearing. From left to right (i.e. in order of unemployment change), the countries are Sweden, the USA, Austria, Italy, Portugal, Germany, France, the Netherlands, the UK, Ireland and Spain. For what it is worth, when Spain is omitted, the regression line is y = 2.9751x + 13.835, with an R2 of 0.612. See Appendix for further details. I am grateful to Chiara Pronzato for constructing the demographic history files combining retrospective and prospective panel information. Thus, there are two groups who are ‘censored’: those who marry childless and those who neither marry, nor have a child before the time they are interviewed last. A survivor function is associated with any hazard function. In this case it is the proportion surviving childless and never-married. Tables 2.3 and 2.4 report the 2 statistic of the log-rank test for the equality of the survivor functions of the two education groups. The value of the test statistic clearly increases for more recent cohorts, indicating bigger differences in their survival functions. Thus, this non-parametric test also supports the emergence of larger educational differences, which may have been driven by social interaction effects. This figure is estimated by the percentage of births outside marriage jointly registered by both parents at the same address. Among women born during 1964–69 in West Germany, 65% of non-marital births were in cohabiting unions (Le Goff 2002). The impact (standard error) of having a higher educational attainment (i.e. above
32
J. Ermisch ‘A-level’ qualifications, usually obtained by 18, or nursing qualifications) on the log odds of cohabiting in one’s first partnership relative to not having one is: 0.64 (0.13), 0.54 (0.12) and –0.38 (0.17) for women born in the 1950s, 1960s and 1970s, respectively. But note that those who have formed a union in 1970s cohort under-samples women who have a higher educational attainment, both because they have not had a union and because some of them have not had time to complete a higher educational attainment.
References Bagnoli, M. and Bergstrom, T. (1993) ‘Courtship as a waiting game’, Journal of Political Economy, 101: 185–202. Boyer, G.R. and Hatton, T.J. (2002) ‘New estimates of British unemployment, 1870–1913’, The Journal of Economic History, 62: 643–75. Bumpass, L. and Lu, H.-H. (2000) ‘Trends in cohabitation and implications for children’s family contexts in the United States’, Population Studies, 54: 29–41. Burdett, K. and Ermisch, J.F. (2002) ‘Single mothers’, Working Paper Institute for Social and Economic Research Working Papers. Paper 2002–30, Colchester: University of Essex. Cook, H. (2004) The Long Sexual Revolution. English Women, Sex and Contraception 1800–1975, Oxford: Oxford University Press. Del Bono, E. (2004) ‘Pre-marital fertility and labor market opportunities: evidence from the 1970 British Cohort Study’, IZA Discussion Paper No. 1320, Bonn. Durlauf, S.N. and Young, H.P. (eds) (2001) Social dynamics, London: MIT Press. Ermisch, J.F. (2001) ‘Cohabitation and childbearing outside marriage in Britain’, in L. Wu and B. Wolfe (eds) Out of Wedlock, New York: Russell Sage Foundation. Ermisch, J.F. (2003) An Economic Analysis of the Family, Princeton: Princeton University Press. Ermisch, J.F. and Francesconi, M. (2000) ‘Cohabitation in Great Britain: not for long, but here to stay’, Journal of the Royal Statistical Society, Series A, 163: 153–71. Goldin, C. and Katz, L.F. (2002) ‘The power of the Pill: oral contraception and women’s career and marriage decisions’, Journal of Political Economy, 110: 730–70. Hair, P.E.H. (1970) ‘Bridal pregnancy in earlier rural England further examined’, Population Studies, 24: 59–70. Humphries, S. (1988) A Secret World of Sex. Forbidden Fruit: The British Experience 1900–1950, London: Sidgwick and Jackson. Laslett, P. (1977) Family Life and Illicit Love in Earlier Generations, Cambridge: Cambridge University Press. Laslett, P. (1980) ‘Introduction: comparing illegitimacy over time and between cultures’, in P. Laslett, K. Oosterveen and R. Smith (eds) Bastardy and Its Comparative History, London: Edward Arnold. Le Goff, J.-M. (2002) ‘Cohabiting unions in France and West Germany: transitions to first birth and first marriage’, Demographic Research, 7: 593–624 (www.demographicresearc.org/Volumes/Vol7/18). Manski, C. (1993) ‘Identification of endogenous social effects: the reflection problem’, Review of Economic Studies, 60: 531–42. Neal, D. (2004) ‘The relationship between marriage market prospects and never-married motherhood’, Journal of Human Resources, 39: 938–57.
An economic history of bastardy
33
Nechyba, T.J. (2001) ‘Social approval, values, and AFDC: a reexamination of the illegitimacy debate’, Journal of Political Economy, 109: 637–72. Nickell, S., Nunziata, L. and Ochel, W. (2005) ‘Unemployment in the OECD since the 1960s. What do we know?’, The Economic Journal, 115: 1–27. Oosterveen, K., Smith, R.M. and Stewart, S. (1980) ‘Family reconstitution and the study of bastardy: evidence from certain English parishes’, in P. Laslett, K. Oosterveen and R. Smith (eds) Bastardy and Its Comparative History, London: Edward Arnold. Retherford, R. and Ogawa, N. (2006) ‘Japan’s baby bust: causes, implications, and policy responses’, in F.R. Harris (ed.) The Baby Bust, Oxford: Rowman and Littlefield. Rosenzweig, M.R. (1999) ‘Welfare, marital prospects and nonmarital childbearing’, Journal of Political Economy, 107: S1–32. Schelling, T.C. (1978) Micromotives and Macrobehavior, New York: W.W. Norton. Shorter, E., Knodel, J. and Van De Walle, E. (1971) ‘The decline of non-marital fertility in Europe’, Population Studies, 25: 375–93. Smout, C. (1980) ‘Aspects of sexual behavior in nineteenth century Scotland’, in P. Laslett, K. Oosterveen and R. Smith (eds) Bastardy and Its Comparative History, London: Edward Arnold. Southall, H. and Gilbert, D. (1996) ‘A good time to wed? Marriage and economic distress in England and Wales, 1839–1914’, The Economic History Review, 49: 35–57. Szreter, S. (1996) Fertility, Class and Gender in Britain, 1860–1940, Cambridge: Cambridge University Press. Willis, R.J. (1999) ‘A theory of out-of-wedlock childbearing’, Journal of Political Economy, 107: S33–64. Wrightson, K. 1980. ‘The Nadir of English illegitimacy in the seventeenth century’, in P. Laslett, K. Oosterveen and R. Smith (eds) Bastardy and Its Comparative History, London: Edward Arnold.
3
Epidemics, demonstration effects, and investment in sanitation capital by U.S. cities in the early twentieth century Louis Cain and Elyce Rotella*
During the years 1899–1929, American cities made enormous capital investments in water and sewage treatment plants. Such investments would accelerate with the availability of New Deal dollars, but this chapter focuses on the earlier period when cities themselves bore the entire cost. By the end of the nineteenth century, most American cities had already built systems to deliver water and remove sewage. Then attention shifted to treatment – especially filtration and purification. In this chapter we examine not the construction of the basic works that supplied water and sewage services, but rather the treatment of what passed through those works. The development of the germ theory provided the intellectual basis for understanding the benefits offered by water and sewage services. We focus on two forces that contributed to the expansion of sanitation treatment – epidemics and demonstration effects. We argue that epidemics have their greatest influence on the demand side of the political market in which the decision to invest in treatment works is made – i.e. epidemics caused urban residents to demand that politicians provide them with better sanitation services. In contrast, demonstration effects, by lowering information costs, have their greatest influence on the supply side.1 The dramatic increase in the quantity of sanitation services provided in this period resulted from changes in both demand and supply. In the first section we show that monies expended on water, sewage, and refuse disposal (both operating and capital expenditures) had a large effect on reducing mortality attributable to waterborne diseases. In the second section, we summarize our previous work on the relationship between epidemics and waterborne disease, which leads to a more extended discussion of demonstration effects in the third section. The fourth section presents evidence on the clustering of extraordinary expenditures on water and sewage treatment capital, while the final section provides a summary and conclusions.
The impact of sanitation expenditures on mortality By the end of the nineteenth century, it was widely accepted that sanitation services reduced mortality from certain causes – among these are typhoid fever, diarrhea, and dysentery. In order to assess the relationship between urban deaths
Municipal investment in sanitation captial
35
and municipal spending on sanitation, we collected mortality and expenditure data for the period 1899–1929 for all cities having a municipal water supply, populations over 100,000 in the 1920 Census, and nearly complete data on mortality experience and sanitation expenditures. This gives a sample of 48 cities. The mortality data come from various Bulletins of the Bureau of Labor Statistics and the Census Bureau and from the Census Bureau’s annual Mortality Statistics of Cities, which provided death-by-cause statistics. We constructed a waterborne death rate series that includes deaths attributed to typhoid fever, diarrhea, and dysentery.2 The data on municipal sanitation expenditures were published in various Census Bulletins and in Financial Statistics of Cities. We used data on annual operating costs and capital acquisition costs of water and sewage works.3 We used these data to estimate a fixed effects model of the determinants of mortality from waterborne diseases. The results, presented in Table 3.1, show Table 3.1 Waterborne disease death rate regression results Dependent variable is natural log of the Waterborne disease death rate Variable
Coefficient
t-ratio
Mean
WATKALL WATERAV3 SEWKALL SEWERAV3 REFUSE YEAR ASSDPC LANDAREA WAR LATE 20 R2 N obs.
–0.0009 0.0413 –0.0120** –0.0812 –0.0603** –0.0788** 0.0061* 0.00019 0.2634** –0.0810** 0.834 1,109
–0.53 1.66 –5.58 –1.17 –2.19 –20.02 2.00 0.81 7.29 –2.10
33.33 1.48 13.92 0.29 1.00 1,915.5 11.03 270.53 0.14 0.11
Notes **statistically significant at the 95% confidence level. *statistically significant at the 90% confidence level. WATKALL WATERAV3 SEWKALL SEWERAV3 REFUSE YEAR ASSDPC LANDAREA WAR LATE20
Sum of all capital expenditures on waterworks prior to the year under observation plus the value of municipal waterworks in 1899 (or in the year acquired) in per capita terms. Average operating expenditures on waterworks and water treatment over the two preceding years and the year under observation in per capita terms. Sum of all capital expenditures on sewage facilities up to the year under observation in per capita terms. Average operating expenditures on the sewer system over the two previous years and the year under observation in per capita terms. Average expenditures on refuse collection and disposal over the two preceding years and the year under observation in per capita terms. A trend variable, the year under observation. Assessed valuation in hundreds of dollars per person. Square miles in hundreds of square miles. A dummy variable equal to 1, if year = 1917–1920. A dummy variable equal to 1, if year = 1925–1927.
36
L. Cain and E. Rotella
that capital and operating expenditures on sanitation had a big effect on reducing mortality from waterborne causes after controlling for the long-term trend and the short-term impact of World War I. Using these regression results we calculate that a 1% increase in all categories of sanitation expenditures would have reduced the annual mortality rate from typhoid, diarrhea, and dysentery by 3% in the average size city in this period. Citizens had good reason to demand sanitation services, and cities received a big payoff to their investments in sanitation capital. Cities could and did buy themselves lower death rates.4
The effects of epidemics In the early nineteenth century, fear of fire promoted demand for urban water supplies. In the early twentieth century, it was fear of typhoid and cholera epidemics that promoted demand for water supply improvements. Widely circulated reports on the relationship between sanitation and disease by Chadwick (1842), Shattuck (reprinted 1948), and others led to an understanding that polluted water was related to disease, even before the germ theory explained the causation. Experiments with filtration at places such as the Lawrence Experiment Station, Massachusetts and experience with filtration in a number of European and North American cities convinced city dwellers that they were right to demand such improvements in their water works and sewage works. In a previous paper (Cain and Rotella, 2001), we employ a counting approach to investigate whether mortality crises (epidemics) caused changes in urban expenditures on sanitation by seeing if mortality shocks in cities were closely followed by notable expenditure increases. We define a mortality shock as a year in which the actual waterborne death rate was more than one standard error above its 1899–1929 trend in that city. The waterborne death rate shocks were grouped into 98 epidemic episodes (some lasting more than one year). They were heavily concentrated in the years 1906–10, with a few during World War I, and only three in the 1920s. An expenditure “response” is an analogously defined shock in the expenditure series (i.e. an expenditure more than one standard error above trend) which occurred within three years of a mortality shock. In 29 of the 98 epidemic episodes, there was no response from the affected city. In all other cases, an expenditure substantially above trend occurred within three years of the mortality shock. We, therefore, count responses in 71% of the observed elevated mortality episodes. We turned to the Engineering News to see if we could find reports of the mortality shocks and/or the municipal expenditure responses that we identified. Engineering News was a widely-read weekly journal of “civil, mechanical, mining, and electrical engineering” which regularly carried news of sanitation developments in the U.S. A typhoid outbreak might result in an article on causes and potential cures. The construction of a new water purification plant might be the subject of an essay complete with drawings and pictures. We found discussions in Engineering News of the waterborne disease shock in 14 of the 69 cases where there was a response (20%), and six of the 29 cases
Municipal investment in sanitation captial
37
where there was none (21%). In each case, the disease was typhoid fever, and the role of pollution was noted. We found a discussion of the response in 13 of the 14 cases where we found an article about the disease shock, and in 28 of the 56 cases where we were unable to find mention of the shock. In some of the six cases where we found mention of a disease shock to which there was no response, we found reasons given for why there was no response. So, although not every epidemic caused an expenditure response, we conclude that epidemics did often provide a “moment of crisis” to which cities responded by investing in sanitation projects. We contend that the success of water and then sewage treatment works in dealing with epidemics accounts for the fact that we observe so few mortality shocks in the later years of our sample. The public demanded continued development of water and sewage treatment works to insure that mortality rates did not return to their nineteenth-century levels.
Demonstration effects While epidemics of waterborne diseases played a critical role in stimulating public demand for sanitation services, demonstration effects caused important changes on the supply side. We define demonstration effects quite broadly as information that is obtained from elsewhere or from pre-construction experimentation within a city itself. There are many forms of demonstration effects – one of the most important being experiences of other cities. A famous example of a dramatic demonstration effect occurred in 1892 when there was a cholera outbreak in Germany. Hamburg experienced 17,000 cases of the disease (with 8,000 deaths) in a population of 640,000. In Altona, an adjoining city, there were only 500 cases (300 deaths) in a population of 150,000. The striking difference means that Hamburg suffered a rate of infection that was eight times the rate in Altona. Both cities drew their water supply from the Elbe, but Altona filtered its supply. As a result of this clear demonstration of the efficacy of filtration, Hamburg and other German cities began to upgrade their waterworks (Melosi 2000: 141). The episode revealed not only that filtration was effective, but it also provided information about the performance of the particular type of filters Altona had installed. Demonstration effects made such information available to cities, lowering the cost of information, thereby causing the supply curve of water and sewage services to shift to the right. The grand tour American engineers in the first half of the nineteenth century either received formal training at a school like the U.S. Military Academy at West Point or received informal, on-the-job training – many of them by working on the Erie Canal. By mid-century, some engineers began specializing in waterworks or, less commonly, sewers. Implementing a sanitation strategy required extremely large expenditures. Given the very high cost of sanitation capital projects, both engineers and the
38
L. Cain and E. Rotella
cities that hired them felt they had to “get it right.” While there was no foolproof way to be sure that a particular technology would be best for a particular set of local conditions, a good way to reduce the risk of making a costly mistake was to look at experiences elsewhere. Engineers and public officials toured sanitation facilities and published reports of their tours, thereby sharing the information more broadly. After he was hired as Chief Engineer to Chicago’s Board of Sewerage Commissioners in 1855, Ellis Sylvester Chesbrough submitted a report in which he referred to the sewers of New York, Boston, and Philadelphia, and showed that he was familiar, through his reading, with the sewers of London, Paris, and other European cities. At this time most U.S. cities had sewers, but none of them had a comprehensive sewerage system, so Chesbrough’s initial recommendations relied on his training and intuition. Very soon, he would add detailed empirical knowledge gathered when he was sent to Europe in 1856 to discover if the sewage disposal techniques used in several cities there were relevant to Chicago’s needs. The 1858 report of this trip, which covered 13 European cities, represents one of the first sanitary engineering treatises (Board of Sewerage Commissioners 1858; Cain 1972). He concluded that none of the cities furnished an exact criterion to judge the effects of disposing of sewage directly into the Chicago River, but he argued that their collective experience indicated that it would be necessary to keep the river free of sewage accumulations. Shortly after the Civil War, James P. Kirkwood, an eminent waterworks engineer, was engaged by the city of St. Louis. In December 1865, Kirkwood, a strong advocate of filtration, then a new technology, was sent to Europe to gather information. His report, published in 1869, details filters and filter galleries in 19 European cities (Kirkwood 1869). It was the only source in any language on municipal water supply filtration until Allen Hazen, another American engineer, published a report of his European trip in 1895 (Hazen 1895). At the time that Kirkwood’s report was published, no U.S. city had constructed a complete water filtration system though a filter basin had been built in Hamilton, Ontario in 1859, and a second was under construction in Newark, New Jersey. Kirkwood submitted a design for filters, but a political change in St. Louis led to a decision against filters. His report was, however, used by other U.S. cities wanting to copy European filtering techniques. Following the publication of his book, Kirkwood was hired as a filtration engineer in Poughkeepsie and Hudson, New York. Later, he became the consulting engineer for the Lowell and Lawrence Waterworks, both of which included filter galleries. Such filtration projects made a strong impression on American engineers, and the technology soon took hold: Immediately following the appearance of the report, a number of filter galleries and two basins were built in America: in 1870 at Whitinsville, Mass.; in 1871, at Schenectady, N.Y., Columbus, Ohio, Indianapolis, Ind., and Des Moines, Iowa; in 1872 at Lowell, Mass., and (a filter basin) at Waltham, Mass.; in 1874 at Decatur, Ill.; in 1875 at Brookline and Lawrence, Mass.;
Municipal investment in sanitation captial
39
in 1878 at Rutland, Vt.; in 1880 at Nashville, Tenn. and Ft. Wayne, Ind.; in 1888 at Green Island and Hoosick Falls, N.Y., and at Springfield, Ill.; in 1891 at Reading, Mass. This is not a complete list. After that, few natural filters were built in the United States. In Canada, Toronto built a filter basin in 1875. (Baker 1948: 279) A second wave of filtration facilities construction took place in the 1890s when perhaps 20 American cities built them. The most influential of these was the one Allen Hazen built in Albany, New York, in 1899, which served as the model for Washington, Philadelphia, and a number of other larger cities. Engineers such as Chesbrough, Kirkwood, and Hazen traveled widely throughout the U.S. in the role of consulting engineers. The information they collected and published was put to use in many projects, and this was made all the easier as transportation costs fell. By 1896, filtration technology was better known and understood when Pittsburgh ordained that a study should be made to investigate the relation of the city’s water supply to public health and to ascertain the desirability of sand filtration. The Filtration Commission hired Hazen, just returned from his European trip, to be the Consulting Engineer. Two gentlemen associated with the Lawrence, Massachusetts, water system were hired to be resident engineer and bacteriologist. Hazen brought his extensive knowledge to bear on the subject, including the influential work that George W. Fuller did on mechanical (rapid sand) filters in Louisville. The reduced cost of transportation and the development of American works meant that it was possible for the Commissioners themselves to make an inspection trip: The Commission as a body, on November 11, 1896, visited the city of Lawrence, Mass., for the purpose of inspecting the filtration beds in operation in that city, and on their return, devoted a day in the city of New York to the inspection of certain plants engaged in mechanically filtering private water supplies. On April 19 and 20, 1898, the Chairman of the Commission, accompanied by the Chairman of the Committee on Water Analysis, the Mayor of Pittsburgh, and the Resident Engineer, visited the cities of Louisville, Ky., and Cincinnati, Ohio, for the purpose of investigating the methods and results of the extensive experimental plants established in these cities, and also visited the city of Covington, Ky., and examined the water works of that municipality. (Filtration Commission of the City of Pittsburgh Pennsylvania 1899: 2) Publications By the 1890s, the weekly Engineering News was an important source of information about sanitation developments in the U.S. The construction of a new water purification plant might be the subject of a long essay complete with many drawings and several pictures. Engineering News was not afraid to take editorial
40
L. Cain and E. Rotella
stances. For example, in a 1907 article on the New York City water supply, the editors wrote: Time was when a city might hesitate before launching upon the expenditure necessary to provide a water purification plant, lest its outlay should not prove to give the desired results, or in giving them be burdensome on account of needless expense that might perhaps have been avoided by a delay that would make available some improved and less costly method of treatment. Fortunately, there is no longer reason for hesitation on this account. (Engineering News 1907: 556) One of the editors of Engineering News and an associate assembled a book entitled Sewage Disposal in the United States. In the Preface, Moses Baker and George W. Rafter note The chief object . . . of this book is to specifically call the attention of sanitary authorities, engineers, and others interested in questions of public sanitation, to the fact that we have already accumulated a considerable stock of experience in sewage disposal in this country, and that for the future Americans, who wish to study the subject in detail, will not be obliged, as until recently was the case, to go abroad for the purpose. (Rafter and Baker 1894: v) The book’s first 20 chapters discuss “Principles,” with numerous examples from American practice. The remaining 25 chapters present descriptions of particular works, with details of plants. Individual chapter titles include “Chemical Precipitation and Mechanical Separation at Long Branch, New Jersey” and “Intermittent Filtration and Broad Irrigation at South Framingham, Massachusetts.” Such books became the texts of the early twentieth century. This greater availability of information reduced the costs associated with developing water and sewage treatment works and thereby led to increased supply. Testing stations A testing station provides full-size, working models of water or sewage treatment technologies. The first was established in 1887 at Lawrence by the Massachusetts State Board of Health. A chemical laboratory was installed initially, and a bacteriological laboratory was added two years later. The construction of the Lawrence testing station marks what has been termed “the era of testing stations” (Greeley 1953); between 1899 and 1929, more than 30 major testing stations were built. The Lawrence Experiment Station had its roots in the work of two scientists involved with the Massachusetts Board of Health – William Thompson Sedgwick and Theobald Smith. Both men recognized that the germ theory tied specific organisms to specific diseases, that humans were carriers of infection, and therefore the state had to enter into a new relationship with individuals because
Municipal investment in sanitation captial
41
of the externalities issues involved (Rosenkrantz 1972). Sedgwick became an Assistant Professor of Biology at MIT in 1883, but there were few university resources for experimental work. In 1886, the Massachusetts legislature adopted a comprehensive water policy which required the Board of Health to adopt water pollution standards. This led to the establishment of the Lawrence Experiment Station because, first, it was recognized that one station would be more economical than multiple investigations by local water boards throughout the state, and, second, uniform scientific standards could be assured through the adoption of new methods using the microscope. The first task of the Lawrence Experiment Station was to determine the effect of filtration as compared to natural oxidation. The Lawrence facilities afforded opportunities that MIT could not provide. In the first two years, Sedgwick and his students were able to develop and apply techniques for identifying and quantitatively analyzing the microorganisms in both water and sewage. The studies set the standards in Massachusetts, other states, and other countries.5 In 1893, when a typhoid epidemic came down the Merrimack River, a slow sand water filter, designed at the Lawrence Experiment Station proved that polluted water could be made potable. Over the next 30 years, testing stations experimented with a wide variety of new technologies for both water and sewage treatment, determined safe operating loads for all sorts of sanitation works, and determined the effect of different raw water and raw sewage characteristics on processes and loads. The individuals who worked at these stations developed, tested, and spread the knowledge of new sanitary technologies and thereby became the successors to Chesbrough, Kirkland, and Sedgwick. Formal demonstration projects The testing station led to formal demonstration projects where the “experiment” was to build full-scale prototypes of particular technologies. Visits to other sites remained important, but field study at a particular site became much more important. The decision reached as a result of this study was that the cost was too high, the risk too great, and previous reported experience too slight for full-scale biological treatment to be recommended at the time. A prototype in the field . . . was to be built and operated for about 6 months to obtain detailed data for the final design and to obtain greater certainty that the earlier findings were valid. (Nemerow 1971: 177) Full-scale tests are demonstration projects that lie in the middle of the spectrum between pilot studies and the construction of multi-million dollar public works. The crucial questions are feasibility, design parameters, reliability, and cost efficiency. These are issues that usually can not be resolved in a laboratory; they require “in the ground” demonstration projects.
42
L. Cain and E. Rotella
The role of demonstration effects The excursion through several forms of demonstration effects indicates that, as time went on, the information used by cities making decisions about investing in water and sewer capital became more extensive and specific. The cost of information was falling, and cities used more information to help them “get it right.” In the middle of the nineteenth century, the Grand Tour enabled engineers to observe technology adopted by other cities. Pioneering cities like Chicago and St. Louis published reports of what their engineers had learned on their Grand Tours, thereby sharing information. By the end of the nineteenth century, publications such as Engineering News served the information dissemination role on a more immediate basis, and universities offered courses about sanitation. Textbooks for such courses were filled with case studies. The number of professionals working in the area increased, and they organized societies for sanitary engineers and public health workers (Melosi 2000: 114–15). The late nineteenth and early twentieth centuries were also the age of progressive reform politics and the municipal housekeeping movement. A large amount of information was created and disseminated in a short period of time. This was also the age of rapid technological change and experimentation. Experiments on new water and sewer technologies were often conducted in public sector facilities by highly qualified scientists, and the results were made available to the world. By the time of World War I, demonstration projects for handling industrial wastes were underway in Chicago as joint ventures of the Chicago Sanitary District and a number of private companies including the major Chicago meatpackers and the Argo Corn Products Company. In the middle of the nineteenth century, cities in search of sanitation information paid to send an expert to find the best practice technology. Fifty years later, they could find excellent information about water and sewage works in a textbook, and they could find much better data to help them decide whether a given technology was “best” practice. As the cost of information fell, cities were ever more willing to supply sanitation works to their citizens. And, as the complexity and cost of these works increased, they were willing to acquire more information to make sure they “got it right.”
The developing technologies of water and sewage treatment Water treatment At the turn of the twentieth century, the three most commonly used technologies to purify water were sedimentation, coagulation, and filtration (Baker 1901; Armstrong 1976; Melosi 2000). Sedimentation is short-term storage to remove suspended matter from water. Either the water rests or passes slowly through shallow settling basins where the sediment precipitates to the bottom of a reservoir from which it is removed. The amount of time involved depends on the quality of the water. Coagulation is the use of chemicals (often alum) to acceler-
Municipal investment in sanitation captial
43
ate sedimentation. It was usual to adopt filtration together with coagulation, but in some cities (e.g. Kansas City between 1900 and 1911) sedimentation and coagulation were used without filtration. Coagulation is never used alone, but always in conjunction with sedimentation and/or filtration. Filtration removes remaining suspended matter. It is not an effective technique by itself if there is a large amount of suspended matter because the filters have to be cleaned often. The popularity of filtration grew rapidly following the 1892 cholera epidemic when the dramatic difference in the experience of Hamburg and Altona convinced many of its effectiveness. Several types of filters were available in the early 1900s, and there was considerable disagreement among experts about the efficacy of the various types. Slow sand filters used beds of sand that rested on beds of gravel. Below the gravel bed was a system of collection pipes. This process eliminated almost all the bacteria from the water. Mechanical (or rapid sand) filtration used coagulants combined with rapid filtration. Mechanical filters processed 100–125 million gallons of water per acre of filter per day as opposed to two to three million gallons for slow sand filters (Melosi 2000: 139–45). Sand filters were laid in the ground, but mechanical filtration generally took place in wooden or steel tanks. Given the amount of water processed, mechanical filters clogged rapidly and required cleaning at least daily. The famed Lawrence Experiment Station began examining slow sand filters as early as 1893.6 Lawrence drew its water from the Merrimac River which was relatively clear but highly polluted. In the early years of the twentieth century, experiments on a scale comparable to actual practice were conducted by Cincinnati, Louisville, Philadelphia, Pittsburgh, New Orleans, and Washington in order to assess the relative merits of the two types of filters. All these cities drew their water supplies from rivers that had large amounts of suspended matter and sewage pollution. Providence, Rhode Island, experimented with mechanical filters. Its water, drawn from the north branch of the Pawtuxet River, was low in suspended matter, but polluted with domestic and industrial wastes. From the results of these experiments, plus evaluation of European experience, a consensus developed that slow sand filtration was more efficient for relatively clear water, while mechanical filtration was more efficient for silt-laden water. There was also general agreement that slow sand filters were preferred for smaller cities, while mechanical filters were preferred for larger cities. In addition to coagulants, chemicals were added to water for a variety of purposes. By the end of the nineteenth century, acceptance of the germ theory and widespread realization that some diseases were waterborne led to the use of chemical sterilization, usually chlorination. The decline in typhoid fever death rates is correlated with the use of hypochlorite (Melosi 2000: 144). Both slow and rapid sand filters made use of oxygen in the water, but heavily polluted water had little free oxygen. Aeration was used to improve the taste and odor of low-oxygen water. Hard water was softened through either chemical precipitation or ion exchange to remove iron, calcium, and magnesium. Table 3.2 uses information on municipally owned water supply systems from
Washington Louisville Council Bluffs
1863 1879 1883 1889 1890 1892 1893 1894
Pittsburgh Cincinnati New Orleans Nashville
Youngstown Washington Harrisburg Columbia
1905
1906 1907 1908
St. Louis
1904
1903
Cincinnati New Orleans
St. Louis Atlanta Harrisburg Columbia Charlotte
Kansas City
1899 1900 1902
Albany Kansas City Philadelphia
Cedar Rapids
Knoxville
Omaha
Coagulation*
1896
KnoxvilleKnoxville
Oshkosh Atlanta
Sedimentation
Year
Table 3.2 Dates of adoption of water treatment techniques
Pittsburgh Wilmington
New York City
Philadelphia Providence Washington Reading Yonkers
Albany
Altoona
Lawrence
Slow sand filtration
San Diego Cincinnati New Orleans Columbus
Youngstown Harrisburg Columbia
New York City
Philadelphia
Cedar Rapids Charlotte Norfolk
Oshkosh
Mechanical filtration
Columbus Omaha Charlotte
Harrisburg
Mobile
Chemical treatment
Springfield
1910
Note *Coagulation is never used alone, but always in conjunction with one or more other processes.
Source: U.S. Department of Commerce, Bureau of the Census p. 45.
1915
Trenton
1914
Baltimore Dallas Trenton St. Louis
Minneapolis
Minneapolis
Atlanta Toledo
1913
Springfield
Louisville Albany
Fort Worth
Dallas Trenton Albany
Louisville Pittsburgh Nashville Springfield Washington
1912
1911
Richmond
1909 Pittsburgh Milwaukee Cincinnati Kansas City Trenton Albany New York City Chicago St. Louis Detroit Wilmington Cedar Rapids Philadelphia Cleveland Louisville Hartford Buffalo Dallas Columbia
Nashville
46
L. Cain and E. Rotella
the General Statistics of Cities, 1915 to report dates at which cities adopted various techniques of water treatment. Cities drawing their water supplies from rivers are overrepresented in the table, while those drawing their water from fresh-water lakes are underrepresented. Cities most likely to be affected by upstream pollution were early adopters of water treatment. In 1900, a little over 6% of the nation’s water supply was filtered. By 1914, over 40% was. Eventually, almost every city would filter its water supply. Notice that in Table 3.2 the clustering moves diagonally from upper left to lower right. This represents the chronology of the development of water treatment technologies, and the larger clusters associated with the newer technologies show the adoption of advanced treatment methods by more and more cities. Of the 33 cities included in Table 3.2 during the years 1899–1912, 22 are among the 48 cities in our sample. For these 22 cities we have information on annual capital expenditures and annual operating expenditures on waterworks and water treatment. By examining the patterns of change over time in per capita outlays on capital and operations, we can observe the effect of adopting the new technologies on expenditures. In most instances we see that adoption of sedimentation and filtration shows up as a notable increase in capital expenditures whereas adoption of coagulation and chemical treatment leads a jump in operating expenditures. Table 3.3 presents water capital expenditures for our sample of 48 cities from 1899 through to 1929. Using annual capital expenditures by urban water departments, we identified those years when cities made particularly big investments. As was true in looking at epidemics, we defined a year of high expenditure as one in which a city reported per capita expenditure on water capital that was more than one standard error above the mean for the period. Often projects took multiple years to complete, but we include in Table 3.3 only the initial year of extraordinary expenditure for multi-year episodes. The capital expenditure data do not reveal what kind of capital the city bought. The money could have been spent on a variety of capital investments – e.g. building new filtration beds, extending the water delivery system to new neighborhoods, enlarging the main waterworks, building new pumping stations. We found notable expenditure increases which match the timing of adoption of new technologies reported in Table 3.2. This gives us confidence that expenditure data reveals dates when cities were adopting new water treatment technologies. We can see clustering patterns in Table 3.3 that suggest the influence of demonstration effects. New England cities are prominent among cities that make large capital expenditures early in the period. These cities were located in the same geographical area, they had easy access to the information being generated at the Lawrence Experiment Station, and they shared water sources and sewage discharge outlets. Two New Jersey cities located on the same river also made large early investments. The first decade of the twentieth century saw heavy investment in water capital by Midwestern cities (Ohio cities are particularly well represented) and by cities located on lakes. Few cities undertook large water capital investments during the years of U.S. involvement in World War I.
1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1923 1924 1924 1925 1926 1926 1927 1928 1929
Cincinnati Columbus Youngstown Chicago Spokane Pittsburgh Portland Jersey City
Seattle
Camden
Denver DATA NOT AVAILABLE DATA NOT AVAILABLE DATA NOT AVAILABLE Atlanta St. Paul Buffalo Toledo Boston Atlanta Reading Louisville Albany Milwaukee
Baltimore Yonkers Camden Washington Dayton Chicago St Louis Newark Fall River Newark
DATA NOT AVAILABLE DATA NOT AVAILABLE Baltimore Cleveland Chicago Dayton Hartford Rochester
Cambridge Worcester Lowell Washington DATA NOT AVAILABLE Atlanta Cambridge Salt Lake Buffalo Atlanta Milwaukee New Bedford Grand Rapids
Minneapolis Newark Kansas City Louisville Yonkers
Jersey City Salt Lake Hartford Worcester Springfield
Fall River
Trenton
Denver Worcester Fall River Detroit Toledo Syracuse Hartford Wilmington
Columbus
Rochester
Cleveland
Youngstown
Hartford
Youngstown
Trenton
Youngstown
Milwaukee
Portland
Nashville
Providence
Grand Rapids Milwaukee
Worcester
Youngstown Wilmington
Cambridge
Fall River Lowell Wilmington
Columbus Louisville Toledo Syracuse Rochester Trenton Minneapolis
Cincinnati Springfield New York Reading Lowell
Philadelphia Seattle
Philadelphia Richmond Washington Yonkers Spokane Wilmington
Kansas City Pittsburgh
New Bedford Trenton
Table 3.3 Years of high expenditure on water capital
Philadelphia
Seattle
Richmond
48
L. Cain and E. Rotella
The 1920s saw many cities make large expenditures updating and enlarging (or replacing) their original waterworks, extending their delivery systems, and adding capacity to water treatment facilities as cities increased in size and prosperity swelled municipal coffers. Sewage treatment Sewage treatment technologies developed at a later date than water treatment technologies (Melosi 2000: 167–73). Initially, sewage was either discharged into water or onto the land. In a world with relatively few industrial wastes, the oxygen contained in a moving body of water will help purify sewage. The same is not true of sewage discharged onto the land. At the turn of the twentieth century, the two major systems of land disposal were sewage farming (irrigation) and intermittent filtration. Sewage farming is very labor intensive and therefore prohibitively expensive in a high wage environment such the U.S. Intermittent filters are prepared beds of sand, cinders, and other porous materials under-drained by open-jointed tile conduits. As sewage passes through the filter, it is operated on by aerobic bacteria. Filtered sewage is discharged continuously for a period onto one bed, and then the flow is diverted to another bed so the first can drain. The sewage flow could be hours if the material in the bed is fine, or days if it is coarse. The Lawrence Experiment Station conducted experiments on intermittent filtration which led to its adoption in Massachusetts and elsewhere. Although intermittent filtration needed only 5% of the land needed by sewage farming, most cities considered the land area required to be a problem. To further reduce the land area used by intermittent filtration, cities increasingly used preliminary processes that would remove most of the suspended matter, whether organic or mineral. Both screening and settling were used for this purpose. Screening was considered to do a reasonably good job at preliminary treatment, but settling created problems because the sewage needed to stand a long time in costly reservoirs or tanks and emitted an offensive odor as the suspended matter decomposed. To deal with this problem, it became common to add chemicals to both accelerate the precipitation process and kill some of the bacteria. Chemicals by themselves left roughly half the organic matter for secondary decomposition and accumulated with the organic and mineral matter in the bottom of the tank. This mélange, known as sludge, still had to be disposed of. One solution was to modify the basic notion of the cesspool in which anaerobic bacteria are the working agent. In a septic tank, sewage passes through a preliminary grit chamber of sand or other material and the suspended organic matter is retained so that anaerobic bacteria can do its work. The (partially) clarified effluent goes on through. Sludge accumulates relatively slowly in the tank because much of the organic matter is either converted to gas, passes out in the water, or dissolves. The resulting sludge was more readily treated via aeration in devices such as the Imhoff tank, introduced in the first decade of the twentieth century.
Municipal investment in sanitation captial
49
In the absence of septic tanks, contact (filter) beds work much faster than intermittent filters. Instead of treating sewage in a single bed, contact beds use a series of multiple beds. First the sewage goes to a bed where the anaerobic bacteria do their work. Then it moves to a second (and perhaps a third) bed where the aerobic bacteria do their work. This cycle is repeated two or three times on a daily basis, with each bed given a period of rest between cycles. It was in the first decade of the twentieth century that many American cities reached a population size where sewage purification was recognized as necessary. Before 1909, most sewage filters were sand filters, although both Columbus and Reading had recently installed sprinkling (trickling) filters in which the sewage is sprinkled onto the bed, and Auburn, New York, was in the process of installing contact filters. Technology was changing rapidly during these years, and new technologies (e.g. activated sludge) were available within the next decade.7 Cities such as Chicago, which began investigating sewage treatment for industrial wastes during these years, did not build major works until after World War I. When they did build, the engineering of the works embodied the experience of the cities that had pioneered sewage treatment technology. In 1900, just 3% of the urban population lived in areas with sewage treatment. Twenty years later, the number had grown to 17.5%. Table 3.4 is constructed in the same way as Table 3.3. It presents years of particularly high sewer capital expenditure for the 48 cities in our sample. Again we see the early leadership of New England cities. The most striking result of the regression model reported in Table 3.1 is that investments in sewer capital had a large and statistically significant effect on reducing urban mortality from waterborne diseases in this period. Many cities responded to this fact by building and extending sewer systems and by adopting new treatment technologies. Over time more and more cities spent to increase their sewer capital although few cities undertook large capital projects during World War I. High expenditures on sewer capital are evident throughout the 1920s as many cities in all regions invested in treatment, particularly the activated sludge technology, which was the last major technology to be innovated.
Conclusion During the first 30 years of the twentieth century, American cities invested heavily in sanitation capital and experienced dramatic declines in deaths from diseases associated with bad water and poor sewage disposal. Death rates from typhoid fever, dysentery, and diarrhea fell by 88% going from 8.9% of urban deaths in 1902 to only 1.4% in 1929 (Cain and Rotella 2001: 139, 151). We have shown that a 1% increase in sanitation expenditure was associated with a 3% decline in the waterborne disease mortality rate. In this period, most sanitation expenditures were directed at the adoption of water and sewage treatment technologies; especially notable are the expenditures on sewer capital. We examined the role of epidemics and demonstration effects as causes of decisions by cities to undertake these expenditures. Empirical examination of
1916 1917 1918 1919 1920 1921 1922 1923 1923 1924 1925 1926 1927 1928 1928 1929
1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915
Lowell Pittsburgh Springfield
Providence
DATA NOT AVAILABLE Boston Cambridge Columbus Columbus Dayton Washington Reading Dayton Denver Jersey City Baltimore Spokane Louisville Seattle Wilmington Grand Rapids Louisville Salt Lake City Atlanta Hartford Portland Kansas City Nashville New Bedford DATA NOT AVAILABLE DATA NOT AVAILABLE Albany Baltimore Cincinnati Washington Wilmington Yonkers Rochester Salt Lake City Albany Columbus Hartford Youngstown Newark Spokane DATA NOT AVAILABLE Nashville Rochester Youngstown Detroit Fall River Jersey City Atlanta Cambridge Denver Richmond Worcester Youngstown Camden Lowell Syracuse Baltimore Buffalo Chicago Cleveland Dayton Grand Rapids Cambridge Chicago Cincinnati Atlanta Buffalo Camden Springfield Wilmington Worcester Cincinnati Reading Seattle
Boston Buffalo Lowell
Seattle St. Louis
New Bedford
Spokane Rochester Newark
Hartford
St. Paul
Youngstown Philadelphia Jersey City Nashville Columbus
Toledo New York Hartford Detroit Chicago
Providence Kansas City Portland Dayton
Minneapolis
Milwaukee
Spokane Louisville
Springfield Louisville Syracuse Louisville
New Bedford
Salt Lake City St. Louis
St. Louis
Youngstown
Youngstown
New Bedford Rochester
Newark
Salt Lake City Wilmington
Reading
Trenton
Worcester
Nashville
Toledo
Lowell
Syracuse
Table 3.4 Years of high expenditure on sewer capital
St. Paul St. Louis Toledo Lowell
Newark
St. Paul
Trenton Pittsburgh
Yonkers
Portland
Washington
Municipal investment in sanitation captial
51
epidemics reveals that many cities experiencing an episode of increased mortality from waterborne diseases responded with large expenditures on sanitation capital. Demonstration effects took place in a number of ways. In the middle of the nineteenth century, prominent engineers traveled to visit sanitation works in Europe and the U.S. Cities issued the engineers’ reports, which spread the news about treatment techniques and their results. Later, publications, such as the weekly Engineering News put out articles and editorials that were read by a growing group of sanitation professionals. In these ways the experience of cities with different techniques received wide publicity. Formal testing sites like the Lawrence Experiment Station undertook careful studies of new techniques as they developed and provided evaluations about what worked and what did not. Later some cities entered into partnerships with private industry to experiment with ways to handle wastes. Formal demonstrations projects have been, and are still, used to try out new methods before they are built on a large scale. It is difficult to find clear evidence of demonstration effects in our data on expenditures because we do not know what type of capital was purchased. Still, our examination of the dates when cities undertook major expenditures on sanitation capital reveals patterns consistent with reasonable conclusions drawn from the many specific instances of information sharing that we detail. Clusters of early adoption by New England cities and somewhat later adoption by Midwestern cities, and the flowering of investments in the 1920s suggest that cities were learning from each other and from published sources. We conclude that both epidemics and demonstration effects led to increases in sanitation services, but through different channels. We see epidemics as having their biggest impact on the demand side. Frightening episodes of high mortality from waterborne diseases caused citizens to demand improvements from their governments. Demonstration effects have their biggest impact on the supply side. Because the cost of sanitation capital was typically very large, cities wanted to “get it right.” Investing in the wrong technique could be an expensive, and politically disastrous, mistake. By lowering the cost of information about the ways that water and sewer treatment technologies work, demonstration effects increased the willingness and ability of cities to adopt those new technologies and thereby increased the supply of sanitation. Together these two forces created the conditions where American cities allocated ample amounts of money to dramatically diminish deaths in the early twentieth century.
Notes * We are grateful for the research assistance of Jaehee Choi, Supriya Mathew, Dong Eun Rhee, and Stacey Tevlin and to financial assistance from Indiana University, Loyola University Chicago, and the Center for Population Economics at the University of Chicago. Our considerable debt to George Alter is gratefully acknowledged. We benefited from the discussant and participants at the 2003 Social Science History meetings. We thank those who participated in both the Seminar on The Economics and Biodemography of Aging and Health Care and the Seminar on Health Promotion Economics at the University of Chicago. We thank those who participated in workshops at
52
1
2
3
4
5
6 7
L. Cain and E. Rotella
Northwestern University, Queen’s University, University of British Columbia, University of Southern California, and University of Toronto. Finally, we thank Thomas Weiss, Joshua Rosenbloom, and the other presenters and participants at the University of Kansas conference celebrating Tom’s retirement. The effects of epidemics and demonstrations of new knowledge and technologies cannot be completely bifurcated into demand or supply influences. The politicians who supplied sanitation services should be as aware of the consequences of epidemics as other citizens, and a city’s populace may be as aware as their politicians of developments in other cities. This group of diseases will be referred to as “waterborne,” even though water is not the exclusive means of transmission. They were spread by impure food, as well as water, and by contact with feces, flies, and other filth. Although much of the historical evidence on death-by-cause is notoriously problematic because the definitions of various diseases changed, as did their diagnoses, these three diseases were well identified in this period. Data on both finances and mortality are contained in Bureau of Labor Statistics Bulletin, #24, 30, 36, and 42, for the years 1899–1902, and Census Bulletin #20 for 1902–03. The Bureau of the Census published Mortality Statistics of Cities annually between 1905 and 1936 and Financial Statistics of Cities more or less annually between 1905 and 1931. Not every series was reported every year. Few direct figures are available for 1903, and Financial Statistics of Cities was not published in 1913, 1914, or 1920. The results presented here are analogous to and use the same techniques as in Cain and Rotella (2001). A similar regression model was estimated using the total death rate as the dependent variable. It did not find large and significant effects of variations in sanitation expenditures on the overall death rate. Before 1890, the relevance of specific pathogenic organisms was not readily apparent, so the Board of Health emphasized eliminating pollution as the proven method of disease prevention. In particular, although the typhoid bacillus was identified in 1880, there was a division among medical practitioners in Massachusetts as to whether the organism caused or merely accompanied the disease. Sedgwick’s team was able to determine how typhoid fever was transmitted (Rosenkrantz 1972: 103–5). This facility, initially under the leadership of Hiram Mills, helped train engineers Allen Hazen, George W. Fuller, and others. This involves combining raw sewage with a concentration of aerobic microorganisms. Air was then pumped into the combination to stimulate bacterial reduction. At the end, the activated sludge and the residual were removed, and the activated sludge recycled to be combined with new sewage. Experiments on this technique were made at the Lawrence Experiment Station in 1912 and at the Chicago Sanitary District in 1914. The first plant was constructed in San Marcos, Texas, in 1916.
Bibliography Armstrong, E.L. (1976) History of Public Works in the United States, Chicago: American Public Works Association. Baker, M.N. (1901) Municipal Engineering and Sanitation, New York: The Macmillan Company. —— (1948) The Quest for Pure Water: The History of Water Purification from the Earliest Records to the Twentieth Century, New York: The American Water Works Association. Board of Sewerage Commissioners (1858) Report on the Results of Examinations Made in Relation to Sewerage in Several European Cities, in the Winter of 1856–57, Chicago.
Municipal investment in sanitation captial
53
Cain, L.P. (1972) “Raising and Watering a City: Ellis Sylvester Chesbrough and Chicago’s First Sanitation System,” Technology and Culture, 13: 353–72. —— (1977) “An Economic History of Urban Location and Sanitation,” Research in Economic History, 2: 337–89. —— (1978) Sanitation Strategy for a Lakefront Metropolis: The Case of Chicago, DeKalb, Northern Illinois University Press. —— (1983) “To Annex or Not? A Tale of Two Towns: Evanston and Hyde Park,” Explorations in Economic History, 20: 58–72. —— and Rotella, E.J. (2001 “Death and Spending: Urban Mortality and Municipal Expenditure on Sanitation,” Annales de Démographie Historiques, 2001–1: 139–54. Chadwick, E. (1842, 1965) Report of the Sanitary Condition of the Labouring Population of Great Britain, Reprinted in Flinn, M.W. (ed.), Edinburgh: University Press. Cutler, D. and Miller, G. (2005) “The Role of Public Health Advances in Health Advances,” Demography, 42: 1–22. —— (2006) “Water, Water Everywhere: Municipal Finance and Water Supply in American Cities,” in Glaeser, E.L. and Goldin, C. (eds) Corruption and Reform: Lessons from America’s Economic History, Chicago: NBER and University of Chicago Press. Engineering News (1907) “Filtration for the Croton Water Supply, New York City,” November 21, 1907. Ferrie, J.P. and Troesken, W. (2008) “Death and The City: Chicago’s Mortality Transition, 1850–1925,” Explorations in Economic History, 45, 1: 1–16. Filtration Commission of the City of Pittsburgh, Pennsylvania (1899) Report, Pittsburgh. Greeley, S.A. (1953) “Testing Stations for Sanitary Engineering – An Outstanding Achievement,” Transactions of the American Society of Civil Engineers, Centennial Transactions, 2610: 574–8. Hazen, A. (1895) The Filtration of Public Water-supplies, New York: J. Wiley. Kirkwood, J.P. (1869) Report on the Filtration of River Water for the Supply of Cities as Practiced in Europe, New York: Van Nostrand. Melosi, M. (2000) The Sanitary City: Urban Infrastructure in America from Colonial Times to the Present, Baltimore: The Johns Hopkins University Press. Nemerow, N.L. (1971) Liquid Waste of Industry: Theories, Practices and Treatment, Reading: Addison Wesley. Rafter, G.W. and Baker, M.S. (1894) Sewage Disposal in the United States, New York: Van Nostrand. Rosenkrantz, B.G. (1972) Public Health and the State: Changing Views in Massachusetts, 1842–1936, Cambridge: Harvard University Press. Shattuck, L. (reprinted 1948) Report of the Sanitary Commission of Massachusetts, 1850, Cambridge: Harvard University Press. U.S. Bureau of Labor Statistics (1899–1902) Bureau of Labor Statistics Bulletin, #24, 30, 36, 42, Washington: Government Printing Office. U.S. Department of Commerce, Bureau of the Census (1902–03) Census Bulletin, #20, Washington: Government Printing Office. —— (1905–1929) Financial Statistics of Cities, Washington: Government Printing Office. —— (1905–1929) Mortality Statistics of Cities, Washington: Government Printing Office. —— (1916) General Statistics of Cities, 1915, Washington: Government Printing Office.
4
Profitability, firm size, and business organization in nineteenth-century U.S. manufacturing Jeremy Atack and Fred Bateman*
This chapter is a part of a broader investigation of the profitability of nineteenthcentury manufacturing, much of it associated with our honoree, Thomas J. Weiss, and his coauthor Fred Bateman (Bateman and Weiss 1975; 1981; Bateman et al. 1975). Our starting point is the observation in A Deplorable Scarcity (Bateman and Weiss 1981) that not only were average rates of return in manufacturing activities high (at least by comparison with rates of return for other sectors at about the same time and in comparison with returns on other kinds of assets) but that the rates of return being earned by larger producers (and especially by the very largest) were systematically lower than those earned by smaller firms. It is this latter relationship between firm size and profit which is our focus here – a topic that has been an enduring issue for historians and economists, particularly those concerned with industrial organization and market structure. In the nineteenth century, populist apologists appeared to assume that firm size was positively correlated with the profit being earned. For example, the Farmers’ Anti-Monopoly Convention meeting in Des Moines in August 1873 adopted a resolution that “all corporations are subject to legislative control; (and) [such control] should be at all times so used as to prevent moneyed corporations from becoming engines of oppression” (Dixon 1896: 25) and the Farmers’ Alliances championed cooperatives as an alternative to corporate control. Within this nineteenth-century environment, the assumed firm size–profit relationship was transformed into a corollary association between different organizational forms and the rate of return. Specifically, corporations, with their potential for enormous scale, were presumed to earn higher profit rates than the traditional proprietorships and partnerships and a new government agency, the U.S. Bureau of Corporations, was created in 1902–3 to deal with the challenges which they posed (Clark 1929: 3, 8). Corporations thus became intertwined with the size–profit connection, often through the nexus of monopoly. Questions of firm size, organizational forms and rates of return survived the populist era reemerging during the Great Depression as exemplified by the landmark work of Adolph Berle and Gardiner Means (Berle et al. 1932) who coined the phrase “the separation of ownership and control.” In their thesis, the dispersion of corporate ownership through public stock sales, partitive inheritance, and
Profitability, firm size and business
55
so forth had allowed corporate managers to usurp control of the firm despite the fact that the interests of those managers were not necessarily aligned with those of the stockholders. In particular, whereas owners preferred that the company profits be returned to them in the form of dividends, managers may have had other interests that led to more direct personal benefits for themselves.
Firm size and profitability: the twentieth-century experience There have been numerous empirical investigations of the relationship between firm size and profitability in the United States beginning after World War I when corporate income data first became available. Conclusions have been mixed. Many of the early studies found an inverse relationship between firm size and profitability akin to that which we find for the mid- to late-nineteenth century (Dewing 1921; Summers 1932; Epstein and Clark 1934). For example, H. B. Summers wrote “no matter where the dividing line is set, small companies, with only 4 exceptions, show higher rates of earnings [than] large, when considered by industries” (Summers 1932: 478), and Ralph Epstein concluded “larger manufacturing enterprises, in the main, do not earn profits at a higher rate than smaller ones” (Epstein and Clark 1934: 45). Studies of manufacturing activities in other countries have reported similar findings. Samuels and Smyth (1968: 139), for example, concluded that “profit rates and firm size are inversely related” based upon a panel of 186 British companies between 1954 and 1963. Various explanations have been proffered for these findings. A leading explanation has involved agency problems created by the separation of ownership from control in the modern large corporation which led to poorer financial performance among managerially-operated firms relative to those operated by their owners as Berle and Means suggested. Another modern-day explanation for the phenomenon is tax avoidance by owner-managers in situations where retained earnings, dividends and wages are taxed at differing rates. As McConnell (1945: 7) noted, “smaller corporations exist as much to provide an income in the form of a managerial wage to corporate officers who are owners as well as workers as to pay dividends to all stockholders.” Consequently, the form in which owner-employees take their compensation has a major impact upon the net revenues of the firm. In the nineteenth century, however, profit taxes and income taxes were non-existent (except during the Civil War decade) so it seems unlikely that tax avoidance would have played a role in encouraging entrepreneurs to favor one form of compensation over another.1 Variations upon this theme have been used as a critique of Bateman and Weiss’s treatment of the census data, for example, by Kenneth Sokoloff who claimed that entrepreneurial labor was not counted as a factor input in the census data and so, by implication, no allowance was made for this factor cost (Sokoloff 1984). However, in A Deplorable Scarcity, Bateman and Weiss (1981) use the same conjecture to justify looking at the large firm profit rates precisely because the rates of return to such firms are not sensitive to the form in which owners received their compensation.
56
J. Atack and F. Bateman
Many of the most recent studies of the relationship between firm size and the rate of return, on the other hand, have concluded either that there is no relationship or that it is positive (Crum 1934; 1939; Alexander 1949; Stekler 1963; 1964; Hall and Weiss 1967; Marcus 1969). Marcus, for example, reports a significant positive relationship between size and profitability in 35 of 118 fourdigit industry groups whereas only nine industries showed evidence of a strong inverse relationship. However, of course, these same results also imply that there was no statistically significant relationship between size and profits for firms in 74 (= 62%) of the 118 industries studied. However, while there is no consensus on the relationship between size and rate of return, virtually all studies find a much greater variability in returns among smaller firms than larger firms (Alexander 1949; Hymer and Pashigian 1962; Mansfield 1962; Samuels and Smyth 1968). We find the same relationship in the nineteenth-century data.
Estimating the profitability of nineteenth-century manufacturing firms The data underlying this study are described in detail elsewhere (Bateman and Weiss 1981; Atack and Bateman 1999) but, briefly, the samples are nationally representative random samples of several thousand industrial firms in each year collected from the extant manuscripts of the censuses of manufactures for the 1850, 1860, 1870 and 1880 census years. Most of the data for 1850, 1860 and 1870 were collected by Bateman and Weiss; those for 1880 were collected by Atack and Bateman.2 The data cover a firm’s activities during a one-year window beginning June 1 of the year that preceded the census year through May of the census year. Two caveats apply. First, not all census records have survived to be sampled. For example, those for many Ohio counties in 1860 and 1870 are missing (Atack 1985). Second, in 1880, the census designated a number of special agents who were industry experts to collect the data for a limited number of important industries – cottons, woolens and worsteds, iron and steel, silk, brewing and liquor distilling, glass, and coke – and to write reports on these industries for the published census. Consequently, enumerators were directed to leave the canvassing of firms in these industries to the special agents – and, for the most part, they did so. Unfortunately, these special agent enumerations have never been found (Delle Donne 1973) and so the industry mix in the 1880 sample is not representative of the population of all manufacturing firms in 1880 and there are too few large firms in the sample as these were heavily concentrated in the special agent industries. This biases our results in a way that only strengthens our argument. Our procedures for estimating profits are the same as those described by Bateman and Weiss (1975; 1981) but expanded to cover 1870 and 1880 and with the samples weighted to be nationally representative rather than simply representative of individual states. These procedures generate estimates of a
Profitability, firm size and business
57
firm’s profit rate that more closely approximate accounting, than economic, profits.3 These profits are expressed as a rate of return per dollar of capital where “capital” is the sum of reported “capital invested” from the census and an imputed estimate of working capital derived from applying the ratio of “live assets” to output in 1890 to output at the earlier censuses.4 Specifically, our rate of return is estimated as: Q–R–w–d–m = KF + KW where: Q = f.o.b. value of product(s); R = c.i.f. value of raw materials; w = annual wage bill; d = depreciation (based upon the 1890 allocation of capital in this industry and state between plant and equipment – plant depreciated over 50 years, machinery over 15 years); m = miscellaneous expenses (based upon the 1890 ratio of miscellaneous expenses to output in this industry and state); KF = invested capital reported by the census; KW = working capital (based upon the 1890 ratio of “live assets” to output in this industry and state). In order to be included in these profit calculations, firms had to report non-zero input and output values, have positive value-added, pay wages, and report the use of capital.
The distribution of profits by firm size After truncating outliers, the distributions of the rates of return in each year are shown in Figure 4.1.5 The distributions are skewed to the left but with a long right-hand tail relative to a normal distribution (shown overlaid) having the same mean and standard deviation as the actual distributions. The mode and the median rates of return are less than the means for each year. Many of the firms in the left-hand tail of the distributions are estimated to have sustained accounting losses on their operations during the census year. Those in the very left-most portion of the left-hand tail – firms whose rate of return was estimated to reached a 100% loss of working and invested capital – should, we claim, have gone bankrupt. Indeed, many businesses would likely cease operations before
58
J. Atack and F. Bateman Year = 1850
Year = 1860
Year = 1870
Year = 1880
0.25 0.2
Relative frequency
0.1 0.05 0 0.25 0.2 0.1 0.05 0 1
0
1
1 0 1 2 3 2 3 4 5 Percentage return on (capital + working capital)
4
5
Figure 4.1 Distribution of unweighted rates of return in manufacturing.
reaching that limit. Doubtless, some of these losses were more apparent than real and an artifact of data limitations. Specifically, in 1850 and 1860, the census reported “the average monthly amount paid for all the labor of all the hands, male and female, employed in the business or manufacture during the course of the year . . . so that by dividing [the average monthly wages] by [the average number of hands] the result will show the average [monthly] earnings of individuals” but without reporting the number of months of operation per year (United States, Census Office, Department of the Interior 1860: 27). Consequently, we estimated the annual cost of labor in these census years by multiplying the reported monthly wage bill by 12. This almost certainly overstates the wage bill of the average firm and especially the wage bill for the smaller firms which, in later years at least, were much less likely to work year-round than larger firms.6 In 1870 and 1880, on the other hand, the census reported the annual cost of labor for the firm. This difference in the way in which the data are reported given how we calculate the rate of return may account for some of the decline in the relative frequency of losses from 1850 and 1860 to 1870 and 1880. While there are differences in the sampling proportions across the census years and problems with missing data, sample sizes are sufficiently large that we have confidence in the results. More than 5,000 observations underlie the rate of return distributions for 1850 and 1860, and more than 7,000 observations in the 1880 distribution. The 1870 distribution has the fewest observations (3,985). These data can be summarized by a single number for each year – the arithmetic mean – shown by the vertical line in distribution of profits in each census year. It is also reported in Table 4.1 for broad census regions as well as for the nation as
Profitability, firm size and business
59
Table 4.1 Average rates of return by region, 1850–1880 (average of the rate of return earned by each firm) Region
1850
1860
1870
1880
Unweighted United States Northeast Midwest South West
0.210 0.177 0.251 0.200 1.672
0.206 0.192 0.203 0.249 0.233
0.376 0.318 0.417 0.470 0.399
0.349 0.301 0.375 0.423 0.348
Weighted by invested capital United States Northeast Midwest South West
0.129 0.104 0.187 0.158 1.092
0.153 0.158 0.131 0.148 0.217
0.113 0.109 0.127 0.063 0.113
0.112 0.096 0.137 0.138 0.083
a whole. Table 4.1 also shows the results of a different calculation in which each firm’s return on capital is weighted by the firm’s relative share of total invested capital. As a pairwise comparison of the weighted with the unweighted rates of returns Table 4.1 reveals the average weighted rate of return is systematically lower than the unweighted average. This implies that the firms with the larger weights – more capital – earned lower returns in general than those with smaller weights – less capital. This same result would also be true if we had weighted the returns by employment or even value of output rather than capital invested: bigger establishments, on average, earned lower returns than smaller plants. It is also robust to industry and location controls.7 One way to show the difference in average rates of return by size of firm is to partition the data between “small” and “large” firms – we picked $5,000 invested capital as our dividing line for reasons we will elaborate below (but the results are not particularly sensitive to the breakpoint) – and perform a t-test for the difference between mean rates of return. The differences between means are relatively large and statistically significant across the board, especially in 1870 and 1880 (Table 4.2). For example, in 1880, the average return for firms with less than $5,000 capital was 0.421 (= 42.1%) with a standard deviation of 0.481 while the average return for larger firms was just 0.145 (= 14.5%) with a standard deviation of 0.255. Assuming these two samples were drawn from distributions with equal variance, this difference between these two estimates has a t-value of 23.99, a result that is most unlikely to arise by chance. The relationship between profitability and size is not as strong in 1850 and 1860. We believe that this is a result of our necessary assumption that firms of all sizes and in all industries worked 12 months per year, thereby paying their workers 12 times the reported monthly wages. The available evidence, however, suggests that there was a secular trend toward year-round work and away from seasonal and
60
J. Atack and F. Bateman
part-time work during the course of the nineteenth century. Even so, in 1870 and 1880, the typical firm averaged only about ten months of work per year but larger firms were much more likely to work 12 months per year than smaller firms.8 As a result our assumption imparts a downward bias to the firm’s estimated profit rate, especially for smaller firms. The underlying relationship between rates of return and firm size is illustrated dramatically by the scatterplots shown in Figure 4.2. It does not require much imagination, especially in 1870 and 1880, to insert an “ocular” regression line that would be negatively sloped between the rate of return and invested capital in these scatterplots. It is also apparent from the heteroskedastic scatter of the Table 4.2 T-test for equality between means Year
Average rate of return (standard deviation in parentheses) “Large” firms (capital ≥ $5,000)
0.222 (0.447) 0.216 (0.426) 0.460 (0.576) 0.421 (0.481)
0.150 (0.297) 0.174 (0.330) 0.176 (0.289) 0.145 (0.255)
4.468 (0.000) 3.192 (0.001) 16.066 (0.000) 23.989 (0.000)
Year = 1850
Year = 1860
Year = 1870
Year = 1880
5 4 3 2 1 0 1
Invested capital (log scale)
Figure 4.2 Distribution of rates of return by invested capital.
100,000
10,000
1,000
100
100,000
10,000
1,000
5 4 3 2 1 0 1 100
Percentage return on (capital + working capital)
1850 1860 1870 1880
“Small” firms (capital < $5,000)
T-test for equal means (Prob > | t |)
Profitability, firm size and business
61
points that the variance of profits among small firms is greater than that among larger ones, with the variance falling sharply in each census year among firms with more than about $5,000 capital.9 We have also indicated the zero profit level represented by the horizontal line; firms below this line – and there were many in 1850 and 1860 but relatively fewer in 1870 and 1880 – are estimated to have sustained accounting losses, earning less from the sale of their output than they spent on wages, raw materials, and miscellaneous expenses, although in 1850 and 1860 this may reflect the attribution of 12-month labor costs to firms that actually incurred lower annual wage bills as discussed above. The capital figure reported in the censuses and shown on the x-axis in Figure 4.2 is “the aggregate amount of the capital, real and personal” (United States, Census Office, Department of the Interior 1860: 25). Unfortunately, the census failed to elaborate exactly how this was to be measured, igniting considerable debate and controversy both at the time (United States, Census Office 1883; United States, Census Office 1990) and since. Robert Gallman’s analysis, however, suggests that the reported figures are most likely market value or net reproduction cost (Engerman and Gallman 1986: 174). This investment could be drawn from a wide variety of sources including personal resources, the resources of family and friends, internally generated by the business itself, supplied by others doing business with the firm, or obtained from banks and investors.
Firm organization, profitability, and firm size Although each investor supplied essentially the same commodity to the firm – dollars to be repaid at some future date – the legal rights and obligations attaching thereto, and resulting from, each of these investments differed markedly depending upon organizational form of the firm and the relationship between the debtor and creditor. In general, banks were protected against legal liability for the actions of their borrowers by the nature of the loan contracts and custom.10 So too were businesses that simply extended trade credit to their customers.11 However, in the absence of state-provided limited liability protection, any personal investment in any business was fraught with financial danger and risk to the investor beyond the initial investment since the courts might find that such investors acted, in essence, as business partners. As “partners,” such investors would be jointly and severally liable for all the debts of the business. The state could offer limited liability protection, but typically did so early on only on a case-by-case basis through a special charter of incorporation, creating what Chief Justice John Marshall (1819) called “an artificial being, invisible, intangible, and existing only in contemplation of law.” These limitations on liability can be interpreted as a simple extension of the legal separation and distinction between persons and the “artificial being” – the business – created by the law and which the law imbued with otherwise personal rights such as legal standing, the right to own and transfer property, and so on (including, indeed, constitutionally protected rights). Other things equal, such protection of investors should have made it easier for businesses that were organized as corporations to raise more capital at lower cost.
62
J. Atack and F. Bateman
Most of the early laws establishing corporations were idiosyncratic and were passed in response to special interests and pleadings before state legislatures. This process inevitably limited access to those to whom this privilege was most valuable and who were able to absorb the costs associated with securing this special right (Hurst 1970; Hamill 1999). Eventually, however, states adopted general incorporation laws. New York was the first to offer a general right to incorporation beginning in 1811 (Howard 1938; Kessler 1940) but other states were slow to follow this lead. Pennsylvania, the next state to follow, waited until 1836 and Connecticut followed in 1837. There was, then, another lull until 1846. By the time the 1850 census was taken, however, 11 states had general incorporation laws on the books (Table 4.3). There were 24 by 1860; 39 at the 1870 census and, by 1880, every state (except Rhode Island) and every organized territory had such a law on its books (Hamill 1999). Although there are a number of studies of the growth of business corporations during the nineteenth century, all are of limited scope and scale (Blandi 1934; Evans 1948; Cadman 1949; Dodd 1954; Kuehnl 1959; Trusk 1960; Eilert 1963; Fundaburk 1963; Wilson 1965). Moreover, despite the growing availability of the corporate organizational form over time, corporations represented only a small fraction of firms – under 8% of a population of over half a million firms – at the end of the nineteenth century when the first official count was taken as part of the Twelfth Census (United States, Bureau of the Census 1902). Despite their small numbers, however, such businesses were disproportionately important, producing more than half of manufacturing output by the end of the nineteenth century (United States, Bureau of the Census 1902). The same was true earlier as we will show. All states imposed a wide variety of restrictions on these businesses. These included the minimum number of individuals who had to come together to establish a corporation (three or five was the most common number), limits upon the type of business which could be conducted (and upon the ability to change or Table 4.3 The growth of general incorporation laws, 1850–1880 Date
States and Territories with statutes on books
By the end of 1850
California, Connecticut, Iowa, Louisiana, Michigan, New Jersey New York, Ohio, Pennsylvania, Tennessee, Wisconsin In addition to those listed by 1850: Alabama, Florida, Illinois, Indiana, Kansas, Kentucky, Maryland, Massachusetts, Minnesota, Mississippi, North Carolina, Vermont, Virginia In addition to those listed by 1860: Arizona, Arkansas, Colorado, Maine, Missouri, Nebraska, New Hampshire, New Mexico, Nevada, Oregon, South Carolina, Washington, Utah, West Virginia, Wyoming In addition to those listed by 1870: Delaware, Georgia, Idaho, Montana, North Dakota, South Dakota, Texas
By the end of 1860 By the end of 1870
By the end of 1880 Source: Hamill 1999.
Profitability, firm size and business
63
add lines of business without securing a new charter), the duration of the corporate charter, stockholder liability for corporate debts, and the minimum capital to be invested (often with stipulations regarding how long the promoters had before all the authorized capital had to be paid in).12 Massachusetts, for example, stipulated that: For the purpose of . . . manufacturing business, except that of distilling or manufacturing intoxicating liquors, . . . three or more persons may associate themselves with a capital of not less than five thousand nor more than five hundred thousand dollars . . . with the intention to constitute a corporation. (Massachusetts 1870) Five thousand dollars seems to have been the most commonly accepted minimum capital requirement – for example, this was the standard in Massachusetts, New Hampshire and Vermont – which is why the distributions were split at this point in Table 4.2 and Figure 4.2 but it was by no means universal. Connecticut and Wisconsin, for example, had a $4,000 minimum, while New Jersey set the minimum capital for incorporation at $10,000 (Cadman 1949; Dodd 1954; Kuehnl 1959). New York’s law, by contrast, set no minimum.13 A detailed summary of the status of corporation law at the beginning of the twentieth century covering 51 states and territories shows that 14 still had minimum capital requirement then – perhaps as a result of a “race to the bottom” in corporate charter-mongering (Grandy 1989). In New York and Vermont, the minimum was set at just $500; Maine, Massachusetts, Michigan, Minnesota, and New Hampshire set the minimum at $1,000; Alabama, Connecticut, Delaware, Missouri, Nevada, and New Jersey set the minimum at $2,000, while Louisiana required a minimum of $5,000 (Frost 1905). The corporation, however, was only one of a variety of different organizational forms that a business could adopt and which the legal system recognized and regulated.14 We have endeavored to identify the organizational form of firms in the census of manufacturing samples based upon their “style” – the manner in which the names of the businesses were recorded. These names are available on the original census schedules but have only been preserved for those firms which were sampled before the early 1980s. Prior to then, the data were transcribed from the original census schedules (or microfilms of the schedules) onto worksheets for coding and eventual data entry onto 80-column Hollerith punch cards. Although the names were not keypunched, we still have the original worksheets and have used these to encode business style. Unfortunately, the 1880 data (collected by Atack and Bateman) as well as subsequent expansions of the earlier datasets to generate the nationally representative samples were entered directly into personal computer databases without the intermediate transcription step. Firm names for these observations were not, for the most part, transcribed and cannot easily be recovered. A firm was classified as a sole proprietorship if the business name was simply that of an individual, such as John Smith. A firm was classified as a partnership
64
J. Atack and F. Bateman
if the names of two or more persons were listed (such as Smith and Jones) or implied (such as Smith and Co. or Smith and Sons), and a firm was classified as a corporation if it had a generic, impersonal name such as Boott Mills, Merrimack Co., or the Prattville Manufacturing Company. This categorization is relatively crude and almost certainly subject to error. For example, in each state in which we know there was a minimum capital requirement for general charter of incorporation, our procedure classifies a small number of firms below these thresholds as corporations (see Figure 4.3) although we clearly do much better than a random assignment of organizational forms. Despite the coarseness of our classification, there seems to be a systematic relationship between organizational form (classified according to the procedure described above) and capital invested (Figure 4.4). Most firms classified as proprietorships were small. Average capital invested in these firms was $2,752 in 1850, $3,478 in 1860, and $3,840 in 1870. Those categorized as partnerships were larger with an average capital approximately three to five times greater ranging from $7,851 in 1850 to $19,435 in 1870 and with average capital increasing as the number of separately listed names increased. Those firms which we have labeled corporations were, on average, the largest with an average capital investment of almost $70,000 in 1860 and over $178,000 in 1870.15 In 1850, we estimate that corporations made up less than 2% of the business population and about 3% in 1860 and 1870. However, they produced about
0.4
0.4
0.3
0.3
0.2
0.2
0.1 0.05
0.1 0.05
100,000 100 1,000 Connecticut and Wisconsin firms classified as corporations
100 1,000 100,000 Massachusetts, NH and VT firms classified as corporations
0.4 0.3 0.2 0.1 0.05 100 1,000 100,000 Size distribution of New Jersey firms classified as corporations
Figure 4.3 Distributions of corporations in states with size limits.
Profitability, firm size and business Partnership
Corporation
0.4 Fraction
Fraction
0.4 0.3 0.2 0.1 0.05
0.3 0.2
0.1 0.05
100 1,000 100,000 Capital Proprietorship 0.4 Fraction
65
100 1,000 100,000 Capital
0.3 0.2 0.1 0.05 100 1,000 100,000 Capital
Figure 4.4 Firms by organizational type and capital.
Table 4.4 Rate of return by implied organizational form and year Year
Sole proprietorship
Partnership
Corporation
Unweighted 1850 1860 1870
0.210 0.209 0.410
0.205 0.200 0.281
0.145 0.196 0.214
Weighted by invested capital 1850 1860 1870
0.140 0.162 0.172
0.145 0.150 0.122
0.089 0.142 0.064
15% of output in 1850 and 1860 with more than a quarter of the capital invested in manufacturing and by 1870 produced almost a quarter of the nation’s manufacturing output with almost 36% of the manufacturing capital. Given the size distributions of firms by organizational form and what has already been said regarding firm size and rates of return, it should not be surprising to find that rates of return also varied systematically by organizational form in each census year (Table 4.4). Average returns were highest for sole proprietorships and lowest for corporations although these differences were small in 1860.
66
J. Atack and F. Bateman
Corporation profits: agency problems or lower capital costs? The lower average profit rates which we estimate were being earned by corporations are consistent with agency theory where the firm’s managers are interested in their jobs and salaries whereas the firm’s owners are concerned with profits and the market value of their business (Kamerschen 1968; Monsen et al. 1968; Palmer 1973). There exist, however, alternative explanations that are equally consistent with this observation. In particular, if larger firms in general and corporations in particular enjoyed preferential access to capital and had lower risks of capital loss which resulted in lower capital costs, then the profit-maximizing firm under these conditions would also earn a lower rate of return. Such preferential access to capital existed because the risks to potential investors from such investments were lower because the investor was less likely to be considered a partner, because larger firms were more likely to be important customers for suppliers and thus better positioned from a bargaining standpoint to secure trade credit and because they were more likely to be owned by the business elite who had better connections to the banking community and others with surplus cash to invest. As a result, these firms would substitute capital for other factors of production and the profit-maximizing firm would increase use to the point where the marginal return per dollar was equal to the capital cost to the firm. Impersonal equity markets for manufacturing enterprises (incorporation notwithstanding) were still in their infancy in this country before the end of the nineteenth century. For example, the premier market for industrials in the twentieth century, the New York Stock Exchange, listed 228 stocks as being traded in 1890, but only a handful of these are identifiable as equity in manufacturing firms (Navins and Sears 1955; Snowden 1987; Baskin 1988) including the American Cotton Oil Company, American Tobacco (preferred), Edison General Electric, Joliet Steel, the National Lead Trust, the National Linseed Oil Company, Pullman, and the Southern Cotton Oil Company. Rather, most of the stocks being traded were railroads. In contrast, as early as 1835, the Boston Stock Exchange listed prices for 17 manufacturing stocks. By the 1850s, prices for 37 companies, most of them cotton textile mills were being reported and by 1869, the number had grown to 48 (Atack and Rousseau 1999). But even as late as 1880, only 50 firms were listed (Rousseau 1999). Moreover, contrary to Berle and Means observation of widely dispersed stock ownership by the 1920s, stock ownership in the nineteenth century remained highly concentrated. A leading Boston broker of the day described the market as an “exclusive” one, “for it is almost exclusively in the hands of certain capitalists, who have no desire to sell when it is up, and can afford to hold it when down” (Martin 1871). The closeness with which most shares were held must have severely constrained management’s discretion in pursuing objectives at wide variance with those of the stockholders. Family and business associates, rather than large institutional investors and an undifferentiated mob, held whatever shares were issued. Moreover, “the small number of shares . . . in each corporation (as par is generally $1,000,) prevents the market from being supplied
Profitability, firm size and business
67
with what is technically called ‘floating stock,’ and completely exempts the securities from any speculative activity” (Martin 1871). Indeed, the various restrictions and privileges conferred by corporate charters created a segmented capital market in which larger firms generally, and corporations specifically, were likely able to borrow at substantially lower cost than self- or family-financed smaller ones. Unfortunately, there is no direct evidence on this in manuscript census samples and only scattered (and often indirect) evidence in general (McGouldrick 1968; Dalzell 1987; Lamoreaux 1994; Homer and Sylla 2005). If, however, larger firms and especially corporations had lower capital costs than other firms, we would expect them to make more intensive use of the relatively cheaper factor of production and thus be more capital intensive. The available statistical evidence is strongly supportive of this proposition. Work by Cain and Paterson (1981; 1986) on the American system of manufacture, for example, concludes that American technology in the late nineteenth and early twentieth century was capital-using and laborsaving and recent work by Atack et al. (2007) has shown that larger firms (“factories”), especially those using steam power, were more capital-intensive than smaller plants. Indeed, the evidence from the manuscript census samples is overwhelming on this point: larger firms in each census had significantly higher capital–labor ratios than smaller ones and those firms which we have identified as corporations had significantly higher capital–labor ratios than partnerships or sole proprietorships (Table 4.5). On average, the difference in the mean capital–labor ratios was at least 2:1 and usually higher. Each pair also fails the test that the sample estimates of the capital–labor ratio were drawn from the same distribution.16 Since sole proprietorships by definition (presumably) relied upon the managerial skill of just one person, individual variations in entrepreneurial skill and talent may explain some of the high variance in the returns to small firms. Moreover, small producers had one distinct advantage over larger firms – they were in a much better position to exploit small, localized economic opportunities than were larger firms. These may have enabled the prescient firms to earn a substantial return to their investment.17 But, since the investment itself was small even the most successful small business owners did not get particularly rich very quickly. Whatever the market signal that was conveyed by high or low profits, for most firms in the nineteenth century, profits often served not as a means of attracting external investment but rather as the source of investment funds themselves, especially for smaller firms as these relied heavily upon self-financing absent access to bank loans, trade credit and impersonal capital markets. High profits thus imply higher rates of growth for the firm; lower profits imply relatively lower growth rates and losses should have shrunk the firm’s capital stock (Baumol 1959; Penrose 1959). If large (and, therefore, longer lived and historically more successful) producers earned lower rates of return than smaller firms, then smaller ones were growing more rapidly and could hope to displace the currently larger firms at some future date, other things equal.
$645 (15.10) $816 (39.20) $1,529 (343.40)
By organizational form Sole proprietorship Partnership Corporation $846 (19.60) $1,196 (54.40) $2,154 (239.70)
$602 (10.20) $2,090 (66.90)
1860
$866 (34.20) $1,406 (71.50) $5,790 (2,645.10)
$526 (9.17) $2,900 (287.90)
1870
$564 (8.20) $2,158 (67.25)
1880
Notes Women are weighted 0.6 of an adult male; children are weighted 0.5 of an adult male. These ratios are approximately equal to the wages of each relative to those of an adult male. Means have been rounded to the nearest dollar; standard errors to the nearest 10 cents.
$487 (8.50) $1,881 (131.50)
By invested capital Under $5,000 capital invested $5,000 and over capital invested
1850
Table 4.5 Capital–labor ratios by firm size and by organizational form, 1850–1880 (dollars per adult male-equivalent) (standard error of mean)
Profitability, firm size and business
69
Other things, however, were not necessarily equal. Leaving aside the proposition that profits might be serially correlated – as would be the case, for example, if the business acumen of the entrepreneur (or managerial team) mattered to the success of the firm – the observed differences in the rates of return and the variability of those returns by firm size have interesting implications for the firm’s growth. If profits were purely stochastic (that is, if firms operated by good entrepreneurs could not consistently earn more than those operated by bad entrepreneurs) and ignoring any role for the business cycle in determining the course of profits over time, we can simulate the cumulative possible effects of the resulting random earnings on a firm’s capital. The model is thus an application of Gibrat’s Law (Kalecki 1945; Simon and Bonini 1958; Evans 1987). It assumes (1) that no outside injections of capital were made (through, for example, the exercise of calls against stockholders), (2) that all earnings were reinvested in the business (so that no dividends were paid out but rather that the firm’s owners received their returns in the form of capital gains), and (3) that once accumulated net losses exhausted invested capital, the firm was declared bankrupt and ceased operation immediately.18 We have simulated firm growth for small and large firms through random draws from a normal distribution of profit rates that have the same means and variances as the profit data shown in Table 4.1 and represented by the overlays in Figure 4.1 to estimate the probability that a firm would be bankrupted. Based upon this simulation, we estimate that a small firm – one with under $5,000 capital in 1850 – had about 0.5% probability of bankruptcy within one year, 5% after five years, and 7% after ten years (Table 4.6). These rates of failure are relatively low – probably too low – reflecting our overly cautious definition of Table 4.6 Probability of bankruptcy and growth among small and large firms, 1850–1880 – an application of Gibrat’s Law Year 1850 After 1 year After 5 years After 10 years 1860 After 1 year After 5 years After 10 years 1870 After 1 year After 5 years After 10 years 1880 After 1 year After 5 years After 10 years
“Small” firms (capital ≤ $5,000)
“Large” firms (capital > $5,000)
0.0045 0.0529 0.0713
0.0009 0.0177 0.0260
0.0045 0.0516 0.0688
0.0002 0.0100 0.0153
0.0032 0.0215 0.0247
0.0002 0.0058 0.0100
0.0009 0.0057 0.0064
0.0000 0.0007 0.0016
70
J. Atack and F. Bateman
bankruptcy as occurring only when accumulated net losses exceeded the firm’s entire invested capital. A less restrictive “stopping rule” would have more firms dropping out sooner. These probabilities are also biased downwards to the extent that profits are positively correlated with entrepreneurial ability and thus are serially correlated. If, for example, we had set the maximum losses before a firm ceased operation at, say, half its invested capital, then failure rate after one year remains almost unchanged, but after five years almost 17% would have gone out of business and the ten-year failure rate would have been about 20%. Among larger firms given the various parameter values from 1850, the one year failure rate was only nine in 10,000; after five years, 1.8%; and after ten years, 2.6%. That is, the probability of failure due to bankruptcy among larger firms was only about a third of that among smaller firms. By 1870, the probability that a larger firm would not be around a decade later was only 1% and only 16 per 10,000 in 1880 – rates that were one-half or less than the rates among smaller firms. On the positive side, small firms in 1850 had a 6% chance in the first year of, at least, doubling their invested capital if they retained all their earnings, a 57% chance of doubling over five years, and an 82% probability over a decade. For larger firms, these probabilities were lower: 1.7% after one year, 53% after five years and 80% after ten years. However, when the average small firm doubled its capital, it remained relatively small whereas a doubling of the capital invested in one of the larger firms was of greater economic consequence and importance helping generate the concentration of economic power to which Berle and Means drew attention (Berle et al. 1932). Bankruptcy, however, was not likely the principal source of firm mortality in the nineteenth century – certainly for sole proprietorships and partnerships. Rather, the greatest risk came from the mortality of the entrepreneur or a partner. Such an event also brought death to the business that was organized as a sole proprietorship or a partnership. In contrast, even the limited 20-year time horizon offered by some of the early corporate charters must have seemed attractive relative to the likelihood of death among the principals of businesses organized in other ways. Indeed, the partnership may have been the “worst of all possible worlds” since many partners simply compounded the likelihood of at least one death among the principals rather than diminishing it. For example, Coale’s model life tables suggest that a 30-year-old man had a 17.3% probability of dying before age 40; at age 35, the probability of dying in the next ten years was 20.7% – better than one-in-five – and almost one-in-four for a man aged 40.19 Compared with these mortality rates, the simulated firm survival rates look positively rosy regardless of firm size. Public policy in the late nineteenth century focused increasingly upon the concentration of economic power and equated the evils of monopoly with excessive rates of returns. Our results however suggest that larger firms had lower rates of return per dollar invested but likely could afford to because they had a lower cost of capital as a result of better access to trade credit, banks, and, above all, private capital as a result of their ability to organize as a corporation with limited liability protection for stockholders. As a result of the lower cost of
Profitability, firm size and business
71
capital, these firms tended to have higher capital–labor ratios and lower capital productivity, observations which are well-documented. As a result of the lower variance in their profit rates, they should have been less likely to experience bankruptcy or other financial embarrassment but, at the same time if they were dependent upon internally-generated financing for growth, they would have grown more slowly than smaller businesses which were, on average, more profitable. Nevertheless, large firms and especially those organized as corporations came to dominate manufacturing industry. One important reason why, lay not in the separation of ownership from control but in the separation of the firm’s mortality from that of its mortal human owners as Chief Justice John Marshall correctly observed. While his definition of the corporation as “an artificial being invisible, intangible, and existing only in contemplation of law” is widely quoted, his decision in Dartmouth College v Woodward continues it possesses only those properties which the charter of its creation confers upon it, either expressly or as incidental to its very existence . . . Among the most important are immortality, and, if the expression may be allowed, individuality; properties by which a perpetual succession of many persons are considered as the same, and may act as a single individual. They enable a corporation to manage its own affairs and to hold property without the perplexing intricacies, the hazardous and endless necessity of perpetual conveyances for the purpose of transmitting it from hand to hand. It is chiefly for the purpose of clothing bodies of men, in succession, with these qualities and capacities that corporations were invented and are in use. (Marshall 1819) This advantage has proved decisive and enduring.
Notes * We thank Lou Cain and Joshua Rosenbloom for their helpful comments on earlier drafts of this chapter. Any remaining errors are ours alone. 1 For example, an income tax was introduced in 1861 and expired in 1871. See Howe (1896). 2 The data from the 1850–70 Census of Manufactures were collected by Fred Bateman, James D. Foust and Thomas J. Weiss under grants from the NSF: GS-2450, GS-2456, SOC 75–18917 and SOC 75–20034. Collection of the 1880 data was funded by NSF grants SES 86–05637 to Jeremy Atack and SES 86–09392 to Fred Bateman. 3 In particular, we make no effort to account for the opportunity cost of capital. Although our definition of profits is closer to an accounting profit than economic profits, we hesitate to label them “accounting profits” since we cannot be sure that all out-of-pocket or traditionally-imputed accounting expenses such as taxes, interest on debt, or depreciation are counted. Certainly, these were not separately reported by the Census prior to 1890. 4 Our use of nationally representative samples which weight locations by the relative importance of manufacturing in each addresses an important complaint by Niemi (1989) and by Vedder and Gallaway (1980) regarding Bateman and Weiss’ estimates for 1850 and 1860.
72
J. Atack and F. Bateman
5 We truncated the profit estimates at –1 (= 100% loss of the reported invested capital plus imputed working capital) and +5 (= 500% return on reported invested capital plus imputed working capital). The number of profit estimates rejected because they fell outside this range was small: 72 (= 1.4%) in 1850, 53 (= 1.0%) in 1860, and 29 in both 1870 (= 0.7%) and 1880 (= 0.4%). The vast majority of these were on the low end: 70 in 1850; 50 in 1860, 27 in 1870, and 24 in 1880; and the net effect of truncating the distributions was to raise the average rate of profit. On the other hand, some of these deleted estimates are clearly in error due to mistakes in the original or transcribed data. For example, in 1850, the losses to one firm are estimated to be more than 52 times as large as the firm’s reported invested capital plus imputed working capital. 6 Elsewhere we have estimated that firms worked only 254 days (out of a possible 309 days) in 1870 and 261 days in 1880. See Atack et al. (2002). 7 For example, in a regression of the rate of return against capital (in thousands of dollars) with industry, state, and census year controls, the regression coefficient on capital was –0.0004 with a standard error of 0.00008. This variable has more explanatory power than any other in the regression save the census year dummies for 1870 and 1880. Including a squared capital term improves the fit and each coefficient separately has a greater statistical significance, with the linear term negative and larger than before and the squared term positive, indicating that the decline in the rate of return with firm size eventually bottoms out and the very largest firms may have somewhat higher returns than those firms which were somewhat smaller. The same holds true if size is measured by employment or gross output but the effect is most pronounced when size is proxied by invested capital. 8 The 1870 and 1880 censuses suggest that firms worked an average of between 9.94 and 10.11 months a year with larger firms much more likely to work year-round than smaller firms, See Atack et al. (2002). If firms in 1850 and 1860 worked similar times (and it is likely that these are upper rather than lower bound estimates at earlier dates), then by assuming that workers worked 12 months per year, we bias estimated profit rates downwards especially for smaller firms thus reducing the differences between the rates of return for small and large firms. 9 Using robust regression to correct for the heteroscedastic errors slightly reduces the negative slope coefficient on size without overturning the “ocular” conclusion that there is a negative relationship between the rate of return and firm size and that this relationship generally becomes more pronounced with time. 10 Indeed, one might usefully think about bank depositors and bank stockholders as different kinds of investors in a mutual fund whose assets were the bank’s loan portfolio. Indeed, according to Lamoreaux (1994) state-chartered banks in New England often served as financing conduits for bank insiders to fund their other, non-bank business activities during the antebellum period. The move to national banking charters after the Civil War encouraged short-term working capital loans, often in the form of selfliquidating commercial paper rather than longer-term fixed capital investment. See Davis (1965). 11 Moreover, the expectation that the business would, in turn, extend trade credit to its customers also probably meant that trade credit for most businesses was, on balance, only a minor source of capital even if it was essential in greasing the wheels of commerce. 12 Although most people immediately think of limited liability as the principal advantage of the corporate form, this principle was not immediately and universally adopted. Rather, the key advantage from the start was the separation of personal mortality from corporate mortality. See quote from Marshall (1819) below. Limited liability simply emphasized and reinforced this separation. 13 Eric Hilt (Department of Economics, Wellesley College), personal communication, 5/1/2007. The requirements of the laws in other states with respect to capital as of the various census dates are not known at this time.
Profitability, firm size and business
73
14 The other, far more common business organizational form that was subject to state regulation was the partnership. See Lamoreaux (1995) and Lamoreaux and Rosenthal (2005). 15 Although corporations were in general larger than sole proprietorships and partnerships, this was not necessarily the case. For example, the giant meat packers, Swift and Armour, remained partnerships until the 1880s and Standard Oil was originally organized as a partnership. 16 Production theory suggests several reasons why firms might use different factor proportions: (1) firms use different technologies, (2) firms face the same production function but the function is non-homothetic so that the expansion path skews, leading to different factor proportions depending upon scale, or (3) firms face different factor prices as a result of market imperfections. By estimating a single production function, one implicitly rejects proposition (1). With factor prices unobserved in our data, we cannot test proposition (3) directly. However, we do observe that firms employed different factor proportions at different scales, consistent with proposition (2) but not a sufficient condition to explain the pattern. If, though, we can show that the production function was homothetic, then the only explanation for the observed difference in factor proportions at different scales given the same technology is that firms of different sizes faced different factor prices. We can test whether or not the production function is homothetic by using a translog specification: ln Y = 0 + 1 ln K + 2 ln L + 31(ln K)2 + 32(ln L)2 + 33(ln L)(ln K) This functional form of the production function is homothetic if:
31 = 32 = – 33/2 Van Gelder (1994) has used this property to argue that large and small Indonesian textile firms must have faced different capital prices since the production function was homothetic and large and small firms used the same technology. Elsewhere (Atack et al. 2004; Atack et al. 2007), however, we have argued that large (factories) and small (artisanal) firms in the same industry during the nineteenth century used different technologies. The latter used hand tools and skilled labor while the former substituted inanimately-powered machinery that was made and maintained by skilled labor but tended by unskilled labor. Absent the same technology and a homothetic production, we cannot make inferences about unobserved factor prices facing firms of different sizes. 17 Firms with greater degrees of monopoly power, defined as firm size relative to the geographic market, had, on average, higher rates of return. 18 Such a bankruptcy rule violates the strict assumptions in Gibrat’s Law of proportionate growth. 19 See Coale et al. (1983), Model West, level 6.
References Alexander, S. S. (1949) “The Effect of Size of Manufacturing Corporation on the Distribution of the Rate of Return,” Review of Economics and Statistics, 31(3): 229–235. Atack, J. (1985) Estimation of economies of scale in nineteenth century United States manufacturing. New York, Garland Pub. Atack, J. and F. Bateman (1999) “Nineteenth-Century American Industrial Development Through the Eyes of the Census of Manufactures: A New Resource for Historical Research,” Historical Methods, 32(4): 177–188.
74
J. Atack and F. Bateman
Atack, J. and P. L. Rousseau (1999) “Business Activity and the Boston Stock Market, 1835–1869,” Explorations in Economic History, 36(2): 144–179. Atack, J., F. Bateman, and R. A. Margo (2002) “Part Year Operation in NineteenthCentury American Manufacturing: Evidence from the 1870 and 1880 Censuses,” Journal of Economic History, 62(3): 792–809. Atack, J., F. Bateman, and R. A. Margo (2004) “Skill Intensity and Rising Wage Dispersion in Nineteenth-Century American Manufacturing,” Journal of Economic History, 64(1): 172–192. Atack, J., F. Bateman, and R. A. Margo (2007) “Steam Power, Establishment Size, and Labor Productivity Growth in Nineteenth-Century American Manufacturing,” Vanderbilt University. Baskin, J. B. (1988) “The Development of Corporate Financial Markets in Britain and the United States, 1600–1914: Overcoming Asymmetric Information,” Business History Review, 62(2): 199–237. Bateman, F. and T. J. Weiss (1975) “Market Structure before the Age of Big Business: Concentration and Profit in Early Southern Manufacturing,” Business History Review, 49(3): 312–336. Bateman, F. and T. J. Weiss (1981) A deplorable scarcity: the failure of industrialization in the slave economy. Chapel Hill, University of North Carolina Press. Bateman, F., J. Foust, et al. (1975) “Profitability in Southern Manufacturing: Estimates for 1860,” Explorations in Economic History, 12(3): 211–231. Baumol, W. J. (1959) Business behavior, value and growth. New York, Macmillan. Berle, A. A. and G. C. Means (1932) The Modern corporation and private property. New York and Chicago: Commerce clearing house inc. Blandi, J. G. (1934) Maryland business corporations, 1783–1852. Baltimore, The Johns Hopkins Press. Cadman, J. W. (1949) The corporation in New Jersey: business and politics, 1791–1875, Cambridge, Harvard University Press. Cain, L. P. and D. G. Paterson (1981) “Factor Biases and Technical Change in Manufacturing: The American System, 1850 1919,” Journal of Economic History, 41(2): 341–360. Cain, L. P. and D. G. Paterson (1986) “Biased Technical Change, Scale, and Factor Substitution in American Industry, 1850 1919,” Journal of Economic History, 46(1): 153–164. Clark, V. S. (1929) History of manufactures in the United States. New York [etc.], Published for the Carnegie Institution of Washington by the McGraw-Hill Book Company Inc. Coale, A. J., P. G. Demeny, and B. Vaughan (1983) Regional model life tables and stable populations. New York, Academic Press. Crum, W. L. (1934) The effect of size on corporate earnings and condition; an analysis of 1931 income tax statistics. Boston, Mass., Harvard University Graduate School of Business Administration Bureau of Business Research. Crum, W. L. (1939) Corporate size and earning power. Cambridge, Mass., Harvard University Press. Dalzell, R. F. (1987) Enterprising elite: the Boston Associates and the world they made. Cambridge, Mass., Harvard University Press. Davis, L. E. (1965) “The Investment Market, 1870–1914: The Evolution of a National Market,” Journal of Economic History, 25(3): 355–399. Delle Donne, C. R. (1973) “Federal Census Schedules, 1850–80: Primary Sources for
Profitability, firm size and business
75
Historical Research,” National Archives and Record Service Reference Information Paper, 67. Dewing, A. S. (1921) “A Statistical Test of the Success of Consolidations,” Quarterly Journal of Economics, 36(1): 84–101. Dixon, F. H. (1896) State railroad control, with a history of its development in Iowa. New York, T. Y. Crowell & Company. Dodd, E. M. (1954) American business corporations until 1860, with special reference to Massachusetts. Cambridge, Mass., Harvard University Press. Eilert, J. W. (1963) Illinois business incorporations, 1816–1870, Urbana, Ill.: University of Illinois Ph.D thesis. Engerman, S. L. and R. E. Gallman (1986) Long-term factors in American economic growth. Chicago, Ill., University of Chicago Press. Epstein, R. C. and F. M. Clark (1934) Industrial profits in the United States. New York, National Bureau of Economic Research in cooperation with the Committee on Recent Economic Changes. Evans, D. S. (1987) “Tests of Alternative Theories of Firm Growth,” Journal of Political Economy, 95(4): 657–674. Evans, G. H. (1948) Business incorporations in the United States, 1800–1943. [New York], National Bureau of Economic Research. Frost, T. G. (1905) A treatise on the incorporation and organization of corporations created under the “business corporation acts” of the several states and territories of the United States. Boston, Mass., Little Brown. Fundaburk, E. L. (1963) “Business Corporations in Alabama in the Nineteenth Century,” Columbus, Oh., Ohio State University Ph.D thesis. Grandy, C. (1989) “New Jersey Corporate Chartermongering, 1875–1929,” Journal of Economic History, 49(3): 677–692. Hall, M. and L. Weiss (1967) “Firm Size and Profitability,” Review of Economics and Statistics, 49(3): 319–331. Hamill, S. P. (1999) “From Special Privilege to General Utility: A Continuation of Willard Hurst’s Study of Corporations,” American University Law Review, 49: 81–180. Homer, S. and R. E. Sylla (2005) A history of interest rates. Hoboken, N.J., Wiley. Howard, S. E. (1938) “Stockholders’ Liability Under the New York Act of March 22, 1811,” Journal of Political Economy, 46(4): 499–514. Howe, F. C. (1896) Taxation and taxes in the United States under the internal revenue system, 1791–1895; an historical sketch of the organization, development, and later modification of direct and excise taxation under the Constitution. New York, T.Y. Crowell & Company. Hurst, J. W. (1970) The legitimacy of the business corporation in the law of the United States, 1780–1970. Charlottesville, Va., University Press of Virginia. Hymer, S. and P. Pashigian (1962) “Firm Size and the Rate of Growth,” Journal of Political Economy, 70(6): 556–569. Kalecki, M. (1945) “On the Gibrat Distribution,” Econometrica, 13(April): 161–170. Kamerschen, D. R. (1968) “The Influence of Ownership and Control on Profit Rates,” American Economic Review, 58(3:1): 432–447. Kessler, W. C. (1940) “A Statistical Study of the New York General Incorporation Act of 1811,” Journal of Political Economy, 48(6): 877–882. Kuehnl, G. J. (1959) The Wisconsin business corporation, Madison, Wis., University of Wisconsin Press.
76
J. Atack and F. Bateman
Lamoreaux, N. R. (1994) Insider lending: banks, personal connections, and economic development in industrial New England, Cambridge, England and New York, Cambridge University Press. Lamoreaux, N. R. (1995) “Constructing Firms: Partnerships and Alternative Contractual Arrangement in Early Nineteenth-Century American Business,” Business and Economic History, 24(2): 43–72. Lamoreaux, N. R. and J.-L. Rosenthal (2005) “Legal Regime and Contractual Flexibility: A Comparison of Business’s Organization in France and the United States during the Era of Industrialization,” American Law and Economics Review, 7(Spring): 28–61. McConnell, J. L. (1945) “Corporate Earnings by Size of Firm,” Survey of Current Business, 25(May): 6–12. McGouldrick, P. F. (1968) New England textiles in the nineteenth century; profits and investment, Cambridge, Mass., Harvard University Press. Mansfield, E. (1962) “Entry, Gibrat’s Law, Innovation and the Growth of Firms,” American Economic Review, 52(5): 1023–1051. Marcus, M. (1969) “Profitability and Firm Size: Some Further Evidence,” Review of Economics and Statistics, 51(1): 104–107. Marshall, J. (1819) Trustees of Dartmouth College v. Woodward, 17 U.S. (4 Wheat.) 518. Martin, J. G. (1871) Seventy-three years’ history of the Boston Stock Market, from January 1, 1798 to January 1, 1871; with the semi-annual dividends paid from commencement of the Boston banks, insurance, railroad, manufacturing, and miscellaneous companies. Also the prices of American gold, government securities, state, city, and railroad bonds, and miscellaneous stocks, Boston, Mass., Joseph G. Martin. Massachusetts (1870) “An Act Concerning Manufacturing and Other Corporations,”, Chapter 224. Monsen, R. J. C., John S. Chiu and D. E. Cooley (1968) “The Effect of Separation of Ownership and Control on the Performance of the Large Firm,” Quarterly Journal of Economics, 82(3): 435–451. Navins, T. and M. Sears (1955) “The Rise of a Market for Industrial Securities, 1887–1902,” Business History Review, 29(2): 105–138. Niemi Jr., A. W., (1989) “Comment and Debate: Industrial Profits and Market Forces. The Antebellum South,” Social Science History, 13(1): 89–107. Palmer, J. P. (1973) “The Profit Performance Effects of the Separation of Ownership from Control in Large U.S. Industrial Corporations,” Bell Journal of Economics, 4(1): 293–303. Penrose, E. T. (1959) The theory of the growth of the firm. New York, Wiley. Rousseau, P. L. (1999) “Share Liquidity and Industrial Growth in an Emerging Market: The Case of New England, 1854–1897,” NBER Development of the American Economy, working paper 117. Samuels, J. M. and D. J. Smyth (1968) “Profits, Variability of Profits and Firm Size,” Economica, 35(138): 127–139. Simon, H. A., C. P. Bonini (1958) “The Size Distribution of Business Firms,” American Economic Review, 48(September): 607–617. Snowden, K. A. (1987) “American Stock Market Development and Performance, 1871–1929,” Explorations in Economic History, 24(4): 327–353. Sokoloff, K. L. (1984) “Was the Transition from the Artisanal Shop to the Nonmechanized Factory Associated with Gains in Efficiency? Evidence from the U.S. Manufacturing Censuses of 1820 and 1850,” Explorations in Economic History, 21(4): 351–382.
Profitability, firm size and business
77
Stekler, H. O. (1963) Profitability and size of firm. [Berkeley, Calif.], University of California Berkeley, Institute of Business and Economic Research. Stekler, H. O. (1964) “The Variability of Profitability with Size of Firm, 1947–1958,” Journal of the American Statistical Association, 59(308): 1183–1193. Summers, H. B. (1932) “A Comparison of the Rates of Earning of Large-Scale and Small-Scale Industries,” Quarterly Journal of Economics, 46(3): 465–479. Trusk, R. J. (1960) “Sources of Capital of Early California Manufacturers, 1850–1880,” Urbana, Ill., University of Illinois PhD thesis. United States, Bureau of the Census (1902) Twelfth census of the United States, taken in the year 1900. Manufactures, Washington, D.C., U.S. Census Office. United States, Census Office (1883) Compendium of the tenth census (June 1, 1880), Washington DC, Government Printing Office. United States, Census Office (1990) The statistics of the wealth and industry of the United States, embracing the tables of wealth, taxation, and public indebtedness; of agriculture, manufactures, mining, and the fisheries: with which are reproduced, from the volume on population, the major tables of occupations: compiled from the original returns of the ninth census (June 1, 1870), under the direction of the Secretary of the Interior, Reprinted: New York, Norman Ross Publ. United States, Census Office, Department of the Interior (1860) Instructions to U.S. Marshals. Instructions to Assistants: Eighth Census, United States. – 1860, Washington DC, George W. Bowman. Van Gelder, L. (1994) “Industrial Agglomeration and Factor Market Segmentation with Empirical Applications to Indonesia,” Cornell University, August: x, 211 leaves PhD thesis. Vedder, R. K. and L. E. Gallaway (1980) “The Profitability of Antebellum Manufacturing: Some New Estimates,” Business History Review, 54(1): 92–103. Wilson, T. L. (1965) “Florida Business Corporations, 1838–1885,” Urbana, Ill., University of Illinois PhD thesis.
5
Railroads and local economic development The United States in the 1850s Michael R. Haines and Robert A. Margo*
The United States experienced a “transportation revolution” in the nineteenth century (Taylor 1951). The development of canals and other navigable waterways and, especially, railroads, linked together far flung factor and product markets and stimulated economic growth through division of labor and exploitation of regional comparative advantage (North 1961; Goodrich 1961; Ransom 1967, 1970; Williamson 1974; Haites et al. 1975). Economic historians have measured the aggregate social savings of the railroads in the nineteenth century (Fogel 1964; Fishlow 1965) but have paid less attention to measuring impacts of gaining rail access at the local economic level, despite the fact that contemporaries at the time were vitally interested in such effects and that economic theory provides useful guidance as to what these effects might have been. In this essay we examine various economic impacts of gaining rail access in the 1850s at the local level, where “local” means “county.” We take as our point of departure previous work by Craig et al. (1998) and Coffman and Gregson (1998), both of whom used cross-sectional antebellum data to show that, at a point in time, proximity to a rail line was positively associated with higher agricultural land values. Two closely related economic frameworks, the Heckscher– Ohlin model of international trade, and the Von Theunen model of agricultural land rents, provide the economic logic to account for this cross-section pattern. These frameworks predict other economic outcomes besides the rise in land rents, and it is on these ancillary predictions that we primarily focus. We measure rail access at the county level using information derived from maps – specifically, whether a rail line passed through the county in 1850 or 1860 (or both). Rail access data are linked to economic outcomes in 1850 and 1860. We use a “difference-in-difference” (DID) approach, comparing outcomes in a treated group (counties that gain rail access in the 1850s) with a control group, before and after treatment (rail access). Our analysis yields diverse results on the impact of the rail in light of the two frameworks just mentioned. As in previous work, we find that rail access did lead to higher land prices, both absolutely and relative to farm wages, but the effects were small. We also find that gaining rail access is associated with a decrease in agricultural yields, a smaller share of improved acreage in total land area, a higher proportion urban, and a lower likelihood of participation in agriculture.
Railroads and local economic development
79
Railroads and American economic development Although plans for railroads were first discussed in the United States in the early 1800s, it was not until the late 1820s that rail expansion actually took place.1 The first American railroads were tramways used in mining and quarrying, such as the so-called “Granite Railroad” that commenced operations in Quincy, Massachusetts in 1826. But railroads in the modern sense really originate with the struggles of port cities like Baltimore, Boston, and Charleston that lacked adequate inland waterway connections that would enable greater volume of trade with the hinterland. By 1840 some 3,300 miles of track had been laid (of which about 2,800 miles were in operation), the majority of it in New England, the mid-Atlantic, and South Atlantic states, and almost all of it involving trips of short duration (“short line”). Further expansion in mileage took place in the 1840s, much of it again in New England, and also in New York. The South and Midwest were largely bypassed during this decade, except for the completion of a rail line linking Savannah and Chattanooga, and a rail line through Ohio from Sandusky to Cincinnati. It was in the 1850s that the United States experienced its first great wave of rail expansion (Stover 1978). Approximately 22,000 miles of track were laid during the decade, bringing the total mileage on the eve of the Civil War to over 30,000. Although the federal government had been involved in railroad expansion prior to 1850 in an indirect way by providing land surveys free of charge (from 1824 to 1838, when the law authorizing the surveys was repealed), direct subsidies in the form of land grants were first voted in 1850, and later extended several times during the decade. By 1860, in addition to substantial coverage in the Northeast, rail lines criss-crossed Illinois, Indiana, and Ohio, with significant penetration into Wisconsin and Iowa. The South was less well served, but it too experienced substantial growth in rail access in the 1850s. As Stover notes: The decade prior to the Civil War had been one of the most dynamic periods in the history of American railroads. In 1850 a broken skein of short iron lines served the area between Maine and Georgia with a few stray strings of rail connecting the Ohio River and the Great Lakes. By the eve of the Civil War more than 30,000 miles of railroad served quite adequately all the states east of the Mississippi and few areas of substantial population were much removed from the Sound of the locomotive whistle. (1978: 232) Economic historians have devoted considerable attention to measuring the aggregate impact of railroads on nineteenth-century American economic growth. Fishlow (1965) focused on the antebellum expansion, particularly on whether railroads were “built ahead of demand” and on total factor productivity growth. Fogel (1964; see also Fishlow 1965; Williamson 1974; Kahn 1988) measured the social savings of the transportation of goods by railroad – that is, the savings in resource cost associated with shipping goods by rail as opposed to the next
80
M.R. Haines and R.A. Margo
best alternative. The consensus of scholarly opinion was that the railroad’s impact, while economically significant, was insufficient to deserve the label of “indispensable,” as Fogel put it. Although contemporaries were certainly aware of the aggregate benefits of transportation improvements, many were more concerned with the impact at the local level. Economic historians have not wholly neglected the measurement of these local effects, although less attention has been paid to them than to the aggregate social savings. The two specific studies that motivate our work are Craig et al. (1998) and Coffman and Gregson (1998). Craig, Palmquist, and Weiss used maps to infer whether a county had water or rail access in 1850 and 1860, and then linked their transportation data to census data, which provided information on land values in agriculture and their potential determinants (other than transportation access). Craig, Palmquist, and Weiss estimated cross-sectional regressions of the log of the average per acre value of farmland on, among other determinants, dummy variables for water and rail access. Controlling for other factors, land values were 15 percent higher in 1850 if the county had rail access; in 1860 the effect was somewhat smaller (around 8 percent). Coffman and Gregson used tax assessment records for Knox County in Illinois in 1855 to measure the impact of distance to a rail line that had recently been completed. Their cross-sectional regression implies that capital gains resulting from rail access equaled about 9 percent of the value of land in the county in 1850. Both studies interpret their findings in light of economic observations from the era and of modern economic models of the impact of transportation access on local economic activity (see also Atack and Passell 1994: 167, 169). Craig et al. (1998: 173) note, for example, that contemporaries “recognized the potential impact of improved transportation of marketable agricultural production on the value of agricultural land” and, in some cases, attempted to measure the impact using the rudimentary techniques then available. Coffman and Gregson (1998: 196) explicitly motivate their analysis using a version of the Von Theunen framework discussed in the next section. However, both of these studies focus on land values and do not assess other impacts; that is, neither study investigates the broad range of effects – for example, the impact on agricultural output or the share of labor in agriculture (see the next section) – that should occur if either the Hekscher-Ohlin or Von Theunen models are, in fact, appropriate frameworks in which to interpret how the railroad expansion affected local economies.
The local economic impact of rail access: frameworks for analysis We sketch two inter-related frameworks for evaluating the impact of a railroad on a local economy. The first framework is simply the familiar Heckscher–Ohlin two sector model of trade, in which there are specific factors to each sector as well as a mobile factor, labor. The second is the Von Theunen model in which
Railroads and local economic development
81
distance to a central location where trade takes place plays a key role in the determination of land rents. Our presentation emphasizes the impacts that would occur if, prior to gaining railroad access, the local economy had a comparative advantage in agriculture. This is consistent with the data on rail access discussed in the next section, which show that the largest gains in access in the 1850s occurred in states that were disproportionately agricultural, such as the Midwest (compared with, for example, New England states). As a point of departure, we imagine a local economy with total land area T, a fraction of which, T, has been “cleared” or is otherwise available to be used with labor, LA, to produce an agricultural good, QA = F(LA, T). There also exists a local stock of capital, K, that can be combined with labor to produce a nonfarm good, QM = G(LM, K). We assume that F and G are constant returns to scale. Households in this local economy are endowed equally with amounts of L, K, and T. Initially the local economy is closed to trade, so equilibrium is defined by local supply equaling local demand. With conventional utility functions this will ensure an interior solution so that positive amounts of both the farm and nonfarm good are produced. Prior to trade we assume that the local economy has a comparative advantage in producing the farm commodity and would, if it could, import the non-farm commodity. That is, letting p be the external (relative) price of the farm commodity and p* be the local relative price, p* < p – the local economy would experience gains from trade if it could trade with other local economies, near or far. Next, imagine that a railroad is built that gives farmers in the local economy access to external trade. Because the relative price of the farm commodity is higher externally, farmers now have an incentive to export their crop and import non-farm goods produced elsewhere. In the short run, output expands in the farm sector as labor shifts from non-farm to farm production. The magnitude of the increase in output, however, is constrained by the fixed supply of T – there are decreasing returns in the short run. The influx of labor into farm production drives up the marginal product of land. Because both the marginal product of land and the relative price of agricultural goods are now higher, the agricultural yield (output per acre) has increased. As a result, the rental price of land is higher. Conversely, the marginal product of capital declines as labor leaves nonfarm production. In the standard model, the gross effect on the rental price of land will dominate the net effect on the mobile factor, and the wage relative to the rental price of land will decline. The standard Heckscher–Ohlin framework takes the stock of land as given. However, our second framework – the Von Theunen model – makes it clear that the rise in relative output price creates an incentive to bring additional land into agricultural production. To illustrate this point, we assume that each farmer has a single unit of land which is combined with labor to produce the farm good, using a fixed coefficient production technology, Q = min (1, L/b). Once produced the good must be transported (by wagon, say) to a “central place” where production of the non-farm good and trade (farm for non-farm) occur. The fixed
82
M.R. Haines and R.A. Margo
coefficient technology will ensure a linear “bid rent” curve giving the maximum willingness to pay for a unit of land at a given distance d from the central place: R = PQ – wL – cdQ By setting R = 0, we define the margin of cultivation, d* = (p – wb)/c. It follows directly that dd*/dp > 0 – the margin of cultivation expands when the relative price of the agricultural good increases, and consequently, land used in agriculture as a share of total land area increases (d > 0). To summarize, our empirical analysis examines the following implications of rail access [d(rail) > 0] as derived from these two frameworks: d(w/r) < 0: the wage/rental ratio declines (Heckscher–Ohlin) d(yield) > 0: agricultural yields increase (Heckscher–Ohlin) d(LA/L) > 0: agriculture’s share of local labor supply increases (Heckscher– Ohlin) d > 0: land in agriculture/total land area increases (Von Theunen)
Data and estimation Our empirical analysis is based on a panel data set of counties for 1850 and 1860. This data set merges the information originally collected by Craig et al. (1998) on railroad access with a revised version of the ICPSR census dataset (Haines and ICPSR 2006) and county-level data on farm wages collected from the manuscript schedules of the federal censuses of social statistics of 1850 and 1860 (Margo 2000). The ICPSR data were augmented by information on urban population, county area, and more complete agricultural census data. Because the wage data are not available for every state (or for every county for states with surviving records) the sample is not fully representative of the antebellum United States in the geographic sense. Within the matched data set we also restrict our attention to counties that did not change land area between 1850 and 1860. The total sample size is 672 counties from 14 states (see Table 5.1, Panel B). Margo (2000) collected wage data from the 1850 and 1860 manuscript censuses of social statistics. The unit of observation in the manuscripts is the minor civil division (or, in the case of some cities, the ward); for each county, we constructed an equally weighted average of the sub-county figures. The monthly wage data pertain to contracts in which board was routinely included. However, the census also collected information on the “weekly cost of board to laboring men” from which it is possible to impute a monthly value to board, which we add to the money wage. Hence, our adjusted farm wage is defined as follows: adjusted farm wage = reported money wage + 4.3*weekly cost of board
Railroads and local economic development
83
The census also reported the daily wage of common labor, “without board” (that is, contracts in which board was not included). Margo (2000: ch. 4) demonstrates that, once one accounts for the unemployment risk premium associated with daily (as opposed to monthly) work, differences in county averages between farm and non-farm wages for unskilled labor appear to have been negligible, consistent with the hypothesis that (free) labor could (and did) move freely between the farm and non-farm sectors at the local level before the Civil War. Craig, Palmquist, and Weiss used maps to determine whether a county had access to water or rail transportation in 1850 and 1860. Here, “access” means that a rail line or navigable waterway passed through the county boundaries in the given year (or, in the case of water transportation, the county bordered on a navigable waterway, such as the Atlantic Ocean or one of the Great Lakes). Panel A of Table 5.1 shows that slightly more than 22 percent of the sample counties had access to a railroad, in this sense. Rail access more than doubled between 1850 and 1860 – nearly 45 percent of the sample counties had access to a railroad on the eve of the Civil War. Rows 3–5 of Panel A show access to the railroad in 1860, conditional on not having access in 1850. Overall, among the 531 counties that lacked rail access in 1850, 30.1 percent gained access by 1860. Rows 4–5 demonstrate that gaining access was correlated with existing access to water transportation; counties that
Table 5.1 Railroad access in 1850 and 1860 Panel A: Full sample of counties (constant land area)
All counties in 1850 All counties in 1860 Counties gaining rail access, 1850–1860 No rail or water access in 1850 Water access but no rail access in 1850
Number of counties in sample
Percent with rail access (counties equally weighted)
Percent with rail access (counties population weighted)
672 672 531 300 221
22.4% 45.8 30.1 33.7 25.2
46.9% [70.8%] 66.3 [82.3%] 38.8 42.9 33.6
Notes Sample consists of counties in merged ICPSR (Haines)-Margo data set with constant land area (square miles) in 1850 and 1860. Equally weighted: unit of observation is the county. Population weighted: counties are weighted by total population in the given year. Sample in row “Gained Access in 1850s” consists of counties with no rail access in 1850. Rail Access = 1 if county had a rail line within its boundaries; water access = 1 if county had a navigable waterway within (or on) its boundaries. [ ]: proportion of population living in counties with rail or water access.
continued
84
M.R. Haines and R.A. Margo
Panel B: Rail access by state: population weighted State
Number of counties
Rail access in 1850 (%)
Rail access in 1860 (%)
Percent change, rail access in 1860, no access in 1850 (%)
Massachusetts Pennsylvania Indiana Michigan Iowa Virginia Georgia Louisiana Mississippi North Carolina South Carolina Texas Kentucky Tennessee
11 58 80 32 46 106 22 25 51 63 28 33 68 49
98.3 75.0 34.1 74.1 0 28.5 76.5 44.3 5.0 16.6 54.6 0 16.7 23.1
99.4 89.6 70.8 78.6 33.0 49.9 84.9 53.2 48.4 40.0 72.0 15.8 36.0 59.5
64.8 58.4 55.7 17.4 33.0 29.9 35.7 16.0 45.7 28.1 38.8 15.8 23.2 47.3
Note See Panel A for sample definition. Column 4: [Column 3 – Column 2]/[100 – Column 2] 100%.
Panel C: Linear probability coefficients: rail access in 1860, no access in 1850
Water Access in 1850 = 1 Percent Urban in 1850 Log (Population/Square Miles) in 1850 State Dummies Included? Adjusted R-square
Equally weighted
1850 population weighted
–0.105 (0.036) 0.424* (0.148) 0.122* (0.021) Yes 0.229
–0.082 (0.038) 0.273* (0.124) 0.049 (0.027) Yes 0.242
Note *significant at 5 percent level.
had access to water transportation were less likely to gain rail access. Below we show that the negative effect of water on gaining rail access survives inclusion of a variety of covariates, as well as state dummies; that is, it is not an incidental consequence of failing to control for other determinants of rail access. In column 3 of Panel A we weight observations by total population in the county in the relevant census year (in the case of rows 3–5, this is population in 1850). The proportion with rail access in both years increases when weighted by population, an indication that populous counties were more likely to have access by 1850 or gain access by 1860 if they did not have it in 1850. An analogous effect is not apparent, however, if we weight instead by land area, the implication being that population density was a factor in having access by 1850, or gaining it by 1860, rather than total population alone. Lastly, the figures in brackets show the proportion of the population with access to either rail or water transportation. In 1850, nearly 71 percent of the American population lived in
Railroads and local economic development
85
counties with either rail or water transportation; by 1860, the figure was 82 percent. As the quote from Stover (1978) suggests, the nineteenth-century transportation revolution was largely over by the eve of the Civil War, if not a decade earlier. This fact may help explain why our estimated impacts of rail access are relatively small (see below). Panel B of Table 5.1 shows population-weighted rail access by state. In the Northeast, states like Massachusetts and Pennsylvania had near total access by 1860. For the three Midwestern states, access varied substantially before the Civil War. Most of the population of Michigan lived in counties with rail access by 1850, and access grew little during the 1850s. In Indiana, by contrast, only about a third of the population had rail access in 1850 but by 1860, nearly 70 percent had rail access. Iowans had no access to rail in 1850, but a third of the state’s population did in 1860. The data also show that there was significant variation in rail access in the South in 1850. A little more than three-quarters of the population of Georgia resided in counties with rail access in 1850 whereas only 28 percent of the population of Virginia did. But access expanded quite rapidly in some southern states in the 1850s, almost doubling in Virginia, and more than doubling in Kentucky and Tennessee. Panel C of Table 5.1 reports the coefficients of three covariates from linear probability regressions of gaining access to rail by 1860 conditional on not having access in 1850. The covariates are a dummy variable for having water access, percent urban, and the log of population per square mile (population density). The regressions also included state dummies. In column 1 the data (counties) are equally weighted whereas in column 2 they are population weighted. The results in the earlier panels suggest that we should observe a negative effect of water access and a positive effect of population density, and this is confirmed by the regression. We also hypothesized that the presence of a sufficiently large urban population, as indicated by the proportion living in incorporated places of at least 2,500 (the usual census definition of “urban”) would enhance the likelihood of gaining access, as the presence of a relatively large “central place” would be complementary to trade – and therefore, provide a convenient place for the railroad to stop to load and unload cargo. This hypothesis is confirmed; an increase in the proportion urban is associated with a positive and highly significant effect on the probability of gaining access. Weighting by population in 1850 slightly attenuates the magnitudes of the coefficients but does not alter their signs. We also estimated regressions excluding the state dummies (not shown) and, again, the substantive findings were not affected. However, F-tests clearly showed that, controlling for the three covariates, the addition of state dummies significantly improved the fit of the regression, indicating that, conditional on the covariates, the likelihood of gaining rail access varied across states. We investigate the impact of the railroad using a DID methodology. Letting yit be an outcome variable and Rit be a dummy variable for rail access, with the
86 M.R. Haines and R.A. Margo “i” subscript referring to counties and the “t” subscript referring to years, our basic model is yit = i + Rit + t + it where is a dummy variable for 1860 and is a random error term. Rather than estimate this directly we estimate the equation in first difference form y = + R + where = i,1860 – i,1850 We exclude two counties that appear to have lost rail access over the 1850s; hence, in our data set, once acquired, rail access is permanent (at least, as of 1860). Therefore, R = 1 if and only if the county did not have rail access in 1850, but acquired it by 1860.2 According to the DID method, the “control” counties consist of those that (1) did not have rail access in 1850 and failed to acquire it between 1850 and 1860 and (2) counties that had access in 1850. In some specifications (see below) we exclude counties that already had rail access by 1850 and focus instead on the “population at risk” of obtaining access. The “treatment” counties are always those that gained access in the 1850s. Thus at issue is whether outcome variables changed in response to the treatment: access to a railroad. This is a “before-andafter” method, so we are comparing changes in outcomes between 1850 and 1860 in the treatment group with the analogous changes in the control group. If OLS (least squares) is used to estimate the regression the assumption must be that the treatment (rail access) and the error term are uncorrelated. In standard DID this assumption can be met in different ways – for example, by preselecting the treatment and control observations so that they are “similar” or by including covariates in the regression that purge any correlation between rail access and the error term. We have already shown that ∆R depended on water access, percent urban and population density in 1850; and, even after controlling for these variables, across states. In certain specifications, therefore, we include these covariates which, in the context of the estimation, amount to interactions between the 1860 time dummy and the covariates. It would be desirable to include two or more pre-treatment values of y, in order to control for the possibility of pre-existing trends. We cannot include any pre-existing trends at present, because our data are not linked to the 1840 census.3 However, we can (and do) include the 1850 values of the outcome variable in some specifications which, in effect, permits “regression to the mean” in the outcome variables. Table 5.2 shows the regression results for our quantity outcome variables and Table 5.3 the results for our factor price outcome variables. In Panel A of Table 5.2 we show the treatment effects of rail access on the percent urban, again measured as the proportion of the population living in incorporated places of at least 2,500 or more. The idea here is that the percent
Railroads and local economic development
87
urban is a proxy for the percent non-farm, therefore, if the marginal county acquiring rail services in the 1850s had a comparative advantage in agriculture the Heckscher–Ohlin framework is correct, and the proportion urban is, indeed, a reliable proxy, we should observe a negative treatment effect. We do not – instead, the effect is positive and statistically significant.4 The treatment effect is smaller if we include counties that already had rail access in 1850 in the control group but still substantial and reliably estimated.
Table 5.2 Treatment effects of gaining rail access: percent urban, output per acre, and improved acres/square mile Panel A: Urban population/total population Sample
Number
State dummies?
1850 covariates?
Lagged dependent variable?
Coefficient
1850 rail access = 0 1850 rail access = 0 1850 rail access = 0 1850 rail access = 0 All All All All
506 506 506 506 655 655 655 655
No Yes No Yes No Yes No Yes
No No Yes Yes No No Yes Yes
No No Yes Yes No No Yes Yes
0.018* (0.005) 0.020* (0.005) 0.018* (0.005) 0.019* (0.006) 0.009 (0.005) 0.011 (0.006) 0.010* (0.005) 0.011* (0.005)
Notes All: all counties. Percent Urban = population in towns, villages, etc. of 2,500 or more/total population. Counties are equally weighted in estimation. 1850 Covariates: From Table 1, Panel C, Percent urban, log (Population/Square Miles), presence of a navigable waterway. *: significant at 5 percent level.
Panel B: Log (value of agricultural output, definition #1/improved acres) Sample
Number
State dummies?
1850 covariates?
Lagged dependent variable?
Coefficient
1850 rail access = 0 1850 rail access = 0 1850 rail access = 0 1850 rail access = 0 1850 rail access = 0 All All All All All
506 506 506 506 506 655 655 655 655 655
No Yes No Yes Yes No Yes No Yes Yes
No No Yes Yes Yes No No Yes Yes Yes
No No No No Yes No No No No Yes
–0.036 (0.040) –0.024 (0.040) –0.027 (0.040) –0.020 (0.041) –0.019 (0.040) –0.033 (0.035) –0.035 (0.035) –0.025 (0.035) –0.031 (0.034) –0.033 (0.033)
Note See Panel A. Definition #1: Value of Crops, orchards, and marketed garden produce.
88
M.R. Haines and R.A. Margo
Panel C: Log (value of agricultural output, definition #2/improved acres) Sample
Number
State dummies?
1850 covariates?
Lagged dependent variable?
Coefficient
1850 rail access = 0 1850 rail access = 0 1850 rail access = 0 1850 rail access = 0 1850 rail access = 0 All All All All All
506 506 506 506 506 655 655 655 655 655
No Yes No Yes Yes No Yes No Yes Yes
No No Yes Yes Yes No No Yes Yes Yes
No No No No Yes No No No No Yes
–0.034 (0.032) –0.019 (0.032) –0.026 (0.033) –0.018 (0.032) –0.020 (0.030) –0.033 (0.029) –0.032 (0.029) –0.027 (0.029) –0.030 (0.029) –0.036 (0.026)
Notes See Panel A. Definition #2: definition #1 + value of slaughtered livestock + value of household manufactures.
Panel D: Log (improved acres/square mile) Sample
Number
State dummies?
1850 covariates?
Lagged dependent variable?
Coefficient
1850 rail access = 0 1850 rail access = 0 1850 rail access = 0 1850 rail access = 0 1850 rail access = 0 All All All All All
506 506 506 506 506 655 655 655 655 655
No Yes No Yes Yes No Yes No Yes Yes
No No Yes Yes Yes No No Yes Yes Yes
No No No No Yes No No No No Yes
–0.212* (0.069) –0.149* (0.045) –0.042 (0.049) –0.075* (0.039) –0.058 (0.035) –0.097 (0.060) –0.099* (0.039) –0.057 (0.043) –0.084* (0.033) –0.072* (0.030)
Note *significant at 5 percent level.
In Panels B and C of Table 5.2 we report treatment effects of rail access on the log of agricultural yields, or output per acre. We use two definitions of output. The first includes the value of crops, orchards, and garden produce intended for market. The second adds to the first the value of slaughtered livestock and household manufactures. “Acre” here refers to “improved acres.” Again, if the Heckscher–Ohlin framework were correct we should observe positive treatment effects but, instead, the effects are negative, although statistically insignificant. Counties that gained rail access in 1850 evidently did not experience rising agricultural output per unit of land.5 In Panel D we test an implication of the Von Theunen model: the share of
Railroads and local economic development
89
land in agriculture should increase. To be precise, the Von Theunen model predicts that land under cultivation should increase; if it did increase, this might be a reason why we do not observe an increase in yields. We do not observe land under cultivation, but we do observe improved land. We find no evidence of a positive treatment effect – indeed, all of the estimate effects are negative, and several are statistically significant. In interpreting Panel D it is important to keep in mind that, by design, our specification holds total land area constant at the county level. In Table 5.3 we report our factor price regressions. We report results for the sub-sample of counties that initially had no rail access in 1850; that is, the control group are counties with no rail access and the treatment group, as before, consists of counties that gained access. Results using the broader control group including counties with rail access in 1850 were similar. The farm wage is adjusted for the imputed value of board, as described earlier. We do not have data on the rental price of land; in its place we use the per acre value of land, computed and adjusted as follows. The census reported the total value of farms and the number of improved and unimproved acres. We compute the average value per acre as the ratio of the total value to the total number of acres. Next, we estimated cross-sectional
Table 5.3 Factor price regressions: counties with no rail access in 1850: difference-indifference estimates Dependent variable
1850 covariates?
Lagged dependent variable?
Coefficient
Log (farm wage with value of board imputed)
Yes
No
0.023 (0.022)
Log (farm wage with value of board imputed)
Yes
Yes
0.035 (0.019)
Log (adjusted per acre land price)
yes
no
0.036 (0.047)
Log (adjusted per acre land price)
yes
yes
0.061 (0.036)
Log (farm wage with value of board imputed/ adjusted per acre land price)
Yes
No
–0.013 (0.050)
Log (farm wage with value of board imputed/ adjusted per acre land price
Yes
Yes
–0.027 (0.038)
Notes Adjusted per acre land price: residual from regression of log (value of farm/total acres) on (Improved Acres/Total Acres) plus state dummies. Farm Wage with value of board imputed = log (monthly money wage + 4.3*weekly value of board).
90
M.R. Haines and R.A. Margo
regressions for both 1850 and 1860 of the log of the per acre value on the percentage of improved acres and state dummy variables. The residuals from this regression are what we mean by the “adjusted” per acre value of land. If the residual is positive, land sold for a higher price than what would be predicted based on the proportion of improved acreage and the state in which the county was located. We find small positive effects of rail access on the farm wage and the adjusted per acre land price but these approach statistical significance only when lagged values of the dependent variables are included. With the lagged value included, the impact of rail access on land price is of the order of 6 percent which, while positive, is less than the impact estimated by Craig et al. (1998). The treatment effect on wages was smaller than on land price, which is consistent with a decline in the wage–rental ratio, as required by the Heckscher–Ohlin model, but the effect is small and not reliably estimated. In Table 5.4 we report results of selected regressions for Northern and SouthTable 5.4 North-south differences: treatment effects of rail access in 1860, no access in 1850: difference-in-difference estimates Dependent variable
Sample size, North
Percent urban
150
Slave population, ages 10+/total population, ages 10+
Na
Log (improved acres/square mile)
139
Log (value of agricultural output #1/improved acres)
Coefficient, North
Coefficient, South
371
0.005 (0.006)
371
0.008* (0.004)
0.046 (0.062)
368
–0.077* (0.034)
140
–0.023 (0.038)
368
–0.008 (0.029)
Log (value of agricultural output #3/improved acres)
140
–0.033 (0.038)
368
–0.007 (0.052)
Log (farm wage w/value of board imputed/adjusted per acre land price)
125
0.006 (0.052)
286
–0.044 (0.050)
Log (farm wage w/value of board imputed)
286
–0.013 (0.020)
286
0.061* (0.026)
Log (adjusted per acre land price)
140
–0.008 (0.052)
368
0.109* (0.045)
Note *significant at 5 percent level.
0.049* (0.011)
Sample size, South
Na
Railroads and local economic development
91
ern states separately. The logic of estimating the regression separately by region is that the crop mixes differed significantly, and because the South made use of slave labor. The vast majority of slaves lived in rural areas and most of these were engaged in agriculture, and so it is useful to determine if gaining rail access altered the share of slaves in the local population. We do find that gaining rail access increased the proportion of slave; in addition, gaining access had a larger impact on land values in the South than in the North, consistent with Fogel and Engerman’s (1980) contention that location rents were a higher share of total land values in the South. We also find a positive effect of rail access in the North on the share of improved acreage in total land area, but the effect is not statistically significant. The impact on urbanization is also larger in the North but in both regions we find no evidence of a positive effect on agricultural yields.
Case studies of Illinois and Indiana: census micro-data We have examined the impact of gaining rail access using county-level data from the published census volumes. In this section, we explore the impact for two states, Illinois and Indiana, using individual-level data from the 1850 and 1860 IPUMS (integrated public use micro-data samples). We focus on these two states because both experienced substantial gains in rail access in the 1850s (see Tables 5.1, 5.5, and 5.6). Indiana was included in our previous analysis but not Illinois because we lack wage data from Illinois for 1850. In this section data for both states are analyzed separately. There are two advantages to using the IPUMS to study the treatment effects of gaining rail access. First, the IPUMS includes additional information on occupation and sector of employment that is not available (at the county level) in the published volumes. Second, because the IPUMS data are individual-level, it is possible to control for characteristics (such as age) that may influence outcome variables but which, again, cannot be controlled for using published data. The data consist of repeated cross-sections (1850 and 1860) of males, ages 20–64, who did not attend school in the previous year. Although we have information on labor force participation for individuals aged 15 and over, the census did not record literacy status until age 20, and this is one of the covariates used in the regression (see below and Tables 5.5 and 5.6). The results for Indiana are displayed in Table 5.5 and those for Illinois in Table 5.6. Panel A shows rail and water access statistics by states. As noted previously in our discussion of Table 5.1, Indiana experienced a substantial rise in rail access during the 1850s. The same was true of Illinois: in 1850 only 11 of the 77 Illinois counties included in the analysis had rail access, but by 1860, 54 did. We explore the impact of rail access using four outcome variables, three of which pertain to agriculture, and one to services. We use information on the type of residence (farm = 1), occupation and industry to indicate agricultural participation, and information on industry to indicate participation in the service sector. Panel B shows the sample statistics by state; in both cases, the proportion
92
M.R. Haines and R.A. Margo
Table 5.5 Preliminary IPUMS results for Indiana Panel A: Sample statistics: rail and water access in the 1850s, counties Rail = 1
Water = 1
No access in 1850 or 1860 No access in 1850, access in 1860 Access in 1850 and 1860
27 28 25
33 5 42
Total
80
80
Source: 1850, 1860 IPUMS matched to Craig, Weiss, Palmquist (1998). Sample consists of adult males, ages 15–64, not attending school during census year. Notes Counties with boundary changes (according to ICPSR coding) or fewer than 10 per year observations are excluded.
Panel B: Sample statistics, outcome variables: regression sample (ages 20–64) Number Farm Agriculture, of residence occupation observations (%) (%) 1850 1,983 1860 2,808
66.7 61.4
61.8 59.2
Agriculture, industry (%)
Services, industry (%)
Rail =1 (%)
Water =1 (%)
61.7 58.7
13.5 15.3
38.2 74.6
44.7 49.6
Notes Farm residence = 1 if IPUMS; Agricultural occupation = 1 if three-digit occupation code = 100, 123, 810, 820, or 970 (laborers, n.e.c.) and farm residence = 1; agricultural industry = 1 if three-digit industry code = 105; services = 1 if industry code = 500, . . ., 946.
Panel C: Treatment effects of rail: indiana in the 1850s: excludes counties gaining water access Number of observations
Farm = 1
Agricultural occupation =1
Agriculture (industry) =1
Services (industry) =1
Baseline
4,550
–0.052 (0.042)
–0.053 (0.034)
–0.061 (0.032)
0.040 (0.026)
With Covariates
4,550
–0.050 (0.040)
–0.050 (0.032)
–0.059 (0.033)
0.038 (0.025)
Notes Sample consists of adult males, ages 20–64, not attending school. Five counties gaining water access during the 1850s are excluded. Baseline: treatment plus year (=1860) dummy. Covariates: baseline plus fourth-order polynomial in age, dummies for literacy, foreign birth, metro area status. Standard errors (in parentheses) are clustered by county.
Railroads and local economic development
93
Panel D: Treatment effects of rail and water: Indiana in the 1850s: all counties in sample Number of observations
Farm = 1
Agricultural occupation =1
Agriculture (industry) =1
Services (industry) =1
Baseline, Rail = 1
4,791
–0.050 (0.040)
–0.060 (0.033)
–0.068* (0.034)
0.042 (0.025)
With covariates, rail = 1
4,791 (0.038)
–0.049 (0.033)
–0.058 (0.033)
–0.067* (0.024)
0.041
Notes See Panel A; includes five counties gaining water access. *: significant at 5 percent level.
engaged in agriculture declined while participation in the service sector increased over the 1850s. Panel C shows the treatment effects of gaining rail access by state when we exclude counties gaining water access. The “baseline” estimates refer to a specification in which there are no covariates; the “with covariate” estimates include a fourth-order polynomial in age, and dummy variables for literacy status, foreign birth, and urban location. Standard errors are clustered by county. Most of the coefficients are not statistically significant at conventional levels but the general patterns are consistent across states: gaining rail access reduced the likelihood of agricultural participation and increased the likelihood of working in the service sector. Including counties that gained water access (Panel D) does not change the substantive findings. To summarize, we find no evidence that individuals in counties that gained rail access were more likely to be engaged in agriculture. Instead, our results suggest that gaining rail access raised the odds the individuals would be engaged in the service sector. To the extent that service sector activities were largely concentrated in urban locations during the antebellum period, the analysis of the IPUMS data is consistent with our finding, using the published census, that railroads promoted urbanization.
Further discussion We have used county and individual level data to examine the economic impact of gaining rail access in the 1850s. Previous studies have found that rail access raised land values, and have interpreted this correlation as evidence that rail access chiefly benefited agricultural land owners in the manner predicted by the versions of the Heckscher–Ohlin or Von Theunen frameworks elaborated earlier in the chapter. In contrast to this previous work, we have used a DID strategy, comparing outcomes in a treated group of counties – those that gained rail access in the 1850s – to a control group – those that either gained access earlier or did not have access before the Civil War.
94
M.R. Haines and R.A. Margo
Table 5.6 Preliminary IPUMS results for Illinois Panel A: Sample statistics: rail and water access in the 1850s, counties Rail = 1
Water = 1
No access in 1850 or 1860 No access in 1850, Access in 1860 Access in 1850 and 1860
24 42 11
39 4 34
Total
77
77
Notes Counties with boundary changes (according to ICPSR coding) or fewer than 10 per year observations are excluded. Source: 1850, 1860 IPUMS matched to Craig, Weiss, Palmquist (1998). Sample consists of adult males, ages 15–64, not attending school during census year.
Panel B: Sample statistics, outcome variables: regression sample (ages 20–64) Number Farm Agriculture, of residence occupation observations (%) (%) 1850 1,817 1860 3,616
67.1 58.5
61.0 55.0
Agriculture, industry (%)
Services, industry (%)
Rail =1 (%)
Water =1 (%)
61.2 55.3
13.4 18.5
23.8 78.4
52.8 58.0
Notes Farm Residence = 1 if IPUMS; Agricultural occupation = 1 if 3-digit occupation code = 100, 123, 810, 820, or 970 (laborers, n.e.c.) and farm residence = 1; agricultural industry = 1 if 3-digit industry code = 105; Services = 1 if industry code = 500, . . ., 946.
Panel C: Treatment effects of rail: Illinois in the 1850s: excludes counties gaining water access Number of observations
Farm =1
Agricultural occupation =1
Agriculture (industry) =1
Services (industry) =1
Baseline
5,141
–0.033 (0.034)
–0.024 (0.038)
–0.025 (0.037)
0.008 (0.025)
With covariates
5,141
–0.016 (0.037)
–0.004 (0.038)
–0.005 (0.037)
0.017 (0.026)
Notes Sample consists of adult males, ages 20–64, not attending school. Four counties gaining water access during the 1850s are excluded. Baseline: treatment plus year (=1860) dummy. Covariates: baseline plus fourth-order polynomial in age, dummies for literacy, foreign birth, metro area status. Standard errors (in parentheses) are clustered by county.
Railroads and local economic development
95
Panel D: treatment effects of rail and water: Illinois in the 1850s: all counties in sample Number of observations
Farm =1
Agricultural occupation =1
Agriculture (industry) =1
Services (industry) =1
Baseline, rail = 1
5,433
–0.042 (0.033)
–0.031 (0.035)
–0.031 (0.055)
0.013 (0.056)
With covariates, rail = 1
5,433
–0.025 (0.035)
–0.012 (0.036)
–0.012 (0.035)
0.020 (0.025)
Notes See Panel A; includes four counties gaining water access.
Most of the estimated treatment effects are small and their signs, taken as a whole, are not readily consistent with either framework. For example, on average we find that rail access appears to have increased the percent urban and the likelihood of participation in the service sector, decreased agricultural yields, and reduced the share of improved acreage in total land area, opposite to the patterns we would expect if the impact on agriculture were as either framework predicts. Some of our results, of course, are consistent with either framework – we find, in particular, positive effects on the price of land, and negative effects on the ratio of farm wages to land price. But even the effects that are consistent are generally quite small and frequently statistically insignificant. When we divide the sample into Northern and Southern states, the predicted impacts are more in line with the theoretical predictions in the South. We find that gaining rail access in the South was associated with an increase in the proportion of slaves in the population, and strongly positive effects on farm wages and on land values. Given the known effects of slavery on agricultural productivity (positive; see Fogel and Engerman 1980) these results suggest that, in the South at least, rail access did benefit agricultural interests. However, even in the South, the results do not seem fully consistent with either the Hekscher-Ohlin or Von Theunen frameworks, because the treatment effect of rail access on the ratio of improved acres in agriculture to total land area is negative whereas, the predicted impact is positive. If, as appears to be the case, the positive cross-sectional correlation of rail access on land prices is robust, why are our DID findings so mixed? One possibility is the frameworks sketched earlier in the chapter are too simple. In particular our presentation of the Hekscher-Ohlin framework omitted any consideration of a second non-farm sector devoted to trade services. Trade involves transactions costs and certainly the type of trade we are considering here – the exporting of agricultural commodities from Midwestern counties via rail – requires a physical location – a “central place” – for trade to take place, such as a town. We could augment the Hekscher-Ohlin framework to include a sector devoted to trade services; in such a framework, we might expect to see a shift in the use of
96
M.R. Haines and R.A. Margo
non-farm labor toward the service sector, which could show up as an increase in percent urban, particularly in counties that were entirely rural (or nearly so) to begin with. However, in such an augmented framework we would still expect to see, for example, increases in yields (or, as in the Von-Theunen model, in acres under cultivation), and we do not. A second explanation may be that the treatment in question – rail access in the 1850s – was not very effective.6 As we show in Table 5.1, a very large fraction of the population lived in counties in the 1850s that were already served by rail or water. The impact of earlier access, which we cannot measure, may have been far greater. Another possibility is that the cross-sectional relationship of rail access on land values is primarily a reflection of omitted variable bias. That this possibility must be taken seriously is suggested by the fact that, while the sign (positive) of the correlation between rail access and land prices is robust, the size of the correlation is readily influenced by the inclusion or exclusion of covariates (see Craig et al. 1998). In this regard, we emphasize again that we are unable to control for preexisting trends. For example, if the percent urban were increasing in the 1840s in counties that acquired rail access in the 1850s – or, alternatively, if agriculture was on the wane – this could account for the positive treatment effect on urbanization that we observe, as well as the mixed effects on agricultural outcomes.7 If omitted variable bias were responsible, the solution would be to use one or more instrumental variables (IVs) to predict rail access. Topographical variables are possible candidate instruments.8 The idea is that local topography influenced the costs of building a railroad – for example, the presence of mountainous terrain may have raised the costs, making it less likely that rail access would be provided, whereas flat terrain made construction easier and maintenance less costly. In the current data set we have one topographical variable available in our dataset, the presence of water transportation. To explore the IV approach, we re-estimated certain regressions treating water access as an instrument (rather than a covariate) for rail access. As already shown in Panel C of Table 5.1, the presence of water access does have a statistically significant, negative effect on the likelihood of gaining rail access in the 1850s – that is, the “first stage” regression looks reasonable. When water access is used as an instrument some results appear to be more favorable – for example, we find a negative treatment effect on percent urban. But none of the more favorable results are statistically significant and, in some cases, the signs of the IV coefficients are still opposite to what the frameworks would predict and are even larger in magnitude. We conclude that, while the IV approach has promise, it will be necessary to find additional instruments other than the presence of water transportation in 1850.
Concluding remarks This chapter has a used a pooled sample of counties from 1850 and 1860 to investigate the economic impact of gaining access to a railroad. Previous work
Railroads and local economic development
97
has shown that, at a point in time, the presence of railroad increased the value of agricultural land. Our chapter takes this correlation as a point of departure and investigates certain ancillary predictions of economic frameworks that predict the rise in agricultural land values. We find that, taken as a group, these ancillary predictions are generally not confirmed – for example, we find no evidence that agricultural yields increase when rail access occurs. In this regard, our results are more consistent with theory for Southern than for Northern states, but even for the South, some results (a negative effect on the share of improved acreage in total land area) are opposite to what theory predicts. Further research is necessary to determine if these findings are robust or whether they are a consequence of various problems and biases in the data.
Notes *Comments from Stanley Engerman, Eric Hilt, Joshua Rosenbloom, Thomas Weiss, conference participants, and an anonymous referee are gratefully acknowledged. 1 See Taylor (1951: ch. 5) for a concise history of early railroads. 2 Two counties that purport to have gained rail access by 1850 and then lost it by 1860 are excluded. There are seven counties in the sample that claimed to have gained access to water transportation in the 1850s. In the regressions reported in this chapter we do not add a dummy for these seven counties but our substantive results would not change if we did or if we excluded them from the sample. 3 Even if our data were linked to the 1840 census our ability to control for pre-existing trends would be extremely limited, because the 1840 census did not collect information on wages, land values, or acreage in agriculture. 4 An alternative interpretation, one consistent with the regression showing that the probability of obtaining rail access in the 1850s was positively associated with the percent urban in 1850, is that the coming of the railroad expanded the demand for trade services – a town of a certain size, in other words, facilitated trade. In future work it may be possible to improve upon this analysis by making use of Craig and Weiss’s (1996) county level estimates of the rural agricultural labor force. These could be combined with estimates of the urban agricultural and total labor forces, by county, using Craig and Weiss’s procedures. However, it should be noted that Craig and Weiss derived their estimates by assuming agricultural participation ratios that vary by age, gender, rural and legal status (slave) across states and over time, but not within states. Thus, for example, a county’s number of farm workers will change if the demographic, rural/urban, or free/slave structure changes but not otherwise. It remains to be seen if this generates sufficient variability to capture any treatment effects of the rail beyond those measured here by percent urban (or percent slave, see below). 5 At present the agricultural output variables are somewhat inconsistently defined; specifically, crop output is valued at 1860 prices, but orchards, garden, and home manufactures are evaluated in current prices (the census did not report quantities for these categories of output) and we have yet to adjust these for price changes. Because our regressions include state dummies, any (state-specific) trends in prices are controlled for implicitly. We also explored the impact of gaining rail access on specific crops (for example, the share of cotton in total agricultural output) but found no consistent patterns. 6 Alternatively, it may be that both rail access and our outcome variables are measured with considerable error, in which case use of DID methods would exacerbate any bias due to measurement error. 7 Responses in anticipation of treatment or lagged responses to treatment might also explain our findings. For example, if the counties gaining rail access in the 1850s
98
M.R. Haines and R.A. Margo
anticipated this in the previous decade, it is possible that the changes we would like to observe already had occurred by the 1840s. Thus in the 1850s, the control group of counties anticipated that the railroad would eventually be extended to [some] of them, and these counties experienced changes (expanding agricultural yields, or investment in clearing land) prior to treatment. The argument can also be reversed; perhaps rail access was provided in anticipation of future development, in which case the subgroup of control counties that already had rail access by the 1850s would show responses in the 1850s, thereby undermining the research design. This is a variant of “building ahead of demand” argument; see Fishlow (1965). 8 Two other candidate instruments are federal land surveys conducted between 1824 and 1838, and the “straight line” method of Banerjee et al. (2006). The United States government conducted numerous land surveys under a law passed in 1824 that was eventually repealed in 1838 (see Taylor 1951). The existence of such a survey would presumably lower the costs of building a railroad; the maintained assumption would be that, at the time, it was not possible to accurately predict differences in the specific outcomes considered in this chapter between the treated counties (those that gained rail access in the 1850s) and the control counties (those that already had access or did not have access by 1860). The use of pre-existing plans as an instrument has been applied in the case of the federal highway system; see Michaels (2005). The “straight line” method posits that, during the initial phase of rail development, rail are most often built on straight lines between existing urban centers; the instrument in question is a dummy for whether the straight line passes through the county.
References Atack, J. and Passell, P. (1994) A New Economic View of American History. New York: W. W. Norton. Banerjee, A., Duflo, E., and Qian, N. (2006, in progress) “The Railroad to Success: The Effect of Infrastructure on Economic Growth,” Department of Economics, Brown University (Power Point Presentation). Craig, L. and Weiss, T. (1996) “The Nineteenth-Century Farm Labor Force and Rural Population: County-Level Estimates and Implications,” unpublished paper, Department of Economics, North Carolina State University. Craig, L., Palmquist, R. B., and Weiss, T. (1998) “Transportation Improvements and Land Values in the Antebellum United States: A Hedonic Approach,” Journal of Real Estate Finance and Economics 16: 173–190. Coffman, C. and Gregson, M.E. (1998) “Railroad Development and Land Values,” The Journal of Real Estate Finance and Economics 16: 191–204. Fishlow, A. (1965) American Railroads and the Transformation of the Ante-Bellum Economy. Cambridge, MA: Harvard University Press. Fogel, R. (1964) Railroads and American Economic Growth: Essays in Econometric History. Baltimore, MD: Johns Hopkins University Press. Fogel, R. and Engerman, S. (1980) “Explaining the Relative Efficiency of Slave Agriculture: A Reply,” American Economic Review 70: 672–690. Goodrich, C., ed. (1961) Canals and American Economic Development. New York: Columbia University Press. Haines, M. R. and ICPSR (2006) Historical, Demographic, Economic, and Social Data: The United States, 1790–2000. Ann Arbor, MI: Inter-University Consortium for Political and Social Research. Haites, E., Mak, J., and Walton, G. (1975) Western River Transportation: The Era of Internal Development, 1810–1860. Baltimore, MD: Johns Hopkins University Press.
Railroads and local economic development
99
Kahn, C. M. (1988) “The Use of Complicated Models as Explanations: A Re-Examination of Williamson’s Late 19th-Century America,” Research in Economic History 11: 185–216. Margo, R. A. (2000) Wages and Labor Markets in the United States, 1820–1860. Chicago, IL: University of Chicago Press. Michaels, G. (2005) “The Effect of Trade on the Demand for Skill: Evidence for the Interstate Highway System,” mimeo, Department of Economics, Massachusetts Institute of Technology. North, D. C. (1961) The Economic Growth of the United States, 1790–1860. Englewood Cliffs, NJ: Prentice-Hall. Ransom, R. (1967) “Interregional Canals and Economic Specialization in the Antebellum United States,” Explorations in Entrepreneurial History 5: 12–35. Ransom, R. (1970) “Social Returns from Public Transport Investment: A Case Study of the Ohio Canal,” Journal of Political Economy 78: 1041–1064. Stover, J. F. (1978) Iron Roads to the West: American Railroads in the 1850s. New York: Columbia University Press. Taylor, G. R. (1951) The Transportation Revolution, 1815–1860. New York: Holt, Rhinehart, and Winston. Williamson, J. G. (1974) Late Nineteenth Century American Development: A General Equilibrium History. New York: Cambridge University Press.
6
Did refrigeration kill the hog–corn cycle? Lee A. Craig and Matthew T. Holt*
Issues In modern, shall we say technical, terms the expression “hog cycle” refers to the correlated, possibly lagged, component of the swings in the prices of corn and hogs over time. Although, in the title of this chapter we pose a single question, in the text we actually answer three. These are: How were hog cycles propagated? How were they eventually ameliorated? What was the social savings from their amelioration? In answering these questions we offer a concise history of the economic relationship between hogs and corn and explain how that relationship manifested itself in the hog cycle. With respect to the propagation of the cycle, we illustrate how, even in the face of rational expectations, risk-averse profit-maximizing farmers took actions that resulted in swings in the hog–corn price ratio. Usually these swings began as the result of a supply-side shock, typically in the corn market, with the archetypal cause being an unusually abundant or an unusually poor crop. Farmers adjusted hog portfolios in response, and the performance of the subsequent crop determined the subsequent direction of the cycle, ceteris paribus, of course. Although instruments to hedge corn price risk date from the mid-nineteenth century,1 we argue that prior to the post-World War II era, the optimal percentage of the expected corn crop that farmers hedged was considerably less than 100. Furthermore, because of the transaction costs, and the constraints associated with collecting and widely disseminating relevant market information (i.e. crop growing conditions, etc.), many farmers did not hedge their crops at all. Thus, at the margin, the hog–corn price ratio was subjected to the vagaries of supply and demand in the spot market. Although the hog–corn cycle continues to this day, the nature of the cycle has evolved over time (Holt and Craig 2006). In particular, the seasonal component of the cycle has been dampened considerably. We show that this amelioration began in the late nineteenth century, largely in response to the advent and adoption of mechanical refrigeration on a large scale. Although the physics of refrigeration had been mastered several millennia ago, and patents on mechanical refrigerators date from the early nineteenth century, it was only in the last decade of the century that the costs of refrigerators had fallen to a point where
Did refrigeration kill the hog–corn cycle?
101
they could be used extensively in the meat trade. Prior to refrigeration, hogs were typically slaughtered and salted or cured once cool weather set in. With the adoption of refrigerators, which initially were used overwhelmingly in cold storage, and which followed by only a decade or so the widespread use of natural refrigeration (ice) in the shipping of meat, farmers could spread their pork production more evenly throughout the year. Finally, turning to the social savings from the adoption of refrigeration, we argue that not only could farmers smooth production across seasons, they also could increase the overall scale of livestock production. They could continue to slaughter large quantities in the fall, while maintaining inventories for slaughter during other times as well. Since the gestation period for hogs was such that sows could easily farrow twice a year, the lifting of the seasonal constraint imposed by the threat of spoilage yielded a net social saving. We estimate this saving, and find that by almost any reasonable standard it was large. The next section documents the history of the affinity of corn and hogs. The third section explains the propagation of the cycle before refrigeration and the amelioration of it by the adoption of refrigeration. The fourth section presents our estimates of the social savings, and the final section offers a summary and conclusion.
Hogs and corn: a concise history Maize (Zea mays), or corn as it is colloquially known in much of the non-British world, represents one of the great elements of the Columbian Exchange. A longdomesticated form of the large wild grass teosinte (Euchleana mexicana), corn was first cultivated more than 5,000 years ago in the river valleys of the Sierra Madre Occidental, in or near the present Mexican state of Michoacan.2 By the time of the Spanish Conquest in the early sixteenth century, corn served as an essential dietary component of the Aztecs, who controlled south-central Mexico at the time. From central Mexico corn cultivation techniques had spread throughout much of the Western Hemisphere, and it ultimately became the dominant food crop throughout Mesoamerica. The cultural and economic importance of corn in pre-Columbian Mexico was reflected in, among other things, the very language of the people. For example, in Nahuatl, the native language of the Aztecs and some neighboring cultures, the root for corn, teo, is also the root for “god,” and the word for maize dough, toneuhcayotl, means literally, “our flesh” (Salvador 1997). Prior to the arrival of Europeans, corn was largely valued as a source of flour. Other than the llama, there were no large domesticated mammalian quadrupeds in the Pre-Columbian Western Hemisphere; thus corn’s value as a feed for livestock was as yet unrealized. That changed with the introduction of hogs and cattle. The Spanish in particular and the Europeans more generally considered corn an inferior foodstuff (Collins 1993). This predominant view was at least partly the result of corn’s biology. Although high in starch (carbohydrates), corn is coarse, and possesses very little glutenin, the key protein in gluten, which
102
L.A. Craig and M.T. Holt
provides the elasticity that allows a wheaten loaf to “rise” when heated, as the gases, generated by yeast fermentation are released. Therefore, maize flour does not rise as, for example, wheat flour does, and corn could not yield anything like a light wheaten loaf of bread, which made corn less palatable to Europeans who were familiar with wheaten bread. Although in Thomas Hariot’s Narrative of the First English Plantation in Virginia (c.1588) he claimed maize “maketh a very good bread,” he also noted that it was most frequently consumed by taking the shelled kernels and “seething them whole until they be broken [into grits]; or boyling the floure with water into a pappe [i.e. porridge]” (quoted in Rasmussen 1960: 14). Even corn consumed in this way provided an inferior diet to one concentrated in animal products and wheat, for although the chemistry of corn – like most other things – was not well understood in the sixteenth century, as it turns out corn is low in niacin, and populations with diets containing large quantities of corn, such as the Mesoamerican Indians, are subject to pellagra, which tends to reduce physical energy and labor productivity (Brinkley 1994). Although the North American Indians encountered by British settlers in the seventeenth and eighteenth centuries ate a more diversified diet than the Aztecs and their Mexican neighbors, they too were unable to exploit corn’s potential as an animal feed. Despite its shortcomings as a staple in human diets, corn proved to be an ideal device for delivering carbohydrates to livestock, and the hog (Sus scrofa domesticus) proved to be particularly efficient in converting carbohydrates into meat. Notwithstanding this effectiveness, hogs initially became an important part of the U.S. farm economy because of their ability to forage – and fend for themselves while doing so. From the time Europeans pushed into the forests beyond the coastal plains of the eastern seaboard, they “turned out” hogs to forage beyond the immediate environs of the homestead. These semi-feral porcines went by any number of names, among which prairie shark, land pike, and razorback were the most common and evocative (Bonner 1964: 145–146; Danhof 1969: 175–178; Gates 1960: 217). Feeding largely on mast, roots, and even the occasional snake, according to one historian, these “[w]ood-ranging hogs became quite wild, hard to brand or mark; it was not unusual for an owner to have to hunt his hogs with the rifle in order to have pork” (Buley 1950: vol. 1: 158). For the slightly more domesticated hogs, it was common to “pen” them for fattening before slaughter, and the corn–hog connection became evident during this process. In the absence of refrigeration, the razorback was typically slaughtered in the fall of the year, with the onset of cool weather.3 Lard was rendered; the hams and tenderloins were salted or smoked; smaller cuts of the prime portions were preserved in mincemeat; and sausage and headcheese were made from the inferior “parts” (Craig 1993). Of the hog, the self-sufficient farm family ate, as Laura Ingall’s mother famously put it “everything but the squeal” (Wilder 1971: 17). While the razorback remained an important source of food for many Americans, particularly those on the ever-changing farming frontier, by the late nineteenth century, once a sufficient combination of urbanization (actually, town
Did refrigeration kill the hog–corn cycle?
103
building) and transportation development had occurred, farmers began producing pork for the market as opposed to solely their own and their family’s consumption. As towns emerged, usually at natural ports or the confluence of rivers, since water – when it was available, of course – generally proved to be the lowcost means of transport for trips beyond the shortest of distances, the agricultural areas from which the burghers were fed began to produce marketable surpluses (Atack and Bateman 1987: 201–224). Except for exotic crops or luxuries with very high value-to-weight or value-to-volume ratios (coffee, for example, or sugar and tobacco), these agricultural areas tended to be located in the lands immediately surrounding the town. Beyond this “belt,” the costs of transport, at least overland, generally exceeded the market value of the goods to be transported. In terms borrowed from economic principles, for goods produced beyond this belt, the supply function lay everywhere above the demand function, and subsistence agriculture was the norm. With further urban growth, industrialization, and improvements in graded roads followed by the emergence of canals, railroads, and improvements in international shipping, farmers further out in the hinterlands had relatively low-cost access to urban consumers and world markets (O’Rourke 1997; O’Rourke and Williamson 2002). Importantly, they also increasingly specialized in a few products and increased their scale of production in those lines.4 In turn, processors of those agricultural products, themselves located in the urban areas to which the raw materials were initially shipped, could exploit economies of scale and scope and become relatively big businesses in their own right. “The hog-corn nexus proved to be a crucial link in this chain” (Holt and Craig 2006: 217). Just as urbanization and transportation improvements influenced industrialization and were in turn influenced by industrialization, this transformation in the market altered the farm economy. In turn, the farm economy influenced the market. Our focus here is mainly on the supply side of the market, that is, with the relationship between corn and hogs in the production process. Since the former was and remains a key input into the production of the latter, it is this relationship that brings about the cyclical phenomena that our broader research of this topic addresses. There was, however, a demand-side feature to this relationship, which was itself not unrelated to the production and transport of the hog. Perhaps nothing reflects this interaction better than the enormous, by the standards of other countries, U.S. consumption of pork. European travelers often commented on the huge quantities of pork consumed in the states. “[T]he national taste certainly runs on pork . . .” observed one nineteenth-century traveler; pork was “the beau ideal of good cheer everywhere,” wrote another. When Americans did not eat pork directly, they used its fat in the preparation of just about everything else, including, somewhat curiously, fish soup (Gates 1960: 215–216). Setting tastes aside, this noted affection for pork was at least partly related to the distances produce often had to travel to reach the market. Prior to the extension of the railroad into the plains, pork dominated beef as a standard component of U.S. diets. In an age with no mechanical and often little natural refrigeration,
104 L.A. Craig and M.T. Holt the only meat that could safely reach a market either had to be driven to market “on the hoof,” slaughtered nearby, or salted or smoked for shipping, with salting being the more prevalent of the two. Although hogs could be and were driven, their physiology and temperaments did not facilitate this strategy, at least over any significant distances.5 In addition, the standard preservation techniques, smoke and salt, “were much more effective with pork than beef” (Cronon 1991: 225) – thus the importance of hogs in the consumption bundles of nineteenthcentury Americans. Furthermore, during the postbellum era, the unit price of pork was roughly half that of beef (Cronon 1991: 235; Shannon 1945: 167). Of course, the price is determined by the interaction of buyers and sellers, and it is the supply-side of the market that ultimately proved essential to the propagation of the hog–corn cycle. To see the economics of hog production and to understand why the hog played a relatively greater role than cattle on nineteenth-century eastern and midwestern farm, consider the following backof-the-envelope rules confronting a farmer at the time. First, the ratio of consumable meat and fat to total body weight was larger for hogs than cattle. Although estimates of this ratio have proven quite controversial in the economic history literature (Cuff 1992; Haines et al. 2003; Gallman 1996; Komlos 1996), some rough parameters prove the point. A steer destined for the slaughterhouse might weigh four or five times as much as a similarly destined hog (1000–1500 lbs versus 200–300 lbs), but the net weight of a steer was roughly 40–50 percent of its gross weight; whereas a hog rendered more than 80 percent of its gross weight (Cuff 1992). Therefore, holding other factors constant, it might take on average two to three hogs to yield the meat of one steer. Advantage beef. However, in terms of productive efficiency – that is, output per unit of input – grain-fed hogs converted corn to meat at a ratio that was two to three times greater than that of either cattle or sheep. Advantage pork. Using 2.5 as a midpoint, cattle and hogs would appear to have been about equally economical on the farm. In fact, given their physiologies and temperaments, cattle could be much more easily driven to market on the hoof than hogs and thus would appear to have been the preferred livestock – as they were in the vast grasslands of the west.6 With good reason it is difficult to imagine John Wayne driving a herd of pigs across Texas. But further east, closer to urban markets, the hog dominated because of three additional factors, all of which favored pork. One of these was the aforementioned ability of the hog to forage. Indeed, it was not uncommon for owners of cattle feed lots to keep hogs among the cattle, because “four cattle would waste enough corn to feed one hog” (Shannon 1945: 165). Second, hogs and cattle did not convert feed to meat at the same speed. Even the slow-maturing, nineteenth-century semi-feral breeds of hog could be marketed typically at nine to 18 months, and with the newer, purer breeds – such as the Berkshire, Duroc, and Chester White – introduced on a large scale later in the century, this figure could be pushed to the six- to eight-month range (Shannon 1945: 166, 168); steers typically took three-and-half to four years from birth to slaughter (Danhof 1969: 35; Gates 1960: 210–211). Although the early “pure-bred” hogs were often characterized as frail and subject to disease,
Did refrigeration kill the hog–corn cycle?
105
farmers found that breeding them with the heartier, older breeds – such as the Tamworth – produced an optimal mix of traits. The resulting animals were so hearty and efficient at converting corn to meat that they were often refered to as “mortgage lifters.” Finally, the reproductive capacity of sows exceeded that of cows. Sows could be bred earlier than cows – at one year of age versus two to three – and sows had much shorter gestation periods. Sows could produce a litter of several piglets in four months; whereas cows took nine months to produce a single calf or occasionally two. By the end of the nineteenth century, it was not uncommon for farmers to farrow sows twice a year. Thus on the mixed farms of the east and the midwest, the economics leaned heavily toward pork. Of course this preference for raising hogs was not unrelated to the fact that much of this same region turned out to be excellent for the production of the very corn that hogs so efficiently converted to meat and fat. Although, today, corn is grown around the world between 60° North and 40° South, the best yields of the nineteenth century could be obtained in thick top soils, with 18 to 25 inches of rainfall, during a growing season of five to six months, in which day-time highs reached 85° (Fahrenheit) and night-time lows reached 55°. These conditions were ideally met throughout a large belt of the United States marked longitudinally by the eastern coastal plain and around 98° longitude through east–central Kansas and Nebraska, and latitudinally by the northern shores of the Great Lakes and the Gulf Coast. Thus, corn could be grown, with varying degrees of success to be sure, from the slopes of the Appalachians to the edge of the Great Plains, from the Ozarks to the banks of the Wabash. As already noted, because of its coarse nature and lack of palatability, corn was never a big cash crop, as were, for example, wheat and cotton. Thus, in the nineteenth century, corn typically went to market in the form of either hogs or whiskey.7 As a contemporary put it: In the hog “[c]orn thus becomes incarnate; for what is a hog, but fifteen or twenty bushels of corn on four legs?” (quoted in Cronon 1991: 226). Still, as the frontier moved west and the country urbanized back east, all that corn in all those hogs had to get to market. Without expeditious and low-cost transportation, early hog cycles were typically quite local in nature, usually centered on a nearby market town, which, depending on its location, might occasionally be tied to a broader market, which itself reflected a trans-village cycle. However, the emergence of Cincinnati – aka Porkopolis – as a meatpacking entrepot early in the nineteenth century signaled the onset of a regional hog cycle of substantial size and duration (Gates 1960: 215). By 1823, the city of 10,000 inhabitants was shipping nearly three million pounds of pork, by 1840 the figure was more than five times as large (Buley 1950: vol. 1: 21–22, 527; Kurlansky 2002: 251). Cincinnati’s reputation as the capital of hogdom was as well recognized as it was earned. Of her visit to the city in the 1820s, the English writer Frances Trollope commented that “I am sure that I would have liked Cincinnati much better if the people had not dealt so very largely in hogs;” when the creator of New York’s Central Park, Frederick Law Olmsted, visited there nearly 30 years later, he observed that there were “as many hogs as trees” (quoted in Cronon 1991: 228). Rudolf Clemen, an early
106
L.A. Craig and M.T. Holt
historian of the livestock industry, quoted a contemporary: “It was Cincinnati which originated and perfected the system which packs fifteen bushels of corn into a pig and packs that pig into a barrel, and sends him over the mountains and over the ocean to feed mankind” (Clemen 1923: 93). As with most aspects of real estate, the key to Cincinnati’s rise as the epicenter of the pork trade was location. Surrounded by fertile river valleys (including the Scioto, both forks of the Miami, the Licking, and the Whitewater) and located on the Ohio, by the 1820s Cincinnati was the Wall Street of pork.8 Of nearly equal importance as its access to transportation was Cincinnati’s access to the salt springs in the nearby Kanawha Valley (Kurlansky 2002: 250–253). As Clemen (1923: 95) notes: “The necessities of the trade required an ample supply of salt, and this could be obtained in Cincinnati” relatively easily. Just as the surrounding river valleys contributed to Cincinnati’s position on top of the meatpacking world, the rise of the railroad ended it. Although Cincinnati would continue to grow, its relative position in meatpacking was forever eroded. As late as 1850, Cincinnati was packing 300,000 hogs a year, between 40 and 50 million pounds of pork; while Chicago shipped 20,000. But the railroad and westward migration gave Chicago a much faster rate of growth. Indeed, it passed Cincinnati during the Civil War, and by the early 1870s, over a million hogs were being slaughtered annually in Chicago (Cronon 1991: 230–231). Chicago’s rise marked the rise of, first, a national and then, ultimately, an international market in meat. It was only with this trans-regional integration of commodity markets that the multitudinous local cycles became singular in the late nineteenth century. Integration itself resulted from an array of long-run economic changes that included, among other things, urbanization, transportation improvements (of which the railroad was the most prominent), and refrigeration (Goodwin et al. 2002; Craig et al. 2004). By the time a national hog–corn cycle emerged with the rise of the Chicago market after the Civil War, marketable corn production was largely concentrated in the midwest. Local corn markets were also arguably integrated with Chicago and, through its connections to rail and water transportation networks, with the wider world. Thus the hog–corn cycle existed on a clearly identifiable if not necessarily consistent pattern from the end of the Civil War. We now turn to the propagation of that cycle.
Propagation and amelioration Figure 6.1 shows the hog–corn price ratio from 1870 through 1940.9 The economics of the hog–corn relationship as discussed above suggests that 20 bushels or so of corn spread over a few months to a year, depending on the breed consuming it, might reasonably be expected to yield 200 pounds of pork of various cuts, which was roughly the average (net) weight for hogs slaughtered during the postbellum era according to Cuff (1992). Thus the “iron rule” of hog–corn economics, as originally posited by Fred Shannon (1945: 165), was that as long as the price of hogs (per hundredweight) was greater than ten times the price of
Did refrigeration kill the hog–corn cycle?
107
corn in bushels it was profitable to feed corn to hogs. Whether the corn was home grown or purchased did not matter, since the market price was the opportunity cost of feeding corn to one’s own hogs. This relationship ultimately led to the hog–corn cycle. When the supply and demand for hogs were such that the price of pork was greater than ten times that of corn, farmers would slaughter mature hogs, feed all of their corn to maturing hogs, and, if possible, purchase more to be fed to their maturing hogs. (Although Shannon put the iron-rule ratio at ten-to-one, the data in Figure 6.1 suggest the mean may have been closer to 11.5). To obtain some insight into the cyclical nature of the data in Figure 6.1, we estimated an autoregressive model that contained 12 lags of the hog-corn price ratio (in levels) and 11 monthly dummy variables. The roots of the associated characteristic polynomial were then obtained. The dominant root is complex, suggesting cyclic behavior. The associated modulus of this root is 0.956, which implies that the model is in fact dynamically stable, that is, the cycle is not explosive, at least not in any meaningful econometric sense. Moreover, the imputed cycle length for the dominant root is 34.76 months or 2.90 years. This estimate suggests a somewhat shorter cycle than the roughly four-to-six-year cycle observed by Fred Shannon for the late nineteenth century (Shannon 1945: 167). It is also shorter than the three-to-four-year cycle observed by Mordecai Ezekiel (1938) based on data for the first three decades of the twentieth century (Ezekiel 1938: 271). Studies of the cycle date from Coase and Fowler (1937) and include Ezekiel’s (1938) classic work on the cobweb theorem. More recently, Chavas and Holt (1991) and Holt and Craig (2006) show that the cycle was non-linear,
1940
1936
1933
1929
1926
1922
1918
1912
Figure 6.1 The hog–corn price ratio, 1870–1940.
1915
1908
1905
1901
1898
1894
1887
1891
1884
1880
1877
1873
1870
20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5
108
L.A. Craig and M.T. Holt
and the parameters that characterized the cycle changed over time. The evolution of the study of the cycle indicates that there are at least two fundamental questions that have not previously been successfully addressed: the changing seasonal structure of the cycle resulting from, among other things, the adoption of mechanical refrigeration, and the social savings attributable to these changes. Turning to the changing seasonal structure and refrigeration first, we note that the markets for corn and hogs experienced substantial structural change during the period in question. In addition to the adoption of mechanical refrigeration, these included the establishment of an integrated transportation and distribution network from farm to kitchen table, continuing mechanization in corn production, as well as changes wrought by war and the Depression. We expect those changes to manifest themselves in some meaningful way in the underlying structure of the hog–corn cycle, and these changes are not necessarily obvious from simply eyeballing the data presented in Figure 6.1. Of the important changes, the one least studied in the literature is mechanical refrigeration. However, recent research on perishable commodity markets suggests that mechanical refrigeration played a particularly important role in changing the seasonal structure of the hog market and thus the hog–corn cycle (Goodwin et al. 2002; Craig et al. 2004). In particular, refrigeration allowed pork dealers to arbitrage over both time and space. Once upon a time hogs were harvested much like field crops: at a single time during the year; however, access to refrigeration ameliorated the seasonality of the hog market. As noted by Goodwin et al. (2002), the physics of refrigeration had been known to the ancients. The mechanics of refrigerated meat cars, even those using ice rather than mechanical refrigeration, were, however, not perfected until the late 1870s and were not used on a large scale until the mid1880s. Subsequent advances in refrigeration were closely associated with changes in shipping, and the midwestern meat barons, George Hammond, Gustavus Swift, and Nelson Morris were all shipping beef in cars refrigerated with natural ice by the late 1870s. The shipping of dressed beef from Chicago to New York became “firmly established on a remunerative basis” by the mid-1880s (Anderson 1953: 51). Once Herman and Phillip Armour joined the “big four,” they formed the “greatest trust in the world” (Russell 1905), and by the early 1890s, the meat barons controlled the market in “shipped perishables” (Aduddell and Cain 1973: 102; Cronon 1991). Although the use of natural ice in the shipping of livestock and dairy products was an important development in the spatial integration of regional markets in the United States, there remained a bottleneck in the storing (at the wholesale and retail levels), and long-distance shipping of perishables. The subsequent developments in those markets were largely the result of the perfection of mechanical refrigeration. The natural laws of refrigeration were well known for at least three millennia prior to the perfection, in the 1890s, of a mechanical refrigerator that was both reliable and affordable. The basic endothermic process requires vaporization of a liquid, which lowers its pressure and which in turn extracts heat from the surrounding environment. The vapor is subsequently condensed (an exothermic
Did refrigeration kill the hog–corn cycle?
109
process), which raises its pressure and results in the release of heat. By enclosing the vaporization process and placing the condenser outside of the enclosed unit, the mechanical refrigerator was born. This process, which ultimately dominated the meatpacking and wholesaling industry by the late nineteenth century, was first patented in the United States in 1853. However, the early machines were costly to build and maintain, and notoriously unreliable. The keys to refrigeration’s widespread adoption were a set of rather mundane improvements in the machine-tool industry, related metallurgical improvements, the development of high-pressure seals, and the addition of the electric motor, all of which occurred later in the century (Goodwin et al. 2002). The cost remained relatively high, and although by the end of the 1890s mechanical refrigeration largely dominated international shipments and cold storage of perishables, the household refrigerator was several decades away. We argue refrigeration, and more specifically, mechanical refrigeration, is the key to the changing seasonal nature and underlying structure of the hog–corn cycle in the late nineteenth and early twentieth centuries. Before reviewing refrigeration’s impact on the hog–corn cycle, we first consider the propagation of the cycle without refrigeration. Although there exists a large number of possible shocks to either the supply or demand functions for hogs or corn, as Ezekiel (1938) noted in his original research on the topic, many of these are not destabilizing. In other words, rather than leading to a series of disequilibria that mapped the large cyclical downturns or upturns in the hog–corn price ratio, they would have simply nudged temporarily the market away from the “iron-rule” ratio, only to see it return by, say, the next season. However, some of the potential shocks would have been destabilizing without refrigeration but not so with it. To see this, assume that initially the hog–corn price ratio is within epsilon of Shannon’s iron rule of ten-to-one, or eleven-toone, or whatever it happened to be at the time, and hence the two markets are in something like a stable equilibrium. Now suppose that good weather in the Corn Belt leads to a bumper crop, which puts downward pressure on the price of corn relative to hogs. Hog farmers, or at least those who did not anticipate the windfall, would attempt to expand their herds on the back of cheap corn. Between one growing season and the next, there is an increase in the supply of maturing hogs, which will eventually put downward pressure on hog prices. As the previous season’s corn is consumed and the new herds mature and are slaughtered, ceteris paribus of course, the market would once again approach the equilibrium price ratio. Thus periodic fluctuations in crop yields, resulting from climatic fluctuations, would have caused fluctuations in the hog–corn price ratio regardless of any other underlying changes in the supply or demand of hogs or corn. Suppose, however, that as this process unfolded, an unusually poor corn crop followed the bumper crop. The resulting negative supply shock would put upward pressure on corn prices and disturb the downward movement in the price ratio toward the iron-rule price ratio. In a world without refrigeration, as input prices rose, hog farmers, caught in an expansionary phase, would increasingly attempt to slaughter and market all that they could, a move that in turn would
110
L.A. Craig and M.T. Holt
exacerbate the cyclic downturn. To maintain the herd as feed prices rose would eventually, quickly in fact, lead to a situation in which the average variable cost of production would exceed price. In short, rather than returning to equilibrium, the combination of expanding herds and scarce corn would drive the price ratio well below the iron-rule level. To wait until the next growing season and the subsequent slaughter season, the farmer would have to run accounting deficits while maintaining herd sizes in the hopes of cheaper corn next year. Depending on what happened in the next corn-growing season, and the relevant price elasticities, the cycle might continue downward or bottom out and rebound; simply put, a farmer who did not slaughter his non-breeding stock ran a tremendous risk of being caught with livestock inventories that he could not afford to feed. Note that these farmers could in fact be operating on the basis of perfectly rational expectations. It is just that sometimes they are wrong. Since livestock portfolios were to an extent a function of expected crop yields, and since actual crop yields were a function of the weather, predicting hog prices was ultimately tied to the prediction of the weather months in advance. Even modern meteorologists have a poor record at that. Of course, farmers, recognizing the relatively large variance of their forecast errors, would have had an incentive to hedge. Although instruments to hedge directly livestock products in the form of pork bellies did not exist until after the period our data cover, farmers could and did hedge their corn crop, which given that most of it went into hogs was essentially a way of hedging pork as well. The extent to which farmers hedged their crops at the time remains unknown, but two factors worked against complete hedging. First, there were transactions costs, which for many farmers could be prohibitive. Not only were there brokerage fees, but one had to go to town to transact, and the town had to be large enough to have a grain trader dealing in futures and related contracts. For many farmers the size and nature of the risk relative to the costs of hedging it simply kept them out of the market for these instruments.10 Those who did hedge, faced the risk of covering their position if their crop failed and prices were high. The national banking system at the time was not well suited to the needs of the nation’s farmers. As Studenski and Krooss note “Rigidly restricted as [the system] was . . . it could not satisfy the farmer’s need for extra funds in the spring for planting and in the fall for moving crops,” or covering short positions (1963: 179). The dearth of credit markets that would have allowed them to cover their position until the next season meant the optimal share of the crop to hedge was considerably below 100 percent. To exacerbate the problem, once hogs were slaughtered in the face of high corn prices, they could not immediately be replaced when the market turned around. In the absence of mechanical refrigeration in cold storage, the slaughtered carcasses could not be stored, and it took several months before the next litter was farrowed and several more months after that before the barrows and gilts from these litters reached a marketable age. Taken together, these characteristics of the market meant that it was subject to frequent swings in prices. The primary ameliorating factor was refrigeration. Consider the same markets we just described, only now with refrigeration. Faced with high corn
Did refrigeration kill the hog–corn cycle?
111
prices and declining pork prices, the hog farmer need not immediately slaughter the entire non-breeding stock of the herd in the fall or early winter, because there would still be a market in the months in which previously few hogs were slaughtered. In short, a farmer did not have to wait until the end of the next growing season or the beginning of the next slaughter season to market pork. Rather, farmers could maintain some of the herd until the spring or summer, and then slaughter when the supply of hogs was scarcer than it otherwise was in late fall. The implication is that the underlying nature of the hog–corn cycle was itself fundamentally changed. The manifestation of this phenomenon was twofold. First, there should have been a smoothing of the hog–corn price ratio throughout the year.11 Second, because refrigeration allowed farmers to maintain herds beyond the previous slaughter season, without reducing slaughter rates overall herd sizes could be increased. In short, refrigeration did more than simply allow a farmer to hold a hog off the market today for future slaughter. It allowed the farmer to slaughter a hog today and hold another hog off the market for later slaughter. In this manner refrigeration helped smooth the hog–corn cycle and increase output in an important product line (pork, and the corn that generated it). This is the social savings from the adoption of refrigeration in the cold storage of meat. In what follows we estimate both this price and output effect.
Social savings To identify the impact of refrigeration on the seasonal behavior of the hog-corn price ratio, we estimated a simple linear model of the form 3
3
11
HCt = iPi + ijD*jPi + t i=1
(1)
i=1 j=1
where HCt denotes the hog–corn ratio, P1 = 1 if t August 1893, 0 otherwise; P2 = 1 if September 1893 t April 1917, 0 otherwise; and P3 = 1 if t May 1917, 0 otherwise. Likwise, D*j are monthly dummy variables such that D*j = Dj – D12, j = 1, . . . , 11; D*j = 1 if the current month is month j and is zero otherwise. In other words, the Pj dummy variables in (1) divide the data into three equal sub-periods: Period 1 = 1870:01–1893:08; Period 2 = 1893:09–1917:04; and Period 3 = 1917:05–1940:12; whereas the D*j variables are the seasonal dummy variables. This particular formulation of the seasonal dummies allows the ij coefficient to be interpreted separately from the corresponding i intercept term. Although the exact dating of the sub-periods may appear somewhat arbitrary, the periods roughly correspond with what the narrative evidence suggests are a pre-mechanical refrigeration period; a period during which refrigeration was being adopted; and a period in which its use was widespread. Furthermore, adjusting the exact beginning and ending of each period has no qualitative impact on what follows.12 By using estimates associated with the model in (1), percentage changes in seasonal effects may be computed between the periods with and without refrigeration. The results are shown in Table 6.1.
112
L.A. Craig and M.T. Holt
Table 6.1 Implied percentage change in seasonal values for the hog–corn price ratio Month
Period 2/Period 1 (%)
Period 3/Period 1 (%)
January February March April September October November December
–1.29 –4.44 –2.87 –1.44 0.48 4.27 3.92 2.95
–7.03 –7.38 –4.37 –3.90 2.80 8.97 11.25 1.40
Note Period 1 = 1870:01–1893:08; Period 2 = 1893:09–1917:04; Period 3 = 1917:05–1940:12.
The results show that the seasonal component of the hog–corn price ratio in the winter and spring months decreased after mechanical refrigeration became available. This suggests an increase in the supply of what was formally “offseason” pork. With refrigeration, farmers could hold off on slaughtering some of their herds in response to higher relative prices for corn. Conversely, the seasonal component of the fall prices increased, suggesting some reduction of supply during what was previously the peak slaughter season. Taken together, this table offers clear evidence of farmers arbitraging across seasons, and thus smoothing seasonal prices, as a result of refrigeration. As noted above, there was also an output effect, as refrigeration reduced the risk to farmers of holding larger livestock portfolios, ceteris paribus, and thus the increase in the size of the spring slaughter was greater than the decrease in the autumn slaughter. Pork was arguably the most important of the perishables impacted by mechanical refrigeration. By the first decade of the twentieth century, when refrigeration enjoyed wide use, pork alone accounted for nearly 4.5 percent of U.S. GDP. Craig and Weiss (1997, 2000) estimate that pork production represented 17.7 percent of agricultural production, and since agricultural production was roughly 25 percent of GDP, pork’s contribution to GDP was roughly 4.5 percent. One can arrive at almost exactly the same figure by an entirely different method. Consider that per capita pork consumption was roughly 165 lbs in 1900 (Cuff 1992); the average price per lb was $0.06 (Holmes 1913); that is, per capita consumption was $9.90, which was 4.4 percent of per capita national income (Romer 1989: 22). Consumption, too, was affected by refrigeration, because the quality of the stored perishables improved substantially as a result of mechanical refrigeration substituting for salting and curing. Furthermore, the elimination of moisture inherently associated with refrigeration via natural ice gave a considerable boost to the quality of refrigerated products. Given the seasonal dynamics of hog (and corn) markets, the key to understanding the effect on real output rests with refrigeration’s differential impact
Did refrigeration kill the hog–corn cycle?
113
across the seasons. At first glance, it is not even obvious that refrigeration would have a positive impact on output. Since refrigeration permits storage, and in the case of pork, hygienic slaughter, it facilitates arbitrage over time, and thus a pound of pork held off the market in November now shows up in March, suggesting no net impact on output. However, the output effect results from the reduction in the risk of maintaining herds when prices fall. Our objective then is to estimate the net change in output (and ultimately consumption) resulting from the adoption of mechanical refrigeration. First, consider a simple model reflecting the relationship between the two primary pieces of information provided by a market, specifically, prices and quantities. These are, after all, the directly observable outcomes of market transactions. Let Pijt stand for the price and Qijt the quantity of the ith perishable commodity (e.g. pork) exchanged in season j (e.g. autumn or spring) in year t. Then the price elasticity of demand for Qijt is defined as: dln Qijt/dln Pijt = ijt and
dln Qijt = ijtdln Pijt
(2)
Thus, if one has an estimate of the elasticity, and if one knows or has an estimate of the change in price, then one can estimate or predict the change in output resulting from a change on the supply-side of the market.13 This estimate provides a first approximation of the impact of refrigeration on the markets for perishable commodities. Since the autumn market was substantially larger than the spring market, roughly 25 to 50 percent larger, it must be the case that in order for refrigeration to generate a net increase in output (Qa/Qs)|dln Pa| < |dln Ps|. However, there is nothing inherent in the market that would make this condition hold. Estimates of the price elasticity of demand for pork in the early twentieth century are in the neighborhood of –0.40, and more recent estimates put the figure around –0.78.14 Using the price data from Figure 6.1, and the mid-point of the elasticity estimates (i.e. –0.59) reveals that while there was a modest 0.19 percent (–0.59 0.325) decrease in the consumption of pork during the autumn, this was overwhelmed by a 5.49 percent (–0.59 –9.30) mean increase in spring consumption.15 However, we must also account for the fact that autumn consumption was considerably larger than spring consumption to begin with. In Table 6.2 we have calculated the change in pork production resulting from the adoption of mechanTable 6.2 Percentage change in U.S. pork production resulting from the adoption of mechanical refrigeration Elasticity estimate
–0.40 –0.59 –0.78
Ratio of “spring” to “fall” production 0.66
0.77
0.88
1.40 2.07 2.73
1.54 2.28 3.01
1.67 2.47 3.26
114
L.A. Craig and M.T. Holt
ical refrigeration. We use three different elasticity figures and three different ratios of spring-to-fall production. Using the middle figure as a benchmark, our approximation of the impact of refrigeration suggests that it resulted in a 2.28 percent net increase in output, ceteris paribus, of course. Goodwin et al. (2002) put total pork consumption in the 1890s at around 75 pounds per capita, which is below Cuff’s estimate of 165 pounds for 1900. Using the middle of the road estimate of an overall 2.28 percent increase in consumption would translate into an increase of 1.67 pounds of pork consumption per capita per annum using the lower Goodwin et al. figure and 3.75 using Cuff’s higher estimate. This would translate into between roughly 2,000 and 4,000 calories and 200 and 400 grams of protein, per capita per annum, or enough to feed the entire population for one to two days.16 While these figures are considerable in and of themselves, extrapolating them across other perishable commodities would yield an even larger overall impact, perhaps three to four times as large, leading to an overall increase in food consumption of 1 or 2 percent. Yet another way to think about these improvements is in terms of aggregate output. The perishables that depended the most on refrigeration were meat and dairy products. Meat and dairy products represented roughly 30 percent of total agricultural output. Since agriculture was 25 percent of GDP in 1900, meat and dairy made up 7.5 percent of GDP, and a 2.28 percent increase in output resulting from refrigeration would be a 0.17 percent increase of GDP, which represents a measure of the social savings from refrigeration. As other scholars have recently observed, refrigeration was truly one of the “great inventions” (Gordon, 2000).
Conclusions Did refrigeration kill the hog–corn cycle? No. But refrigeration certainly changed the cycle, and in attempting to answer this question we identified three related ones: How were hog cycles propagated? How were they eventually ameliorated? What was the social savings from their amelioration? We argue that the hog cycle was propagated by the combination of weather shocks to the corn market and, in the absence of refrigeration, the inability of hog farmers to smooth production in response to feed price shocks. This failure to smooth was not the result of irrational expectations, but rather the result of the inability of farmers to perfectly predict the weather or its impact on the corn market and their inability to store inventories of slaughtered hogs. Some variation of unusually good crop years followed by bad ones could lead to large swings in the hog–corn price ratio. The dynamics of animal populations – i.e. animals could be slaughtered over night, but it took time to produce a new herd – caused the market to occasionally swing from one extreme to another. The seasonal cycle was ameliorated by the advent of mechanical refrigeration, which allowed farmers to both smooth and increase production. Refrigeration allowed farmers to hold livestock off the market when prices were relatively
Did refrigeration kill the hog–corn cycle?
115
low during the prime slaughter season. Thus, fall prices rose in response to a reduction in the fall supply, and spring prices fell, as hogs previously marketed in the fall could now be marketed in the spring. There was also an output effect from refrigeration. Since refrigeration allowed farmers to arbitrage over time and space, it reduced their exposure to price swings and they could accordingly increase herd sizes. This increase in production represented a social saving. We estimate this saving was in the neighborhood of 0.17 percent of aggregate output, which in turn corresponded with a 1 to 2 percent increase in food consumption. We argue that these improvements were large by most reasonable standards.
Notes * The authors thank Jeremy Atack, Joshua Rosenbloom, participants of the “Putting Things in Perspective: Hogs, Sewage, Bastards and Other Cliometric Issues” Conference in honor of Thomas Weiss, held at the University of Kansas, April 2006, and seminar participants at Purdue University, the Stockholm School of Economics, Ball State University, Eastern Carolina University, the Triangle Universities‚ Economic History Workshop, and the Washington, DC-area Universities‚ Economic History Workshop. 1 In fact the advent of the instruments themselves dates from centuries earlier; however, the market in corn futures and options dates only from the middle of the nineteenth century (Ferris 1988). 2 The origin of maize is much debated among scientists. For a fine account of the history of the debate, see Kahn (1985: 1–82). 3 Wendell Berry describes in a rather vivid albeit fictional account a fall slaughter scene on a Kentucky farm circa 1891 (Berry 2004: 12–24). 4 While the long-run trend was clearly toward specialization, there was considerable variance in farmers’ responses in the short run. See Bogue (1993), Craig and Weiss (1997), and Gregson (1993, 1994). 5 Hence, among other things, Hollywood has generally not focused on making “Midwesterns” wherein the driving of hogs to market would supposedly have been romanticized and idealized. For similar reasons “Pigboys” have not become part of the cultural mystique of early frontier life in America. 6 On the typical drive, cattle lost more weight by volume than hogs did, but the difficulty of herding hogs offset this advantage (Gates 1960: 210). 7 There is also evidence of “whiskey cycles,” but that is another topic. 8 As well, the introduction of steam navigation on the Ohio River in 1811 and the completion of the Miami and Erie Canal in 1828 effectively linked Cincinnati by water to the rest of the world, thereby making transport of salted and cured pork economically feasible. 9 The corn price used is the wholesale price of corn in Chicago, dollars per bushel, and the hog price is the wholesale price of hogs in Chicago, dollars per hundredweight (Chicago Board of Trade various years; and Wallace 1920). 10 In a more modern setting, Berck (1981) concludes that due to transactions costs and yield risks California cotton farmers have only limited incentives to use production hedges. Presumably similar arguments would apply to a midwestern corn farmer circa 1890. 11 The ratio would not be expected to be perfectly smooth throughout the year, because in both markets there were other exogenous factors, all of which followed their own economic logic.
116
L.A. Craig and M.T. Holt
12 Furthermore, equation (1) nests a model wherein there is no change in seasonality across periods. However, an F-test of the model with no structural change vis-à-vis equation (1) yielded a p-value of 0.0015, leading us to conclude that the structural change implied in equation (1) is supported by the data. 13 This assumes that the short-run supply function is perfectly inelastic beyond some point, which seems reasonable given that once the herd is slaughtered it cannot be reconstituted for several months or more. 14 For example, Deaton and Muellbauer (1980: 67) cite a figure of –0.40. The original source is Stone (1954). More recently, Brester and Wohlgenant (1991) put it at –0.78. 15 The percentage change in mean autumn pork prices between periods 1 and 3 in Table 6.1 was 0.325; the percentage change in mean spring pork prices between periods 1 and 3 was –9.30. 16 The pounds, calories, and protein estimates are calculated using the conversion rates in Craig et al. (2004).
References Aduddell, R. M. and Cain, L. P. (1973) “Location and Collusion in the Meat Packing Industry.” In L. P. Cain and P. J. Uselding, Eds. Business Enterprise and Economic Change. Kent, OH: Kent State University Press. Anderson, O. E. (1953) Refrigeration in America: A History of a New Technology and Its Impact. Princeton, NJ: Princeton University Press. Atack, J. and Bateman, F. (1987) To Their Own Soil: Agriculture in the Antebellum North. Ames, IA: Iowa State University Press. Berck, P. (1981) “Portfolio Theory and the Demand for Futures: The Case of California Cotton.” American Journal of Agricultural Economics 63: 466–474. Berry, W. (2004) That Distant Land: The Collected Stories. Washington, DC: Shoemaker & Hoard. Bogue, A. G. (1993) “Communication.” Agricultural History 67: 105–107. Bonner, J. C. (1964) A History of Georgia Agriculture 1732–1860. Athens, GA: University of Georgia Press. Brester, G.W. and Wohlgenant, M.K. (1991) “Estimating Interrelated Demands for Meat Using New Measures for Ground and Table Cut Beef.” American Journal of Agricultural Economics 73: 1182–1194. Brinkley, G. (1994) The Economic Impact of Disease in the American South, 1860–1940. Unpublished Ph.d. Dissertation, University of California, Davis, CA. Buley, R.C. (1950) The Old Northwest: Pioneer Period, 1815–1840, Two volumes. Bloomington, IN: Indiana University Press. Chavas, J. P. and Holt, M.T. (1991) “On Nonlinear Dynamics: The Case of the Pork Cycle.” American Journal of Agricultural Economics 73: 819–828. Chicago Board of Trade (various years) Annual Report. Chicago, IL: Board of Trade. Clemen, R. A. (1923) The American Livestock and Meat Industry. New York: Ronald Press. Coase, R.H. and Fowler, R.F. (1937) “The Pig-Cycle in Great Britain: An Explanation.” Economica 4: 55–82. Collins, E. J. T. (1993) “Why Wheat? Choice of Food Grains in Europe in the Nineteenth and Twentieth Centuries.” Journal of European Economic History 22: 7–38. Craig, L. A. (1993) To Sow One Acre More: Childbearing and Farm Productivity in the Antebellum North. Baltimore, MD: The Johns Hopkins University Press. Craig, L.A. and Weiss, T. (1997) “Long Term Changes in the Business of Farming:
Did refrigeration kill the hog–corn cycle?
117
Hours at Work and the Rise of the Marketable Surplus.” Paper presented at the International Business History Conference, Glasgow, Scotland, July 1997. Craig, L.A. and Weiss, T. (2000) “Hours at Work and Total Factor Productivity Growth in 19th-Century U.S. Agriculture.” Advances in Agricultural Economic History 1: 1–30. Craig, L. A., Goodwin, B., and Grennes, T. (2004) “The Effect of Mechanical Refrigeration on Nutrition in the United States.” Social Science History 28: 325–336. Cronon, W. (1991) Nature’s Metropolis: Chicago and the Great West. New York: W. W. Norton and Company. Cuff, T. (1992) “A Weighty Issue Revisited: New Evidence on Commercial Swine Weights and Pork Production in Mid-Nineteenth Century America.” Agricultural History 66: 55–74. Danhof, C. H. (1969) Change in Agriculture: The Northern United States, 1820–1870. Cambridge, MA: Harvard University Press. Deaton, A. and Muellbauer, J. (1980) Economics and Consumer Behavior. Cambridge: Cambridge University Press. Ezekiel, M. (1938) “The Cobweb Theorem.” Quarterly Journal of Economics 53: 255–280. Ferris, W. (1988) The Grain Traders: The Story of the Chicago Board of Trade. East Lansing, MI: Michigan State University Press. Gallman, R. E. (1996) “Dietary Change in Antebellum America.” Journal of Economic History 56: 193–201. Gates, P. W. (1960) The Farmer’s Age: Agriculture, 1815–1860. New York: Holt, Rinehart, and Winston. Goodwin, B. G., Grennes, T., and Craig, L. A. (2002) “Mechanical Refrigeration and the Integration of Perishable Commodity Markets.” Explorations in Economic History 39: 154–182. Gordon, R. J. (2000) “Does the ‘New Economy’ Measure up to the Great Inventions of the Past?” NBER Working Paper No. 7833. Cambridge: National Bureau of Economic Research. Gregson, M. E. (1993) “Specialization in Late Nineteenth-Century Midwestern Agriculture.” Agricultural History 67: 16–35. Gregson, M. E. (1994) “Reply to Professor Bogue.” Agricultural History 68: 127–128. Haas, G. C. and Ezekiel, M. (1926) “Factors Affecting the Price of Hogs.” U.S. Department of Agriculture Bulletin No. 1440. Haines, M. R., Craig, L. A., and Weiss, T. (2003) “The Short and the Dead: Nutrition, Mortality, and the ‘Antebellum Puzzle’ in the United States.” The Journal of Economic History 63: 382–413. Holmes, G. K. (1913) “Cold Storage and Prices.” Bureau of Statistics Bulletin No. 101, United States Department of Agriculture, Washington, DC. Holt, M. T. and Craig, L. A. (2006) “Nonlinear Dynamics and Structural Change in the U.S. Hog–Corn Cycle: A Time-Varying STAR Approach.” American Journal of Agricultural Economics 88: 215–233. Kahn, E. J. (1985) The Staffs of Life. Boston, MA: Little, Brown. Komlos, J. (1996) “Anomalies in Economic History: Toward a Resolution of the ‘Antebellum Puzzle’.” Journal of Economic History 56: 202–214. Kurlansky, M. (2002) Salt: A World History. New York: Walker and Company. O’Rourke, K. H. (1997) “The European Grain Invasion, 1870–1913.” Journal of Economic History 57: 775–801.
118
L.A. Craig and M.T. Holt
O’Rourke, K. H. and Williamson, J. G. (2002) “When did Globalisation Begin?” European Review of Economic History 6: 23–50. Rasmussen, W. D. (1960) Readings in the History of American Agriculture. Urbana, IL: University of Illinois Press. Romer, C. D. (1989) “The Prewar Business Cycle Reconsidered.” Journal of Political Economy 91: 1–37. Russell, C. E. (1905) The Greatest Trust in the World. New York: Ridgway-Thayer. Salvador, R. J. (1997) “Maize.” In M. S. Werner, ed. The Encyclopedia of Mexico: History, Culture, and Society. Chicago, IL: Fitzroy Dearborn Publishers. Shannon, F. (1945) The Farmer’s Last Frontier: Agriculture, 1860–1897. New York: Farrar and Rinehart. Stone, J. R. N. (1954) “Linear Expenditure Systems and Demand Analysis: An Application to the Pattern of British Demand.” Economic Journal 64: 511–527. Studenski, P. and Krooss, H. E. (1963) Financial History of the United States, 2nd Edition. New York: McGraw Hill. Wallace, H. A. (1920) Agricultural Prices. Des Moines, IA: Wallace Publishing. Wilder, L. I. (1971) Little House in the Big Woods. New York: Harper and Row.
7
Measuring the intensity of state labor regulation during the Progressive Era Rebecca Holmes, Price Fishback and Samuel Allen*
Labor regulations were the lynchpins of the Progressive movement. During the late nineteenth and early twentieth centuries, there was a tremendous expansion in the role that state governments played in regulating labor markets and labor conditions. In the 1870s, 1880s, and 1890s most states established bureaus to collect labor statistics and regulatory bodies to inspect boilers, factories, and mines. Many passed employer liability laws that served to expand the liability of employers for workplace accidents and the vast majority eventually regularized the accident compensation process by establishing strict liability in the form of workers’ compensation laws. Limits were established for child labor and women’s hours. Some states passed laws that promoted unionization by outlawing “yellow dog” contracts and protecting union trademarks and labels. On the other hand, other states seemed bent on limiting unionization with the passage of anti-enticement laws and laws that limited picketing which were specifically targeted at reducing intimidation of non-union workers. There has been a growing literature examining the quantitative impact on labor markets of the leading progressive laws in the late 1800s and early 1900s.1 While each of the studies provides invaluable evidence on how the individual laws influenced specific aspects of the labor market, they do not capture the broad range of labor laws that were established during the period. On several occasions the U.S. Commissioner of Labor and later the Bureau of Labor Statistics documented the extent of state labor legislation in the various states in a series of reports. In these reports, the Labor Department reported on roughly 135 laws that influenced labor markets and workplace conditions. The thought of examining the impact and political economy associated with 135 separate laws when we have only 48 states is mind-boggling. In terms of degrees of freedom, we are starting 87 degrees below zero, which is probably a record, even for the Arctic Circle. Therefore, we need to find ways to lift the “curse of dimensionality.” How can we accumulate information on these laws into a handful of conceptual measures that capture the essence of the regulatory climate created by the laws? We explore two ways of aggregating the laws based on summing the number of laws and creating employment weighted indices.2 After showing the correlations between these measures, we examine the relationships that the different
120
R. Holmes et al.
measures have with manufacturing value added per worker. After showing the results from OLS estimation, we are able to reduce some of the potential endogeneity because we can control for a variety of time-invariant features of the states using state fixed effects. The preliminary results suggest that the labor regulations were not associated with higher or lower labor productivity.
State labor legislation: patterns and sources State labor legislation came in several waves. As seen in Table 7.1, soon after the Civil War a number of mining states, led by Pennsylvania, began passing rudimentary mine safety regulations. Meanwhile, Massachusetts in 1869 was the first to establish a general bureau of labor. The early bureaus in the late 1870s had little coercive power. They primarily surveyed workers about wages and working conditions and made recommendations about improvements. In 1878 and 1879 New Jersey and Massachusetts led the way in establishing factory inspectors to enforce and administer regulations. In the latter part of the nineteenth century a number of states passed early child labor laws, refined the nature of accident liability for employers, provided political protection for workers as voters, and established a series of laws that gave unions more legal status. In the first decade of the twentieth century, the early forms of labor legislation spread to a majority of states, and many of the existing laws were refined and updated. The next wave of legislation followed in the 1910s as states became more involved with social insurance, introducing mothers’ pensions and replacing the employer liability system with the statutory rules of workers’ compensation. Nearly half of the states passed women’s hours laws during this period and about 20 percent established some form of minimum wages for women and children. At the same time the leading labor legislative states reorganized their state labor bureaucracies into industrial commissions and some established child labor commissions. The waves of legislation can be seen in Table 7.2 which shows the number of states that had adopted each of the 135 laws in four years: The year 1894 follows the establishment of many state labor bureaus, 1908 shows the status of legislation on the eve of the expansion in workers’ compensation and social insurance legislation that followed in the 1910s, the year 1918 shows the situation as the U.S. entered World War I, and the year 1924 shows the setting after the vast majority of Progressive Era legislation had been passed. A law is listed in the table if it was adopted in at least one state. To collect the information on the laws, page counts for the laws, and the appropriations for labor legislation were started with reports of the U.S. Commissioner of Labor (1896, 1904, 1908) and the U.S. Bureau of Labor Statistics (1914, 1925) in their volumes on “Labor Laws in the United States.” These volumes contained full texts for all of the labor legislation that the compilers could find for each state, collected within a uniform framework and print font. We then went back to the original state legislative acts to fill in gaps in our knowledge about the timing of the introduction of the laws. When we saw anomalies in the evidence we investigated further using
Measuring state labor regulation
121
Table 7.1 Year of introduction of labor commission, factory inspectors, department of labors, and industrial commissions State
First labor bureau
Factory inspection adopted
Alabama Arizona Arkansas California Colorado Connecticut Delaware Florida Georgia Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota Mississippi Missouri Montana Nebraska Nevada New Hampshire New Jersey New Mexico New York North Carolina North Dakota Ohio Oklahoma Oregon Pennsylvania
1907a 1925b 1913 1883 1887 1887 1893 1893d 1911 1890f 1879 1879 1884 1885 1892h 1900 1887 1888i 1869 1883 1887j 1914 1879 1893l 1887n 1915 1893 1877
1907a
Rhode Island South Carolina South Dakota Tennessee Texas Utah Vermont Virginia Washington
b
Industrial commission introduced
Extent of code-writing by industrial commissions
1925
Few
1913 1915
Extensive No codes
1917
No codes
1928 1913
No codes Extensive
1915 1929 1919 1917
No codes No codes Few No codes
1913 1931 1919 1913
Extensive
1920 1913 for mines only
Few Extensive
c
1885 1911 1887 1893 e
1916 g
1893 1899 1897 1901 1903 1908 1887 1898 1879 1893 1891 1914 1891k l m
1895n 1915 1917 1878
o
o
1882 1887 1899 1877 1907 1903 1872
1883
1887 1912 1890 1881–1884q 1911 1892 s 1912 1897 1903
1894 1912
e
1905 1884 1910 1907 1889
No codes Extensive
p
1897r 1911 1917 1912 1919 1910
1923
Few
1917
Extensive
1919
Few
Permanent workers’ compensation law 1919 1913 1939 1911 1915 1913 1917 1935 1920 1917 1911 1915 1913 1911 1914 1914 1915 1912 1911 1912 1913 1948 1926 1915 1913 1913 1911 1911 1917 1913 1929 1919 1911 1915 1913 1915 1912 1935 1917 1919 1913 1917 1915 1918 1911
122
R. Holmes et al.
Table 7.1 continued State
First labor bureau
Factory inspection adopted
Industrial commission introduced
Extent of code-writing by industrial commissions
Permanent workers’ compensation law
West Virginia Wisconsin Wyoming
1890t 1883 1917
1899 r 1883 1917
1911
Extensive
1913 1911 1915
Sources: First Labor Bureau refers to the introduction of either a commissioner of labor, a bureau of labor statistics, or a factory inspector. Factory inspection adopted refers to the first statutory provision for a factory inspector. For dates of adoption of inspectors and departments of labor we started with evidence from Brandeis (1966 [1935]: 628–645) and the U.S. Commissioner of Labor (1896). When the precise date of introduction was unknown, the microfiche for the State Session Laws of American States and Territories was searched until the original act was found. The earliest commissioner of labor was in Massachusetts in 1869 and the earliest factory inspector was in Massachusetts in 1879. The initials n.a. under the coal laws means that there was so little coal mining that regulation of coal was not applicable. Year of coal mining law adoption is from Aldrich (1997: 70). Information on Industrial Commissions is from Brandeis (1966 [1935]: 654), who was citing work of John Andrews of the American Association of Labor Legislation. The information on the adoption of workers’ compensation is from Fishback and Kantor (2000: 103–4). Notes a Alabama had a mine inspector and later a board of arbitration but no official department of labor. b Arizona had a mine inspector as of 1908. c Arkansas had an inspector of mines in 1894 or earlier. d The Florida Agriculture department was given the responsibility to collect statistics on manufactures. e No law as of 1924. f Idaho established Commission in Constitution. No record of laws passed between 1879 and 1890. g Idaho had an inspector of mines in 1893 or earlier. h The Kentucky commissioner was to devote efforts to collect statistics on agriculture, manufacturing, and mining. i The initial Maryland Law in 1868 was for agriculture and industry with most of the focus on agriculture. The code of 1888 with amendments in 1892 is more specific to industry. j The Minnesota Law included language about enforcing laws and prosecuting violations by the commissioner but only funds for the commissioner were provided. k Missouri statute for inspector in 1891. Not found in earlier years. l The Montana act established a bureau of agriculture, labor, and industry. m Montana had a mine inspector in 1895 or earlier. n Nebraska gave the commissioner the power to inspect workplaces. o New Mexico had a mine inspector as of 1908. p South Dakota had a mine inspector as of 1903. q The Tennessee Law called for the Bureau of Agriculture, Mining, and Statistics to collect information on labor. The original Bureau of Agriculture was established in 1871, became the Bureau of Agriculture, Mining, and Statistics in 1875, but appears to have obtained the role of collecting labor statistics sometime between 1881 and 1884. We have had trouble pinning down the date. r In Tennessee and West Virginia there were no regular inspectors. Commissioner merely had the power to inspect. s The Utah legislature had authorized a bureau of labor statistics or labor department earlier. t West Virginia gave the commissioner the power to inspect workplaces but only to report on findings there.
Measuring state labor regulation
123
Table 7.2 Number of states with each type of law, 1895, 1908, 1918, and 1924 Law
1895
1908 1918 1924
Employer Liability Law Restates Common Law General Railroads Street Railroads Mines Can not make employees to sign contracts waiving damages
15 21 16 1 1 14
28 47 31 8 4 25
23 48 32 7 4 28
21 47 32 7 4 28
0 0
0 4
37 30
42 43
Factory safety Rehabilitation commission Industrial safety commission Sanitation/bathroom regulations Ventilation Guards required on machines Electrical regulations Building regulations Other Bakery regulations Sweatshop regulations Fire escapes Factory inspector Occupational disease reporting Steam boiler inspector/violation of safety laws
0 0 11 10 12 0 5 1 7 9 23 15 1 15
0 0 22 22 22 0 13 3 14 11 30 29 1 17
0 9 34 25 34 6 23 10 27 14 36 39 16 15
4 17 35 26 35 8 24 11 32 14 37 41 17 17
Reporting of accidents Mine accidents Railroad accidents Factory accidents
19 3 10
26 21 14
33 36 22
32 39 23
Railroad regulations Safety regulations Street railroad safety regulations Railroad inspectors
20 7 4
32 28 7
45 30 6
45 30 6
Mining Regulations Mine inspectors Mine safety regulations: employees/individuals Mine safety regulations: companies Fine for failure to weigh coal-no screening Fine for mine inspector failing to do his job Miners’ hospital and or home No women and children in mines
23 18 22 14 9 4 25
30 23 30 21 13 5 31
33 30 33 22 17 5 35
33 32 35 23 18 5 35
Child labor Child safety commission Child labor inspector Children in manufacture/mercantile/mechanical jobs Minimum Age Penalty for false certificate of age
0 13 20 17 16
1 30 42 33 36
10 40 42 40 38
14 41 44 42 38 continued
Social insurance Workers’ Compensation Law Mother’s pension
124
R. Holmes et al.
Table 7.2 continued Law
1894
1908 1918 1924
Certificates of age required for employment Fine for children working to support idle parents No children cleaning or handling moving parts No children in immoral jobs (acrobat) Is this street job? No women and children in bars
19 1 10 25 6
38 7 20 30 23
45 7 36 34 5
46 7 39 34 6
Child hours law General Mercantile Mechanical Textile Other Minimum age for night labor for children
7 6 18 15 2 7
18 15 30 27 8 29
30 22 30 27 10 42
35 22 28 26 10 45
Women’s regulation Special accommodations (seats) Earnings of married women belong to her
23 31
33 43
44 46
44 46
Women’s hours Night labor General/all employment Mercantile Mechanical Textiles
3 2 3 12 8
4 6 8 16 13
11 24 24 25 25
13 28 27 28 27
Holidays No work on legal holidays Labor Day a holiday Sunday labor fines
0 29 43
0 48 48
3 51 49
3 51 50
Hours laws Textiles Mines Manufacture Railroads Street railroads Public employment Other General hours Law Public roads One hour for meals
6 5 7 8 8 14 5 13 2 6
6 13 7 26 10 22 5 12 23 9
6 15 8 27 10 29 11 11 16 17
6 15 9 27 10 30 11 11 16 19
1 9 5 11
10 9 5 11
12 10 10 11
13 11 14 11
25 14 4 4
11 9 1 6
9 9 1 7
9 9 2 7
15
16
19
21
Unionization and bargaining False use of union cards or employers’ certificates Incorporation of labor unions Labor organizations exempt from antitrust Enticement fines Interfere with or intimidate in railroads or workers abandoning trains or obstructing track Interference with railroad employees Interference with street railroad employees No intimidation of miners Illegal to interfere with a business or the employment of others
Measuring state labor regulation
Law Anti-picketing Anti-boycott Strikes: agreements not to work allowed Conspiracy vs. workmen (conspiring to prevent someone from working Labor agreement is not a conspiracy Anti-intimidation No blacklisting Yellow dog contract (not allowed to join a union as a condition of employment) (illegal for anyone to coerce, to join, or not to join a union) Prohibition on hiring armed guards Industrial police are legal Misrepresentation about a strike or other job characteristics Limits on injunctions Criminal syndicalism (advocating violence or sabotage for political or industrial ends) Labor organizations–embezzlement of funds by officers Combinations of employers to fix wages illegal Trespass on mines, factories, without consent of owner Union trademark fine
1894
125
1908 1918 1924
0 3 3
2 7 5
2 5 7
6 5 10
11 6 19 14
14 8 23 23
15 8 26 25
17 10 27 25
11 17 1 2 0
16 12 9 7 1
12 9 19 12 4
12 9 22 13 8
0 2 0 1 25
0 2 1 1 42
7 3 0 0 43
19 3 0 0 44
Convict labor Convict labor regulations
22
27
32
33
Bribery, coercion, or gouging Foreman accepting fees for employment illegal Bribing employees Coercion of employees is illegal Company stores cannot gouge
1 0 10 6
4 13 13 8
12 17 19 8
14 17 19 8
Political protections Coercing the votes of employees illegal Time off to vote
30 18
33 22
38 24
38 24
Administrative Bureau of labor statistics or department of labor State board of arbitration Free employment offices
28 20 0
34 26 14
43 32 23
44 33 32
2 5 3
1 12 3
0 14 3
0 17 3
3 11
6 25
11 35
12 42
3 9 2
1 19 5
1 22 6
2 23 6
Alien labor Importing alien labor illegal No aliens in public employment Chinese labor illegal Employment agents Emigrant agent license Regulation of Employment Agencies Occupational licensing Railroad telegraph operators (also minimum age) Plumbers Horseshoers
continued
126
R. Holmes et al.
Table 7.2 continued Law
1894
1908 1918 1924
Chauffers Aviators Other Motion picture operator Barbers Steam engineers (firemen) Mine manager Elevator operators Railroad employees Electricians Stevedores
0 0 0 0 1 11 7 1 1 0 2
1 0 0 0 13 16 11 2 1 1 2
27 2 2 8 16 17 13 2 1 2 2
36 6 2 8 16 17 16 2 1 4 2
Anti-discrimination Cannot fire due to age only Sex discrimination Antidiscrimination
0 3 1
1 3 1
1 4 1
1 6 1
1 19 20 1 5 1
1 29 26 9 6 2
3 28 32 9 7 2
4 30 37 12 8 2
6
9
10
10
Minimum wages Minimum wage for public work Minimum wage for women/children (<18) Minimum Wage Commission
1 0 0
4 0 0
9 12 9
10 14 10
Miscellaneous Illegal to desert a ship
5
1
0
0
Wage Payments Nonpayment Wages in cash Wage payment frequency Repayment of advance made by employer No forced contributions by employers Railroad workers: notice of reduction of wages required Fine for no notice of discharge if employee has to give notice too
Source: Holmes (2003: chapter 3). Includes all states as of 1912, Alaska, Hawaii, and the District of Columbia.
the state codes, legislative acts, and other secondary sources, particularly Elizabeth Brandeis’s (1966) discussions of labor legislation from around 1936. This method leads to some margin for error. Laws that were passed and stricken from the books, either because they were repealed or vacated by court decision, between the dates of the volumes will be missed. After working through the legislative acts, their indexes, and the reports of the U.S. Labor Commissioner and Bureaus, we believe that this is not a serious problem. The laws are listed under a variety of headings, ranging from employer liability and workers’ compensation laws to collective bargaining legislation to minimum wages. Even grouping them this way suggests as many as 24 natural headings of types of laws, which is still a large degree of dimensionality. One
Measuring state labor regulation
127
warning for scholars interested in using the information for further research on specific legislation: the information for the overall index in many cases is just a signal that a law of this type exists. Many of the specific dimensions of the laws are still not included. For example, there is one listing for workers’ compensation laws, but Fishback and Kantor (2000) documents quite a bit of variation in the generosity of the benefits and in other features established by those laws. We will take that variation into account in some of the indices. Similarly, the mine safety regulations include a wide range of clauses on specific aspects of mine safety described in much greater depth in work by Fishback (1992, 1987), Graebner (1976), and Aldrich (1997). We know that similar stories can be told for many of the other specific topics listed here.3 Therefore, scholars interested in particular regulations are encouraged to use these measures as a starting point for further investigation rather than a substitute for in-depth work on each particular aspect of labor laws.
Measures of labor regulation We explore two single-index aggregation measures of labor market regulation that have been used by quantitative social scientists: (1) a raw sum of the total number of laws of all types in each state with laws that harm labor given a negative value; and (2) an employment-share weighted index of labor laws that creates separate indices for different types of labor laws and then accumulates them into an overall index. In the course of constructing the index, we also weight the laws by the share of workers who are affected by the laws.4 Law sums One method of aggregating state labor laws is to sum the total number of laws in each state at each point in time. Each law with subscript i is given a value List, which is 1 when the law is in force in state s in year t, and zero when it is not. Then a raw sum Rst for state s in year t is calculated for all of the laws. Rst = ∑ List, over all i This simplifies the process greatly and can work effectively in situations where (1) it is difficult to establish the relative importance of laws, (2) some laws might well be important but not studied very carefully by earlier scholars and therefore might be underweighted by focusing on only the well-studied laws, and (3) a simple sum might serve as a first look at the breadth of coverage of the laws. The raw sum measure has obvious flaws because it treats all laws as equal in importance. In some cases a difference in the number of laws inaccurately measures the true strength of labor regulation. As one example, a state with separate laws for hours of labor in railroading, manufacturing, and mining would have a law count of three, yet a single law covering all workers in another state is a stronger law. We can resolve this problem by using an employment-share-
128
R. Holmes et al.
weighted index. Before summing, the laws are weighted by the industry’s share of employment. In this specific setting the employment-share-weighted sum and the raw sum are very highly correlated, so we focus on the raw sum in the rest of the chapter.5 Laws that sound good on paper might not have had much impact. Under employer liability Fishback and Kantor’s (2000) research on the origins of workers’ compensation laws shows that some employer liability laws simply restated the common law and therefore really did not change the de jure or the de facto liability of employers. This type of law is counted in the raw sum, but when we move to indexes below, we will change the way we deal with employer liability and thus it will no longer play a role. In some cases the laws might make life more difficult for workers. For example, Table 7.1 includes a set of labor laws targeted at limiting the effectiveness of labor unions by treating unions as illegal, limiting picketing and boycotts, and making it illegal to trespass without an employer’s permission, among several. Others limited labor mobility by creating fines for employers who tried to entice workers to leave their current places of work. Therefore, we assigned these antilabor laws negative values. Even here it is sometimes difficult to determine which laws have negative and which have positive consequences for workers. Economists note that minimum wage laws, for example, benefit some workers who receive wage increases without having to exert any new effort while harming workers who become unemployed or find it harder to find new work. The raw indices for the 48 continental states and territories in 1899 are plotted against the 1919 indices in Figure 7.1. There are several lessons that we can learn from this graph. First, all but one of the states are located in the upper 80
CA MD
Raw sum 1919
60
OR
OK NV
AZ
40
UT NDME ID KY VA AR
SC DENC VT MS
KS
TX
FL
MT WA CT TN LA EWY NH WV RI
NY CO WI
MA
OH IN MO
PA
NJ ILIA
SD GA AL
NM
20
0
20
40 Raw sum 1899
60
Figure 7.1 Plots of raw sum of labor regulations in 1919 and 1899.
Measuring state labor regulation
129
left portion of the figure, which implies that they have more laws in 1919 than they did in 1899. Generally, this is consistent with the narrative evidence from the Progressive Era, which suggests that regulations had expanded in scope and depth. However, it should be noted that there will be inertia or path dependence in the number of laws because states find it easier to enact legislation than to repeal it. Thus, ineffective laws or ones no longer enforced often remained on the books. The one state that experienced a net decline in the number of laws was New Jersey, which ranked second in terms of laws to Pennsylvania in 1899.6 Second, the states are aligned in such a way that their relative rankings in 1919 are generally similar to their rankings in 1899. The cross-sectional correlation between the law sum in 1919 and in 1899 is 0.73, and the 1899–1909 and 1909–1919 correlations are both above 0.82. Generally, the Mid-Atlantic states of Pennsylvania, New York, and New Jersey, the New England states and the Midwestern progressive states where John R. Commons and his academic colleagues played such a role, Wisconsin and Minnesota were among the leaders in both periods.7 The western states of California, Colorado, and Montana were grouped around the edge of the top ten. Indexes In the indexing method we use some of the knowledge that we have accumulated in studying the laws to take more control of the weighting by assigning the laws to various categories. We then create two types of indexes with different weights, a raw index and an employment-share weighted index. Raw index The first is a basic index where laws fitting a specific category are summed together and then the sum is adjusted so that the index for that category lies between zero and one generally by dividing by the number of laws in the category. The Index Ijst for the jth category of laws in year t for state s is the sum of the law variables Lijst in category j divided by the number of laws in category j, nt,8 Ijst = ∑ Lijst/nj for i = 1 to nj For example, we identified 12 factory safety laws that deal with different aspects of factory safety. We created a factory safety subcategory such that for each state and year the laws were summed and then turned into an index by dividing by 12. Therefore, the maximum value that could be reached in this subcategory was one and the minimum was zero. There were no states that had all 12 laws in 1899 or in 1919, so the maximum in this category was 0.917.9 There are two categories in which the index can be negative. In the miscellaneous category laws designed to prevent out-migration by licensing emigrant agents were given a negative value, so the miscellaneous category might have a
130
R. Holmes et al.
negative value in some states. The union regulations include both measures favoring unions and limiting union activity. Within that category we create an index that sums the nine pro-union regulations and divide by nine and another index that sums the eight anti-union regulations and divide by eight. The union regulation index is calculated by subtracting the anti-union index from the pro-union index. The overall union regulation index in some cases might be negative. To create the overall index, the indexes are summed together and the total is divided by the 24 categories of laws that were used. Ist = ∑ Ijst/24 for j = 1 to 24 = ∑ (∑ Lijst/nj for i = 1 to nj)/24 for j = 1 to 24 The raw index was built on the following 24 categories: employer liability and workers’ compensation; steam boiler inspections; factory regulations; mining regulations; railroad regulations; street railroad regulations; miscellaneous industry regulations; child regulations; women’s regulations; occupational licensing; women’s hours regulations; children’s hours regulations; men’s hour regulations; holiday regulations; union regulations; regulations dealing with bribery, coercion, and political freedoms of workers; presence of a department of labor or bureau of labor statistics; boards of arbitration, mediation, and conciliation in labor disputes; a rehabilitation commission; free employment offices; anti-discrimination regulations; payday regulations; an industrial commission; and minimum wage regulations and enforcement. (Note that the 24 categories above are separated by semi-colons.) We treated the industrial safety commission as a separate category because several of them obtained rule-making authority that had not been awarded to factory regulators. Two sub-category indexes were not based on pure law counts. In the employer liability and workers’ compensation category, we used Fishback and Kantor’s (2000) expected benefits measures as a percentage of annual income as the basis for the index. For years when workers’ compensation was in place the expected benefits were calculated using the specific parameters for the laws. The typical workers’ compensation value was above 1 and ranged up to 2.82. For nonworkers’ compensation years, we started with an expected benefit to annual income ratio of 0.5 based on Fishback and Kantor’s comparisons of expected payouts under employer liability and workers’ compensation. The expected benefit/income ratio for states without workers’ compensation was then adjusted upwards from 0.5 in states where the actuaries suggested higher employer liability premiums to insurers based on employer liability laws that would lead to higher payouts to workers. The expected benefit/income ratio was lowered from 0.5 in states where lower premiums were suggested by the actuaries.10 To index the values we divided by the maximum value of 2.82 so the value would range between zero and one. In the male hours category, a number of states passed general maximum hours laws in the late 1800s that stated that the official work day was an eighthour day but typically added the proviso, “unless specified otherwise by con-
Measuring state labor regulation
131
tract.” We gave them a maximum value of 0.1. The U.S. Bureau of Labor Statistics suggested that these laws had little effect because of the contracting out clause and lack of enforcement. The laws weakened in effect still further after the U.S. Supreme Court in 1905 declared that New York state regulation of bakers’ hours violated the U.S. Constitution.11 We decided to assign them some positive value because some employers might have routinely followed the law even without much enforcement and because the laws likely triggered some negotiations of contracts that might have led to hours law reductions. Once the new hours structure was in place, these employers might have decided to leave them in place even after the 1905 decision.12 Employment-share weighted indexes Our most preferred measures of all of the measures based on summing the labor legislation are employment-share weighted indexes (ESWI). Weighting by employment share allows us to avoid the problem of putting too much weight on labor legislation that is specific to only one category of k workers in the state’s labor force. At the same time we are able to avoid double-counting regulations that deal with only specific groups. Therefore, each category index for state s becomes ESWIjst = ∑ (Lijstk,/mj)*(Ss1900k) for i = 1 to mj The employment weights (Ss1900k) have very small values when the subscript k refers to a small subgroup of professionals, as was the case for some occupational licenses. The employment weight is 1 when the k subscript refers to regulations that apply to all workers.13 The weights for each employment category k in each state are based on information from the 1900 census on the number of gainfully employed men, women, and children in occupations in 1900 and the average numbers of each type employed in various manufacturing industries in 1899 (U.S. Bureau of the Census, 1904, 1902: volumes 7 and 8; Haines, 2006). We chose the 1900/1899 weights to reflect the situation in the first year for which we make cross-state comparisons of the labor laws. We then continue to use the 1900/1899 weights for all other periods to avoid feedback problems where changes in regulation are causing changes in the shares of people in industries. This weighting still potentially has some problems with feedback effects for the 1899 ESWI. We chose this period to develop the employment shares because we thought that the situation in 1890 would not be reflective of employment in the twentieth century because the late 1800s were a period of industrial change. As another way to avoid problems with specific state distributions being influenced by the labor laws in the state, we have also developed an index where the weights are based on the national distributions across occupations and industries (and thus the s subscript is dropped for the employment share). In calculating the employment share we use as the denominator an “employment base” composed of the number of gainfully employed in trade, transport,
132
R. Holmes et al.
manufacturing, and mechanical pursuits from the 1900 Census of Occupations. These categories include miners and construction workers and were the focus of nearly all of the labor regulations. The denominator probably overstates the number of people likely to be affected by regulation because workers in wholesale and retail trade were less affected by the labor regulations. The workers left out of the base are professional workers, domestic and personal service workers, and agricultural workers. The agricultural, domestic, and personal service workers were often explicitly left uncovered by many regulations. Some professionals were licensed and we do include in the numerator a licensing measure for specific occupations but the workers in these occupations account for a very small share of the professionals and an even smaller share of workers subject to regulation; therefore, we do not see that their absence in the denominator will be a serious problem.14 As in the case of the raw index, we start with the purpose of adding up various regulations in subcategories to create indices for those subcategories. The difference is that we weight each law in the index for the subcategory by the shares of the workforce affected by the group of regulations as of 1900. In the factory regulation situation we described above, 11 of the 12 factory regulations were broad manufacturing regulations and one was a broad measure focused on bakers. We summed the general factory regulations and then divided by 11 to create an index with a maximum value of one if a state had all 11. We then weighted that index by the share of workers in manufacturing. We took the bakery regulation measure and then weighted it by the share of workers in baking. The result was an employment-share-weighted index for factory regulations that had a maximum value of 0.59 (compared to 0.91 for the raw index) and a minimum value of zero. Once we start weighting by employment shares, the number of laws is reduced in situations where the topics of the law are similar but are applied to different industries. To make this more obvious, we changed the symbol for the number of laws in category j to mj. Therefore, in this manufacturing regulation category the number of laws in the category for the raw sum (mj = 12) was reduced to 11 (mj = 11) to prevent the bakers law, which dealt with similar issues to the general factory law, from being counted as a separate law.15 The same issue of double-counting arises when summing the category indexes to obtain the overall ESWI. Starting with the 24 categories described in the raw index above, we can group factory regulations, mining regulations, railroad regulations, street railroad regulations, and occupational licensing into one overall heading of economic activity regulation. Each has a separate index that is multiplied by its employment share and then summed up to get an employment-share weighted index of general regulations. Similarly, we can group women’s hours regulations, men’s hours regulations, and children’s hours regulations into a single category for hours regulations. We group women’s and children’s activity regulations into another heading because the laws for them were often changed on top of the normal regulations. As a result we end up with 17 categories: workplace activity regulation; hours regulation; women and children
Measuring state labor regulation
133
activity regulations; miscellaneous regulations; employer liability and workers’ compensation; steam boiler inspections regulations; union regulations; regulations dealing with bribery, coercion, and political freedoms of workers; presence of a department of labor or bureau of labor statistics; boards of arbitration, mediation, and conciliation in labor disputes; a rehabilitation commission; free employment offices; anti-discrimination regulations; payday regulations; an industrial commission; and minimum wage regulations and enforcement. After summing across the 17 categories we divide by 17 and get a summary employment-share-weighted index of labor regulation.16 Thus, the formula for the overall ESWI for state s in year t is ESWIst = ∑ ESWIjst/17 for j = 1 to 17 = ∑ (∑ (Lijstk,/mj)*(Ss1900k) for i = 1 to mj)/17 for j = 1 to 1717 The analysis that we will perform on manufacturing value added per worker uses an index limited to general regulations and regulations that would apply to manufacturing workers. In the raw manufacturing index the 19 subcategories specific to manufacturing are: employer liability and workers’ compensation; steam boiler inspections; factory regulations; children’s regulations; women’s regulations; women’s hours regulations; children’s hours regulations; men’s hours regulations; holiday regulations; union regulations; regulations dealing with bribery, coercion, and political freedoms of workers; presence of a department of labor or bureau of labor statistics; boards of arbitration, mediation, and conciliation in labor disputes; a rehabilitation commission; free employment offices; anti-discrimination regulations; payday regulations; an industrial commission; and minimum wage regulations and enforcement. The final raw index was then created by summing the indexes for each of the 19 categories and then dividing by 19, so that the index could not exceed one. When we created the ESWI, the women’s and children’s activity regulations are grouped together in one subcategory, as are the men’s, women’s, and children’s hours regulations. The number of categories then becomes 15. The manufacturing indexes are also highly correlated with the total indexes even though the total indexes contain extensive coverage of mining and railroads. All correlations with the same weighting schemes for employment shares were above 0.97 in both 1909 and 1919, and above 0.96 in 1899. The correlations for changes in the indices were above 0.96 for 1919–1899, above 0.91 for 1909–1899, and 0.96 for 1919–1909. When making comparisons of the indices with the other measures, we will focus on the ESWI using state weights, which is very highly correlated with the ESWI using national weights. Correlations between the two indexes were above 0.99 in every year, and correlations between the changes in the indices between 1899 and 1919, 1899 and 1909, and 1909 and 1919 were greater than 0.98. In the rest of the chapter we will focus on the state ESWI rather than the raw index. The raw index and the ESWI have correlations above 0.96 in 1919, above 0.97 in 1909, and above 0.96 in 1899. The correlations of the changes in these
134
R. Holmes et al.
two types of indices were above 0.89 for 1919–1899, above 0.87 for 1909–1899, and above 0.89 for 1919–1909.18 As is the case for the raw index, the states with high and low ESWIs in 1899 also tended to be ranked similarly in 1919 with some slight reshuffling. The correlations for the ESWI are above 0.82 for the 1899/1909 and 1909/1919 comparisons and 0.71 for the 1899/1919 comparisons. The maps in Figures 7.2 and 7.3 are based on the same legend categories; therefore, they show the relative intensity of labor regulation across states within the map and across time in comparing the maps. Pennsylvania and New Jersey, which collectively had the broadest range of industries, displayed the highest labor regulation intensity in 1899 with values above 0.4. The Midwest, California, Colorado, Massachusetts, and Montana were next in line. By 1919 a substantially higher level of labor intensity darkens the map considerably around the nation. Massachusetts and New York led the way with Pennsylvania; the famously progressive midwestern states of Minnesota, Wisconsin, Michigan, and Ohio, and a swath of western states from Colorado west to California close behind.
Legend Under 0.081282553 0.081282553 to 0.160375469 0.160375470 to 0.245034207 0.245034208 to 0.349425930 0.349425931 or more No data
Figure 7.2 ESW index, 1899.
Legend Under 0.081282553 0.081282553 to 0.160375469 0.160375470 to 0.245034207 0.245034208 to 0.349425930 0.349425931 or more No data
Figure 7.3 ESW index, 1919.
Measuring state labor regulation
135
There was clearly a national trend toward increased labor regulation. Figure 7.4 shows national ESWIs for all industries and for manufacturing industries. We developed the national index by weighting each of the state ESWIs in that year by that state’s share of national employment in that year.19 The national index for all industries displays an 0.2 rise in the extent of regulation from 1900 through 1924 from a weighted index of 0.31 in 1899 to over 0.51 in 1924. There was a relatively sharp rise between 1899 and 1903 and then the trend was relatively flat with a dip in 1908. Between 1908 and 1914 there was a substantial rise in the index from about 0.35 to about 0.47 that coincided with the introduction of many employer liability and workers’ compensation laws. The trend stayed relatively flat through the end of World War I, followed by a rise to around 0.51 in the aftermath of the War. The ESWI for manufacturing followed a nearly parallel pattern throughout. The regulatory rise was a nation-wide phenomenon with the west and the northeast leading the way, as seen in Figure 7.5. Every state except New Jersey
ESW index (max = 1)
0.6 All regulations
0.5 0.4
Manufacturing regulations
0.3 0.2 0.1 0 1900
1905
1910
1915
1920
1925
Year
Figure 7.4 National labor regulation ESW-indexes, 1899–1924, all labor and manufacturing labor.
Legend Under 0.081282553 0.081282553 to 0.160375469 0.160375470 to 0.245034207 0.245034208 to 0.349425930 0.349425931 or more No data
Figure 7.5 Change in ESW index, 1919–1899.
136
R. Holmes et al.
and Florida increased the extent of its regulation; New Jersey had less room to move than many because it was a leader in 1899. The ten largest increases in the ESWI (above 0.23) occurred in the western states of Nevada, Oklahoma, Oregon, Arizona, and Washington, and the northeastern states of New York, Massachusetts, and New Hampshire, with Indiana and Maryland rounding out the group. The second ten was largely composed of western and midwestern states. Most of the southern states ranked in the third and fourth groupings of ten with increases in their indexes between 0.07 and 0.21. Virtually none of the rise in the national regulatory index is attributable to shifts in employment between states. When we recalculate the national ESWI using the 1900 state total employment weights for each year, the recalculated ESWI rises by the same amount as the index seen in Figure 7.4 does.
Labor productivity and labor regulation Even as the extent of labor regulation increased during the Progressive Era, labor regulations were often controversial issues. Opponents of regulations typically argued that they would diminish the competitiveness of industry by restricting production in ways that raised costs and/or lowered productivity. Some regulation supporters were willing to cede the point but argued that the costs were exceeded by the benefits to workers associated with improved safety, lower hours, higher wages, and better working conditions in general. Others argued that any losses in productivity associated with the restrictions would have been partially offset or more than fully offset by reductions in turnover by workers in response to better working conditions. Reduced turnover cut training costs and meant that the workers would tend to be more experienced and thus more productive. Despite their extensive differences over the proper means for improving working conditions, both union leaders and the large-scale employers who became known as “welfare capitalists,” made these arguments. The welfare capitalists improved their own working conditions in part to cut turnover and raise productivity, as well as to prevent the organization of unions in their workplaces. The Carnegies, Rockefellers, and other members of the National Association of Manufacturers were not interested in having unions or governments tell them how they should operate their businesses. Yet, once the leading employers had established a practice, they were willing to have government tell everybody else to follow suit. A number of progressive pieces of legislation ultimately succeeded when the welfare capitalists joined the reformers in passing regulations with which both could live. The welfare capitalists benefited because the field of competition was “leveled” at the spot where they were comfortable, which really meant it was likely tilted in their favor. This benefit was weakened to the extent that their product competitors were located in other states, but the National Association of Manufacturers and other groups of large firms often waged their campaigns in multiple states simultaneously. The reformers entered these compromises with their eyes wide open. The new rules often fell short of their original goals but were typically improvements over the prior status quo.20
Measuring state labor regulation
137
As a result, there is no single prediction of the relationship between state labor regulations and labor productivity. Here is a case where empirical work is valuable because it helps sort out the different stories told today and in the past about labor regulations. If there is a negative relationship, the critics of the regulations were right. To do a full cost–benefit analysis we need also to examine the benefits to workers in terms of wages, safety, hours, and employment. If there is a positive relationship or no relationship, it lends credence to the supporters of regulation, with one caveat. If more productive states were the ones that chose regulation, then a positive or zero coefficient might still have been consistent with a negative impact of regulation. The estimated effect of regulation on productivity will be biased in a positive direction unless we can control for this possible endogeneity. To examine the relationship between labor productivity and labor regulation, we develop a panel data set for the 48 continental states and territories for the years 1899, 1904, 1909, 1914, and 1919. The best productivity information we have available is in manufacturing in factories at the state level in the form of manufacturing value added per wage earner in 1967 dollars. All manufacturing information for those years comes from the Manufacturing Censuses conducted every five years during that time.21 Although the estimates we have described in the rest of the chapter were derived for regulation in all industries, we developed a new set of estimates, where feasible, that focused specifically on manufacturing. The Law Sums and ESW Law Index were confined to laws that affected all workers in manufacturing. Table 7.3 shows the raw cross-sectional correlations between the law measures and real value added per wage earners in 1899, 1909, and 1919. At the turn of the century there was virtually no correlation between the law indices and the labor productivity measure. However, by 1909 and again in 1919 a positive correlation appears. The correlations in Table 7.4 show that changes in the law indices and changes in the measures of productivity over time are not positively correlated across states. The strongest correlation is –0.21 for the changes between 1919 and 1899. Table 7.3 Cross-sectional correlations of manufacturing real value added per wage earner with measures of labor regulation for years 1919, 1909, and 1899
EWLI Sum Law Index
1919
1909
1899
0.3077 0.3943
0.3003 0.1703
0.0449 –0.0334
Sources: For sources and construction of labor regulation measures, see text. Notes Value added per wage earner was calculated for manufacturing in each state-year (value of production minus cost of materials) divided by average wage earners for 1899, 1904, and 1909 comes from United States Department of Commerce (1914: 208–213), for 1914 and 1919 from United States Department of Commerce (1924: 315–322. The measure is converted to 1967 dollars using the CPI, series E-135 from U.S. Bureau of Census (1975: 211).
138
R. Holmes et al.
To examine the relationships more carefully, we estimate a reduced-form relationship between the natural log of real value added per wage earner and the natural log of the employment weighted labor index. Table 7.5 shows the results for several specifications: Ordinary Least Squares (OLS) estimates of the simplest relationship, OLS estimates after controlling for other correlates that might influence productivity, and then estimates incorporating year and state fixed effects both with and without the correlates. We chose natural logs so that the coefficients would be in the form of elasticities. The OLS estimates without fixed effects show the basic relationships.22 As in the cross-sectional correlations in Table 7.3 for 1909 and 1919, the coefficient in the simplest relationship in column 1 is positive and statistically significant, suggesting an elasticity of labor productivity with respect to the law index of 0.1091. This positive relationship appears to have been driven by the omission of standard factors that are related to productivity. When the additional correlates are included in the analysis, the law elasticity falls sharply to a statistically insignificant 0.0267. Meanwhile, the capital per wage earner and union measures are both positively correlated with labor productivity. Since firms with more capital per worker and unionized states were more likely to press for legislation, it seems reasonable to conclude that their exclusion from the simplest equation contributed to biasing the law coefficient toward being positive in the simplest regression. Since larger and more productive employers tended to aid in the passage of labor regulations, some of the positive relationships we capture here may be a result of endogeneity in the political process. One method to control for the timeinvariant portion of this endogeneity is to include fixed effects. The inclusion of the fixed effects focuses the analysis on changes over time within the states in their labor regulations and in the labor productivity of their firms. The inclusion of fixed effects causes the elasticity of labor law index to switch signs and become negative, although the elasticity is very small and statistically insignificant.
Conclusions and plans for future research In seeking to measure state labor regulations during the Progressive Era, we have explored two simple aggregate measures of the overall intensity of regulation: raw sums of the number of laws and employment-share-weighted indices Table 7.4 Cross sectional correlations of changes in manufacturing real value added per wage earner with changes in measures of labor regulation for periods 1919–1909, 1919–1899, and 1909–1899
EWLI Sum Law Index Source: See Table 7.3.
1919–1909
1919–1899
1909–1899
0.0288 –0.0899
–0.068 –0.2112
–0.0835 –0.0342
8.5940 0.1091
Constant ln (employmentweighted law index) ln (capital per wage earner in 1967$) ln (horsepower per wage earner) Pct. workers employed in plants with less than 20 workers Pct. workers employed in plants with more than 500 workers Union index Year = 1904 Year = 1909 Year = 1914 Year = 1919 State Effects R2
153.63 2.69
t-statistic
1.20 –1.28 3.45
0.3195 –5.0840 0.0233
0.5959
–1.19
5.41
0.00005 –0.0625
33.94 0.61
t-statistic
7.6445 0.0267
Coefficient
–0.0021 0.0216 0.0828 0.0545 0.1757 included 0.8619
5.5575
–0.3582
–0.1334
0.00003
8.1852 –0.0444
Coefficient
–0.15 0.62 1.47 0.76 2.86
0.46
–0.45
–0.95
3.34
11.66 –1.30
t-statistic
Notes The estimation was performed on a panel data set composed of the continental 48 states (and territories) for the years 1899, 1904, 1914, 1919. The panel has a total of 240 observations. In a specification with the natural log of the law index and only the state and year fixed effects, the elasticity was –0.0343 with a t-statistic of –0.89.
Sources: For state labor regulation measures, see text. Information on manufacturing value added, value of production, cost of materials, wage earners, horsepower, and capital for each state for 1899, 1904, 1909 comes from United States Department of Commerce (1914): for 1914, 1919 from United States Department of Commerce (1924: 315–322). The census data we used was confined to “factory system” industries, and excludes the household, hand trades, and neighborhood industries. We also double checked the information with the manufacturing censuses for those years. All dollar values were converted to 1967 dollars using the CPI, series E-135 from U.S. Bureau of Census (1975: 211). The union index implicitly assumes that the national unionization rates for each industry in 1899, 1909, 1919, and 1929 were the same across states (see Fishback and Kantor (2000: 263) for a description of how it was constructed from data in Whaples (1990b: 434–47), Wolman (1936), Troy (1965), U.S. Bureau of Census (1902, volume 7; 1913, volume 9; 1923, volume 8; 1933, volume 3), and U.S. Bureau of the Census (1975: 137).
0.0543
Coefficient
Variable
Table 7.5 OLS and fixed effects estimates of coefficients in regressions of the natural log of manufacturing value added in 1967 dollars on the variables listed
140
R. Holmes et al.
that take more information into account. Both measures are highly correlated with each other in cross-sectional comparisons. Our ultimate goal is to use these measures to answer a range of questions about the impact of Progressive Era state regulations. There have been a number of studies that have focused on individual regulations, like workers’ compensation. Yet, none of these studies have tried to examine the effects of those laws while controlling for the variety of other laws that were introduced in the Progressive Era at the time. In our view it is important to assess not only the impact of specific laws but also the impact of the broad regulatory climate in which employers and workers operated in each state. These measures are not a substitute for studying the impact of specific laws. However, they can be used to get a sense of the amorphous concept of what the overall labor regulation climate was in the states. And they can be useful in settings where scholars studying the impact of specific laws seek to control for other aspects of the labor climate. As a first look at the impact of these laws we developed estimates of the reduced-form relationship between the various measures of regulation and labor productivity, measured simply as real value added per wage earner. Simple correlations of the levels of regulation and labor productivity across time and space suggest a positive relationship between regulation and labor productivity. This is consistent with the reformers’ contention that the regulations did not impose significant costs on employers through declines in productivity. It is also consistent with a view that states where manufacturing was more productive chose to adopt regulation. After controlling for unmeasured factors across states that did not vary across time with fixed effects, the results show that the elasticity of labor productivity with respect to labor regulation was very close to zero. To the extent that the fixed effects have eliminated endogeneity in the adoption of the labor regulations, the results might be taken as a positive sign for the reformers at the time who claimed that the labor regulations would not necessarily lower productivity. If the fixed effects do not fully control for endogeneity, the labor law coefficient is biased toward not finding a negative impact of regulation because high productivity states might have been more likely to adopt the labor regulation. We plan to explore this issue more fully in further papers, while also exploring other ways of measuring the labor legislation climate.
Notes * The authors would like to thank Robert Margo, Paul Rhode, Andy Daughety, Claudia Goldin, Lawrence Katz, Sumner LaCroix, Jeremy Atack, Shawn Kantor, Bill Collins, Ronald Oaxaca, Joshua Rosenbloom, Lou Cain, and Tom Weiss for helpful comments. 1 For example, see Moehling (1999), Sanderson (1974), Osterman (1979), and Brown et al. (1992) on child labor, Goldin (1990) and Whaples (1990a, 1990b) on women’s hours laws, Fishback and Kantor (2000), Buffum (1992), Chelius (1976, 1977), Fishback (1986, 1987, 1992), and Aldrich (1997) on workers’ compensation and
Measuring state labor regulation
2
3 4
5
6
7
8 9
10
141
employer liability laws, Fishback (1986, 1992) on coal mining regulations, Aldrich (1997) on safety regulations in manufacturing, mines, and railroads. For a summary of the research, see Fishback (1998). Child labor legislation had little impact on employment of children, but Margo and Finegan (1996) find that school attendance legislation did significantly raise the rate of school attendance. In Holmes, Fishback, and Allen (2007) we examine other measures, including the number of pages of text of the labor legislation and the funds per worker appropriated by state legislatures to administer, monitor, and enforce labor legislation. We also develop alternative multi-dimensional measures based on data mining that determine which states look most alike in the policies that they choose. Two methods are used: the principal component analysis that Holmes (2003, 2005) used in her dissertation on state labor laws and the Nominate analysis developed by Poole and Rosenthal to develop a nonparametric measure of locations in an undefined policy space. Each of the methods have pluses and minuses that we discuss more fully in that paper. For example, see Margo and Finegan (1996) and Moehling (1999) on child labor laws, Goldin (1990) and Whaples (1990a and 1990b) on womens’ hours laws. We have performed a variety of combinations of the raw sums, weighting by employment shares and indexing. As discussed below, the various combinations are highly correlated with each other, so we focus on the raw index and the employment-share weighted index as the two extremes of complexity. The correlations between the raw sum and the employment-share weighted sum was 0.93 in 1899, 0.90 in 1909, and 0.81 in 1919. The correlation of the changes in the raw sums from 1899 to 1919 and the employment-share weighted sums for the same period was 0.77, for 1899 to 1909 it was 0.79, and for 1909 to 1919 it was 0.80. Most states dropped some laws by repeal or because state courts declared them unconstitutional. Between 1899 and 1919 each state except Montana eliminated at least one law that was in place in 1899. The states that eliminated the most were New Jersey (18), Illinois (11), Missouri (9), Iowa (8), Ohio (8), Maryland (8), California (7), South Dakota (7), Nebraska (7), Pennsylvania (7), and Connecticut (6). The laws most commonly eliminated by states were restatements of the common law for employer liability (6), boiler inspection (5), railroad inspectors (4), deserting ship illegal (10), hours for public employment (4), hours for other work (4), general hours laws (4), hours for public roads (2), fines for enticement of employees (17), prevention of interference with railroad employees (5), prevention of interference with street railroad employees (3), labor agreements not defined as conspiracies (4), coercing employees to trade with stores illegally (3), company store anti-price gouging (3), licenses for railroad telegraph operators (5), licenses for steam railroad engineers (4), and requirements of cash wage payments (5). In many of the cases the laws were rendered moot by later legislation. See Kaufman (1993) and Moss (1996) for discussions of the roles played by John R. Commons, Richard Ely, Edwin Witte, and a large number of institutional economists in developing the field of industrial relations and in pressing for state labor legislation. The number of laws in the category is fixed over time and thus nj does not have a time subscript. Other sub-categories where the index reached a maximum below 1 in 1919 were railroad safety laws (0.8), miscellaneous laws (–0.167), child labor laws, men’s hours laws, pro-union laws, laws relating to bribery, voting, coercion, and licensing. The miscellaneous laws index was negative with a value of –0.167 in several southern states that had restrictions on emigrant agents that helped limit black out-migration. These states had no other miscellaneous laws with positive values that offset the emigrant agent law. The variable is expben10 in the wcdata.xls Excel spreadsheet under Workers’ Compensation Project Data at http://www.u.arizona.edu/~fishback/ used in March 2006.
142
11 12 13 14
15 16 17 18
19 20
21 22
R. Holmes et al.
Descriptions of the construction of the variable are in Fishback and Kantor (2000: Appendix B) and at the same website. The 1905 Lochner v. New York Supreme Court decision that declared unconstitutional limits on bakers’ hours in New York chilled efforts to establish wages and hours limits for male workers for some time. We also need to consider the issue of what weight to give to other hours laws for men in other industries after the Lochner decision. We plan to work on this issue further. For the licensing categories we used the shares of people in the profession from that national census. We explored use of two other denominators, the total number of workers in gainful employment in 1900 as the base and the number of workers gainfully employed in mining and manufacturing. These two measures were correlated with the employment base in the text at levels above 0.8 and we do not believe they will change the conclusions much. In constructing the indexes for the union relations category, worker’s compensation and employer liability category, and the general men’s hours law category, we followed the same procedure as for the Raw Index. The coverage of the general hours laws and industry-specific hours laws is extremely confusing and we are still working on better ways to understand situations where there appears to be overlap in coverage. We do all of the weighting by employment share within the subcategory indexes, so we do not need to weight again when summing across subcategories to develop the general index. The Raw Index and Raw Sums were highly correlated, the levels at 0.94 in 1919, 0.95 in 1909, and 0.96 in 1899. The correlations for changes were 0.84 for the change between 1919 and 1899, 0.84 for 1909–1899, and 0.88 for 1919–1909. The state employment share weighted index and state employment weighted raw sum were also highly correlated. The levels were correlated at 0.82 in 1919, 0.83 in 1909, and 0.92 in 1899. The correlations for changes were 0.71 for the change between 1919 and 1899, 0.67 for 1909–1899, and 0.81 for 1919–1909. The national employment share weighted index and employment share-weighted sum levels were correlated at 0.92 in 1919, 0.91 in 1909, and 0.94 in 1899. The correlations for changes were 0.80 for the change between 1919 and 1899, 0.75 for 1909–1899, and 0.81 for 1919–1909. The manufacturing law and overall law indexes with the same weighting schemes were also very strongly correlated. In terms of levels, the correlations were all above 0.91 in 1919, above 0.97 in 1909, and 0.96 in 1899. The correlations for changes were all above 0.92 for the change between 1919 and 1899, above 0.91 for 1909–1899, and 0.96 for 1919–1909 (U.S. Bureau of the Census, 1903, 1904, 1913, 1923, and 1933). The employment shares came from the occupational censuses of 1900, 1910, 1920, and 1930 with interpolations based on national employment totals between the census years. It should be noted that there were also times when employers actively sabotaged regulatory attempts where they opposed them. For extended discussions of the roles played by employers, unions, and reformers, and the ultimate impact of labor legislation, see Aldrich (1997), (Brandes (1976), Fishback (1998, 2006), Fishback et al. (2007), Fishback and Kantor (2000), Glaeser and Shleifer (2003), Goldin (1990), Lubove (1967), Moehling (1999), Moss (1996), Weinstein (1967, 1968), Wiebe (1962), and Whaples (1990b). Specific numbers are from U.S. Department of Commerce (1914, 1924) and were rechecked against U.S. Bureau of the Census (1902, 1917, 1923, and 1928). We perform this estimation for exploratory purposes. Holmes (2003, 2005) and Holmes and Fishback (2005) use the four factors from the Principal Component
Measuring state labor regulation
143
Analysis in a system of equations derived from a translog production function to estimate the impact of regulation on input demands and average costs. We then estimate reduced form equations for wages and employment to try to tease out the effects on labor supply.
References Aldrich, M. (1997) Safety First: Technology, Labor, and Business in the Building of American Work Safety, 1870–1939. Baltimore, MD: Johns Hopkins University Press. Brandeis, E. (1966 [1916–1935]) “Labor Legislation,” in John R. Commons and Associates, History of Labor in the United States, 1896–1932, volume 3, reprint. New York: Augustus Kelley Publishers. Brandes, S. (1976). American Welfare Capitalism, 1880–1940. Chicago, IL: University of Chicago Press. Brown, M., Christiansen, J., and Phillips, P. (1992) “The Decline of Child Labor in the U.S. Fruit and Vegetable Canning Industry: Law or Economics?,” Business History Review 66 (Winter): 723–770. Buffum, D. (1992) “Workmen’s Compensation: Passage and Impact.” Ph.D. dissertation, University of Pennsylvania. Chelius, J. R. (1976) “Liability for Industrial Accidents: A Comparison of Negligence and Strict Liability Systems,” Journal of Legal Studies 5 (June): 293–309. —— (1977) Workplace Safety and Health: The Role of Workers’ Compensation. Washington, DC: American Enterprise Institute. Commons, J. R. and Associates. (1966) History of Labor in the United States, 4 volumes, reprint of material published between 1916 and 1935. New York: Augustus Kelley Publishers. Fishback, P. V. (1986) “Workplace Safety During the Progressive Era: Fatal Accidents in Bituminous Coal Mining, 1912–1923,” Explorations in Economic History 23 (July): 269–298. —— (1987) “Liability Rules and Accident Prevention in the Workplace: Empirical Evidence from the Early Twentieth Century,” Journal of Legal Studies 16 (June): 305–328. —— (1992) Soft Coal, Hard Choices: The Economic Welfare of Bituminous Coal Miners, 1890 to 1930. New York: Oxford University Press. —— (1998) “Operations of ‘Unfettered’ Labor Markets: Exit and Voice in American Labor Markets at the Turn of the Century,” Journal of Economic Literature 36 (June): 722–765. —— (2006) “The Irony of Reform: Did Large Employers Subvert Workplace Safety Reform, 1869 to 1930,” Corruption and Reform: Lessons from American Economic History. Chicago, IL: University of Chicago Press. Fishback, P. V. and Kantor, S. E. (2000) A Prelude to the Welfare State: The Origins of Workers’ Compensation. Chicago, IL: University of Chicago Press. Fishback, P., Higgs, R., Libecap, G., Wallis, J., Engerman, S., Hummel, J. R., LaCroix, S., Margo, R., Sylla, R., Alston, L., Ferrie, J., Guglielmo, M., Pasour, Jr., E. C., Rucker, R., and Troesken, W. (2007) Government and the American Economy: A New History. Chicago, IL: University of Chicago Press. Glaeser, E. and Shleifer, A. (2003) “The Rise of the Regulatory State,” Journal of Economic Literature 41 (June): 401–425. Goldin, C. (1990) Understanding the Gender Gap: An Economic History of Women. New York: Oxford University Press.
144
R. Holmes et al.
Graebner, W. (1976) Coal-Mining Safety in the Progressive Era. Lexington, KY: University of Kentucky Press. Haines, M. (2006) Historical, Demographic, Economic, and Social Data: The United States: 1790–2000. Inter-University Consortium for Political and Social Research. ICPSR 2896. Holmes, R. (2003) “The Impact of State Labor Regulations on Manufacturing Input Demand During the Progressive Era.” Unpublished Ph.D. dissertation, University of Arizona, 2003. —— (2005) “The Impact of State Labor Regulations on Manufacturing Input Demand During the Progressive Era, Dissertation Summary,” Journal of Economic History 65 (June); 531–532. Holmes, R. and Fishback, P. (2005) “The Impact of State Labor Regulations on Manufacturing Input Demand During the Progressive Era.” Unpublished working paper, University of Arizona. Holmes, R., Fishback, P., and Allen, S. (2007) “Lifting the Curse of Dimensionality: Measuring State Labor Legislation During the Progressive Era.” NBER working paper. Kaufman, B. (1993) The Origins and Evolution of the Field of Industrial Relations in the United States. Cornell Studies in Industrial and Labor Relations Number 25. Ithaca, NY: ILR Press. Lubove, R. (1967) “Workers’ Compensation and the Prerogatives of Voluntarism,” Labor History 8 (Fall): 227–254. Margo, R. and Finegan, A. (1996) “Compulsory Schooling Legislation and School Attendance in Turn of the Century America: A ‘Natural Experiment’ Approach,” Economic Letters 53 (October): 103–110. Moehling, C. (1999) “State Child Labor Laws and the Decline of Child Labor,” Explorations in Economic History 36 (January): 72–106. Moss, D. (1996) Socializing Security: Progressive-Era Economists and the Origins of American Social Policy. Cambridge, MA: Harvard University Press. Mulligan, C. and Shleifer, A. (2004) “Population and Regulation.” NBER working paper no. 10234. Cambridge, MA: National Bureau of Economic Research, January. Osterman, P. (1979) “Education and Labor Markets at the Turn of the Century,” Politics and Society 9(1) (March): 103–122. Sanderson, A. (1974) “Child-Labor Legislation and the Labor Force Participation of Children,” Journal of Economic History 34 (March): 297–299. Troy, L. (1965) Trade Union Membership, 1897–1962, National Bureau of Economic Research Occasional Paper No. 92. New York: Columbia University Press. U.S. Bureau of Labor Statistics (1914) “Labor Laws of the United States, with Decisions of Courts Relating Thereto,” Bulletin of the United States Bureau of Labor Statistics No. 148, 2 parts. Washington, DC: Government Printing Office. —— (1925) “Labor Laws of the United States, with Decisions of Courts Relating Thereto,” Bulletin of the United States Bureau of Labor Statistics No. 370. Washington, DC: Government Printing Office. U.S. Bureau of the Census (1902) Twelfth Census of the United States: 1900, Manufactures, volume 7. Washington, DC: Government Printing Office. —— (1903) Twelfth Census of the United States, 1900, Population Part 1, volume 1. Washington, DC: Government Printing Office. —— (1904) Twelfth Census of the United States, 1900, Special Report – Occupations. Washington, DC: Government Printing Office. —— (1913) Thirteenth Census of the United States: 1910, Population, volume 1;
Measuring state labor regulation
145
Occupations, volume 4, Manufactures, volumes 8 and 9. Washington, DC: Government Printing Office. —— (1917) Abstract of the Census of Manufactures, 1914. Washington, DC: Government Printing Office. —— (1923) Fourteenth Census of the United States: 1920, Population, volumes 1 and 4; Manufactures, volume 8. Washington, DC: Government Printing Office. —— (1928) Biennial Census of Manufactures, 1925. Washington, DC: Government Printing Office. —— (1933) Fifteenth Census of the United States: 1930, Manufactures, volumes 1 and 3; Population, volumes 2 and 5. Washington, DC: Government Printing Office. —— (1975) Historical Statistics of the United States, Colonial Times to 1970. Washington, DC: Government Printing Office. U.S. Commissioner of Labor (1986) “Labor Laws of the United States,” Second Special Report of the Commissioner of Labor. Washington, DC: Government Printing Office. —— (1904). “Labor Laws of the United States with Decisions of Courts Relating Thereto,” Tenth Special Report of the Commissioner of Labor. Washington, DC: Government Printing Office. —— (1908) “Labor Laws of the United States with Decisions of Courts Relating Thereto, 1907,” Twenty-Second Annual Report of the Commissioner of Labor. Washington, DC: Government Printing Office. U.S. Department of Commerce (1914) Statistical Abstract of the United States, 1913, Thirty-Six Number. Washington, DC: Government Printing Office. —— (1924) Statistical Abstract of the United States, 1923, Fourty-Sixth Number. Washington, DC: Government Printing Office. U.S. Department of Labor (1962) Growth of Labor Law in the United States. Washington, DC: Government Printing Office. U.S. Supreme Court (1905) Lochner v. New York. 198 U.S. 45. Downloaded from http://straylight.law.cornell.edu/supct/cases/name.htm, on August 18, 2005. Weinstein, J. (1967) “Big Business and the Origins of Workmens’ Compensation,” Labor History 8 (Spring): 156–174. —— (1968). The Corporate Ideal in the Liberal State: 1900–1918. Boston: Beacon Press. Whaples, R. (1990a) “Winning the Eight-Hour Day, 1909–1919,” Journal of Economic History 50 (June): 393–406. —— (1990b) “The Shortening of the American Work Week: An Economic and Historical Analysis of Its Context, Causes, and Consequences.” PhD. dissertation, University of Pennsylvania. Wiebe, R. (1962) Businessmen and Reform: A Study of the Progressive Movement. Cambridge, MA: Harvard University Press. Witte, E. (1932) The Government in Labor Disputes. New York: McGraw-Hill. Wolman, L. (1936) The Ebb and Flow of Trade Unionism. New York: National Bureau of Economic Research.
8
Reexamining the distribution of wealth in 1870 Joshua L. Rosenbloom and Gregory W. Stutes*
The marked rise in income inequality in the United States over the past two decades has prompted a renewed interest in the history of both income and wealth distribution. Several recent studies have sought to construct consistent measures of inequality across most of the twentieth century.1 Evidence about either income or wealth distribution before the twentieth century is quite limited, but it is important to be able to place twentieth century trends in a broader context. The federal censuses of 1850, 1860 and 1870 offer a rare glimpse of patterns of property ownership in the United States during the nineteenth century. In 1850 census enumerators gathered information on the value of real property and in 1860 and 1870 they collected data on the value of both real and personal property holdings of every individual. These mid-century data offer a snapshot of wealth holding prior to the late nineteenth-century acceleration of industrialization. In this chapter we make use of data from the 1870 census contained in the Integrated Public Use Microdata Series (IPUMS) to examine the distribution of wealth at a relatively disaggregated level. A number of previous studies have used these mid-nineteenth-century census wealth data to explore a variety of issues related to wealth accumulation and inequality in the nineteenth century. But these earlier efforts have been based on relatively small samples or focused on particular sub-groups within the population.2 The large size of the IPUMS sample allows us to extend our understanding of mid-nineteenth-century wealth holding by exploring in much greater detail differences in the level of wealth holding and inequality both geographically and within a variety of population sub-groups. Based on a much smaller national sample drawn from census manuscripts Soltow (1975) found that wealth was much more unequally distributed in the South than elsewhere. We too find a pronounced North–South difference in the distribution of wealth, but by disaggregating within these regions we find that property was nearly as unequally distributed in some parts of the Northeast, and in the Pacific and Mountain regions as it was in the South. Decomposing wealth holding by race, residence, occupation, nativity and age, we find that inequality was higher in urban than rural areas, higher among Blacks than Whites, and varied with occupation and age. In light of the property requirements for entry into farming, it is not surprising that wealth was relatively equally distributed
Reexamining distribution of wealth in 1870
147
within this occupation group, but we also find relatively low levels of inequality among professionals, and clerical and kindred workers, while those in sales occupations displayed the highest level of inequality. Breaking the data down by age we show, consistent with Atack and Bateman’s (1981) results for rural households, that inequality was highest among the young, and declined for successively older groups. In contrast to these between group differences, however, we find that there was little difference in inequality between the native born and the foreign born in 1870. Beginning with Kuznets (1955) economic historians have been intrigued by the relationship between inequality and economic development. In his seminal article Kuznets conjectured that income inequality would likely follow an inverted U-shaped path. In support of this hypothesis he noted that inequality was higher in the urban and industrial sectors of the economy than in the rural and agricultural sectors, and showed that given this differential inequality the movement of population from the agricultural to the industrial sector would be (other things being equal) expected to cause income inequality to increase during the early stages of industrialization. Williamson and Lindert (1980) have argued that movements of skilled/unskilled pay ratios – which they interpret as a proxy for income inequality – in nineteenth-century United States were consistent with this conjecture. More recently Steckel and Moehling (2001) have used state property tax data to show that in Massachusetts the distribution of wealth became increasingly unequal from 1800 to the early twentieth century. Like these earlier studies we find support for the view that the early phases of U.S. industrialization were associated with rising inequality. Constructing a measure of wealth inequality comparable with that employed by Kopczuk and Saez (2004), we find that aggregate wealth inequality in 1870 was substantially lower than it would be half a century later. At the same time, by exploiting cross-state variation in the IPUMS data we show that in 1870 wealth inequality varied systematically with the level of urbanization and industrialization.
Characteristics of the data The 1870 census IPUMS contains a 1 percent random sample of the population drawn from the original census manuscripts. In total there are data for 383,308 individuals, with a combined aggregate wealth of $250.7 million. Many of these individuals were part of larger households, whose assets were likely to be reported as belonging to the head of the household. Analyzing wealth distribution across individuals thus may produce misleading results about the concentration of property ownership. Therefore, in the subsequent analysis we focus on wealth holding of household heads.3 Household heads accounted for 75,567 observations or about 20 percent of the individuals in the IPUMS sample, but held close to 90 percent of the reported wealth in 1870. The information on the value of real and personal property collected by Census enumerators was self reported, and the instructions to enumerators acknowledged that “exact accuracy may not be arrived at, but all persons should
148
J.L. Rosenbloom and G.W. Stutes
be encouraged to give a near and prompt estimate for your information” (quoted in Soltow 1975: 1). In 1870 enumerators were instructed to record information on personal property only if its aggregate value was $100 or greater. As a result there is some understatement of property ownership among the poorer segments of the population. In 1860, however, no such limitation was imposed and information from this year can be used to draw inferences about the extent of censoring in the 1870 data. In 1860, approximately one-third of household heads with personal property valued at less than $100 reported non-zero values of personal property. Given the small amounts involved, however, the impact of this truncation in personal wealth is likely to be small.4 Because the data on the value of real and personal property were selfreported the resulting figures are unlikely to be entirely accurate, but previous researchers have concluded that the discrepancies do not create large systematic biases. Analysis of the distribution of reported values clearly reveals a tendency toward heaping on round numbers. Matching census manuscripts with tax lists, Steckel (1994) found that census wealth figures often exceeded taxable wealth levels, but that there was no systematic association between such discrepancies and socioeconomic variables such as age or occupation. He also reported that differences in the Gini coefficients computed from the two sources were small and not statistically significant. The first column of Table 8.1 summarizes a number of the personal characteristics of the full IPUMS population sample, while the next three columns provide comparable information for all household heads, and for male and female household heads separately. Compared to the general population household heads were considerably older, more likely to be foreign born and to be employed in manufacturing. As previously noted, their average wealth level was substantially higher than the population as a whole, and they were much more likely to own any property. On the other hand, regional and urban–rural distributions were quite similar for the population as a whole and the household heads. The racial breakdown of the two groups was also quite similar. Reflecting typical gender roles of the time there were relatively few female headed households. Only about 11 percent of household heads were female in 1870, and it is likely that in most cases these women were recorded as heads because they had been widowed. The average female head was nearly five years older than her male counterpart, and almost twice as likely to be Black. She was also more likely to be native-born and to reside on a farm. Given the adverse events that were likely to have preceded their ascendance to the role of household head and their limited economic prospects it is not surprising that female household heads reported owning substantially less property on average and were more likely to report owning no real or personal property.
An overview of wealth holding and inequality in 1870 In 1870 there were pronounced differences across states and regions in both average wealth levels and in the distribution of wealth. The large size of the
Reexamining distribution of wealth in 1870
149
Table 8.1 Summary statistics, 1870 IPUMS and selected sub-samples Household heads
Number of Observations Personal Characteristics Age Female Black Employed in manufacturing Living on farm In city with population ≥100,000 In city with 25,000 ≤population <100,000 Foreign born Has disability Is literate Property Ownership Value of real property Value of personal property Value of total property Has any property Geography New England Mid Atlantic East North Central West North Central South Atlantic East South Central West South Central Mountain Pacific
All
Male
Female
383,308
75,567
66,825
8,742
23.5 0.496 0.126 0.073 0.586 0.105 0.044 0.144 0.001 0.578
42.3 0.116 0.126 0.199 0.620 0.106 0.044 0.254 0.001 0.791
41.8 0.000 0.117 0.219 0.584 0.102 0.043 0.261 0.001 0.810
46.7 1.000 0.193 0.047 0.889 0.136 0.054 0.199 0.001 0.648
$444 $210 $654 0.156
$2,038 $920 $2,958 0.689
$2,141 $966 $3,107 0.714
$1,251 $565 $1,816 0.505
0.089 0.225 0.239 0.100 0.152 0.116 0.053 0.008 0.017
0.096 0.230 0.234 0.096 0.150 0.111 0.054 0.010 0.020
0.095 0.231 0.242 0.101 0.142 0.105 0.053 0.010 0.021
0.100 0.223 0.173 0.057 0.209 0.153 0.064 0.009 0.012
Source: Ruggles et al. (2004).
IPUMS sample makes it possible to characterize these differences much more clearly than has heretofore been possible. We summarize these patterns at the state and regional level in Table 8.2. The table shows separate tabulations for real property, personal property and total property (equal to the sum of real and personal property). For each category of wealth we report the average value held by all household heads, as well as two indicators of inequality: the share of wealth owned by the top 1 percent of wealth holders, and the share of the population reporting any property ownership. These statistics are reported for the country as a whole, for census divisions and for individual states within each census division. Regional differences in average wealth were quite substantial, ranging from a high of $4,935 in the Pacific to just $957 in the Mountain states. Excluding these two recently settled areas there was a clear North–South gap in wealth levels: the average value of property holding in the northern states was about two to
USA New England Connecticut Maine Massachusetts New Hampshire Rhode Island Vermont Mid Atlantic New Jersey New York Pennsylvania East North Central Illinois Indiana Michigan Ohio Wisconsin West North Central Iowa Kansas Minnesota Missouri Nebraska South Dakota South Atlantic
75,567 7,225 1,092 1,242 3,017 734 435 705 17,351 1,829 8,847 6,675 17,702 4,923 3,233 2,360 5,198 1,988 7,226 2,211 752 858 3,122 234 47 11,351
Number of observations
2,038 2,207 3,138 1,341 2,161 1,963 2,688 2,440 2,740 2,876 2,857 2,547 2,693 2,990 2,408 2,497 2,931 2,031 2,123 2,476 1,484 1,791 2,157 1,926 638 972
0.268 0.268 0.267 0.132 0.337 0.156 0.517 0.134 0.271 0.220 0.294 0.241 0.220 0.280 0.163 0.177 0.197 0.133 0.248 0.145 0.117 0.210 0.344 0.273 0.267 0.364
0.483 0.539 0.536 0.747 0.418 0.659 0.400 0.657 0.467 0.439 0.458 0.487 0.622 0.574 0.609 0.699 0.595 0.745 0.610 0.686 0.585 0.717 0.526 0.697 0.596 0.325
920 1,651 3,068 753 1,694 1,283 1,701 1,207 1,230 1,045 1,298 1,191 918 1,079 815 866 919 743 872 993 647 710 906 721 266 417
0.383 0.497 0.581 0.270 0.457 0.308 0.478 0.169 0.402 0.294 0.418 0.371 0.339 0.420 0.288 0.278 0.293 0.252 0.255 0.184 0.168 0.260 0.332 0.244 0.200 0.455
Share of top 1%
Average
Any property
Average ($)
Share of top 1%
Personal property
Real property
0.628 0.614 0.616 0.746 0.508 0.726 0.563 0.745 0.643 0.648 0.583 0.720 0.743 0.740 0.760 0.736 0.734 0.751 0.792 0.826 0.771 0.780 0.781 0.799 0.511 0.442
Any property 2,958 3,858 6,205 2,093 3,855 3,246 4,389 3,647 3,970 3,921 4,156 3,738 3,610 4,068 3,223 3,363 3,850 2,774 2,995 3,469 2,131 2,501 3,063 2,646 905 1,388
Average ($) 0.279 0.327 0.406 0.155 0.346 0.200 0.474 0.128 0.263 0.223 0.288 0.242 0.217 0.291 0.178 0.188 0.204 0.144 0.229 0.139 0.113 0.206 0.317 0.253 0.106 0.354
Share of top 1%
Total property
0.690 0.696 0.701 0.831 0.599 0.779 0.667 0.803 0.705 0.700 0.663 0.764 0.815 0.802 0.825 0.824 0.805 0.849 0.840 0.879 0.828 0.840 0.815 0.880 0.702 0.497
Any property
Table 8.2 Average value of property owned, share of property owned by top 1% of wealth holders, and share owning any wealth, by state and region, 1870
269 388 2,334 1,387 2,050 1,547 2,347 789 240 8,375 2,040 2,393 1,702 2,240 4,076 958 1,582 1,536 761 23 94 39 66 199 186 134 20 1,500 1,264 187 49
2,161 335 536 1,760 455 583 1,066 1,654 4,098 976 400 1,722 541 1,035 769 638 881 734 462 378 1,188 1,047 87 186 440 488 13 3231.13 3,568 1,570 886
0.370 0.258 0.338 0.258 0.292 0.551 0.323 0.277 0.341 0.338 0.370 0.295 0.404 0.296 0.475 0.406 0.635 0.270 0.323 0.460 0.358 0.490 0.350 0.230 0.159 0.459 1.000 0.505 0.481 0.131 0.184
0.242 0.271 0.293 0.334 0.384 0.223 0.310 0.504 0.458 0.362 0.279 0.486 0.250 0.388 0.318 0.400 0.198 0.391 0.449 0.435 0.404 0.385 0.121 0.492 0.683 0.336 0.050 0.459 0.426 0.663 0.510
749 342 295 771 229 321 298 790 1,460 531 265 866 371 536 385 436 288 453 496 424 533 1,117 1,040 228 338 717 35 1,705 1,813 1,120 1,149
0.476 0.460 0.271 0.368 0.259 0.628 0.324 0.497 0.362 0.340 0.315 0.382 0.361 0.303 0.322 0.411 0.397 0.219 0.313 0.513 0.119 0.321 0.175 0.599 0.239 0.375 0.571 0.393 0.412 0.182 0.213
0.461 0.376 0.452 0.500 0.429 0.301 0.399 0.705 0.654 0.552 0.427 0.679 0.429 0.625 0.504 0.624 0.346 0.593 0.432 0.522 0.479 0.436 0.455 0.241 0.629 0.433 0.100 0.656 0.633 0.813 0.653
2,910 677 831 2,531 684 903 1,364 2,444 5,559 1,507 665 2,588 912 1,572 1,154 1,074 1,170 1,187 957 802 1,721 2,164 1,126 414 778 1,204 48 4,936 5,381 2,690 2,035
0.397 0.295 0.284 0.256 0.235 0.562 0.300 0.324 0.349 0.312 0.304 0.287 0.331 0.249 0.367 0.392 0.517 0.199 0.274 0.379 0.278 0.379 0.169 0.419 0.138 0.341 0.684 0.385 0.400 0.084 0.201
Note For personal property the share holding any property reflects the fraction of responses indicating ownership of 100 or more worth of personal property.
Source: Ruggles et al. (2003).
District of Columbia Florida Georgia Maryland North Carolina South Carolina Virginia West Virginia Delaware East South Central Alabama Kentucky Mississippi Tennessee West South Central Arkansas Louisiana Texas Mountain Arizona Colorado Idaho Montana New Mexico Utah Nevada Wyoming Pacific California Oregon Washington
0.487 0.443 0.490 0.553 0.512 0.346 0.448 0.782 0.733 0.593 0.471 0.730 0.456 0.663 0.550 0.664 0.399 0.633 0.556 0.522 0.596 0.513 0.455 0.508 0.720 0.507 0.100 0.705 0.681 0.856 0.735
152
J.L. Rosenbloom and G.W. Stutes
three times as high as in the South.5 But broad regional averages mask significant within region variations. Within the South, wealth levels were generally higher in border states – Maryland, West Virginia and Kentucky, than in the deep South. In the North, there was an especially wide dispersion in average wealth holding across the New England states with the industrialized states of southern New England – Connecticut, Rhode Island and Massachusetts – having much higher levels of wealth holding than the more rural northern states – Vermont, New Hampshire and Maine. The same regional patterns are also apparent when real and personal property ownership are considered separately. But it is interesting to note that in New England real property accounted for an unusually small share of total wealth, while personal property holding was correspondingly more important. In New England personal property accounted for almost 43 percent of total wealth while it amounted to just 30 to 35 percent of wealth in most other regions. Turning to the distribution of wealth, the two summary measures we report show a high degree of correlation; the greater the concentration at the top the lower the fraction of the population reporting any property ownership. There is again a pronounced North–South division, with wealth holding more highly concentrated in the South than in the North. In the North 70 to 80 percent of household heads reported positive levels of property ownership. In comparison, in the South the share reporting any property ranged from a high of 59 percent in the East South Central region to a low of 50 percent in the South Atlantic. The data on wealth concentration among the top 1 percent of wealth holders also suggest that inequality was greater in the South than in the North. Across the southern census divisions the share of wealth owned by the top 1 percent of wealth holders ranged from 31.5 to 36.7 percent; and in some southern states the figure was dramatically higher, reaching more than 50 percent in South Carolina and Louisiana. In the Mid Atlantic and North Central census divisions the share of wealth owned by the top 1 percent was substantially lower, varying from 21.7 to 26.3 percent. Departing from this regional pattern the New England states displayed a much greater level of inequality, with shares owned by the top 1 percent in Massachusetts (35 percent), Connecticut (41 percent) and Rhode Island (47 percent) comparable to those in many southern states. Data on the share of wealth held by the top 1 percent in 1870 can be compared directly to data gathered by Kopczuk and Saez (2004). In 1916 – the first year covered by their data – they found that the top 1 percent of households held close to 40 percent of total wealth. According to Kopczuk and Saez the share of wealth held by this wealthiest group fell sharply between 1930 and 1932, and continued to decline until by 1949 they held just 22.5 percent of the nation’s wealth. Despite some subsequent fluctuations in wealth inequality Kopczuk and Saez did not find any long-run trend in the share held by the top 1 percent since 1950. In comparison, at the national level in 1870 the top 1 percent of wealth holders accounted for just 27.9 percent of total wealth, a figure substantially lower than Kopczuk and Saez found early in the twentieth century. Indeed the figure for 1870 is close to the current level of wealth concentration. Thus it
Reexamining distribution of wealth in 1870
153
appears that inequality increased substantially over the course of the halfcentury between 1870 and 1916 as the nation industrialized
Determinants of individual wealth accumulation The state, regional and national data discussed so far reflect the aggregation of the experiences of thousands of individuals. Differences in wealth accumulation across these individuals reflect both systematic differences associated with observable characteristics, like age, occupation and race, and the influence of random shocks and unobservable differences. Because the IPUMS combines individual level data on wealth holding with a wide range of other individual characteristics we can examine the extent to which systematic differences in personal characteristics can account for differences in wealth accumulation. Since a large number of household heads in 1870 were recorded as possessing no property we proceed in two stages. In the first stage we use a probit regression to examine factors associated with whether a person reported owning any property. Here the dependent variable is equal to 1 if the individual was recorded as having any property (for personal property it is equal to 1 if they had more than $100 of property), and 0 otherwise. In the second stage we limit our analysis to individuals reporting positive amounts of property (more than $100 for personal property), and regress the log of the level of wealth on personal characteristics. Table 8.3 reports the results of the probit regressions converted to marginal probabilities, so that each coefficient shows how changes in the dependent variable affected the probability of reporting any wealth.6 Table 8.4 reports the results of Ordinary Least Squares (OLS) regressions of the log of wealth on individual characteristics for those household heads reporting positive (greater than $100 for personal property) levels of property ownership. Because several of the characteristics we consider, such as location and occupation, are endogenous and may be dependent on wealth our analysis should be interpreted as descriptive of patterns of correlation, rather than as a causal model of wealth accumulation. Although the coefficients vary somewhat depending on which category of property we are analyzing the results are generally quite similar across the different categories of property ownership. To simplify the discussion we will focus on describing the regressions for total property but will mention areas in which results differ substantially. Considering first the probability of property ownership, Blacks were about 32 percent less likely to own property than Whites. Women (11 percent less likely to own property), the foreign born (8 percent) and the disabled (20 percent) were also less likely to own property. On the other hand, literacy increased the probability of owning property by about 9 percent. The age coefficients indicate that wealth holding followed an inverted-U shape with age, rising until the late 1850s before beginning to decline. For real property the peak was reached a few years later, at age 62.
Personal Characteristics Black Female Foreign born Literate Disability Age Age squared Urbanizationa City 25–100 thousand City 100 thousand + Occupationb Professional Farmer Clerical Sales Managerial Operative Laborer Service Non-occupational Regionc Mid-Atlantic East North Central West North Central South Atlantic
0.0066 0.0108 0.0052 0.0063 0.0568 0.0009 0.0000
0.0091 0.0058
0.0113 0.0065 0.0228 0.0177 0.0094 0.0084 0.0071 0.0174 0.0122
0.0078 0.0078 0.0094 0.0087
–0.1334 –0.2796
0.2876 0.2682 –0.0241 –0.0416 0.1540 –0.0982 –0.2058 –0.0922 0.0140
–0.0046 0.0512 0.0133 –0.0990
Standard error
–0.3455 –0.0638 –0.0033 0.1297 –0.1374 0.0359 –0.0003
dF/dx
Probability real property > 0
0.558 0.000 0.155 0.000
0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.2510
0.000 0.000
0.000 0.000 0.522 0.000 0.024 0.000 0.000
P > |z|
■
0.0759 0.1061 0.1366 –0.0320
0.2183 0.2473 0.0175 0.0244 0.1920 –0.0588 –0.1508 –0.0756 –0.0260
–0.0811 –0.1141
–0.3211 –0.1208 –0.1170 0.1073 –0.2409 0.0218 –0.0002
dF/dx
0.0067 0.0066 0.0074 0.0084
0.0074 0.0056 0.0200 0.0156 0.0066 0.0082 0.0072 0.0156 0.0110
0.0095 0.0070
0.0080 0.0102 0.0050 0.0061 0.0626 0.0008 0.0000
Standard error
0.000 0.000 0.000 0.000
0.000 0.000 0.387 0.123 0.000 0.000 0.000 0.000 0.000
0.000 0.000
0.000 0.000 0.000 0.000 0.000 0.000 0.000
P > |z|
Probability personal property > 100
Table 8.3 Probit estimates of the determinants of property ownership, 1870 ■
0.0585 0.0973 0.1061 –0.0613
0.1786 0.2108 0.0028 0.0027 0.1555 –0.0693 –0.1552 –0.0759 –0.0340
–0.0766 –0.1576
–0.3261 –0.1080 –0.0833 0.0971 –0.2072 0.0218 –0.0002
dF/dx
0.0063 0.0061 0.0069 0.0081
0.0064 0.0053 0.0189 0.0149 0.0060 0.0079 0.0070 0.0148 0.0104
0.0091 0.0069
0.0082 0.0099 0.0048 0.0058 0.0624 0.0007 0.0000
Standard error
Probability total property > 0
0.000 0.000 0.000 0.000
0.000 0.000 0.884 0.858 0.000 0.000 0.000 0.000 0.001
0.000 0.000
0.000 0.000 0.000 0.000 0.000 0.000 0.000
P > |z|
–0.1108 –0.1169 –0.0211 –0.0219 0.4825 0.4498 2613
0.0091 0.0111 0.0204 0.0154
0.000 0.000 0.302 0.157
0.0415 0.0155 –0.1453 0.0820 0.6377 0.6660 0.2319
0.0085 0.0104 0.0205 0.0128
0.000 0.452 0.000 0.000
Notes a Excluded category is places with population less than 25,000. b Excluded category is craft workers. c Excluded region in New England. Coefficients are from transformed probits and show the change in probability of a change in the independent variable.
Source: Ruggles et al. (2004).
East South Central West South Central Mountain Pacific Obs. P Pred. P (at x-bar) Pseudo R2
–0.0024 –0.0213 –0.1011 0.0485 0.6895 0.7330 0.2554
0.0083 0.0101 0.0194 0.0122
0.769 0.032 0.000 0.000
156
J.L. Rosenbloom and G.W. Stutes
The two measures of urban location – one for residents of cities between 25,000 and 99,999 population; and the other for residents of cities with populations of 100,000 or more – indicate that the probability of owning property declined with city size. Residents of medium-sized cities were about 8 percent less likely to own property than were otherwise comparable individuals in smaller localities. For residents of the largest cities the probability of owning property was reduced by about 15 percent. Property ownership also depended on occupation, though it is likely that these effects may in part reflect entry requirements into certain occupations. Here we employ the IPUMS recoding of the original occupational responses based on the 1950 census occupational classification scheme. The excluded category in all of the regressions is skilled craft workers, so the coefficients reflect differences in wealth accumulation relative to this group. Laborers were the occupational group least likely to have accumulated any property, and the wealth of those who had property was lower than for any other group. Farmers, were the most likely to own property, being about 20 percent more likely to do so than the excluded category of skilled craft workers. They were followed closely by individuals in managerial and professional occupations. Clerical and sales workers were about as likely to own property as craft workers, while operatives and laborers were substantially less likely to own property. The final section of the table reports the effects of region of residence on the probability of owning property. These coefficients capture locational effect after controling for the other demographic and occupational characteristics included in the regression. Comparing these estimated regional effects with the unconditional shares reporting any wealth in Table 8.2 differences in population characteristics across regions can account for an appreciable fraction of the regional differences noted earlier. For example, Table 8.2 shows that residents of the South Atlantic region were 19.4 percent less likely (49.6 percent versus 69 percent) to own property than were residents of New England. In Table 8.3, the regional effects imply a difference of just 6 percent. Turning now to the results in Table 8.4, which show the relationship between personal characteristics and the value of property owned for those reporting positive wealth holding, the results parallel, in many ways, the impacts on the likelihood of property ownership. Blacks, women, the foreign born and the disabled were not only less likely to own property, but those who did own property owned less of it. Blacks, for example, owned about 50 percent less property than comparable non-Blacks.7 Literacy, which increased the probability of property ownership by about 9 percent was associated with an approximate doubling in the value of property owned. Looking at the relationship of property values with age, they again follow an inverted-U shape, with the peak occurring at age 61. While the ranking of occupations in Table 8.4 resembles that in Table 8.3, there are some notable differences. Although farmers were more likely to own property than individuals in professional and managerial occupations, the amounts they owned were on average substantially lower. Professional and managerial workers reported owning nearly four times as much property as craft
Reexamining distribution of wealth in 1870
157
workers, while farmers reported wealth was approximately double that of craft workers. While the direction of many of the effects are the same in Tables 8.3 and 8.4, one area where they are different is in the effect of urban residence. While residents of medium- and large-sized cities were less likely to own property than residents of smaller places, the amount that those who owned property reported was on average larger in bigger cities. For both real and personal property considered separately, the value of property owned increased with city size, with the effects being especially large for real property ownership. For total property ownership, changes in composition produced a slightly higher level of property ownership in medium-sized cities, but the average wealth of residents in the largest cities is no greater than for those living in places smaller than 25,000 population. Again, controlling for personal characteristics helps to explain some of the regional differences in average wealth holding reported in Table 8.2. This is especially true for the North–South differences, which are reduced though not eliminated; and for differences between the Pacific region and the Northeast, which is entirely eliminated.
The sources of inequality Despite the evident correlation of property ownership with a variety of personal characteristics, these observable factors can account for at best a small fraction of total inequality. The R2 values reported in Table 8.4 indicate that observable characteristics can explain only about 30 percent of the variance in the value of property reported. No matter how the population is divided the vast majority of variation in wealth levels occurred within groups rather than between them. In this section we formalize this observation making use of the Theil inequality index to partition inequality into components attributable to inequality within and between groups. The measures of inequality that we have considered so far reflect only one part of the wealth distribution, and do not capture variations across the entire population. To address this problem, scholars have developed a number of statistics intended to summarize the full range of income or wealth distribution. Prominent among these are the Gini and Theil indexes. Here we use the Theil index because it has the desirable characteristic that it can be linearly decomposed to express the relative contributions of inequality within and between different subgroups of the population being studied. For a selected population the Theil index is calculated as:
1 n wi wi T = ln n i=1
(8.1)
where n represents the number of observations, wi represents the wealth of individual i, µ represents mean wealth, and 0 ln (0) is defined to be equal to zero. In the case of perfect equality the index is equal to zero. When wealth is perfectly
Personal Characteristics Black Female Foreign born Literate Disability Age Age squared Urbanizationa City 25–100 thousand City 100 thousand + Occupationb Professional Farmer Clerical Sales Managerial Operative Laborer Service Non-occupational Regionc Mid-Atlantic East North Central West North Central South Atlantic 0.0452 0.0338 0.0144 0.0203 0.1932 0.0026 0.0000 0.0337 0.0288 0.0433 0.0183 0.0747 0.0631 0.0287 0.0282 0.0257 0.0722 0.0404 0.0212 0.0205 0.0244 0.0253
–0.4779 –0.5366 –0.1660 0.6809 –0.8604 0.0732 –0.0006
0.6846 0.9900
1.2589 0.6035 0.5027 0.5004 1.0193 –0.0689 –0.4422 0.1922 0.7321
0.3254 0.1403 –0.0798 –0.5663
0.000 0.000 0.001 0.000
0.000 0.000 0.000 0.000 0.000 0.014 0.000 0.008 0.014
0.000 0.000
0.000 0.000 0.000 0.000 0.000 0.000 0.000
0.0429 –0.1999 –0.1429 –0.5204
1.3570 0.6131 0.5046 0.5298 1.5474 0.0297 –0.3655 0.2157 0.9201
0.2100 0.5157
–0.4549 –0.7607 –0.3188 0.4846 –0.8103 0.0636 –0.0006
0.0244 0.0232 0.0256 0.0280
0.0466 0.0198 0.0986 0.0750 0.0344 0.0329 0.0289 0.0865 0.0483
0.0468 0.0402
0.0469 0.0448 0.0149 0.0182 0.1812 0.0028 0.0000
Standard error
Coefficient
P > |t|
Coefficient
Standard error
Personal property
Real property
Table 8.4 OLS estimates of the determinants of the value of property owned, 1870
0.079 0.000 0.000 0.000
0.000 0.000 0.000 0.000 0.000 0.366 0.000 0.013 0.000
0.000 0.000
0.000 0.000 0.000 0.000 0.000 0.000 0.000
P > |t|
0.0546 –0.0441 –0.2083 –0.6843
1.4809 0.8310 0.4693 0.4054 1.3661 –0.1364 –0.5179 –0.0044 0.8722
0.1724 0.0216
–0.6976 –0.6188 –0.1271 0.7028 –0.7486 0.1032 –0.0008
Coefficient
0.0209 0.0201 0.0232 0.0249
0.0402 0.0183 0.0725 0.0571 0.0288 0.0266 0.0224 0.0608 0.0419
0.0344 0.0273
0.0288 0.0384 0.0141 0.0178 0.1730 0.0025 0.0000
Standard error
Total property
0.009 0.028 0.000 0.000
0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.942 0.000
0.000 0.282
0.000 0.000 0.000 0.000 0.000 0.000 0.000
P > |t|
0.2845 36,462
Adj. R2 Number of observations
0.0267 0.0350 0.0611 0.0448 0.0690
0.000 0.000 0.000 0.225 0.000 0.2306 32,117
–0.2838 –0.4317 –0.2681 0.1431 4.0638
0.0283 0.0355 0.0738 0.0564 0.0715
0.000 0.000 0.000 0.011 0.000 0.3277 52,103
–0.6155 –0.7129 –0.6418 –0.0092 3.5891
0.0252 0.0313 0.0622 0.0480 0.0625
0.000 0.000 0.000 0.849 0.000
Notes a Excluded category is places with population less than 25,000. b Excluded category is craft workers. c Excluded region is New England. The dependent variable in each regression is the log of the value of property owned. Regressions estimated for those reporting positive property values (values greater than or equal to $100 for personal property). Standard errors are corrected for heteroscedasticity using the Huber-White method.
Source: Ruggles et al. (2004).
–0.5719 –0.7640 –1.1262 –0.0544 4.0618
East South Central West South Central Mountain Pacific Constant
160
J.L. Rosenbloom and G.W. Stutes
unequally distributed – which is the case if one individual owns all the wealth – the index equals ln(n). The Theil index can be decomposed for any exhaustive set of population subgroups into the contributions attributable to inequality within each subgroup and across subgroups. If there are G population subgroups and Tj denotes the Theil index calculated using equation (1) for individuals within subgroup j. Then aggregate inequality can be expressed as: G G njj njj i T = Tj + ln j=1 n j=1 n
(8.2)
where nj is the number of observations in subgroup j, and µj is the mean wealth of subgroup j. Notice that the first term in each summation is the same and is equal to subgroup j’s share of total wealth. Thus, the first term in the decomposition is a weighted sum of the within subgroup inequalities where the weights are subgroup shares of total wealth. This is the measure of within group inequality. The second term is a weighted sum of the log of the ratios of subgroup average wealth to the mean wealth of the entire population. This is the measure of between group inequality. Variations in the Theil index across states closely resemble the pattern of variation in the measure of inequality we considered in Table 8.2, the share of wealth owned by the top 1 percent of wealth holders. Figure 8.1 plots the Theil index for each state as a function of the corresponding share of total wealth owned by the top 1 percent. The fact that the two measures are not perfectly correlated reflects the additional information about other points in the wealth distribution that is captured by the Theil index but ignored when we look only at wealth holding of the very rich.
Theil index for total wealth
3.5000 3.0000
y = 4.0087x + 0.395 R2 = 0.8186
2.5000 2.0000 1.5000 1.0000 0.5000 0.0000 0.0000 0.1000 0.2000 0.3000 0.4000 0.5000 0.6000 0.7000 0.8000 Share of wealth owned by top 1%
Figure 8.1 Relationship between the share of wealth owned by the top 1 percent and the Theil index of inequality.
Reexamining distribution of wealth in 1870
161
To begin our examination of the sources of inequality we first look at inequality within a variety of population subgroups. We report in Table 8.5 Theil indexes calculated for subgroups of the population broken down by race, nativity, age, occupation, urban residence and region. These decompositions reveal a number of interesting features of wealth accumulation patterns. First, while real property ownership became increasingly equitable with age, personal property ownership became increasingly unequally distributed. When these patterns are combined there is relatively little relationship between age and inequality. Second, inequality was substantially greater among Blacks than among Whites. Third, on the other hand, there was little difference in inequality between the native and foreign born. Fourth, inequality was greater in larger cities – those over 25,000 population – than in smaller places. Fifth, there were marked differences in inequality across different occupation groups. As one might expect, farmers had the most equal distribution of property ownership. Interestingly, however, laborers were among the occupations with the most unequal distribution of property. Finally, regional patterns of inequality parallel those noted earlier – with real property inequality highest in the South, and personal property inequality highest in New England and the Mid Atlantic. For each of the decompositions reported in Table 8.5 it is possible to apportion the aggregate inequality in the total population using equation (8.2) into the contributions of inequality within each subgroup and inequality between the different subgroups. These respective contributions are reported in Table 8.6, which shows that almost all inequality arose within subgroups rather than between them. For the entire population the Theil index for total property inequality is 1.60. Taking, for example, the regional decomposition, aggregating the inequality within each of the census divisions accounts for 94.7 percent of this total, while the differences in average wealth across the census divisions contribute only 5.3 percent. It is apparent that almost all of the inequality occurred with groups rather than between them. In all but one case 90 percent or more of total inequality was attributable to within group variations in wealth holding. The sole exception is the decomposition by occupation groups, where between group inequality accounts for about 20 percent of total inequality.
The correlates of geographic variation in inequality One motivation for studying variations in wealth and income inequality is to gain a better understanding of the mechanisms that have contributed to historical variations in the level of inequality produced by the American economy. The presence of substantial cross-sectional variation in levels of wealth inequality in 1870 provides an opportunity to examine the relationship between inequality and the structural changes in the economy that were associated with the process of industrialization during the nineteenth century. Over the course of the nineteenth century the process of economic transformation that accompanied American industrialization proceeded at different rates in different parts of the country. Industrialization began much earlier, for
162
J.L. Rosenbloom and G.W. Stutes
Table 8.5 Within group inequality, selected population groups, 1870 Number of observations
By Age 0–19 534 20–29 13,854 30–39 20,616 40–49 18,115 50–59 12,699 60–69 9,749 By Race White 66,069 Black 9,498 By Occupation Miscellaneous 8,442 Professionals 1,838 Farmers 27,673 Managers 4,375 Clerical and Kindred 573 Salesmen and clerks 987 Craftsmen 9,216 Operatives 6,311 Service Workers 1,460 Laborers 14,692 By nativity Native 56,405 Foreign 19,162 By urbanization Less than 25,000 64,247 Cities 25,000–100,000 3,330 Cities larger than 100,000 7,990 By Region New England Division 7,225 Middle Atlantic Division 17,351 East North Central Division 17,702 West North Central Division 7,226 South Atlantic Division 11,351 East South Central Division 8,375 West South Central Division 4,076 Mountain Division 761 Pacific Division 1,500 Source: Ruggles et al. (2004). Notes See text for a Theil Index formula.
Within group Theil index Real property
Personal property
Total property
3.388 2.045 1.701 1.536 1.408 1.502
2.777 1.472 1.524 1.988 1.929 2.220
2.697 1.563 1.433 1.507 1.400 1.542
1.563 3.697
1.890 2.299
1.482 2.698
2.174 1.499 0.980 1.566 1.502 2.744 1.588 2.033 2.570 2.535
2.600 1.327 1.017 1.740 1.932 1.831 1.604 1.941 2.277 1.827
2.086 1.234 0.876 1.446 1.352 2.221 1.370 1.741 2.130 1.956
1.641 1.839
1.951 2.137
1.560 1.724
1.436 2.283 2.791
1.692 2.892 2.858
1.345 2.271 2.567
1.564 1.624 1.260 1.379 2.255 2.070 2.686 1.877 2.464
2.405 2.035 1.568 1.200 2.216 1.812 1.738 1.852 2.014
1.732 1.555 1.195 1.180 2.069 1.797 2.101 1.610 2.045
Reexamining distribution of wealth in 1870
163
Table 8.6 National inequality arising from within and between group inequality, for selected population subgroups, 1870 Real property
By age By race By occupation By nativity By urbanization By region
■
Personal property
■
Total property
Within Between Within group group group inequality inequality inequality
Between group inequality
Within group inequality
Between group inequality
1.554 1.572 1.414 1.684 1.678 1.598
1.896 1.893 1.566 1.985 1.972 1.903
0.103 0.105 0.432 0.013 0.027 0.095
1.477 1.488 1.289 1.594 1.586 1.516
0.123 0.112 0.311 0.006 0.014 0.084
94.8 94.7 78.4 99.3 98.7 95.2
5.2 5.3 21.6 0.7 1.3 4.8
92.3 93.0 80.6 99.6 99.1 94.7
7.7 7.0 19.4 0.4 0.9 5.3
0.133 0.115 0.273 0.003 0.009 0.089
As a percentage of total Inequality By age 92.1 7.9 By race 93.2 6.8 By occupation 83.8 16.2 By nativity 99.8 0.2 By urbanization 99.5 0.5 By region 94.7 5.3 Source: Ruggles et al. (2003). Notes See text for additional information.
example, in New England and the Mid Atlantic regions, than in the North Central and Southern regions. By 1870, close to 35 percent of the population in Massachusetts and New York lived in places with population of 25,000 or more, more than three times the national average of 11 percent. Similarly, while manufacturing accounted for only 7 percent of employment nationally, more than 20 percent of the population of Massachusetts and Rhode Island was employed in manufacturing. Industrialization and urbanization were also closely linked to high rates of immigration, although many of the foreign born could also be found in more agricultural regions. It is inappropriate of course to equate the results of such cross-section comparisons with genuine time-series observations. On the one hand it is possible that patterns of within group inequality changed over time. On the other hand, there have been interactions between states at a point in time – arising from interstate migration and trade – that caused cross-section and time series relationships to differ. Nonetheless, in the absence of time series data on inequality over the course of the century it is illuminating to explore the cross-section relationship. Using the full IPUMS population sample for 1870 we have constructed measures of a number of demographic characteristics for each state. These include: the average age of the population, the share of the population that was Black,
164
J.L. Rosenbloom and G.W. Stutes
foreign born, literate, living in a city with population 25,000 or greater or employed in manufacturing. Several of these characteristics are highly correlated with each other, and it does not make sense to include all of them in a regression model. After some experimentation we found that we could account for a large fraction of the across state variation in inequality with a small number of state characteristics. The top three panels of Table 8.7 report the results of several OLS regressions estimated across states where the dependent variable is the Theil inequality index calculated for, respectively, real, personal and total property. In these regressions we have dropped the four smallest states (those with less than 50 heads of household in the 1870 IPUMS sample) to reduce errors arising from very small sample sizes. The bottom panel of the table reports summary statistics for the dependent and independent variables in the regressions. State characteristics can account for close to two-thirds of the variation across states in real and total property variation, and about half of the variation in personal property inequality. Which specification fits best, and the relationship between inequality and the various explanatory variables differs depending on which type of wealth we are considering. Our first specification (Specification 1) includes the share of Blacks in the population (a proxy for the legacy of slavery), along with the share employed in manufacturing and the share living in large cities (those with populations of 25,000 or more), which can be interpreted as proxies for industrialization and urbanization, respectively. Urbanization and the fraction Black are consistently positive and statistically and economically significant, but the share employed in manufacturing is significant only in the regression for real property inequality.8 The effect of the fraction Black on inequality is not simply capturing North–South differences in inequality. When we replace the share of Blacks with a dummy variable for southern states that dummy variable is indeed positive and significant, but when we include both the dummy variable and the share of Blacks, the dummy variable loses its significance, indicating that the relationship between the share of Blacks and inequality is being identified largely on the basis of variations within the South. Adding the fraction of the population that is literate (Specification 2) substantially increases the explanatory power of our model, especially for the case of personal property wealth inequality. In addition the size and significance of the fraction Black declines, so that this variable is statistically significant in only one case – for real property inequality. There was a strong negative relationship between literacy and the fraction Black across states – the simple correlation coefficient between these two variables is –0.78 – but it is clear that the fraction literate is more closely related to inequality than the fraction Black. Adding literacy also increases the size and significance of the share in manufacturing, which is now positive and statistically significant for all three measures of inequality. Adding the average age of the population (Specification 3) only marginally increases the explanatory power of the model, and this variable is only statisti-
Reexamining distribution of wealth in 1870
165
cally significant in the regression for personal property inequality. While including age does not greatly affect the magnitude of the estimated effects of the other explanatory variables it does increase the standard errors for several of them. The regression results in Table 8.7 suggest several conclusions. First, consistent with Kuznets (1955) hypothesis, increasing urbanization and industrialization are positively related to the level of inequality. The effect of urbanization was consistently strong for all measures of inequality across all three specifications. The impact of industrialization is not as consistently significant, but after controlling for literacy we find that the share in manufacturing had a positive and statistically significant relationship with all three measures of inequality. This relationship is not simply a compositional effect arising because inequality was higher in urban areas. Restricting the analysis to residents of rural counties (those with populations of less than 2,500) we find that the positive relationship between inequality and each state’s level of urbanization and industrialization is, if anything, stronger, than for the population as a whole. In other words, inequality among a state’s rural population was increased by the extent of urbanization and manufacturing in the state. Second, even, after emancipation, the legacy of slavery continued to exert an important influence on wealth distribution in 1870. This is clearly true for real property ownership, where after controlling for urbanization and industrialization the states with the largest fraction of Blacks in their population had the highest rates of inequality. It is less evident in the distribution of personal property. That the relationship between inequality and the share of Blacks weakens with the inclusion of the literacy measure suggests that this is one important mechanism through which slavery may have affected wealth accumulation.
Conclusions Information on real and personal property ownership collected in the federal population censuses of 1850 through 1870 offer one of the few opportunities to study patterns of wealth accumulation and inequality in the nineteenth-century United States. Comparing the aggregate level of wealth inequality in 1870 to recently constructed time series for the twentieth century reveals that, at that aggregate level, wealth was much more equally distributed in 1870 than it was early in the twentieth century. Thus, consistent both with Kuznets’ conjecture that inequality follows an inverted-U shape pattern over time, and the time series evidence for Massachusetts in the nineteenth century, it appears that inequality was rising during the late nineteenth century. Taking advantage of the large size of the IPUMS sample we are able to examine patterns of spatial variation in wealth inequality at a relatively disaggregated level. As we have shown, inequality varied considerably across states. Exploiting this cross-sectional variation we have shown that it supports a link between economic development and rising inequality. In cross-state regressions, measures of industrialization and urbanization are positively correlated with
Total property inequality Fraction Black Fraction in city >25,000 Fraction in manufacturing Fraction literate Average age (years) Constant Adjusted R2
Personal property inequality Fraction Black Fraction in city >25,000 Fraction in manufacturing Fraction literate Average age (years) Constant Adjusted R2
Real property inequality Fraction Black Fraction in City >25,000 Fraction in manufacturing Fraction literate Average age (years) Constant Adjusted R2
0.355 0.365 0.756 0.114
1.042 0.514
0.148
1.367 0.237
1.985 1.185 1.152
0.461 0.475 0.983
0.114
0.979 0.638
1.098 1.361 0.922
0.354 0.365 0.755
Standard error
2.805 0.916 2.856
Coefficient
Specification 1
Table 8.7 OLS estimates of determinants of state inequality, 1870 ■
2.251 0.625
0.744 1.418 2.685 –2.120
2.961 0.418
–0.539 1.668 2.944 –2.796
1.859 0.677
1.901 1.085 3.973 –1.544
Coefficient
Specification 2
0.359
0.471 0.327 0.795 0.604
0.464
0.610 0.424 1.028 0.782
0.385
0.506 0.351 0.853 0.649
Standard error
■
0.619 1.344 2.036 –2.514 0.047 1.449 0.631
–0.844 1.489 1.363 –3.758 0.114 1.005 0.488
2.010 1.149 4.537 –1.201 –0.041 2.557 0.677
Coefficient
Specification 3
0.479 0.330 0.947 0.679 0.038 0.739
0.585 0.404 1.157 0.830 0.046 0.903
0.518 0.357 1.024 0.734 0.041 0.799
Standard error
Obs 42 42 42 42 42 42 42 42
Mean 1.712 1.747 1.546 0.139 0.110 0.085 0.563 23.672
Std. Dev. 0.637 0.572 0.551 0.187 0.168 0.089 0.174 2.619
Notes Coefficients in bold are statistically significant at the 95% confidence level or greater.
Source: Ruggles et al. (2004).
Summary statistics Variable Real property inequality Personal property inequality Total property inequality Fraction Black Fraction in city >25,000 Fraction in manufacturing Fraction literate Average age (years) Min 0.757 0.843 0.721 0.000 0.000 0.010 0.135 20.534
Max 3.668 3.104 2.986 0.590 0.857 0.488 0.850 29.570
168
J.L. Rosenbloom and G.W. Stutes
state inequality measures. Industrialization was not the only factor that mattered, however. Some of the highest levels of inequality were found in the South, which remained in 1870 highly rural and agricultural. This exception is explained, however, by the legacy of slavery. Despite the recent abolition of slavery, the inequality that it had bred managed to survive after the Civil War. In addition to examining spatial variation in inequality we also consider patterns of property ownership across a variety of different population subgroups. This examination shows both that wealth accumulation followed certain predictable patterns associated with personal characteristics, but that even within homogenous population groups wealth varied substantially. Even after controlling for a wide range of personal characteristics we can account for only about one-third of the variation in wealth holding across individuals.
Notes * We thank Tom Weiss, Joseph Ferrie and Lee Craig for many helpful suggestions. We are also indebted to participants in the Northwestern University Economic History Seminar, the All-UC conference on the new history of economic inequality and the NBER-DAE Summer Institute for their assistance. This research would have been much more difficult to complete without access to the IPUMS sample of the 1870 census. 1 Using data from the Internal Revenue Service, Piketty and Saez (2001) have shown that income inequality followed a roughly U-shaped pattern: falling sharply during the Great Depression and World War II before beginning to increase. At first, inequality rose gradually, but over the past several decades income dispersion has grown rapidly, so that by the end of the century it had returned to levels comparable to those at the beginning of the century. As with income distribution, inequality in wealth distribution declined dramatically during the 1930s and 1940s. But, in contrast to income, there has been no corresponding rise in wealth inequality in the recent past according to the evidence compiled by Kopczuk and Saez (2004). 2 Soltow (1975) contains a relatively comprehensive discussion of wealth accumulation and distribution based on a national sample of census returns at all three dates. His sample is, however, considerably smaller than that collected in the IPUMS thus limiting his ability to disaggregate the data across different demographic groups or geographic areas. Steckel (1990) used a sample of about 1,500 observations matched from the 1850 to 1860 censuses to examine wealth accumulation in the 1850s, while Ferrie (1999) used samples of immigrants and natives in 1850 and 1860 to trace the impact of changes in occupation and location and wealth accumulation. Atack and Bateman (1981) analyzed wealth accumulation over the life-cycle based on a sample of approximately 21,000 rural northern households in 1860. 3 In 1870 family interrelationships were not recorded by enumerators, but their instructions specified that the household head’s name should be entered first in the record for each family recorded, with other members following. Using this fact the compilers of the IPUMS have constructed the family relationship variable for record locations for each individual in the family along with other demographic data. 4 To assess the impact of truncation on the data we constructed a hypothetical personal property variable in which we used the 1860 distribution of wealth holding for those with less than $100 of wealth to assign non-zero values to a portion of those recorded as having no personal property in 1870. We then compared measures of aggregate wealth and the distribution of wealth in each state for the actual and hypothetical data and found that they were quite similar.
Reexamining distribution of wealth in 1870
169
5 Space does not permit us to report tests of the statistical difference of these average values, but differences between the three southern census divisions and the northern ones are all statistically significant at the 95 percent confidence level. 6 For continuous variables the transformed coefficient is the slope of the probability function calculated at the means of the independent variables. For zero-one dummy variables we report the change in probability resulting from changing the value of the particular dummy variable from zero to one. 7 Because the dependent variable in the regressions is the natural logarithm of the wealth reported it is necessary to exponentiate the coefficient values to calculate the precise percentage differences in value. 8 Our assessment of economic significance is based on calculating the implied effect of a one standard deviation change in each variable. For total wealth, a one standard deviation increase in the share employed in manufacturing would have increased the Theil index by 0.18, or a bit more than 10 percent of the unweighted average of the index across states.
References Atack, Jeremy and Bateman, Fred (1981) “Egalitarianism, Inequality, and Age: The Rural North in 1860”, Journal of Economic History, 41, no. 1: 85–93. Ferrie, Joseph P. (1999) Yankees Now: Immigrants in the Antebellum United States, 1840–1860, New York and Oxford: Oxford University Press. Kopczuk, Wojciech and Saez, Emmanual (2004) “Top Wealth Shares in the United States, 1916–2000: Evidence from Estate Tax Returns”, National Tax Journal, 57, no. 2, 445–87. Kuznets, Simon (1955) “Economic Growth and Income Inequality”, American Economic Review 45, no. 1: 1–28. Piketty, Thomas and Seaz, Emmanual (2003) “Income Inequality in the United States, 1913–1998”, Quarterly Journal of Economics 118, no. 1: 1–39. Ruggles, Steven and Sobek, Matthew, Alexander, Trent, Fitch, Catherine A., Goeken, Ronald, Hall, Patricia Kelly, King, Miriam and Ronnander, Chad (2004) Integrated Public Use Microdata Series: Version 3.0, Minneapolis: Historical Census Projects, University of Minnesota, 2003 http://www.ipums.org. Soltow, Lee (1975) Men and Wealth in the United States, 1850–1870, New Haven, CT and London: Yale University Press. Steckel, Richard (1990) “Poverty and Prosperity: A Longitudinal Study of Wealth Accumulation, 1850–1860”, Review of Economics and Statistics, 72, no. 2: 275–85. Steckel, Richard H. (1994) “Census Manuscript Schedules Matched with Property Tax Lists: A Source of Information on Long-Term Trends in Wealth Inequality”, Historical Methods, 27, no. 1: 71–85. Steckel, Richard H. and Moehling, Carolyn M. (2001) “Rising Inequality: Trends in the Distribution of Wealth in Industrializing New England”, Journal of Economic History 61, no. 1: 160–183. Williamson, Jeffrey G. and Lindert, Peter (1980) American Inequality: A Macroeconomic History, New York: Academic Press.
The publications of Thomas J. Weiss
Work on the service industries The Service Sector in the United States, 1839 through 1899. New York: Arno Press, 1975. “The Service Sector in the U.S., 1839-1899,” Journal of Economic History 28, no. 4, Dec., 1967, 625–28. “The Service Industries in the 19th Century,” with Robert Gallman, in Production and Productivity in the Service Sector, Victor R. Fuchs ed., Studies in Income and Wealth, vol. 34. New York: NBER, 1969. “U.S. Transport Advance and Externalities: A Comment,” Journal of Economic History 28, no. 4, 1968, 625–28. “Urbanization and Growth of the Service Workforce,” Explorations in Economic History 8, no. 3, Spring 1970, 631–34. “The Impact of the Rural Market on the Growth of the Urban Workforce, U.S. 1870-1900,” with John Ermisch, Explorations in Economics History 11, no. 2, Winter 1974, 137–53. “The Service Sector in Economic Growth,” in The Encyclopedia of American Economic History. New York: Charles Scribner’s Sons, 1980. “The 19th Century Origins of the American Service Industry Workforce,” Essays in Economic and Business History, 1984, 48-68. “Tourism in America before World War II,” Journal of Economic History 64, no. 2, June 2004, 289–327. “The Service Industries,” in Susan B. Carter, Scott Sigmund Gartner, Michael R. Haines, Alan L. Olmstead, Richard Sutch, and Gavin Wright, eds. Historical Statistics of the United States, Earliest Times to the Present. New York and Cambridge: Cambridge University Press, 2006, Vol. 4, chap. Dh.
Work on southern industrialization “Large Scale Manufacturing in the South and West, 1850 and 1860,” with Fred Bateman and James Foust, Business History Review 45, no. 1, June 1971, 1–17. “The Participation of Planters in Manufacturing in the Antebellum South,” with Fred Bateman and James Foust, Agricultural History 48, no. 2, April, 1974, 277–97. “Profitability in Southern Manufacturing: Estimates for 1860,” with Fred Bateman and James Foust, Explorations in Economic History 12, no. 3 July 1975, 211–31. “Comparative Regional Development in Antebellum Manufacturing,” with Fred Bateman, Journal of Economic History 35, no. 1, March 1975, 182–208.
Publications of T.J. Weiss
171
“Market Structure Before the Age of Big Business: Concentration and Profit in Early Southern Manufacturing,” with Fred Bateman, Business History Review 49, no. 3 Autumn 1975, 312–36. “Manufacturing in the Antebellum South,” with Fred Bateman, Research in Economic History. Greenwich, CT: JAI Press, 1976. A Deplorable Scarcity: The Failure of Industrialization in the Antebellum South, with Fred Bateman. Chapel Hill: University of North Carolina Press, 1981. “Southern Business Never Had It So Good! A Look at Antebellum Industrialization” in Business in the New South. Sewanee, TN: The University Press, 1981. “Risk, the Rate of Return and the Pattern of Investment in Nineteenth Century American Manufacturing,” with Jeremy Atack and Fred Bateman, Southern Economic Journal 49, no. 1, July 1982, 150-61.
Work on the U.S. labor force “The Industrial Distribution of the Urban and Rural Workforce: Estimates for the United States, 1870-1910,” Journal of Economic History 32, no. 4, December 1972, 919–37. “Revised Estimates of the United States Workforce, 1800 to 1860,” in Long-Term Factors in American Economic Growth, Stanley L. Engerman and Robert E. Gallman, eds. Studies in Income and Wealth, vol. 51. Chicago, IL: University of Chicago Press, 1986. “Demographic Aspects of the Urban Population, 1800 to 1840,” in Essays in U.S. Economic History in Honor of Stanley Lebergott, Peter Kilby, ed. Middletown, CT: Wesleyan University Press, 1987. “Labor Force Changes in the Old Northwest,” with Eleanor von Ende, in Essays on the Economy of the Old Northwest. Athen, OH: Ohio University Press, 1987, 103–30. “Estimates of White and Non White Gainful Workers in the United States by Age Group and Sex, 1800 to 1900,” Historical Methods 32, no. 1, Winter 1999, 21–36.
Work on U.S. economic growth in the nineteenth century “U.S. Labor Force Estimates and Economic Growth, 1800 to 1860,” in The Standard of Living in Early Nineteenth Century America, R. Gallman and J. Wallis, eds. Chicago, IL: University of Chicago Press, 1992. “Consumption of Farm Output and Economic Growth in the Old Northwest, 1800 to 1860,” with Terry von Ende, Journal of Economic History 53, no. 2, June, 1993, 308–18. “Economic Growth Before 1860: Revised Conjectures,” in Economic Development in Historical Perspective, D. Schaefer and T. Weiss, eds. Stanford, CA: Stanford University Press, 1994. “Economic Growth in the United States: Antebellum Period,” in The Oxford Encyclopedia of Economic History, Joel Mokyr, ed. Oxford and New York: Oxford University Press, 2003.
Work on U.S. agriculture “Long Term Changes in U.S. Agricultural Output per Worker, 1800 to 1900,” Economic History Review 46, no. 2, May, 1993, 324–41. “Agricultural Productivity Growth During the Decade of the Civil War,” with Lee Craig, Journal of Economic History 53, no. 3, September 1993, 527–48.
172
Publications of T.J. Weiss
“Transportation Improvements and Land Values in the Antebellum United States: A Hedonic Approach,” with L. Craig and R. Palmquist, Journal of Real Estate Finance and Economics 16, no. 2, March 1998, 173–90. “Nutritional Status and Agricultural Surpluses in the Antebellum United States,” with Lee Craig, in Studies on the Biological Standard of Living in Comparative Perspective: Proceedings of a Conference Held in Munich, January 18–23, 1997, J. Komlos and J. Baten eds. Department of Economics, University of Munich, no. 6, March 1998, 190–207. “Hours at Work and Total Factor Productivity Growth in 19th-Century U.S. Agriculture,” with Lee Craig, Advances in Agricultural Economic History, vol. 1, 2000, 1–30.
Work on colonial America “Was Economic Growth Likely in Colonial British North America,” with Peter C. Mancall, Journal of Economic History 59, no. 1, March 1999, 17–40 “Slave Prices in the Lower South, 1722–1809,” with Joshua L. Rosenbloom and Peter C. Mancall, Journal of Economic History 61, no. 3, September 2001, 616–39. Agricultural Labor Productivity in the Lower South, 1720–1800,” with Joshua L. Rosenbloom and Peter C. Mancall, Explorations in Economic History 39, no. 4, October, 2002, 390–424. “Conjectural Estimates of Economic Growth in the Lower South, 1720 to 1800,” with Joshua L. Rosenbloom and Peter C. Mancall, History Matters: Economic Growth Technology, and Population, Bill Sundstrom and Tim Guinnane, eds. Stanford, CA: Stanford University Press, 2004. “Indians and the Economy of Eighteenth-Century Carolina,” with Joshua L. Rosenbloom and Peter C. Mancall, in The Atlantic Economy during the Seventeenth and Eighteenth Centuries: Organization, Operation, Practice and Personnel, Peter A. Coclanis, ed. Columbia, SC: University of South Carolina Press, 2005. “Slave Prices, the African Slave Trade, and Productivity in Eighteenth-Century South Carolina: A Reply,” with Joshua L. Rosenbloom and Peter C. Mancall, Journal of Economic History 66, no. 4, December 2006, 1066–71. “Exports and the Economy of the Lower South, 1720–1800,” with Peter C. Mancall and Joshua L. Rosenbloom, Research in Economic History. Greenwich, CT: JAI Press, 2008.
All other publications Economic Development in Historical Perspective, editor (with D. Schaefer). Stanford, CA: Stanford University Press, 1994. “The Use of Simulation Techniques in Historical Analysis: Railroads Versus Canals,” with Donald Schaefer, Journal of Economic History, 31, no. 4, December 1971, 854–84. “The Regional Diffusion and Adoption of the Steam Engine in American Manufacturing,” with Jeremy Atack and Fred Bateman, in Journal of Economic History 40, no. 2, June 1980, 281-308. “Development, Health, Nutrition and Mortality: The Case of the Antebellum Puzzle in the United States,” with Michael R. Haines and Lee A. Craig, Journal of Economic History 63, no. 2, June, 2003, 382–413.
Index
Baker, M.N. 38–9, 40 bankruptcy 69–70 bastardy see non-marital childbearing Bateman, Fred xv, xvi, 2, 54, 55, 56, 147 benefit systems: and non-marital births 19 Berle, Adolph 54, 70 BHPS (British Household Panel Survey) 24, 26, 29–30 birth control 8–9, 11, 18 births: outside marriage see non-marital childbearing Boyer, G.R. 13 Brandeis, Elizabeth 126 British Household Panel Survey (BHPS) see BHPS Bureau of Labor Statistics 35
Chadwick, E. 36 Chavas, J.P. 107 Chesbrough, Ellis Sylvester 38, 39 Chicago: meat market 106; water and sewage 38, 42, 49 child-bearing decisions 18–20 child-bearing: outside marriage see nonmarital childbearing child labor laws 119, 120, 123 chlorination 43 cholera 36, 37, 43 Cincinnati: meatpacking 105–6 Clemen, Rudolf 105–6 cliometric revolution xv, 1–2 coagulation 42–3 Coase, R.H. 107 Coffman, C. 78, 80 cohabiting unions: and non-marital childbearing 11, 18–19, 25–8 Commons, John R. 129 contact filters 49 contraception 8–9, 11, 18 contraceptive pill 18 Cook, H. 10 corn: hog–corn cycle 101–2 corporate ownership 54–5, 61–2, 71 corporation profits 66–71 courtship 9–10 Craig, L. xvi, 78, 80, 82, 83, 90 Cuff, T. 106, 114
Cain, L.P. 67 capital–labor ratios 67, 68, 71 case studies: railroads 91–3, 94–5 Census Bulletins 35 census data xv–xvi, 2 censuses: federal 82, 146; ICPSR data set 82; of manufactures xvi, 2, 56, 137; population 6, 146–9, 150–1; and rail access 91, 92, 94
data sources: and quantitative economics 2; see also censuses death rates 49 demonstration effects: and sanitation expenditures 34, 37–42 demonstration projects: sanitation expenditures 41 depression (1837) 2 diarrhea 34–6, 49
abortion 18, 31n10 abstinence: as birth control 8–9 A Deplorable Scarcity 54, 55 age: and distribution of wealth 147, 153, 161 age-specific fertility rates 1938–2003 14–16 agricultural labor xvi agriculture: and distribution of wealth 146–7, 157, 161; impact of railroads 81–3, 88–9 Aldrich, M. 127 Atack, Jeremy xvi, xviin3, 2, 56, 67, 147
174
Index
DID (“difference-in-difference”) approach 78, 85–6, 93, 95 dysentery 34–6, 49 educational differences: non-marital childbearing 23–5 employment-share weighted indexes (ESWI) see ESWI Engerman, S. 91 Engineering News 36, 39–40, 42 epidemics: effects on sanitation expenditures 34, 36–7 Epstein, Ralph 55 equity markets: for manufacturing enterprises 66 ESWI (employment-share weighted indexes) 131–6, 138–40 Ezekiel, Mordecai 107 Family Planning Association (FPA) 18 Farmers’ Anti-Monopoly Convention meeting 54 farming see agriculture fertility rates 8–9 filtration systems 38–9, 42–8 Financial Statistics of Cities 35 firm mortality 70 Fishlow, A. 79 Fogel, R. 79–80, 91 Foust, James xviin3 Fowler, R.F. 107 FPA (Family Planning Association) see Family Planning Association France: and non-marital childbearing 26 Fuller, George W. 39 Gallman, Robert xv, 61 General Fertility Rate (GFR) 8–9 General Statistics of Cities, 1915 46 germ theory 34, 40 Germany: cholera outbreak 37, 43 GFR (General Fertility Rate) see General Fertility Rate Gibrat’s Law 69 Gilbert, D. 12 gold 2 Goodwin, B.G. 114 Graebner, W. 127 “Granite Railroad” 79 Gregson, M.E. 78, 80 Hamburg: cholera outbreak 37, 43 Hariot, Thomas 102 Hatton, T.J. 13
Hazen, Allen 38, 39 Heckscher–Ohlin model 78, 80–1, 95 hog–corn cycle: amelioration of 100–1, 108, 110–11, 114–15; history of 101–6; propagation 106–10; social savings 111–14; as a term 100 hogs 102–4 household heads 148 Humphries, S. 10 ICPSR census data set 82 Illinois: railroad case study 91–3, 94–5 income distribution see wealth distribution income inequality see inequality Income Support 19 incorporation laws 61–3 Indiana: railroad case study 91–3 industrialization 161–3, 165 inequality: determinants of 166–7; distribution 146–7; and economic transformation 147, 161–5; sources of 157–61 integrated public use micro-data samples (IPUMS) see IPUMS Integrated Public Use Microdata Series 2 intermittent filtration: sewage treatment 48 IPUMS (integrated public use micro-data samples) 6, 91, 92, 94, 146–9, 150–1 “iron-rule”: of hog–corn economics 106–7 Jackson, Andrew 2 Japan: non-marital childbearing 18 Kantor, S.E. 127 Kirkwood, James P. 38, 39 Kopczuk, Wojciech 147, 152 Krooss, H.E. 110 Kuznets, Simon xv, 147, 165 labor inputs xvi labor productivity: and labor regulations 136–8, 140 labor regulations: expansion 119, 120–6; influence 119; and labor productivity 136–8, 140; measures of 127–36; patterns and sources 120–7 land values: and rail access 78, 80, 90, 91, 93, 95, 96–7 Law Omsted, Frederick 105 Lawrence Experiment Station 40–1, 43, 48 Lebergott, Stanley xvi liability laws 119, 123, 128 limited liability protection 61 Lindert, Peter 147
Index 175 McConnell, J.L. 55 maize see corn Marcus, M. 56 marriages: delay of 18; shift from 17–18; of single women 11–12 Marshall, John 61, 71 Means, Gardiner 54, 70 meat barons 108 mine safety regulations 120, 123 Moehling, Carolyn M. 147 monetary gold 2 mortality crises see epidemics mortality rates 49 mortality shock 36 Mortality Statistics of Cities 35 mortality: and sanitation expenditure 34–6 National Association of Manufacturers 136 national income: measuring 1 nineteenth-century firms: corporation profits 66–71; distribution of profits 57–61; estimating profitability 56–7; investments 61; organization 61–5 non-marital childbearing: age-specific fertility rates 1938–2003 14–18; background 8–11; and cohabiting unions 25–8; and contraception 18–20; educational differences 23–5; as rational choice 18–20; and social influence 8–10, 20–3; and unemployment 10, 11–14, 20, 21–2, 23; before World War II 11–14 North–South difference: distribution of wealth 146, 152–3, 157 Norway: and non-marital childbearing 26 occupation: and distribution of wealth 146–7, 156–7, 161 Palmquist, R.B. 80, 83 Parker, William xv partnerships 63–4, 70 Paterson, D.G. 67 personal property 149–52, 153–7 population density: and rail access 84–7 pork see hog–corn cycle Portugal: and non-marital childbearing 26 premarital sexual intercourse 9–10, 30n10 profitability, and firm size: and apologist populist view 54; and corporate ownership 54–5; and nineteenth-century firms see nineteenth-century firms; twentieth-century experience 55–6 Progressive Era: labor legislation see labor regulations
property ownership 153–7, 158–9, 168 race: and distribution of wealth 146, 153, 161 Rafter, George W. 40 rail access 78, 83–8 railroads: aggregate impact 79–80; case studies 91–3, 94–5; local level impact 80–91; origins 79 rates of failure 69–70 rates of return: and firm size see profitability real property 149–52, 153–7 refrigeration: and hog–corn cycle 108–9, 114–15; and pork production 100–1 Saez, Emmanual 147, 152 Samuels, J.M. 55 sanitation capital projects 37–9 sanitation expenditures: and demonstration effects 37–42, 46, 51; demonstration projects 41; effect on mortality 34–6; and epidemic effects 36–7, 49, 51; sewage treatment technologies 48–9, 50; testing centers 40–1; water treatment technologies 42–8 screening: sewage treatment 48 Second Bank of the United States 2 Sedgwick, William Thompson 40–1 sedimentation 42–3 septic tanks 48–9 settling: sewage treatment 48 sewage farming 48 sewage treatment expenditures see sanitation expenditures sexual restraint: culture of 8–9 Shannon, Fred 106, 107 Shattuck, L. 36 slavery 2, 165, 168 Smith, Theobald 40–1 Smyth, D.J. 55 social equilibrium 20–1 social interaction effect 20–2, 26, 28 Sokoloff, Kenneth 55 sole proprietorship 63, 67, 70 Soltow, Lee 146 Southall, H. 12 state labor legislation see labor regulations Steckel, Richard 147, 148 stock ownership 66–7 Stover, J.F. 79, 84 Studenski, P. 110 Summers, H.B. 55 Sweden: and non-marital childbearing 26
176
Index
Szreter, S. 8–9 tax avoidance 55 testing centers: sanitation expenditures 40–1 Trollope, Frances 105 typhoid fever 34–6, 41, 43, 49 unemployment rate: and births outside marriage 10, 11–14, 20, 21–2, 23 unionization 119, 124–5 urbanization: and railroads 93, 96; and wealth distribution 164, 165 urban–rural areas: and distribution of wealth 146, 156 U.S. Bureau of Corporations 54 USA: and non-marital childbearing 26 Von Theunen model: of agricultural land rents 78, 80–2 wage data 82–3 water supplies: and mortality 34–6
water transportation 83–4, 96 water treatment technologies 42–8 waterborne death rate 36, 49 waterborne diseases 34–6 wealth distribution: data characteristics 147–8; data sources 146; decomposition 146–7; determinants of 165–8; and economic transformation 161–5; and personal characteristics 153–7; regional differences 149–52; Theil index 157–61 Weiss, Thomas: A Deplorable Scarcity 54, 55; career overview xv–xvii; censuses xv, 2, 56; pork production 112; water and rail access 80, 83 “welfare capitalists” 136 Williamson, Jeffrey G. 147 women: and distribution of wealth 153; household heads 148; marriages of single 11–12; working hours 119 women’s hours laws 120, 124 the workhouse 10 Wrightson, K. 9–10