The International Structure of Income: Its Implications for Economic Growth

The International Structure of Income Its Implications for Economic Growth

Salvatore J. Babones

for my mother

Contents Preface ............................................................................................................xiii 1. Introduction.................................................................................................... 1 2. Trends in International Income Inequality...................................................... 7 3. From International Inequality to Global Inequality....................................... 35 4. The Structure of National Income ................................................................ 63 5. Neoclassical Economic Growth in a Structured World-Economy............... 103 6. Investment and Growth: Disentangling Causal Paths ................................. 143 7. Conclusion ................................................................................................. 163 References...................................................................................................... 169 Index .............................................................................................................. 177 About the Author ........................................................................................... 181

vii

viii

List of Tables Table 1. Three methodologies for international income inequality .................. 19 Table 2. National inequality by decade and region (SD of Logs).................... 48 Table 3. Summary of the origins of cases for national inequality samples....... 50 Table 4. Modes of the global income distribution (logged) ............................. 58 Table 5. Modes of the global income distribution (1995 dollars)..................... 59 Table 6. Annual levels of PP and PC boundaries............................................. 87 Table 7. Position and mobility in the world-economy, 1960-1999................... 89 Table 8. Observed mobility across boundaries of the world-economy............. 96 Table 9. Organic zone membership (benchmark countries only) ..................... 97 Table 10. Regression of log(GNP) on neoclassical inputs, 1995 data ............ 128 Table 11. Long-run neoclassical growth model (1960-1999)......................... 129 Table 12. Long-run neoclassical model, LABOR excluded (1960-1999)....... 131 Table 13. Stratified long-run model (1960-1999) .......................................... 133 Table 14. Stratified medium-run model (1975-1995) .................................... 135 Table 15. Medium-run models (1975-1995).................................................. 138 Table 16. Stratified, differentiated medium-run model (1975-1995).............. 139 Table 17. Full model for growth and investment, 1975-1995 all avail. .......... 156 Table 18. Full model for growth and investment, 1975-1995 periphery ........ 157 Table 19. Panel model for growth and investment, 1975-1995...................... 159 ix

x

List of Figures Figure 1. FX versus PPP methodology (schematic)......................................... 14 Figure 2. Real Local Currency (RLC) methodology (schematic)..................... 16 Figure 3. International income inequality -- weighted VarLogs....................... 20 Figure 4. International income inequality, weighted Ginis .............................. 21 Figure 5. International inequality -- weighted VarLogs (China excluded) ....... 23 Figure 6. International inequality -- weighted Ginis (China excluded) ............ 24 Figure 7. Trajectory of Chinese growth since 1960 (FX, LCR, PPP)............... 26 Figure 8. Detail of Chinese growth since 1990 (FX, LCR, PPP) ..................... 29 Figure 9. Illustration of a Lorenz curve ........................................................... 38 Figure 10. Quintile-based (dashed) versus actual (solid) Lorenz curves .......... 40 Figure 11. Observed Ginis versus SDs of logs, "high quality" data ................. 42 Figure 12. Illustration of a lognormal distribution of income .......................... 44 Figure 13. Gini and SD of the logarithms for a lognormal distribution............ 45 Figure 14. Fit of national inequality to GNP per capita ................................... 48 Figure 15. Illustration of a gaussian kernel (United States) ............................. 52 Figure 16. Global distribution of income, 1960s (Tier II sample)................... 54 Figure 17. Global distribution of income, 1970s (Tier II sample) ................... 54 Figure 18. Global distribution of income, 1980s (Tier II sample)................... 55 Figure 19. Global distribution of income, 1990s (Tier II sample) .................... 55 xi

Figure 20. Two-population model of the global income distribution ............... 57 Figure 21. Distribution of countries in the world-economy (1965) .................. 83 Figure 22. Distribution of countries in the world-economy (1975) .................. 83 Figure 23. Distribution of countries in the world-economy (1985) .................. 84 Figure 24. Distribution of countries in the world-economy (1995) .................. 84 Figure 25. Average internal income inequality by zone, 1960s-1990s........... 100 Figure 26. Investment differentiation growth model...................................... 147 Figure 27. Investment differentiation model with reciprocal effects .............. 148 Figure 28. Structural model of the investment-growth nexus......................... 150 Figure 29. Investment-growth nexus, all countries, 1975-1995 ..................... 152 Figure 30. Investment-growth nexus, peripheral countries, 1975-1995 ......... 154

xii

Preface This book is a new and extensively edited version of my dissertation of the same title, which was deposited at The Johns Hopkins University in September, 2002. I have expanded and updated the introduction and conclusion, reorganized the substantive chapters, edited the whole text for readability, and added a detailed index. I have also eliminated the extensive technical appendices that, in the original, ran longer than the main text itself. Those seeking highly technical details like lists of the particular countries included in each analysis, tests of the robustness of the numerical results, and multiple alternate specifications of the statistical models are directed to the dissertation, available electronically in facsimile form from ProQuest. Most readers, however, will find the current volume much more useful and accessible. The material is better organized, the chapters are better laid out, and the index is invaluable. That said, most of the actual text of the main body of this volume is drawn directly from the 2002 original. Of course, much has changed in the world since the composition of that original text, but strangely little that would affect the interpretation of the results presented in this volume. It is, after a fashion, a striking confirmation of the ideas expressed here that they require little revision upon publication six years later. The simple fact is that the structure of world income has changed very

The International Structure of Income little in decades (or, for that matter, centuries) and is changing very little today despite the massive dislocations associated with the current global financial crisis. Once one looks beyond the topical ripples to anything approaching the longue durée, the structure of the world-economy has been remarkably stable over time. In terms of income, the countries of the world today rank pretty much where they did 100 years ago, a few historical oddities notwithstanding. The ultimate implication of this for economic growth may simply be to imply that growth is not very relevant for understanding our world. Structure clearly dominates. It was the problem of growth, however, that brought me to the topic. I first became interested in issues of economic development as a senior sociology major at the University of Montevallo. My advisor, guide, and confidant there was Jeri Bullard, now of the University of North Alabama. I did not go into sociology in order to study development, but when I "discovered" development I immediately saw the importance of studying it from a sociological perspective. There is no pure economics of growth and development; economics can help elucidate the relationship between economic allocation and economic growth, but it cannot, in the end, solve the root problems of allocation. The global allocation of resources, both within and among nations, is fundamentally a sociological phenomenon. This book is concerned not so much with the causes of the international structure of the allocation of economic resources as with its implications for growth.

Neoclassical and "new" growth theory in economics have largely

ignored the implications of sociological theory for understanding economic growth. Economists working in these traditions have not largely, but completely ignored the sociological literature on growth.

The result has been the

promulgation and proliferation of formal growth models that do not correspond xiv

Preface in any meaningful way to empirical reality. Contrary to established theory, capital investment does not flow to the poorest countries; international immigration is dominated by the relatively rich, rather than the relatively poor, and flows in the "wrong" directions; trade liberalization is not turning out to be the panacea it promised to be; convergence between rich and poor countries, if it is occurring at all, is occurring very, very slowly. To the credit of the economics profession, a vibrant tradition of criticism and debate among economists has led to a general awareness of these empirical shortcomings in their dominant theoretical perspectives.

This awareness,

however, has not led them to look farther afield to ideas emanating from other disciplines. However, economists are not the only students of economic growth who insist on wearing disciplinary blinders. A general aversion to the political and policy stances of the mainstream of the economics profession has led sociologists to devalue and deride the important orienting influence of formal modeling in economic theories of growth. Sociologists also rarely reference the current empirical literature in economics, preferring instead to get their economics from policy books and popular media such as magazines and newspapers.

As a result, although

sociologists do not operate in complete ignorance of economics, their understanding of economists' work on growth is generally topical and often a decade out-of-date. This is unacceptable. One need not accept the failed policy prescriptions of neoclassical economists in order to adopt and adapt the genuine theoretical accomplishments of neoclassical economics. I take neoclassical economics, and specifically the neoclassical growth model, as an orienting framework, which I then proceed to fill in with sociological data and insight. In my opinion, such an approach best leverages the strengths of the two fields: economics for formal mathematical modeling and xv

The International Structure of Income sociology for applied empirical analysis. Were I a sociological purist, I could have run the same models without locating their origins in economic theory, and were I an economist, I could have experimented with the non-traditional specifications found below without referring to their sociological inspirations. By taking an explicitly interdisciplinary approach, I have gained enormous insights from work in both fields. I hope that my readers, be they economists, sociologists, other social scientists or laypersons, will similarly benefit from this cross-fertilizing approach. As I said at the opening, this book has grown out of my dissertation research. Accordingly, most of the required acknowledgements are related to that period of my life. First and foremost, while I certainly owe my thanks to all of the members of my dissertation committee at Johns Hopkins, I owe a much greater measure of thanks to one scholar in particular. This is my academic advisor and personal friend, Christopher Chase-Dunn. Although this book does not flow directly from his work, it does flow directly from his unwavering support and encouragement, from my first week as a new graduate student at Johns Hopkins clear through to today. Quite simply, without his input, this book, and the dissertation it is based on, would not exist. Thank you, Chris. I also owe a great debt, both intellectual and personal, to my second reader (and de facto second advisor), Giovanni Arrighi. Much of Chapters 3, and 4 of this book is a replication, confirmation, and extension of his work. After joining the faculty at Hopkins, he graciously took on responsibility for helping oversee my dissertation work, going well beyond the call of duty to add a late-stage dissertation student to his docket of finishing students at Binghamton and new students at Hopkins. My thanks to him and to my third departmental reader at Johns Hopkins, Robert Gordon, for cheerfully pitching in when called upon to add this substantial burden to their already busy schedules. xvi

Preface One other sociologist deserves special mention with regard to the successful completion of this project.

Without the constant support and

encouragement of Binnie Bailey, my dissertation likely never would have been started, and certainly would not have been completed, and thus this book would not exist I would like to take the opportunity to thank her for all of her help over those challenging years. I would also like to thank the members of the Johns Hopkins sociology office staff, including Vonnie Wild, Pam Stichion, Mary Ann Zeller, and especially Amy Cline for her help in the home stretch. If Binnie has been the godmother to this project, they have been the midwives. Of course, in a much deeper sense, I "owe it all" to my mother. This book is dedicated to her, as was my dissertation. I hope she likes it.

Salvatore Babones Sydney, Australia November 1, 2008

xvii

The International Structure of Income

xviii

1. Introduction A decade ago in 1998 an interdisciplinary team of four economists and sociologists published a general review of the state of the literature on economic growth. Reviewing articles in both the sociology and the economics literatures, they found an almost complete disjunction between the two camps. The authors concluded: For more than two decades, economists and sociologists have pursued parallel cross-national quantitative investigations of the determinants of economic development. These investigations have proceeded in mutual ignorance despite the often large overlap in statistical methods and data employed. Apparently contradictory findings have resulted . . . . What the studies we have reviewed herein highlight, perhaps more than anything else, are the need to avoid simplistic theoretical formulations and the virtue of drawing upon the existing and related work of various disciplines. (Crowly et al 1998, p. 45) I agree wholeheartedly with the sentiments expressed by Crowley and her colleagues, but must protest that if there is a tendency to ignore literatures outside one's own discipline, it is much more marked among economists than

The International Structure of Income among sociologists. The present volume is a case in point: though a sociologist, I take the economics literature very seriously. Throughout I review results from economics alongside those from sociology, and I go so far as to frame my theoretical models as extensions of models from the economics literature. Though many of the sociologists I cite also cite economists, in the course of researching this book I never once came across an economist who cited the sociology literature. Interdisciplinarity seems to be a one-way street. It is none the less important for that fact. World-systems sociology in particular could benefit from importing tools and perspectives from neoclassical economics.

The fact that the assumptions and conclusions of neoclassical

economists tend to conflict with those of world-systems sociologists does not in any way invalidate the efficacy of the tools that neoclassical economists have developed to get from assumptions to conclusions.

In particular, the

neoclassical growth model is a flexible, robust framework for understanding how national-level variables are related to national-level economic growth. The neoclassical growth model depicts growth as the result of having capital inputs (both physical and human) higher than is typical for countries of a given level of national income. This base model seems reasonable, and there is no reason why variables of sociological interest (like dependency or demographics) shouldn't be added to it. The model is sufficiently flexible to allow for this, and provides a convenient framework for comparing the performance of theories from different disciplines. The neoclassical growth model is especially useful for helping us understand the degree to which countries are or are not growing more alike over the years. A major point of contention between economists and sociologists throughout the 1960s, 1970s, and 1980s was that economists asserted that countries' levels of national income would converge over time, while 2

Introduction sociologists observed that levels of national income were in fact diverging. This should not be a theoretical debate: the data exist to test the proposition. Both sociologists and economists have tested it, but have come, each within their own literatures, to conflicting conclusions. In Chapter 2 below I attempt a definitive treatment of this conflict, explaining the reasons for the differences in results reported by different authors. The balance of evidence seems to suggest no change: no convergence, but no major divergence either. I argue, however, that the problem is not the answer but the question. The question presumes the existence of a single economy in which the countries of the world compete on equal terms in a single economic space. The reality is that the world-economy is highly structured, with individuals and countries playing highly circumscribed structural roles. This is why economics needs sociology. One key element of the sociological world-systems approach to the study of economic development is the realization that economic opportunities are not smoothly and seamlessly distributed across all the areas of the world, but are strongly patterned by location in the world-economy. Economic activity is organized in global commodity chains linking ultimate producers with ultimate consumers. As formally theorized by Arrighi and Drangel (1986), there are two general types of locations in the commodity chains that make up the worldeconomy: core nodes, in which economic actors are able to extract a monopoly rent through the reduction or elimination of competitive pressures, and peripheral nodes, in which near-perfect competition drives economic rents toward zero. As a result, theory predicts that the population of the world is roughly divided into two broad income levels, with a great trough separating them. This prediction is at odds with distribution theory in economics, which generally supports the presumption of a unimodal, lognormal distribution of income. 3

The International Structure of Income The bimodal distribution of income predicted by Arrighi and Drangel (1986) refers to the global distribution of individual incomes. At the national level, countries are hypothesized to fall into one of three broad income categories. Core countries are those for which most economic activity occurs within core nodes of global commodity chains; they are thus high-income countries. Peripheral countries are those for which most economic activity occurs within peripheral nodes of global commodity chains; they are thus lowincome countries. Some countries, however, display a mix of core-type and peripheral-type nodes; these countries find themselves in an uneasy tug-of-war between relative wealth and poverty. They do not possess the critical mass of core nodes that would allow them to reduce competitive pressures in all of their nodes of economic activity, but they do possess sufficient core nodes to prevent a complete descent into peripheral-type activity.

These countries form a

permanent "semiperiphery" within the world-economy. The world-systems theory of the structure of the world-economy thus predict a bimodal global distribution of income at the level of the individual, but a trimodal global distribution of levels of national income per capita at the country level. The next two chapters of this book confirm these two predictions in turn. In Chapter 3, I use newly available data to show for the first time that the global distribution of income at the individual level is, in fact, strikingly bimodal. In Chapter 4, I update and improve on earlier work by world-systems sociologists showing that the distribution of the national incomes per capita of the countries of the world is, in fact, trimodal. These two confirmational studies lay the groundwork for the construction of sociologically-informed neoclassical models of economic growth. Returning to the neoclassical growth model, the original formulation of the model is based on the input-output model of Solow (1956), in which 4

Introduction aggregate economic output is modeled as a multiplicative function of the level of technology and levels of inputs of labor, physical capital, and, in recent specifications, human capital. Since inputs of labor, physical capital, and human capital are subject to diminishing marginal returns on investment, permanent, steady-state growth can only be attributed to improvements in technology. In this sense, the neoclassical model can be seen as a formalization of Schumpter's (1951) theory of economic growth. Since technological change is exogenous to the model, the neoclassical model is sometimes referred to as an exogenous growth model. Over the very long term, potential rates of growth in national income are limited by the rate of technological change. Over shorter time frames, however, economic growth (or decline) from one equilibrium level of output to another through changes in the levels of inputs is possible. This kind of growth is termed "disequilibrium growth" because it represents the movement of an economy in disequilibrium towards a new equilibrium level of output. There is every reason to believe that the specific mechanisms of disequilibrium growth differ systematically across the three zones of the worldeconomy. Levels of human capital may be more important for growth outcomes in the core of the world-economy, where technological utilization is most intense; demographic factors may be more important in the periphery, where production is more labor-intensive; the effects of physical capital may differ according to source (domestic or foreign) in the differing political environments prevalent in the core, semipheriphery, and periphery. In short, the coefficients of the neoclassical inputs probably differ by zone. In Chapter 5, I estimate a series of neoclassical models for the world as a whole and for each zone of the world-economy, uncovering important differences in the growth process across the zones of the world-economy. In Chapter 6, I concentrate on an important theoretical debate in the dependency literature: the relationship between foreign 5

The International Structure of Income investment, domestic investment, and growth. Operating within the context of a neoclassical growth model, I am able to show that the previously reported beneficial short-term impact of foreign direct investment is, in fact, spurious. Investment as a whole, in fact, seems to have no causal impact on observed rates of economic growth. This very non-neoclassical result is a striking confirmation, using neoclassical tools, of sociological research in the dependency theory tradition. That this result is inspired by and based on models developed in the economics literature of the past five years demonstrates that powerful new insights can be generated through the cross-fertilization of sociological and economic thought. How much more can be accomplished pursuing this strategy is an open question, though we can hardly do worse than our current understanding of the growth process. In the Conclusion, I propose some immediate next steps that can be taken to apply sociological insights to further improve the specification of formal models of economic growth. In addition, although the focus of this book is on the patterning of growth processes by income zone of the world-economy, in the Conclusion I briefly discuss the implications of patterns of economic growth for our understanding of the global income distribution.

6

2. Trends in International Income Inequality One of the most striking features of the global social geography is the degree to which the world's income is concentrated in just a few countries, while the bulk of the world's population subsists on incomes that do not even exist, as conceivable incomes, in the developed world.

The 15% of the world's

population living in the "developed" countries of the world account for over 78% of global economic output. (World Bank 2001b) While we possess a plethora of theories that attempt to account for this pattern, no one theory or theoretical tradition has emerged to gain wide acceptance as an explanation. Moreover, though the specter of European colonialism may be invoked to explain the geography of global income at the end of the second world war, it implies the paradox that the global income distribution has remained equally wide, and by some accounts has widened much further, in the subsequent era of independent states.

Even dependency theory must admit that political

independence should have yielded (slightly) better economic performance than colonial rule. In this chapter, I present new results on the international distribution of income, working within the summary statistic tradition of existing international income inequality studies; in Chapter 3 I lay out a new course new course of inequality measurement based on the full distribution of income among the

The International Structure of Income whole population of the world. I begin this chapter with a review of the running debate on the international distribution of income. Since the main sources of dispute in this debate center on the choice of data series to be analyzed, I go into some detail on the origins of the relevant data. Then, I replicate and update the trends in inequality as reported in the literature, displaying inequality as computed from each of the three major data sources in parallel. This exercise clarifies the reasons for the wide variety of results reported in the literature, as well as pointing the way forward for a deeper understanding of inequality.

I then

examine the impact of China on international inequality measures. I conclude this chapter with a critique of the practice of using summary measures of international inequality to encapsulate the complexity of the global income distribution.

THE INCOME CONVERGENCE DEBATE Recent years have witnessed a renaissance in comparative empirical research on the international distribution of income. This research has taken a variety of forms, ranging from regression-based panel studies to comparative studies of national income distributions to longitudinal descriptions of between-country inequality.

It also straddles several literatures, including the economics,

development, and sociology literatures. The theoretical fount of much of this recent interest is the debate over neoclassical convergence. Writing after the second world war, neoclassical economists predicted the convergence in incomes of poorer with richer economies, as a result of capital movements from areas of relatively high capital utilization to those where capital intensity is relatively lower. This body of 8

Trends in International Income Inequality theory, based on the work of Samuelson (1948) and Solow (1957), was complemented by the work of heterodox theorists like Gerschenkron (1962), who emphasized the importance of technology transfer in contrast and complement to that of capital movement. Whatever the mechanisms, the new countries of the south were expected to grow rapidly in the decades following formal political independence. Empirically, however, this growth did not materialize, and a sociological critique emerged that capital transfers, far from promoting convergence in income between rich and poor countries, might actual promote and perpetuate divergence.

Building on the theoretical work of Frank (1969), Emmanuel

(1972), and Amin (1974), Chase-Dunn (1975) demonstrated empirically that international economic dependence -- the flip-side of capital movement -actually retarded economic growth. The ensuing empirical debate on foreign capital dependence continues to this day, with recent contributions from Kentor (1998), Dixon and Boswell (1996a, 1996b), and Firebaugh (1992, 1996). Only a hiatus in the production of new data seems to have slowed down the controversy. Although rarely citing the sociology literature on dependence, economists have now largely abandoned the concept of convergence driven by exogenous capital flows. The "new growth theory" of Romer (1986) and Lucas (1988), essentially an elaboration on Gerschenkron, implies that convergence between low and high income countries will only occur where sufficient human capital exists in the poorer countries to take advantage of the technological advances made in the richer countries.

Empirical studies of "conditional

convergence" -- convergence conditional on levels of educational attainment -have generally supported this view. Barro (1991), Barro and Lee (1993a) and Barro and Sala-I-Martin (1995) find that educational levels are significantly 9

The International Structure of Income related to growth rates, and that once education is controlled statistically, low income countries do grow faster than high income countries. More recently, economists have begun to question even the limited finding of conditional convergence, especially over the long-term. (Easterly 2001) More importantly for the present purpose, when no controls are made, Barro and his colleagues find no evidence of absolute convergence, but instead find a measurable (but statistically non-significant) divergence between high and low income countries over the past few decades. This is consistent with Romer's (1987) seminal finding of unconditional divergence, and has since been confirmed by Pritchett (1997). The long-term divergence between rich and poor countries over centuries is even more certain -- and much more striking. (Maddison 1995) In short, the rich are getting richer, and while the poor can't get much poorer, they are not keeping up.

Measures of International Income Inequality Taking a more direct approach to measuring changes in the global distribution of income, a different tradition has emerged in the sociology and development literatures. In a duo of roughly parallel articles, Korzeniewicz and Moran (1997) and Schultz (1998) both demonstrated that by far the largest contributor to overall global inequality is international, or between-country, inequality. By contrast, within-country inequality contributes a much smaller share of total global inequality, although estimates range widely depending on methodology. Essentially, this means that, in the aggregate, the income gaps between rich and poor countries far outweigh the income gaps between rich and poor people within each country. At issue, however, is the trend in between-country inequality itself in the years since 1960. While Korzeniewicz and Moran reported a rising trend in 10

Trends in International Income Inequality inequality from 1960-1992, Schultz reported a relatively stable level of inequality. The Korzeniewicz and Moran findings were based on levels of gross national product (GNP) evaluated in dollars at current contemporary exchange rates (in shorthand, a GNP/FX methodology), while Schultz's conclusions were based on levels of gross domestic product (GDP) evaluated in dollars at purchasing power parities (a GDP/PPP methodology).

That the two

methodologies yield conflicting results was pointed out by Firebaugh (1999), and has been confirmed by Babones (2002). The quantitative difference between GNP and GDP for a given country is usually minor (less than 10%), and since the ratio of GNP to GDP for a given country tends to be stable over time, the choice of one series over the other is not of critical importance for shaping the resulting trend in international income inequality. The choice of methodology for conversion to a common currency is, however, critical. Stepping into the debate, Firebaugh (1999) came down firmly on the side of the GDP/PPP methodology, asserting that: Though early studies in economics used FX estimates because PPP estimates were unavailable, PPP-based income is now the industry standard . . . FX rates are highly flawed calibrators of currencies for two reasons. First, many goods and services are not traded on the international market, so exchange rates are based on a restricted bundle of goods and services . . . Second, FX measures are not totally "free" but are routinely distorted by government policy and speculative capital movement. (pp. 1604, 1609) Of these two objections to the use of the GNP/FX methodology, the first is rather dubious.

Rather than seeing it as an objection, the thrust of

Korzeniewicz and Moran is to view it as a positive virtue. The whole point of comparisons of incomes across countries, according to Korzeniewicz and 11

The International Structure of Income Moran, is to compare purchasing power on the world market. Goods not traded on the world market are irrelevant to this purpose, and the emphasis in the GDP/PPP methodology on domestic consumption is thus misplaced for this purpose.

If the debate between the two methodologies depended only on

Firebaugh's first objection, there would be no debate: the GNP/FX methodology could be used for some purposes, the GDP/PPP methodology for others. This is the compromise solution offered by Korzeniewicz and Moran (2000) in a later exchange with Firebaugh on the issue -- a solution, however, scorned by Firebaugh (2000) in his portion of the exchange. The second objection raised by Firebaugh -- distortions in exchange rates -- is more serious, especially for the years before the early 1990s, when most exchange rates were fixed by governments rather than determined in open markets. Balancing this objection, however, are Korzeniewicz and Moran's objections to the methods underlying the computation of PPP conversion factors, in particular, the paucity of benchmark countries and years and the lack of accounting for the quality of goods and services. Moreover, the compilers of the main GDP/PPP data source, the Penn World Table (PWT), themselves assign "quality ratings" of "C" or better to only 55 of 139 countries in the PWT Mark 5, and of 47 African countries only Kenya even reaches the "C" level. (Summers and Heston 1991) Unfortunately, the choice of data series makes all the difference in determining the trend in between-country inequality in the four decades since 1960. Babones (2002) demonstrates one solution to this dilemma: the exclusion of the non-market economies of China and the Soviet bloc from the analysis. When analysis is restricted to market economies only, Babones reports rising trends in inequality using both the GNP/FX and GDP/PPP methodologies. The inclusion of China in particular is hotly debated on all sides: in population12

Trends in International Income Inequality weighted analyses, the values assigned to China are, for obvious reasons, of critical importance.

Both Firebaugh (1999) and Korzeniewicz and Moran

(2000) make a particular example of China in criticizing the other's methods.

An Alternative Currency Conversion Methodology Although they differ in the conversion factors used, the FX and PPP methodologies are relatively parallel in general outline. In each case, national income in local currencies (LCs) is converted to U.S. dollars by applying a conversion factor of some kind for that year, a conversion factor based either on exchange rates or on purchasing power. The reality is a little more complicated, in that the valuation of outputs into dollars occurs at different stages of the process for the FX and PPP methodologies, but the basic pattern is the same: income for a given (benchmark) year is converted to dollars for that year according to some conversion formula. Once this has been accomplished, the resulting dollar-denominated data can be used for comparisons across all countries for the given year. Both the FX and the PPP methodologies are particularly designed for comparing incomes across countries in a given year. They are not intended for comparing incomes over time. It is possible, however, to make the resulting income measures comparable from year to year. Once national income for each country in each year is expressed in the U.S. dollars of that year, it can be deflated to a common benchmark year using a U.S. dollar price series, such as the U.S. Consumer Price Index or the U.S. GDP deflator. This procedure is outlined schematically in Figure 1. The resulting "real" national income figures (expressed in constant U.S. dollars) can then be used for inter-temporal comparisons of incomes as well as international comparisons. There is, however, one major drawback to this methodology.

The 13

The International Structure of Income computation of inflation rates is plagued by the problem of shifting consumption patters.

This problem is particularly insidious due to the endogeneity of

consumption patters to the inflation rate: people reduce consumption of goods that rise in price relative to other goods, substituting goods that decline in relative price.

Such shifting consumption patterns are not parallel across

countries. In short, inflation rates for one country cannot simply be applied to another country with any kind of validity. Practically speaking, even assuming perfect foreign exchange markets or an impeccable PPP implementation, a country's GNP or GDP growth rate when computed using a constant U.S. dollar income series will not equal its growth rate when computed using a constant LC income series. Adjustment for inflation is currency-specific.

Figure 1. FX versus PPP methodology (schematic) Year

pounds

1995 (base year)

real real GNP$ GNP$ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ nominal --------> nominal nominal <-------- nominal GNP£ GNP$ GNP$ GNP¥

1990 1985 1980 1975

dollars

dollars

yen

To perform comparisons that are at the same time international and intertemporal, compromises must be made. Both the GNP/FX and the GDP/PPP methodologies for studying international income inequality implicitly accept less-than-optimal implementations for converting current incomes to real 14

Trends in International Income Inequality incomes using U.S. dollar price deflators. They make this sacrifice because both methodologies were developed for comparing national incomes across countries at a point in time. International income inequality estimates for any particular year arrived at using these series are perfectly valid (from the perspective of inflation adjustment). However: income inequality estimates based on both the GNP/FX and the GDP/PPP methodologies are not strictly comparable across time periods. Thus, although the GNP/FX and GDP/PPP methodologies provide valid estimates of the mean and variance of the global income distribution at any point in time, it is not strictly speaking appropriate to compare those estimates from year to year. This suggests that an alternative methodology is necessary for studying changes in the global distribution of income over time. The only obvious alternative using readily available data is to reverse the order of the steps in the GNP/FX and GDP/PPP methodologies. The existing methodology involves first converting LC figures to current dollars, and then deflating those dollars to a common year using a U.S. dollar-based price index. In an alternative methodology, it would be technically feasible first to deflate all LC figures to a common base year, then to convert the resulting real LC figures to dollars at the exchange rate for that base year. This procedure is outlined schematically in Figure 2. This alternative, real local currency (RLC) methodology would not be appropriate for estimating the mean and variance of the global income distribution for any given year, since the resulting values would be biased by any random disturbances in the exchange rates in the base year. However, any bias thus introduced into the distribution of global income would be identical in every year studied. The income inequality debate centers not on determining the absolute level of international inequality (which is agreed by all to be 15

The International Structure of Income substantial), but on determining its trend. Any biases introduced by the RLC methodology will be consistent for all years, and thus will not affect the trend. Moreover, similar trends in global inequality should be found using the RLC methodology regardless of the base year or currency conversion factor (FX or even PPP) chosen for the analysis.

Figure 2. Real Local Currency (RLC) methodology (schematic) Year

pounds

dollars

dollars

1995 (base year)

real --------> GNP£ ^ ^ ^ ^ ^ ^ nominal GNP£

real GNP$

real GNP$

1990 1985 1980 1975

yen <--------

real GNP¥ ^ ^ ^ ^ ^ ^ nominal GNP¥

The RLC methodology also helps answer Firebaugh's two critiques, that most goods are not traded on the world market and that exchange rates are highly artificial and are often distorted by governments. The RLC methodology relies on measured exchange rates only to locate each country's relative income level in the base year. Any changes over time in inequality are realized through the fluctuations of domestic price and output series, not through fluctuations in exchange rates. Although the RLC methodology may fail to satisfy Firebaugh's critique as regards the parameters of the global income distribution, observed changes in those parameters over time should satisfy Firebaugh's objections. Similarly, when used in conjunction with base-year FX currency 16

Trends in International Income Inequality conversion factors, the RLC methodology should satisfy Korzeniewicz and Moran's (2000) call for a "relational indicator of command over world income." (p. 212) In theory, the international purchasing power of a real unit of currency (adjusted for inflation) should be a constant over time, and thus in theory the FX and RLC methodologies satisfy the same purposes. To the extent that annual FX conversion factors do in practice deviate from this rule, the question of which series to use (FX or RLC) reduces to a question of faith. Choosing the FX series implies a faith that the reported official exchange rates for each year do represent the actual exchange rates at which the nationals of a country were able to exercise command over the global economy in those years. Choosing the RLC series implies more faith in the accuracy of the measurement of domestic inflation over time than in the accuracy of officially reported exchange rates ("accuracy" meaning that the officially reported rates truly reflect open market rates). Given the fact that in most of the countries of the world before around 1990 (and in many of the countries of the world today) government-issued licenses are required to obtain foreign exchange at official rates, and that the allocation of such licenses is considered by the World Bank to be a major source of corruption, the accuracy of officially reported FX rates must at least be called into question.

One advantage of the RLC methodology is that it can be

implemented using as its base year that year during which exchange rates are judged to have been (historically) the least distorted. Over time, as more and more countries move from official to market exchange rates, the RLC methodology will become more and more reliable. In contrast to the FX and PPP methodologies, better exchange rate data in the future will improve the reliability of the RLC figures not only for those future years, but for all past years as well, since the reliability of all of the RLC estimates is influenced by 17

The International Structure of Income the reliability of the base year exchange rates. For the base year, a GNP/RLC inequality calculation using FX currency conversion factors would be identical to Korzeniewicz and Moran's GNP/FX measure. For other years, it would diverge only if reported exchange rates deviated from their theoretically-appropriate values, at which goods with carry identical real prices in all markets values. It is at least a reasonable suggestion that the contemporaneous conversion factors implied through the RLC methodology are better estimates of the exchange rates actually experienced by the public than are the reported official FX rates. If this is the case, then the RLC methodology is a more faithful application of Korzeniewicz and Moran's principles than is the FX methodology. At worst, from the Korzeniewicz and Moran perspective the RLC methodology (using base-year FX conversion factors) is an inferior operationalization driven by the same theoretical principle that inequality should be measured with reference to international, rather than domestic, purchasing power.

EMPIRICAL RESULTS FOR INTERNATIONAL INCOME INEQUALITY I have computed the variance of the international income distribution to measure between-country inequality for each of the thirty-nine years 1960-1998 (inclusive) using each of the three methodologies discussed above. Details on the specific series used are given in Table 1 below. Data was available for all years for all three series for a total of 82 countries, representing approximately 71% of world population in both 1960 and 1998. I operationalize international income inequality in two ways, both common in the inequality literature. First (and primarily) I use the weighted 18

Trends in International Income Inequality variance of the logged per capita incomes (VarLog) of the 82 countries for each year. In this approach, incomes are first logged so that any general percentage change in incomes across all cases does not inflate the variance. I use base 10 logarithms, and plot all three time series on the same scale. On a log scale, if all incomes grow by, say, 10% between time periods, the entire distribution would shift by 0.041 to the right, but the relative positions of the cases would not change.

Table 1. Three methodologies for international income inequality Methodology

Series used

Data source

GDP/PPP

RGDPCH (chain-indexed real GDP in 1996 constant dollars)

Heston et al 2001 (Penn World Table)

GNP/FX

World Bank 2001b NY.GNP.MKTP.CD* (current GNP in dollars deflated to 1995 using the U.S. GDP deflator)

GNP/RLC

NY.GNP.MKTP.KD** (real GNP in 1995 local currency converted to dollars at 1995 rates)

World Bank 2001b

* In approximately 4% of cases series NY.GDP.MKTP.CD (current dollar GDP) is substituted due to missing data in the primary series. ** In approximately 5% of cases series NY.GDP.MKTP.KD (constant dollar GDP) is substituted due to missing data in the primary series. As a confirmation of the VarLog results, I also calculate international income inequality using weighted Gini coefficients.

Gini coefficients are

described in more detail below (see Figure 9). 19

The International Structure of Income In absolute terms, the GNP/RLC generally returns the highest levels of inequality, and the GDP/PPP series the lowest.

The GNP/FX series is

sandwiched in between. The results appear as Figures 3 and 4 below. In the VarLogs implementation, the GNP/RLC and GDP/PPP series show a similar trend over time, although swings in the RLC series are much more pronounced. This would be expected, since the annual PPP observations from the Penn World Table are smoothed between benchmark years.

The

GNP/FX series follows an unrelated trajectory, until 1992. From 1992 forward, the trend of the FX series closely matches the trend of the RLC series. Note that this is not a constraint of the methodology: the two series are constrained to cross in 1995, but not to converge.

Any observed convergence is a real

manifestation of the data, not a mere methodological artifact.

Figure 3. International income inequality -- weighted VarLogs

20

Trends in International Income Inequality The three series, then, report dramatically conflicting trends in inequality before 1992, but, incredibly, they line up almost perfectly from 1992 onwards. This suggests that the major factors that distinguish the three methodologies are technical, not ideological, since the data underlying all three series can be presumed to more reliable for the later years. In particular, the worldwide move towards market-based exchange rates after 1990, combined with the massive growth in foreign exchange markets over the course of the decade, suggest that accurate comparative figures for the decade of the 1990s should be easier to come by than figures for earlier decades.

Figure 4. International income inequality, weighted Ginis

In the Gini implementation, the situation is somewhat different. Here, the GNP/RLC and the GNP/FX series both show a rising trend from 1960 through the early 1990s, with a small reduction in inequality since then. The 21

The International Structure of Income GDP/PPP series, on the other hand, roughly parallels its VarLogs implementation. The differences in results between the two implementations are likely due to tail behavior. When incomes are lognormally distributed, the two methods should yield identical results, since the Gini is a constant function of the VarLog in this special case.

If, however, the right tail of the income

distribution is much heavier than in the lognormal, the two methods diverge. The Gini then becomes disproportionately sensitive to changes in the far right tail. The results observed here would indicate that in the RLC series growth in the rich right tail of the international income distribution counterbalanced any catch-up efforts by the poorer countries of the world.

In the Gini

operationalization, this shows up as a flat inequality series over time; in the VarLog operationalization, which is less sensitive to this tail behavior, catch-up among the poorer countries is recorded as a decrease in measured international inequality. That these differences show up only in the RLC series suggests that measured exchange rates (whether FX or PPP) are inflated for poorer countries for the 1960s and 1970s. This interpretation is in line with the historical record. In any case, it is clear from the extreme sensitivity of the inequality results to the implementation chosen that none of the resulting trends can be taken to be definitive.

THE CHINA SYNDROME One major debating point suffusing the income inequality literature is the effect of China on the trajectory of weighted inequality over time. China's enormous impact on the trend in inequality (as opposed to merely the level) is due only in part to China's huge proportion of total world population. After all, India's population weight is almost as large. India's historical rate of economic growth, 22

Trends in International Income Inequality however, has been unspectacular (2.6% annually from 1960-1998 in real local currency), while China's has been astronomical (5.4% in real local currency). As a result, India has moved very little in the scale of relative economic wellbeing over time, while China has moved up the scale aggressively.

Since

China's income per capita is well below the world population-weighted average, Chinese economic growth implies reduced international income inequality. Because of China's gargantuan effect on measured inequality, the accuracy of the income figures used for China is of paramount importance. Unfortunately, as Korzeniewicz and Moran (2000) emphasize, no reasonably reliable income estimates exist for China, even today. Removing China from the analyses results in measured income inequality trends as depicted in Figures 5 and 6.

Figure 5. International inequality -- weighted VarLogs (China excluded)

23

The International Structure of Income Figure 6. International inequality -- weighted Ginis (China excluded)

When China is removed, movements in the GNP/LCR series closely track movements in the GDP/PPP series in both implementations (though the LCR series still measures inequality at a much higher average level). The LCR series also still coincides with the GNP/FX for the period after 1992, when exchange rates become more reliable. In the Gini implementation the LCR and FX series seem to coincide. In terms of overall trends, the VarLog LCR and PPP series now show marginally increasing inequality throughout the 1960s and 1970, with marginally decreasing inequality in the 1980s and 1990s, but with no really strong movements in either direction. The Gini series, however, are less volatile, and all show rising inequality through the early 1990s, with a leveling out or decline since then. The trajectory of the FX series, interestingly, remains fundamentally unchanged by the exclusion of China in both implementations. This is an unexpected result. It implies that, when the FX series is used, China is not, in fact, a growth outlier. In fact, Chinese growth in GNP per capita 24

Trends in International Income Inequality in real U.S. dollar terms, using contemporary FX rates to convert Chinese figures into dollars, averages only 3.0% over the study period -- similar to India's dollar growth rate over the same period (2.6% using the FX methodology). Thus, just as India has little effect on any of the inequality series due to its unexceptional historical growth rates, China has little long-run effect on the FX series.

Whether or not China's purchasing power in the world

economy has, in fact, increased at roughly the same rate as India's since 1960 is of course open to debate. Examining the various trajectories of Chinese growth over time implied by the three series helps illustrate the differences among the series. The three annual growth series are plotted in Figure 7. At the initiation of the series, the impact of the failure of the "Great Leap Forward" can be seen, with all three series beginning with strongly negative income growth in 1961, although, as expected, the effect is muted somewhat in the (smoothed) GDP/PPP series from the Penn World Table. Over the ensuing four decades, the high annual volatility of the GNP/FX series is immediately apparent. This volatility is due only in part to swings in exchange rates. In the years 1960-1971, 1973-1977, and again from 1994 to the end of the series, the Yuan was fixed to the dollar. The high volatility of the FX series relative to the LCR series during these periods can only be attributed to the deflation of nominal to real GNP using a U.S. dollar deflator (in the FX series, all annual values are expressed in same-year dollars, then deflated to 1995). The repeated sharp spikes of negative growth are what drive the FX series down to such a low period-average growth level. On the other hand, the 1994 and 1995 growth rates seem excessively large, despite a massive devaluation of the Yuan from 1993 to 1994 (which should have retarded the growth rate) and a fixed exchange rate thereafter.

25

The International Structure of Income Figure 7. Trajectory of Chinese growth since 1960 (FX, LCR, PPP)

A closer look at the behavior of the three series in the 1990s is plotted in Figure 8. In this period, when the official exchange rate can be accepted more meaningfully than in earlier decades, and when (after 1994) exchange rates are fixed, the swings in the FX series must either be accepted as being real (implying that China experienced two consecutive years of growth over 20%) or as being an artifact of deflating Chinese economic performance in U.S. dollars. Moreover, the 2.6% Chinese real (U.S. dollar) growth rate implied for 1998 seems rather low, especially considering that it is followed by a 2.0% growth rate for 1999 (not shown in the chart). Once again, it is worth noting that these low growth figures are not affected by exchange rate considerations, since the exchange rate for the period was (and remains) fixed. The GNP/FX series clearly implies a total compound growth in global purchasing power of less than 26

Trends in International Income Inequality 5% for China from 1997 to 1999. The PPP series for China also shows problems in the 1990s.

The

GDP/PPP series implies annual Chinese growth rates of only 4-6% for the period after 1993. While the annual growth rate implied by the PPP series is relatively stable over the period, it seems to err on the low side, considering the dramatic Chinese economic expansion of the late 1990s.

While the PWT

methodology is not meant to generate reliable annual time series (it is designed for international, rather than inter-temporal, comparisons), it should generate meaningful growth rates over periods as long as five years. Instead, the PPP series records a Chinese economic boom in the early 1990s, which clearly stalls later in the decade -- the reverse of what anecdotal evidence would suggest. Though certainly not conclusive on its own, this result does add to the arguments against the use of the GDP/PPP methodology, at least for the world's largest country. Perhaps the strongest arguments against the use of data from the Penn World Table for studying weighted international income inequality come from the lead researcher of the PWT himself. Commenting on the new PWT Mark 6.0 estimates for China, he states: In earlier versions of PWT we have made estimates for th PRC based upon quasi-benchmark estimates of individual researchers including Irving Kravis, who, in the early 1980s, made PPP estimates for one year, 1975. Because of its large population, much interest has attached to real product estimates for China. . . . The wide range of PPP estimates for China and the large size of their differences from the exchange rate suggest that substantial uncertainty is associated with these numbers. . . . Put another way, the basis for purchasing power estimates for China is very little improved over previous versions of PWT. (Heston 2001: 1) 27

The International Structure of Income To be fair, the dismal scientists who put together the PWT have few kind words for the official Chinese income estimates, either. In their notes to the PWT Mark 5.6, they wrote: Finally, a note for the reader on the present treatment of China's growth rates. Anyone who reads on will be struck by the very large and seemingly arbitrary adjustments (40% in some cases) that have been made in China's growth rates. It is widely felt that these growth rates are too high as is evidenced in a recent World Bank study. But politically it is convenient to have high growth rates, and associated lower inflation rates, so there has been little incentive for China to officially lower its growth or raise its inflation estimates. While we do not claim to know very accurately how to correct these growth rates we certainly know the right direction and we doubt that we can have overshot truth in the direction of understatement by more than truth is presently being overshot by official Chinese growth rates. (Summers et al 1994: unpaginated) At least one major argument can be made in support of the (official) GNP/RLCR figures over the PWT figures for computing international income inequality.

In the context of the convergence debate, the key issue is the

distribution in nations' levels of command over global resources. In this context, annual Chinese growth rates averaging around 10% in the 1990s sound about right. China may not be growing at that rate, but it seems like China's command over the global economy is.

Chinese government policy may be to buy

international economic clout at the expense of impoverishing the Chinese rural population, but whatever the source that clout is clearly emerging, and has been throughout the 1990s. Thus, it may be that the PPP and LCR figures tell the same story from different perspectives -- that of China as a whole and that of 28

Trends in International Income Inequality China's international face, respectively. This interpretation received additional support from Li et al (1998), who report that income inequality within China has increased faster since 1947 than for any other country in the world. (p. 27) If the two-Chinas explanation is correct, the LCR is the proper series to use in describing the trajectory of international inequality, China included.

Figure 8. Detail of Chinese growth since 1990 (FX, LCR, PPP)

Whatever the cure to the China syndrome, the remarkable parallelism of measured inequality using the LCR and PPP series when China is excluded and the near coincidence in measured inequality in the 1990s using the LCR series and the FX series argue well for the reliability of the LCR series.

The

GNP/RLCR series implies that inequality was relatively stable over the period 1960-1998 in the world x-China, with a downward trend in inequality in recent 29

The International Structure of Income years. With China included in the analysis, weighted inequality has declined precipitously, but only due to China's expanded role in the global economy, surely a reasonable conclusion given that caveat. Incidentally, between-country inequality not weighted by population (results not shown here) has risen consistently over the period under all three methodologies. This may account for some of the anecdotal evidence of the increasing impoverishment of many parts of the world as the U.S. has prospered: with weighted inequality, the impoverishment of small countries can go unnoticed. For example, the population of China is larger than that of the continents of South America and Africa combined.

Weighted international

inequality is largely determined by the relative trajectories of China, India, and the West.

It may not correspond to common-language notions of what

"inequality" should mean.

PROBLEMS IN INTERPRETING SUMMARY MEASURES OF INTERNATIONAL INEQUALITY On that note, a strong argument can be made that the entire income inequality debate is based on a false premise. As illustrated in Figures 3 and 4, the debate has centered on the behavior over time in summary measures of the spread in incomes between the countries of the world. Summary measures, however, may obscure the more meaningful mechanics operating just below the surface. As of 1998, China's GNP per capita (RLC methodology) was $708. The global mean of the logged GNP/RLC per capita was 3.15 in 1998, equivalent to $1413. At historical trend growth rates, China will overtake the mean of the logged series in about two decades. Up till that point, economic growth in China will effectively reduce measured world income inequality. 30

Trends in International Income Inequality Then comes the great reversal.

Around the year 2020, China will

surpass the mean logged income level. From that point on, if China continues to grow faster than the global average, it will be effectively increasing global income inequality as measured here. To take another case, Guatemala in 1998 recorded a GNP/RLC per capita of $1514, just above the global mean of the logged series. Guatemala is not a rich country. Yet if Guatemala were to achieve growth rates higher than the global mean, according to accepted measures this growth would make the world more unequal. Guatemalans might respond "if that is inequality, please do give us more of it." Similarly, higher than world average population growth rates in countries close to the mean of the logged income series effectively reduce measured income inequality. Thus, more births in Egypt, Jamaica, and the Philippines would reduce measured inequality, while more births in the United States and Europe would increase inequality.

Moreover, the emigration of

economic refugees from poor countries to rich countries increases measured inequality, by increasing the population of the rich countries. Interestingly, the steady relative economic decline of Argentina since 1960 has helped reduce global inequality, as Argentina slowly recedes toward the mean world income. Clearly, summary measures of inequality do not lend themselves to the natural-language implication that more inequality is a social bad. It may be a moot point whether, following Firebaugh: [T]he centuries-old trend of rising inequality leveled off from 1960 to 1989. [Thus] the dependency theory thesis of a polarizing world system received no support. (Firebaugh 1999: 1597) 31

The International Structure of Income or whether, as Korzeniewicz and Moran conclude: [E]fforts to account for patterns of development in the world-economy must include polarization as a crucial component of recent trends. Findings in this respect are rather clear. The world distribution of income became more unequal over the 1965-1990 period, and inequality accelerated during the 1980s. (Korzeniewicz and Moran 1997: 1031) Both Firebaugh and Korzeniewicz and Moran consider international income inequality to be a measure of the "polarization" of the global economy. If explosive growth in Guatemala would be polarizing, this would be true. If the massive economic success of the four Asian "tiger" economies has been polarizing, then this would be true. Polarization would simply mean an increase in the measured dispersion of income across countries, however counterintuitive the implications. On the other hand, if by "polarization" one means the emergence of a barrier between the rich and the poor countries of the world that is increasingly difficult to cross, conventional summary measures of income inequality are largely irrelevant. Since this is the context in which Firebaugh, Korzeniewicz and Moran, and other authors in the development literature use the term, better measures are obviously necessary to resolve the debate. A profound argument can be made against any use of summary measures for global income inequality. All summary measures, such as the variance of the logged income used above and the commonly used Gini and Theil coefficients, are meaningful only insofar as they describe the distribution of a variable (income) among a population (the population of the world). Their validity is ultimately predicated on the assumption that the population of the world constitutes, statistically, a single population with respect to the 32

Trends in International Income Inequality distribution of income. In the literature this basic assumption has (naturally) been made implicitly, and as a result has never been verified empirically. We now have the data and the computing power to perform this verification. To do this it is necessary to chart the course, not just of a summary measure of inequality, but of the aggregate global income distribution in its entirety. This can now be done over a four-decade time period, 1960-1999. Doing so may lead participants on both sides of the income inequality debate to question their conclusions.

33

The International Structure of Income

34

3. From International Inequality to Global Inequality It is shocking that until 2002 we really had no firm empirical grasp of the shape of the global distribution of income. In that year, Milanovic (2002) published estimates of the global income distribution for two years, 1988 and 2003, based on data from 91 countries. The following year I presented similar estimates based on worldwide panels for 1960-2000 (Babones 2003). In this chapter, I describe in further detail the methodology I used then for mapping the aggregate global distribution of income.

The goal of this effort is to move beyond

summary statistics for international inequality to view the full distribution of income in the world, treating persons rather than national populations as the cases to be analyzed. This moves us out of the old paradigm of summary measures and into a new paradigm of full information on global income inequality. Below I estimate the distribution of world income for the four decades of the 1960s, 1970s, 1980s, and 1990s.

I begin by developing a parametric

methodology for aggregating the panoply of raw income distribution available for the countries of the world into a unified global income distribution. I then apply this methodology to producing four snapshots of the global income

The International Structure of Income distribution and analyze and interpret the results. I conclude the chapter by discussing the implications of the results of this chapter for research on international inequality.

METHODS FOR UNCOVERING THE GLOBAL DISTRIBUTION OF INCOME In order to map the aggregate global distribution of income it is first necessary to know the full distribution of income in countries accounting for substantially all of the population of the world. Most studies of national income inequality report not full distributions, but a summary measures, and all compendia of national income distributions by necessity use summary measures. Converting such summary measures into reasonable estimates of the actual distributions of income in a large number of countries is a multi-step process. First, summary measures of national income distributions must be selected, and all nonparametric summary measures (such as the Gini coefficient or the quintile distribution) must be converted into parametric measures (such as mean and variance). Second, study time periods must be chosen and estimates of income inequality made for countries that are missing data for those time periods. Third, the resulting observed and estimated parametric summary measures must be applied to an assumed functional distribution of income to yield a synthetic sample of representative incomes for each country in each time period. The full global distribution of income is simply the re-aggregation in one pool of all of the disaggregated national income distributions. Details on the operationalization of each of the three steps in aggregating the global distribution of income are given in the three sub-sections below.

36

From International Inequality to Global Inequality Sources and Methods for National Income Distributions The most comprehensive pooled source of data on national income distributions available at the time these analyses were undertaken was Deininger and Squire's (1996) massive collation from primary sources of 2,621 observations on income inequality for 108 countries for various time points, mainly covering the period 1960-1995. (More recently these data have been incorporated into the fuller World Income Inequality Database produced by the World Institute for Development Economics Research). Deininger and Squire rate 682 of these observations as "high quality," meaning that they were based on household surveys with comprehensive coverage of the population and comprehensive measurement of income or expenditure (pp. 567-571).

The

database has since been expanded (Deininger and Squire 2001) to 2633 observations, of which 693 are classified as high quality. Deininger and Squire, however, do not provide the full distribution of income for the countries in the database. They could not have provided such information, since their sources do not. The standard summary measure of income inequality in the literature, and that adopted by Deininger and Squire, is the Gini coefficient. The Gini coefficient derives from the Lorenz curve depiction of the distribution of income. The Lorenz curve is a graphical representation of the cumulative distribution of a society's population, ordered as to income. An illustrative example of a Lorenz curve is plotted in Figure 9. The solid line represents the Lorenz curve, which by construction will always be concave up. The dashed line represents the limiting case of the Lorenz curve, perfect equality in incomes among all observational units in a society.

37

The International Structure of Income Figure 9. Illustration of a Lorenz curve

Area "P"

Area "Q"

With reference to Figure 9, the Gini coefficient can be expressed as: Gini = P / (P + Q) Since (P + Q) is, by construction, equal to 1/2, the Gini coefficient can be reduced to: Gini = P / (1/2) = 2P The Gini coefficient is a non-parametric summary statistic bounded by 0 and 1 that is strongly Lorenz-consistent (Fields 2001: 31-32), which means that any Lorenz curve that falls completely to the left of a another Lorenz curve will have a higher Gini coefficient. Since they are by nature non-parametric statistics, Gini coefficients alone cannot be used to reconstruct the actual distributions of income within nations. However, where data were available Deininger and Squire also recorded the income shares of each ranked quintile of the population (950 observations, of 38

From International Inequality to Global Inequality which 621 are high quality). Quintile shares, of course, are also non-parametric measures, but they can be used to recover actual grouped figures from the true distribution.

This grouped data can then be used to estimate parametric

summary statistics describing the national income distribution. The quintile data give the proportion of national income controlled by each of five evenly-sized groups in society. Representing the proportion of national income in each ranked quintile, poorest to richest, as Q1 through Q5, total national income with the symbol Y, and population with P, the implied mean income per capita in each quintile can be written as: Mi = (Y * Qi) / (P / 5) That is, the mean income per capita for each quintile is equal to that quintile's share of national income divided by its share (1/5) of the population. This simple procedure spreads observed national income out over five summary groups of equal population size. Although it would be preferable to have more than five pooled observation points, five is sufficient for computing a variance statistic. Five pooled observation points are in fact sufficient to give a very close approximation of an actual income distribution. In Figure 10, I have reproduced the sample Lorenz curve from Figure 9 (solid line). Superimposed on the same graph, I have plotted the Lorenz curve for the same population, grouped into quintiles (dashed line). Each quintile has been represented solely by the mean income level within the quintile. It is clear from Figure 10 that the resulting five-point distribution closely approximates the true (quasi-continuous) distribution. Any residual error does not lead to a gross misrepresentation of the shape of the income distribution.

39

The International Structure of Income Figure 10. Quintile-based (dashed) versus actual (solid) Lorenz curves

Given the quintile means, it is a simple matter to compute an estimate of the variance of the logarithms (VarLogs) of the national income distribution for each of the 950 cases for which Deininger and Squire report quintiles. I first logged the implied average income (per capita) levels for each income quintile observations for each country (base 10), then computed the mean logged income across the five quintiles. Using this mean, I computed the variance and standard deviation of the logs for each country, using the n=5 quintile means as if they were five individuals composing a population. Working with the standard deviations of the logs, I aggregated the results for each country over each of the four decades, 1960-1969, 1970-1979, 19801989, and 1990-1998. National income inequality is extremely stable over time, changing glacially if at all. (Li et al 1998) For 32 of 49 countries in their study, internal inequality did not measurably change over the period 1947-1994, while in an additional 10 cases trends were statistically detectable, but minimal in 40

From International Inequality to Global Inequality magnitude. The pooling of cases into decade averages produces reasonably continuous series (most countries do not have an observation in the Deininger and Squire database for any given year), while leaving sufficient scope to capture trends in national inequality over time. Separating the 621 high-quality observations from the other 329 cases, I computed the simple average of all high-quality cases of the standard deviations of logged income for each country for each decade. I then separately computed the average of the low-quality observations. In decades where high-quality observations were available, I used the average of the high-quality series in further analyses. In decades where there were no high-quality observations, I used the average of the low-quality observations. In total, there was enough data to compute decade-average standard deviations for 43 countries in the 1960s, 51 in the 1970s, 65 in the 1980s, and 60 in the 1990s. While this is a substantial number of countries for each time period, only 20 countries in the Deininger and Squire database yield at least one standard deviation estimate in each of the four decades. More observations are clearly necessary than can be obtained from the quintile data alone. The obvious source for additional information is the Gini data provided by Deininger and Squire for almost three times as many cases as the quintile data. There is, however, one major obstacle to using this data: the Gini coefficient is completely nonparametric. Unlike the quintile data, it does not implicitly embed any actual income data. The Gini route would seem to be a blind alley. In fact, however, there is a way out. Deininger and Squire report corresponding Gini coefficients for all 950 cases for which income quintiles are available, as well as for an additional 1671 cases. I computed decade-average Gini coefficients in the same manner as the decade-average standard deviations of logs, separating out and giving 41

The International Structure of Income precedence to high-quality observations. I paired each decade-average standard deviation of logs with its corresponding decade-average Gini. The results are graphed in Figure 11.

Figure 11. Observed Ginis versus SDs of logs, "high quality" data

Clearly, the relationship between the standard deviation of logged income and the Gini coefficient in the observed range is very nearly linear, with over 90% of the variance explained. The correlation between the two series is greater than .95, virtually perfect. The two variables seem almost perfectly substitutable, at least within the observed range (at higher levels they must diverge, since the Gini is bounded at 1.0). Why should the two series be so closely related? theoretical justification for the empirical result.

There is some

As noted above, the Gini

coefficient is strongly Lorenz-consistent; the standard deviation of logged 42

From International Inequality to Global Inequality income is not. If, however, in all cases the distribution of income were to take the same functional form, the standard deviation of the logs would be Lorenzconsistent. Thus, under the assumption that the distribution of income in all cases was of the same functional form, the standard deviation of the logs and the Gini coefficient would both be strongly Lorenz-consistent. This means that any increase in the standard deviation of logged income, holding the functional form of the income distribution constant, entails a shift to the right in the Lorenz curve. Since the Gini coefficient is also strongly Lorenzconsistent, such a shift would necessarily imply an increase in the Gini coefficient.

Thus, there will be a monotonic relationship between the two

measures. To emphasize, this is only the case if a constant functional form for the income distribution is assumed. Though monotonicity between the two measures is thus theoretically reasonable, linearity is pure luck. One distribution that is commonly used for modeling national income distributions is the lognormal distribution, which "describe(s) many empirical frequency distributions closely over the whole range of incomes." (Creedy 1985: 4) In fact, many if not most national income distributions are visually (though not necessarily statistically) indistinguishable from the lognormal. The lognormal distribution is the equivalent of a normal distribution drawn on a logarithmic scale. In addition to its good empirical fit to actual incomes, several theoretical arguments nave been made that incomes within a single, interacting population should fall into a lognormal distribution. (Aitchison and Brown 1957: Chapter 11; Cowell 1995: 77-87; Champernowne and Cowell 1998: Chapter 11). The lognormal distribution includes as special cases what are sometimes called "diamond-shaped" and "triangular" distributions of income. When the standard deviation of the lognormal distribution is very large relative to the 43

The International Structure of Income mean, the resulting income distribution is "triangular," with many poor, no distinct middle class, and a few very rich. When the standard deviation is moderate relative to the mean, the income distribution is more "diamondshaped,: with a medium number of poor, a large middle class, and a few rich. At very low standard deviations relative to the mean, the lognromal distribution approaches the normal distribution, in which the number of rich equals the number of poor. A typical lognormal distribution of income (mean $1000) in a population of 100 million is plotted in Figure 12

Figure 12. Illustration of a lognormal distribution of income

I have computed numerically the relationship between the standard deviation of the logarithms and the Gini coefficient under the assumption of lognormality. The mean of the distribution is irrelevant to the relationship (Aitchison and Brown 1957: 113) The results are shown in Figure 13. For standard deviations between .2 and .7, the relationship is almost perfectly linear. 44

Conveniently, all of the standard deviations in the observed

From International Inequality to Global Inequality sample of decade-average standard deviations fall within this range.

The

observed linear relationship between the standard deviation of logged income and the Gini coefficient for the 193 "high-quality" cases graphed in Figure 10 is: SD logs = .824(Gini) -- .018 I use the observed (rather than theoretical) relationship, since the observed relationship carries only the weak assumption that all national income distributions follow the same functional form; the theoretical relationship is based on the strong assumption that the functional form is specifically lognormal. Figure 13. Gini and SD of the logarithms for a lognormal distribution

Applying the empirical relationship to those cases for which decadeaverage standard deviations are not available (but Gini coefficients are) yields an additional 19 cases for the 1960s, 21 for the 1970s, 17 for the 1980s, and 16 45

The International Structure of Income for the 1990s. In these totals I include only those cases for which GNP/LCR per capita is also available, since the level of inequality must be combined with the income level to give a full distribution. This brings the number of countries for which a full four-decade observed inequality data series and World Bank GNP per capita data are available up to 37. I will refer to this 37-country, four-decade data set as the Tier I sample.

Estimating Missing Cases Clearly, it would be desirable to conduct analyses using more countries, although the 37 countries in the Tier I sample do account for around 45% of world population. In the construction of the Tier I sample only 37 countries were found to have inequality observations for all four decades, but standard deviation of logged income estimates for at least one decade of the four are available for an additional 51 countries. Relying on the relative stability of national income inequality over time (Li et al 1998), it should be possible to reasonably estimate income inequality for those 51 countries. Among these 51 countries, 25 have data for three decade-averages, 16 have data for two, and 10 have data for just one, for an average of 2.3 observations per country. If this is accepted as a reasonable basis for linear interpolation/extrapolation, modeled decade-averages can be computed for all 51 of these countries (countries with just one observation have that observation applied to all four decades). While this would not be an acceptable means of generating missing data for the study of trends in inequality within countries, minor errors in the levels of national inequality will not substantially affect the resulting global distribution of income, for two reasons. First, the differences in income levels between countries are much larger than those within countries, meaning that the possible levels of error in within-country are dwarfed by the 46

From International Inequality to Global Inequality level of between-country inequality. Second, the global distribution of income is an aggregate of many (if these cases are accepted, 88) national distributions. Extrapolation/interpolation error in the spreads of national income distributions will tend to cancel each other out as those national distributions are aggregated to a global distribution. I will refer to the set of 88 countries for which observed, interpolated, or extrapolated decade-average standard deviations (and associated income per capita) can be found for all four decades as the Tier II sample. The Tier II sample comprises approximately 90% of world population. This is the sample that I will primarily use in estimating the global income distritution. Using regression models, it is possible to generate some estimate of the level of inequality in the remaining countries of the world.

There are an

additional 20 countries, not in the Tier II sample, for which GNP/RLCR data is available. covariates.

Inequality estimates for those countries can be made based on One such covariate is income per capita.

National inequality

(measured as the decade-average standard deviations of the logs) is quadratically related to the log of income per capita, though the relationship is far from reliable, showing an R2 of only .220 for the 352 decade-averages (4 per country) in the Tier II sample. The relationship is graphed below as Figure 14. It follows the classic quadratic Kuznets curve of relatively high inequality in middleincome countries. National inequality is also strongly patterned by geography and somewhat trended over time. Mean inequality levels by decade and region for the 88 countries in the Tier II sample are shown in Table 2 below. I use the seven standard regions defined by the World Bank (2001).

47

The International Structure of Income Figure 14. Fit of national inequality to GNP per capita

Table 2. National inequality by decade and region (SD of Logs) Region (n) Developed (26) East Asia (7) E. Europe (1) Latin Am. (22) Mid. East (4) South Asia (5) Sub-Sah. Afr. (23) All countries (88)

1960s 0.30 0.29 0.18 0.39 0.32 0.32 0.41 0.35

1970s 0.28 0.30 0.17 0.40 0.35 0.27 0.39 0.34

1980s 0.27 0.31 0.17 0.38 0.29 0.24 0.36 0.32

1990s 0.26 0.32 0.23 0.38 0.27 0.19 0.35 0.32

All years 0.28 0.30 0.19 0.39 0.31 0.25 0.38 0.33

National inequality levels are highest in Latin America and Sub-Saharan Africa, while in south Asia they are exceptionally low. There is also a slight trend towards lower within-nation inequality over the study period. Nonetheless, an ANOVA model of inequality by decade and region returns an R2 of only .383. Adding the log of income per capita as a quadratic covariate 48

From International Inequality to Global Inequality only raises this to an R2 of .419. Although the F statistics for both of these ANOVAs are highly significant, the poor overall levels of fit militate against using these models to estimate missing values. Nonetheless, it is useful to have values, even poor ones, for the additional 20 cases. I have estimated values for these cases based on decade and region. Adding these 20 countries to the Tier II sample yields a Tier III sample of 108 countries, comprising all of the population of the world living in countries for which income data is available.

I will use this sample for

comparative purposes only. By way of review, a summary of the origins of the cases contained in the Tier I, II, and III samples is presented in Table 3. Every country in a sample has exactly one case (observation) for each of four decades. The origin of each case is broken down by the following case origin codes: QUINT-HQ: Observed income quintiles converted to standard deviations of logs (high quality). QUINT-LQ:

Observed income quintiles converted to standard deviations of logs (low quality).

GINI-HQ:

Observed Gini coefficients converted to standard deviations of logs (high quality).

GINI-LQ:

Observed Gini coefficients converted to standard deviations of logs (low quality).

LIN-FIT:

Standard deviations of logs interpolated or extrapolated from observed values for the same country from other decades (fit on the basis of both high and low quality observations).

GEO-YEAR: Standard deviations of logs estimated based on decade and geographical location using a regression model (fit on the basis of both high and low quality observations). 49

The International Structure of Income Table 3. Summary of the origins of cases for national inequality samples Tier I sample (37 Countries)

Tier II sample (88 Countries)

Tier III sample (108 countries)

Origin of case QUINT-HQ QUINT-LQ GINI-HQ GINI-LQ Total

# Cases 106 13 10 19 148

% 72 9 7 13 100

Origin of case QUINT-HQ QUINT-LQ GINI-HQ GINI-LQ LIN-FIT Total

# Cases 175 26 19 45 87 352

% 50 7 5 13 25 100

Origin of case QUINT-HQ QUINT-LQ GINI-HQ GINI-LQ LIN-FIT GEO-YEAR Total

# Cases 175 26 19 45 87 80 432

% 41 6 4 10 20 19 100

Over half of the cases in the Tier II sample are based on actual highquality observations in the Deininger and Squire database. In addition, the bulk of the linear fit cases are ultimately based on high-quality original observations. The main weakness of the Tier II sample is the paucity of observations for eastern Europe; as noted in Table 2, only one country from eastern Europe (Hungary) is represented. The Tier III sample does not improve this situation, since it merely applies the observed Hungarian inequality levels to the 50

From International Inequality to Global Inequality remainder of eastern Europe. Thus, the Tier II sample seems the best available compromise solution for studying a constant set of countries over several decades.

AGGREGATING NATIONAL DISTRIBUTIONS TO A GLOBAL DISTRIBUTION Having developed parametric summary measures of the national income distributions of a sufficiently large sample of countries, the next step in charting the global income distribution is to disaggregate the national income distributions into simulated populations of individuals, which can then be reaggregated into a unified global distribution. Since full national distributional data is not available on a wide cross-national basis, this requires the application of summary parametric statistics to an assumed functional form for each national distribution. The lognormal distribution is both a good empirical fit to most actual income distributions (Creedy 1985: 4) and computationally convenient.

Using the means and standard deviations of logged national

incomes derived above, the full distribution of income for each country in the Tier I, II, and III samples can be simulated. To simulate the distribution of income in each country, I used a gaussian kernel with mean and standard deviation equal to the mean and standard deviation of the logarithms of the national income distribution. This results in a simulated income distribution that is normal on a logarithmic scale, or a lognormal distribution.. For the Tier II sample, this results in a total of 88,000 "clumps" of population, with clumps ranging in size from a billion persons each for China to a few hundred for Fiji or Lesotho. The purpose is to spread each country's 51

The International Structure of Income population in small increments over a common (logarithmic) scale.

The

gaussian kernel ensures a smooth simulated distribution, as illustrated in Figure 15 below. In Figure 15 I plot an example of using a gaussian kernel (on a logarithmic scale) to represent the United States household income distribution from 1998. Though not a perfect representation of US household incomes, the guassian kernel representation is close enough to reality for use in compiling the global income distribution.

Figure 15. Illustration of a gaussian kernel (United States) U.S. Distribution of Household Income (modeled as lognormal)

Frequency

$29,021

$0

$20

$40

$60

$80

$100 $120 $140

$160 $180

$200 $220

$240 $260 $280 $300 $320

$340 $360 $380 $400

Income (nominal scale '000s)

Moving from smooth country curves like that depicted in Figure 15 to a full global distribution of income requires only that the relevant country curves for all countries of the world be aggregated into a single global distribution.

52

From International Inequality to Global Inequality

RESULTS Strictly speaking, the graphs that result from applying the methodology outlined above represent the aggregate global distribution of population by income level, although "global distribution of income" serves as a useful shorthand. Using this methodology, I have aggregated the national income distributions of the 88 countries in the Tier II sample into global income distributions for each of the four study decades. The resulting global income distributions are plotted below in Figures 16-19. Similar plots for the Tier I and Tier III samples and for samples of all available Deininger and Squire observations by decade are not shown. Results for other samples are near-identical to those reported here. Note that the standard deviation estimate used for China for the 1960s is almost certainly too small, as it results from the extrapolation of a linear trend back in time from the observed 1990s and 1980s values.

A larger initial

standard deviation for China would have resulted in a more rounded primary peak for the 1960s. Figures 16-19 reveal a clearly bimodal global distribution of income for all four study decades. This is a shocking result: nothing in income distribution theory or neoclassical economics more generally would lead one to suspect a bimodal global distribution of income. Moreover, the secondary (high-income) mode of the distribution is roughly twice as high as the trough between the modes.

This suggests that perhaps the data are not drawn from a single

population with a bimodal distribution but from two distinct populations. Some support for this conjecture can be drawn from the fact that the two distinct population clusters in the global income distribution are each roughly normally distributed (on the log scale).

53

The International Structure of Income Figure 16. Global distribution of income, 1960s (Tier II sample)

Figure 17. Global distribution of income, 1970s (Tier II sample)

54

From International Inequality to Global Inequality Figure 18. Global distribution of income, 1980s (Tier II sample)

Figure 19. Global distribution of income, 1990s (Tier II sample)

55

The International Structure of Income The global income distributions plotted in Figures 16-19 would be very roughly consistent with a model that posits the existence of two distinct economic populations in the world, each possessing a lognormal income distribution. Where the income distributions of the two populations overlap, a "shoulder" results, linking the two distributions. Relatively poor people in the high-income population would earn incomes that overlap with those of relatively rich individuals in the low-income population. Incidentally, such a model would provide a fruitful foundation for analyzing the distinction between "relative" and "absolute" poverty. Is this two-population model truly consistent with the results reported above? It is possible to estimate a two-population model from the empirical aggregated income distribution.

Assuming lognormality of the income

distribution for each of the two populations, the mean of each (logged) population distribution can be estimated with the mode of the empirical distribution. Estimating the standard deviations is more problematic, since the two distributions overlap.

However, using the symmetry of the normal

distribution, it would be possible to split each distribution at the mean and estimate its standard deviation from just its extreme tail (the left tail for the lowincome population and the right tail for the high-income population). Given the means and standard deviations of each income distribution, they could be plotted separately and in aggregate, and superimposed on the observed aggregate distribution. I have done exactly this in Figure 20 for the 1990s income distribution (Tier II sample). I chose the 1990s distribution because real national income figures in dollars are most reliable for the 1990s. The 1990s sample also has the highest percentage of "high-quality" inequality data of the four decades. Figure 20 does suggest that the empirical data are relatively consistent with the two56

From International Inequality to Global Inequality population model, though the model distribution is underweight in the "shoulder" between the two modes. Similar plots for the previous three decades (not shown) fit less well with each decade further back in time. It should be noted that the use of PPP rather than RLC data would vastly improve the fit of this model, since the (log scale) distance between the two modes is much narrower (and thus the shoulder firmer) using PPP data. In any case, the shape of the distribution of income within each of the two conjectured populations is very nearly normal -- giving strong circumstantial support for the model

Figure 20. Two-population model of the global income distribution

Whether or not the two modes of the global distribution of income are accepted as representative of the means of separate populations, the mere fact that the global income distribution is strongly bimodal invalidates much of the reasoning implicit on both sides of the income inequality debate. None of the 57

The International Structure of Income summary measures of income inequality (Gini, VarLog, etc.) has much intuitive meaning outside the context of a single, unimodal income distribution, although their analytical meanings still apply. However, the absolute regularity of the bimodal pattern does suggest a new way to approach the international income inequality debate: the distance between the two modes of the global income distribution could be used as a rough indicator of the degree of global inequality. Low- and high-income modes and the differences between them are summarized in Table 4 below. The steadily declining intermodal distance on the logged scale is consistent with the strong decline in GNP/RLC inequality as measured by the variance of the logarithms of incomes. It is also consistent with the decline in the Gini coefficient. As illustrated in Figure 13, the Gini coefficient is roughly linearly related to the VarLog at lower levels. Thus, although it is formally nonparametric, empirically the Gini coefficient is relatively consonant with changes in the distribution of income on the log scale, rather than the absolute.

Table 4. Modes of the global income distribution (logged) 1960s 1970s 1980s 1990s

Low mode 1.9 2.2 2.4 2.7

High mode 4.0 4.2 4.2 4.3

Difference 2.1 2.0 1.8 1.6

On an absolute scale the trend in the intermodal distance is strikingly different. Converting the logged modes to absolute dollar figures yields the values given in Table 5. In real dollar terms, the low-income mode rises by a factor 6.3 from the 1960s to the 1990s, while the high-income mode just about doubles. Thus, in geometric terms, the low-income mode is catching up to the 58

From International Inequality to Global Inequality high-income mode rather quickly (though at historical rates, even assuming no diminishing gains, it would still take 130 years for the two modes to reach parity) . Nonetheless, in absolute terms the two modes are diverging -- sharply. The absolute difference between the two modes has nearly doubled over the four decade period.

Table 5. Modes of the global income distribution (1995 dollars) Low mode $79 $158 $251 $501

1960s 1970s 1980s 1990s

High mode $10,000 $15,849 $15,849 $19,953

Difference $9,921 $15,690 $15,589 $19,451

In conclusion, while international income inequality at the national level may have declined since 1960, it is patently not true that world has become less polarized. In fact, the rich are getting much richer while the poor get . . . slightly richer. The absolute gap between the rich and the poor has never been wider than it is right now.

DISCUSSION AND INTERPRETATION The results presented above represent an intriguing mix of confirmations and contradictions of existing theories. Clearly, the gap between the global rich and poor, measured in absolute terms, has increased dramatically over the past forty years. This confirms the conclusions of Korzeniewicz and Moran (1997), and indeed the general orientation of much of radical sociology. At the same time, I suggest that Korzeniewicz and Moran's methods are somewhat flawed, and find that measured "inequality" (using the measures common in the literature) has 59

The International Structure of Income been declining over most of the period 1960-1999.

This suggests that

"inequality" may not be what most of the literature assumes it to be. Measures like the Gini and the VarLog, developed for measuring income inequality among individuals within a population, may not be appropriate for measuring income inequality across nations. One major argument against using standard within-population income inequality measures to study international income inequality is that the residents of the world, taken together, do not constitute a single unified "population" from the perspective of income. Instead, we find that there are two empirically distinct locations in the global income distribution, a low-income population and a high-income population, with a non-arbitrary dividing line separating the two. While there seem to be more medium-income individuals in the world than would be predicted by the overlap in two-population model, the fit of this model to the data is remarkably good, considering its simplicity. Moreover, the excess of middle-income individuals may be a methodological artifact: errors at every stage in the process of determining the global distribution of income would tend to accumulate "in the middle," since they would presumably be random with regard to the two-population model. A theoretical basis for the two-population model can be found in Arrighi and Drangel (1986). Arrighi and Drangel developed a theory of core-periphery hierarchy based not on countries, but on nodes of activity within global commodity chains. In this model, economic actors continuously endeavor to shift, and some succeed in shifting, the pressure of competition from themselves onto other actors. As a result, the nodes or economic activities of each and every commodity chain tend to become polarized into positions from which the pressure of competition has been transferred elsewhere (core-like activities) and 60

From International Inequality to Global Inequality positions to which such pressure has been transferred (peripheral activities). It follows that aggregate rewards in peripheral activities will tend to approach levels of remuneration that are only marginally higher than what the factors of production engaged in them would collectively fetch outside the overarching world division of labor. In contrast, aggregate rewards in core-like activities will tend to incorporate most if not all of the overall benefits of the world division of labor. (Arrighi and Drangel 1986: 17) The Arrighi-Drangel model implies, to some extent, a dichotomous distribution of world income, which is consistent with the results presented above. On the other hand, Arrighi and Drangel do not extend this line of reasoning to argue that the countries of the world should fall into a dichotomous set of income zones, with core-type activities concentrated in high-income countries while peripheral-type activities concentrate in low-income countries.

Quite the

contrary: while admitting that such concentrations are likely to emerge (pp. 2226), they argue for the existence of a third group of states, the "semiperiphery," in which a mix of core- and periphery-type activities coexist side by side (pp. 26-30). It is the purpose of the next chapter to explore the empirical basis of this conjecture.

61

The International Structure of Income

62

4. The Structure of National Income A key contribution of Wallerstein's (1974) Modern World System, Volume I is its identification of three broad zones in the world-economy, the core, the semiperiphery, and the periphery. Embracing this structured view of the worldeconomy, several groups of scholars have attempted to properly allocate the countries of the world among the three Wallersteinian zones. These attempts have in general been successful, with three-zone structures emerging from network analyses of patterns of trade, network analyses of economic, political, and military relationships, and distributional analyses of income levels. All of these analyses, using very different methods and data, yield roughly similar groups of countries for each of the three structural zones. This tends to confirm the basic validity of the model of a world-economy divided into three structural zones Given that the existence of a three-zone structure to the world-economy has been several times confirmed by a wide variety of studies, it is indeed ironic that we do not possess a widely-accepted, up-to-date set of benchmarks for position in the world-economy. Network-based methods are too data-intensive to be updated conveniently, require a high degree of technical expertise, and (due to data constraints) typically cannot be applied to a sufficiently large sample of countries to create true "benchmark" series.

Moreover, it is

The International Structure of Income impossible to extrapolate the results of network studies out-of-sample, meaning that network-derived benchmarks can not be applied to new countries as data becomes available.

To top it all off, the data required for network-based

methods generally becomes available several years after the study period itself. Income-based methods are more promising on all of these criteria. They are not very data-intensive; they require little specialist knowledge; they can be applied to most of the countries of the world for which statistics of any kind are reported. Benchmark boundary points in the global income distribution can be applied out-of-sample with little loss of validity, especially if the income level of the country in question is not very near a boundary point. Income-based methods carry the additional advantage that they can be updated annually, usually about 18 months after the end of the study period. However, the two existing income-based trichotomizations of the worldeconomy (Arrighi and Drangel 1986; Korzeniewicz and Martin 1994) suffer from some drawbacks as benchmarking studies. They eyeball troughs between zones, rather than establishing a consistent definition that can be automatically applied. Also, they both use all available countries for each time period that they study, reducing the inter-temporal comparability of their zonal boundaries. They do not report annual results and, of course, they are both now rather outof-date. Major improvements in the income-based approach are possible, which would give the resulting benchmarks much wider and simpler application. To be fair, computational methods are widely available today that were not conveniently available to either Arrighi and Drangel or even Korzeniewicz and Martin. In this chapter I develop a new set of income-based benchmarks for the zones of the world-economy. I open this chapter with a brief review of the theoretical literature on the classic three-zone partition of the world economy. 64

The Structure of National Income In the next section, I review the major approaches taken in the literature to delineating the three theorized tiers of the world-economy. In the third section of this chapter, I develop a refined methodology for partitioning the worldeconomy, based on the income approach. In the fourth section of this chapter, I report the results from my income-method partitioning of the world-economy. In this section, I detail trends in inequality within and between zones. I close this chapter with a post-script in which I revisit the regression models of convergence found in the economics literature, adding a structural dimension drawn from world-systems theory.

THEORY OF A THREE-TIERED WORLD-ECONOMY At least since Wallerstein's (1974) seminal work on the origins of the modern world-system, mainstream sociologists have recognized the long-standing existence of a relatively stable structure to the international system of states. Wallerstein defined a world-system as a social system, one that has boundaries, structures, member groups, rules of legitimation, and coherence . . . life within it is largely self-contained, and . . . the dynamics of its development are largely internal. (p. 347) According to Wallerstein, there have historically existed two types of worldsystems:

world-empires

and

world-economies.

World-empires

are

characterized by the organization of an entire world-system under a single state structure, while world-economies are characterized by the existence of competing states within the system. Today's modern world-system is postulated to be a world-economy, comprising all of the countries and areas in the world today. 65

The International Structure of Income Structure of the Post-WWII World-Economy In Wallerstein's model, all world-systems are structured around a system-wide division of labor: This division is not merely functional -- that is, occupational -- but geographical . . . it is a function of the social organization of work, one which magnifies and legitimizes the ability of some groups within the system to exploit the labor or others, that is, to receive a larger share of the surplus. (p. 349) The modern world-economy is no exception. Wallerstein (1974: 347-357, 1979: 66-73) and theorists following in the Wallersteinian tradition (e.g., Chase-Dunn 1998: 201-214; Arrighi & Drangel 1986: 9-30) have recognized a strong international component to the functional division of labor in the modern worldeconomy. They claim that countries at the "core" of the world-economy tend to specialize in "core production": the production of core commodities using relatively . . . capital intensive technology and relatively skilled and highly paid labor" (Chase-Dunn 1998: 346), while countries in the "periphery" of the world economy tend to specialize in "peripheral production": the production of peripheral commodities using technology which is relatively low in capital intensity and labor which is paid low wages and is usually politically coerced compared to labor in core areas (Chase-Dunn 1998: 347). Economic activity in the modern world-economy, however, is not structured neatly within national borders.

The very essence of the world-

economy as such is that commodity chains cut across national borders, tying the entire interstate system into a single world-system. Thus, in the Wallersteinian 66

The Structure of National Income tradition, core-periphery hierarchies are embedded in commodity production chains as much as in state-to-state relations. In this view, Core activities are those that command a large share of total surplus produced within a commodity chain and peripheral activities are those that command little or no such surplus. All states enclose within their boundaries both core and peripheral activities. Some (core states) enclose predominantly core activities and some (peripheral states) enclose predominantly peripheral activities. As a consequence, the former tend to be the locus of world accumulation and the latter the locus of exploitation and powerlessness. (Arrighi and Drangel 1986: 12-13) Arrighi and Drangel further argue that the self-reinforcing advantages that come from having a high concentration of core activities within a state (and the

complementary self-reinforcing

disadvantages

that

come

from a

concentration of peripheral activities) tends to polarize the world-economy into core and peripheral states. (p. 26)

However, in between the core and the

periphery there exists a group of states which contain an "even mix" (Arrighi and Drangel 1986: 26) or "balance" (Chase-Dunn 1998: 210) of core and peripheral type production. These are the states that Wallerstein (1974) termed semiperipheral (p. 349).

In the Wallersteinian tradition these states are

postulated to represent a kind of "safety valve" that is necessary for the perpetuation of core-country capitalism. (Wallerstein 1979: 70) As formerly leading industrial sectors mature, declining profit margins push production out of core countries in search of lower-cost environments. The countries of the semiperiphery use their cost advantages vis-a-vis the core to attract these declining industries, at the same time extending core capitalists' ability to maintain the profitability of a given production process. The relationship is thus 67

The International Structure of Income symbiotic, and self-propagating. (Arrighi and Drangel 1986: 26-27) In some sense, the states of the semiperiphery can be thought of as a sort of second-run theater for what were formerly core activities.

Location of the Semi-Periphery On the location of the semiperiphery, Arrighi and Drangel (1986) take Wallerstein to task for inconsistent, vague, and even contradictory depictions (pp. 13-14). Arrighi and Drangel themselves, in a section titled "The Concept of Semiperiphery," offer a highly theorized account of how core-periphery hierarchies are created and perpetuated (pp. 16-26), complemented by a short and tentative description of how semiperipheries might come to exist (pp. 2628). Chase-Dunn (1998) lauds the concept of the semiperiphery as "one of the most fruitful concepts introduced by Immanuel Wallerstein" (p. 210) only to cast out entirely the idea that there are discrete zones in the world economy in favor of arguing for a hierarchical continuum of power among countries (pp. 210-214).

He concludes that "the vocabulary of zones is simply a useful

metaphor." (p. 214) Terlouw (1993) goes even further to conclude that the semiperiphery is "a blurred zone on the continuum between core and periphery." (p. 87) On the other hand, anecdotal evidence suggests that a qualitatively intermediate zone exists somewhere between the core and what are conventionally considered peripheral countries.

Wallerstein's own list of

semiperipheral countries is a case in point. According to Wallerstein (1979): It includes the economically stronger countries of Latin America: Brazil, Mexico, Argentina, Venezuela, possibly Chile and Cuba. It includes the whole outer rim of Europe: the southern tier of Portugal, Spain, Italy, and Greece; most of Eastern Europe; parts of the northern tier such as Norway 68

The Structure of National Income and Finland. It includes a series of Arab states: Algeria, Egypt, Saudi Arabia; and also Israel. It includes in Africa at least Nigeria and Zaire, and in Asia, Turkey, Iran, India, Indonesia, China, Korea, and Vietnam. And it includes the old white Commonwealth: Canada, Australia, South Africa, possibly New Zealand. (Wallerstein 1979: 100) Arrighi and Drangel make clear that they do not see much science in Wallerstein's diverse list of countries, accounting, as Arrighi and Drangel (p. 13) point out, for well over half of the world's population. They suggest that: As a matter of fact, the list simply includes all states that seem to occupy an intermediate position in the worldeconomy from the point of view of either their income levels or their power in the interstate system. The connection between such positions and the structure of the world-economy, as spelled out in the concept of semiperiphery, is completely lost, and the list could have been drawn up without any reference to such a concept. (Arrighi and Drangel 1986: 14) Arrighi and Drangel met this need for a formalization of the concept of the semiperiphery by positing that each national economy is composed of a mix of core-type and peripheral-type activity.

Countries hypothetically can be

ranked on the basis of the percentage of core-type activity in their economies. Below a certain percentage, the PC (perimeter of the core) boundary, countries have little or no power to upgrade their mix of activities; below an even lower percentage, the PP (perimeter of the periphery) boundary, countries have little or no power even to prevent the downgrading of their mix of activities. In this conceptualization, the two boundaries delineate the three zones of the worldeconomy. The semiperiphery is composed of those countries that lie between the PC and PP boundaries. 69

The International Structure of Income

THREE APPROACHES TO STUDYING THE STRUCTURE OF THE WORLD-ECONOMY Several attempts have been made over the past twenty years to operationalize empirically the concept of world-system position. In one group of studies, methods developed for social network analysis have been applied to world trade and other international data to delineate structurally equivalent blocks of countries(Snyder and Kick 1979; Nemeth and Smith 1985; Kick 1987; Smith and White 1992; Van Rossem 1996). In a second group of studies, countries have been clustered by income level (Arrighi and Drangel 1986; Korzeniewicz and Martin 1994). Studies in both traditions consistently uncover a three-zone partition of the world-economy. A third group of studies, however, is premised on the idea that the countries of the world fall on a continuum from core to periphery, with no attempt made at partitioning into zones. (Terlouw 1992; Van Rossem 1996; Kentor 2000) Chase-Dunn (1998) argues strongly for this third approach that operationalizes world-system structure as multi-dimensional convergence of hierarchies of "political, military, and economic types of power/dependence relations" (p. 215). In this approach, world-system position is not a discrete role variable, but is more of a continuous status variable. Higher-status countries do not directly exploit specific lower-status countries, but are simply more able to gain advantages in the global economy than are their competitors. This critique, however, does not invalidate the idea of a discrete partitioning of the world-economy, but complements it. The division of the world-economy into discrete zones, however, is probably best thought of in terms of roles played by states, contra groups of specific other states. Status in the world-system, however, is more of an absolute measure of a state's ability to project its will in the global arena. Thus, while India and China are clearly not 70

The Structure of National Income core countries on the basis of their roles in the world-economy, they are highstatus countries due to their large populations, activist international policies, and nuclear capabilities. On the other hand, while the Netherlands and Switzerland play unambiguously core country roles in the world-economy (both being wealthy centers of trade and administration), they carry less weight in the international arena than, say, China. The two perspectives are not mutually exclusive. The arguments presented in this volume are concerned more with differences between zones of the world-economy in the relationships among economic variables than with relative state status or power. Consequently, I will use a zonal characterization of the world-economy, rather than a continuous one. For a variety of reasons, however, network studies of world-system structure have yet to yield a convincing partition of the world-economy, while data limitations remove many countries from the network analyses, especially for periods before 1980.

Thus, I turn to a third tradition of world-system

classification for my methodological inspiration. This tradition, represented by two benchmarking studies employing near-identical methodologies, maintains that world-system zones can be differentiated on the basis of income alone. (Arrighi and Drangel 1986; Korzeniewicz and Martin 1994) In what follows, I discuss in more detail the literature in the network and continuum traditions. I then move on to review the income tradition. My own operationalization of the structure of the world-economy, developed in the context of the income tradition, will be presented in the following section.

The Network Tradition In some ways, the network analysis tradition comes the closest to capturing the idea of world-system zones as roles in the world-economy. Early 71

The International Structure of Income studies, such as Snyder and Kick (1979) and Nemeth and Smith (1985) have been largely superceded by similar work by the same authors. Of the later studies, Kick (1987) and Van Rossem (1996) both build their blockmodels on multiple networks of dominance in world trade, military power, and political memberships. Although they use similar variable lists, they use very different methodologies, and so arrive at very different results. Kick (1987) builds a structural equivalence network model on eight overlapping networks of relationships among countries: transportation/communication

treaties,

trade, aid treaties,

sociocultural

treaties,

administrative/diplomatic treaties, armaments transfers, and miliary conflict. For each network, Kick codes the ties between each pair of countries into a 0/1 dichotomy. His data represent 130 countries for roughly the period 1970-75. Kick finds 11 structural blocks. His first block is an obvious world-systems "core," but his other blocks represent various other groupings of countries. Kick's analysis is highly idiosyncratic, suffers from necessary but arbitrary dichotomization, and does not yield obvious world-systems categories. While it is interesting to compare Kick's results to those of other studies, Kick's methodology does not provide a model for basic world-systems benchmarking. Smith and White (1992) build their model around the concept of the "regular" or role equivalence of patterns of world trade. Role equivalence is an elaboration of the more familiar structural equivalence concept in network analysis.

Structural equivalence categorizes subjects on the basis of the

similarity of their relationships with specific blocks of other subjects. Role equivalence goes one step further to group together blocks of subjects that have similar sets of structural relationships. Thus, if former French colonies form a structural block in relation to France and former British colonies another structural block in relation to Britain, under regular equivalence former colonies 72

The Structure of National Income of all countries would form a role block in relation to all former colonists. As a result of their use of role equivalence, Smith and White find fewer blocks than other network studies of world-system structure.

In fact, they

confirm the three-tier structure common to most theoretical models, although they detect some splitting of blocks two and three into sub-blocks. They find one block (core) with an overwhelming volume of in-block trade, a second block that trades heavily with the core but not within itself, and a third block with very little trade of any kind. (p. 882, table 6) There are three major drawbacks to the Smith and White methodology. One is the limited number of countries for which data is available. Their study was limited to just 63 nations, and new data for most of the remaining countries of the world is not forthcoming. Second, their results may be clouded by currency issues. The United Nations commodity trade statistics used by Smith and White are denominated in dollars at official exchange rates, and thus grossly misrepresent the true volumes of trade for all countries before the early 1970s and for most countries even today. Third, network analysis has not progressed to the point where weighted analyses are possible. Smith and White mitigate this problem by excluding countries with fewer than 1 million population, but all remaining countries are weighted equally in the analysis. Van Rossem's (1996) methodology is similar to Kick's. It involves the network analysis of dichotomized measures of import dependence, export dependence, diplomatic ties, arms trade, and troop presences. Van Rossem, however, uses role equivalence as his primary measure, rather than structural equivalence. However, his classification of world-system "roles" lacks face validity when he places China, Brazil, and Saudi Arabia in the 1993 core while Sweden and Switzerland are placed in the semiperiphery and Norway, Ireland, and Israel are relegated to the periphery. These odd results are probably an 73

The International Structure of Income artefact of how he chose to operationalize his variables: he measures "export dependence," for example, as having exports greater than 1% of GDP. Such high levels of dependence do not, of course, represent "dependence" at all, but are more likely a sign of economic strength. His measures of international prominence are more meaningful, however, and will be discussed in the section on world-system continua below. All network-based methodologies suffer in varying degrees from the same shortcomings as Smith and White (1992).

Until these major

methodological problems are solved, studies in the network tradition will be most useful as confirmatory checks on benchmarks produced through other means. It will not be possible in the foreseeable future to produce a sound set of world-system benchmarks using network techniques.

The Continuum Tradition The continuum tradition, built on the theoretical work of Chase-Dunn (1998), emphasizes the comparative ranking of states on a continuum or multiple continua of status and power.

Three major studies to date have

operationalized such continuous status hierarchies: Terlouw 1992, Kentor 2000, and Van Rossem 1996. I discuss each of them in turn. Terlouw (1992) uses the mean level of six indicators to operationalize what he calls "mean coreness." These are level of trade, stability of trade, GDP per capita, military power, embassies sent and received, and diplomats sent and received. Terlouw's coreness measure is a good first approximation of worldsystem status, but it suffers from three key drawback: the even weighting of each of these factors in the final measure, the big country bias, and the lack of geographic controls. On the first issue, it is difficult to equate the combined importance of 74

The Structure of National Income embassies and diplomats, on the one hand, with GDP and military power on the other. Second, his measures seem to be highly correlated with country size. This is not a problem for a measure of status, or "punch." It is a problem for the study of role position -- which is not, to be fair, what Terlouw sets out to study. Third, Terlouw makes no accommodation for geography, and it would be difficult to see how he could. Countries with hostile neighbors, for example, will have larger militaries, but will in fact be less secure, not more secure, than isolated countries with small militaries. In any case, Terlouw's work has been largely superceded by work in the same tradition by Kentor (2000). Kentor takes a much longer view than any of the other studies considered here, attempting to measure world-system status over the entire 20th century. He starts with measures of countries' positions on each of ten status variables, grouped into three dimensions: economic power, military power, and global dependence. He was not able to assemble data for all variables for all countries at all time periods, but used mean z-scores within each dimension to cover for missing data as long as one measure existed for each dimension. He weights economic and military power equally, but gives dependence only half weight, on the basis of the low face validity of his results when dependence was weighted fully. Kentor's measures of world-system status are reported for 1900, 1930, 1950, 1970, and 1990. Although he musters an impressive 98 cases for 1998, he has only 52 countries in his database for 1970 -- and this despite a relatively liberal attitude towards missing data. Kentor's is probably the most careful study to date in the tradition of measuring world-system status on a continuum of relative power, but the unlikelihood of ever assembling the data for status in 1970 or earlier makes it difficult to use as a standard for studying long-term shifts in status. Kentor's method will probably prove most useful for studying 75

The International Structure of Income changing relative strengths within the core. An interesting and generally overlooked operationalization of worldsystem status is provided by Van Rossem's (1996) ranking of network prominence for 163 countries and territories in 1993. Van Rossem's prominence rankings suffer from the same variable measurement drawbacks as his network study (prominence is a measure of network centrality arising from his network analyses), but the idea is intriguing. A better-designed study using network prominence as a measure of status would nicely tie together the network and continuum traditions operationalizing hierarchy in the world-economy.

The Income Tradition World-systems sociologists generally agree that all states contain some mix of core and peripheral type activities within their borders. If it is the case that core type activities are vastly more remunerative than peripheral type activities, it should be possible to use national GNP per capita as a proxy for the level of core activity in the productive mix of an economy. As Arrighi and Drangel argue in their landmark 1986 article: The greater the weight of peripheral activities in the mix falling within the jurisdiction of a given state, the smaller the share of the total benefits of the world division of labor commanded by the residents of that state. The differences in the command over total benefits of the world division of labor must necessarily be reflected in commensurate differences in the GNP per capital of the states in question. We can therefore take GNP per capita expressed in a common monetary unit as an indirect and approximate measurement of the mix of core-peripheral activities that fall within the jurisdiction of a given state. (p. 31)

76

The Structure of National Income Arrighi and Drangel thus plotted share of world population by decile of logged GNP per capita, and took a three-interval moving average, to arrive at a graphical representation of the distribution of the world's population by national income bracket.

In general, they found a three-peaked aggregation of

population into distinct zones that was relatively stable over time.

They

delineated the zones using a formula that segmented zones at the troughs, with troughs defined as the mid-points between the peaks of adjacent zones. Korzeniewicz and Martin (1994) follow a virtually identical methodology, but use data for a larger number of countries and estimate zonal boundaries on an annual basis. The income-based methodology is a sound approach to delineating the zones of the world-economy, and is firmly rooted in theory, but it is possible to substantially improve upon these two existing income-based studies. Below, I develop my methodology for creating a new set of income-based benchmarks of position in the world-economy.

AN UPDATED INCOME APPROACH In their income-based approach to delineating the zones of the world-economy, Arrighi and Drangel (1986) and Korzeniewicz and Martin (1994) plot histograms of the sum of the populations of nations falling into national income bins of .1 points on a logarithmic scale (base 10). Thus, all nations with logged income between 2.0 and 2.1 ($100 and $125.89) have their populations assigned to the 2.0/2.1 bin; all nations with logged income between 2.1 and 2.2 ($125.89 and $158.49) have their populations assigned to the 2.1/2.2 bin; etc. Due to the fine level of the bins and the relatively small number of countries in the world, the resulting histograms are very erratic. The histograms average around 3 77

The International Structure of Income countries per bin, depending on the year, which means that many bins end up with no countries at all. At the other end of the extreme, one bin ends up with China, which puts the (population-based) histogram literally off the chart. The authors solve this problem by using a three-bin moving average to smooth out the results across adjacent bins. Even so, the resulting histograms are quite coarse. Peaks and troughs of the distributions are not always welldefined, and countries with stable income levels that are close to bin boundaries may randomly disturb results from year to year by arbitrarily jumping boundaries. As a result, the histograms do not show strong period-to-period continuity of form. Vast improvements in computing power in recent years have made feasible the use of stochastic plotting methods that would not have been available to Arrighi and Drangel and Korzeniewicz and Martin (collectively, ADKM). I set up population-weighted histogram of national income levels in much the same way as these earlier studies, categorizing national incomes along a base 10 logarithmic scale in bins of .1 on the log scale. However, I make two major improvements to the methodology. First, I improve on the current GNP series used in ADKM, substituting a GNP series deflated in local currency units for the dollar-deflated measure used by ADKM.

Second, instead of using

moving averages to smooth a final histogram, I smooth the underlying data itself by spreading each national observation over a range of bins. Below, I detail the operationalization of national income used in this study, followed by a summary of the countries and periods covered. I close this section with a detailed discussion of the smoothing method used to generate final histograms before moving on to the results.

78

The Structure of National Income National Income Data Both ADKM studies operationalized national income as gross national product (GNP) per capita, expressed in dollars at current (contemporaneous) exchange rates, then deflated to constant 1970 dollars using a U.S. dollar price index. The resulting "real" national income figures (expressed in constant U.S. dollars) can then be used for inter-temporal comparisons of incomes as well as international comparisons. I refer to this above as the FX methodology. The major drawback with the FX methodology is that reported exchange rates before the early 1990s are largely official (rather than market) rates, which are often highly distorted. There is an alternative methodology discussed above that also uses readily available data. It is possible to reverse the order of the steps in the FX methodology. The FX methodology involves first converting local currency figures to current dollars, and then deflating those dollars to a common year. The alternative is to first to deflate all local currency figures to a common base year, then to convert the resulting "real" local currency figures to dollars at the exchange rate for that base year. This real local currency (RLC) methodology substitutes reliance on domestic local currency price series for reliance on currency exchange rate series for earlier years. Choosing the RLC series implies more faith in the accuracy of the measurement of domestic inflation over time than in the accuracy of officially reported exchange rates ("accuracy" meaning that the officially reported rates truly reflect open market rates). Given the fact that in most of the countries of the world before around 1990 (and in many of the countries of the world today) government-issued licenses are required to obtain foreign exchange at official rates, and that the allocation of such licenses is considered by the World Bank to be a major source of corruption, the accuracy of officially reported FX rates must at least be called 79

The International Structure of Income into question.

One advantage of the RLC methodology is that it can be

implemented using as its base year that year during which exchange rates are judged to have been (historically) the least distorted. Over time, as more and more countries move from official to market exchange rates, the RLC methodology will become more and more reliable.

In contrast to the FX

methodology, better exchange rate data in the future will improve the reliability of the RLC figures not only for those future years, but for all past years as well, since the reliability of all of the RLC estimates is influenced by the reliability of the base year exchange rates. It is at least a reasonable suggestion that the contemporaneous conversion factors implied through the RLC methodology are better estimates of the exchange rates actually experienced by the public than are the reported official FX rates. If this is the case, then the RLC methodology is a more faithful application of ADKM's principles than is the FX methodology. At worst, from the ADKM perspective the RLC methodology (using base-year FX conversion factors) is an inferior operationalization driven by the same theoretical principle that inequality should be measured with reference to international, rather than domestic, purchasing power. For more details on FX versus RLC conversion factors, see Chapter 2.

Period and Country Coverage In the AKDM studies, strict inter-temporal comparability was sacrificed, in return for maximizing the number of cases available for analysis. To some extent this was a necessary choice, given their study periods that extended to the early part of the twentieth century, when data for most countries (and indeed most countries themselves) did not exist. In this study, I use a constant set of countries over the entire study period. Since consistent World Bank (2001b) 80

The Structure of National Income data is only available from 1960-1999, I use those forty years as my study period. This beginning of the period is dictated by historical constraints as well as by data sources. Before the major wave of decolonization in the 1950s, most of today's countries of the world did not exist as independent statistical entities. The year 1960 marks approximately the earliest year for which a sample covering the full spectrum of countries in the world-economy report national income. The end of the study period is in some ways also problematic, coming as it does after the dissolution of the USSR and Czechoslovakia and the reunification of Germany. These countries, however, are not critical for the determination of the boundaries between zones of the world-economy, since in fact over the study period they fell towards the middle of their respective zones. For the period 1960-1999 a continuous local currency income per capita series is available for 100 countries accounting for approximately 80% of the global population. The preferred income measure for the purpose of this study is gross national product (GNP), but GNP is not available for all countries for all periods. Some countries in some years only report gross domestic product (GDP). The two measures, however, are typically very similar, since the differ only by the amount of net international transfers. Empirically, the two measures are within 10% of each other for nearly all countries at all times. Therefore, in the 5% of cases for which local currency GDP data is available but GNP data is not, I use GDP. The constant-country sample, therefore, consists of 100 countries over the period 1960-1999. I will refer to these countries as the benchmark countries, since the zonal boundaries derived from the analysis of these countries will be used as benchmarks for placing other countries into zones in years when their income data is available. 81

The International Structure of Income Aggregating to a Smooth Histogram To spread the national income observation for each country over a range of bins, I again use a gaussian kernel with mean equal to the country's national income and a standard deviation of .1 (on the log scale). I have used a bin size of 0.1 units (of logged income) for aggregating the simulated income distribution.

This is also the bin size used by AKDM.

This bin size is

sufficiently granular to return a seemingly continuous histogram. In compiling each annual histogram, all population "clumps" that fall within a given bin are aggregated, regardless of the countries to which they belong. The resulting total population in each bin is then graphed to reveal a smoothed distribution of national income levels.

RESULTS Together, these methodological improvements yield dramatically richer histograms of the distribution of national incomes. Partly due to improved methodology and partly due to the use of a constant country set, the plotted histograms form a continuously interpretable series from year to year over the entire 40-year study period. Samples of the resulting histograms, for the years 1965, 1975, 1985, and 1995 appear below as Figures 21-24.

82

The Structure of National Income Figure 21. Distribution of countries in the world-economy (1965)

Figure 22. Distribution of countries in the world-economy (1975)

83

The International Structure of Income Figure 23. Distribution of countries in the world-economy (1985)

Figure 24. Distribution of countries in the world-economy (1995)

84

The Structure of National Income The existence of a trimodal distribution of countries (weighted by population) in the distribution of national incomes is unmistakable. In all forty study years, a smooth and continuous metamorphosis of one year's histogram into another's is maintained. In early years (before 1985) there are slight signs of a fourth mode, but this mode is never as pronounced as the other three, and is not stable over time.

An examination of the raw data reveals that the

prospective fourth mode mainly represents the rise of Egypt and Thailand out of the low-income zone. Egypt's move fails, while Thailand's move succeeds, by the late 1980s. Clear troughs in the histograms for every year mark the perimeter of the core (PC) and perimeter of the periphery (PP). I use the mid-point of each of the two major relative minimums in each histogram to define these two boundary points. The resulting boundary points for each year are reported below in Table 6. The time series of the estimated PP and PC boundaries are remarkably smooth, never increasing or decreasing by more than the .1 point minimum interval (determined by the bin size). Additional confirmation for the reliability of the PC and PP boundary points comes from a replication of the study, using data for all countries for which data is available for the single year 1995. More countries (178) have data available for 1995 than for any other year (more recent years show a drop-off as some countries report with a lag of several years). Analyses of the 1995 country set yields exactly the same boundary points as do the 1995 results based on the 100 countries in the 1960-1999 constant country set. The results reported in Table 6 are thus very robust with respect to country inclusion. Applying the boundary points in Table 6, I then determined the country zone assignments at five-year intervals. These are listed in Table 7 below. Each country is indexed as falling into one of three national income zones -- H, M, or 85

The International Structure of Income L -- corresponding to the core, semiperiphery, and periphery, respectively. For convenience of reading, I have below highlighted mobility between zones by displaying previous and current zonal assignments in the cells where zonal assignments have changed from the previous cell.

Thus, 1985 record for

Argentina ("HM") implies that prior to 1985 Argentina fell into the highest income grouping, but as of 1985 it was in the middle income grouping. The 100 countries that formed the benchmarking sample are identified in the column headed "BM."

The zonal boundaries, however, have been applied to all

countries for which data is available in a given year Table 7 shows considerable evidence of mobility within the worldeconomy, despite the impressive stability of the zonal boundaries. Working from the annual data (not shown), 28.5% of all countries have at one time or another over the 40-year study period experienced a zonal transition. This figure (28.0%) holds even for the 100 benchmark countries, on which the zonal boundaries themselves were constructed.

Even with over a quarter of all

countries experiencing transitions at one time or another, the overall structure of the system has remained unchanged. The direction of movements between zones is relatively symmetrical. Again working off the annual data (not shown), I have tabulated the frequencies of movements of countries between pairs of adjacent zones. The results are displayed in Table 8 below. While rates of upward and downward mobility across the PP and PC boundaries are roughly equal, the overall level of mobility across the PP (lower) boundary is about twice as high as that across the PC (upper) boundary. This is likely due to the fact that there are roughly twice as many countries in the low national income group than in the high.

86

The Structure of National Income

Table 6. Annual levels of PP and PC boundaries YEAR

1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989

Perimeter of the Periphery (PP) 2.95 2.95 2.95 2.95 3.05 3.05 3.05 3.05 3.05 3.15 3.15 3.15 3.15 3.15 3.15 3.25 3.25 3.25 3.25 3.25 3.35 3.35 3.25 3.35 3.25 3.25 3.25 3.25 3.35 3.35

Perimeter of the Core (PC)

PP (1995$)

PC (1995$)

3.55 3.55 3.55 3.55 3.55 3.55 3.55 3.65 3.65 3.65 3.65 3.75 3.75 3.75 3.75 3.75 3.85 3.85 3.85 3.85 3.85 3.85 3.85 3.85 3.85 3.95 3.95 3.95 3.95 3.95

$891 $891 $891 $891 $1,122 $1,122 $1,122 $1,122 $1,122 $1,413 $1,413 $1,413 $1,413 $1,413 $1,413 $1,778 $1,778 $1,778 $1,778 $1,778 $2,239 $2,239 $1,778 $2,239 $1,778 $1,778 $1,778 $1,778 $2,239 $2,239

$3,548 $3,548 $3,548 $3,548 $3,548 $3,548 $3,548 $4,467 $4,467 $4,467 $4,467 $5,623 $5,623 $5,623 $5,623 $5,623 $7,079 $7,079 $7,079 $7,079 $7,079 $7,079 $7,079 $7,079 $7,079 $8,913 $8,913 $8,913 $8,913 $8,913

87

The International Structure of Income

YEAR

1990 1991 1992 1993 1994 1995 1996 1997 1998 1999

Perimeter of the Periphery (PP) 3.25 3.25 3.25 3.25 3.25 3.25 3.25 3.25 3.25 3.25

Perimeter of the Core (PC) 3.95 3.85 3.85 3.85 3.85 3.95 3.95 3.95 3.95 3.95

PP (1995$)

PC (1995$)

$1,778 $1,778 $1,778 $1,778 $1,778 $1,778 $1,778 $1,778 $1,778 $1,778

$8,913 $7,079 $7,079 $7,079 $7,079 $8,913 $8,913 $8,913 $8,913 $8,913

These results tend to confirm Arrighi and Drangel's (1986) argument that the CP and PP boundaries are essentially one-way gates, allowing movement downward but rarely movement upward.

Although the raw frequency of

movement is roughly the same in both directions, in terms of percentages the likelihood of a semiperipheral country moving downward is much greater than that of a peripheral country moving upward, and similarly for the CP boundary. Many of the movements recorded in Table 3 are repeated movements across a boundary by countries with national income levels near the boundary itself. Nonetheless, a summary analysis of mobility based on the 5-year mobility table (in which virtually all movements represent permanent changes in status) yields identical results (not shown). Thus, I am able to confirm that the structure of the world-economy has been remarkably stable since 1960, despite the frequent movement of individual countries between zones over the forty-year period.

88

The Structure of National Income

Table 7. Position and mobility in the world-economy, 1960-1999 COUNTRY Aruba Andorra Afghanista Angola Albania Netherland United Ara Argentina Armenia American S Antigua an Australia Austria Azerbaijan Burundi Belgium Benin Burkina Fa Bangladesh Bulgaria Bahrain Bahamas, T Bosnia and Belarus Belize Bermuda Bolivia Brazil Barbados Brunei

BM 1960 Y H Y H Y H Y L Y H Y L Y L Y L Y H Y M Y L Y M Y M -

1965 H H H L H L L L H ML L M MH -

1970 H H H L H L L L H L L M H -

1975 -H H H H L H L L L H L L M HM -H

1980 -L -L H H -M H H L H L L L -L -H H L L M M H

1985 L L H HM M H H L H L L L L HM H L L M M H

1990 L L H M -L M H H -L L H L L L L M H -M LM L M M H

1995 L L H M L M H H L L H L L L L M H -L M M L M M H

1999 L L HM L M H H L L H L L L L MH L M M L M M H-

89

The International Structure of Income

COUNTRY Bhutan Botswana Central Af Canada Switzerlan Channel Is Chile China Cote d'Ivo Cameroon Congo, Rep Colombia Comoros Cape Verde Costa Rica Cuba Cayman Isl Cyprus Czech Repu Germany Djibouti Dominica Denmark Dominican Algeria Ecuador Egypt, Ara Eritrea Spain Estonia

90

BM 1960 Y L Y L Y H Y M Y L Y L Y L Y L Y M Y M Y H Y L Y M Y L Y L Y H -

1965 L L -H H M L L L L M M H L ML L L H -

1970 L L H H M L L L L ML M H L L L L H -

1975 L L H H M L L L L L M -M H L L L L H -

1980 -L L L H H M L L L L L -L M M -L H L L L L H -M

1985 L LM L H H M L L L L LM L -L M M LM H L LM L L H M

1990 L M L H H M L L L L M L L M MH -M M H L ML L L H M

1995 L M L H H M L L L L M L L M H M -H -L M H L L L L -L H M

1999 L M L H H M L L L L M L L M H M H LM H LM L L L L H M

The Structure of National Income

COUNTRY Ethiopia Finland Fiji France Faeroe Isl Micronesia Gabon United Kin Georgia Ghana Guinea Gambia, Th Guinea-Bis Equatorial Greece Grenada Greenland Guatemala Guam Guyana Hong Kong, Honduras Croatia Haiti Hungary Indonesia Isle of Ma India Ireland Iran, Isla

BM 1960 Y H Y M Y H Y M Y H Y L Y M Y M Y L Y M Y L Y L Y M Y L Y L Y H -

1965 H M H M H L MH ML L MH L L M L L H -

1970 H M H M H L -L -L H L L H L L M L L H -

1975 H M H MH H L L L H L L H L L M L L H -M

1980 H M H HM H L L L H -L L L H L L M L L H ML

1985 -L H M H M H L L L -L H LM L L H L L M L L H L

1990 L H M H -M M H L -L L L L H M L L H L -M L M L L H L

1995 L H M H M M H -L L L L L L H M L L H L M L M L L H L

1999 L H M H M M H L L L L L L H M L L H L M L M L L H L

91

The International Structure of Income

COUNTRY Iraq Iceland Israel Italy Jamaica Jordan Japan Kazakhstan Kenya Kyrgyz Rep Cambodia Kiribati St. Kitts Korea, Rep Kuwait Lao PDR Lebanon Liberia Libya St. Lucia Liechtenst Sri Lanka Lesotho Lithuania Luxembourg Latvia Macao Morocco Monaco Moldova

92

BM 1960 Y H Y H Y H Y M Y H Y L Y M Y L Y L Y H Y L -

1965 H H H M H L M -H L L H -L L -

1970 H H H M H L -L M H L L H L L -

1975 H H H M -L H L LM M H L L H L L -

1980 H H H ML L H L ML -M M H -L L L H L L -L

1985 H H H L LM H L L M M H -L LM L L H LM -H L L

1990 H H H L ML H -M L -L -L L M M H L -M M L L -M H M H L L

1995 H H H L L H ML L L L L M MH H L M M L L ML H M H L L

1999 H H H L L H L L L L L M H HL M M L L LM H M H L L

The Structure of National Income

COUNTRY Madagascar Maldives Mexico Marshall I Macedonia, Mali Malta Myanmar Mongolia Northern M Mozambique Mauritania Mauritius Malawi Malaysia Mayotte Namibia New Caledo Niger Nigeria Nicaragua Netherland Norway Nepal New Zealan Oman Pakistan Panama Peru Philippine

BM 1960 1965 Y L L Y M M Y M M Y L L Y M M Y L L Y M M Y H H Y L L Y L L Y L L Y H H Y H H Y L L Y H H L L Y L L Y M M Y M M Y L L

1970 L M -L M L ML L ML H L L L H H L H LM L M M L

1975 L M L M L L L L H L L L H H L H M L M M L

1980 L M L M -L L L L LM -L H L L L H H L H M L M M L

1985 L -L M L M -L L L LM L M L H L L L H H L H M L M M L

1990 L L M -L L M L L L M L M LM H L L L H H L H M L M M L

1995 L L M L L M L L L M L M M H L L L H H L H M L M M L

1999 L L M -L L L MH L L L M L M M H L L L H H L H ML M M L

93

The International Structure of Income

COUNTRY Palau Papua New Poland Puerto Ric Korea, Dem Portugal Paraguay French Pol Qatar Romania Russian Fe Rwanda Saudi Arab Sudan Senegal Singapore Solomon Is Sierra Leo El Salvado San Marino Somalia Sao Tome a Suriname Slovak Rep Slovenia Sweden Swaziland Seychelles Syrian Ara Chad

94

BM 1960 Y L M Y M Y L Y H Y L Y M Y L Y M Y L Y M Y L Y H Y L Y M Y L Y L

1965 L MH MH L H -M L MH L M L M L H L M L L

1970 L H H L H -H M L H L MH -L L M L H L M L L

1975 L H H L H H -L M L H L H L L ML L H L M L L

1980 L HM H L H H L M L H L H L L L L H L M L L

1985 L M HM L H H LM M L HM L H L L L L -M H L M L L

1990 L -M M MH LM H H ML M L M L H L L L -L L M -H H L M L L

1995 L M M H M H H L M L M L H L L L L L M H H L M L L

1999 L M MH ML H HL M L M L H L L L L L M H H L M L L

The Structure of National Income

COUNTRY Togo Thailand Tajikistan Turkmenist Tonga Trinidad a Tunisia Turkey Tanzania Uganda Ukraine Uruguay United Sta Uzbekistan St. Vincen Venezuela Virgin Isl Vietnam Vanuatu West Bank Samoa Yemen, Rep Yugoslavia South Afri Congo, Dem Zambia Zimbabwe

BM 1960 1965 Y L L Y L L Y M M -L Y H H Y H H Y H H Y M M L L Y L L Y L L

1970 L L M L -M HM H HM M L L L

1975 L L M L M M H M M L L L

1980 L L M L ML M H -L M -L -L M L L L

1985 L L -L M L LM -L M H L M -L L L M L L L

1990 L LM -M L M L M -L L -M M H -L LM M L L L -L M L L L

1995 L M -L ML L M LM M L L ML M H L M M L L -L L L M L L L

1999 L M L L L M M M L L L M H L M M L L L L L M LL L

95

The International Structure of Income

Table 8. Observed mobility across boundaries of the world-economy Cases

Percent

L->M M->H Total

61 26 87

1.1% 0.5% 1.6%

H->M M->L Total

23 57 80

0.4% 1.0% 1.5%

Stability

5284

96.9%

Total all repeated observations of countries' zone assignments

5451

100%

Upward mobility

Downward mobility

Considering the existence of such a long-run stability of the structure of the world-economy, it is meaningful to speak of sets of countries that are "organically" core, semiperipheral, or peripheral countries. These are countries that, in terms of income, typify each of the three zones (respectively). Using the set of 100 benchmark countries for which data are available for all 40 study years, I have compiled a list of the countries that have been consistently classified into a single income zone (high, medium, or low) over the entire 40 year study period. There were 72 countries that fit this definition of "organic" zone membership. These are recorded in Table 9.

96

The Structure of National Income

Table 9. Organic zone membership (benchmark countries only) CORE Australia Austria Belgium Denmark Finland France French Polynesia Iceland Ireland Israel Italy Japan Luxembourg Netherlands New Caledonia New Zealand Norway Spain Sweden Switzerland United Kingdom United States

SEMIPERIPHER Y Brazil Costa Rica Hungary Mexico Panama Seychelles South Africa Trinidad & Tob

PERIPHERY Bangladesh Benin Bolivia Burkina Faso Burundi Cameroon Central Afr. Rep. Chad China Congo, Rep. Cote d'Ivoire Ecuador Egypt, Arab Rep. Ghana Guyana Haiti Honduras India Indonesia Kenya Lesotho

Madagascar Malawi Mauritania Morocco Nepal Nicaragua Niger Nigeria Pakistan Pap. New Guinea Philippines Rwanda Senegal Sierra Leone Sri Lanka Suriname Swaziland Syrian Arab Rep. Togo Zambia Zimbabwe

Clearly, the striking feature of Table 9 is the shortage of "organically" semiperipheral countries.

This is, however, not so surprising.

Since the

semiperiphery is an intermediate category, it is possible for countries to move through it in both directions, whereas both the core and the periphery have "hard" boundaries on one side. Countries never rise above the core, nor do they fall below the periphery. In any case, the three big classically semiperipheral countries -- Mexico, Brazil, and South Africa -- do fall into the organic semiperiphery as operationalized here. In an additional confirmation of face 97

The International Structure of Income validity, the one historically socialist country in the benchmark data set, Hungary, also falls into the organic semiperiphery, in accordance with most theoretical categorizations. Similarly, the organic core reported here, although large by comparison to some other studies, passes face validity (with the exception of two low-population island groups, for which national income data are problematic). Thus, in general outline the zones of the world-economy uncovered in the forgoing analyses are consistent with accepted notions of the meanings of the terms "core," "semiperiphery," and "periphery."

DISCUSSION AND INTERPRETATION Whereas Chapter 3 was concerned with mapping the distribution of the world's people by income level, Chapter 4 has been concerned with mapping the distribution of the world's countries by average income level. The results are in some ways similar, but are analytically distinct. Taken together, the results of Chapters 3 and 4 clearly confirm Arrighi and Drangel's (1986) model of a world-economy in which there are two basic economic positions, but three distinct proportions in which those activities are likely to be mixed at the country level: core- and peripheral-type activities (or at least income levels) do in fact seem to be organized in countries that can be characterized as core, peripheral, and semiperipheral in average income. Moreover, the semiperipheral average income level is not merely the middle of a single distribution of countries; it is a distinct zone with a clearly definable and empirically estimable boundaries. This is a somewhat problematic conclusion. If, as researchers such as Korzeniewicz and Moran (1997) and Schultz (1998) suggest, global income inequality primarily reflects the between-country component of inequality, how 98

The Structure of National Income is it possible that an entire zone of the world economy, the semiperiphery, disappears in the graphs of global inequality presented in Chapter 3? This disappearing act is only possible if levels of within-country inequality in the semiperiphery are extremely high, in fact so high as to be comparable in magnitude to the observed differences in between-country inequality. Is this really the case? A long research tradition suggests that it is. Arguing from 19th and early 20th century data from the U.S., U.K., and Germany, Kuznets (1955) postulated an "inverted U" shaped relationship between income and inequality. In his model, countries began their modern developmental trajectories at relatively low levels of inequality, then passed through a high-inequality phase of economic expansion, after which inequality declined as the economies matured. This Kuznets Hypothesis has informed most of the inter-temporal and cross-national research on within-country inequality in the past half-century.

For a brief

review of this literature see Deininger and Squire (1998: 275:276). Although the Kuznets Hypothesis is inter-temporal in focus, much the evidence mustered in its support has been cross-national, substituting several countries at different levels of development for single countries over time. Whether or not these studies do in fact lend support to the Kuznets Hypothesis, they certainly have confirmed the fact that middle-income countries tend to have higher levels of income inequality than both high- and low-income countries. Combining the results from Chapter 3 with those reported above, it is possible to measure the (population-weighted) average level of inequality for the "organic" members of the core, semiperiphery, and periphery. I do this in Figure 25. I graph the average levels of within-country inequality by zone of the world-economy and decade. Using only organic zone members gives 72 cases. Inequality levels are taken from the Tier III sample of Chapter 3 (to ensure that 99

The International Structure of Income observations are available for all countries). The results graphed in Figure 25 are weighted by population.

Figure 25. Average internal income inequality by zone, 1960s-1990s

Average levels of income inequality are roughly twice as high, on average, in the semiperiphery as they are in the other two zones. As a result, it turns out that income inequality across the semiperiphery are similar in magnitude to the overall level of international income inequality for the world as a whole, as reported in Chapter 3. Average levels of the standard deviation of income on the order of .45 (on the log scale) imply that the plus or minus two standard deviation range of the incomes within semiperiphery countries (the range within which roughly 95% of households are likely to fall) is enormous: 1.8 points on the log scale. To put this in perspective, this is roughly the difference between the high and the low income modes reported in Chapter 3. 100

The Structure of National Income Or, in other words, the richest 2.5% of the population in semiperipheral countries can afford to live in typical core-like conditions, while the poorest 2.5% of the population in semiperipheral countries must live in typical periphery-like conditions.

Semiperipheral countries truly do straddle two

different worlds.

101

The International Structure of Income

102

5. Neoclassical Economic Growth in a Structured World-Economy The descriptive results of Chapters 3 and 4 established that the world-economy is structured in discrete levels of output per capita. Taking the individual person as the unit of analysis, the population of the world is divided into two broad, but distinct, groups based on income level, a low-income group centered around $500/year annual income and a high-income group centered around $20,000/year annual income.

However, taking the country as the unit of

analysis yields three broad national income levels, introducing a middle level that has very high internal income inequality compared to low- and high-income countries. Both of these global patterns in the world-economy -- the bimodal distribution of individual income and the trimodal distribution of national income -- have been remarkably stable over the four decades 1960-2000, even as many countries experienced substantial mobility within and between zones. The existence and, especially, the persistence of both of these patterns are inconsistent with accepted theories of income distribution in a competitive, open world-economy.

This suggests that, at a minimum, models of economic

development should be adapted to take into account the general structural traits of the world-economy.

The International Structure of Income Since Solow's (1957) seminal article on the role of technological innovation in determining the long-run growth potential of an economy, empirical development economics has evolved a common theoretical language and methodological toolkit, built on a neoclassical framework that models output as a function of capital, labor, and technology. In contrast to the classical framework in which land, labor, and capital were the only inputs, the neoclassical model emphasizes the long-run dominance of technology in determining the level of economic output. In the first section of this chapter, I review the neoclassical model, highlighting its departures from the classical model and examining the nature and measurement of each of its components in detail. The

neoclassical

model,

however,

nowhere

accounts

for

the

differentiation of the countries of the world on the basis of their positions in the world economy. On the contrary, it presumes that all countries have access to the same level of technology and that capital is equally mobile across all borders.

Thus, in the neoclassical model all countries possess identical

production possibilities. This is almost certainly not true in the world as it exists today, or ever has existed.

To address these issues, I propose two major

improvements to the standard neoclassical growth model in the second section of this chapter. First, I argue that the model should be estimated separately for each zone of the world-economy, allowing for different coefficients on labor and capital in each of the three zones. Second, I adopt the key insight from the world-systems and dependency literature that in addition to responding to levels of capital investment, rates of economic growth depend on the source of the capital invested in an economy as well. I further suggest that the effect of the source of capital should itself be contingent on a country's position in the worldeconomy. 104

Neoclassical Economic Growth in a Structured World-Economy In the third section of this chapter I estimate a series of growth models. I start with a base neoclassical model of long-run growth. I then differentiate this model by zone of the world-economy. Using a similar methodology, but a shorter time frame (due to data limitations), I estimate a series of models that examine the effect of investment source by zone. I conclude this chapter with a section discussing and interpreting the results of these regression models.

THE NEOCLASSICAL FRAMEWORK FOR GROWTH In the classical economics of the late eighteenth and the nineteenth centuries, all economic output was hypothesized to result from the interaction of just three scarce inputs: labor, land, and capital. Labor was conceived as including both the number of workers and their skills -- early on, a distinction was made between skilled and unskilled labor. Capital generally was held to include all forms of physical capital (fixed and moveable), but not financial capital. The elevation of land (as distinct from human-made capital) as a special category of input reflected the importance of agriculture in the economies of the day. Neoclassical growth theory still builds on this basic framework, with some modifications. Human capital has become such an important input in modern economies that it is now typically operationalized as a separate input, distinct from both labor and from physical capital. Land, on the other hand, has been subsumed within the more general category of physical capital. Two key twentieth century innovations in modeling the determinants of economic output, however, completely changed how we account for the relationship between economic inputs and economic output.

In the neoclassical framework, the

classical inputs of labor and capital are modeled as being subject to diminishing returns on each marginal unit of input, with the resulting level of output 105

The International Structure of Income depending, in addition, on the level of technology available. Though both of these "new" concepts (diminishing returns and the importance of technology) have pedigrees reaching back to the nineteenth century, their specific application in a model of aggregate output can be traced to the work of Solow (1956, 1957). Solow's synthesis of the neoclassical output model is the foundation on which most subsequent development economics has been built. In the fullest development of the model originally proposed by Solow, Mankiw et al (1992) model aggregate output as a product of the levels of labor, physical capital, and human capital inputs, scaled according to the level of technology in use in an economy. The use of each input factor, though, is subject to diminishing returns -- at the margin, each additional unit of a given input results in less increase in output than did the previous unit of input. Under the simplest model, in which the marginal utility of each input factor decays exponentially, the contribution of each input factor is raised to a (positive) fractional power.

Using T to stand for the (exogenously given) level of

technology and LABOR, CAPITAL, and HUMAN to stand for (respectively) the labor supply, the aggregate level of physical capital, and the aggregate level of human capital in the economy, output in the neoclassical model is given by: OUTPUT = T * LABORα1 * CAPITALα2 * HUMANα3 where α1 + α2 + α3 = 1 The sum of the coefficients of labor, physical capital, and human capital (α1 + α2 + α3) is specifically constrained to equal one, since multiplying all three inputs by a constant must result in output rising by that same constant (e.g., if all three inputs are doubled, output must also double).

Mankiw et al (1992)

estimate that a good approximation for α1, α2, and α3 in this model is α1 = α2 = α3 = 1/3. In other words, all three input factors have about the same rate of 106

Neoclassical Economic Growth in a Structured World-Economy diminishing returns to additional inputs. The neoclassical model implies the existence of two kinds of economic growth, equilibrium growth and disequilibrium growth. Growth in output per unit of labor, or in output per capita more generally, can only result from increases in the relative intensity of use of other factors of production: more technology, physical capital, or human capital per person means more output per person. Of three potential substitutes for labor, however, two -- physical capital and human capital -- are subject to diminishing returns. Consequently, capital intensity per capita would have to increase exponentially in order to maintain a linear increase in output per capita. Clearly, in an economy that maintains a static equilibrium in the ratios of the factors of production, continuous annual increases in output per capita must be attributed to increases in the level of technology used. Since output is a linear, multiplicative factor of technology, there is no limit (in the model) to the level of output per person, so long as technological innovation continues at a steady pace. Technological innovation leads to a steady rate of "equilibrium growth." If the ratios of the factors of production are not constrained to be constant, however, increases in capital intensity (physical and human) can lead to economic growth. This type of growth is sometimes termed "disequilibrium growth," since it results when the level of capital intensity changes from its equilibrium level. The economy grows (or shrinks) until it reaches the level of output per capita commensurate with the new level of capital intensity. For example, the neoclassical convergence debate (see Chapter 2) has been predicated on the principle that capital should continually be moving from highincome countries to low-income countries, balancing levels of capital intensity around the world. In order to test this hypothesis, it is possible to construct a model of so-called "unconditional convergence," in which growth in income per 107

The International Structure of Income capital over a long time frame is regressed on initial income per capita. A negative coefficient on initial income per capita indicates slower (faster) growth in initially high (low) capital intensity economies.

The best-known

representative of this type of model, estimated over the very long term of over a century, is Baumol (1986), famously debunked by De Long (1988). Better specified disequilibrium growth models also take into account existing domestic levels of capital intensity, both physical and human. Human capital, in particular, may not be geographically mobile.

In these models,

growth in output per capita is regressed on measures of capital intensity, while controlling for the initial level of output.

Models that include both initial

income per capita and measures of human and physical capital intensity are thus known as "conditional convergence" models, since the predicted convergence effect is conditional on domestic circumstances as well as on . The models found in Barro (1991) and the subsequent empirical growth literature are of this type. Disequilibrium growth can be modeled in a linear regression framework, with the constant forming part of the equilibrium, the coefficient on initial output per capita representing the convergence effect, and the levels of the neoclassical inputs included as controls. Such a generic neoclassical growth model can be written as: GROWTH = A + B0*INITIAL + B1*LABOR + B2*CAPITAL + B3*HUMAN This model can be estimated using standard linear least squares techniques, though some care must be shown due to the strong colinearities among the predictors and the systematic relationship between initial output and the error variance. Measuring the initial level of output per capita and its growth rate are 108

Neoclassical Economic Growth in a Structured World-Economy problematic enough, as evinced in Chapter 2. Initial output is generally logged, both to compress the scale and to eliminate period effects due to the fact that growth in output is exponential. I follow this convention. For estimating all of the models in this volume, I adopt the GNP/RLC series, described in Chapter 2, to measure output per capita. For growth, this is equivalent to using real local currency denominated figures, since the same currency conversion factor is used for all years. I operationalize growth as the geometric mean growth rate in output per capita. Arriving at measures for the three neoclassical inputs is considerably more difficult. Below, I discuss how labor, physical capital, and human capital have been operationalized in the literature to date, and detail my own chosen measures.

For an alternative review of the use of these variables in the

sociology and economics literatures, see Crowly et al (1998). Note that the star of the neoclassical model, the level of technology, drops out of the growth regression entirely. Accounting for differences in the level of "technology," broadly construed, across zones of the world-economy is the subject of the next section of this chapter.

Labor and the Age Structure of the Population Incredibly, no major contribution to the empirical growth literature in economics in the last twenty years includes demographic factors as predictors of growth.

On the contrary, in a benchmarking sensitivity analysis of some 50

variables mentioned in the empirical growth literature, Levine and Renelt (1992) include only one demographic variable, population growth. Occasionally output growth is alternatively operationalized as growth in output per worker (as opposed to per capita), but nowhere is the age structure of the population considered as a fundamental control variable in growth regressions. This is all 109

The International Structure of Income the more surprising in context of the fact that reasonable demographic estimates are available for virtually all countries for all periods. One partial exception to this general trend in economics is a paper by Williamson (1998) on the growth-distorting effects of demographic transitions. A "demographic transition" from a state of high fertility and mortality to a state of low fertility and mortality has occurred in all countries that have grown from low- to middle-income status. The transition in all cases follows a similar pattern. First, the mortality rate drops, as improved medical care, immunization, and most of all better nutrition lower mortality at all age levels. This means that people live longer, often a generation longer than their parents did. As a result, the population begins to grow rapidly. Second, after a lag of several decades, the birth rate drops to developed-country levels. This returns population growth rates back to normal levels. At the beginning of a demographic transition, per capita growth rates are depressed by the shift of the age structure toward older, unproductive ages. This effect is short-lived, however, as the lower mortality rates at all ages allow more children live to adulthood, meaning that in the population as a whole there is a greater percentage more working-age adults than there was before the transition. This favorable demographic climate is improved further when, later in the transition, the birth rate drops. This leads to a bulge generation, the last large generation of children, maturing through the age structure, with smaller numbers of dependents behind them. This is the period of highest economic output. Finally, there is a demographic bust, as the bulge generation matures into old age.

With the passing of the bulge generation, the country returns to a

demographic equilibrium, at a higher proportion of working adults in the population than there were before the transition, but a lower level than during the "golden years" of the bulge generation's middle age. 110

Neoclassical Economic Growth in a Structured World-Economy Bloom and Williamson (1998) document the effects of the demographic transition on the east Asian "miracle" of the past thirty years, finding that demographic factors accounted for as much as one-third of total east Asian growth, and full half of "excess" east Asian growth (growth in excess of the world-wide mean of 2%). Williamson (1998) further argues that demographic factors -- mainly migration -- accounted for substantially all of the difference in economic growth rates between North America and western Europe in the period 1870-1913. The massive migration of working-age adults from western Europe to North America over this period kept the American population profile continuously over-weight in the productive range, and the European population correspondingly under-weight. The current paucity of demographic variables in growth regressions belies a long history of demographic theories of growth, beginning with Malthus and the classical economists and extending through the mid-century work of Leibenstein (1954, 1957).

Leibenstein (1954) argued that under conditions

where the fertility rate (and thus the population growth rate) was positively related to income over a range of incomes, any absolute economic expansion would simply feed back into population growth, potentially precluding any increase in per capita incomes. This is a realistic picture of the first stage of the demographic transition, and Leibenstein's model warned that countries may be permanently trapped in the first stage of an economic transition if their economies were not shocked over the hump to a level of income commensurate with lower fertility rates. Liebenstein (1957) formalized and generalized this observation into the theory of low-level equilibrium trap (LLET). China's one-child policy can be interpreted as an attempt to escape a LLET by shifting the age structure of the population upwards through direct action on population, rather than indirectly and automatically through 111

The International Structure of Income experiencing a natural demographic transition. In this interpretation, economic growth in China may, over time, make the one-child policy superfluous, as Chinese fertility rates begin to decline naturally as a result of higher incomes. Assuming that China's strategy for escaping a low-level equilibrium is ultimately successful (and assuming that the Leibenstein's theory is itself correct), it highlights the exceptional level of policy effort that may be required to break through the LLET. The existence of a LLET is, however, by no means proved. More careful research on the interaction between demographic factors and economic development is necessary before we will have even a basic understanding of the dynamics of demography and growth. As a preliminary step in this direction, I account for the labor market impact of the age structure of the population on growth in my implementation of the neoclassical growth model. To account for the level of labor available for the production of economic output, I use the percent of population that is of working age. I do not adjust for labor force participation, male or female, since this is a poorly defined and poorly measured concept. Data for the percent of population that is of working age is derived from World Bank (2001b) series SP.POP.DPND, the age dependency ratio (ratio of dependents to working-age population). The percent of population that is of working age is extracted from the dependency ratio by adding one and then taking the reciprocal: LABOR = 1 / [DEPENDENCY_RATIO + 1] = 1 / [(dependent/working) + 1] = 1 / [(dependent/working) + (working/working)] = 1 / [(dependent + working) / working] = 1 / [total / working] = working / total 112

Neoclassical Economic Growth in a Structured World-Economy I use the period average of the percent of population that is of working age to represent LABOR in all of the regression models below and in Chapter 6.

Investment as a Proxy for Capital Intensity Much more prominent in empirical growth models is the inclusion of a measure of capital intensity.

In principle, capital intensity should be

operationalized through some direct measure of a country's total capital stock. In practice, however, the level of capital stock in a country is unknown. On the argument that, at equilibrium, a constant rate of new investment is required to maintain a constant level of capital (due to depreciation), investment rates (investment as a proportion of output) are generally used as proxies for capital intensity. Almost all empirical growth regressions in the economics literature adopt the use of investment as a proxy for capital intensity. To address the robustness of investment and other variables as predictors of growth, Levine and Renelt (1992) conducted a sensitivity analysis of a wide variety of predictors of economic growth. They find that there is "a positive and robust correlation between average growth rates and the average share of investment in GDP." (P. 959) This is in sharp contrast to the large number of variables -- population growth, government expenditure, political stability, civil liberty, defense spending, etc. -- for which the relationship to growth was sensitive to the presence or absence of specific controls. Investment is not only widespread in the growth literature, but consistently a powerful predictor of growth as well. A major refinement to the investment -- growth argument is to differentiate between types of investment. De Long and Summers (1991) take up this challenge: We believe that many previous studies have been carried 113

The International Structure of Income out at an inappropriate level of aggregation. We see no reason to expect that investments in structures should carry with them the same external effects as plausibly attach to investments in equipment. We are not aware of previous work that has separated the components of aggregate investment and studied their differential impacts on growth in a cross section of nations. Given the clear differences in the composition of investment . . . it is not surprising that studies that have focused on total capital accumulation have understated the potential contribution of investment to growth. (p. 480) De Long and Summers find that equipment investment is the key component of investment for the purpose of promoting economic growth. In their words: Equipment investment has far more explanatory power for national rates of productivity growth than other components of investment, and outperforms many other variables included in cross-country equations accounting for growth. (p. 446) Unfortunately, De Long and Summers base their arguments on the study of a limited number of countries. Although their full sample contains 61 cases, they find that their results in this sample are highly sensitive to outliers. They thus focus their efforts on a sub-sample of 25 high-income countries. The generality of their results is thus unknown, and, unfortunately, later work on investment and growth has not followed up on their arguments. In any case, a reliable break-down of investment by type is not available for a sufficient number of countries, especially poorer countries, to make the general use of equipment in growth regressions investment possible. Two short studies by economists, however, call into question the direction of causality between investment and growth. Using similar data and 114

Neoclassical Economic Growth in a Structured World-Economy methods, Blomstrom, Lipsey, and Zejan (1996) and Barro (1997, pp. 32-35) both find that rates of economic growth strongly predict rates of investment. Further, they agree in detecting no reciprocal causality: they find that the direction of causation is one way only, from growth to investment. The empirical evidence for this view seems strong, although it is at odds with most theoretical arguments. Offering an alternative to orthodox accounts, Carroll, Overland, and Weil (2000) present a model that can explain growth's effect to promoting savings (which is construed to equal investment), but casts no light on investment's corresponding lack of effect on growth.

One

explanation may be De Long and Summers' (1991) argument that total investment is irrelevant for growth, and that machinery investment (the critical component) is not adequately captured in the investment figures used by Blomstrom et al and Barro. Whether or not the heterodox view that levels of investment have no causal effect on growth is correct, it is a fact that investment still figures prominently in most growth theories and empirical growth studies as a proxy for capital intensity.

It is thus essential to include it in a replication of the

neoclassical growth model, however questionable its pedigree. Annual investment is, of course, a flow variable, while the neoclassical model formally incorporates a stock variable, the level of physical capital. Over a period of several years, however, any initial stock of capital, foreign or domestic, must depreciate to zero. As Scott (1989: 171-179) argues, the stock of capital in an economy is reduced to the total of new or recent investment, since the high depreciation rate on capital is constantly making the existing capital stock obsolete. Thus, long-term growth can be modeled based solely on rates of new investment, without accounting for the initial capital stock. See Lal and Myint (1996: 76-82) for a review and endorsement of this view. While is it 115

The International Structure of Income feasible to separate stocks from flows in a short-term growth study, it is not feasible to do so in a long-term study, since today's flows become tomorrow's stocks. In my implementation of the neoclassical growth model, I adopt the general practice of the economics literature of using investment flows as a growth predictor, making no effort to estimate directly levels of capital stock. For the regression models estimated in this chapter, I operationalize the variable CAPITAL using period average levels of investment as a percentage of output. Specifically, for total investment I use World Bank (2001b) series NE.GDI.TOTL.ZS (gross capital formation as a percentage of GDP), and for foreign investment I use World Bank (2001b) series BX.KLT.DINV.DT.GD.ZS (net inflows of foreign direct investment as a percentage of GDP). Domestic investment is simply the difference between these two series, where both are available.

Education as a Proxy for Human Capital Under the influence of emerging endogenous growth theories, measures of human capital have routinely been included in growth regression models since the early 1990s. However, as Mankiw et al (1992) clearly demonstrate, human capital is most appropriately incorporated in traditional neoclassical growth models, since human capital is presumably subject to diminishing returns in the same fashion as physical capital. In almost all implementations, human capital is operationalized through some measure of average education levels, with some interesting exceptions noted in Crowly et al. Human capital has been variously operationalized as educational attainment (Barro 1991; Barro and Lee 1993; Barro and Sala-I-Martin 1995), fertility (Barro 1997: 22-25), and life expectancy (Sachs and Warner 1997). 116

Neoclassical Economic Growth in a Structured World-Economy Barro and Lee (1996) have compiled a composite database of cross-national measures of education, including years of schooling, attainment rates, teacherstudent ratios, teacher salaries, etc., all broken down by level into primary, secondary, and higher education. In his own work focusing on education and growth, Barro (1997) uses average years of attainment for males in secondary and higher education. An alternative to Barro's operationalization is to use the secondary school enrollment rate (the percentage of the age-appropriate population that is enrolled in secondary school) as a proxy for a society's annual level of investment in human capital. A major advantage of this measure is that it is parallel in construction to the use of investment flows to represent physical capital intensity at equilibrium. An additional advantage of the enrollment rate operationalization is that Levine and Renelt (1992) use this variable to represent human capital in sensitivity analyses, so it is known to be robust with respect to a large number of sample configurations and conditioning variables.

The

enrollment rate operationalization also has a longer pedigree in the literature, having been used, seminally, by Barro (1991) himself. It is also the measure adopted by Mankiw et al (1992). A potential confounding factor in estimating the relationship between education and growth is the close correlation between educational attainment and other psychometric variables.

Both personal value sets and individual

intelligence are highly correlated with educational attainment within and between groups. Other omitted variables that are uncorrelated with education may also be at work. Provocatively, McClelland (1961) correlated motifs in children's stories with subsequent economic growth rates, convincingly demonstrating that the direction of causality ran from what he termed "need for achievement" to subsequent development, and not the other way around. These 117

The International Structure of Income observations should counsel caution in the interpretation of education coefficients in growth regressions.

The general lack of data on other

psychometric quantities, however, precludes their inclusion in the conditioning set. Following the mainstream of the empirical growth literature, I operationalize the variable HUMAN using period average levels of the secondary school enrollment rate in all of the growth regressions estimated below. Data are taken from World Bank (2001b) series SE.SEC.ENRR, gross secondary school enrollment as a percent of secondary school age population.

Omitted Variables Appearing in the Growth Literature The neoclassical model variables of initial income per capita, investment, and education are standard predictors that now appear in virtually all empirical growth studies. Demographic predictors occur only sporadically, but are clearly have a strong theoretical mandate in terms of the model. Many other variables, however, have at one time or another appeared in empirical growth regressions, with or without theoretical justification. Levine and Renelt's (1992) sensitivity analyses, however, suggests that none of the additional variables that were in use at the time are robustly correlated with economic growth. Specifically, they conclude that: We found that a large variety of trade policy measures [including the black market premium] were not robustly correlated with growth when the equation included the investment share. We found that none of the broad array of fiscal indicators that we studied is robustly correlated with growth. We found that a large assortment of other economic and 118

Neoclassical Economic Growth in a Structured World-Economy political indicators [including inflation and political stability] are not robustly correlated with growth. (Levine and Renelt 1992: 959) One potentially important variable not included in Levine and Renelt's study, geography, has been analyzed in detail by De Long and Summers (1991). To their surprise, they found no spatial correlation in residual growth rates in their sample (pp. 487-490).

THE NEOCLASSICAL GROWTH MODEL IN A STRATIFIED WORLD-ECONOMY A major limitation embedded in the standard neoclassical framework is that it differentiates among countries only through their levels of inputs.

This

limitation is reflected in models of both equilibrium and disequilibrium growth. Regarding equilibrium growth, in the neoclassical framework all countries benefit equally from a constant accrual of technological progress. In reality, however, it seems likely that new technologies disproportionately benefit those countries already that are already close to the technological frontier. Regarding disequilibrium growth, in the neoclassical framework the relative prominence of excess levels of the various neoclassical inputs are equal for all countries (i.e., there are no interaction effects).

Without making any a priori arguments

regarding differential impacts of individual inputs, it would be interesting from an exploratory point of view to uncover any major departures from the implicit neoclassical assumption of an undifferentiated world. A second implicit difficulty with the neoclassical model is its assumption international capital flows are a major force for equalizing levels of income per capita across countries. As discussed above, the sociology literature on capital 119

The International Structure of Income and investment dependence and economic growth casts serious doubts on this optimistic view of the role of international flows. Amazingly, capital flows have never been explicitly accounted for in a neoclassical growth model in the economics literature, despite the centrality of capital flows in the neoclassical framework for growth. In sociology, Dixon and Boswell (1996a) and Firebaugh (1992) incorporate capital flows into their growth models, but these models are not fully-specified neoclassical models, and are estimated over a very limited time period (1965-1973).

Clearly, a more comprehensive treatment is

warranted. In this chapter, I lay out some first steps in this direction. In this section, I specify the details of introducing a structural component to the neoclassical growth model, stratifying countries by position in the worldeconomy. I then discuss my operationalization of foreign capital flows in the context of this stratified model.

I close this section with some notes on

estimation issues.

Stratifying the Neoclassical Model The clearest way to integrate the observed structure of the worldeconomy into a neoclassical growth model is simply to estimate the neoclassical model independently for each zone of the world-economy. Trichotomizations of the countries of the world by national income level for all years 1960-1999 were derived in Chapter 4, based on a benchmark sample of 100 countries. All other countries for which income data are available for any year were assigned a zone based on where their incomes fell relative to the observed perimeter of the periphery (PP) and perimeter of the core (PC) for the 100 benchmark countries. A cross-validation was made on the reliability of this mapping using data for 1995, the year for which data are available for the largest sample of countries. The 1995 histogram of the income distributions of the 179 countries for which 120

Neoclassical Economic Growth in a Structured World-Economy data are available in that year yielded PP and PC figures that were identical to those derived from the benchmark series. In the stratified models below, I divide the countries of the world into three sets according to their initial zones in the world-economy, and estimate separate neoclassical growth models for each zone.

A comparison of the

resulting estimates for the three different zones highlights differences in the growth process between zones. It also serves to restrict the regression models to comparisons of somewhat similar countries. This is important from a policy development perspective, since the countries within each zone are likely to face more or less the same universe of possible policy actions. This is patently not the case for countries in different zones of the world-economy.

Differentiating Investment by Source A second structural dimension that is missing from the standard neoclassical model is the differentiation of domestic and foreign sources of capital. International capital flows are the lynchpin of the neoclassical model of growth, in which income convergence between economies is driven by flows of investment from high-income countries to low-income countries. This leads to a potential confounding of investment as a proxy for capital intensity within countries and investment as a mechanism for capital equalization across countries. Domestic and foreign investment are rarely if ever differentiated in the economics literature. Instead, foreign investment flows have mainly been studied by sociologists within the broader context of the effects of foreign capital penetration and foreign economic control. A long tradition of empirical work in sociology, dating from ChaseDunn (1975), has studied the growth-retarding impact of economic structures that are based on a high level of dependence on foreign capital. Researchers 121

The International Structure of Income working in this tradition claim that "foreign capital dependence diminishes economic growth, enhances income inequality, and very probably impairs domestic capital formation" (Dixon and Boswell 1996a: 544). The most recent evidence for the negative effects of foreign capital penetration and a review of the literature on investment source and growth comes from a long-term longitudinal study by Kentor (1998): The results of this study confirm that peripheral countries with relatively high dependence on foreign capital exhibit slower economic growth rates than those that are less dependent. These findings have been replicated using different measures of foreign investment dependence, GDP data, countries, time periods, and statistical methods. This is a significant and persistent negative effect, lasting for decades. Further, a structure of dependency is created that perpetuates these effects. (p. 1042) In his models, Kentor shows that a small, short-term positive effect of foreign investment (flows) on economic growth is counterbalanced by a large, long-term negative effect of foreign capital penetration (stocks). Kentor's results are consistent with interpretations by Dixon and Boswell (1996a, 1996b) and Firebaugh (1992, 1996), who conclude that domestic investment has a stronger positive effect on growth than does foreign investment. If this is the case, then those economies that rely more heavily on foreign investment will display lower overall growth rates than those that do not. In the short run, however, any new (non-equilibrium or non-replacement) investment causes a growth shock, since investment is itself a major component of national income. Unlike the economics literature, the sociology literature makes a theoretical and empirical distinction between foreign investment flows and 122

Neoclassical Economic Growth in a Structured World-Economy foreign capital stocks. However, the major sociological results differentiating the effects of foreign and domestic stocks and flows are based on a short sevenyear time frame for studying growth (Dixon and Boswell 1996a). As argued above for investment generally, the distinction between stocks and flows of foreign investment disappears as the study period lengthens, due to the rapid depreciation of capital stocks.

Thus, in the models below I consider only

period-average flows of foreign and direct investment. With respect to the dependent variable, the impact of this compromise solution is to average out the effects of the levels of initial stocks and subsequent flows. For the longer (40year) study period, the coefficient for foreign investment can be interpreted as being primarily a stock effect, while for the shortest (10-year) study periods, the effect can be interpreted as a pure flow effect. For the 20-year periods it represents some combination of the two. Because my operationalization of investment -- both foreign and domestic -- does not distinguish between stocks and flows, my models do not reflect the traditional theoretical meaning of "foreign investment dependence" as implying foreign control of the domestic economy. In my opinion the issue of foreign control is a red herring. In the traditional dependency theory view, foreign investment gives a country's economy an immediate boost, only to come back to haunt the economy later with slower growth due to the extraction of surplus value from the domestic economy by foreign firms. Foreign investment is thus cast in the role of forbidden fruit -- tasty now, but poisonous in the long run. What I intend to show in this volume, however, is that the perceived immediate boost to be gained from foreign investment is not in fact grounded in empirical reality of the past forty years. If there is no immediate gain from foreign investment, there is no fruit to be forbidden. Thus, I am not here concerned with the long-term effects of dependency as such. Instead, taking 123

The International Structure of Income them to be negative, I show that the expected short-term boost does not appear in the available data. The specific operationalization that I adopt for investment source is based on control over the management of investments, rather then the ultimate source of the funds. I operationalize domestic investment rates with the variable INV-DOM and foreign investment rates with the variable INV-FOR, respectively the annual levels of domestic and foreign investment as a percentage of GNP. By construction, CAPITAL = INV-DOM + INV-FOR. In my differentiated growth models, I simply substitute INV-DOM and INV-FOR in the place occupied by CAPITAL in the neoclassical model. This has the effect of decomposing the effect of CAPITAL into its two components. Data for INV-FOR are taken directly from the World Bank (2001b) series BX.KLT.DINV.DT.GD.ZS, net inflows of foreign direct investment as a percentage of GDP. Since this series only includes foreign direct investment (as opposed to portfolio investment), it includes only investments that are associated with foreign managerial control. The variable INV-DOM is calculated as the residual level of CAPITAL after INV-FOR is subtracted out.

It implicitly

includes net foreign portfolio investment, which is subject to domestic managerial control. As with CAPITAL, I use period average values of INVDOM and INV-FOR in all regression analyses. Not only is this parallel with the treatment of other variables in the neoclassical model, but it also dampens the high annual volatility of INV-FOR.

Study Periods and Estimation Issues There are two natural periods over which to estimate the base neoclassical growth model and its structured variants. The first is the maximum time period over which data are available. Around the year 1960 is the earliest 124

Neoclassical Economic Growth in a Structured World-Economy point in time for which models can be constructed that can fairly be said to represent the entire population of the world, since before 1960 much of the world was ruled as colonies by the European powers. Coincidentally, 1960 is the first year contained in the World Bank (2001b) database, which makes possible the sourcing of all data from a single, homogeneous source. The final year in the current World Bank (2001b) database is 1999. Long-run growth models will thus be estimated over the 40-year period 1960-1999. I also estimate growth regression models over a more restricted period, 1975-1995. Several factors point to 1975 as an appropriate start date. First, by 1975 the process of new nation-creation through decolonization was virtually complete. This means that from 1975 on data exist for a maximum number of countries. Second, in the emerging consensus view of economic historians, the reconstruction of Europe from the Second World War was not complete until around 1975. (Temin 1997) The same may also be true for the reconstruction of the former Japanese conquests, though the evidence is that Japan itself completed its reconstruction earlier. (Arrighi et al 1993) Taken together, all this implies that post-war growth rates for many major countries are substantially tainted by war-related factors until around 1975. Third, by 1975 the collapse of the Bretton Woods system had brought to a close the post-war period of fixed exchange rates tied to the dollar. In a development organically related to this event, the mid-1970s marked the beginning of a new phase of financial expansion, following the exhaustion of the post-war expansion in commodity trade. (Arrighi 1994: 298-299) In short, the overarching political economy in which the countries of the world operated changed dramatically in the years just before 1975. Similar, though less cataclysmic, factors suggest 1995 as a reasonable end date. Economic data series for many countries in the World Bank (2001b) 125

The International Structure of Income database thin out beginning in 1996. An end date of 1995 excludes both the emerging market crises and the massive U.S.-led asset bubble of the late 1990s. Last, and critically, due to data constraints the investment differentiation models can only easily be estimated over the period 1975-1995. For these reasons, all of the models in this chapter are estimated over the two periods, 1960-1999 and 1975-1995. The differing period lengths, start dates, and end dates also serve as a robustness check on the resulting estimates. Countries that experienced major changes in borders at the end of the Cold War (Germany, the USSR, and Yugoslavia) are not included in the dataset for either period. Note that China is present in the 1975-1995 regressions but absent for the 1960-1999 period due to missing data. Taiwan is absent from the database altogether, as the World Bank does not report data for Taiwan. All other major countries are in the regression datasets for both periods. All regression equations are estimated using both ordinary least squares (OLS) and weighted least squares (WLS). In the WLS analyses the initial populations of the countries are used as the regression weights.

Initial

population is taken from World Bank (2001b) series SP.POP.TOTL, total population. The variance of observed growth rates (like most other economic variables) is known to be systematically related to income per capita (variance declines as income rises). Whether this is due to increased economic stability or to better measurement at higher income levels, some adjustment for this heteroskedasticity should be made to ensure the reliability of the resulting regression estimates. The estimation of separate regression equations for the countries in each zone of the world-economy solves this problem, by associating countries in each sample only with other countries at a similar level of income per capita. Thus, the models for peripheral countries can be expected to have 126

Neoclassical Economic Growth in a Structured World-Economy higher error variances than the models for core countries, even though within each zone error variances are relatively constant. The larger number of cases in the periphery should help offset the larger error variances to be expected there.

RESULTS In this section I estimate the basic, stratified, and investment flow variants of the neoclassical growth model, using the specifications developed above. Before doing so, however, I think it would be worthwhile to verify some of the general assumptions of the neoclassical growth framework. First, equilibrium growth presumes that there is a general rate of positive economic growth, due to technological progress. In fact, geometric mean annual economic growth rates over the period 1960-1999 for the 100 countries for which we have data have averaged 1.90%, with a standard deviation of 1.68%.

The fastest average

growth rate was that recorded by Botswana (6.37%, beginning from a very low base), followed closely by Singapore and South Korea. The worst performance was turned in by Niger (-1.72%). In all, 88 out of 100 countries showed positive growth. Weighting growth rates by population in 1960 gives a higher global mean of 3.29% (SD 1.66%). This is an even stronger confirmation of the general assumption of positive growth. Disequilibrium growth is based on the assumption that each of the neoclassical inputs -- labor supply, physical capital, and human capital -- tend to increase with increasing levels of output per capita. When a country exhibits a level of inputs that is incommensurate with its current level of output per capita, the economy grows (shrinks) until a new, higher (lower) equilibrium is established. This model only makes sense if there is a positive correlation between levels of economic output and levels of the neoclassical inputs 127

The International Structure of Income (regardless of the potential for reverse causality). Regressions of the log of GNP per capita on the three neoclassical inputs are reported in Table 10 below. Data are from 1995; results for other years are similar.

Table 10. Regression of log(GNP) on neoclassical inputs, 1995 data dep. var.: LOG(GNP) (constant) LABOR CAPITAL HUMAN R2 (n)

Unweighted Coeff. (SE) .690 (.499) .040 (.010) -.007 (.005) .011 (.002) .680 (118)

t-stat 1.384 3.392 -1.365 6.178

Population weighted Coeff. (SE) t-stat .548 (.572) 0.958 .041 (.013) 3.199 -.037 (.006) -6.636 .016 (.002) 7.483 .787 (118)

In a positive sign, the three neoclassical inputs, taken together, account for 68% -- 78% of the variance in levels of output per capita, depending on whether the unweighted or the weighted results are adopted. The variables LABOR and HUMAN are strongly and positively related to output. But there is one major departure from the neoclassical assumptions: relationship between CAPITAL and output.

the negative

Even worse, in the weighted

analysis, this relationship is highly significant statistically.

The negative

relationship between CAPITAL and output shows up in the bivariate correlation as well. This is clearly at odds with the basic assumption of the disequilibrium model, that the (known positive) investment -- growth correlation is due the overall correlation between investment and output per capita. This results is, however, consistent with Blomstrom et al's (1996) and Barro's (1997, pp. 32-35) arguments that growth leads to investment, not the other way around. In any case, in the growth models estimated below, I will continue to include the 128

Neoclassical Economic Growth in a Structured World-Economy variable CAPITAL, but in light of its anomalous behavior here, the coefficients for CAPITAL below must be interpreted with care.

The Base Neoclassical Model A first step in building a stratified model of economic growth is to estimate a base neoclassical model, including all countries for which data is available. I have estimated base neoclassical models for geometric mean growth rates over both the long term (1960-1999) and the medium term (1975-1995). The results are reported in the tables below. In Table 11, I regress growth in GNP per capita over the period 19601999 on initial GNP per capita and the period averages of the three neoclassical inputs. Only 63 countries displayed full data on all required variables, for both the unweighted and the population weighted models.

The percent of the

variance in growth outcomes explained by the neoclassical predictors is well over 50% for both models. Similar coefficients are obtained when the model is estimated over the shorter period 1975-1995, the period that maximizes the available sample of countries (n=88; see Table 15).

Table 11. Long-run neoclassical growth model (1960-1999) dep. var.: GROWTH (constant) INITIAL LABOR CAPITAL HUMAN R2 (n)

Unweighted Coeff. (SE) -10.99 (2.89) -1.54 (0.54) .256 (.059) .147 (.037) -.001 (.014) .551 (63)

t-stat -3.799 -2.852 4.355 3.985 -0.068

Population weighted Coeff. (SE) t-stat -8.33 (2.45) -3.397 -1.02 (0.26) -3.952 .182 (.046) 3.943 .150 (.032) 4.726 -.002 (.012) -0.190 .626 (63)

129

The International Structure of Income The first major result from Table 11 is a strong confirmation of the conditional convergence hypothesis: after controlling for the neoclassical inputs, growth is negatively and significantly related to initial GNP per capita. This implies that poor countries would tend to grow faster than rich countries if they had equal levels of the three neoclassical inputs. This result is completely consistent with the economics literature on convergence, and with the current state of the convergence/divergence debate discussed in Chapter 2. Moving on to the three neoclassical inputs, LABOR displays a positive, significant relationship with growth, implying that demographic factors are, in fact, important for development. This is a novel, if not surprising, result. It is, moreover, a large result. Holding the level of GNP per capita constant, a move of just 4 or 5 percentage points, from a South Asian or Middle Eastern mean of around 55.5% of population that is of working age to a Latin American mean of 60%, would imply an increase in average growth rates of around 1%. To take a case where the cultural similarities are stronger, the six point difference between Latin America and Europe could account for substantially all of the growth differential between the two regions over the period 1960-1999. Results of this magnitude certainly argue for deeper analysis. CAPITAL also displays a positive relationship to growth, as expected. Both the size and the statistical significance of the coefficient are consistent with previous findings. As noted above, however, the lack of a correlation between CAPITAL and total output makes this result difficult to interpret in the context of the standard neoclassical growth framework. Moreover, CAPITAL does not vary systematically by geography as LABOR does. Nonetheless, the inclusion of LABOR as a control does nothing to diminish the predictive value of CAPITAL in the neoclassical growth model. For education, on the other hand, Table 11 reveals a strikingly 130

Neoclassical Economic Growth in a Structured World-Economy unconventional result: the coefficient for HUMAN is non-significant, and even negative! The explanation for the observed departure from the literature is the introduction of LABOR in the model reported in Table 11. The age structure of the population is not found as a variable in the empirical economics literature, and LABOR and HUMAN are highly correlated (r > .8). Removing LABOR from the equation gives the results reported in Table 12.

Table 12. Long-run neoclassical model, LABOR excluded (1960-1999) dep. var.: GROWTH (constant) INITIAL LABOR CAPITAL HUMAN R2 (n)

Unweighted Coeff. (SE) -.172 (1.694) -.853 (.590) -.160 (.042) .029 (.014) .405 (63)

t-stat -0.101 -1.445 -3.809 1.992

Population weighted Coeff. (SE) t-stat .906 (.809) 1.120 -1.245 (.287) -4.345 --.175 (.035) 5.014 .031 (.009) 3.342 .525 (63)

The results reported in Table 12 are consistent with well-confirmed results from the empirical economics literature. This confirms that it is, in fact, the inclusion of LABOR that partials out the marginal impact of HUMAN on growth. This suggests that the growth-enhancing effects of education touted in the growth literature of the past dozen years may have been due to nothing more than education's correlation with the excluded variable, percent of population of working age. Countries that have relatively fewer children in the population are better able to educate those children that they do have. The epiphenomenality of education's effect on growth is consistent with recent work by Easterly (2001) and Pritchett (2001). Easterly's (2001: Chapter 4) review of the effect of education on growth draws mainly on recent work on 131

The International Structure of Income changes over time in education levels and rates of economic growth, including Pritchett's study. Easterly concludes that: Despite all the lofty sentiments about education, the return to the educational explosion of the past four decades has been disappointing. ... Education is another magic formula that failed us on the quest for growth. (Easterly 2001: 84) Easterly's work was motivated by the observation that although education levels in the world's poorer countries have improved dramatically over the past four decades, growth in those countries has in general declined over the same period. If, growth regression results like those reported in Table 12 to the contrary, education itself actually has no effect on growth, education levels must be correlated to something that does. The evidence of this chapter points to the conclusion that that something is the age structure of the population. Taken in combination with the suspect status of investment in the neoclassical model (also highlighted by Easterly 2000: Chapters 2 and 3), the lack of any impact of education on growth vitiates the standard neoclassical growth model, as it is generally formulated. However, as I report below, fresh life may be blown into the neoclassical model when it is restructured to accommodate differences between zones of the world economy.

Stratification of the Neoclassical Model by Zone of the World-Economy In this section, I re-estimate the base neoclassical growth model of Table 11, this time dividing the sample into three groups based on countries' zones in the world-economy as of 1960. The 63 countries of Table 11 break down to 9 core countries, 20 semiperipheral countries, and 34 peripheral countries. Regression results for each zone are reported in Table 13.

132

Neoclassical Economic Growth in a Structured World-Economy

Table 13. Stratified long-run model (1960-1999) dep. var.: GROWTH Core (constant) INITIAL LABOR CAPITAL HUMAN R2 (n) Semiperipher y (constant) INITIAL LABOR CAPITAL HUMAN R2 (n) Periphery (constant) INITIAL LABOR CAPITAL HUMAN R2 (n)

Unweighted Coeff. (SE)

t-stat

Population weighted Coeff. (SE) t-stat

9.23 (12.73) -3.44 (2.92) -.040 (.117) .118 (.056) .080 (.033) .895 (9)

0.725 -1.178 -0.344 2.119 2.442

-1.28 (11.13) -2.51 (2.69) .101 (.088) .109 (.029) .056 (.023) .984 (9)

-0.115 -0.932 1.150 3.816 2.220

-14.42 (4.52) -2.87 (1.41) .456 (.082) .071 (.051) -.039 (.025) .800 (20)

-3.190 -2.042 5.564 1.406 -1.568

-2.01 (5.50) -4.50 (1.74) .313 (.084) .005 (.060) .025 (.023) .746 (20)

-0.365 -2.580 3.731 0.076 1.092

-3.86 (6.24) -2.07 (1.10) .125 (.108) .196 (.059) -.001 (.021) .460 (34)

-0.620 -1.889 1.161 3.344 -0.069

-8.80 (4.90) -1.11 (0.79) .155 (.077) .270 (.064) -.014 (.015) .681 (34)

-1.799 -1.398 2.006 4.208 -0.918

As Table 13 demonstrates, the structure of neoclassical economic growth differs dramatically across zones of the world-economy. The pattern of the results is near-identical in the unweighted and population-weighted regressions. 133

The International Structure of Income Inputs of capital, both human (HUMAN) and physical (CAPITAL), are the main independent contributors to growth rates in the core.

CAPITAL is also

important in the periphery, but HUMAN is not. And in the semiperiphery, neither form of capital has a significant independent effect on growth. Instead, growth is driven almost exclusively by LABOR; demographic effects predominate. In addition, the neoclassical conditional convergence effect is significant only within the semiperiphery.

This suggests the increasing

consolidation, over time, of a semiperipheral level within the world-economy; in other words, it suggests that the PC and PP boundaries are both difficult to cross, at least from within. One shortcoming of the regression results reported in Table 13 is that they are estimated over a 40-year period, a period during which the values of each of the neoclassical inputs may change substantially. The LABOR input is particularly troublesome in this regard, since countries' demographic profiles do not just evolve over generations, but evolve in a somewhat deterministic fashion. Re-estimating the equations from Table 13 over a shorter time frame may capture more causal detail, especially considering that the model underlying the regression equations is a "disequilibrium" growth model. Using a shorter time frame also increases the number of countries for which data are available. Changing the time frame to the medium run of 1975-1995 yields the estimates reported in Table 14 below. Again, the unweighted population-weighted regressions yield parallel results.

Strikingly, in both the core and in the semiperiphery, the only

independent predictor of economic growth is LABOR.

In the periphery,

LABOR is joined by CAPITAL. Interestingly, in the periphery the coefficient on HUMAN is negative, in the weighted model significantly so. The results reported in Table 14 might be taken to suggest that the growth process is not 134

Neoclassical Economic Growth in a Structured World-Economy strongly differentiated by zone of the world-economy, after all.

Table 14. Stratified medium-run model (1975-1995) dep. var.: GROWTH

Unweighted Coeff. (SE)

Core (constant) INITIAL LABOR CAPITAL HUMAN R2 (n)

5.08 (8.11) -6.21 (1.84) .307 (.104) -.023 (.076) .038 (.021) .540 (26)

Semiperipher y (constant)

-11.08 (14.34)

t-stat

Population weighted Coeff. (SE) t-stat

0.627 -3.368 2.938 -0.307 1.798

-10.20 (6.89) -1.42 (1.39) .263 (.086) .032 (.036) .002 (.020) .545 (26)

-1.481 -1.018 3.061 0.906 0.113

-0.773

-8.71 (10.63)

-0.820

INITIAL LABOR CAPITAL HUMAN R2 (n)

-3.78 (4.74) .362 (.175) .207 (.137) -.010 (.042) .543 (17)

-0.799 2.073 1.517 -0.230

-5.22 (3.78) .422 (.109) .120 (.148) .013 (.040) .754 (17)

-1.380 3.857 0.812 0.334

Periphery (constant) INITIAL LABOR CAPITAL HUMAN R2 (n)

-17.24 (5.67) -0.62 (1.10) .300 (.094) .201 (.057) -.034 (.022) .478 (45)

-3.039 -0.569 3.212 3.544 -1.513

-22.42 (5.29) -0.71 (0.72) .393 (.085) .268 (.038) -.056 (.020) .922 (45)

-4.241 -0.985 4.637 7.102 -2.777

But this would ignore the fact that, of the three neoclassical inputs, 135

The International Structure of Income LABOR is the one that differs most in its internal dynamics across zones of the world-economy.

As Williamson (1998) argues, the effect of demographic

variables on growth warrants greater attention. Contrasting the stratified results reported in Tables 13 and 14 with those found when all countries were included in a single sample suggests that the level of CAPITAL is, in fact, an important driver of convergence between zones of the world-economy. CAPITAL, after all, is a significant independent covariate of growth rates when all countries of the world are analyzed together, but not a stable and significant predictor of growth rates within each zone of the world-economy. Much of the power of CAPITAL as a predictor, therefore, must come from its correlation with growth rates across zones. This is consistent with the neoclassical convergence model in which capital is internationally mobile but labor, and thus implicitly human capital, is not. It is possible to test this conjecture explicitly by separating domestic from foreign capital flows in the growth regressions, both at the aggregate level and at the stratified level. I do this in the next section.

Investment Differentiation in a Medium-Run Growth Model Levels of domestic and foreign investment as proportions of national income are practically uncorrelated (r = .179; ns). This means that the effects of foreign and domestic investment are additive, making it possible to decompose the relationship between investment and growth into its foreign and domestic constituents.

Making this decomposition more appealing is the fact that

investment, domestic, foreign, or total, is not significantly correlated with any of the other variables in the neoclassical growth model. Colinearity is thus not a major issue in interpreting the regression coefficients for investment. In Table 15, I investigate effects of investment on growth, differentiating 136

Neoclassical Economic Growth in a Structured World-Economy total CAPITAL into its components INV-DOM and INV-FOR. In the top panel, I estimate the medium-run (1975-1995) base neoclassical growth model for the sample of all countries for which data are available. In the bottom panel, I estimate an equivalent model with CAPITAL broken into domestic and foreign components. The samples are slightly different, due to the absence of eight countries from the differentiated model for which data on foreign investment were not available, but the R2 statistics and regression coefficients for the two models are almost identical for all non-investment variables. In both the unweighted and the population-weighted models, the coefficient for INV-DOM is nearly identical to that for CAPITAL. This is not surprising in light of the fact that, on average, domestic investment accounts for well over 90% of all investment; the two variables are nearly synonymous. The coefficients on INV-DOM are, like those for CAPITAL, highly significant statistically. The results for INV-FOR reported in Table 15 are broadly consistent with the assumptions of neoclassical growth theory.

INV-FOR displays a

strong, positive relationship with growth rates, after controlling for the levels of other neoclassical inputs. This relationship is statistically significant in the population-weighted model, but not in the unweighted model. Dollar for dollar, the impact of foreign investment is more than twice as large as that of domestic investment in the weighted model. This directly contradicts Dixon and Boswell (1996a), who find that the effect of foreign investment is only half as large as that of domestic investment. Dixon and Boswell's estimates, however, are based on growth rates over the very restricted time period 1965-1973, a time period, moreover, which does not overlap that for which the results reported in Table 15 are estimated.

137

The International Structure of Income Table 15. Medium-run models (1975-1995) dep. var.: GROWTH Base (constant) INITIAL LABOR CAPITAL HUMAN R2 (n)

Unweighted Coeff. (SE)

t-stat

Population weighted Coeff. (SE) t-stat

-11.47 (2.83) -1.89 (0.55) .267 (.061) .124 (.038) .009 (.014) .420 (88)

-4.047 -3.469 4.400 3.286 0.622

-18.08 (2.38) -1.68 (0.30) .362 (.046) .199 (.026) -.012 (.012) .892 (88)

-7.586 -5.570 7.820 7.618 -0.991

-13.27 (3.05) -1.80 (0.60) .291 (.062) .141 (.039) .236 (.150) .004 (.015) .496 (80)

-4.346 -3.020 4.710 3.637 1.579 0.238

-17.42 (2.22) -1.79 (0.28) .349 (.044) .195 (.026) .472 (.160) -.005 (.012) .919 (80)

-7.833 -6.297 7.968 7.497 2.952 -0.418

Differentiated (constant) INITIAL LABOR INV-DOM INV-FOR HUMAN R2 (n)

Table 16 breaks out the lower panel of Table 15 into separate regression models for each of the three zones of the world-economy. In no zone is INVFOR significant for predicting differences in growth outcomes among the countries within the sample.

The coefficients for INV-FOR are, however,

positive in all three zones in both the unweighted and the population-weighted models. The coefficient is particularly strong in the semiperiphery, suggesting that mobility for semiperipheral countries may be closely tied to foreign investment. The results can only be characterized as "suggestive" on account of the large standard error of the estimate. 138

Neoclassical Economic Growth in a Structured World-Economy Table 16. Stratified, differentiated medium-run model (1975-1995) dep. var.: GROWTH Core (constant) INITIAL LABOR INV-DOM INV-FOR HUMAN R2 (n)

Unweighted Coeff. (SE)

-14.26 (13.14)

t-stat

-1.085

Population weighted Coeff. (SE) t-stat

-14.63 (7.77)

-1.883

-0.02 (3.31) .179 (.149) .045 (.081) .354 (.201) .033 (.041) .523 (20)

-0.007 1.202 0.560 1.759 0.810

2.17 (1.67) .146 (.096) .033 (.033) .150 (.152) -.032 (.022) .565 (20)

1.295 1.522 1.002 0.986 -1.444

Semiperipher y (constant) INITIAL LABOR INV-DOM INV-FOR HUMAN R2 (n)

-9.03 (14.19) -4.46 (4.70) .355 (.172) .232 (.135) .782 (.565) -.020 (.042) .593 (17)

-0.637 -0.948 2.063 1.712 1.385 -0.469

-2.82 (11.78) -8.08 (4.59) .450 (.112) .272 (.195) 1.223 (1.124) -.022 (.050) .780 (17)

-0.240 -1.761 4.007 1.395 1.088 -0.445

Periphery (constant) INITIAL LABOR INV-DOM INV-FOR HUMAN R2 (n)

-19.84 (5.67) -0.70 (1.09) .353 (.093) .185 (.057) .304 (.265) -.023 (.022) .544 (43)

-3.5000 -0.646 3.785 3.261 1.148 -1.033

-27.92 (4.51) -0.05 (0.61) .462 (.074) .230 (.039) .346 (.210) -.033 (.017) .951 (43)

-6.186 -0.088 6.227 5.953 1.643 -1.874

139

The International Structure of Income

DISCUSSION AND INTERPRETATION The single most important predictor of neoclassical disequilibrium growth seems to be the supply of labor. Amazingly, although the supply of labor is central to the neoclassical model, empirical growth regressions in the neoclassical tradition have not to date included LABOR as a predictor. The omission of LABOR in the literature has contributed to the consistent finding of a now seemingly spurious correlation between education and growth.

The

strong correlation between the age structure of the population (reflected in LABOR) and countries' abilities to educate their populations (reflected in HUMAN) has allowed HUMAN to usurp the explanatory power of LABOR in growth regressions from which LABOR was excluded. While more work is certainly warranted to more clearly differentiate the effects of LABOR and HUMAN, it is clear that LABOR or a similar variable should be included in neoclassical growth regressions in the future. The behavior of investment, the second major input in the neoclassical growth model as commonly estimated, is also problematic. Although associated with growth, especially in the periphery, CAPITAL is not associated with GNP per capita itself. This means that any impact that CAPITAL has on economic growth cannot operate through the disequilibrium model that forms the theoretical foundation for neoclassical growth regressions. More likely, as Blomstrom et al (1996) and Barro (1997) argue, the level of investment is itself determined by the rate of economic growth.

This

conclusion is even more likely in the case of foreign investment. If rapid economic growth creates incentives that encourage investment (such as higher returns on capital), those incentives will have a greater impact on the flow of international investment than on the flow of domestic investment.

This is

because the elasticity of domestic investment depends largely on its 140

Neoclassical Economic Growth in a Structured World-Economy attractiveness relative to domestic consumption, while the elasticity of foreign investment in a given country depends largely on the attractiveness of investment in that country relative to other countries. A higher growth rate than other countries directly implies a higher attractiveness for investment. Despite the poor theoretical showing of the neoclassical growth model, its empirical performance is quite strong. Consistent with the mainstream of the empirical growth literature, I find that the base neoclassical model accounts for around 42% of the variance in growth outcomes over the twenty years 19751995. This rises to 55% of the variance in growth over the longer term of 19601999. Weighting on population gives stronger results still: 63% for the long term and 89% for the medium term.

Thus, in my theoretically preferred

operationalization, the base neoclassical model accounts for almost nine-tenths of all of the weighted cross-national variation in growth outcomes. Similarly high percents of variance explained obtain for the stratified models. Specific coefficients, however, vary dramatically across zones of the world-economy. LABOR is the key input in the semiperiphery. This suggests that the successful navigation of the demographic transition from high birth and death rates to low birth and death rates is crucial for the long-term economic prospects of middle-income countries. London (1988) finds that high levels of dependence can impede the completion of a successful demographic transition; combined with London's result, this suggests a previously unexplored pathway for the effect of dependence on growth. LABOR is also an important indicator for the periphery, though not as strong and consistent an indicator as investment. Although Blomstrom et al (1996) and Barro (1997) find that investment is only a lagging indicator of growth for the countries of the world taken as a whole, it is possible that this would not be the case for the periphery taken in isolation. Given the prominence given to investment in international development policy, 141

The International Structure of Income this is an important topic for future research. In contrast to the results from the semiperiphery and periphery, regression results for the core are highly unstable. For the small sample of nine 1960 core countries in the long-run growth model, CAPITAL and HUMAN are the two significant independent predictors of economic growth.

For the

expanded core of 26 countries studied over the period 1975-1995, however, LABOR displayed the only significant coefficient. The latter result, based on better data and a larger sample of countries, is probably the more reliable. For core countries, however, the interpretation of LABOR is different than it is in the semiperiphery and periphery.

All core countries have already passed

through the demographic transition, and are facing problems arising from their high numbers of aged adults relative to workers, rather than a high number of children. The only major solution to this problem is immigration. This suggest that immigration rates should be studied in more detail as predictors, rather than consequences, of economic growth. To summarize, the major novel finding of this chapter is that demographic factors are much more important for predicting disequilibrium economic growth than previously thought. While this is especially true for the periphery and semiperiphery of the world-economy, it is even, in some models, true for the core. Capital -- physical and human -- is much less important than its prominence in the literature would suggest. I demonstrated in this chapter that the relationship between human capital and growth seems to be spurious. In Chapter 6 I demonstrate that the observed correlation between physical capital (investment) and growth also does not represent a causal relationship.

142

6. Investment and Growth: Disentangling Causal Paths The neoclassical growth model estimated in Chapter 5 clearly demonstrated that cross-national differences in levels of human capital, measured in terms of educational attainment, have little or no direct role in explaining cross-national differences in rates of economic growth. Levels of physical capital, on the other hand, seem to be more powerful in predicting disequilibrium growth, particularly when represented by domestic investment flows, and particularly in the periphery of the world-economy. Recent work in economics suggests that this observed positive relationship between investment and growth may also be spurious.

It is possible that the observed relationship is a case of reverse

causality, in which growing economies attract more investment, while increased investment contributes little or nothing to increased growth. In the case of human capital, it was possible to identify the confounding variable that accounted for the spurious correlation between education and growth.

Human capital is highly correlated with the percentage of the

population that is of working age. When this variable (LABOR) was omitted from the regression models for growth, human capital (HUMAN) had a strong and significant coefficient. When LABOR was included, the apparent effect of

The International Structure of Income HUMAN disappeared. In dealing with the relationship between physical capital and growth, the causal relationships are more difficult to resolve.

Here, the potential

confounding variable is not an alternative independent variable, but the dependent variable itself. Previous attempts to separate the effect of growth on investment from that of investment on growth have relied on time-series methods. (Blomstrom, Lipsey, and Zejan 1996; Barro 1997) Both of these articles use the Granger test of causality, finding that prior period levels of growth predict future investment (even when prior period investment is controlled for), while prior period levels of investment do not predict future growth (when prior period growth is controlled). The interpretation is that growth holds predictive value for future investment above and beyond its current relationship to investment, while investment holds no such additional predictive value for growth. There are at least three shortcomings to this approach. First, the Granger test is, statistically, extremely conservative. statistically significant results whatsoever.

Granger tests often return no The finding that growth holds

predictive power for future investment even after controlling for prior investment thus a strong result, which should be accepted at face value. On the other hand, the investment's lack of predictive power for future growth is less convincing, since the bar of statistical significance is, in this case, set very high. The result is thus suggestive, rather than conclusive. The second and more important limitation of the Granger test approach is that it does not consider potential contemporaneous effects of investment on growth (or vice versa). Instead, contemporaneous effects are controlled in an attempt to isolate lagged effects.

This approach is reasonable when the

"contemporaneous" period is short enough that it can be assumed that no 144

Investment and Growth: Disentangling Causal Paths important causal effect is occurring over the time period. In the present case, however, both Blomstrom et al and Barro use "contemporaneous" periods that are a decade in length, in order to smooth growth rates over the course of the business cycle. Substantial contemporaneous causality may occur over such a long period. Ironically, the decade-long period averages used by Blomstrom et al and Barro are also, from another perspective, too short. This leads to the third problem with their causality tests: they are fundamentally mis-specified. Investment, as such, does not appear in the neoclassical growth model. Investment over time is instead a stand-in for the level of physical capital, which is difficult to measure directly. Over twenty- or forty-year periods, like those studied in Chapter 5, the total or average level of new investment can be taken to be equivalent to the level of physical capital, since any initial stock of capital would depreciate to nothing over the course of the period of study. A ten-year period, on the other hand, may be too short for this purpose. In this chapter, I take a different approach to measuring the causal relationship between physical capital and growth. In place of the time series based techniques generally found in the economics literature, I specify two distinct kinds of structural equation models to elucidate the causal paths. First, I formulate a reciprocal effects model specifying domestic investment, foreign investment, and growth all as endogenous variables.

The model is

contemporaneous; all variables are measured over the same period (1975-1995). Both domestic and foreign investment are allowed to have reciprocal effects with growth -- that is to say, growth can affect them and they can affect growth, contemporaneously. I separate the effects of domestic and foreign investment in order to allow them to have different causes and different relationships to growth. The specifics of model specification, the estimation results, and a 145

The International Structure of Income discussion of those results appear in the first two sections below. Second, as a check on the results reciprocal effects model, I estimate a series of simple panel regression models of the relationship between investment and growth. In these models, I regress current period growth on both prior period investment and prior period growth; a significant coefficient for investment would indicate the presence of a causal path. Similarly, I regress current period investment on both prior period investment and prior period growth, to see whether growth affects investment. I perform these regressions for both domestic and foreign investment rates, using the periods 1975-1985 and 1985-1995 as the prior and current periods, respectively. These models are presented and discussed in the third section below. In a final concluding section, I summarize the implications of the empirical analyses conducted in this chapter for our understanding of the causal paths connecting physical capital and growth.

A RECIPROCAL EFFECTS MODEL FOR INVESTMENT AND GROWTH The first step in constructing a reciprocal effects model to elucidate the causal relationships between foreign and domestic investment and growth is specify models for each of the three dependent variables.

Below, I begin with the

investment differentiation growth model from Chapter 5, then add additional instrumental variables to supplement the models for foreign and domestic investment. The full structural model that results is sufficiently over-identified to permit the estimation of reciprocal effects.

146

Investment and Growth: Disentangling Causal Paths Implementation of the Neoclassical Model for Growth In the investment differentiation growth model of Chapter 5, GROWTH is regressed on the four inputs LABOR, HUMAN, INV-DOM, and INV-FOR, while also controlling for INITIAL (initial GNP per capita). In graphical terms, this model can be represented with the diagram below (Figure 26). The arrows in the model represent causal effects; the epsilon contributing to GROWTH represents the effect of all omitted variables, or, from the perspective of the model, error.

The variables LABOR, HUMAN, and INITIAL have been

grouped in a common box to simplify the display.

Figure 26 exactly

corresponds to the models estimated in Table 15 (lower panel) and Table 16.

Figure 26. Investment differentiation growth model INV-DOM

e LABOR GROWTH HUMAN INITIAL

INV-FOR

The reciprocal effects model endogenizes the two investment variables. In addition to being independent variables in the model for investment, they are also dependent variables in their own right. The resulting model is depicted in Figure 27. The dashed arrows represent new causal paths that have been added to the model. Since INV-DOM and INV-FOR are now dependent variables in 147

The International Structure of Income regression models of their own, they now have their own corresponding error components (epsilons). Figure 27 conceptually represents the causal paths to be estimated.

Figure 27. Investment differentiation model with reciprocal effects e

INV-DOM

e LABOR GROWTH HUMAN INITIAL

e

INV-FOR

Although this model could be estimated as depicted (using reiterative maximum likelihood techniques), the reciprocal effects between growth and the two investment variables are relatively poorly identified. The stability and reliability of estimates can be improved through the addition of instrumental variables that are correlated with INV-DOM and INV-FOR. Two such variables are proposed below.

Instrumental Variables Identification of the structural model for investment and growth will benefit most from the addition of variables to the models for investment. The 148

Investment and Growth: Disentangling Causal Paths economics literature suggests that a key determinant of domestic investment is the level of domestic financial development. (King and Levine 1993; Levine 1997) Conceptually, financial development represents the scale of domestic financial intermediation. The more national financial systems operate on the basis of commercial credit, the higher the level of financial development. Recent research has shown that the relationship between financial development and investment is causal and unidirectional. (Levine, Loayza, and Beck 2000) The correlation between financial development and domestic investment is on the order of .3 or more. An even stronger case can be made for the addition of merchandise trade to the model for foreign investment. Foreign investment is often organically linked to foreign trade, as when a manufacturer sets up an offshore assembly plant. Although the direction of causality may between investment and trade may be difficult to establish, it is not directly relevant to the use of trade as an instrumental variable, since trade will remain exogenous to the structural model. The important consideration is whether foreign investment and trade are sufficiently highly correlated.

The correlation between trade and foreign

investment is on the order of .7 or more. I operationalize financial development with the variable FIN_DEV, defined as the period average level of domestic credit provided by the banking sector as a percentage of GDP. Data are taken from World Bank (2001b) series FS.AST.DOMS.GD.ZS. This operationalization is consistent with the recent literature on financial development, investment, and economic growth. I operationalize the variable TRADE as the period average level of imports plus exports as a percentage of GDP. Data are taken from World Bank (2001b) series NE.TRD.GNFS.ZS.

This choice of operationalization is

supported by Frankel and Romer's (1999) finding that observed levels of trade 149

The International Structure of Income are unbiased for the purpose of estimating the effect of trade on growth.

The Full Structural Model The addition of FIN-DEV and TRADE completes the structural model for growth, domestic investment, and foreign investment. These two variables are added to the models for INV-DOM and INV-FOR, respectively, as well as to the model for growth.

Additional causal paths are possible, but are not

necessary for identifying the relationships of interest among the variables. The resulting structural model is depicted in Figure 28. The arrow connecting the error terms for INV-DOM and INV-FOR represents a correlation of residuals. Since a wide number of factors may affect both domestic and foreign investment, the residuals are allowed to correlate in order to account for these potentially omitted variables.

Figure 28. Structural model of the investment-growth nexus e

FIN-DEV

INV-DOM

e LABOR GROWTH HUMAN INITIAL

TRADE

150

e

INV-FOR

Investment and Growth: Disentangling Causal Paths The final structural model, with six degrees of freedom, is comfortably over-identified. Data exists to estimate this model on a sample of 78 countries for the period 1975-1995. Although there are too few core or semiperipheral countries in this sample to permit estimation for those zones separately, there are enough peripheral countries (49) to estimate a periphery-only version of the model.

Results on the all-country and periphery-only relationship between

investment and growth are detailed in the next section.

RECIPROCAL MODEL RESULTS In this section I estimate the structural model represented by Figure 28 for all countries for which data are available (n=78) and for all countries that are initially members of the periphery (n=49). For each sample, I estimate structural models using both the unweighed and the population-weighted correlation matrix. Results are detailed below. I begin by concentrating on the revealed relationship between investment and growth, after which I highlight the role of the instrumental variables in the model for growth. I have also estimated the equivalent structural model without instrumental variables (the model represented by Figure 27). The results (not reported) for investment and growth are similar in this model to those of the fully specified model. The introduction of the instrumental variables does not change the general direction or strength of the results.

Relationship between Investment and Growth The results of the structural models confirm that growth has a strong and generally significant causal impact on both domestic and foreign investment flows, while investment flows have virtually no effect on growth. In Figures 29 151

The International Structure of Income and 30 below, I detail the investment-growth nexus from the causal model of Figure 28.

Coefficients have been standardized for ease of comparison.

Statistically significant coefficients (∀ = .05) are marked (*).

Full results,

including causal paths not shown in these detail figures, will presented in full later on in Tables 17 and 18. In the full-sample estimation (Figure 29), the weighted results (right panel) clearly show the positive effect of the growth rate in attracting and promoting investment, both domestic and foreign.

The relatively lower

coefficient for foreign investment is surprising. In the unweighted model, the coefficient is small enough to lose statistical significance (though it is still positive and still five times the effect of foreign investment on growth). This suggests, counterintuitively, that domestic investors are more sensitive to rates of economic growth than are foreign investors.

Figure 29. Investment-growth nexus, all countries, 1975-1995 Unweighted

Weighted by Population e

e

Domestic Investment .104

.372*

Domestic Investment e

Growth rate

.128

0.25

Foreign Investment

152

-.121*

.083

.862*

e

Growth rate

e

.426*

.009

Foreign Investment

e

.059

Investment and Growth: Disentangling Causal Paths An alternative interpretation is that economic growth does not just attract domestic investment, but also generates it directly. It must be remembered that "investment" is measured on the consumption side, not on the income side. It represents outlays on additions to fixed assets (plus net changes in inventories). According to the U.N. system of national accounts, Fixed assets include land improvements (fences, ditches, drains, and so on); plant, machinery, and equipment purchases; and the construction of roads, railways, and the like, including schools, offices, hospitals, private residential dwellings, and commercial and industrial buildings. (World Bank 2001b) Economic growth is likely to generate the need for such outlays organically, and as domestic firms expand their very expansion would be recorded as domestic investment. In contrast, foreign investment is not organically tied to economic performance, although it is apparently attracted to growing economies. This is especially true in the periphery of the world-economy, where the effect of growth on foreign investment is much stronger than in all-nation sample. (Figure 30) When the sample is restricted to peripheral countries only, the relationship of growth with foreign investment comes close to that of growth with domestic investment, in the weighted model. Also in the weighted model, there is a suggestion -- though well below the bar of statistical significance -- that, if foreign investment has an effect on growth, it is negative rather than positive. One surprising aspect of both the all-country and peripheral models is the significantly negative correlation of the residuals of domestic and foreign investment in the unweighted analyses. A positive correlation would have been expected, since any omitted variables that affect investment would be expected to affect domestic and foreign investment in the same direction. The fact that 153

The International Structure of Income this surprising result disappears in the weighted models indicates that it is primarily due to factors affecting (demographically) smaller countries. Further investigation would be necessary to establish the reasons for this anomalous result.

Figure 30. Investment-growth nexus, peripheral countries, 1975-1995 Unweighted

Weighted by Population e

e

Domestic Investment -.095

.612*

Domestic Investment e

Growth rate

.230

-.048

Foreign Investment

-.202*

.037

.765*

e

-.029

Growth rate

e

.680*

-.134

e

Foreign Investment

Full Model Results The results above focus narrowly on the investment-growth nexus. Here, I briefly review the remaining coefficients from the structural equation models. In general, the models for growth are similar to those reported in Chapter 5. In every model, the percent of population that is of working age (LABOR) is the only statistically significant neoclassical input. Of the two instrumental 154

Investment and Growth: Disentangling Causal Paths variables, trade has a significantly positive impact on growth in both of the weighted models, while the effect of financial development on growth is unstable and not significant. Models for domestic and foreign investment came out as anticipated, with FIN-DEV positively correlated with INV-DOM and TRADE positively correlated with INV-FOR. Full results for all models are reported in Table 17 (all countries) and Table 18 (peripheral countries) below. The results for trade are consistent with a growing body of literature in economics on the small but positive contemporaneous effect of trade on growth. One open question, however, is the long-term effect of greater trade openness. It is conceivable that, like foreign investment, foreign trade has a negative longterm effect on growth. This is all the more likely given the strong relationship between foreign trade and investment: if foreign trade contemporaneously promotes foreign investment, and foreign investment has a long-term negative impact, then trade may ultimately have a long-term negative net effect. Careful work is required to trace out these possibilities. The results for financial development are strikingly at odds with the larger economics literature on financial development. In this literature, financial development is consistently shown to have a significantly positive effect on growth. My own experiments show this result to be highly contingent on the choice of period and sensitive to the presence of control variables. While financial development as typically measured is clearly an important variable for predicting some economic outcomes (such as investment rates), it is not clear that it has important implications for growth.

155

The International Structure of Income

Table 17. Full model for growth and investment, 1975-1995 all avail. (1) dep. var.: GROWTH (constant) INITIAL LABOR HUMAN INV-DOM INV-FOR FIN-DEV TRADE R2

Unweighted Coeff. (SE) t-stat -12.96 (3.19) -4.068 -2.13 (0.63) -3.382 .345 (.068) 5.043 .012 (.016) 0.721 .046 (.068) 0.679 .044 (.407) 0.108 -.291 (.273) -1.069 .004 (.010) 0.380 .477

Population weighted Coeff. (SE) t-stat -7.96 (2.05) -3.877 -2.99 (0.41) -7.369 .308 (.041) 7.441 -.003 (.009) -0.346 .038 (.043) 0.880 .015 (.110) 0.139 .056 (.162) 0.346 .006 (.003) 2.112 .879

(2) dep. var.: INV-DOM (constant) FIN-DEV GROWTH R2

Unweighted Population weighted Coeff. (SE) t-stat Coeff. (SE) t-stat 19.49 (1.34) 14.593 18.637 (0.69) 26.995 .951 (.574) 1.657 .674 (.286) 2.361 .831 (.364) 2.286 1.926 (.126) 15.293 .249 .806

(3) dep. var.: INV-FOR (constant) TRADE GROWTH R2

Unweighted Coeff. (SE) t-stat -0.55 (0.18) -3.117 .023 (.002) 11.227 .072 (.064) 1.128 .662

Population weighted Coeff. (SE) t-stat -0.37 (0.24) -1.586 .017 (.002) 7.012 .241 (.054) 4.471 .457

Error cov. for (2) and (3) Error cov.

Unweighted Coeff. (SE)

Population weighted Coeff. (SE) t-stat

156

-.897 (.469)

t-stat -1.912

-.436 (.285)

1.531

Investment and Growth: Disentangling Causal Paths

Table 18. Full model for growth and investment, 1975-1995 periphery (1) dep. var.: GROWTH (constant) INITIAL LABOR HUMAN INV-DOM INV-FOR FIN-DEV TRADE R2 (2) dep. var.: INV-DOM (constant) FIN-DEV GROWTH R2

Unweighted Coeff. (SE) -5.54 (12.19) -0.51 (1.32) .472 (.144) .028 (.028) -.041 (.127) -.105 (.641) -8.359 (4.837) .008 (.021) .369

t-stat -0.447 -0.386 3.291 1.011 -0.320 -0.164 -1.728 0.382

Population weighted Coeff. (SE) t-stat -38.52 (8.58) -4.491 -1.76 (0.76) -2.305 .658 (.117) 5.647 -.008 (.010) -0.775 .016 (.068) 0.235 -.296 (.195) -1.519 3.327 (2.640) 1.260 .030 (.009) .914

3.197

Unweighted Population weighted Coeff. (SE) t-stat Coeff. (SE) t-stat -24.72 -1.443 -12.82 -1.092 (17.136) (11.74) 21.359 2.680 15.664 2.834 (7.971) (5.527) 1.422 (0.433) 3.282 1.776 (0.197) 9.023 .429 .894

(3) dep. var.: INV-FOR (constant) TRADE GROWTH R2

Unweighted Coeff. (SE) t-stat -0.82 (0.32) -2.532 .026 (.004) 6.440 .104 (.083) 1.260 .525

Population weighted Coeff. (SE) t-stat -1.01 (0.33) -3.068 .026 (.004) 6.434 .308 (.055) 5.636 .578

Error cov. for (2) and (3) Error cov.

Unweighted Coeff. (SE)

Population weighted Coeff. (SE) t-stat

-1.52 (0.74)

t-stat -2.061

-0.22 (0.27)

-0.806

157

The International Structure of Income

CONFIRMATION: A PANEL MODEL OF INVESTMENT AND GROWTH The reciprocal effects model seem to definitively preclude any contemporaneous causal path from investment to growth, but this leaves open the possibility of a lagged relationship.

The twenty-year period averages used in the

"contemporaneous" model should, to some extent, capture lagged effects as well, but it is possible as a confirmation to explicitly model the lagged relationships among the variables. In this section, I estimate such lagged effects between investment and growth using the two periods 1975-1985 and 19851995 as prior and current periods, respectively.

The method, as described

briefly in the introduction to this chapter, is to estimate three models of the forms: GROWTHt = A + B0*GROWTHt-1 + B1*INV-DOMt-1 + B2*INVFORt-1 INV-DOMt = A + B0*INV-DOMt-1 + B1*GROWTHt-1 INV-FORt = A + B0*INV-FORt-1 + B1*GROWTHt-1 The B0 coefficients are the stability coefficients for each model, and represent the stability of the dependent variable from its previous period value. The B1 (and B2) coefficients represent the effects of the independent variables, after controlling for the stability of the dependent variable over time. The effects of all other variables are presumed to be subsumed in the stability effect and in the model error. Statistically significant values for the B1 (and B2) coefficients indicate causal relationships. The models detailed above are estimated for the current (t) period 19851995 and prior (t-1) period 1975-1985 in Table 19 below. Results are reported for all available countries (n = 96). Similar results were found for a sample of peripheral countries only (n = 55; not shown). The coefficients for the model 158

Investment and Growth: Disentangling Causal Paths for GROWTH did not change substantially when the effects of INV-DOM and INV-FOR were estimated individually in separate models (not shown).

Table 19. Panel model for growth and investment, 1975-1995 (1) dep. var.: GROWTHa (constant) INV-DOMb INV-FORb GROWTHb R2

Unweighted Coeff. (SE) 0.23 (1.08) .017 (.048) .120 (.220) .335 (.108) .145

(2) dep. var.: INV-DOMa (constant) INV-DOMb GROWTHb R2

Unweighted Coeff. (SE) 7.35 (2.00) .537 (.087) .561 (.194) .439

t-stat 0.213 0.347 0.546 3.093

Population weighted Coeff. (SE) t-stat 0.06 (0.88) 0.064 .059 (.042) 1.410 -.858 (.330) -2.601 .850 (.087) 9.794 .769

t-stat 3.681 6.166 2.891

Population weighted Coeff. (SE) t-stat 6.27 (1.46) 4.305 .592 (.069) 8.522 1.305 (.144) 9.040 .874

(3) dep. var.: Unweighted a INV-FOR Coeff. (SE) t-stat (constant) 0.52 (0.16) 3.180 b INV-FOR .954 (.105) 9.115 GROWTHb -.105 (.049) -2.121 R2 .476 a Current period (1985-1995) b Prior period (1975-1985)

Population weighted Coeff. (SE) t-stat 0.19 (0.14) 1.403 .942 (.162) 5.827 .190 (.028) 6.820 .426

The results correspond closely in direction and magnitude to those reported for the contemporaneous model. Growth in each case significantly predicts investment flows, though in the unweighted model for INV-FOR the coefficient is negative (this indicates that, controlling for 1975-1985 levels of INV-FOR, high growth is correlated with lower later investment). This anomaly 159

The International Structure of Income disappears in the weighted results. On the other hand, nowhere does investment have a significant impact on GROWTH (controlling for prior period GROWTH), except in the case of INV-FOR in the weighted model, where the effect is negative. This echoes the results of the peripheral contemporaneous model (Table 18), in which foreign investment was found to have a negative impact on growth among countries in the peripheral zone. In short, the crosslagged panel models, though less stable than the contemporaneous models, yield generally the same results. The poor stability of the panel model can be ascribed to massive colinearity between current and previous period values of the dependent variables. This problem also affects the Granger tests found in the economics literature. Only the contemporaneous, reciprocal effects model (with adequate instrumentation) fully eliminates this difficulty.

DISCUSSION AND INTERPRETATION The results of this chapter unequivocally confirm the thesis that the arrow of causality runs from growth to investment, with no measurable return causation. Development policy that aims to promote growth by promoting investment is probably just pushing on a string. Policies should aim to promote growth directly, and an appropriate level of new investment will follow. Despite its consistency with emerging results from the empirical economics literature, this finding is likely to be controversial, both because it is counterintuitive and because it flies in the face of several decades of received wisdom about the predictors of economic growth.

The present results

notwithstanding, it is still possible to argue that investment does indeed spur growth, but that the measures used here (and elsewhere) are not reliable enough 160

Investment and Growth: Disentangling Causal Paths to allow detection of the relationship.

The high raw correlation between

investment and growth makes this argument difficult, though not impossible, to sustain. One would have to argue that although exogenous investment shocks do cause economic growth, such shocks are relatively rare. Instead, most of the variation in investment (and relative capital intensity) actually observed among the countries of the world is due to endogenous factors. By this argument, although exogenously-originating changes in the level of physical capital do cause growth, they are swamped in the data by the much more common endogenously-originating changes in the level of physical capital. It is thus possible to square the circle of the investment-growth nexus, but only just. In any case, if exogenous shocks to investment flows are empirically so rare as to be undetectable in the 40 years of data under study here, it begs the policy question of whether or not to bother to pursue them. What the present results leave more obscure than ever is the nature and identity of the variables that do cause growth. Clearly, demographic effects are important. None of the other neoclassical inputs, now studied in a variety of periods on a variety of samples and with a variety of conditioning variables, have such a consistently positive measured effect on growth. Nonetheless, in the unweighted models, about half the variance in growth outcomes remains unexplained.

This optimistic note must be tempered, however, by the

observation that in the weighted models, virtually all of the variance in growth outcomes is accounted for. The structural models specified and estimated in this chapter are a first step toward building a full structural model of the growth process. An early attempt at a full causal model of the development process was made by Hout and Meijerink (1996), and Kentor (2000) uses causal models to study the term structure of the effects of foreign investment, but neither of these articles 161

The International Structure of Income attempts to model the effects of demographic variables.

A marriage of

sociology, economics, and demographics in this area is surely long overdue.

162

7. Conclusion In the Introduction, I laid out my ambition to take some first steps towards integrating the sociology and economics literatures on economic growth. In studying the form of the global income distribution, the differential performance of the neoclassical model by zone of the world-economy, and the (spurious) roles of domestic and foreign investment in fostering economic growth in all three zones of the world-economy, I have attempted to pursue an integrated approach that draws on the strengths of both fields. The new insights generated from the integration of world-systems theory into the neoclassical growth model suggests that similar success might result from adding a sociological dimension into economists' models. In addition, the surprising importance of the labor supply in explaining growth calls for increased attention to the interaction of economic variables with demographic variables such as fertility, migration, and mortality. In short, the general implication of this study is that social policy probably matters much more than economic policy in determining countries' growth rates. Growth may not, however, be very relevant in the broad scheme of things. For the past four decades, individual and national incomes across the world-economy have exhibited a stable pattern that is consistent with the

The International Structure of Income predictions of sociological world-systems theory. At the individual level, there exists a strikingly bimodal pattern in the logged income distribution. The world can truly be said to consist of "haves" and "have-nots," with relatively few people in between. At the national level, there are three discernable income levels, corresponding to the world-systems notions of "core," "semiperiphery," and "periphery." Extreme levels of internal inequality in the semiperiphery of the world-economy indicate that countries falling into this zone of the worldeconomy do, as anticipated, comprise both core- and peripheral-type activities within their borders. These overall patterns in the global distribution of income have been a constant feature of the post-WWII world-economy, even as individual countries have moved up and down the national income ladder. It seems fair to say that these patterns represent the general "structure" of the world-economy. It is at the same time both optimistic and a pessimistic to observe that most meaningful upward mobility in national income levels has come from reductions in birth rates expanding the proportions of populations in poor countries that are of working-age. It is pessimistic because so few countries have historically been able to achieve demographic shift of sufficient scale to bring about transformative economic growth.

The four Asian "Tiger"

economies of South Korea, Taiwan, Hong Kong, and Singapore are the main examples. Moreover, one might assert that these four aren't even "countries" in the statistical sense: all are city-states, or nearly so. Hong Kong and Singapore are clearly cities, Singapore only recently separated from its parent nation of Malaysia and Hong Kong more recently reunited with its parent nation of China. Taiwan and South Korea, however, are also problematic cases. Over half the population of Taiwan lives in the metropolitan area of the capital, Taipei, and almost half the population of South Korea lives in greater Seoul. Throwing in 164

Conclusion Pusan cements the South Korean case: well over half the national population lives in these two urban areas. Why does this matter? It matters because urban areas worldwide are areas of low fertility. Reducing fertility in a small fully-urbanized country is much easier than reducing fertility in a continental nation-state. Consider the case of China today. Shanghai exhibits income per capita levels not too far behind those of Seoul and Taipei. Unlike those latter two cities, however, Shanghai possesses an infinite (for all practical purposes) demographic hinterland.

Native Shanghai residents may adopt urban lifestyles with low

birthrates, but there will always be new immigrants from the hinterland to take up the demographic slack. So despite China's dramatic success in reducing birth rates (which predates the one-child policy by at least a decade), China still has a vast rural population (over 500 million people). Moreover, rural areas nearly everywhere are less productive (economically) than urban areas.

China is

growing rapidly, but does anyone really think that it will grow to developedcountry levels of national income per capita? This would make China four times as rich and powerful as the United States. Such a scenario seems unlikely for this century, or even for the next. The four Asian Tigers seem more an exception than a rule to be followed. Nonetheless, it is to some extent optimistic to observe that reducing the birth rate is the key to transformative growth because at least we know how to reduce the birth rate.

Governments can promote reductions in people's

preferred family sizes through education efforts, encouragement of female labor market participation, and enforcement of women's human rights. They can also help people achieve the smaller families they want through financial support for family planning. These steps are difficult, but at least they are well-understood. Attempting to promote growth through macroeconomic management, on the 165

The International Structure of Income other hand, has proved completely futile. No country under IMF tutelage has done any better than average in growth outcomes. We simply don't know what macroeconomic policies (if any) promote growth. Even the neoinstitutionalist arguments dominating the growth literature today do little more than explain how to avoid outright catastrophe. Neoinsitutionalism has emerged as a newly hegemonic approach in both the sociology and economics literatures on growth. Once again, the two literatures are developing in parallel, with little or no cross-referencing. In any case, neither sociologists nor economists have much to show for their efforts. Dysfunctional national institutions can be disastrous for national economies; that's little more than common sense. How does a country with poor institutions get better ones? We don't know. Do countries with "good" institutions grow remarkably faster than average?

No.

The promotion of healthy national

institutions and good governance is certainly a worthy aim, but even if successful it will not lead to economic growth on such a scale as to transform the peripheral countries of today into the core countries of tomorrow. Formally speaking, the growth models estimated in this volume are "disequilibrium" models, in which increases (decreases) in input levels are hypothesized to lead to growth (decline) towards a new equilibrium level of output. By achieving an ideal mix of labor, physical capital, and human capital countries can maximize their overall levels of national income. Ultimately, however, output is limited by the level of technology available. "Technology" here is to be construed in the broadest sense, as the sum total of all knowledge used in the production process. For some countries in the core zone of the world-economy, which already have access to the most advanced productive technologies available anywhere in the world, long-term growth prospects may indeed be limited by the rate of advance of technology. For most of the nations 166

Conclusion of the world, however, the hard output barrier represented by the level of available technology in the world at a given date in history is so far from present levels of output as to be irrelevant for assessing current growth potential. In these countries, substantial opportunities for equilibrium growth must exist. Equilibrium growth is driven by changes in technology, again, broadly construed. I suspect that the most important element in this "technology" of economic production may be something we rarely think of under that rubric: the social organization of society itself. After all, most of the physical technology of the world is available anywhere in the world for a market price. There is certainly some increased burden associated with peripheral locations, but it beggars belief that barriers to physical technology transfer could be so high as to account for the 100-fold national income differential between, say, western Europe and west Africa.

Such a massive gap in productivity can only be

accounted for socially: people in Europe are far better organized (from an economic standpoint) then people in Africa. In this sense, the argument harkens all the way back to Adam Smith. The main message of The Wealth of Nations is not the mysterious wisdom of the invisible hand, but the irrefutable logic of the division of labor. Adam Smith is invoked tirelessly by economists, but is much more often read today by sociologists. Whatever its determinants, truly transformative economic growth, growth on a scale to make poor countries rich, the kind of growth that is the only fair outcome for the 80% of the world's population living outside today's core of the world-economy, is extraordinarily rare. Today's economic fault lines were drawn in the nineteenth century and through two world wars, the Great Depression, and the rise and fall of Communism have changed very little since then.

Granted, any real income convergence bringing today's peripheral

countries up to core consumption levels will devastate the world's environment 167

The International Structure of Income through runaway global warming and natural resource depletion, but it is inconceivably selfish to use this as an excuse to keep 80% of the world in poverty. Given that we will not reduce our consumption levels to theirs, we must work to raise theirs to ours, whatever the ecological consequences. The only thing delaying (though in the long run certainly not preventing) complete ecological catastrophe is that we don't know how to do it. That is perhaps the most important lesson to be learned from this study: we know very little about any of the topics covered here. The dynamics of the global income distribution; how growth is patterned by income levels; how growth itself affects income levels, economic growth itself ... despite decades of intensive academic research, all remain mysterious. Existing theoretical models do not stand up to a careful examination of the empirical facts. Clearly, new models are needed. Where they will come from is anyone's guess, but it seems unlikely that they will come from research that continues in existing disciplinary traditions. The facts cry out for radically new interpretations. These may arise out of today's separate disciplinary debates, but both sociologists and economists have much to learn from each other about the mechanics of income and growth, if they are willing to listen. That said, the discourse between them will be most profitable if it operates in both directions. Hopefully, we will pay greater heed to each others' ideas in the future than we have in the past.

168

References Ahmad, Sultan. 1994. "Improving Inter-spatial and Inter-temporal Comparability of National Accounts." Journal of Development Economics 44: 53-75. Aitchison, J. and J.A.C. Brown. 1957. The Lognormal Distribution: With Special Reference to its Uses in Economics. London: Cambridge U.P. Alderson, Arthur S., and Francois Nielsen. 1999. "Income Inequality, Development, and Dependence: A Reconsideration." American Sociological Review 64: 606-631. Amin, Samir. 1974. Accumulation on a World Scale. New York: Monthly Review Press. Arrighi, Giovanni. 1994. The Long Twentieth Century: Money, Power, and the Origins of Our Times. New York: Verso. Arrighi, Giovanni, and Jessica Drangel. 1986. "The Stratification of the WorldEconomy: An Exploration of the Semiperipheral Zone." Review 10: 9-74. Arrighi, Giovanni, Satoshi Ikeda, and Alex Irwan. 1993. "The Rise of East Asia: One Miracle or Many?" Pp. 41-65 in Ravi A. Palat, ed., Pacific-Asia and the Future of the World-System. Westport, CT: Greenwood Press. Babones, Salvatore J. 2002. "Population and Sample Selection Effects in Measuring International Income Inequality." Journal of World-Systems Research 8: 1-22.

The International Structure of Income Babones, Salvatore J. 2003. "One World or Two? A Snapshot of the Global Income Distribution." Paper presented at the 98th Annual Meeting of the American Sociological Association. Barro, Robert J. 1991. "Economic Growth in a Cross-Section of Countries." Quarterly Journal of Economics 106: 407-443. Barro, Robert J. 1997. Determinants of Economic Growth: A Cross-Country Empirical Study. Cambridge, MA: The MIT Press. Barro, Robert J., and Jong-Wha Lee. 1993a. "International Comparisons of Educational Attainment." Journal of Monetary Economics 32: 363-394. Barro, Robert J., and Jong-Wha Lee. 1993b. "International Measures of Schooling Years and Schooling Quality." American Economic Review 86, Papers and Proceedings: 218-223. Barro, Robert J., and Xavier Sala-I-Martin. 1995. Economic Growth. New York: McGraw-Hill. Benhabib, Jess, and Mark M. Spiegel. 1994. "The Role of Human Capital in Economic Development: Evidence from Aggregate Cross-Country Data." Journal of Monetary Economics 34: 143-173. Blomstrom, Magnus, Robert E. Lipsey, and Mario Zejan. 1996. "Is Fixed Investment the Key to Economic Growth?" Quarterly Journal of Economics 111: 269-276. Bloom, David, and Jeffrey G. Williamson. 1998. "Demographic Transitions and Economic Miracles in Emerging Asia." World Bank Economic Review 12: 419-455. Carroll, Christopher D., Jody R. Overland, and David N. Weil. 2000. "Saving and Growth with Habit Formation." American Economic Review 90: 341-355. Champernowne, D.G., and F.A. Cowell. 1998. Income Distribution. Cambridge: Cambridge U.P.

Economic Inequality and

Chase-Dunn, Christopher. 1975. "The Effects of International Economic Dependence on Development and Inequality: A Cross-National Study." American Sociological Review 40: 720-738.

170

References Chase-Dunn, Christopher. 1998. Global Formation: Structures of the WorldEconomy. Oxford: Basil Blackwell. Chase-Dunn, Christopher, and Thomas D. Hall. 1997. Rise and Demise: Comparing World-Systems. Boulder, CO: Westview Press. Cowell, F.A. 1995. Measuring Inequality, 2nd ed. New York: Prentice Hall. Creedy, John. 1985. Dynamics of Income Distribution. New York: Blackwell. Crowly, Angela Martin, James Rauch, Susanne Seagrave, and David A. Smith. 1998. "Quantitative Cross-National Studies of Economic Development: A Comparison of the Economics and Sociology Literatures." Studies in Comparative International Development 33/2: 30-57. Deininger, Klaus, and Lyn Squire. 1996. "A New Data Set Measuring Income Inequality." The World Bank Economic Review 10: 565-591. Deininger, Klaus, and Lyn Squire. 1998. "New Ways of Looking at Old Issues: Inequality and Growth." Journal of Development Economics 57: 259-287. Deininger, Klaus, and Lyn Squire. 2001. "A New Data Set Measuring Income Inequality." MRDF, The World Bank. De Long, J. Bradford, and Lawrence H. Summers. 1991. "Equipment Investment and Economic Growth." Quarterly Journal of Economics 106: 445502. Dixon, William J., and Terry Boswell. 1996a. "Dependency, Disarticulation, and Denominator Effects: Another Look at Foreign Capital Penetration." American Journal of Sociology 102: 543-562. Dixon, William J., and Terry Boswell. 1996b. "Differential Productivity, Negative Externalities, and Foreign Capital Dependency: Reply to Firebaugh." American Journal of Sociology 102: 576-584. Easterly, William R. 2001. The Elusive Quest for Growth: Economists' Adventures and Misadventures in the Tropics. Cambridge, MA: MIT Press. Emmanuel, Arghiri. 1972. Unequal Exchange: A Study of the Imperialism of Trade. New York: Monthly Review Press. Fields, Gary S. 2001. Distribution and Development: A New Look at the Developing World. New York: Russell Sage. 171

The International Structure of Income Firebaugh, Glenn. 1992. "Growth Effects of Foreign and Direct Investment." American Journal of Sociology 98: 105-130. Firebaugh, Glenn. 1996. "Does Foreign Capital Harm Poor Nations? New Estimates Based on Dixon and Boswell's Measure of Capital Penetration." American Journal of Sociology 102: 563-575. Firebaugh, Glenn. 1999. "Empirics of World Income Inequality." American Journal of Sociology 104: 1597-1630. Firebaugh, Glenn. 2000. "Observed Trends in Between-Nation Income Inequality and two Conjectures." American Journal of Sociology 106: 215-221. Frank, Andre Gunder. 1969. Latin America: Underdevelopment or Revolution. New York: Monthly Review Press. Frankel, Jeffrey A., and David Romer. 1999. "Does Trade Cause Growth?" American Economic Review 89: 379-399. Good, David F., and Tongashu Ma. 1999. "The Economic Growth of Central and Eastern Europe in Comparative Perspective, 1870-1989." European Review of Economic History 2: 103-137. Gustafsson, Bjorn, and Mats Johansson. 1999. "In Search of Smoking Guns: What Makes Income Inequality Vary over Time in Different Countries?" American Sociological Review 64: 585-605. Harberger, Arnold C. 1998. "A Vision of the Growth Process." American Economic Review 88: 1-32. Heston, Alan. 1994. "National Accounts: A Brief Review of Some Problems in Using National Accounts Data in Level of Output Comparisons and Growth Studies." Journal of Development Economics 44: 29-52. Heston, Alan, Robert Summers, and Bettina Aten. 2001. Penn World Table (PWT) Mark 6.0 (MRDF). Center for International Comparisons, University of Pennsylvania. Hout, Will, and Frits Meijerink. 1996. "Structures in the International Political Economy: World System Theory and Unequal Development." European Journal of International Relations 2: 47-76.

172

References Kentor, Jeffrey. 2000. Capital and Coercion: The Economic and Military Processes that Have Shaped the World Economy 1800-1990. New York: Garland Publishing. Kick, Edward L. 1987. "World-System Structure, Nationalist Development, and the Prospects for a Socialist World Order." Pp. 127-155 in Terry Boswell and Albert Bergesen, eds., America's Changing Role in the World-System. New York: Praeger. King, Robert G., and Ross Levine. 1993. "Finance and Growth: Schumpeter Might Be Right." Quarterly Journal of Economics 108: 717-737. Korzeniewicz, Roberto Patricio, and William Martin. 1994. "The Global Distribution of Commodity Chains." Pp. 67-91 in Gary Gereffi and Miguel Korzeniewicz, eds., Commodity Chains and Global Capitalism. Westport, CT: Praeger. Korzeniewicz, Roberto Patricio, and Timothy Patrick Moran. 1997. "WorldEconomic Trends in the Distribution of Income, 1965-1992." American Journal of Sociology 102: 1000-1039. Korzeniewicz, Roberto Patricio, and Timothy Patrick Moran. 2000. "Measuring World Income Inequalities." American Journal of Sociology 106: 209-214. Kuznets, Simon. 1955. "Economic Growth and Income Inequality." American Economic Review 45: 1-28. Lal, Deepak, and H. Myint. 1996. The Political Economy of Poverty, Equity, and Growth: A Comparative Study. Oxford: Clarendon Press. Leibenstein, Harvey. 1954. A Theory of Economic-Demographic Development. Princeton, NJ: Princeton UP. Leibenstein, Harvey. 1957. Economic Backwardness and Economic Growth. New York: John Wiley. Levine, Ross. 1997. "Financial Development and Economic Growth: Views and Agenda." Journal of Economic Literature 35: 688-726. Levine, Ross, and David Renelt. 1992. "Sensitivity Analysis of Cross-Country Growth Regressions." American Economic Review 82: 942-963. 173

The International Structure of Income Li, Hongyi, Lyn Squire, and Heng-fu Zou. 1998. "Explaining International and Intertemporal Variations in Income Inequality." The Economic Journal 108: 2643. London, Bruce. 1988. "Dependence, Distorted Development, and Fertility Trends in Noncore Nations: A Structural Analysis of Cross-National Data." American Sociological Review 53: 606-618) Lucas, Robert E. 1988. "On the Mechanics of Economic Development." Journal of Monetary Economics 22: 3-42. Lucas, Robert E. 1993. "Making a Miracle." Econometrica 61: 251-271. Maddison, Angus. 1995. Monitoring the World Economy, 1920-1992. Paris: OECD. Mankiw, N. Gregory, David Romer, and David N. Weil. 1992. "A Contribution to the Empirics of Economic Growth." Quarterly Journal of Economics 107: 407-437. McClelland, David C. 1961. The Achieving Society. Princeton: Van Nostrand. Meyer, John W., John Boli-Bennett, and Christopher Chase-Dunn. 1975. "Convergence and Divergence in Development." Annual Review of Sociology 5: 223-246. Milanovic, Bronko. 2002. "The World Income Distribution, 1998 and 2003: First Calculation Based on Household Income Surveys Alone." The Economic Journal 112: 51-92. Nemeth, Roger, and David A. Smith. 1985. "International Trade and WorldSystem Structure: A Multiple Network Analysis." Review 8: 517-560. Pritchett, Lant. 1997. Perspectives 11: 3-17.

"Divergence, Big Time."

Journal of Economic

Romer, Paul M. 1986. "Increasing Returns and Long-Run Growth." Journal of Political Economy 94: 1002-1037. Romer, Paul M. 1987. "Crazy Explanations for the Productivity Slowdown." NBER Macroeconomics Annual 2: 163-202. Romer, Paul. 1994. "The Origins of Endogenous Growth." Economic Perspectives 8: 3-22. 174

Journal of

References Sachs, Jeffrey D., and Andrew M. Warner. 1997. "Fundamental Sources of Long-Run Growth." American Economic Review 87, Papers and Proceedings: 184-188. Samuelson, Paul A. 1948. "International Trade and the Equalization of Factor Prices." Economic Journal 58: 163-184. Schultz, T. Paul. 1998. "Inequality in the distribution of personal income in the world: How it is changing and why." Journal of Population Economics 11: 307344. Schumpeter, Joseph A. 1951. The Theory of Economic Development: An Enquiry into Profits, Capital, Credit, Interest, and the Business Cycle. Cambridge, MA: Harvard UP. Scott, Maurice F. 1989. A New View of Economic Growth. Oxford: Oxford UP. Smith, David A. and Douglas R. White. 1991. "Structure and Dynamics of the Global Economy" Network Analysis of International Trade 1965-80." Social Forces 70: 857-893. Snyder, David, and Edward Kick. 1979. "Structural Position in the WorldSystem and Economic Growth, 1955-1970: A Multiple-Network Analysis of Transnational Interactions." American Journal of Sociology 84: 1096-1126. Solow, Robert M. 1956. "A Contribution to the Theory of Economic Growth." Quarterly Journal of Economics 70: 65-94. Solow, Robert. 1957. "Technical Change and the Aggregate Production Function." Review of Economics and Statistics 39, 312-320. Summers, Robert, and Alan Heston. 1991. "The Penn World Table (Mark 5): An Expanded Set of International Comparisons, 1950-1988." Quarterly Journal of Economics 106: 327-368. Summers, Robert, Alan Heston, Bettina Aten, and Daniel Nuxall. 1994. Penn World Table (PWT) Mark 5.6a Data (MRDF). Center for International Comparisons, University of Pennsylvania. Temin, Peter. 1997. "The Golden Age of European Growth: A Review Essay." European Review of Economic History 1: 127-149. 175

The International Structure of Income Terlouw, Cornelius P. 1992. The Regional Geography of the World System: External Area, Periphery, Semi-periphery, Core. Utrecht: Faculteit Ruimtelijke Wetenschappen. Terlouw, Cornelius P. 1993. "The Elusive Semiperiphery: A Critical Examination of the Concept Semiperiphery." International Journal of Comparative Sociology 34: 87-102. Van Rossem, Ronan. 1996. "The World-System Paradigm as General Theory of Development: A Cross-National Test." American Sociological Review 61: 508-527. Wallerstein, Immanuel M. 1974. The Modern World-System, Vol. I: Capitalist Agriculture and the Origins of the European World-Economy in the Sixteenth Century. New York: Academic Press. Wallerstein, Immanuel M. 1979. The Capitalist World-Economy. Cambridge: Cambridge U.P. Wallerstein, Immanuel M. 1989. The Modern World-System, Vol. III: The Second Era of Great Expansion of the Capitalist World-Economy, 1730-1840s. New York: Academic Press. Williamson, Jeffrey G. 1998. "Growth, Distribution, and Demography: Some Lessons from History." Explorations in Economic History 35: 241-271. World Bank. 2001a. Finance for Growth: Policy Choices in a Volatile World. Oxford: Oxford UP. World Bank. 2001b. World Development Indicators. Washington: World Bank.

176

Index Asian Tigers, 164 Capital, 2, 107

conditional, 9 Core, 4, 5, 61, 63, 86, 134

domestic, 121

Demographic transition, 111

flows, 120, 121

Dependency theory, 6, 7, 9, 104, 141

foreign, 121

Diminishing returns, 106, 107

human, 5, 9, 106, 116, 131, 134,

Disequilibrium growth, 107, 119,

140, 143 physical, 5, 106, 113, 130, 134,

127, 143, 166 Divergence, 3, 10, 59

140, 143

Division of labor, 167

stocks, 113

Education, 9, 116, 131

China, 126, 165 effect of, 22 one-child policy, 111

Equilibrium growth, 107, 119, 127, 167 Exchange rates

Colonialism, 7

market, 12, 22

Commodity chains, 3

purchasing power parity, 12, 22

core nodes, 3

Financial development, 149, 155

peripheral nodes, 3

Gaussian kernel, 51, 82

Convergence, 2, 8, 20, 59, 107 absolute, 10

Gini Definition, 38

The International Structure of Income Relationship to Lorenz curve, 37 relationship to VarLog, 42, 44

153, 155 Invisible hand, 167

Global warming, 168

Kuznets curve, 47, 99

Heteroskedasticity, 126

Labor, 5, 106, 109, 130, 134, 140,

Income

143, 154

bimodal distribution of, 4, 53, 103

Lorenz curve, 37

global distribution of, 10, 35, 53

Low-level equilibrium trap, 111

lognormal distribution of, 43, 51

Mobility, 86, 103

lognormal distribution of, 3

National income, 2

trimodal distribution of, 4, 64, 85, 103 Income quintiles, 38 Inequality between-country, 10, 12, 18

GDP/PPP, 11, 13 GNP/FX, 11, 13, 79 GNP/RLC, 15, 79, 109 Neoclassical economics, 2 Neoclassical growth model, 2, 4,

Gini, 19, 21, 41

104, 105, 119, 129, 141, 143, 145,

global, 10, 35

147

international, 10

Neoinstitutionalism, 166

national, 47

Network analysis, 71

VarLog, 19, 20, 40

blockmodeling, 72

within-country, 10

role equivalence, 72, 73

Instrumental variables, 146, 148, 151 Investment, 6, 113, 123, 143, 145, 151 domestic, 123, 136, 140, 143, 146, 153, 155 foreign, 6, 123, 136, 140, 146, 178

structural equivalence, 72 New growth theory, 9 Organic

zones

of

the

world-

economy, 96 Perimeter of the core, 85, 120, 134 Perimeter of the periphery, 85, 120, 134

Index Periphery, 4, 5, 61, 63, 86, 134, 151, 153 Reciprocal effects, 146 Reverse causality, 143, 160 Semiperiphery, 4, 61, 63, 68, 86, 98,

Technology, 5, 9, 106, 166 Trade, 149, 155 VarLog relationship to Gini, 42, 44 World-systems theory, 2, 3, 63, 104

134

179

The International Structure of Income

180

About the Author Salvatore Babones is a Lecturer in the Department of Sociology and Social Policy at The University of Sydney.

From 2003 - 2008 he was Assistant

Professor of Sociology, Public Health, and Public and International Affairs at the University of Pittsburgh. He is the editor of Social Inequality and Public Health (2009 The Policy Press) and co-editor with Christopher Chase-Dunn of Global Social Change: Historical and Comparative Perspectives (2006 The Johns Hopkins University Press). His work has been published in sociology, political science, public health, and interdisciplinary journals and he is currently the editor for global development studies of the International Studies Association's Compendium of International Studies. His research focuses on income inequality, its roots in our globalizing world, and its consequences for human health. He holds both the MSE in mathematical sciences (2002) and the Ph.D. in sociology (2003) from The Johns Hopkins University. He now resides in Sydney, Australia.