E77q7 gsct fndings New ways of looking at inequality and growth Contrary to popular belief, economic growth does not systematically increase inequality. But very unequal land distribution does hinder growth. The challenge is to address this problem without undermining investment, which is crucial to growth and to improving the incomes of the poor. Research using greatly improved data has States). Subsequent research used larger data drawn some surprising conclusions about the sets but included many observations of ques- BECAUSE relationship between inequality and growth. tionable quality-for example, synthetic esti- Contrary to the famous proposition first put mates derived from national accounts data or forward by Simon Kuznets in 1955-that the surveys that covered urban but not rural INCOME early stages of growth exacerbate inequality- areas. Some surveys were flawed because they INEGUALITY new research shows that growth does not covered only cash income, excluding produc- consistently affect inequality either way and tion for household consumption. To avoid TEND TO BE that changes in inequality tend to be small. these problems, we set three minimum stan- As a result growth almost always improves the dards for data quality: SMALL, GROWTh incomes of the poor. But while growth does * Observations should be based on surveys ALMOST not consistently affect inequality, inequality * The surveys should be nationally repre- , , ~~~~~~~~~~~~~ALWAYS does affect growth. In general, developing sentative countries with a more equal distribution of * The surveys should encompass all impor- BENEFITS THE assets-especially land-have grown more tant types of income rapidly than countries with a less equal distri- Applying these tests to existing data sets bution of assets. reveals problems of sufficient magnitude to This note explains how the data used to call into question previous results. For exam- reach these conclusions differ from those used ple, data assembled by Shail Jain have been in previous research on these issues, explains widely used to support the Kuznets hypothe- the findings, and discusses the implications of sis, but only 61 of 405 observations (15 per- these findings for development policy. cent) satisfied our criteria (figure 1). Similarly, of the 55 data points collected by Felix Better data on inequality Paukert that have recently been used to demonstrate a negative relationship between Previous empirical work on the relationship initial inequality and subsequent growth, only between inequality and growth was ham- 18 (33 percent) are of acceptable quality. The pered by inadequate data. Kuznets himself data assembled by Gary Fields used standards relied on historical data from the first half of broadly similar to ours, and 73 of 105 (about the nineteenth century from just three coun- 70 percent) meet our quality standards. Yet tries (England, Germany, and the United few of the acceptable observations are from FROM THE DEVELOPMENT ECONOMICS VICE PRESIDENCY OF THE WORLD BANK NO. 28 FEBRUARY 1997 the 1960s, and there are very few comparable stable within regions and countries over time (fig- observations for a single country over time. ure 2). Income inequality is much greater in Compiling additional high-quality data from Latin America and Sub-Saharan Africa, where other research and official statistical publications Gini coefficients are in the upper forties, than in greatly expanded the data set; however, we were East and South Asia, where coefficients are in the still left with few observations from the 1960s. middle to upper thirties. OECD countries gener- Partly to address this problem, we next com- ally have egalitarian distributions of income, with piled data on the distribution of land, which are Gini coefficients just above thirty. Income more readily available and subject to fewer data inequality was lowest in Eastern European coun- problems than are income distribution data. We tries under socialism; the region's Gini coeffi- collected 261 observations of acceptable quality cients began to rise in 1980. The Gini coefficients from 103 countries, including 73 countries for for land distribution show similar variations which we had observations in 1960. Practical across regions and are even more stable over time. advantages aside, we anticipated that land distri- The new data provide a much improved bution might have a greater impact on future empirical basis for an investigation of the rela- growth than income distribution, for reasons tionship between growth, inequality, and pover- described below. ty. We examined two questions: how growth In measuring income distribution and land affects inequality and how inequality affects distribution we used the Gini coefficient, a 100 growth. Distributions of point scale in which 1 represents perfect equali- land and income ty and 100 represents absolute inequality (that can be very is, all income or land belongs to a single house- How does growth affect hold). Two characteristics immediately became inequality? different evident when we compiled the data set. First, distributions of land and income can be very Much of the vast empirical literature on the different. For example, India, Indonesia, and the Kuznets hypothesis was motivated by concern Republic of Korea all have Gini coefficients for that the poor might lose out from the early the distribution of income in the thirties, but stages of development, and many researchers the coefficients for the distribution of land range believed that they found evidence that this is the from 63 for India to 55 for Indonesia to 35 for case. Yet the analyses were doubly flawed. In Korea. This finding suggests that tests of the addition to the data quality problems described relationship between initial inequality and sub- earlier, the lack of data from the 1960s meant sequent growth may yield different results that researchers had to use variations in per capi- depending on whether the definition of initial ta income across countries to represent increases inequality is based on income or land. in per capita income over time within a country. Second, measures of inequality differ widely across regions and countries but remain relatively Figure 2. Income inequality differs across regions but changes little over time Figure 1. Few data sets provide adequate Gini coefficient for income data 60 Percent 100 Latin America 500 50 and the Cacibbean 80 3 S.b-Shr 60 40 ~- 1 c 40 30 20 uoead Cenrr Asi,a 0 20 Jain (I1975) Paukert (I1973) Fields (I1989) 1960 1970 1980 1990 Source: Authors' calculations. Using our new data set, it is possible for the no impact on subsequent growth. Higher first time to test the Kuznets hypothesis within inequality in land distribution, on the other countries over time. We found that growth does hand, does result in lower subsequent growth not consistently affect equality either way. Of the (figure 4). Only two of the fifteen developing eighty-eight cases where there was economic countries with a Gini coefficient for land distri- growth for a decade, income equality improved bution above 70 managed to grow by more than slightly in about half the cases and worsened 2.5 percent a year during 1960-92. slightly in the other half; changes over time were Why is a more equal distribution of land good much smaller than the differences across coun- for growth? One likely explanation involves cred- tries. More important, we found a strong system- it. We know that investment is crucial for growth. atic relationship between aggregate growth and The poor often cannot invest because they lack growth in the income of the poorest fifth of the capital and the collateral to borrow. In countries population, with this group's income rising in 85 with a very unequal distribution of land, poor percent of the eighty-eight growth episodes. In people may find it difficult or impossible to general, the higher is the rate of economic invest-even in their own health and education. growth, the larger is the improvement in the Two pieces of evidence support this explanation. incomes of the poor (figure 3). This finding sug- First, although land distribution is a significant gests that even when inequality worsened, its neg- determinant of future growth in developing ative effect on the poor was usually outweighed countries, this is not the case in high-income _ by the positive effect of aggregate growth. countries, where there are alternative forms of Poverty reduction is collateral. Second, among developing countries, better served by the more unequal is the land distribution, the How does inequality affect lower is the average educational attainment. increasing growth? investment than by Recent empirica work has suggested anegative Redistribution or investment? redistributing land Recent empirical work has suggested a negative association between initial inequality and future These findings present policymakers with a growth. If confirmed, this would imply that dilemma. On the one hand, since growth bene- unequal economies will experience lower rates of fits the poor and productive investment is neces- growth and, in general, lower rates of poverty sary for growth, policies that encourage invest- reduction. As we have seen, however, the data ment are crucial to reducing poverty. On the underlying these results are of questionable other hand, growth is constrained by an unequal quality. distribution of land; thus, where land distribu- Using the new database, we find that higher tion is highly unequal, redistribution also should inequality in income distribution seems to have help reduce poverty. Figure 3. The faster a country grows, the Figure 4. Countries with more equal land more the poor benefit distribution grow faster Percentage change in the income of the poor Average annual growth, 1960-90 4 Japan. Tha,land *2Malays,a 3 - 4 2 2 B I _ _ I~~~~~~~~~~~~~~~~~~~~~~~~ * +Togo _ _ - O Venezuelal aq Senegal, O L I S5 25 35 45 SS 65 < I.5 1.5-2.5 > 2.5 Initial land distribution Average GDP growth of three country groups (percent) M More equal Source: Authors' calculations, Source: Authors' calculations. While investment and redistribution need not income, can impede rapid growth, implying that be mutually exclusive, it is useful to have some redistributive policies could enhance growth. sense of how much each can contribute to the Third, the potential of redistributive policies to reduction of poverty. We investigated this issue benefit the poor can be realized only if redistribu- by examining how investment and land distribu- tion does not jeopardize productive investment. tion affect income growth for different income -Lyn Squire and Klaus Deininger quintiles, especially the poorest fifth. We found that while investment and redistribution are Further reading both potentially good for the poor, investment is much more powerful: increasing investment by a Alesina, Alberto, and Dani Rodrik. 1994. "Distributive standard deviation improves the incomes of the Politics and Economic Growth." Quarterly Journal of poor by almost twice as much as reducing land Economics 108: 465-90. inequality by a standard deviation. Anand, Sudhir, and S. M. R. Kanbur. 1993. "Inequality What does this mean for policymakers? In and Development: A Critique." Journal of Development countries where land distribution is very Economics41(1): 19-43. unequal, the ideal solution would be land redis- Clarke, George R. G. 1993. "More Evidence on Income tribution combined with increased investment. Distribution and Growth." Journal of Development But redistribution that comes at the expense of Economics 47(2): 403-27r investment may have a negative overall effect on Deininger, Klaus, and Lyn Squire. 1996a. "A New Data Set Appropriate data the poor by reducing growth. Where land redis- Measuring Income Inequality." The World Bank can provide tribution cannot be implemented without nega- Economic Review 10(3): 565-91. uive consequences for investment, a better - . 1996b. "New Ways of Looking at Old Issues: important insights approach would be to improve poor people's Inequality and Growth." World Bank, Policy Research into the dynamics access to productive assets through education, Department, Washington, D.C. of growth and health care, and microcredit schemes. . Forthcoming. "Does Inequality Matter? Re-exam- ining the Links between Growth and Inequality." Three conclusions Finance and Development. Fields, Gary. 1989. "A Compendium of Data on Inequality Appropriate data, which are increasingly becom- and Poverty for the Developing World." Cornell ing available, can provide important insights into University, Ithaca, N.Y. the dynamics of growth and poverty reduction. Jain, Shail. 1975. Size Distribution of Income: A These insights have led us to three broad conclu- Compilation of Data. Washington, D.C.: World Bank. sions. First, although policymakers should pay Paukert, Felix. 1973. "Income Distribution at Different attention to the distributional consequences of Levels of Development: A Survey of the Evidence." different policy options, the fear that economic International Labor Review 108: 97-125. growth will have a systematically negative effect Persson, Torsten, and Guido Tabellini. 1994. "Is Inequality on the distribution of income is ill-founded. Harmful for Growth?"AmericanEconomic Review 84(3): Second, unequal distribution of assets, more than 600-21. This DECnote was prepared by Lyn Squire and Klaus Deininger in the Policy Research Department of the World Bank. DECnotes trans- mit key research findings to Bank Group managers and staff They are drawn from the work of individual Bank researchers and do not necessarily represent the views of the World Bank and its member countries-and therefore should not be attributed to the World Bank or its affiliates. DECnates are produced by the Research Advisory Staff. We welcome your questions and comments; please e-mail them to the authors or to Evelyn Alfaro, RAD. Prepared for World Bank staff