The Middle Class Consensus and Economic Development Abstract: A middle class consensus is defined as a high share of income for the middle class and a low degree of ethnic divisons. The paper links a middle class consensus to resource endowments, along the lines of the provocative thesis of Engerman and Sokoloff 1997 and 2000. This paper exploits this association using tropical resource endowments as instruments for inequality. A higher share of income for the middle class and lower ethnic divisions are associated with higher income and higher growth, as well as with more education, better health, better infrastructure, better economic policies, less political instability, less civil war and ethnic minorities at risk, more social “modernization” and more democracy. Keywords: Inequality, Ethnic Divisions, Economic Development, Economic Growth, Human Capital JEL Codes: O1, O4, D31, I0, Z13 William Easterly* World Bank July 2001 * Views expressed here are not necessarily of the views of the World Bank or its member governments. I am grateful for comments by two anonymous referees, the editor and associate editor, Thorsten Beck, Stan Engermann, Karla Hoff, Aart Kraay, Ross Levine, Branko Milanovic, Guy Pfeffermann, Maurice Schiff, and Ken Sokoloff and from participants in seminars at Georgetown University, the Inequality Workshop at the World Bank, the University of Copenhagen, and for comments on related work by participants in the Munich CESifo/University of Washington conference on inequality. 2 “Thus it is manifest that the best political community is formed by citizens of the middle class, and that those states are likely to be well-administered, in which the middle class is large .. where the middle class is large, there are least likely to be factions and dissension.” Aristotle 306 BC (quoted in Decornez 1998) “So strong is this propensity of mankind to fall into mutual animosities, that where no substantial occasion presents itself, the most frivolous and fanciful distinctions have been sufficient to kindle their unfriendly passions and excite their most violent conflicts. But the most common and durable source of factions has been the various and unequal distribution of property.” (James Madison 1787, Federalist Papers No. 10) “Many of the world's problems stem from the fact that it has 5,000 ethnic groups but only 190 countries.”1 (Rodger Doyle, Scientific American, September 1998) I. Literature review and discussion Many explanations of the cross-country differences in economic growth and development only lead to further questions. If differences in saving rates explain cross-country income differences (Mankiw, Romer, and Weil 1992), then why do some societies save more than others? If national policies explain much of the differences in growth rates across countries (Barro and Sala-I-Martin 1995, Easterly and Levine 1997), then why do some nations have worse government policies than others? The dissatisfaction with explanations of cross-country development differences by endogenous variables has led to a recent search in the literature for more fundamental characteristics of nations that determine development outcomes. This paper puts forward the existence of a middle class consensus as a critical determinant of development differences. A middle class consensus is defined as a national situation where there are neither strong class differences nor ethnic differences. The paper links the existence of a middle class consensus to exogenous country characteristics like resource endowments and ethnolinguistic diversity. Political economy explanations of development outcomes usually focus on “society’s polarization and degree of social conflict” (Alesina 1994, p. 38). Societies that are polarized tend to focus on redistribution between polarized factions that alternate in power; societies that are not polarized are able to reach a consensus on public goods and overall economic development. Casual observation and previous literature suggests that among the most common polarizing 3 forces are differences between classes and differences between ethnic groups. The ethnic or class elite in power at any given moment may not want to invest in the human capital of the majority because that would increase the political voice of the majority and weaken the elite’s hold on power. Two recent strands of the literature in particular motivate this paper. First, there is the literature on resource endowments, the size of the middle class (or conversely, the absence of high inequality), and economic development. Second, there is the literature on ethnic diversity and growth and development. Economic historians have pointed out the importance of a large middle class for economic development. Landes 1998 says the “ideal growth and development society” would have “a relatively large middle class” (pp. 217-18). He cites “the great English middle class” as a reason for England’s being first at industrialization (p. 221). Adelman and Morris 1967 noted that “in the economic development of Western Europe, the middle classes were a driving force”. Moreover, they presciently said that “it is clear from many country studies that the growth of a robust middle class remains of crucial importance in contemporary low-income nations.”2 A large theoretical literature also links a small middle class – or other measures of inequality -- to low growth and low human capital accumulation. Galor and Zeira 1993 postulate that the poor are liquidity-constrained from accumulating human capital; higher inequality implies a greater share of the population will be liquidity constrained and thus the society accumulates less human capital. Alesina and Rodrik 1994 and Persson and Tabellini 1994 link high inequality to low growth through the poor majority imposing a tax on the rich. Perotti 1996 did not find direct evidence that inequality increased taxes, but did find evidence for the effect of inequality on growth due to the effect through human capital and through political instability. 4 A number of empirical cross-country studies find high inequality linked to poor growth outcomes (see in addition to others mentioned above Clarke 1995 and Deininger and Squire 1998, the latter using land inequality).3 The papers that directly inspired the present paper are Engerman and Sokoloff 1997 and Sokoloff and Engerman 2000. Engerman and Sokoloff link tropical commodity factor endowments (which had significant scale economies) in Latin American countries to the concentration of wealth in the hands of a small elite, which in turn led to their entrenchment in power. Once entrenched, the elite was reluctant to invest in mass human capital for fear that they would be displaced from power (Bourguignon and Verdier 2000 also described a similar mechanism in a theoretical paper). So Latin America was condemned to low human capital and low development. In contrast, the non-tropical land in North America lent itself to family farms, which implied greater equality and greater investment in mass human capital.4 The second strand of the literature links ethnic divisions to poor growth and public good outcomes. While violence directed at or by ethnic groups is well-known, the more subtle economic effects of ethnic conflict have only recently attracted attention in the economics literature. The mechanism could be similar to that in the previous paragraph – an ethnically distinct group in power is reluctant to invest in public services for the other ethnic groups for fear that the other ethnic groups will be enabled to displace the first group from power. Easterly and Levine 1997 find that Africa’s high linguistic diversity helps explain the continent’s poor policies, including low public goods, and poor economic growth. Alesina, Baqir, and Easterly 1999 find that more ethnically diverse US cities and counties devote less resources to public goods than more ethnically homogeneous cities and counties. Goldin and Katz 1999 find lower public support for higher education in states with more religious - ethnic heterogeneity. Goldin and Katz 1997 likewise find lower high school graduation rates in states that had higher religious-ethnic diversity. Miguel 1999 likewise finds lower primary school funding in more ethnically diverse districts in Kenya. Mauro 1995 and La Porta, Lopez de 5 Silanes, Shleifer and Vishny 1999 find that ethnic diversity predicts poor quality of government services. Alesina, Baqir, and Easterly 2000 find a link from ethnic diversity to bloated government payrolls in US cities. Rodrik 1999 noted that ethnically divided nations react more adversely to external terms of trade shocks. Svensson 2000 finds that more foreign aid proceeds are diverted into corruption in more ethnically diverse places. Annett 1999 finds that linguistic or religious diversity leads to greater political instability, which Annett finds in turn leads to higher government consumption. Knack and Keefer 1997 find that ethnic homogeneity raises “social capital” or “trust,” which in turn is associated with faster growth and higher output per worker. Adelman and Morris 1967 also noted that “cultural and ethnic heterogeneity tend to hamper the early stages of nation-building and growth.”5 This paper brings together these two strands of the literature. I call a situation of relative equality and ethnic homogeneity a “middle class consensus.” I argue that this middle class consensus facilitates higher levels of income and growth, as well as higher levels of public goods. Like Engerman and Sokoloff 1997 and Sokoloff and Engerman 2000, I link the existence of a middle class consensus to initial factor endowments, mainly a tropical endowment that lent itself to production of primary commodities, but I test their hypothesis with cross-country data.6 I find that a middle class consensus provides a remarkably parsimonious explanation of development outcomes. The middle class consensus also may affect the likelihood that a country is democratic. One might think that a rich elite would in fact have an incentive to suppress democracy. Democratic voting would result in higher mass education than the oligarchic elite would choose on its own, because the elite fear that a more educated majority will displace them from power. Similarly, in an ethnically diverse society, the most powerful ethnic group may want to suppress democratic rights of the other groups because they are threatened by the human capital accumulation those groups would vote for themselves. Of course, under both high inequality and high ethnic diversity, the elite fear that democracy will redistribute resources away from the 6 themselves to a new majority coalition. A large and homogeneous middle class would not have anything to lose in a democracy and so would be more likely to grant universal suffrage.7 We will test this prediction in the empirical section. So far, I have been referring to human capital as one of the main social choices, broadly construed to include education and health outcomes. The same predictions would go through if we were discussing publicly provided infrastructure capital, so I will also test various infrastructure measures in the empirical section. We should also expect that consensual societies will favor growth of future production over redistribution of existing resources. Societies lacking a middle class consensus will see the economic/political/ethnic elite underinvest in human and infrastructure capital because they fear empowering the opposition. We also may see more political instability as the elite is sometimes successfully – but often temporarily -- displaced from its access to the spoils of power. II. Empirical testing on commodity endowment, middle class consensus, and development In this section, I test some of the propositions advanced by the previous literature and by the story in this paper. The previous literature and this model suggests that the middle class share and ethnic diversity are fundamental determinants of incentives to invest in the future, and so would determine many of the right-hand side variables in growth or income regressions. I will run parsimonious regressions of growth, income, human capital accumulation, and infrastructure on ethnic diversity and inequality. Given the auxiliary predictions for democracy and political instability, I will also relate those variables to the middle class consensus. Table 1 reports summary statistics on the variables in the paper. The data on inequality are the broadest possible sample from Deininger and Squire 1996. The ethnolinguistic fractionalization, which varies from 0 to 100, is from Easterly and Levine 1997. A. Tropical endowments, commodity exporting, and inequality I first test the hypothesis of Engerman and Sokoloff 1997 and Sokoloff and Engerman 2000 that a tropical endowment leads to commodity production, and that commodity production 7 is associated with higher inequality. Their hypothesis has not been systematically tested with cross-country data as far as I am aware.8 Establishing these facts will make these variables candidates to be instruments for inequality. I use the World Bank World Development Report classification of countries as non-oil commodity exporters. I should clarify that the commodities are not necessarily "tropical" like bananas; they could just as well be coal or iron ore that are also produced in industrial countries. Hence, the hypothesis that tropical location is associated with commodity exporting is far from tautological. For tropical location, I construct a dummy that takes on the value one if the country’s mean absolute latitude is less than 23.5 degrees and 0 otherwise. Table 2 shows a probit equation for commodity production on tropical location: Not too surprisingly, commodity exporting is strongly associated with the tropics. Table 3 classifies countries by whether they are commodity exporters and by whether they are tropical: The vast majority (85%) of commodity exporting nations are in the tropics. Tropical nations are 5 times more likely to be commodity exporters than temperate nations. The next step is to see whether being commodity exporting is associated with higher inequality, as hypothesized by Engerman and Sokoloff. Here is a simple regression of the share of the middle three income quintiles on the commodity exporting dummy (in light of the foregoing regression, TROPICS is an instrument for COMMOD) and a dummy for oil exporting nations: Confirming the Engerman-Sokoloff hypothesis, commodity production (including oil production) is associated with a lower share of income of the middle quintiles. The effect of commodity exporting is enormous, equal to nearly 3 standard deviations of the middle income share. Going from being a commodity exporter to being a non-commodity exporter explains two-thirds of the entire range of the middle class share variable, which only varies between 30 and 58 percent. Oil production also moves the middle income share by a sizeable amount, more than one standard deviation. 8 B. The middle class consensus and per capita income and growth I now have potential instruments for the middle income share to use in a regression of income or growth on the middle income share and the ethnic fractionalization index. Of course, these instruments must not only be correlated with the instrumented variables, they also must be uncorrelated with the error term. TROPICS, for example, must enter the income equation only through its effect on the middle income share. This is a strong assumption, which I will test below. I first use 1990 per capita income as the most general measure of economic development. I adopt a very parsimonious specification that features only the middle class share (suitably instrumented) and ethnic heterogeneity. We can think of this as a reduced form, where all the variables that usually appear in income or growth regressions are endogenous outcomes of the middle class consensus (many of them will indeed be dependent variables below). I estimate the system of the inequality equation and the income equation jointly using three stage least squares (Table 5). As always, there are benefits and costs from using a system estimator. The benefit is that a system estimator is more efficient than a single equation estimator. It also gives a more accurate estimate of the standard errors when we control for a possible association of the error term in the middle class share equation and that in the per capita GDP equation. The disadvantage of the system estimator is that any specification error in one equation will be propagated to the other equation. I will attempt to deal with this problem in the robustness checks. Per capita income is strongly influenced by the middle class share and by ethnic fractionalization. A one standard deviation increase in the middle class share (7 percentage points) is associated with an enormous movement of 1.2 standard deviations in log per capita income (equivalent to an income increase by a factor of 3.4). The effect of ethnic diversity is not as strong but still important: a one standard deviation increase in ethnic diversity lowers log income by one quarter of a standard deviation.9 Figure 1 shows the fall in log income as one 9 moves from high to low terciles of the middle class share, as well as the fall in income from low to high terciles of ethnic diversity. It is also of interest to do a minimalist growth regression, using only the middle class income share and ethnic fractionalization. The exercise is once again how much can be explained by the middle class consensus hypothesis. Once again I use a system estimator, as shown in Table 6. A one standard deviation increase in the middle class income share is associated with a growth increase of .42 standard deviations, equivalent to one additional percentage point of per capita growth. A movement from the minimum middle class income share to the maximum in the sample is associated with an enormous increase in growth -- 3.8 percentage points. A one standard deviation increase in ethnic fractionalization is associated with a growth decrease of .21 standard deviations, equivalent to half of a percentage point of growth. A movement from the minimum ethnic fractionalization to the maximum is associated with a fall in growth of 1.5 percentage points. Figure 2 shows the fall in the per capita growth rate as one goes from high to low middle class share, and from low to high ethnic diversity. The highest growth rate is with a high middle income share and low ethnic diversity; growth miracles Japan and Korea are in this group. The lowest growth is with a low middle income share and high ethnic diversity. Guatemala, Sierra Leone, and Zambia are examples of countries that fall in the low middle class share, high ethnic diversity part of the sample. How robust are these result to other exogenous factors that have been mentioned in the literature? Bloom and Sachs (1998) and Sachs and Warner (1997) argue that being landlocked is a geographic disadvantage for development. When I introduce a landlocked dummy as an exogenous variable into either the income or growth regressions (and add it as an instrument in this and the following system regressions), it is insignificant and the middle class share and ethnic fractionalization remain significant. 10 These authors also argue that tropical location is a development disadvantage. I agree with this thesis, but provide a structural explanation for why it matters -- through the effect on inequality. I introduce the tropics dummy directly into the income or growth equation controlling for middle class share and ethnic fractionalization. Although the tropics dummy weakens the significance of the ethnic fractionalization variable in the income equation and of the middle class share in the growth equation, the tropics dummy itself is not significant in either equation. Still, insignificance does not prove the coefficient is zero, and the dependence of my results on the exclusion restriction on the tropics dummy should be kept in mind. I may be overstating the role of the middle class share if the "true" equation does contain a direct effect for the tropics dummy. Another possibility is that commodity-exporting is directly detrimental to growth because it sets off a frenzy of rent-seeking, overindebtedness, and under-investment in human capital (Manzano and Rigobón 2000, Gylfason 2000, Lane and Tornell 1996, Tornell and Lane 1999, Sachs and Warner 1995). I also try introducing the commodity exporting dummy directly into the income and growth equations. In the income equation, it leaves the significance of middle class share and ethnic fractionalization unchanged, while commodity-exporting is itself insignificant in these equations. In the growth equation, the significance of the middle class share is weakened, but commodity exporting itself remains insignificant. Although the original Engermann-Sokoloff hypothesis was about inequality driven by economies of scale in cash crops for export, I test whether inequality is robust to the choice of structure of production rather than the structure of exports as the determining variable. I construct data on the share of agriculture and mining value added in GDP by country, averaged over 1960-98. I estimate the system of equations substituting this variable for the commodity- exporting and oil-exporting dummies, and find that a high share of agriculture and mining indeed predicts a low share for the middle class. Although ethnic fractionalization becomes insignificant in the second equation for income, the share of the middle class is still a significant causal determinant of both income and growth in this system. 11 Some of my illustrative examples were about land inequality. Does land inequality also have a negative effect on income and growth like income inequality? It turns out that land inequality is not as well explained by the commodity and oil exporting dummies, but it does seem to be directly related to the tropics dummy. Instrumenting with tropics (and the other exogenous characteristics), higher land inequality significantly reduces both income and growth in three- stage least squares regressions identical to those above except substituting the land Gini coefficient for the middle class share. The source of the land Gini data is Deininger and Olinto 2000, where I average over all available years for each country. Another robustness check I pursue is to apply the hypothesis to a completely different dataset -- the dataset of US counties in 1990 studied by Alesina, Baqir, and Easterly 1999. I am unable to instrument with a tropical dummy for inequality in US counties, so I just describe the results of an OLS regression as a stylized fact. More unequal (measured by a higher ratio of mean to median income, as in Alesina, Baqir, and Easterly 1999) and more ethnically diverse US counties have lower average per capita incomes compared to more homogeneous and equal counties, controlling for whether the county is urban or rural. Although not causal, these results are consistent with the middle class consensus hypothesis as applied to intraregional variation in the US. As a final robustness check, I investigate whether ethnic fractionalization enters in a quadratic fashion, as some authors have suggested (Collier 1999, Collier and Hoeffler 1998). This possibility might also be suggested by some non-monotonicity in Figures 1 and 2. I find some evidence for a quadratic effect for the income regression, with a positive sign on the linear term and a negative term on the quadratic term for ethnic fractionalization. The turning point is at Ethnic Fractionalization = 28 on a scale of 0 to 100, so for most of the range of the variable, the relationship is still negative. I find no evidence for a quadratic effect in the growth regression. To summarize, the result that per capita income and growth depends positively on middle class share and negatively on ethnic fractionalization is robust to instrumenting for endogeneity, 12 alternative control variables, and choice of datasets. I regard this result as the key result of the paper, subject to the usual caveats about adequacy of the instruments, suggesting that the middle class consensus is a deep fundamental determining development outcomes. III. The middle class consensus and other development outcomes A. The middle class consensus and human and infrastructure capital accumulation I now look directly at whether the middle class share and ethnic fractionalization are related to human capital accumulation and other public goods. Table 7 shows the results from system estimations – of the exact same form as for income and growth -- for different dependent variables, showing only the coefficients for each dependent variable on the middle class share and ethnic fractionalization. These can be thought of as joint reduced form equations with the income and growth equations. I will test later how controlling for per capita income (which is difficult to do because of its endogeneity) affects these results. Starting first with the education variables, we see that the middle class share has a weak association with primary enrollment, and a strong association with secondary and tertiary enrollment. Ethnic diversity does not have much of an association with tertiary enrollment, but is related to primary and secondary enrollment. To get an idea of the magnitude of these effects, a one standard deviation increase in ethnic diversity lowers secondary enrollment by .26 standard deviations, while a one standard deviation decrease in the middle income share lowers secondary enrollment by .93 standard deviations. On health, the middle income share affects all the indicators: life expectancy, infant mortality, low birth weight of infants, percent of children immunized against DPT, and percent of children immunized against polio. Ethnic diversity also significantly affects virtually all the indicators, with the expected sign: higher ethnic diversity leads to worse health outcomes and lower levels of publicly provided health services.10 A one standard deviation increase in middle class share lowers infant mortality by .73 standard deviations, while a one standard deviation fall in ethnic diversity lowers infant mortality by .25 standard deviations. 13 On infrastructure, the results are less uniform. The middle class income share does not affect access to clean water or faults per telephone line, but increases percent of roads paved, access to sanitation, and telephones. Ethnic diversity does not affect percent of roads paved, but it lowers access to clean water, access to sanitation, telephones, and increases telephone faults per line. A one standard deviation increase in the middle class share raises access to sanitation by .53 standard deviations, while a one standard deviation fall in ethnic diversity raises access to sanitation by .27 standard deviations. There is some variation as to which kind of polarization – by class or by ethnic group – matters for the different indicators. Overall, however, these results are supportive of the hypothesis that a middle class consensus -- measured by share of the middle class and ethnic homogeneity -- is associated with higher levels of human and infrastructure capital accumulation. B. Economic policies and the middle class consensus The existence of a middle class consensus also affects the choice of economic policies. Societies with a middle class consensus will choose policies to promote growth, while societies polarized by class and ethnic group will opt for redistributive policies. The following table shows the effect of the middle class share and ethnic diversity on four key policy indicators – the black market premium, real overvaluation, financial depth, and trade openness. The policy indicators respond to different measures of group polarization, as shown in Table 8. Financial depth (reflecting the absence of a redistributive policy like interest rate controls that yield negative real interest rates) is positively related to the middle class share. The overvaluation index (the deviation from Purchasing Power Parity estimated by Dollar 1992, extended for the whole sample 60-98) and consumer price inflation are negatively related to middle class share. We can interpret this finding as inflation and an overvalued exchange rate being used as a redistributive device in an unequal society. The black market premium is positively related to the degree of ethnic diversity.11 The trade share in GDP is negatively related 14 to ethnic diversity. Again, we can see trade distortions and the black market premium being used as redistributive devices in an ethnically-divided society. C. Democracy, Political Instability, and Middle Class Consensus As mentioned in the discussion section, we might expect polarized societies to be less democratic-- the most powerful group may attempt to suppress democracy so as not to vote for “excessive” (i.e. outside the group) human capital accumulation. We will use the well-known Freedom House measures of political rights and civil liberties to test this prediction. We might expect that societies that opt for redistributive policies would also have more unstable governments, as different factions fight for the spoils of power. This could show up most overtly as civil war, or less violently as revolutions, coups, constitutional changes, and cabinet changes. Struggles over redistribution may also put minority groups at risk of economic or political discrimination, or even violent oppression. This type of political instability is captured well by the “Minorities at Risk” measure of Gurr (1993), which measures the percent of the population belonging to minorities at risk. Table 9 shows the results of system estimation with democracy and political instability variables as the dependent variable in the second equation (the first equation as always determines the middle class share endogenously as a function of commodity-exporting and oil- exporting, instrumenting for commodity exporting with tropical location). Suppression of civil liberties decreases with middle class share, but is not related to ethnic diversity. The measure of suppression of political rights increases with ethnic diversity and decreases with the share of the middle class. A middle class consensus is good for democracy. A one standard deviation increase in the middle class share raises political rights by .57 standard deviations, while a one standard decrease in ethnic diversity raises political rights by .21 standard deviations. As far as political instability, one or the other polarization measure is statistically significant for civil war, revolutions and coups, constitutional changes, and minorities at risk, 15 while cabinet changes do not appear to be related to these polarization measures. More ethnic diversity is associated with more time in civil war, greater share of the population belonging to minorities at risk, and more constitutional changes, while a greater share for the middle class is associated with fewer revolutions and coups and fewer constitutional changes.12 D. “Modernization” indicators and the middle class consensus We can also examine the effect of the middle class consensus on other indicators of a society’s development or “modernization.” More developed societies move away from agriculture towards industry and services (see Kongsamut, Rebelo, and Xie 2001 for a recent treatment). In Table 10, I use the share of agriculture in GDP as the dependent variable in the second equation of the 3SLS system. I find that societies with a larger middle class and more ethnolinguistic homogeneity have smaller agriculture shares. Another indicator of “modernization” is the share of the population that lives in cities. In Table 10, I show the coefficients of the urbanization ratio regressed in the 3SLS system on the middle class share and ethnolinguistic heterogeneity. A larger and more homogenous middle class is associated with more urbanization. A one standard deviation increase in the middle class share raises urbanization by .54 standard deviations, while a one standard deviation fall in ethnic diversity increases urbanization by .34 standard deviations. The middle class consensus is associated with these two well-known indicators of greater societal modernization. E. Robustness for controlling for income The above regressions do not control for per capita income. As I mentioned, we can think of all of them as reduced forms for different measures of development outcomes. However, the question arises whether the relationship of per capita income to the middle class consensus is accounting for all of the other results in the paper, since many of these other variables plausibly depend on per capita income. Unfortunately, ir is not easy to identify separately the effect of income, since per capita income is endogenous. I add log per capita income as another right-hand-side variable and 16 instrument for it with the same instrument set. We are now asking the exogenous instruments to simultaneously identify the middle class share and per capita income. The system is identified, but we can expect a high degree of correlation between the instrumented values of per capita income and the middle class share. Despite these problems, we still find an effect after controlling for income of the middle class share and/or ethnic divisions on a few important variables: civil liberty, political rights, tertiary enrollment, DPT immunization, polio immunization, life expectancy, infant mortality, and financial depth. For the other variables, neither income nor the middle class consensus variables were significant, suggesting a problem with correlation of the instrumented values for income and the middle class share. For these variables, we need to qualify the conclusion that the middle class consensus is directly determining these outcomes, as it may be working through per capita income. The regressions remain valid expressions of the reduced form relation between the middle class consensus and the development outcomes. V. Conclusions Countries with a middle class consensus have a higher level of income and growth. We can see why relatively homogenous middle-class societies have more income and growth, they have more human capital and infrastructure accumulation, they have better national economic policies, more democracy, less political instability, more “modern” sectoral structure, and more urbanization. These effects are causal, assuming the identifying restrictions hold, as I am able to instrument for middle class share with tropical commodity exporting. In future research, I plan to investigate the effects of endowments of particular commodities on inequality and increase yet further the instrument set. Readers of previous papers on difficulties created by ethnic heterogeneity often ask what policy implications follow. Surely we do not want to give intellectual comfort to those who engage in “ethnic cleansing.” However, the result on the poor development outcomes associated with ethnic heterogeneity only says that, on average, politicians exploit ethnic divisions to the 17 detriment of growth. It remains a choice for individual politicians whether they seek to divide and conquer, or to promote interethnic consensus. Easterly 2001 also suggests that good institutions eliminate the adverse effects of ethnic conflict, although again good institutions are less likely a priori with high ethnic diversity. The results in this paper are consistent with stories in which societies dominated by an elite class or ethnic group will accumulate less human and infrastructure capital for the majority because of the fear of empowering groups outside the elite’s own class or ethnic group. I relate the degree of middle class consensus to tropical endowments which led to commodity-exporting (as in the Engermann-Sokoloff hypothesis) and to ethnolinguistic fragmentation. This paper's argument suggests that rich societies are rich, not because of superior culture as Landes 1998 would argue, but more because of accidental geographic and demographic make-up as argued by Diamond 1997. 18 Figure 1: Per capita income as function of middle income share and ethnic diversity 9.50 9.00 8.50 Log of per capita income 8.00 7.50 low medium 7.00 ethnic diversity high high medium low share of middle class 19 Figure 2: Growth as function of middle income share and ethnic diversity 4.0% 3.5% 3.0% 2.5% Per capita growth 1950- 2.0% 92 1.5% 1.0% low 0.5% medium 0.0% ethnic diversity high high medium low middle income share 20 Bibliography Acemoglu Daron and James A. Robinson (2000) “Why Did the West Extend the Franchise? Democracy, Inequality and Growth in Historical Perspective,” Quarterly Journal of Economics v115, n4: 1167-99 Adelman, Irma and Cynthia Taft Morris (1967), Society, politics, and economic development: a quantitative approach, Johns Hopkins Press, Baltimore,. Alesina, Alberto(1994), “Political Models of Macroeconomic Policy and Fiscal Reforms,” in Stephan Haggard and Steven Webb, eds. Voting for Reform: Democracy, Political Liberalization, and Economic Adjustment, New York, NY: Oxford Univ. Press. Alesina, Alberto, Reza Baqir, and William Easterly (1999), “Public Goods and Ethnic Divisions,” Quarterly Journal of Economics, November, Volume CXIV, Issue 4, 1243-1284. Alesina, Alberto, Reza Baqir, and William Easterly (2000), “Redistributive government employment,” Journal of Urban Economics, Vol. 48, No. 2, 219-241 Alesina, Alberto and Dani Rodrik (1994), “Distributive politics and economic growth,” Quarterly Journal of Economics 108: 465-90. Annett, Anthony (1999). “Ethnic and religious division, political instability, and government consumption,” IMF mimeo, March. Barro R. J. (2000). “Inequality and Growth in a Panel of Countries,” Journal of Economic Growth, 5, 5-32. Barro, Robert and Xavier Sala-I-Martin (1995), Economic Growth, McGraw-Hill,. Berge, Kersti and Adrian Wood (1997), "Exporting Manufactures: Human Resources, Natural Resources, and Trade Policy", Journal of Development Studies v34, n1: 35-59 Bloom, David E. and Jeffrey D. Sachs (1998). “Geography, Demography, and Economic Growth in Africa”, Brookings Papers on Economic Activity 1998:2 Bourguignon, François and Thierry Verdier (2000), Oligarchy, democracy, inequality and growth, Journal Of Development Economics (62)2, 285-313 .Bourguignon François(1993). "Croissance, distribution et ressources humaines: Comparaison internationale et specificites regionales." Revue d'Economie du Developpement v0, n4, 3-35 Bourguignon François and C. Morrison (1990), "Income Distribution, Development and Foreign Trade: A Cross-sectional Analysis", European Economic Review v34, n6, 1113-32 Clarke, George (1995). “More evidence on income distribution and growth,” Journal of Development Economics Vol. 47, 403-427. 21 Collier, Paul (1999). “The Political Economy of Ethnicity,” in Boris Pleskovic and Joseph E. Stiglitz, eds. Annual World Bank Conference on Development Economics 1998, 387- 99 Collier, Paul and Anke Hoeffler (1998), “On economic causes of civil war,” Oxford Economic Papers 50, 563-573. Decornez, Shubhasree Seshanna (1998). “An empirical analysis of the American middle class (1968-1992),” Ph.D. Dissertation, Vanderbilt University. Deininger, Klaus W and Pedro Olinto (2000). “Asset distribution, inequality, and growth Washington, DC : World Bank, Development Research Group, Rural Development, Policy research working paper 2375. Deininger, Klaus and Lyn Squire (1998), “New ways of looking at old issues: inequality and growth,” Journal of DevelopmentEconomics;57(2), 259-87 Deininger, Klaus and Lyn Squire (1996), “A New Data Set Measuring Income Inequality,” World Bank Economic Review;10(3), 565-91. Diamond, Jared (1997). Guns, germs, and steel: the fates of human societies. New York: W. W. Norton. Easterly, William and Ross Levine (1997), “Africa’s Growth Tragedy: Policies and Ethnic Divisions,” Quarterly Journal of Economics, CXII (4), 1203-1250. Easterly, William and Mirvat Sewadeh (2001), “Global Development Network Growth Database,” http://www.worldbank.org/research/growth/GDNdata.htm Easterly,William (2001). “Can institutions resolve ethnic conflict?” Economic Development and Cultural Change, forthcoming. Engermann, Stanley and Kenneth Sokoloff (1997), “Factor Endowments, Institutions, and Differential Paths of Growth Among New World Economies: A View from Economic Historians of the United States,” in Stephen Haber, ed. How Latin America Fell Behind, Stanford CA: Stanford University Press. Filmer, Deon, and Lant Pritchett (1997), “Child mortality and public spending on health : how much does money matter?”, World Bank Policy research working paper 1864. Forbes, Kristin (2000). “A Reassessment of the Relationship Between Inequality and Growth”, American Economic Review, v90, n4, 869-87 Galor, Oded and Joseph Zeira (1993), “Income distribution and macroeconomics,” Review of Economic Studies 60: 35-52. Goldin, Claudia and Lawrence Katz (1999), “The shaping of higher education: the formative years in the United States, 1890 to 1940”, Journal of Economic Perspectives, Volume 13, No 1, 37-62. 22 Goldin, Claudia and Lawrence Katz (1997), “Why the United States led in education: lessons from secondary school expansion, 1910 to 1940,” NBER Working Paper 6144. Gradstein M. and M. Justman (1995), A Political Interpretation of the Kuznets Curve, Mimeo. Gurr, Ted Robert (1993). Minorities at risk : a global view of ethnopolitical conflicts with contributions by Barbara Harff, Monty G. Marshall, James R. Scarritt. Washington, D.C. : United States Institute of Peace Press. Gylfason, Thorvaldur (2000). "Natural resources, education, and economic development," Center for Economic Policy Research Working Paper No. 2594, October. Inter-American Development Bank (1999), Economic and Social Progress in Latin America: Facing up to inequality in Latin America, 1998-1999 report. Knack, Stephen, and Philip Keefer (1997), “Does Social Capital Have an Economic Payoff? A Cross-country Investigation,” Quarterly Journal of Economics, Vol. CXII, Issue 4, 1251-1288 Kongsamut, Piyabha, Sergio Rebelo, and Danyang Xie (2001). “Beyond Balanced Growth,” Review of Economic Studies, forthcoming. Lal, Deepak (1998). Unintended consequences : the impact of factor endowments, culture, and politics on long run economic performance. Cambridge, Mass. : MIT Press, Landes, David (1998). The Wealth and Poverty of Nations. New York NY Norton. Lane, Philip R. and Aaron Tornell (1996), Power, Growth, and the Voracity Effect, Journal of Economic Growth v1, n2, 213-41 La Porta, Rafael, Florencio Lopez-de-Silanes, Andrei Shleifer, and Robert Vishny (1999), “The quality of government,” Journal of Law, Economics and Organization. Mankiw, N. Gregory, David Romer, and David Weil (1992), “A Contribution to the Empirics of Economic Growth,” Quarterly Journal of Economics 101, 407-437. Manzano, Osman and Roberto Rigobón (2000). "Resource Curse or Debt Overhang?", paper presented at MIT Energy Lab conference. Miguel, Ted (1999). “Ethnic diversity and school funding in Kenya.” Mimeo, Harvard University. Moore, Barrington Jr. (1966), The Social Origins of Dictatorship and Democracy. Boston: Beacon Press Perotti, Roberto (1996), “Growth, Income Distribution, and Democracy: What the data say,” Journal of Economic Growth 1, 149-188. 23 Persson, Torsten and Guido Tabellini (1994), “Is inequality harmful for growth?”, American Economic Review 84: 600-621. Sachs, Jeffrey and Andrew Warner (1997), “Fundamental sources of long-run growth,” American Economic Review Papers and Proceedings, May, Vol 87 No 2, 184-188. Sachs, Jeffrey and Andrew Warner (1995). "Natural resource abundance and economic growth," NBER Working Paper 5398. Sokoloff, Kenneth L. and Stanley L. Engerman (2000), Institutions, Factor Endowments, and Paths of Development in the New World, Journal of Economic Perspectives v14, n3, 217-32 Svensson, Jakob (2000) “Foreign aid and rent-seeking,” Journal of International Economics v51, n2, 437-61 Tornell, Aaron and Philip Lane (1999). “The Voracity Effect.” American Economic Review, v89, n1, 22-46 24 Table 1: Statistics on variables used in this paper Variable Mean Maximum Std. Obser- Median Minimum Dev. vations Cabinet changes per year 60-88 0.39 0.36 1.07 0.00 0.22 168 Civil liberty, 1998 3.57 4.00 7.00 1.00 1.79 190 Constitutional changes per year 60-88 0.10 0.10 0.36 0.00 0.09 168 Dummy for non-oil commodity 0.22 0.00 1.00 0.00 0.42 175 exporting Dummy for oil exporting 0.09 0.00 1.00 0.00 0.28 175 Dummy for Tropical Location 0.50 0.00 1.00 0.00 0.50 229 Ethnic diversity,1960 41.47 42.00 93.00 0.00 29.79 113 Faults per phone line 67.60 47.50 350.00 2.00 74.23 62 GDP Per capita 60 2247 1316 9895 257 2186 124 GDP Per capita 90 5825 3780 22660 400 5719 152 Growth Per Capita 1950-92 (Summers- 0.02 0.02 0.07 -0.08 0.02 146 Heston), average of available data Immunization DPT (%) 74.76 82.00 100.00 13.00 21.66 145 Immunization Polio (%) 75.68 83.00 100.00 13.00 21.85 145 Infant mortality 1990 86.45 89.00 194.00 3.00 56.36 141 Infants, low birth weight,1990 11.95 10.00 50.00 4.00 6.53 111 Life expectancy 1997 66.37 69.80 79.99 37.51 10.23 194 Black market premium 1997 0.704 0.04 46.6 -0.89 4.517 142 Log inflation 1960-98 0.15 0.08 1.42 0.03 0.21 136 M2 to GDP 1997 43.90 35.12 206.25 6.09 32.63 140 Middle class share (share of quintiles 46.71 48.04 57.70 30.00 7.11 103 2-4), average 60-96 Percent of population with access to 68.32 72.00 100.00 12.00 24.34 104 clean water 1990 Percent of population with access to 56.75 56.00 100.00 1.00 33.35 120 sanitation 1990 Percent of roads paved 1990 45.55 42.00 100.00 0.00 31.61 191 Political rights,1998 3.47 3.00 7.00 1.00 2.24 190 PPP Trade Share in GDP 1997 35.69 25.72 290.71 3.37 37.89 133 Primary enrollment, 1990 0.86 1.00 1.00 0.15 0.22 120 Real exchange rate overvaluation 117.1 107.45 381.94 50.47 41.55 104 (100=PPP) 1960-98 6 Revolutions and coups per year, 60-88 0.19 0.07 1.00 0.00 0.24 168 Secondary enrollment, 1990 0.49 0.44 1.00 0.03 0.31 118 Share of agriculture in GDP, 1990 20.4 17.5 65.5 0.3 15.8 162 Share of pop. in minorities at risk, 1990 0.28 0.17 1.00 0.01 0.27 111 Share of time at civil war 60-89 0.07 0.00 0.80 0.00 0.15 135 Telephones per capita, 1994 82.36 63.03 293.83 8.27 67.75 189 Tertiary enrollment, 1990 0.12 0.07 0.58 0.00 0.12 123 Urbanization ratio, 1990 51.0 49.7 100 5.2 24.0 197 For sources see Easterly and Sewadeh 2001. 25 Table 2: Commodity Exporting and Tropical Location Dependent Variable: COMMODITY EXPORTING DUMMY Method: ML - Binary Probit Included observations: 175 Covariance matrix computed using second derivatives Variable Coefficient Std. Error z-Statistic Prob. C -1.471424 0.205586 -7.157227 0.0000 TROPICS 1.130729 0.245913 4.598093 0.0000 Mean dependent var 0.222857 S.D. dependent var 0.417357 S.E. of regression 0.391207 Obs with Dep=0 136 Obs with Dep=1 39 26 Table 3: Commodity Exporting and Tropical Location # Countries Tropical Non-tropical Total Commodity exporter 33 6 39 Non-commodity exporter 58 78 136 Total 91 84 175 Percent of row totals Commodity exporter 36% 7% Non-commodity exporter 64% 93% Percent of column totals Commodity exporter 85% 15% Non-commodity exporter 43% 57% 27 Table 4: Middle Class Share and Commodity Exporting Dependent Variable: MIDDLE CLASS INCOME SHARE Method: Two-Stage Least Squares Included observations: 102 White Heteroskedasticity-Consistent Standard Errors & Covariance Instrument list: C TROPICS DUMMY, OIL DUMMY Variable Coefficient Std. Error t-Statistic Prob. C 51.63167 1.058569 48.77496 0.0000 COMMODITY DUMMY -19.62242 4.996424 -3.927293 0.0002 OIL DUMMY -10.88073 3.315207 -3.282066 0.0014 S.E. of regression 9.308190 Mean dependent var 46.75934 F-statistic 9.590131 S.D. dependent var 7.121557 Prob(F-statistic) 0.000156 28 Table 5: Income per capita and middle class consensus Estimation Method: Three-Stage Least Squares Instruments: Ethnic Fractionalization, Oil Dummy, Tropics Dummy, Constant Coefficient Std. Error t-Statistic Prob. C(1) 50.8239 1.7441 29.14 0.000 C(2) -18.7833 5.3571 -3.51 0.001 C(3) -8.0868 3.6927 -2.19 0.030 C(4) 2.3079 1.4033 1.64 0.102 C(5) 0.1402 0.0292 4.80 0.000 C(6) -0.0098 0.0035 -2.81 0.006 Equation: Middle Class Share =C(1)+C(2)*COMMODITY DUMMY +C(3)*OIL DUMMY Observations: 83 Equation: LOG(GDP Per Capita 90)=C(4)+C(5)*Middle Class Share +C(6)*Ethnic Fractionalization Observations: 81 29 Table 6: System estimation for Per Capita Growth as Dependent Variable Estimation Method: Three-Stage Least Squares Instruments: Ethnic Fractionalization, Oil Dummy, Tropics Dummy, Constant Coefficient Std. Error t-Statistic Prob. C(1) 51.0619 1.7443 29.27 0.000 C(2) -19.1717 5.2902 -3.62 0.000 C(3) -9.1976 3.7984 -2.42 0.017 C(4) -0.0314 0.0238 -1.32 0.189 C(5) 0.0012 0.0005 2.51 0.013 C(6) -0.0001 0.0001 -2.48 0.014 Equation: Middle Class Share =C(1)+C(2)*COMMODITY DUMMY +C(3)*OIL DUMMY Observations: 80 Equation: Per Capita Growth (1950- 92)=C(4)+C(5)*Middle Class Share +C(6)*Ethnic Fractionalization Observations: 80 30 Table 7: Results of 3SLS regressions of human capital and infrastructure related variables on share of the middle class and ethnic diversity Dependent variable: RHS variable Coefficient T-statistic #observations Education: Primary enrollment Middle class share 0.0075 1.42 76 Ethnic diversity -0.0021 -3.05 Secondary enrollment Middle class share 0.0402 5.15 76 Ethnic diversity -0.0027 -2.91 Tertiary enrollment Middle class share 0.0170 4.51 79 Ethnic diversity -0.0005 -0.95 Health: Life expectancy Middle class share 1.0794 4.18 83 Ethnic diversity -0.1353 -4.56 Infant mortality Middle class share -5.1633 -4.08 77 Ethnic diversity 0.4551 2.69 Infants, low birth weight Middle class share -0.3825 -2.55 72 Ethnic diversity 0.0755 3.81 Immunization DPT (%) Middle class share 1.7796 3.04 80 Ethnic diversity -0.1048 -1.55 Immunization Polio (%) Middle class share 1.5629 2.78 80 Ethnic diversity -0.1518 -2.32 Infrastructure: Percent of roads paved Middle class share 4.2969 5.52 81 Ethnic diversity -0.1099 -1.16 Access to clean water (%) Middle class share -0.4882 -0.25 51 Ethnic diversity -0.3446 -3.62 Access to sanitation (%) Middle class share 2.5587 2.54 77 Ethnic diversity -0.2989 -2.50 Log(telephones per capita) Middle class share 0.1206 4.81 82 Ethnic diversity -0.0078 -2.60 Faults per phone line Middle class share 0.6374 0.26 35 Ethnic diversity 1.3833 3.65 Notes: each equation contains a constant (not shown) and is part of a system with one other equation, which gives the middle class share as a function of a commodity-exporting dummy and oil-exporting dummy (both of which are generally significant and similar to the results in Tables 5 and 6). Instruments for the whole system are the oil dummy, tropical location, and ethnic diversity. 31 Table 8: Results of 3SLS regressions of policy-related variables on share of the middle class and ethnic diversity Dependent variable: RHS variable Coefficient T-statistic #observations Log black market premium Middle class share -0.0466 -1.23 71 Ethnic diversity 0.0126 2.49 Log overvaluation index Middle class share -3.40 -2.53 79 Ethnic diversity -0.21 -1.26 Log CPI inflation Middle class share -0.016 -2.42 79 Ethnic diversity -0.001 -1.33 Financial depth Middle class share 3.7164 3.27 74 Ethnic diversity -0.0942 -0.73 PPP Trade share of GDP Middle class share -0.7032 -0.44 75 Ethnic diversity -0.4821 -2.32 32 Table 9: Results of 3SLS regressions of democracy and political instability variables on share of the middle class and ethnic diversity Dependent variable: RHS variable Coefficient T-statistic #observations Political rights (1-7 where 1 is Middle class share -0.1577 -2.77 82 most free) Ethnic diversity 0.0146 2.02 Civil liberties (1-7 where 1 is most Middle class share -0.1359 -3.14 82 free) Ethnic diversity 0.0069 1.22 Percent of period in civil war Middle class share -0.0072 -1.40 76 Ethnic diversity 0.0014 2.02 Revolutions and coups per year Middle class share -0.0153 -2.01 82 Ethnic diversity 0.0009 0.96 Constitutional changes per year Middle class share -0.0065 -3.00 82 Ethnic diversity 0.0007 2.51 Cabinet changes per year Middle class share 0.0115 1.57 82 Ethnic diversity 0.0001 0.08 Minorities at Risk (% of Middle class share -0.009 -1.02 60 population) Ethnic diversity 0.003 2.50 33 Table 10: Results of 3SLS regressions of “modernization” variables on share of the middle class and ethnic diversity Dependent variable: RHS variable Coefficient T-statistic #observations Share of agriculture in GDP Middle class share -1.0740 -2.75 78 Ethnic diversity 0.1680 3.46 Urbanization ratio Middle class share 1.8197 2.71 83 Ethnic diversity -0.2807 -3.28 34 Endnotes 1 I am indebted for this quote to Miguel 1999 2 p. 30, Adelman and Morris 1967. 3 Forbes 2000 challenges these findings by using a GMM estimator that removes country effects and focuses on short-term intertemporal variation (across 5-year periods). A recent paper by Barro 2000 also uses panel methods to dispute the effect of inequality on growth, and argues that it only holds for the poorest countries. However, since the argument of this paper and many of those listed are on inequality as a long-run source of polarization and underdevelopment, Forbes’ and Barro’s findings are not at the relevant frequency. Moreover, Deininger and Olinto 2000 find that even in Forbes’ econometric methodology land inequality has a causal negative effect on growth. 4 Another author who emphasizes the importance of factor endowments is Lal 1998. 5 p. 41, Adelman and Morris 1967. 6 Berge and Wood 1997 relate primary commodity exporting to an abundance of natural resources relative to skill, which may be another mechanism by which they are related to inequality. Bourguignon and Morrison 1990 directly relate inequality to mineral exporting and land concentration in agricultural exports. Bourguignon 1993 uses mineral exporting as an instrument in a regression of growth on inequality. 7 This may be related to the famous thesis of Barrington Moore (1966) that (to simplify a little) when the commercial bourgeoisie (read middle class for our purposes) is strong, democracy emerges, whereas when landowners are dominant, dictatorship emerges. An alternative hypothesis for the motivation of extending the franchise is that the rich elite fears revolution (Acemoglu and Robinson 1998). Gradstein and Justman 1995 have voting determined by a minimum level of income, hence the franchise expands as income grows. 8 After completing a previous draft, I became aware of the 1998-99 report of the InterAmerican Development Bank (1999), which graphically shows correlations between commodity exports and income inequality and between latitude and income inequality. The advantage of my approach compared to theirs is that I make the endogenous variable (commodity exporting) respond to the exogenous variable (tropical location). 9 Easterly and Levine 1997 also found an effect of ethnic diversity, measured the same way, on income. 10 Filmer and Pritchett 1997 also found that higher ethnic diversity increases infant mortality. 11 Easterly and Levine 1997 also found an effect of ethnic diversity on the black market premium. 12 Annett 1999 also finds higher political instability with more ethnic diversity. Collier and Hoeffler 1998 also find a relationship between ethnic diversity and civil war but find it to be of an inverted U-shape – I use here a different measure of civil war (Sivard 1993) than theirs.