WPS7153 Policy Research Working Paper 7153 Long-Run Effects of Democracy on Income Inequality Evidence from Repeated Cross-Sections Carlos Felipe Balcázar Poverty Global Practice Group January 2015 Policy Research Working Paper 7153 Abstract This paper assesses the link between democracy and inequality. that long-run historical patterns in the degree of democracy Inequality is measured at the cohort level with pseudo-panel relate to income inequality. However, this relationship is data built from nine Latin American countries’ household non-monotonic: inequality first increases with the stock of surveys (1995–2009, biannual). Democracy is measured as democracy before falling. The paper also presents evidence a stock during long periods of time both before and after that education may be a mechanism explaining this result. each cohort’s year of birth. The paper presents evidence This paper is a product of the Poverty Global Practice Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The author may be contacted at cbalcazarsalazar@ worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Long-Run Effects of Democracy on Income Inequality: Evidence from Repeated Cross-Sections azar* Carlos Felipe Balc´ Keywords: Inequality, democracy, Latin America, pseudo panel JEL Classification: C23, D31, N36 * I thank Leopoldo Fergusson, Camilo Garcia-Jimeno, Ana Mar´ ıa Ib´ an ˜ ˜ ez, Hugo Nopo, ın, Jose Cuesta, Pablo Querub´ ırez and Javier Rojas for their comments and suggestions. Martha Schmitz, Jaime Ram´ 1. Introduction This paper assesses the relationship between democracy and income inequality. It is reasonable to think that democracies reduce income inequality by allowing citizens to vote for political parties that privilege redistributive platforms. Thus, societies with a more egalitarian distribution of polit- ical power may have lower income inequality. Nonetheless, Asian countries, for example, display low income inequality and low-rating democratic institutions while Latin America, a region with higher-rating democratic institutions, displays extreme income inequality (Savoia et al., 2009). The lack of a clear relationship between democracy and income inequality has two popular ex- planations. The first one is the political Kuznets curve. Chong (2004) argues that democratization first produces an increase in inequality before it produces an improvement in the distribution of income. However, Palma (2011) states that inequality does not decline after democratization in countries with historically high levels of inequality. Acemoglu and Robinson (2002) clarify that there are exceptions to the political Kuznets curve, for example, inequality might fluctuate if coun- tries switch between less and more democratic regimes (e.g., Latin America). A second explanation suggests that democracies that begin with high inequality may develop poor economic and political institutions that perpetuate the unequal distribution of income. Indeed, recent research attributes high and persistent inequality to the presence of “bad” economic and political institutions, which are in turn a result of high initial inequality (Savoia et al., 2009).1 Gerring et al. (2005; 2012) argue that that the best way to think about the relationship between democracy and social outcomes is as a time-dependent, historical phenomenon. Most recent empir- ical studies assess the link between democracy and inequality using cross-country panel data sets, traditionally: the Deninger and Squire data base; UNU-WIDER’s World Income Inequality data base; and the World Bank’s World Development Indicators.2 In general they conceptualize this link as the effect of regime type on inequality at time t plus some specified period. However, regimes are historically informed phenomena rather than contemporary variables. Gerring et al. find, for ex- ample, that a country’s level of democracy in a given year is not likely to affect its level of human development –but its stock of democracy over a very long-period of time does. On top of the previous problem “traditional” cross-country studies have other considerable prob- lems: i) Some surveys report data on income and some on expenditure, making cross-country com- parisons difficult, ii) The degree of accuracy is still a problem; e.g., some surveys undertaken in the midst of civil wars in sub-Saharan Africa have “national” coverage, but polities were fragmented 1 See Acemoglu et al. (2013) for a further discussion on why inequality might not decline under a democratic regime. 2 For example: Li et al. (1998), Gradstein et al. (2001), Reuveny and Li (2003), Chong (2004), Hamanaka (2008) and Timmons (2010), to name a few. 2 during those events, iii) None of the data sets provide relevant distributional information given that they are composed of country-level aggregates, iv) Democracy is endogenous –it might be a result of many socioeconomic phenomena (omitted variables bias) and may vary between groups within countries (measurement error), v) Although many studies resort to instrumental variables to solve the problem of endogeneity, the variables related to the quality of democratic institutions relate to either a prior or subsequent redistribution of income, violating the exclusion restriction.3 I assess the link between democracy and inequality using pseudo-panel data built from nine har- monized Latin American countries’ household surveys (1995-2009, biannual), conceiving within- cohort inequality as a function of the accumulated level of democracy (or its stock) around the cohort’s year of birth. My approach has some advantages over cross-country studies, in that: i) I explore a relatively homogenous region, ii) surveys report the same sources of income, making surveys comparable between countries and over time, iii) I do not rely upon country-level measures of income inequality but upon cohort-level inequality (somewhat accounting for heterogeneity be- tween groups within countries), iv) I measure democracy over long periods of time –I explore the accumulated effect of historical legacies on the contemporary level of inequality. The biggest empirical challenge to overcome in this study is endogeneity in the proxy for democracy. To address endogeneity I include a set of fixed-effects that capture cohort specific time-invariant characteristics, gender differences, long-run and short-run country-level economic differences, and changes in survey lifting. Therefore, my identification strategy, although it does not completely solves the problem of endogeneity, attenuates it. To reassure the reader about the validity of my results, I test whether my results survive statistical scrutiny. I check if my results hold after controlling for periods of wars and economic crises since both can lead to regime transitions and affect socioeconomic outcomes (Bethell, 1997; Piketty and Saez, 2003; Hagopian and Mainwaring, 2005; UNESCO, 2010; Fernandez and Lopez-Calva, 2010; Torche, 2010; Acemoglu et al., 2013). A number of messages emerge from this paper. First, cohorts that “experienced” long periods of quality democratic institutions have lower inequality. I find that 20 years of high quality democratic institutions could reduce the dispersion in labor income (on average) by around 25 %. Second, I find suggestive evidence that this relationship is non-monotonic: The level of inequality increases with the democracy stock and then starts decreasing with long periods of quality democratic institutions. Third, cohorts that “experienced” long periods of quality democratic institutions show lower dis- persion in educational attainment. I find that 20 years of high quality democratic institutions could 3 See Atkinson and Brandolini (1999), Timmons (2010), Palma (2011), Niemeyer et al. (2008), Wejnert (2005), Savoia et al. (2009), Bardhan (2005), Nikoloski (2009, 2010), Engermann and Sokoloff (1997, 2000) and Acemoglu et al. (2013) for further discussion. 3 reduce the dispersion in the years of education (on average) by around 15 %. This relationship also appears to be non-monotonic. These results give credence to the hypothesis that education maps democracy onto long-run changes in the distribution of human capital and then onto contemporary changes in income inequality. The rest of the paper proceeds as follows. The next section briefly presents a review of four mechanisms that link democracy and inequality, their limitations, and presents an alternative ap- proach. Section three describes the data. Section four describes the econometric approach and ad- dresses the methodological challenges. Section five presents the main results and shows suggestive evidence that education may be one mechanism explaining thee link between democracy and in- equality. The last section concludes. 2. Do democracies breed more egalitarian societies? Meltzer and Richard (1981), Alesina and Rodrik (1994), and Persson and Tabellini (1994) pro- pose that utility-maximizing individuals redistribute based on rational choices; if the median in- come lies below the mean income, the median voter chooses redistribution and higher taxation for rich people. This model predicts that democracies have lower levels of inequality than non- democracies (Acemoglu and Robinson, 1998). A second mechanism alleges that democracy lowers the costs of political participation of orga- nized labor, allowing labor unions to obtain a privileged position in the policy process (Schumpeter, 1942; Rodrik, 1999). As a result, democracy encourages unionization, centralized wage bargaining, and minimum wages, which reduce wage dispersion (Katz and Autor, 1999; Rodrik, 1999). Democracies should also guarantee broad access to property rights. With well-defined property rights and broad access to them, the poor have the possibility to gain access over improved or produced assets by facilitating the development of efficient market-based economies and opening up markets and institutions, which also prevent the elite from erecting entry barriers and enjoying markets with monopoly power (Gerring et al., 2005; Acemoglu, 2008). A fourth mechanism suggests that democracy increases competition among politicians for citi- zen support. This causes governments to invest more in public services, such as education (Saint- Paul and Verdier, 1993). Education, in turn, acts as a redistribution channel reducing the dispersion of human capital and increasing a generation’s human capital relative to the previous generation. I find these four mechanisms the most relevant.4 However, recent empirical findings reveal 4 See Thorbecke and Charumilind (2002) and Savoia et al. (2009) for further discussion. 4 caveats in most of these mechanisms. In a well regarded paper, Milanovic (2000), using micro- level data from the Luxembourg Income Studies, finds weak evidence for redistribution through the median voter channel. Indeed, he finds that the middle classes are not net beneficiaries from redistributive transfers. Timmons (2010), using the University of Texas Inequality Project and the United Nations Industrial Development Organization (UTIP-UNIDO) data set, finds no evidence for the existence of the second channel. He shows that although democracies may pay higher av- erage wages in manufacturing, democracy does not dampen wage dispersion between industries. Amendola et al. (2013), using the UNU-WIDER Income Inequality data set, find evidence that democracy is not a sufficient condition to reduce income inequality in the presence of strong proper- ty rights. They find that in presence of weak democratic institutions, strong property rights actually lead to an increase in the level of inequality. Education presents an interesting alternative. The historical record offers many cases where the birth of democracies opened up new possibilities for the scaling up of public education (Sachs, 2012). Engerman et al., (2000), Baum and Lake (2001), Lindert (2004) and Glaeser et al. (2004) suggest that democracy increases the education output, mapping democracy onto long-run changes in the distribution of human capital and then onto contemporary changes in income inequality. However, if the quality of education is low for the poor or the social payoff from education di- minishes as a consequence of deeply ingrained patterns of social exclusion and discrimination, educational attainment may not reduce inequality (Perry et al., 2003). A considerable issue that democracy faces to achieve redistribution is inequality itself: If democ- racies display high initial levels of inequality, it is difficult for the poor to hold the rich and powerful accountable. Such a state of affairs affects the social norms about the legitimacy of rules and in- stitutions. On the one hand, the wealthy minority may be inclined and able to establish a legal framework to ensure themselves a disproportionate share of political power, and use it to establish rules, laws, and government policies to give themselves broader access to economic opportunities. On the other hand, citizens may feel discouraged to press for redistributive changes that may not ultimately benefit them. Hence, property rights, legal systems, and fiscal and economic institutions may perpetuate the unequal distribution of income. (Engerman and Sokoloff, 2000; You and Ka- gram, 2004; Savoia et al., 2009; Acemoglu and Dell, 2010.) It is important that democracy works as a commitment device that prevents ruling minorities from erecting barriers to the excluded. 2.1. Conceptualizing democracy over long periods of time Works on democracy and inequality have usually focused on the possible causal effect of a coun- try’s contemporary level of democracy on its subsequent level of inequality. However, institutional 5 effects unfold over time, sometimes a great deal of time, and are cumulative. It is the long-term effect of these historical legacies the central concern if we wish to comprehend the causal effect of democracy on social outcomes. (Gerring et al., 2005; 2012.) It is reasonable to think that children that enjoyed higher quality democratic institutions during their childhood enjoyed a more appropriate environment for individual development. For exam- ple, Kudmatsu (2011) provides suggestive evidence that democratization was associated with an increase in the use of some health inputs on children in Sub-Sharan Africa. Also, higher quality democratic institutions might benefit children and teenagers during the school years if we con- sider that more democratic countries are more likely to spend more in school education (Brown and Hunter, 2004; Ansell, 2008; Huber and Stephens, 2012; Harding and Stasavage, 2014). Maybe democratic transitions are also important when entering the labor markets. On the one hand, work- ers might enjoy more bargaining power as a result of the political power of unions (Rodrik, 1999). On the other hand, workers might have access to job opportunities that were only available to the elite. It is also likely that the benefits from democracy have intergenerational impacts, as in the case of education (Saint-Paul and Verdier, 1993). In other words, democracy is likely to have long-run and cumulative impacts on social outcomes (Baum and Lake, 2003; Gerring et al. 2012). 3. Data The data come from several harmonized household surveys of nine Latin American countries: Argentina, Brazil, Bolivia, Colombia, Honduras, Peru, Panama, Paraguay and Uruguay. To max- imize the number of countries and periods I use surveys between 1995 and 2009, biannual, and restrict myself to urban areas in Bolivia due to the lack of rural data for the 1995 and 2009 surveys. Peru and Honduras surveys for 1995 are not harmonizable, but the countries represent sources of variation too important to exclude them; therefore, my pseudo-panel is unbalanced. Table A1, in the appendix, reports the data sources. I restrict myself to individuals born between 1936 and 1977. However, I drop individuals be- tween 18 and 24 years of age in 1995 and between 21 and 24 years of age in 1997 to consider that by 2009 life expectancy in the region was about 74 years –to keep a representative sample of “old” individuals- and that a person’s educational attainment is likely to remain unchanged after age 25 –to avoid truncation in the distribution of educational attainment. The source of income for every observation is monthly monetary labor income (constant prices of 2005, US$, adjusted by purchasing power parity), which is the sum of earnings from wages, tips (as applicable), over-time payments, commissions and bonuses. I drop monthly monetary labor 6 income outliers.5 The previous procedure comprises 2 % of the sample with positive labor income.6 I measure educational attainment as the number of years of education attained in the year previ- ous to the survey. Missing values for the variable of educational attainment comprise an additional 3 % of the sample. The pseudo panel When using repeated cross-sections ordinary least squares estimators are likely to be biased, however we can use cohort aggregates to obtain consistent estimators (Deaton, 1985). In other words, we can use a pseudo-panel. A pseudo-panel consists of cohorts (synthetic individuals) that we are able to follow over time. However, to obtain consistent estimators from a pseudo-panel, grouping variables –those that aggregate individuals into cohorts- must i) not have missing val- ues for any individual in the sample, ii) not vary over time, and iii) be exogenous and relevant (Verbeek, 2008). It is also important that the number of cohorts is large enough to avoid small sam- ple size problems, that cohort-sizes are large enough to avoid measurement error problems, and that cohorts somehow minimize within-cohort heterogeneity and maximize heterogeneity between cohorts (Verbeek and Nijman, 1993; McKenzie, 2000). I use gender, country and birth year in three-year spans as grouping variables –which should be exogenous and relevant (Verbeek, 2008). I group individuals in three-year-span birth-cohorts to reduce within cohort heterogeneity and obtain high heterogeneity between cohorts. This also provides me a relatively large number of cohorts to mitigate sample size problems and cohorts that, due to their large sizes, should not present considerable measurement error problems. Table A2, in the Appendix, shows the distribution of individuals and of cohorts. Income inequality estimates consist of Gini indices at the cohort level using survey weights. However, one potential problem of using labor income to compute inequality is that of self-selection into the labor markets, which leads to self-selection bias. However, if grouping variables are exoge- nous and relevant, they should act as instruments for self-selection into the labor markets (Moffitt, 1983; Moscarini and Vella, 2003). Therefore self-selection is not a concern. I proxy educational attainment as the average number of years of education attained and its dispersion as the standard deviation of the number of years of education attained. 5I detect outliers using the blocked adaptive computationally efficient outlier nominators algorithm (Billor et al., 2000), using the first percentile of the chi-squared distribution as a threshold to separate outliers from non-outliers. 6 The sample with positive labor income comprises 59 % of the individuals born between 1936 and 1977. 7 Measuring democracy I measure democracy as stock rather than a level variable. This implies conceptualizing democ- racy as an asset. It implies that democracy is a fungible resource that may accumulate over time, promising increased returns in the future. Along these lines, if a democratic regime endures it is likely to yield some return, for example: market-augmenting economic policies, political stability (understood as a reduction of uncertainty), rule of law, and efficient public bureaucracies (Gerring et al., 2005). But, why should we expect better results from a long-term democracy? All polities are subject to the “liability of newness” (Stinchcombe, 1965). New polities face a constellation of problems associated with their newly founded status which renders them particu- larly prone to failure. In democratic regimes the policy-making process, by definition, involves a considerable num- ber of players. Policy-making is a continual back and forth between different interest groups (e.g., legislators, citizens, political and economic organizations, etc.). Each of these players must learn to anticipate the goals, interests, and special sensitivities of the other players. Hence, the process of defining a “good policy” is likely to take considerable time; “lessons” arise only after many miscues. Not only must governing politicians learn what constitutes good policy; voters must also learn to recognize good policy. Only as democratic experience accumulates, we can expect a tran- sition away from a populist style of politics and policy-making to the institutionalization of “good” policies. (Gerring et al., 2005.) Off course, there can be reversals. In authoritarian regimes, in contrast, this process is generally monopolized by a small number of elite actors and has few mechanisms of accountability. Naturally, the longer these elites are in power, the greater their opportunities for gaining experience in the diverse tasks of governance. But, since the political environment is highly constrained, the development of legal-bureaucratic authority is virtually impossible. Therefore, it is less likely that “good” policies institutionalize. (Gerring et al., 2005.) To measure democracy I employ the Polity2 democracy index, which I draw from the Polity IV data set. This variable measures the extent to which “authority patterns are institutionalized” in a given country. It takes into account how the executive is selected, the degree of checks on executive power, and the form of political competition (Marshall et al., 2010). This indicator is highly sensitive, it employs a twenty-one-point scale (-10 to 10) from more autocratic to more democratic, and offers good historical coverage (1800-2013). To create a stock measurement of democracy I first add up each person’s Polity2 score from 1905 to one year before their year of birth and also up to when they are 20 years of age. I also 8 consider the stock of democracy between 30 years before their year of birth and 20 years after, and between their year of birth and 20 years after, as benchmarks. I assume that the stock of democracy depreciates at a rate of r per year, with r ∈ [0, 1]. This captures the possibility that democratic stock does not accumulate linearly over time. Now, given that for every individual born in any given year (t ) belonging to any given country ( p) the value of the stock would be the same, I compute the simple average of the proxy for the three birth years that compose the cohort to obtain the cohort average. To estimate r it is precise to specify how a particular historical path in country p up to year t maps into a value of the stock of democracy S p,t . The simplest way to do this might be to assume that democratic capital accumulates and is discounted geometrically at the rate (1 − r): S p,t = D p,t + (1 − r) × S p,t −1 , where D p,t denotes the value of Polity2 for the year t and country p. We can solve backwards to obtain t S(r) p,t = ∑ D p,s(1 − r)t −s, s=t0 where t0 = 1905. Thus, democratic experience is more valuable the closer to the present it is. Person and Tabellini (2009) find that r lies between 0.01 and 0.06; that is, the stock of democracy depreciates at a rate between 1 % and 6 %. I use the lower value of the depreciation rate for my main exercises, although for values greater than 1 % my conclusions do not change. Summarizing, my main variables of interest are the depreciated democracy stock of county p in the birth-year t , averaged for cohort c, defined by: t −1 1 3 t∑ ∑ Dc, p = 0,99t −s D p,s , (a) ∈c s=1905 t +20 1 3 t∑ ∑ Dc, p = 0,99t −s D p,s , (b) ∈c s=1905 t +20 1 3 t∑ ∑ Dc, p = 0,99t −s D p,s , (c) ∈c s=t −30 t +20 1 3 t∑ ∑ 0,99t −sD p,s Dc, p = . (d) ∈c s=t Figure 1, panel a, illustrates that younger cohorts have “experienced” a higher stock of democ- 9 racy. However, note that the confidence intervals are much wider when considering a shorter time span (Figure 1, panel b). Moreover, the average magnitude of the stock and its variation differ. Variation is much smaller for the shorter time span. Therefore, it is reasonable to think that the relationship between democracy and inequality is not likely to be stable when considering “short” time spans. It is important to note that I choose 1905 as the starting period to avoid considering the one thousand days war.7 I define the upper limit to be at most 20 years after the year of birth to con- sider all the years of individual formation (childhood and teenage years), and to avoid dropping the youngest cohorts from the sample in any of my exercises. For the benchmark, defining the lower limit of the span “30 years before the year of birth” allows me to avoid loosing the oldest cohorts given that I am not considering the one thousand days war period. Figure 1: Stock of democracy by birth cohort (a) (b) 0 100 birth year to 20 years after birth Average stock of democracy Average stock of democracy 1905 to 20 years after birth −20 50 −40 0 −60 1936−38 1939−41 1942−44 1945−47 1948−50 1951−53 1954−56 1957−59 1960−62 1963−65 1966−68 1969−71 1972−74 1975−77 1936−38 1939−41 1942−44 1945−47 1948−50 1951−53 1954−56 1957−59 1960−62 1963−65 1966−68 1969−71 1972−74 1975−77 Birth cohort Birth cohort 95% confidence intervals 95% confidence intervals Source: Author’s calculations based on data from Latin American household surveys, 1995-2009, biannual and Polity IV Project. 7A civil armed conflict between the liberal and conservative parties in the newly created Republic of Colombia. After the peace treaty Panama separated from Colombia. 10 4. Approach Given the panel settings of my data (panel of successive cross-sections), we can model the relationship between democracy and inequality as follows: Yc,g, p,e = α + β Dc, p + ρc,g, p,e + ε c,g, p,e , (1) where Yc,g, p,e denotes the Gini index of birth-cohort c, of gender g, of country p and survey e; D is a proxy on the accumulated quality of democratic institutions; ρ corresponds to cohort fixed effects. ε denotes the idiosyncratic error. This equation, however, is misspecified. The relationship between democracy and inequality is non-monotonic (Figure 2) –it shows the form of an inverted U. Acemoglu and Robinson (2002) suggest that, “economic development tends to increase inequality, but this inequality contains the seeds of its own destruction, because it in- duces a change in the political regime towards a more redistributive system”. In this setup, where I measure contemporary inequality, this is a plausible mechanism if it acts through an intergener- ational channel –a channel that can be affected by long-periods of democracy, such as education (Gerring et al., 2012). Indeed, the historical evidence suggests that democratization led to a surge in redistribution. The increased supply of educated workers that came to be as a result, and the direct impacts of these redistributive efforts, are key factors in the non-monotonic pattern of inequality (Acemoglu and Robinson, 2002). Figure 2: Stock of democracy vs. labor income Gini (a) (b) 70 70 2 2 R =0.08 R =0.02 Labor income Gini (0−100 scale) Labor income Gini (0−100 scale) 60 60 50 50 40 40 30 30 −250 −200 −150 −100 −50 0 50 100 150 200 −250 −200 −150 −100 −50 0 50 100 150 200 Democracy stock between 1905 and 20 years after birth Democracy stock between birth year and 20 years after birth Source: Author’s calculations based on data from Latin American household surveys, 1995-2009, biannual and Polity IV Project. 11 D is endogenous. Indeed, some researchers attribute the persistently high level of inequality in Latin America to the long lasting effects of colonial institutions. The encomienda (which gave Spanish conquistadors the right to Amerindian labor), the mita (a system of forced labor used in the mines), the repartimiento (the forced sale of goods to Indians, typically at highly inflated prices), the resguardo (which organized the Amerindian labor to be provided in the Haciendas and mines) and slavery, established important sources of economic and political inequality that endured after the region gained independence from their colonial masters and despite significant social, economic and political changes during the 20th century (Perry et al., 2003). This would likely lead to omitted variable bias, E (Dc, p , ε c,g, p,e ) = 0. To address the potential misspecification problems I rewrite Equation (1) as follows: 2 Yc,g, p,e wc,g, p,e = [α + β Dc, p + δ Dc, p + ρc,g, p,e + ηg + φ p + ϕe + ψ p,e + ε c,g, p,e ]wc,g, p,e , (2) where η , φ , ϕ and ψ denote fixed effects by gender, country, survey, and an interaction between survey and country fixed effects respectively. The gender fixed effects capture the differences in socioeconomic conditions between men and women, for example gender roles and other gender ˜ differences which are prevalent in the region (Nopo, 2012). The country fixed effects, survey fixed effects, and their interaction allow to control for long-run and short-run country-level economic changes, such as GDP growth, social public spending, inflation, etc., changes in survey lifting, and other possible effects of long-lasting legacies. In other words, this set of fixed effects should attenuate endogeneity by eliminating unobserved country, survey, gender, country level trends and cohort specific unobservables. Also, note that I multiply both sides of the equation by wc,g, p,e , which is equal to Nc,g, p,e , with N the cohort size. The reason for this is to correct for underlying 2 heteroscedasticity due to variation between periods in cohort sizes. Finally, Dc, p captures the non- monotonic relationship between democracy and inequality. Given that I have a number of time invariant variables I use ordinary least squares (OLS) to estimate my main regression. On this regard, if grouping variables are exogenous and relevant and cohort sizes are large enough (as they indeed are) I can estimate Equation (1) using OLS (Moffitt, 1983; Verbeek and Vella, 2005).8 8 Girma (2000) and Collado (1997) provide other methods to estimate pseudo-panel models that are useful in the presence of time-invariant variables. However, these do not offer any gain in terms of consistency, and provide less efficient estimators (Verbeek and Vella, 2005). Thus, I do not consider these approaches. 12 5. Results Table 1 shows the estimates of democracy on inequality for several specifications of the pseudo- panel model. Each duplet of columns corresponds to computing Equation (2) using equations (a), (b), (c) and (d) respectively to measure the stock of democracy. The first column of each duplet shows the results obtained after estimating Equation (2) not including gender, country, survey and survey-country fixed effects; the second column adds the complete set of fixed effects. In general, columns (1), (3), (5) and (7) suggest that democracy is negatively related to inequality, and that this relationship is non monotonic (δ < 0 and statistically significant). Nonetheless, by controlling for the set of fixed effects I obtain a lower value for β for (2) and (4); for columns (6) and (8) the value of β is larger, and in column (8), it is actually positive. On the one hand, these results provide evidence that benchmark specifications do not provide stable results (as expected). On the other hand, focusing on columns (1) to (4), we have a sense of the direction of the bias due to non-observable covariates: β is likely to be biased upwards –it is likely that the region’s democratic institutions are failing as a commitment device that prevents ruling minorities from erecting barriers to the excluded (Perry et al., 2003), and as a result the effect of democracy on inequality is under- 13 estimated in absolute value. I test for the robustness of all these specifications in section 5.2. Table 1: Estimates of the pseudo-panel model, income inequality 2 Equation: Yc,g, p,ewc,g, p,e = [α + β Dc, p + δ Dc, p + ρc,g, p,e + ηg + φ p + ϕe + ψ p,e + ε c,g, p,e ]wc,g, p,e Dependent variable: Income inequality (Gini, 0-100 scale) (1) (2) (3) (4) (5) (6) (7) (8) Stock of democracy From 1905 to 1 year -0.0406*** -0.0537*** before birth (0.000) (0.003) From 1905 to 20 -0.0497*** -0.0657*** years after birth (0.000) (0.004) From 30 years -0.0251*** -0.0025** before birth to 20 (0.000) (0.001) years after it From the year of -0.0399*** 0.0375*** birth to 20 years (0.000) (0.000) after it -0.0003*** -0.0003*** -0.0004*** -0.0004*** -0.0002*** -0.0001*** -0.0004*** -0.0005*** Squared value (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Constant 55.4062*** 58.6357*** 55.4062*** 58.6357*** 55.9672*** 50.2152*** 53.1822*** 42.9768*** (0.000) (0.364) (0.000) (0.364) (0.000) (0.368) (0.000) (0.431) Full set of fixed No Yes No Yes No Yes No Yes effects R-squared 0.783 0.907 0.783 0.907 0.783 0.907 0.783 0.907 Clusters 126 126 126 126 126 126 126 126 Observations 1850 1850 1850 1850 1850 1850 1850 1850 Note: Robust standard errors clustered by country and birth-cohort in parentheses. * Significant at ten percent; ** significant at five percent; *** significant at one percent. Specifications (1), (3) and (5) only include cohort fixed effects (ρ ). Specifications (2), (4) and (6) include the following fixed effects: cohort (ρ ), gender (η ), country (ϕ ), survey (φ ), and an interaction between survey and country fixed effects (ψ ). Source: Author’s calculations based on data from Latin American household surveys, 1995-2009, biannual and Polity IV Project. Now I focus on columns (1) to (4). In fact, let us consider the results of columns (2) and (4), in which I control for various fixed effects. Within the parameters of these specifications, a cohort with no existing stock of democratic capital experiences the following democratic inequality effect: for each full decade of high-quality democracy (Polity2 = 10), the democracy stock increases by approximately 100. To estimate the predicted effect of this change on inequality, I simply multiply this change by the coefficient on democracy stock, -0.0537 –column (2)- or -0.0657 –column (4). The predicted impact on inequality of a decade of high-quality democracy is then approximately between -5.4 and -6.6. Considering that the average labor income Gini is around 48, this means that by having 20 years of high quality democratic institutions we could reduce (on average) labor income inequality by 25 % –a sizeable improvement considering the high and persistent level of inequality in the region (Perry et al., 2003). 14 5.1. The role of education It is plausible to regard democracy as an important institutional factor in the development of education. Political elites in a democracy have electoral incentives to improve the quality of life for the least advantaged; these incentives are present to a much smaller degree in authoritarian systems. Indeed, studies have shown that democratic rule translates into improvements in education (Baum and Lake, 2003; Brown and Hunter, 2004; Ansell, 2008; Huber and Stephens, 2012; Harding and Stasavage, 2014). Because of the long-lived nature of human capital, the longer a democracy is in place, the more pronounced we can expect its impact on human capital to be (Gerring et al., 2012). Democracy might change the education output, mapping democracy onto long-run changes in the distribution of human capital and then onto contemporary changes in income inequality. Education, therefore, might be a suitable mechanism to explain the link between democracy and income inequality. This is indeed plausible, the level and the dispersion of educational attainment and income inequality are highly correlated (Figure 3). The pseudo-panel models for the level and dispersion of educational attainment follow that of Equation 2. Although for these equations D is also endogenous,9 the set of fixed effects should deal with endogeneity. It is also important to note the number of years of education has serial correlation. Although I restrict myself to individuals ages 24 years and older to avoid truncation in the distribution of educational attainment, individuals might still attain more education with time (e.g., graduate studies). Therefore, I also introduce a lag of the dependent variable as a regressor to control for serial correlation when analyzing the level of educational attainment.10 9 Governments in Latin America, particularly democratic governments, conducted substantial efforts to expand primary education during the 20th century, the elite was able to procure schooling services for their own children and resist subsidize services for others (Bethell, 1997; Perry et al., 2003; Schiefelbein, 2007). Indeed, governments did not provide adequate access to education to most people until the second half of the 20th century. Furthermore, the differences in the quality of education between the wealthy and the poor of the population prevailed. This would likely lead to omitted variables bias. 10 Verbeek and Vella (2005) show that if grouping variables are exogenous and relevant and cohorts sizes are large enough, it is possible to obtain and appropriate measure of Y c,g, p,e−1 , equivalent to the lagged value of Y c,g, p,e . 15 Figure 3: Educational attainment vs. labor income Gini (a) (b) 70 70 2 2 R =0.30 R =0.34 Labor income Gini (0−100 scale) Labor income Gini (0−100 scale) 60 60 50 50 40 40 30 30 2 3 4 5 6 7 0 5 10 15 Standard deviation of years of education Average years of education Source: Author’s calculations based on data from Latin American household surveys, 1995-2009, biannual and Polity IV Project. Similar to Table 1, in Table 2 each duplet of columns corresponds to computing Equation (2) using equations (a), (b), (c) and (d) respectively. Once again, it seems that the benchmark specifica- tions are not stable. The main specifications: columns (1) to (4), however, show consistent results. They show that democracy is positively related to the level of educational attainment (panel a) and negatively related to dispersion of educational attainment (panel b). The direction of the bias for β appears to be positive for educational inequality and negative for educational attainment. In both equations democracy and inequality show a slight non-monotonic relationship. Let us consider the results of columns (2) and (4) of panels a and b. A cohort with no existing stock of democratic capital experiences the following democratic educational effect: for each full decade of high-quality democracy (Polity2 = 10), democracy stock increases by approximately 100 points. Thus, for the level of educational attainment, by multiplying this change by the coefficient on democracy stock, 0.0252 –column (2)- or 0.0308 –column (4), I find that the predicted impact on educational attainment of a decade of high-quality democracy is approximately between 2.5 and 3.1 years of education. For the dispersion of educational attainment, by multiplying this change by the coefficient on democracy stock, -0.0036 –column (2)- or -0.0044 –column (4), I find that the predicted impact on educational inequality of a decade of high-quality democracy is approximately between 0.3 and 0.4 years of education. Considering that the average dispersion is almost 4.7 years, this means that by having 20 years of high quality democratic institutions we could reduce the dispersion in the years of education (on average) by 15 %. However, the coefficients for educational attainment are too big. Having 20 years of high quality democratic institutions would increase educational attainment on average by around 5.6 years; this does not seem plausible. 16 Table 2: Estimates of the pseudo-panel model, educational attainment 2 a. Equation: Yc,g, p,ewc,g, p,e = [α + θ Yc,g, p,e−1 + β Dc, p + δ Dc, p + ρc,g, p,e + ηg + φ p + ϕe + ψ p,e + ε c,g, p,e ]wc,g, p,e Dependent variable: Average years of education (1) (2) (3) (4) (5) (6) (7) (8) Stock of democracy From 1905 to 1 year 0.0100*** 0.0252*** before birth (0.001) (0.002) From 1905 to 20 0.0122*** 0.0308*** years after birth (0.002) (0.002) From 30 years 0.0100*** 0.0143*** before birth to 20 (0.001) (0.001) years after it From the year of 0.0135*** -0.0257*** birth to 20 years (0.002) (0.001) after it 0.0001*** 0.0001*** 0.0001*** 0.0002*** 0.0001*** 0.0002*** 0.000*** 0.000*** Squared value (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Constant 6.875*** 6.683*** 6.875*** 6.686*** 5.407*** 8.074*** 6.018*** 13.901*** (1.093) (0.412) (1.093) (0.365) (0.869) (0.521) (0.935) (0.822) Full set of fixed No Yes No Yes No Yes No Yes effects R-squared 0.961 0.985 0.961 0.985 0.961 0.985 0.961 0.985 Clusters 126 126 126 126 126 126 126 126 Observations 1598 1598 1598 1598 1598 1598 1598 1598 2 b. Equation: Yc,g, p,ewc,g, p,e = [α + β Dc, p + δ Dc, p + ρc,g, p,e + ηg + φ p + ϕe + ψ p,e + ε c,g, p,e ]wc,g, p,e Dependent variable: Standard deviation of years of education (1) (2) (3) (4) (5) (6) (7) (8) Stock of democracy From 1905 to 1 year -0.0047*** -0.0036*** before birth (0.000) (0.000) From 1905 to 20 -0.0057*** -0.0044*** years after birth (0.000) (0.000) From 30 years -0.0018*** -0.0005*** before birth to 20 (0.000) (0.000) years after it From the year of -0.0004*** 0.0030*** birth to 20 years (0.000) (0.000) after it -0.0000*** -0.0000*** -0.0000*** -0.0000*** -0.0000*** -0.0000 -0.0000*** -0.0000*** Squared value (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Constant 4.7907*** 4.0742*** 4.7907*** 4.0742*** 5.0486*** 4.2659*** 4.2966*** 3.9724*** (0.000) (0.023) (0.000) (0.023) (0.000) (0.028) (0.000) (0.032) Full set of fixed No Yes No Yes No Yes No Yes effects R-squared 0.820 0.893 0.820 0.893 0.820 0.893 0.820 0.893 Clusters 126 126 126 126 126 126 126 126 Observations 1850 1850 1850 1850 1850 1850 1850 1850 Note: Robust standard errors clustered by country and birth-cohort in parentheses. * Significant at ten percent; ** significant at five percent; *** significant at one percent. Specifications (1), (3) and (5) only include cohort fixed effects (ρ ). Specifications (2), (4) and (6) include the following fixed effects: cohort (ρ ), gender (η ), country (ϕ ), survey (φ ), and an interaction between survey and country fixed effects (ψ ). Source: Author’s calculations based on data from Latin American household surveys, 1995-2009, biannual and Polity IV Project. 17 5.2. Robustness In this section I verify if my results survive statistical scrutiny. I check whether my findings hold after controlling for periods of wars and economics crises. Both wars and economic crises can lead to regime transitions and affect socioeconomic outcomes. They can destroy or deplete capital assets, reducing top capital incomes (Piketty and Saez, 2003). They can also raise the income of the poor, for example, by employing them in the arms industry (as in the case of the United States during the Second World War) or reduce it, by leaving poor people jobless. These phenomena can also have effects on human capital. For example, UNESCO (2010) re- ported that conflict often results in smaller shares of the population with formal schooling, fewer average years of education, and decreased literacy rates, which persist over time and might affect particularly marginalized groups. Economic crises might also affect human capital. In Latin Amer- ica, for example, the economic crisis and structural adjustment during the 1980s economic crisis led to declines in family income, which may have resulted in growing inequality of educational opportunity (Fernandez and Lopez-Calva; Torche, 2010). Wars and economic crises can also hasten regime transitions by destabilizing political regimes. In Latin America, in many opportunities regime changes were result of a war that replaced the incumbent head of government with another. In Brazil, for example, after the end of World War II, Get´ulio Vargas regime became unsustainable because during the period of the war the threat of a German attack did not materialize, leading him to be swiftly overthrown in a military coup that “re- stated” democracy. In the Argentine Revoluci´on Libertadora (1955), the Peron regime was deposed etat by military forces. Economic crisis also preceded regime for a military dictatorship in a coup d’´ transitions. The economic depression of the 1930s weakened the legitimacy of emerging democrat- ic institutions, which led to social and political unrest between liberal movements and oligarchs and eventually to a military dictatorship in most countries backed up by the right-wing (which sought to protect the status quo). The economic crisis of the 1980s brought important economic problems for dictatorships to deal with: high unemployment, inflation, and growth stagnation, which had deligitimazing effects. Eventually the economic crisis of the 1980s led non-democratic regimes to an end and to the birth of representative democracies. (Bethell, 1997; Hagopian and Mainwaring, 2005.) The effects of wars and economic crises can be far reaching, and although we can consider them as shocks, they might have long-lasting consequences. Given that these two phenomena can precede regime transitions I test for the stability of my proxy of democracy by including proxies that account for these events in my main regressions. However, given their transitory nature it would not be wise to measure wars and economic crisis as stock variables. Instead, I compute several proxies 18 as level variables at different periods around a person’s year of birth. For both wars and economic crisis I compute the percentage of years of war or crisis that an individual born in year t experienced between t − 30 and t − 1; t − 20 and t − 1; t − 10 and t − 1; t and t + 10; t and t + 20 ; t − 30 and t + 20 ; t − 20 and t + 20, and t − 10 and t + 10. These comprise a comprehensive set of proxies. Table A3, in the appendix, presents the periods of wars and economic crisis that took place in the countries of interest. Given that for every individual born in any given year (t ) belonging to any given country ( p), and for a given time interval, the percentage of years of war or crisis would be the same, I compute the simple average of the proxy for the three birth years that compose the cohort to obtain the cohort average. That is 1 H c, p = ∑ Ht ,c, p, (3) 3 t ∈c where Ht ,c, p denotes the percentage of years of war or economic crisis for the birth year t belonging to cohort c and country p, and for a given time interval. I run 216 different regressions, which come from including each proxy in my main regressions with replacement. To save space, I include the regressions in an online appendix.11 My main re- sults hold for the relationship between inequality in labor income and educational attainment and democracy; that is, β < 0 and statistically significant. However, this parameter is not stable when using equations (c) and (d) –my benchmark specifications. This comes as no surprise because when considering shorter time periods we reduce variation in the independent variable. Moreover, we no longer consider the very long-run and cumulative effect of historical legacies. Thus I ask the read- er to consider only the results I obtain when using equations (a) and (b) to measure the stock of democracy. For the related specifications I find robust evidence of a non-monotonic relationship between democracy and labor income inequality and educational attainment inequality (although subtle for the latter one). For the case of the level of educational attainment, I not only find that the absolute value of β is much smaller, but I also do not find robust evidence that β > 0 (or the opposite). In other words, I find that there is no robust relationship between the stock of democracy and the average years of education. 11 https://sites.google.com/site/cfbalcazars/misc 19 6. Conclusions Despite many research documents assessing the link between democracy and inequality, empir- ical research is still not conclusive on whether and how democracy and inequality relate. This rela- tionship is a complex and multidimensional process that involves many socioeconomic phenomena. Empirical research usually resorts to cross-country panel data sets, but these impose several mea- surement error and endogeneity problems that are hard to solve. This paper tries to overcome many of these difficulties using pseudo-panel data built from several household surveys of nine Latin American countries, exploiting panel data techniques to control for endogeneity due to unobserved factors, and proposing a renewed approach which seeks to determine if long-run historical patterns in the degree of democracy relate to contemporary income inequality. My findings show that democracy is negatively related to income inequality, and that it is al- so related to lower dispersion in educational attainment. My interpretation of these results is: It seems plausible to regard democracy as an important institutional factor in the development of education. Political elites in a democracy have electoral incentives to improve the quality of life for the least advantaged; incentives that are present to a much smaller degree in authoritarian sys- tems. By changing the education output, political regimes map democracy onto long-run changes in the distribution of human capital and then onto contemporary changes in income inequality in a non-monotonic fashion. One of the concrete implications of these results is that if democratic institutions contribute to secular-historical changes in policy outcomes, then it is inappropriate to judge the results of institutional reforms on the basis of immediate policy gains. The immediate effects of institutional change are often negative since such change introduces uncertainties and information costs in the short run-positive changes are likely to take longer to materialize, since they depend upon the establishment of a new equilibrium. (Gerring et al., 2005.) It is unrealistic to expect such reforms to show instantaneous results. Many important points remain to be considered, I focus on a few important ones. First, what would happen if we consider younger cohorts by having access to recent household survey data – those born after the third democratic wave (Mainwaring and Hagopian, 2005). These cohorts should show larger democracy stocks. Based on the results shown above, we could guess that the estimated effect of democracy on inequality is likely to be larger than the one I estimate here. Second, I did not spend much time analyzing the non-monotonic relationship between the stock of democracy and inequality from the theoretical perspective. It is precise to rethink Acemoglu and Robinson’s theory of the political Kuznets curve from the inter-generational perspective, introducing the role of political change (Person and Tabellini, 2006). However, this is a task that I defer to future work. 20 References Acemoglu, D. (2008). Oligarchic vs. democratic societies. Journal of the European Economic Association, 6(1):1–44. Acemoglu, D., Daron, Naidu, S., Restrepo, P., and Robinson, J. (2013). Democracy, redistribution and inequality. National Bureau of Economic Research, Working Paper 19746. Acemoglu, D. and Dell, M. (2010). Productivity differences between and within countries. Ameri- can Economic Journal: Macroeconomics, 2(1):169–88. Acemoglu, D. and Robinson, J. (1998). Why did the west extend the franchise? democracy, in- equality, and growth in historical perspective. Quarterly Journal of Economics, 115(4):1167–69. Acemoglu, D. and Robinson, J. (2002). The political economy of the kuznets curve. Review of Development Economics, 6(2):183–203. Alesina, A. and Rodrik, D. Distributive politics and economic growth. The Quarterly Journal of Economics, 109(2):465–490. Amendola, A., Easaw, J., and Savoia, A. (2013). Inequality in developing economies: the role of institutional development. Public Choice, 155(1):43–60. Ansell, B. (2008). Traders, teachers, and tyrants: Democracy, globalization, and public investment in education. International Organization, 62(2):289–322. Atkinson, A. and Brandolini, A. (2001). Promise and pitfalls in the use of ’secondary’ data-sets: Income inequality in oecd countries as a case study. Journal of Economic Literature, 39(3):771– 799. Bardhan, P. (2005). Scarcity, conflict and cooperation: Essays in political and institutional eco- nomics of development, chapter History, Institutions, and Underdevelopment, pages 1–27. Mas- sachusetts: MIT Press. Baum, M. and Lake, D. (2001). The invisible hand of democracy: Political control and the provision of public services. Comparative Political Studies, 34(6):587–621. Baum, M. and Lake, D. (2003). The political economy of growth: Democracy an human capital. American Journal of Political Science, 47:333–334. erica Latina: Pol´ Bethell, L. (1997). Historia de Am´ ıtica y Sociedad desde 1930. Barcelona: Cam- bridge University Press. Billor, N., Hadi, A., and Velleman, P. (2000). Bacon: Blocked adaptive computationally efficient outlier nominators. Computational Statistics and Data Analysis, 34(3):279–298. Brown, D. and Hunter, W. (2004). Democracy and human capital formation education spending in 21 latin america, 1980 to 1997. Comparative Political Studies, 37(7):842–864. Chong, A. (2004). Inequality, democracy, and persistence: Is there a political kuznets curve? Eco- nomics and Politics, 16(2):189–212. Collado, M. D. (1997). Estimating dynamic models from time series of independent cross-sections. Journal of Econometrics, 82(1):37–62. Deaton, A. (1985). Panel data from time series of cross-sections. Journal of Econometrics, 30:109– 126. Engerman, S., Mariscal, E., and Sokoloff, K. (2000). Political Institutions and Economic Growth in Latin America, chapter Schooling, Suffrage, and the Persistence of Inequality in the Americas, 1800-1945, pages 159–217. Stanford: Hoover Institution Press. Engerman, S. and Sokoloff, K. (1997). How Latin America Fell Behind, chapter Factor endow- ments, institutions, and differential paths of growth among new world economies: a view from economic historians of the United States. Stanford: Stanford University Press. Engerman, S. and Sokoloff, K. (2000). History lessons: Institutions, factor endowments, and paths of development in the new world. Journal of Economic Perspectives, 14(3):217–232. Fernandez, A. and Lopez-Calva, L. (2010). Transitory shocks, permanent effects: impact of the economic crisis on the well-being of households in latin america and the caribbean. Estudios Economicos, 1(49):3–35. Gerring, J., Bond, P., Barndt, W., and Moreno, C. (2005). Democracy and economic growth: A historical perspective. World Politics, 57:323–336. Girma, S. (2000). A quasi-differencing approach to dynamic modelling from a time series of independent cross-sections. Journal of Econometrics, 98(2):365–383. Glaeser, E., Porta, R. L., de Silanes, F. L., and Schleifer, A. (2004). Do institutions cause growth? Journal of Economic Growth, 9(3):271–303. Gradstein, M., Milanovic, B., and Ying, Y. (2001). Democracy and income inequality: an empirical analysis. CESifo Working Paper No. 411. Harding, R. and Stasavage, D. (2014). hat democracy does (and doesn’t do) for basic services: School fees, school inputs, and african elections. The Journal of Politics, 76(1):229–245. Huber, E. and Stephen, J. (2012). Democracy and the left: social policy and inequality in Latin America. University of Chicago Press. Katz, L. and Autor, D. (1999). Handbook of Labor Economics, volume 3, chapter Changes in the wage structure and earnings inequality, pages 1463–1555. San Diego, CA: North-Holland. 22 Li, H., Squire, L., and fu Zou, H. (1998). Explaining international and intertemporal variations in income inequality. The Economic Journal, 108:26–43. Lindert, P. (2004). Growing Public: Social Spending and Economic Growth since the Eighteenth Century. Cambridge: Cambridge University Press. Mainwaring, S. and Hagopian, F., editors (2005). The Third Wave of Democratization in Latin America: Advances and Setbacks. Cambridge: Cambridge University Press. Marshall, M., Jaggers, K., and Gurr, T. (2010). Political regime characteristics and transitions, 1800-2010: Dataset users’ manual. Technical report, Center for Systemic Peace. McKenzie, D. (2004). Asymptotic theory for heterogeneous dynamic pseudopanels. Journal of Econometrics, 120(2):235–262. Meltzer, A. and Richard, S. (1981). A rational theory of the size of government. Journal of Political Economy, 89(5):914–927. Milanovic, B. (2000). The median-voter hypothesis, income inequality, and income redistribution: an empirical test with the required data. European Journal of Political Economy, 16(3):367–410. Moffit, R. (1993). Identification and estimation of dynamic models with a time series of repeated cross-sections. Journal of Econometrics, 59:99–123. Nikoloski, Z. (2009). Economic and political determinants of income inequality. Master’s thesis, University College London. Nikoloski, Z. (2010). Democracy and income inequality: Revisiting the long- and short-term rela- tionship. University of College London. ˜ Nopo, H. (2012). New Century, Old Disparities: Gender and Ethnic Earnings Gaps in Latin Amer- ica and the Caribbean. New York: Inter-American Development Bank and The World Bank. Palma, J. (2011). Homogeneous middles vs. heterogeneous tails, and the end of the ’inverted-u’: It’s all about the share of the rich. Development and Change, 42(1):87–153. Perry, G., Ferreira, F., and Walton, M. (2003). Inequality in Latin America and the Caribbean, Breaking with history. The World Bank. Persson, T. and Tabellini, G. (1994). Is inequality harmful for growth. The American Economic Review, 84(3):600–621. Persson, T. and Tabellini, G. (2009). Democratic capital: The nexus of political and economic change. American Economic Journal: Macroeconomics, 1(2):88–126. Piketty, T. and Saez, E. (2003). Income inequality in the united states , 1993-1998. The Quarterly Journal of Economics, 518 (1):1–39. 23 Reuveny, R. and Li, Q. (2003). Economic openness, democracy and income inequality: An empir- ical analysis. Comparative Political Studies, 36(5):575–601. Rodrik, D. (1999). Democracies pay higher wages. Quarterly Journal of Economics, 114(3):707– 738. Sachs, J. (2012). Reply to acemoglu and robinson’s response to my book review. Retrieved from: http://jeffsachs.org/2012/12/reply-to-acemoglu-and-robinsons-response-to-my-book-review/. Saint-Paul, G. and Verdier, T. (1993). Education, democracy and growth. Journal of Development Economics, 42(2):399–407. Sarkees, M. and Wayman, F. (2010). Resort to war: 1816 - 2007. CQ Press. Savoia, A., Easaw, J., and Mckay, A. (2010). Inequality, democracy, and institutions: A critical review of recent research. World Development, 38(2):142–154. Scheina, R. (2003). Latin America’s Wars Volume II: The Age of the Professional Soldier, 1900- 2001. Dulles: Potomac Books, Inc. Schiefelbein, E. (2007). Universalization of Primary Education in the Historical and Develop- mental Perspective, chapter Universalization of Primary Education in Latin America: The Poor Results and Their causes, pages 141–177. IDE-JETRO. Schumpeter, J. (1942). Capitalism, Socialism and Democracy. New York: Harper Brothers. Stinchcombe, A. (1965). Handbook of Organizations, chapter Social Structure and Organizations. Chicago: Rand McNally. Thorbecke, E. and Charumilind, C. (2002). Economic inequality and its socioeconomic impact. World Development, 30:1477–1495. Timmons, J. (2010). Does democracy reduce economic inequality? British Journal of Political Science, 40(4):741–757. Torche, F. (2010). Economic crisis and inequality of educational opportunity in latin america. Sociology of Education, 83 (2):85–110. UNESCO (2010). The quantitative impact of conflict on education. Think piece prepared for the Education for All Global Monitoring Report 2011. Verbeek, M. (2008). The Econometrics of Panel Data: Fundamentals and Recent Developments in Theory and Practice, chapter Pseudo panels and repeated cross-sections, pages 369–384. Berlin: Springer. Verbeek, M. and Nijman, T. (1993). Minimum mse estimation of a regression model with fixed effects from a series of cross-sections. Journal of Econometrics, 59(ss):125–136. 24 Verbeek, M. and Vella, F. (2005). Estimating dynamic models from repeated cross-sections. Jour- nal of Econometrics, 127:83–102. You, J. and Khagram, S. (2004). Inequality and corruption. Harvard University, Kennedy School of Government, RWP04-001. 25 Table A1: Data Sources Country Name of survey Year Coverage Country Name of survey Year Coverage Argentina Encuesta Permanente de Hogares 1995 Urban Panama Encuesta de Hogares 1995 National 1997 Urban 1997 National 1999 Urban 1999 National 2001 Urban 2001 National 2003 Urban 2003 National Encuesta Permanente de 2005 Urban 2005 National Hogares-Continua 2007 Urban 2007 National 2009 Urban 2009 National Bolivia Encuesta Integrada de Hogares 1995 Urban Paraguay Encuesta de Hogares (Mano de Obra) 1995 National Encuesta Nacional de Empleo 1997 National Encuesta Integrada de Hogares 1997 National 26 Encuesta Continua de Hogares 1999 National Encuesta Permanente de Hogares 1999 National 2003 National Encuesta Integrada de Hogares 2001 National 2001 National Encuesta Permanente de Hogares 2003 National 2005 National 2005 National 2007 National 2007 National Encuesta de Empleo 2009 Urban 2009 National Brazil Pesquisa Nacional por Amostra de 1995 National Peru Encuesta Nacional de Hogares 1997 National Domicilios 1997 National 1999 National 1999 National 2001 National 2001 National 2003 National 2003 National 2005 National 2005 National 2007 National 2007 National 2009 National 2009 National (Continues on next page) Country Name of survey Year Coverage Country Name of survey Year Coverage Colombia Encuesta Nacional de Hogares-Fuerza 1995 National Uruguay Encuesta Continua de Hogares 1995 Urban de Trabajo 1997 National 1997 Urban 1999 National 1999 Urban Encuesta Continua de Hogares 2001 National 2001 Urban 2003 National 2003 Urban 2005 National 2005 Urban Gran Encuesta Integrada de Hogares 2007 National 2007 Urban 2009 National 2009 Urban Honduras Encuesta Permanente de Hogares de 1997 National ositos M´ Prop´ ultiples 1999 National 2001 National 2003 National 2005 National 27 2007 National 2009 National Source: Author’s compilations based on data from Latin American household surveys, 1995-2009, biannual. Table A2: Sample Size Birth cohort Number of cohorts by year Number of individuals (years) 1995 1997 1999 2001 2003 2005 2007 2009 Unweighted Weighted 1936-38 14 18 18 18 18 18 18 18 29237 10034004 1939-41 14 18 18 18 18 18 18 18 42596 14354638 1942-44 14 18 18 18 18 18 18 18 54768 18299629 1945-47 14 18 18 18 18 18 18 18 75225 25036860 1948-50 14 18 18 18 18 18 18 18 95509 32201858 1951-53 14 18 18 18 18 18 18 18 117188 39522084 1954-56 14 18 18 18 18 18 18 18 139507 47440640 1957-59 14 18 18 18 18 18 18 18 159606 54468349 1960-62 14 18 18 18 18 18 18 18 175293 60275779 1963-65 14 18 18 18 18 18 18 18 191295 66839428 1966-68 14 18 18 18 18 18 18 18 188610 66279462 1969-71 14 18 18 18 18 18 18 18 198280 68779300 1972-74 0 18 18 18 18 18 18 18 178815 62720401 1975-77 0 0 18 18 18 18 18 18 165060 57798437 Total 168 216 216 216 216 216 216 216 1467114 503532031 Source: Author’s calculations based on data from Latin American household surveys, 1995-2009, biannual. 28 Table A3: Periods of Economic Crisis and Wars (1905-2013) a. Periods of war Country Years Type of war 1955 Intra-state Argentina 1975-1977 Intra-state 1982 Inter-state Bolivia 1932-1935 Inter-state 1912-1916 Intra-state Brazil 1932 Intra-state 1944-1945 Inter-state 1948-1958 Intra-state Colombia 1951-1953 Inter-state 1989-2013 Intra-state 1906 Inter-state 1907 Inter-state Honduras 1924 Intra-state 1969 Inter-state Panama n.a. n.a. 1911-1912 Intra-state Paraguay 1932-1935 Inter-state 1947 Intra-state 1932 Intra-state Peru 1982-1992 Intra-state 1995 Inter-state Uruguay n.a. n.a. b. Periods of economic instability Name of the crisis Years Economic recession 1928-1939 Latin American debt crisis 1982-1992 Latin American financial crisis 1998-1999 Global financial crisis 2008-2013 Note: Intra-state wars encompass wars that predominantly take place within the recog- nized territory of a state. Inter-state wars en- compass wars that take place between or among the recognized states. Source: Author’s compilations and Correlates of War. 29