wr$ Pz561 POLICY RESEARCH WORKING PAPER 2561 Democracy and Income Ideology, as proxied by a country's dominant religion, Inequality seems to be related to inequality. In Judeo-Christian An Empirical Analysis societies increased democratization appears to lead to lower inequality; in Mark Gradstein Muslim and Confucian Branko Milanovic societies it has an insignificant Yvonne Ying effect. One reason for this difference may be that Muslim and Confucian societies rely on informal transfers to reach the desired level of inequality, while Judeo-Christian societies, where family ties are weaker, use political action. The World Bank Development Research Group Poverty and Human Resources March 2001 POLICY RESEARCFI WORKING PAPER 2561 Summary findings Standard political economy theories suggest that The authors' cross-country empirical analysis, covering democratization has a moderating effect on income 126 countries in 1960-98, confirms the hypothesis: inequality. But the empirical literature has failed to ideology, as proxied by a country's dominant religion, uncover any such robust relationship. Gradstein, seems to be related to inequality. In addition, while in Milanovic, and Ying take another look at the issue. Judeo-Christian societies increased democratization The authors argue that prevailing ideology may be an appears to lead to lower inequality, in Muslim and important determinant of inequality and that the Confucian societies it has an insignificant effect. The democratization effect "works through" ideology. In authors hypothesize that Muslim and Confucian societies societies that value equality highly there is less rely on informal transfers to reach the desired level of distributional conflict among income groups, so inequality, while Judeo-Christian societies, where family democratization may have only a negligible effect on ties are weaker, use political action. inequality. But in societies that value equality less, democratization reduces inequality through redistribution as the poor outvote the rich This paper-a product of Poverty and Human Resources, Development Research Group-is part of a larger effort in the group to study inequality and income redistribution. The study was funded by the Bank's Research Support Budget under the research projects "Democracy, Redistribution, and Inequality" (RPO 683-01) and "Deriving World Income Distribution in 1988 and 1993" (RPO 683-68). Copies of the paper are available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contactPatricia Sader, room MC3-556, telephone 202-473-3902, fax 202-522-1153, email address psaderCworldbank.org. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at bgumail.bgu.ac.il, bmilanovic@worldbank.org, or yvonne yingCahotmail.com. March 2001. (44 pages) The Policy Research Working Paper Series disseminates the findings of sork in progress to eincourage 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, intterpretations, antd conclusions expressed int this paper are entirely those of the autbors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the countries they represent. Produced by the Policy Research Dissemination Center DEMOCRACY AND INCOME INEQUALITY: AN EMPIRICAL ANALYSIS Mark Gradstein*, Branko Milanovic** and Yvonne Ying***' JEL classification: D3 1 Key words: inequality, democracy, religion. * Department of Economics, Ben Gurion University (grade@bgumail.bgu.ac.il) ** Research Department, World Bank (bmilanovic@worldbank.org) *** Research Department, World Bank (yvonne_ying@hotmail.com) We would like to thank Prem Sangraula for excellent research assistance. The project was financed in part by the World Bank Resaeach Grants RPO 683-01 ("Democracy, redistribution and inequality") and RPO 683-68 ("Deriving world income distribution in 1988 and 1993"). 1 1 Introduction It has long been recognized that the distribution of income in an economy may, to a large extent, depend on political factors. More specifically, a natural hypothesis is that a more egalitarian distribution of political rights in the form of a political democracy should be accompanied by a more equal income distribution. Indeed, this hypothesis has a celebrated tradition in social sciences (Lipset,1959, Lenski, 1966, Meltzer and Richard, 1981). Empirical research testing this hypothesis has also loomed large. Political scientists, economists, and sociologists alike have devoted a great deal of energies arguing whether or not the hypothesis holds. The existing evidence, however, does not find any robust relationship between democracy and inequality in a cross-country regression analysis. Thus, Bollen and Jackman, 1985, fail to detect such a relationship; Li et al., 1998, find some limited support for a negative relationship between democracy and inequality; Rodrik, 1999, presents evidence that democracy is associated with a higher share of wages in GDP and thus lower inequality.' Indeed, a casual inspection of recent events in East Europe as well as in East Asia casts doubts that any such simple relationship may exist. It has been argued that, in the East European countries, democratization of the 90's actually resulted in an increase in income inequality - for the review of findings supporting this claim see Fleming and Micklewright, 2000. Similarly, some of the East Asian countries such as South Korea, Taiwan, Singapore have had among the most egalitarian income distributions in the world, yet their political record is far from democratic. These observations lead us to consider additional factors, which may affect income inequality alongside democracy. A clue in the search for such factors is provided 2 by the experience of the East European countries under the communist regime. There is little doubt that in this era, the political rights' record in these countries was especially miserable. Yet, the distribution of income was quite egalitarian, especially when compared with other countries with similar per capita income levels. This is most likely true even when one takes into account the fact that income was derived from non-market transactions, that some of it was given in kind, and a substantial portion of income was in fact determined by a person's status (see Atkinson and Micklewright, 1992, Milanovic, 1998, 1999, and the review of Fleming and Micklewright, 2000). One reason for this could be that the prevailing political ideology of these countries was deeply rooted in the egalitarian tradition. To be sure, differences in political power as well as in social status existed and were at least as powerful as everywhere else, bul income differences were not much approved by the populace. Ostentatious display of the rich was frowned upon and very uncommon; modest material life and the ability to make ends meet with little means were praised. This may imply that in a cross-country comparison, ideology needs to be taken into consideration when examining income distribution and its relationship to democracy. While the concepts of political culture and ideology have been controversial and elusive in the social sciences, they seem too important to be neglected altogether. Cultural values may play an important role in shaping policies and the resulting cross- country differences in economic outcomes may to a large extent hinge on different ideologies. Indeed, a substantial body of political literature exists (e.g., Almond and Verba, 1963, and Abramson and Inglehart, 1996) which emphasizes the relationship between culture and ideology on the one hand and political economic institutions on the 3 other hand. More specifically, Granato et al., 1996, studies how the former may affect economic development;2 Gradstein and Justman, 2000, investigate the socializing role of education and its effect on the efficiency of resource allocation; Bisin and Verdier, 2000, offer a dynamic evolutionary model of cultural transmission. This paper is an attempt to employ the insights derived from that literature to study the effect of democracy on income inequality. We stipulate that the outcome of redistribution, and hence inequality, hinges on the details of the political process: when it is democratized, the resulting level of income inequality is expected to be lower than when it is controlled by a rich oligarchy. But it also depends on the society's predisposition towards equity: if equity is valuable in itself, then even the rich oligarchy will avoid extreme inequality. This implies that democratization process in societies which value equity will result in only marginal further reduction in inequality. This is in contrast to societies for which equity in itself is immaterial, and where transition of political power to the majority results in much more aggressive redistribution and inequality reduction. This view of the effect of democracy on inequality through the prism of ideology is tested in the empirical part of the paper. Over the last decades, several reasonable measures of the degree to which countries are democratic have been generated. A subset of these measures is used in this paper. For inequality we use the high quality Gini data set compiled by World Institute of Development Economics Research (WIDER, 1999). While a proxy for ideology may be difficult to construe, as a first approximation we use the dominant religion in a country, whereas the predominantly atheist post communist countries form a separate category.3 Our analysis of unbalanced panel data for the period 4 1960-98 and 126 countries provides support for the hypothesis that ideological factors are important determinants of income inequality. While in certain countries, mostly in the Judeo-Christian tradition, the expansion of democracy is likely to result in substantial inequality reduction, in other countries (Buddhist/Hindu, Confucian, Communist) such effect is negligible or absent altogether. These findings are obtained even when controlling for other variables, which traditionally have been thought as affecting inequality. Another, surprising, finding of this paper is the different impact of political institutions on the relationship between democracy and inequality. Specifically, we present evidence that the negative relationship between democracy and inequality is more likely to hold-for a given level of democracy-in countries with parliamentary rather than presidential system. Although this finding was not anticipated by the empirical design, we consider it to be interesting enough to be reported here. The plan of the paper is as follows. Section 2 describes a very simple theoretical framework and its analysis; section 3 presents the data to be used for the empirical part, the analysis of which is undertaken in section 4; finally, section 5 closes with brief concluding remarks. 2 Theoretical framework 2.1. Description of the economy The model economy consists of a finite but large number of agents indexed by i =1,..,N. The agents are initially endowed with an exogenously given income yio. The distribution of initial income in the population is assumed to be skewed, so that the median income YmO lies below its mean Y0, and we let F denote the cdf of income. All individuals share 5 identical preferences. These are derived over an individual's income ex post taxes and transfers, as well as over the. distribution of this income across the individuals in the economy. Letting y,I denote individual i's ex post income, we write his overall utility as: U(yil) + W(y11, . .,YNI) (1) where both U and W are continuous, U is monotonic and concave and W is symmetric and quasiconcave. Note that the first term in the above expression is the utility from own income, whereas the second term represents the utility an individual derives from income distribution in the population. The assumptions on W guarantee that it captures preference for equality in the sense that its value increases as a result of an equalizing transfer from a richer to a poorer individual - see Rothschild and Stiglitz, 1973. It will be convenient to relate W to a standard inequality measure, such as the Gini coefficient, or a coefficient of variation and to think of utility as a linear combination between own income and such inequality measure; this then would allow us to write individual utility as: Yjl - a I(AYI,. . ., YNI) (1') where a > 0 is interpreted as the parameter related to disutility from income inequality, called equality preference, and I is the inequality measure. The above depicted economy is a special case of the one considered in, for example, Arrow, 1983, Becker, 1974, and Hochman and Rodgers, 1969. Existence of the second term in the utility specification implies that the individuals may want to voluntary transfer part of their initial endowment in order to decrease income inequality. As is shown in Arrow, 1983, however, the amount of voluntary transfers is typically inefficient because of the free riding effect: every giver hopes that the additional transfers will be 6 made by the other potential givers, hence not enough income will be transferred.4 The implication of this argument in the present context is that, typically, the equilibrium income inequality is excessively high.5 Alternatively, income redistribution could also be implemented through a formal redistribution mechanism. Following the long-standing tradition in the literature (see e.g. Meltzer and Richard, 1981, and the survey in Persson and Tabellini, 1999, part I), we assume that this is given by a combination of a proportional income tax and a lump-sum transfer. In addition, we assume that such redistribution is associated with a deadweight loss. The deadweight loss may stem from adverse labor supply effects, or from the hindering of innovative activity. Specifically, let t denote the income tax rate. The amount of the deadweight loss then is B(t) Yo, where B', B ">O, B(O) = 0, B(1) = 1. Taking the deadweight loss ilto consideration, the relationship between the ex ante and ex post income of individual i is as follows: yi,=(I-t)y0o y (t-B(t)) Yo (2) Note that Y1 = (1 - B(t))Yo, so that implementation of this redistribution mechanism results in average income loss, whose magnitude is positively related to the tax rate. The tax rate is chosen by a majority of votes. We assume that the population of voters consists of all individuals whose income exceeds a minimal threshold y: the lower is the threshold the more democratic the society is, as it allows a larger fraction of the population to be enfranchised. Finally, we denote I(t) = I((J-t) ylo + (t-B(t))Yo,...,(1-t) YNO + (t-B(t))Yo) the indirect utility from a more equal income distribution as a function of the tax rate. 7 2.2 Analysis In order to characterize the voting equilibrium of this mechanism, we first turn to the optimal tax rate from the viewpoint of voter i. The interior optimal tax rate is determined from the following FOC: - yio + (I -B'(t))Yo - a dI/dt = 0 (3) Differentiation reveals that the second order conditions are satisfied provided that d2IIdt2 2 0, which holds true for such inequality measures as the Gini coefficient and the coefficient of variation.6 Thus, individual preferences are single peaked, and a majority voting equilibrium exists. Differentiating (3) with respect to initial income yio, we obtain, given the assumptions, dt/dy,o -l /[B "(t) Yo + a d2Ildt2] < 0 (4) Thus, the preferred tax rate is a decreasing function of income implying that the politically decisive voter is the one with the median income among the voters. Differentiation of (3) with respect to a reveals that the higher the equality preference the higher is the chosen tax rate. Moreover, differentiating (4) with respect to a we obtain that d2t/dy1oda > 0, implying that the greater is the equality preference, the less steep is the negative relationship between income and the preferred tax rate, i.e., the more willing are the rich to bear high taxes. The implicit characterization of the interior equilibrium tax rate is as follows: - YdO + (I-B '(t))Yo - a dI/dt = 0 (5) where ydo denotes the income level of the decisive voter, which has the median income in the voters' population. The tax rate chosen by the median voter is the one that optimally 8 balances reduction in inequality and the deadweight loss of taxation; clearly, the equilibrium tax rate is an increasing function of the median voter's income. Now, democratization enfranchises some of the poor, lowering the minimal franchise requirement, y. As a result, political power is shifted to a poorer coalition of voters, so that a poorer voter becomes decisive (Ydo in (5) is decreased), which in turn results in a higher tax rate and a lower level of inequality (this is why the slope of the line in Figure 1 must be negative). As we have seen, this effect is weaker when there is concern for equality, so that a is large (which means that line BB in Figure 1 must throughout the whole range lie above the line AA and be flatter). The implication is that, when democratization is enhanced, its inequality reducing effect may be difficult to detect because of the intervening factor of the social norm of inequality intolerance. The implication is that, when democratization is enhanced, its inequality reducing effect may be difficult to detect because of the intervening factor of the social norm of inequality intolerance. Summing up, Proposition 1. Under formal redistribution, democratization results in higher taxes and transfers, hence in lower inequality. This effect is stronger when under the prevailing ideology concern for equality is small and is weaker when such concern is substantial.7 9 Figure 1. Optimal tax rate and income of the decisive voter in two societies Optimal tax rate B Society with a high a B Society with a low a A Income of the decisive voter We now turn to the empirical test of the hypothesis summarized in the above Proposition, beginning with the description of the data - the variables and the sample - and then proceeding with the estimation. 10 3. The data 3.1. Variables As argued in the previous part, the effect of democracy on inequality is mediated through prevailing ideology. We approximate ideology by dominant religion except in the case of Communist countries where the dominant ideology is, of course, Communism. Appendix 1 shows the ideology data for all the countries in the sample.8 Our rule in deciding what is a dominant religion in a given country was that at least 40 percent of the population had to have the same religion, with the second most numerous religion not exceeding 25 percent of the population. In a number of cases, however, two or even three religions have similar number of adherents, and a single dominant ideology could not be defined. In these cases, we have created hyphenated groups. They are African traditional/Christian which includes countries where Afri an traditional religion and Christianity claim about the same share of the population (Botswana, Cameroon, Central African Republic, Kenya, Madagascar, and Zimbabwe). In these countries, between 33 and 55 percent of the population profess Christianity (of all denominations), and between 25 and 50 percent of the population follow the traditional African religions. Another hyphenated category is Buddhist/Hindu. It includes India, Mauritius, Mongolia (after the end of Communism), Nepal, Sri Lanka, and Thailand. It was based on the assumption that cultural similarities between Buddhism and Hinduism are sufficiently great to treat them as one group. Arguing that Christian practices differ significantly between the countries that were converted to Christianity relatively recently, that is over the last two centuries as the result of European expansion, 11 we have created a category of "New Christian" countries that includes African (e.g. Gabon, South Africa, Uganda, Zambia), Caribbean (Trinidad and Tobago), and Oceanic (Papua New Guinea) countries. In these countries, the dominant religion is Christian but the population has been relatively recently converted (as opposed to, say, Australia or Canada which are what Maddison called "European off-shoots"). Finally, in some countries, more than two religious groups claim sizeable percentage of the population (e.g. Nigeria with more than 40 percent of the population following Islam, about a third Christian, and about a fifth professing traditional African beliefs). These countries where there is no single dominant religion (Burkina Faso, Cote d'Ivoire, Fiji, Guinea-Bissau, Guyana, Nigeria, Sierra Leone, Tanzania, and Trinidad and Tobago) were classified as 'mixed religion." For democracy, we use five variables. Two variables come from the Polity98D data source (version June 2000). 9 The two variables we use are the extent of democracy (Dem) and openness of the political system or party competitiveness (Parcomp). Dem defined as "general openness of political institutions" ranges from 0 to 10; parcomp, defined as "extent to which non-elites are able to access institutional structures for political expression" ranges from 0 to 5 (definitions taken from the codebook of Polity database by Jaggers [1996110). Values of both variables increase as level of democracy and political openness increase. The advantage of the Polity98D is that it provides a long series of data stretching in some cases back to the 19th century. The disadvantage, a serious one, is lack of transparency in how the scores are calculated. While the authors mention a number of checkpoints which they follow, they are very broad so that it very unclear how, in practice, they are instrumentalized, and the judgments are, of course, 12 subjective."' The problem with Polity database is that there are neither objective criteria used for measuring democracy, nor can a user see and check for himself how the authors have arrived to their judgments. Basically, one needs to accept the authors' judgments on faith. These drawbacks are remedied by the newly created Database of Political Institutions (DPI), which is explained and discussed in Beck et al. (2000). We use three variables from DPI. They are type of political system (System) which ranges from 0 to 2, with 0 indicating a presidential system, 1 assembly-elected president, and 2 a parliamentary system. Two points are important to underscore with respect to this variable. First, regimes with a low level of Executive Political Competitiveness (in other words, authoritarian or dictatorial regimes) score 0 on the System index; and the same holds for regimes where presidents are elected directly or by an electoral college (whose only function is to elect the president), and where there is no prime minister. Second, DPI uses a set of clear rules to distinguish between the parliamentary and presidential systems, such as presidential veto power, presidential appointment of ministers and dissolution of the parliamet. Thus, the French system is classified as parliamentary, because the Prime Minister depends only on parliamentary majority and not on president's will, and the Russian system as presidential since the Prime Minister is both proposed by the President and needs to be rejected by the Duma, which obviously requires a lower level of parliamentary support than the need to muster parliamentary majority. As can be seen from this example, the DPI database's main advantage over other measures of democracy is its transparency: the rules in classifying the regimes are very clear and are based on "objective" indicators. 13 The second DPI variable we use is Executive Index of Electoral Competitiveness (EIEC) index. The index ranges from 1 to 7, with competitively elected presidents or prime ministers depending on who is assigned the Chief Executive title (e.g., in the US, it would be president, in the UK, it would be prime minister) getting 6 or 7. For example, thr chief executives of Communist nations (the chairman of the Communist Party) are given a 3, because they are elected by the Party Congress, electing bodies which they do not appoint. Executives elected by small, appointed juntas or by appointed electoral colleges get a 2. Rival chief executives in one country, particularly in the setting of armed conflicts, are counted as No executives, and thus score a 1; see Beck et al., 2000, for a more complete elaboration. Very similar rules and ranking are used for Legislative Index of Political Competitiveness (LIEC). The scale also ranges from 1 to 7. The rules are: if there is no legislature LIEC scores 1, if there is an unelected legislature 2; elected legislature with single candidates (like in many Communist countries) scores 3; single party with multiple candidates scores 4; if multiple parties are legal but only one party won seats (like in many Arab countries), the score is 5; if some parties had won seats but the largest party received more than 75 percent of all seats, the score is 6; and finally, if there are multiple parties and none holds more than 75 percent of all seats, the score is 7. Table I shows simple correlations between the five measures of democracy. It can be easily noticed that the two measures from Polity98D (Democracy and Party Competitiveness) are very strongly correlated (0.93). One of the DPI measures (EIEC) also seems to measure similar aspects of democracy as Polity98D variables. The correlation between EIEC and the Polity measures is about 0.80. The correlation is 14 weaker (a little over 0.70) between the Legislative Index of Electoral Competitiveness and the two Polity98D measures. Finally, the correlation between System and all other measures of democracy is relatively weak. Basically, we can conclude that Democracy and Party Competitiveness seem to measure the same thing (which is not surprising in light of lack of transparent and objective criteria used in the derivation of the Polity98D data), and what they measure seems to be similar (close to) competitiveness in elections for the executive office. The variables LIEC and System do measure, as explained in the DPI manual, competitiveness in legislative elections, and the type of political system (presidential vs. parliamentary). Table 1. Sim le correlations between the various democracy measures System EEC LIEC Dem EIEC 0.z 4_____ LIEC 0.39 0.86 Dem 0.53 0.80 0.73 Parcomp 0.48 0.79 0.75 0.93 In addition to these five measures of democracy, we also create two interacted variables, by interacting respectively EIEC and LIEC with the political system. The last control variable we use is the level of economic development which we approximate by GDP per capita expressed in international dollars of equal purchasing power parity (PPP). The benchmark year is 1995 for which we have the actual $PPP levels for more than 100 countries. 12 We then use real GDP changes to derive the GDPPPP levels in the previous years going all the way to 1960. Thus all the GDP per capita data are expressed in PPP levels using the international prices of the year 1995. Most of the data are obtained from the on-line World Bank data base (called SIMA), 15 while for some countries-mostly transition economies, and in particular the former republics of the USSR-we had to calculate their real per capita growth rate using the countries' statistical yearbooks. 13 Finally, for the dependent variable, we use Gini coefficients as reported in the most recent (June 2000) WIDER database.'4 It contains 909 high-quality Gini observations from 126 countries. The data are described in detail in WIDER (1999). The Gini coefficients reported vary as to the recipient (household or individual), welfare indicator (income or expenditures), and net or gross measurement (net is after deduction of personal taxes). Following the approach taken Li, Squire and Zou (1998), we use dummy variables to adjust for each of these characteristics. 15 3.2. Sample Our data are longitudinal. We have a total of 126 countries in the sample.16 For most of these countries GDPPPP per capita data go back to the early 1960's. For example, from 1965 forward, in no year are there GDPPPP per capita data for fewer than 105 countries.'7 Note, however, that to conduct a panel analysis for all years since 1960, the limiting factor is the availability of the Gini data. In only a few exceptional cases (United States, UK, Taiwan, Bulgaria, India), inequality data are available for most years. For the vast majority of countries, such data are available in time intervals of several years, and for many countries, the WIDER data.base gives only a few observations.18 All currently existing countries (year 2000) are "projected back" into the past. We mean by that that individual data are collected for all former USSR republics, for the Czech republic and Slovakia separately, for all five successor states of the former 16 Yugoslavia, and for Pakistan and Bangladesh (until 1972, East Pakistan).'9 For all the "current" countries (formerly republics/parts of larger entities), we use the republican GDP's, population, or Gini coefficients. The political variables for each of the republics are often, but not always, the same as for the entire country where they belonged. While the GDPPPP data vary a lot in time, the variability is less for the political variables, either in Polity98D or DPI database. However, they too are time-variant as shown in Tables 2 and 3 on the example of Dem variable from Polity98D, and EIEC from the DPI database. We see from the last columns in Table 2 that only the least democratic (value 0) and the most democratic (value 10) observations tend to stay in the same group: conditional on having democracy level 0 or 10, respectively 66 and almost 70 percent of cases (or time) such countries remain in the same group. The constancy of the other levels of democracy is much less (between 11 and 23 percent). The same regularity is observed for EIEC: the stability of political arrangements is the greatest at the extremes. 64 percent of the time, countries where the chief executive is elected by a party congress or by referendum or "popular acclamation" stay in the same category; 52 percent of time, countries whose chief executive is competitively elected and gets less than 75 percent of the vote, remain in the same group. 17 Table 2. Variable Democracy from Polity98D Estimated level Number of Percent Number of Percent of Percent of of democracy observations countries with countries with observations in at least 1 such at least 1 same group observation: observation: 0 2417 46.2 86 71.1 66.1 1 260 5.0 32 26.5 18.4 2 109 2.1 18 14.9 14.1 3 142 2.7 29 24.0 11.4 4 86 1.6 17 14.1 11.4 5 86 1.6 19 15.7 10.2 6 170 3.3 26 21.5 15.0 7 220 4.2 21 17.4 22.6 8 379 7.2 48 39.7 18.0 9 256 4.9 25 20.7 22.9 10 1108 21.2 34 28.1 69.7 Total 5233 100 355 293.4 33.6 Table 3. Variable Executive Index of Electoral Competitiveness from DPI Definition of variable values Number of Percent Number of Percent of Percent o observations countries with at countries with at observations in least 1 such least 1 such same group observation: observation: Elected by a junta, or by an 485 17.5 51 41.5 42.0 appointed electoral college; or unilateral extension of the term of office; or elected for life (2) Elected by a party congress, 669 24.1 46 37.4 64.1 referendum or acclamation (3) 3.5 15 0.5 2 1.6 32.6 Several candidates from one party 54 1.9 7 5.7 36.7 (4) Several parties field candidates, 8 0.3 3 2.4 11.6 only one gets all votes (5) 5.5 5 0.2 1 0.8 21.7 Competitively elected with more 254 9.1 31 25.2 35.7 than 75% of vote (6) 6.5 94 3.4 20 16.3 20.5 Competitively elected with less 1196 43.0 101 82.1 52.3 than 75% of vote (6) Total 2780 100 262 213.0 46.8 The situation, however, is very different as far as religious variables are concerned. That variable itself is much more sluggish since religious composition of population does not change fast. Practically, the only source of variability is the change from Communist 18 to whatever the dominant religious affiliation may be in the case of countries that have abandoned Communism. Table 4 shows that out of 4861 observations (of countries in different points in time) some 22.7 percent are Catholic, followed by 17.1 percent Communist, 16.7 percent Muslim etc. The last column show the extent of variability in the ideology variable. We see that in 5 out of 11 religious affiliations, the variable is time- invariant. For example, the Christian mixed, New Christian, Confucian, or Buddhist observations are in 100 percent of cases constant; countries that are once classified as Catholic, remain so in 85.6 percent of cases, Protestant/ Evangelical Protestant remain so in 86.8 percent etc. On average, once a country is given a certain affiliation, in 85.6 percent of the cases it remains within that affiliation. The only reason why the variable is not entirely time-invariant is that Communist countries have, after the end of the Cold War, changed their dominant ideology. Table 4. Religious com position of the sample Number of Percent Number of Percent Percent of observations countries time to the ________ _ _________ _________ sam e group Catholic 1101 22.2 33 26.2 85.6 Protestant/Evangelical 406 8.2 12 9.5 86.8 Orthodox 142 2.9 10 7.9 36.4 Christian mixed/Judaism 195 3.9 5 4.0 100.0 New Christian 351 7.1 9 7.1 100.0 Muslim 812 16.4 24 19.1 86.8 Buddhism 203 4.1 6 4.8 86.8 African Christ/traditional 234 4.7 6 4.8 100.0 Confucian 205 4.1 5 4.0 100.0 Communist 861 17.4 27 21.4 83.7 Mixed 440 8.9 1 0 7.9 100.0 Total 4950 100 147 116.7 85.65 Note: Number of countries shows all countries that have, at least once, had a given religion. Thus, for example, if a country changes from Communism to Orthodoxy, the country would be included in both groups (Communism and Orthodoxy). This explains why there are 147 countries in the sample, and why the percentage column gives 116.7 percent. 19 4. Empirical estimation 4.1. Methodology We use the following specification: GINIJ, = fct(GDPi,t; DEMit; IDEOLOGY; IDEOLOGY * DEMti) (6) where subscript i refers to country and t to year (from 1960 to 1998). In the empirical estimation, GDP per capita expressed in the 1995 dollars of equal purchasing parity (PPP) enters both linearly and squared, as is conventionally done to reflect some Kuznets-type movement of inequality. Democracy (DEM) is approximated by the five variables described above. The IDEOLOGY dummies test for the possibility, explained earlier, that some ideologies may be more sensitive to equality than others. Finally, the interaction between ideology and democracy tests for our hypothesis (Proposition 1) that democracy may exert a differential impact on inequality depending on the prevailing religious/political affiliation of the country. In other words we posit that the effect of religion or ideology on inequality is exerted through two channels: directly (as reflected in the ideology dummies) and through differential effect of democracy on inequality depending on the religious-ideological context within which democratization occurs. We expect the, effect of GDP per capita to be of the usual inverted-U shape, the impact of democracy to be negative, while the coefficients on the IDEOLOGY dummies and the interaction term between ideology and democracy are not determined on an a priori basis. The data are an unbalanced panel covering 38 years and 126 countries, primarily because of unevenly spaced observations on the Gini coefficient. IDEOLOGY (religion) 20 is clearly exogeneous. While the effect of inequality on growth has recently been hotly debated, with arguments put forward that the effect is both positive and negative (for an excellent review of the literature and testing of the hypotheses, see Perotti, 1996), the influence of inequality on GDP levels is unlikely, so that reverse causality is not a problem. Reverse causation can, however, be a problem with the DEMOCRACY variable. Inequality can influence the level of democracy, and we address this problem by lagging DEMOCRACY. The use of lagged DEMOCRACY, however, is not only an econometric expedient. It has a substantive role too. It is reasonable to assume that the effect of democracy on inequality, if any, is unlikely to be instantaneous. While political changes can be fast (a country can move swiftly from a, dictatorship to a democracy, and perhaps within a relatively short time span from democracy to dictatorship), they are unlikely to immediately affect the relatively stable economic forces which underlie inequality. This is why in addition to one-period (one-year) lagged democracy, we also use the average of the DEMOCRACY values for years t-l to t-3.20 We also need to control for country effects. Without this adjustment it is quite conceivable, for example, that what is retrieved as a religious effect is in reality a country effect. This may be particularly a problem for religious practices that exist in only a few countries. For example, is relatively low inequality in Taiwan due to ideological preferences for equality, or to the fact that a successful agrarian reform and privatization were conducted in the 1950's which in turn, derive from the past of Japanese occupation and the Communist threat from the Mainland? All regressions therefore include country dummies. 21 4.2. Effect of democracy The first four columns in Tables 5 and 6 show the results of estimating equation 1 with four formulations of the DEMOCRACY variables; the next two columns include interaction of DEMOCRACY with the system variable. 21 Table 5 presents the results using a one-year lag formulation for DEMOCRACY, while in Table 6 we use the average value of DEMOCRACY during the three previous years. As expected, the effect of democracy is negative in all formulations but is not statistically significant (at 5 percent level) anywhere. However, the significance is almost always stronger when we use the three-year lagged formulation. This conforms with Muller, 1988, who similarly finds that the stability of democracy is a better predictor of inequality that the one time level of democracy. However, democracy may affect inequality not only directly but through the type of political system. In other words, democracy in a parliamentary system may have a different impact on inequality than democracy in a presidential system. The former is closer to a direct democracy and, by giving a greater role to the political parties and formation of coalition governments, may stimulate redistributionist policies of the type that we generally associate with the median voter behavior. The effect of a democratic presidential system on inequality is more difficult to gauge on an a priori ground. A strong president, once elected, is not subject to the day-to-day "control" of the political parties, and ultimately, voters which is a key characteristic of parliamentary regimes. President can thus pursue a wider range of distributional policies; in some cases, he/she may opt for policies that increase (e.g. Salinas in Mexico), and in other cases, for policies that reduce (e.g. Chavez in Venezuela) inequality. To account for the political system, we 22 interact political system (system variable from DPI) with competitiveness in election for the executive office and legislature (respectively EIEC and LIEC). The results are shown in columns 5 and 6 (Tables 5 and 6). We see that the parliamentary and the mixed system (strong president elected by parliament) are associated with reduced inequality compared to the presidential system. This effect is particularly strong and significant when we interact the type of polical system with competitivness in the election for the executive office: the Gini coefficient is some 0.3 points less, controlling for the level of democracy. We also interact each year during which country is in transition from Communism (that is, all years after 1990 for all formerly Communist countries) with democracy in order to sweep the already-noted paradoxical effect (see Gradstein and Milanovic, 2000) of post-Communist transition during which increase in democracy was associated with increase in inequalit). The coefficient is always positive, and in 11 out of 12 cases statistically significant with various (Polity- or DPI-measured) improvements in democracy adding beWween 1 and 3 Gini points. We conclude that the effect of democracy on inequality is negative but very weak. It becomes statistically more significant when we assess country's democracy by looking at its level over a longer (three-year) time period. With one-year lag, the effect of democracy almost vanishes. In addition, democracy reduces inequality more in parliamentary and mixed systems than in presidential systems, while during the transition from Communism, democratization is associated with increased inequality. 23 Table 5. Inequality, ideology and democracy (lagged by a year) Fixed effect regressions. Dependent variable: Gini coefficient Polity98D variables DPI variables | Definition of democracy Democracy Party Executive Legislative Executive Legislative competiti. competit. compet competit. compet Formulations 1 2 3 4 5 Ln gdp per capita 15.5 15.0 -5.9 -6.4 -4.5 -6.5 (0.00) (0.00) (0.51) (0.48) Ln gdp per capita squared -0.97 -0.94 0.36 0.42 0.3 0.42 (0.00) (0.00) (0.50) (0.43) (0.60) (0.44) Democracy interacted with: -0.37 -0.19 Presidential-assembly system ( (0.20) Parliamentary system -0.27 -0.22 (0.05) (0.10) Transition from Communism 0.74 2.9 1.2 1.6 1.3 1.7 (0.02) (0.00) (0.03) (0.06) (0.03) (0.05) One-year lagged democracy -0.13 -0.58 -0.07 -0.16 -0.09 -0.19 (0.26) (0.08) (0.82) (0.56) (0.76) (0.48) Religious dummy variables: (Catholic omitted) 5.2 (dropped) 18.7 12.5 -4.8 -12.0 Protestant (0.15) (0.01) (0.13) (0.85) (0.62) Orthodox 3.4 -1.1 3.7 -1.6 4.7 -0.53 (0.27) (0.76) (0.43) (0.82) (0.33) (0.94) Christianity mixed/Judaism (dropped) (dropped) (dope) (dropped) (dropped) (dropped) "New" Christianity -9.5 -7.0 11.5 -8.8 -4.7 -7.7 (0.15) (0.34) (0.03) (0.40) (0.51) (0,46) Muslim -11.6 -12.7 -9.0 -9.1 -9.3 -9.8 (0.00) (0.00) (0.01) (0.0 1) (0.01) (0.00) Buddhism/Hinduism -14.2 -17.9 2.7 -9.6 -10.7 (0.00) (0.00) (0.67) (0.26) (dropped) (0.21) African Christian/traditional religion -0.01 (dropped) (dropped) 7.1 -7.4 (dropped) 24 (0.997) (0.53) (0.35) Confucianism -24.4 -24.7 -13.7 -18.9 -19.4 -18.6 (0.00) (0.00) (0.02) (0.00) (0.00) (0.00) Communism 1.3 5.9 -4.1 -30.3 -2.3 -24.0 (0.66) (0.08) (0.49) (0.00) (0.70) (0.02) Mixed 4.3 15.0 16.2 12.6 15.1 15.5 (0.22) (0.00) (0.00) (0.01) (0.00) (0.00) Democracy interacted with: 0.21 1.8 -2.0 -2.2 1.6 1.6 Protestant (0.52) (0.01) (0.05) (0.05) (0.65) (0.65) Orthodox 0.48 2.9 0.0 0.11 0.1 0.11 (0.17) (0.01) (0.999) (0.92) (0.90) (0.92) Christianity mixed/Judaism -1.6 -3.8 -3.1 -3.2 -2.8 -2.9 (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) "New" Christianity 1.3 2.1 1.6 2.1 1.6 2.2 (0.03) (0.12) (0.07) (0.14) (0.07) (0.13) Muslim 0.4 0.10 -0.12 0.1 0.02 0.2 (0.07) (0.89) (0.79) (0.82) (0.96) (0.56) Buddhism/Hinduism -0.06 0.81 -1.5 0.3 -1.1 0.7 (0.87) (0.21) (0.08) (0.79) (0.18) (0.56) African Christian/traditional religion 0.73 -2.4 0.5 0.7 0.6 0.7 (0.25) (0.58) (0.55) 0.65) (0.53) (0.64) Confucianism 0.73 1.6 0.4 0.4 0.4 0.6 (0.01) (0.00) (0.63) (0.32) (0.65) (0.20) Communism 0.68 0.57 2.3 9.0 1.8 7.7 (0.29) (0.58) (0.16) (0.00) (0.28) (0.00) Mixed -1.2 -3.3 -1.7 -1.3 -1.7 -1.3 (0.00) (0.02) (0.00) (0.01) (0.00) (0.01) Dummy income (vs. expenditure) 5.6 5.5 5.9 6.6 5.9 6.5 (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) Dummy household (vs. individual) 0.5 0.2 0.7 0.7 0.7 0.6 (0.31) (0.58) (0.23) (0.27) (0.26) (0.28) 25 Constant -12.9 -9.5 71.2 70.8 65.2 71.9 (0.54) (0.65) (0.07) (0.07) (0.10) (0.07) Adjusted R2 0.88 0.88 0.92 0.90 0.90 0.90 (F) (48.3) (48.5) (41.0) (42.7) (41.6) (42.3) Number of observations 815 816 608 607 603 602 Note: P-values between brackets (if P<0.05 shaded). The omitted ideology is Catholic (both as the dummy, and in the interaction between democracy and religion); omitted system is presidential. The omitted country is Mexico which was Catholic and presidential throughout the entire period. Coefficients for country dummies not shown. 26 Table 6. Inequality, ideology and democracy (average level over the past three years) Fixed effects regressions. Dependent variable: Gini coefficient Polity98D variables DPI variables Definition of democracy Democracy Party Executive Legislative Executive Legislative competit. competit. compet competit. compet Formulations 1 2 3 4 5 Ln gdp per capita 18.1 19.0 -5.0 -1.4 -3.6 -1.1 (0,00) (O 00) (0.60) (0.88) (0.70) (0.91) Ln gdp per capita squared -1.1 -1.1 0.27 0.12 0.19 0.10 (0.00) (0.00) (0.63) (0.83) (0.72) (0.86) Democracy interacted with: -0.31 -0.10 Presidential-assembly system (0.06) (0.50) Parliamentary system -0.32 -0.19 ___ __ _ _ (0.02) (0.17) Transition from Communism 1.0 3.6 1.7 2.7 1.8 2.8 (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) Three-year lagged average democracy -0.10 -0.60 -0.15 -0.48 -0.17 -0.50 (0.40) (0.08) (0.63) (0.14) (0.58) (0.12) Religious dummy variables: (Catholic omitted) 8.9 24.2 21.0 18.9 1.8 -22.3 Protestant (0.01) (0.00) (0.00) (0.01) (0.94) (0.75) Orthodox 1.1 0.28 -0.84 -1.8 1.8 1.8 (0.68) (0.92) (0.84) (0.76) (0.67) (0.86) Christianity mixed/Judaism (dropped) (dropped) (dropped) (dropped) (dropped) (dropped) "New" Christianity 16.7 12.2 3.7 (dropped) -1.9 1.8 (0.00) (0.00) (0.45) (0.8l) (0.86) Muslim -10.7 -10.4 -11.4 -12.8 -12.0 -13.6 (Q.00) (0.00) (0.00) (0.00) (0.00) (0.00) Buddhism/Hinduism (dropped) -16.6 (dropped) (dropped) (dropped) (dropped) I (0.00) I_I_I African Christian/traditional religion 8.9 (dropped) (dropped) 31.3 (dropped) 31.8 27 (0.02) (0.25) (0.24) Confucianism -25.7 -25.2 (dropped) (dropped) -28.1 (dropped) _____________________________ _ (0.00) (0.00) _________ (0.14) ________0 Communism 2.8 7.0 1.6 -28.9 6.1 -20.0 (0.18) (0.04) (0.77) (° °°) (0.29) (0.08) Mixed 10.4 (dropped) 11.6 11.7 2.5 11.6 (0.01) (0.04) (0.03) (0.67) (0.03) Democracy interacted with: -0.19 -4.5 -2.7 -3.5 0.4 2.7 Protestant (0.57) (0.01) (0.01) (0.00) (0.91) (0.79) Orthodox 1.0 2.9 0.9 0.2 0.6 -0.1 (0.Q34 (0.00) (0.20) (0.81) (0.42) (0.92) Christianity mixed/Judaism -1.6 -3.8 -3.0 -2.8 -2.7 -2.9 ( (0.00) (0.00) (0.00) (0.00) (0.00) "New" Christianity 1.6 4.5 1.0 0.5 1.1 0.6 (0.02) (0.02) (0.32) (0.70) (0.29) (0.67) Muslim 0.19 -0.9 0.2 0.5 0.4 0.7 (0.60) (0.29) (0.75) (0.32) (0.45) (0.20) Buddhism/Hinduism -0.01 0.9 2.6 2.8 2.6 2.9 (0.98) (0.15) (0.00) (0.00) (0.00) (0.00) African Christian/traditional religion -0.3 0.7 -0.4 -2.9 -0.3 -2.8 (0.66) _ (0.92) (0.90) (0.44) (0.90) (0.46) Confucianism 0.81 1.6 1.8 0.9 1.7 0.9 (0.01) (0.00) (0.51) (0.11) (0.53) (0.09) Communism 1.7 1.0 1.1 10.2 -0.4 8.1 (0.24) (0.63) (0.50) (0.00) (0.82) : Mixed -1.9 -3.0 -1.6 -1.7 -1.6 -1.6 (0.0 1) (0.08) (O.U (01) (.1) (O 01) Dummy income (vs. expenditure) 6.5 6.5 5.5 6.1 5.5 6.0 (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) Dummy household (vs. individual) 0.52 0.50 0.51 0.45 0.39 0.42 (0.26) (0.28) (0.42) (0.48) (0.53) (0.50) Dummy gross (vs. net) 0.55 0.47 -0.05 0.02 0.22 0.12 28 (0.23) (0.30) (0.92) (0.96) (0.67) (0.81) Constant -26.1 -29.5 71.4 52.8 64.6 52.3 (0.20) (0.15) (0.09) (0.20) (0.11) (0.21) Adjusted R2 0.89 0.89 0.93 0.91 0.91 0.91 (F) (52.1) (51.6) (41.1) (44.0) (42.1) (43.6) Number of observations 795 796 560 559 556 555 Note: P-values between brackets (if P<0.05 shaded). The omitted ideology is Catholic (both as the dummy, and in the interaction between democracy and religion); omitted system is presidential. The omitted country is Mexico which was Catholic and presidential throughout the entire period. Coefficients for country dummies not shown. 29 4.3. Effect of religion Countries with Muslim, Confucian and Buddhist/Hindu ideology consistently show, both in Tables 5 and 6, a statistically significant lower inequality that Catholic countries (Catholicism, the most represented religion in terms of countries, is the omitted category). Since we do not know what really explains lower inequality in these societies, we call them "intrinsically more equal" (than Catholic), using "intrinsic" as a technical term, and implying thereby that there are certain preferences for equality which may be due to the differences in family formation (fewer nuclear households) or to greater informal transfers-points raised in Section 1 above. Looking at Table 5 and 6 results, the intercept (dunmmy variable) shows that the inequality reducing effect of Islam-which is the most consistent and statistically significant throughout-amounts to between 10 and 14 Gini points; the effect of Confucianism which is also statistically significant in all but one case ranges between -14 and -25 Gini points, while the effect of Buddhism/Hinduism is between -11 and -14 Gini points. Communism too shows a statistically significant negative effect in four regressions. At the other end of the spectrum, countries without a dominant religion (Burkina Faso, Cote d'Ivoire, Fiji, Guinea-Bissau, Guyana, Nigeria, Sierra Leone, Tanzania, and Trinidad and Tobago) consistently display greater intrinsic inequality (the dummy variable is statistically significant in almost all formulations both in Tables 5 and 6). Protestant countries and the "new" Christian countries also show, in some instances, a positive coefficient on the dummy variable. The intercept term for all other religions does not differ from the one for the Catholic countries. The implication of our finding is that Muslim and Confucian societies very strongly, and somewhat more tentatively . 30 Buddhist/Hindu, and Communist societies, exhibit certain features, independent of whether they are democratic or not, which make then more equal than other societies. This effect has been, in some previous empirical work on inequality, established for Communist societies (Kaelble and Thomas, 1991; Ahluwalia, 1976; Milanovic 1996) but not for the other three. 4.4. Interaction between democracy and religion It is also possible that ideology exerts an impact on inequality indirectly, that is in "determining" how a given level of democracy is "translated" (reflected) on inequality. This effect comes in addition to the direct effect captured by the religion dummies. To account for it, we interact religion dummy variables with democracy. We thus allow for religion to affect both the intercept and the slope coefficients. The results here somewhat vary between the two regressions. With a one-year lagged DEMOCRACY, only Confucianism and Communism show a positive (inequality increasing) effect of democracy (in at least two formulations out of six). The result for Communism is not unexpected because greater political liberalization in Communist countries in Eastern Europe (up to 1990), and in contemporary China and Vietnam, was accompanied by economic liberalization and increased income differences. Thus a combination of intrinsic inequality-reducing effect of Communism (as reflected in the intercept term), and increasing inequality with democratization makes intuitive sense. On the other hand, the group of Christian mixed societies composed of Canada, Switzerland, Germany, Israel, and the Netherlands shows a very strong inequality-reducing effect of democracy present in all formulations (see Table 5). Nations with mixed religion too 31 show that democracy reduces inequality: this effect is both strong and present in all the formulations. For example, a one standard deviation increase in Polity98D democracy index (equivalent to an improvement from the level of Armenia to that of Australia, both in 1998), reduces on average the Gini coefficient in mixed-religion societies by 5.6 points; similarly, a one standard deviation increase in the EIEC index (equivalent to an improvement from Vietnam to Zimbabwe in 1997), reduces the Gini coefficient by 3.6 points. Other religions show no statistically significant effect in more than one case. 22 When we approximate DEMOCRACY using its three-year average value, the number of religions with inequality-increasing effect of democracy goes up. In addition to Confucian and Communist countries, there is now a strong evidence of a positive relationship for Buddhist/Hindu societies, and somewhat weaker evidence for "New" Christian and Orthodox countries. On the other side of the spectrum, the group of Christian mixed/Judiasm and countries without a dominant religion is joined by countries where Protestantism is the dominant religion. They exhibit the same inequality reducing effect, although it is not statistically significant in all the formulations. Table 7 summarizes the effect of religion on inequality, by combining the "intrinsic" effect of religion on inequality, and the additional effect working through the interaction term. For the effect to be deemed significant, we request that the sign of the coefficient be the same (positive or negative) throughout all the formulations of the regression, and that it be statistically significant at the 5 percent level in at least two cases out of six. 32 Table 7. Effect of different religions on inequality 1/ Additional effect of democracy on inequality 3/ (depending on the religious context) Intrinsic effect of More equal than Same as under Positive effect (offsets religion 2/ under Catholicism Catholicism direct negative effect .of democracy) More egalitarian Muslim Confucianism than Catholic I Same as under Christian mixed/Judeo Orthodox Buddhist/Hindu Catholicism "New" Christian Communism African Christian/ traditional Less egalitarian Protestant than Catholic Mixed religion 1/ Based on the results from Table 6. 2/ As reflected in the religion dummy variable. 3/ As reflected in interaction between democracy and religion. Table 7 classifies all societies into six (out of possible nine) groups. Note first that empty cells are the "extreme" societies. There are, according to our results, no religions that are both intrinsically more egalitarian than Catholicism and where democracy has more of a pro-equality effect (than under Catholicism); nor are there religions that are intrinsically less equal than Catholic and where democracy increases inequality. Thus all religions cluster in the "middle" cells. Second, we note that in all cases where societies are predominantly Judeo-Christian greater democracy either reduces inequality-significantly as in the case of Protestant and mixed Christian societies, or mildly so as in the case of Catholic, Orthodox, "new" Christian, and African Christian/traditional societies. Unlike the Judeo-Christian societies, other cultures, notably Confucian and less so Buddhist/Hindu and Communist, 33 show that while the effect of increased democracy on inequality is positive, they apparently have other "intrinsic equalizers" independent of democracy which reduce inequality. The same "intrinsic equalizers" are very strongly present in Muslim societies too. Therefore, more equality seems, in the Judeo-Christian context, and this particularly in Protestant and mixed Christian societies, to be achieved through democracy, and presumably, the ability which democracy gives to poorer segments of society to redistribute some income (via government transfers and taxes) away from the rich.23 In the other societies (Muslim, Buddhist/Hindu, Confucian and Communist), the effect of democratization on inequality is small or even positive, but inequality is reduced, we surnise, through other tools like religious alms, broadly provided state-sector employment, private transfers between the generations, and generally stronger family, and perhaps ethnic, ties. Whether such societies are democratic or not does not seem to matter, as far as equality is concerned. Third, an interesting case is offered by societies without a dominant religion. These are in all but two cases (Fiji, and Trinidad and Tobago) African nations where traditional African beliefs, Christianity and Islam each appeal to a broad segment of the population. These religiously fragmented societies seem to possess certain features that make them intrinsically more unequal than other societies. However, on a positive note, democracy there is very strongly associated with reduced inequality, the way it is associated in mixed Christian and Protestant nations. Thus, democracy, in addition to its positive freedom-expanding effect, may also exert a desirable effect on inequality. To the extent that inequality stimulates inter-ethnic and inter-religious conflict, one may speculate that 34 democracy in such fractious settings may be a good tool for lessening inequalities and thus the underlying tensions.24 5. Conclusion In conclusion, we find that it is not democracy per se that matters for inequality-in fact, its direct net effect appears quite weak. Our findings suggest rather that democratization affects inequality indirectly. First, through the social context and societal values within which it takes place. For the Judeo-Christian societies, democratization is generally associated with reduction in inequality. For Muslim, Buddhist/Hindu and Confucian societies, democracy has either hardly discernible, or even a positive, effect on inequality. Yet these societies seem to possess some features which make them intrinsically more equal than the Judeo-Christian societies. It could be - although our empirical test does not account for that - that, the same "desired" level of inequality which in the Judeo-Christian societies is achieved through expanded franchise and government-sponsored redistribution, is implemented in the Muslem, Buddhist/Hindu, and Confucian societies informally, through family and ethnic ties. Second, our empirical analysis indicates that democracy "works" through the type of political system: controlling for the level of democracy, parliamentary systems are more likely to generate lower inequality than presidential systems. While this aspect was not perceived by us as the main motivation for pursuing this work, the robustness of the finding begs further empirical analysis and the development of theoretical foundations for the study of the effect of political institutions on inequality. 35 References Abramson, P. and R. Inglehart, 1995, Value Change in Global Perspective, University of Michigan press, Ann Arbor, MI. Ahluwalia, Montek S. (1976), "Inequality, Poverty and Development", Journal of Development Economics, No.3. Almond, G. and S. Verba, 1963, The Civic Culture: Political Attitudes and Democracy in Five Nations, Princeton University Press, Princeton, NJ. Arrow, K.J., 1983, "Optimal and voluntary income distribution," ch. 15 in Collected Papers of Kenneth J Arrow, vol. 1, The Belknap Press of Harvard University Press, Cambridge, MA. Atkinson, A.B. and J. Micklewright, 1992, Economic Transformation in Eastern Europe and the Distribution of Income, Cambridge University Press, Cambridge. Beck, T., G. Clarke, A. Groff, P. Keefer, and P. Walsh (2000), "New tools and new tests in comparative political economy: The Database of Political Institutions, World Bank Policy Research working Paper No. 2283, February 2000. Becker, G.G., 1974, "A theory of social interactions," Journal of Political Economy, 82, 1063-93. Bisin, A. and T. Verdier, 2000, "A model of cultural transmission, voting and political ideology," European Journal of Political Economy, 16, 5-29. Boix, C., 2000, "Democracy and inequality," working paper, http://home.uchicago.edu/-cboix/dem-ineq.pdf Bollen, K. and R.W. Jackman, 1985, "Political democracy and the size distribution of income," American Sociological Review, 50, 43 8-457. Collier, P., 1998, "The political economy of ethnicity," Proceedings of the Annual World Bank Conference on Development Economics, Washington DC. Deininger, K. and L. Squire (1996), "A new data set measuring income inequality", World Bank Economic Review, vol. 10:565-91. Deininger, K. and L. Squire (1998), New ways of looking at old issues: inequality and growth. Journal of Development Economics, vol, 57:259-87. 36 Easterly, W. (2000) "The Middle Class Consensus and Economic Development," World Bank Policy Research Department Working Paper 2346. Flemming, J.S. and J. Micklewright, 2000, "Income distribution, economic systems and transition," Ch. 14 in Handbook of Income Distribution, A.B. Atkinson and F. Bourguignon, eds. Elsevier, North-Holland. Gradstein, M. and M. Justman, 2000, "Human capital, social capital, and public schooling," European Economic Review. Gradstein, M. and B. Milanovic, 2000, "Does Liberte=Egalite?" Granato, J., Inglehart, R. and D. Leblang, 1996, "The effect of cultural values on economic development: Theory, hypotheses and some empirical tests," American Journal of Political Science, 40, 607-631. Hillman, Arye and H.-W. Urspung (2000), "Political culture and economic decline", European Journal of Political Economy, vol. 16, No. 2, July. Hochman, H.M. and Rodgers, J.D., 1969, "Pareto optimal redistribution," American Economic Review, 59, 542-547. Jackman, R.W. and R.A. Miller, 1996, "A Renaissance of political culture?" American Journal of Political Science, 40, 632-59. Kaelble, Hartmut and Mark Thomas (1991), "Introduction", in Y.S. Brenner, Hartmut Kaelble and Mark Thomas (eds.), Income Distribution in Historical Perspective, Caambridge and Paris:Cambridge University Press and Editions de la Maison des Sciences de l'Homme. Lenski, G., 1966, Power and privilege: A theory of social stratification, New York: Mc- Graw Hill. Li, Hogyui, Lyn Squire and Heng-fu Zou (1998), "Explaining intemational and intertemporal variations in income inequality", The Economic Journal, vol. 108: pp. 26- 43, January 1998. Lipset, S.M., 1959, "Some social requisites of democracy: Economic development and political development," American Political Science Review, 53, 69-105. Meltzer, A. and S. Richard, 1981, "A rational theory of the size of government," Journal of Political Economy, 89, 914-927. 37 Milanovic, Branko (1994), "Determinants of Cross-country Income Inequality: An Augmented Kuznets' Hypothesis", World Bank PRE Working Papers Series, No. 1246, January 1994. Milanovic, B., 1998, Income, Inequality, and Poverty during the Trnasition from Planned to Market Economy, The World Bank, Washington DC. Milanovic, B., 1999, "Explaining the growth in inequality during the transition," Economics of Transition, 7. Milanovic, Branko (2000), "The median voter hypothesis, income inequality and income redistribution: An empirical test with the required data", European Journal of Political Economy, September 2000. Muller, E.N., 1988, "Democracy, economic development, and income inequality," American Sociological Review, 53, 50-68. Perotti, Roberto (1996). "Growth, income distribution, and democracy: What the data say", Journal ofEconomic Growth 1, 149-187. Persson, T. and G. Tabellini, 1999, "Political economics and public finance," NBER WP 7097, http://www.nber.org/rapers/w7097. Rodrik, D., 1999, "Democracies pay higher wages," Quarterly Journal of Economics, 114, 707-738. Rothschild, M. and J.E. Stiglitz, 1973, "Some further results on the measurement of inequality," Journal of Economic Theory, 6, 188-203. Sirowy, L. and A. Inkeles, 1990, "The effects of democracy on economic growth and inequality: A review," Studies in Comparative International Development, 25, 126-157. WIDER (1999), World Income Inequality Database, World Institite for Development Economics Research, United Nations Development Programme, version Beta 3, 8 November 1999. 38 APPENDIX: RELIGIOUS COMPOSITION OF THE COUNTRIES Country Religious composition Assigned dominant religion Armenia Largest share Armenian Orthodox (94%), Russian Orthodox, Protestant, Islamic, Yazdi communities Communist; Orthodox Australia 26% Roman Catholic, 50% various Protestant Protestant Austria 77% Roman Catholic, 5% Protestants Catholic Belgium Roman Catholic 75% Catholic Burkina Faso Majority animist belief (65%), between 25% and 30% Muslims, some 10% Christian-mainly Roman African traditional Catholic Bangladesh 85% Muslim, 16% Hindu, some Buddhist and Christian Muslim Bulgaria Most are Bulgarian Orthodox Church (85%), and 13% Muslim. Communist; Orthodox Bahamas Most inhabitants profess Christianity, largest are Anglican, Baptist, Roman Catholic & Methodist Protestant Churches, overall various Protestants 52%, Catholics 19%. Belarus Major religion Eastern Orthodox (90%), and few Muslim and Jew Communist; Orthodox Bolivia Almost all Christianity, majority are Roman Catholic (95%) Catholic Brazil Almost all Christianity, 90% Roman Catholic Catholic Barbados Almost all Christianity, largest is Angelican Church, small group of Hindus, Muslims & Jews; overall 67% Protestant Protestant, 4% Catholic. Botswana 50% Christian; 50% traditional animist beliefs. African Christian/traditional Cent. Af. Rep. Christian between 33% and 50%, Muslims between 5% and 15%, traditional beliefs about 25% African Christian/traditional Canada 45% Roman Catholic, 26% Protestants Churches, other numerous religions Christianity mixed/Judeo Switzerland 48% Protestants, 44% Catholic. Christianity mixed/Judeo Chile 89% Roman Catholic, 11% Protestant Catholic China Confucianism, Buddhism, Daosim, also small Muslim and Christian minority (officially Communist) Communist Cote d'Ivoire 63% traditional beliefs, 25% Muslims, 12% Roman Catholic African traditional Cameroon 51% traditional beliefs, 33% Christian, 16% Muslims African Christian/traditional Colombia 95% Roman Catholic, also Protestant and Jewish. Catholic Costa Rica 95% Roman Catholic, remaining are also Christian Catholic Czechoslovakia Communist Communist Czech Rep. 40% Roman Catholic, 5% Protestant Catholic 39 FRG/Germany 45% Protestant, 39% Roman Catholic Christianity mixed/Judeo Djibouti More than 90% Muslims, 6% Christian Muslim Denmark 87% Evangelical Lutheran Church, small percentage other Protestent, and Catholic Protestant Dominican Republic More than 90% Roman Catholic, small communities of Protestants and Jews Catholic Algeria Almost all Muslims (99%) Muslim Ecuador More than 90% Roman Catholic Catholic Egypt Between 80% and 90% Muslim, others Christians mainly Copts Muslim Spain Almost all Roman Catholic (99%) Catholic Estonia 62% Evangelical Lutheran Church, 30% Orthodox Communist; Protestant Ethiopia Nearly half of the population are members of the Ethiopian Orthodox Church (North and Southern Orthodox plateau), the rest: Christian, Muslim and Animist Finland 85.7% Evangelical Lutheran Church, almost all Christian Protestant Fiji 53% Christians(mainly Methodists), 38% Hindus, 8% Muslims Mixed France 81% Roman Catholic, other Christians, some Muslims and Jews Catholic Gabon 60% Christians, mainly Roman Catholic "New" Christian Gambia 85% Muslims, 10% Christians, small numbers of Animists Muslim UK The Church of England dominant, some Roman Catholicism, Methodists and Baptists Protestant Ghana 38% traditional beliefs, 30% Muslim, 24% Christians Mixed Guinea Most inhabitants (85%) are Muslims, 8% Christian, some traditional animist beliefs Muslim Guinea-Bissau Animism (65%) and Islam (30%) are principal religions, 5% Roman Catholics and other Christians African traditional Greece 97% Greek Orthodox Orthodox Guatemala About 80% Roman Catholic, about 20% Protestants Catholic Guyana 57% Christianity, about 30% Hindu, Islam about 10% Mixed Hongkong Confucian Honduras 97% Roman Catholic Catholic Hungary 65% Catholic church, 20% Calvinist, 5% Lutheran Church Communist; Catholic Indonesia 87% Islam, about 10% Christians, remainders are Hindus and Buddhists Muslim India 80% Hindu, 11% Muslim, others are Christians, Sikhs, Buddhists, Jains and other minorities Hindu/Buddhist Ireland About 95% Roman Catholics, 5% Protestants Catholic Iran Great majority (95%) Muslims Muslim Israel 81% Judaism, 14% Muslims Christianity mixedWJudeo Italy About 90% Roman Catholic Catholic Jamaica Church of God' is the most numerous, majority population (56%) are various Protestant "New" Christian Jordan More than 90% Muslims, 8% Christians Muslim Japan Major religion: Shintoism and Buddhism; also Christian minority Confucian Kazakhstan Predominant religion is Islam (47%), 15% Eastern Orthodox Communist; Muslim Kenya 28% Catholic, 20% Protestants, about 20% African traditional beliefs, 8% Muslim African 40 Christian/traditional Kyrgyzstan The major religion is Islam (70%) Communist; Muslim Korea, South Confucianism, Buddhism and Chundo Kyo are principal traditional religions; large percentage of Confucian Christians (between 25% and 45%) Laos Communist Communist Sri Lanka Nearly 70% Buddhists, 15% Hindus, 8% Muslim, 8% Christian Hindu/Buddhist Lesotho 90% Christians, largest denominations are Catholic, Lesotho Evangelical and Anglican Churches "New" Christian Lithuania Predominantly Roman Catholic (80%), 10% Orthodox, small minorities of Lutherans and Calvinists Communist; Catholic Luxembourg 95% Roman Catholic, and small minority of Protestant Catholic Latvia Most are Lutherans or Roman Catholics, 37% Russian Orthodox or Old Believers Communist; Protestant Morocco Vast majority Muslim (99%) Muslim Moldova Largest denomination is Eastern Orthodox Church Communist; Orthodox Madagascar More than 50% Animist beliefs, 43% Christians, remainders are Muslims African Christian/traditional Mexico 90% Roman Catholic Catholic Mali About 80% Muslims, 18% traditional Animist beliefs, 1.2% Christians Muslim Mongolia No state religion but Buddhism is being encouraged Communist; Hindu/Buddhist Mauritania Almost all Muslims Muslim Mauritius 50% Hindu, about 30% Christians, 17% Muslims Buddhist/Hindu Malawi 75% Christianity, 10% traditional beliefs, 10% Muslims, Hindu minority "New" Christian Malaysia Between 55% and 60% Muslims, 19% Buddhists, some Hindus, Christians and some traditional beliefs Muslim Niger 95% Muslims, most remainders follow traditional beliefs Muslim Nigeria 47% Muslims, 35% Christians, 18% Animist beliefs Mixed Nicaragua Almost all are Roman Catholics (95%) Catholic Netherlands 33% Roman Catholics, 25% Protestants Christianity mixed/Judeo Norway 86% Evangelical Church, almost all profess Christianity Protestant Nepal 90% Hindu, 5% Buddhist, 3% Muslims Hindu/Buddhist New Zealand 42% various Protestants (18% Angelican Churchs, 13% Presbytaerians), 13% Catholic Protestant Pakistan Islam-State Religion, 97% Muslims Muslim Panama 85% Roman Catholics, 15% Protestants Catholic Peru 90% Roman Catholic Catholic Philippines 84% Roman Catholics, 4% Protestants, 6%Philippine Independent Church, 5% Muslims Catholic Papua New Gui 90% profess Christianity (most Protestant) "New" Christian Poland More than 90% Roman Catholic Church Communist; Catholic Puerto Rico Most Roman Catholic Catholic Portugal Almost all are Roman Catholic (97%) Catholic Paraguay Almost 90% Roman Catholic, small Protestants minority Catholic 41 Romania 83% Romanian Orthodox Churches, 6% Catholic Communist; Orthodox Russia Largest religion Russian Orthodox (85%), some Muslims, and Buddhists Communist; Orthodox Rwanda Most Christians (74%), 25% traditional Animist beliefs "New" Christian Sudan About 70% Muslims, 25% Animists, 5% Christians Muslim Senegal 94% Muslims, 4% Christians, mostly Roman Catholic, and a small number of Animists Muslim Singapore Principal religions are Taoism, Confucianism, Buddhism, Islam, Christianity and Hinduism Confucian Sierra Leone About 30% each of Muslim and traditional beliefs, 10% Christians Mixed El Salvador 88% Roman Catholic Catholic Soviet Union Communist Communist Slovakia 60% Roman Catholic Church, 8% Protestants (Evangelical Church) Catholic Slovenia Largest religion Roman Catholic Communist; Catholic Sweden Almost 90% Evangelical Lutheran Church Protestant Seychelles Almost all are Christians (90% Roman Catholics) Catholic Thailand Predominantly Buddhism (95%), 4% Muslims, and small Christian minority Hindu/Buddhist Turkmenistan Most population (87%) profess Islam Communist; Muslim Trinidad and Tobago 30% Roman Catholics, almost 30% various Protestant, 24% Hindus, 6% Muslims Mixed Tunisia State religion-Islam; almost all inhabitants (98%) are Muslims, Muslim Turkey 99% Muslims Muslim Taiwan Predominantly Buddhism, some Muslims, Daoists, Christians (both Catholic and Protestant). Confucian Tanzania 40% Christians, 33% Muslim, 25% African traditional beliefs, some Hindus Mixed Uganda More than 60% Christians, 16% Muslims "New" Christian Ukraine Ukranian Orthodox Church (750/O), 14% Catholics, some Muslims Communist; Orthodox USA Christianity is predominant religion (various Protestants a plurality) Protestant Uzbekistan Islam-predominant religion; some Orthodox Christian Communist; Muslim Venezuela 92% Roman Catholics Catholic Vietnam Communist Communist Yemen Almost all Muslims Muslim Yugoslavia Serbian Orthodox Church (80%), Muslims Communist; Orthodox South Africa Most inhabitants profess Christianity, some traditional African religions, small minority of Hindus and "New" Christian some Muslims Zambia More than 60% Christians, others profess traditional beliefs, some Muslims and Hindus "New" Christian Zimbabwe 55% Christians, a large numbers are in traditional beliefs, minority of Muslims and Hindus Aftrican _Christian/traditional Sources: Most of the data come from the Europa Yearbook (BM: what edition), complemented wth the Almanach and a number of Internet sources for more "difficult" countries (e.g. for Ethiopia, http://www.africanconnection.ore/docs/factsheets/ethiopia.html, for Sudan, http://w.wwv.sufo.deinon.co.uk/reli003.htm, for South Korea and Ivory Coast, http://atheism.about.com/reliuion/atlieisni/librarv/world/KZ/bl SKoreaReligion.htm, for Fiji, From http://wvw.fiii.gov.fi/aboutIhist.html). 42 ' See Sirowy and Inkeles, 1990, for a survey of earlier literature There also exists a related literature, which examines the reverse causal link, from inequality to democracy, see Boix, 2000, and references therein. While this paper generally abstracts from this direction, the empirical analysis below takes the possibility of such reverse link into account. 2 See also Jackman and Miller, 1996, for a dissenting view. 3 The dominant religion or ideology usually provides core values for a culture, although other social factors can obviously also shape peoples' values. 4 A central result in Arrow's paper is that, for a large enough economy, all efficient allocations of income are egalitarian in the sense that the difference between the minimal income level and the average one becomes negligible. 5 More precisely, this is so unless there is a single individual whose final income is above the minimal income of the receivers; a formal proof is available from the authors. 6 Specifically, suppose that I is one of these inequality measures and let Io refer to its pre-tax value and I,(t) refer to its post-tax value. Straightforward calculations reveal then that I(t) = (J-t) Io /(J-B(t)), so that dll(t) /dt = [-(I-B) +B '(I-t)]I0 /(J-B(t))2 < 0, and d211(t) Id? > 0. Thus, in both these cases inequality decreases as a result of increased redistribution, albeit at a diminishing pace. 7 One long run implication of the above proposition, which we, however, do not test directly here, is that democratization should result in a convergence in inequality levels across countries. s The terms "ideology" and "religion" will be used interchangeably. 9 The Polity98D database (produced in June 2000) can be downloaded from the Internet at httP:/lk- (zleditsch.socsci.ila.ac.uk/Polity.html-. 10 Available on the Internet at ftp://isere.colorado.eduipub/datasets/volity3/po]itv3.codebook. " For example, in explaining how the variable "Democracy" is constructed, Gurr (1997) writes that the variable is a sum of three elements: competitivness of political participation (coded 1 to 3), competitivness of executive recruitment (coded I to 2), and constraints on chief executive (coded I to 4). Political participation can be competitive (3 "democracy points"), transitional (2) or factional (1). However, it is left unclear what exact requirements need to be fulfilled in order for political participation to be deemed "competitive"; apparently this is left to the judgment of the authors. 12 The data come from the 1997 World Bank's World Development Indicators. 13 The calculation is performed as follows. For most countries, we have (from the World Bank sources) GDP per capita in the 1995 US$ for all the years going back to 1960. We also have the benchmark 1995 GDP per capita in $PPP. We then use the relationship GDP$t,95 GDPPPPt,95 GDP$95, p5 GDPPPP95,95 where GDP$,95 = dollar per capita GDP for year t expressed in dollars of 1995, and GDPPPP,t95= PPP per capita GDP for year t expressed in international prices (PPP) of 1995, to derive the GDPPPP for the years before 1995. 4 Kindly provided to the authors by Sampsa Kiiski. The data are also available at '5 Household dummy is derived from the Reference unit variable in WIDER ; welfare indicator dummy is derived from Income definution variable in WIDER. We also distinguish between gross and net income or expenditures. 16 Note that this number is larger than the number of countries with Gini or GDP data alone, because of the countries that may have one or another variable, but not all. 17 Between 1960 and 1964, the number of countries with GDPPPP per capita data varies between 84 and 90. After 1984, it is never less than 120 countries in each year. 18 For 19 countries, there is only one observation per country. 19 There are some exceptions. Ethiopia includes Eritrea until the separation of the two; Pakistan includes both West Pakistan and Bangladesh until 1971. Also, former East Germany (German Democratic Republic) is ignored, and is included only after the Unification in 1991. 20 Even dramatic regime changes like the Cuban revolution, or democratic changes in Eastern Europe took several years to "percolate" to the level of income distribution. Less radical political changes (partial democratization, strengthening of the opposition parties, greater openness of the media etc.) need even more time to affect inequality. 21 The country dummies are not shown since we are not per se interested in them. 22 Protestant countries show a statistically significant negative effect in two cases, and positive effect in one. 23 See the relationship between level of factor-income inequality and redistribution in Milanovic, 2000. The only non-Judeo-Christian society in that sample is Taiwan (Province of China), and it exhibits features that are markedly different from the rest: there is almost no redistribution and no reduction in inequality through the action of the government tax-and-transfer system. And Taiwan has the most equal factor-income level distribution of all the countries in the sample. 24 A similar point with respect to the role of democracy in prevention of violent conflict in fractionalized societies is made in Collier, 1999, and Easterly, 2000. Policy Research Working Paper Series Contact Title Author Date for paper WPS2542 Checks and Balances, Private Philip Keefer February 2001 P. Sintim-Aboagye Information, and the Credibility of David Stasavage 37644 Monetary Commitments WPS2543 When Do Special Interests Run Philip Keefer February 2001 P. Sintim-Aboagye Rampant? Disentangling the Role in 37644 Banking Crises of Elections, Incomplete Information, and Checks and Balances WPS2544 The Uniqueness of Short-Term Leora Klapper February 2001 A. Yaptenco Collateralization 31823 WPS2545 Financing the Future: Infrastructure Marianne Fay February 2001 A. Francois Needs in Latin America, 2000-05 37841 WPS2546 Gender Dimensions of Pension Paulette Castel February 2001 J. Smith Reform in the Former Soviet Union Louise Fox 87215 WPS2547 The Design of Incentives for Health Jeffrey S. Hammer February 2001 H. Sladovich Care Providers in Developing William G. Jack 37698 Countries: Contracts, Competition, and Cost Control WPS2548 International Provision of Trade Alan V. Deardorff February 2001 L. Tabada Services, Trade, and Fragmentation 36896 WPS2549 Measuring Poverty Dynamics and Erzo F. P. Luttmer February 2001 C. Wodon Inequality in Transition Economies: 32542 Disentangling Real Events from Noisy Data WPS2550 Measuring Equity in Health Care Adam Wagstaff February 2001 A. Maranon Financing: Reflections on (and 38009 Alternatives to) the World Health Organization's Fairness of Financing Index WPS2551 Infrastructure Coverage and the Kristin Komives February 2001 S. Minovi Poor: A Global Perspective Dale Whittington 30012 Xun Wu WPS2552 Inventories in Developing Countries: J. Luis Guasch February 2001 J. Troncoso Levels and Determinants-a Red Flag Joseph Kogan 37826 For Competitiveness and Growth WPS2553 The Value of Relationship Banking Giovanni Ferri February 2001 A. Yaptenco during Financial Crises: Evidence Tae Soo Kang 31823 from the Republic of Korea In-June Kim Policy Research Working Paper Series Contact Title Author Date for paper WPS2554 Administrative Costs and the Estelle James February 2001 A. Yaptenco Organization of Individual James Smalhout 31823 Retirement Account Systems: Dimitri Vittas A Comparative Perspective WPS2555 Implicit Pension Debt, Transition Yan Wang February 2001 A. Datoloum Cost, Options, and Impact of Dianqing Xu 36334 China's Pension Reform: Zhi Wang A Computable General Equilibrium Fan Zhai Analysis WPS2556 Household Strategies for Coping with Michael M. Lokshin February 2001 P. Sader Poverty and Social Exclusion in Ruslan Yemtsov 33902 Post-Crisis Russia WPS2557 Decentralization and Accountability: Stuti Khemani February 2001 H. Sladovich Are Voters More Vigilant in Local than 37698 in National Elections? WPS2558 Growth, Inequality, and Poverty: Martin Ravallion February 2001 P. Sader Looking beyond Averages 33902 WPS2559 Deposit Insurance as Private Club: Thorsten Beck February 2001 P. Sintim-Aboagye Is Germany a Model? 38526 WPS2560 Catastrophe Risk Management: John D. Pollner February 2001 J. Pollner Using Alternative Risk Financing 30079 and Insurance Pooling Mechanisms