inequality and democracy ：why inequality harms consolidation but does not affect democratization

7/29/2019 Inequality and Democracy Why Inequality Harms Consolidation but Does Not Affect Democratization

1/35

Inequality and Democracy: Why Inequality Harms Consolidation

but Does Not Affect Democratization

Christian Houle

World Politics, Volume 61, Number 4, October 2009, pp. 589-622 (Article)

Published by Cambridge University Press

DOI: 10.1353/wp.0.0031

For additional information about this article

Access Provided by Stanford University at 01/26/13 3:11PM GMT

http://muse.jhu.edu/journals/wp/summary/v061/61.4.houle.html
http://muse.jhu.edu/journals/wp/summary/v061/61.4.houle.htmlhttp://muse.jhu.edu/journals/wp/summary/v061/61.4.houle.html


2/35

INEQUALITY AND DEMOCRACY

Why Inequality Harms Consolidation but

Does Not Aect Democratization

By CHRISTIAN HOULE*

IntroductIon

Why do some autocracies democratize while others do not? Why

do some democracies endure and consolidate while others back-slide to dictatorship? Until recently most political scientists believedthat economic development drives both the emergence and the con-solidation o democracies. Przeworski and his coauthors revolution-ized the eld o comparative politics by showing that wealth has noeect on democratization but instead promotes consolidation.1 Sincethen scholars have shited their ocus away rom wealth level towardwealth distribution, raising the question o the eect o inequality ondemocracy. Two schools o thought have been particularly infuential.

The rst, advanced notably by Boix, argues that inequality harms bothdemocratization and consolidation.2 The second, proposed by Acemo-glu and Robinson, agrees with the rst that inequality inhibits consoli-dation but also predicts that it relates to democratization through aninverted U-shaped curve.3

Although these ideas are well developed theoretically, they haveyet to nd strong empirical support. Acemoglu and Robinson do notperorm statistical analysis but rather only present our case studies:

* Earlier versions o this article were presented at the 2008 annual meeting o the Midwest PoliticalScience Association, Chicago, at the Comparative Politics Workshop at the University o Rochester,and at the 2008 annual meeting o the American Political Science Association, Boston. I thank KevinClarke, John R. Freeman, Hein Goemans, Stuart Jordan, Jeremy Kedziora, Bonnie Meguid, MichaelPeress, Curtis Signorino, Randall Stone, Jay Ulelder, and especially Mark Kayser, Gretchen Helmke,

Alexandre Debs, G. Bingham Powell, and three anonymous reerees or excellent comments andsuggestions. I am also grateul to Francisco Rodriguez and Arjun Jayadev or sharing data. All errorsare mine.

1 Przeworski et al. 2000.2 Boix 2003.3 Acemoglu and Robinson 2006.

World Politics61, no. 4 (October 2009), 589622

Copyright 2009 Trustees o Princeton University

10.1017/S0043887109990074


3/35

590 worldpolItIcs

4 Acemoglu and Robinson 2006.5 One published paper (Burkhart 1997) does test a nonlinear relationship between inequality

and the level o democracy, but not between inequality and the probability o democratization. It doesnot distinguish between democratization and consolidation. Several unpublished papers test thisrelationship (or example, Epstein et al. 2004; Papaioannou and Siourounis 2005). Ziblatt 2008 testsor a curvilinear relationship at the subnational level, within Germany.

6 Boix 2003; Barro 1999.7 Deininger and Squire 1996.8 Deininger and Squire 1996. There are many reasons suggesting that the exclusion o some

regions could aect the results. For example, Middle Eastern and Arican countries tend to have bothintermediate levels o inequality and nondemocratic regimes. Their exclusion could lead one to nd

support or the inverted U-shaped relationship o Acemoglu and Robinson 2006, only because someo the less democratic countriesthat also have middle inequality levelswould be omitted rom theanalysis. The more recent version o the data set o Deininger and Squire 1996, available rom the

World Bank, is subject to the same problems.9 Ortega and Rodriguez 2006.

Singapore, Britain, Argentina, and South Arica.4 To my knowledge,there is to date no published article testing a nonlinear relationshipbetween inequality and democratization in a cross-national setting.5

Those who test the linear hypotheses, like Boix and Barro, do not dis-tinguish between democratization and consolidation, and/or they relyon databases subject to severe selection bias.6 For example, the widelyused data set o Deininger and Squire includes less than 11 percento the country-years during the period covered (195096).7 The prob-lem is that the available observations are unlikely to be representativeo the overall population. Thus, or example, while about one-third othe observations o Deininger and Squire are rom developed West-ern countries, less than 8 percent are rom sub-Saharan Arica.8 In the

whole population 30 percent o the observations are rom sub-SaharanArica.

This article argues and empirically demonstrates that the relation-ship between inequality and democracy parallels the one betweenwealth and democracy: inequality harms consolidation but has no neteect on democratization. It rst shows that democratization theoriessuer rom serious limitations, which in turn do not aect theories oconsolidation. The article then bridges the gap between the theoretical

and empirical literatures by conducting the most comprehensive em-pirical test to date o the relationship between inequality and democra-cy. Unlike most other empirical studies, I employ a methoddynamicprobitthat enables me to distinguish between the impact o inequal-ity on democratization and its impact on consolidation.

The measure o inequality employed, namely, the capital shares da-tabase o Ortega and Rodriguez, contains about 3500 observations orthe period between 1960 and 2000 in 116 countries.9 It has more than


4/35

InequalIty & democracy 591

10 Deininger and Squire 1996.11 Regime types are taken rom the extended data set o Przeworski et al. 2000.12 ODonnell and Schmitter 1986; Huntington 1991.13 Przeworski et al. 2000.

67 percent o the possible observations, a substantial increase over the11 percent o Deininger and Squire.10 It is also much more representa-tive. For example, 25 percent o its observations are rom sub-Saharan

Arica. Moreover, contrary to alternative measures o inequality, capitalshares are consistent with the theoretical literature, which ocuses oninterclass inequality rather than on overall inequality.

I nd no support or the two main theories that link inequality todemocratization; estimation suggests neither a monotonic negative re-lationship nor an inverted U-shaped relationship between inequalityand the probability o democratization. Yet consistent with the logic omy argument, inequality increases the probability o backsliding romdemocracy to dictatorship.

Taken together, these ndings have important implications or un-derstanding the conditions that promote democracy. Since World WarII many poor but equal countries, such as Costa Rica, India, and Mau-ritius, have successully established and sustained stable democracies.At the same time similar but unequal countries, like Nigeria, Peru, andTurkey, have oscillated between dictatorship and democracy. The keydierence between these two groups o countries is not the inabilityo the latter to create democratic regimesthey have done so several

timesbut their inability to maintain them. For instance, between1946 and 2000 Peru experienced our democratic breakdownsandthus several democratizationsin only twenty-three years o democ-racy.11 What characterizes Peru thereore is not its inability to institutedemocracy but rather its inability to preserve democracy.

This article not only contributes to the question o the relationshipbetween inequality and democracy, but it also addresses questions aboutthe undamental dierences between transitions to and rom democ-racy. Many authors, like ODonnell and Schmitter, and Huntington,

argue that the actors that infuence these two types o transitions arenot necessarily the same.12 In an interesting parallel to the results oPrzeworski et al., this study nds strong evidence that the relationshipbetween inequality and democracy mirrors the one between wealth anddemocracy.13 Additionally, it oers the rst explanation o why inequal-ity has more infuence on the consolidation o democratic regimes thanon their creation.


5/35

592 worldpolItIcs

InequalIty, democratIzatIon,and consolIdatIon

InequalItyand democratIzatIon

There are two main schools o thought which argue that inequalityis associated with regime type. The rst, advanced most recently byBoix, claims that unequal authoritarian countries are less likely to tran-sition to democracy.14 According to these authors, the eliteswho holdpolitical power in a dictatorshipare less likely to democratize wheninequality is high, because they ear redistribution.15 Others argue thateducation promotes democracy, because educated people are less proneto espouse extremist political positions.16 Because, ceteris paribus, thepopulation in egalitarian countries is typically more educated, inequal-

ity also harms the prospects or democracy by inhibiting education.Further, Lipset argues that a large middle classand thus a relative-ly equal income distributionalso promotes democracy, because themiddle class rarely supports extremist politics.17

The second school o thought is most notably associated with theseminal book o Acemoglu and Robinson.18 These authors suggest thatthe relationship between inequality and the probability o democra-tization ollows an inverted U-shaped curve. Equal countries do not

democratize because the potential redistribution and expropriationgains are small and thereore the population does not threaten to re-volt. The elites can thus maintain the regime without acing the threato a revolution. At intermediate levels o inequality, however, revolutionbecomes appealing to the population. The elites are unwilling to userepression, because redistribution is relatively inexpensive. Thereore,they democratize. At higher levels o inequality, the cost o redistribu-tion surpasses that o repressing revolts. The elites hence repress thepopulation and there is no democratization. It is thus at intermediate

levels o inequality, where the poor are willing to revolt and the elitespreer not to repress, that democratization is most likely. It is the cred-ible threat o a revolution by the population that ultimately pushes theelites to democratize.

There are at least three problems with the current theories that linkinequality to democratization. First, they are useul only or understand-

14 Boix 2003. See, or example, Lipset 1959; Dahl 1971; Muller 1995.15

This ollows rom the model o Meltzer and Richard 1981, according to which unequal democraciesredistribute more.16 See, or example, Lipset 1959.17 Lipset 1959.18 Acemoglu and Robinson 2006.


6/35


ing transitions rom below.19 In these theories, demand or democracyis always assumed to originate within the population. However, in real-ity democratization is oten driven rom above. Collier, and Llavador

and Oxoby show, or example, that in many West European and LatinAmerican countries democracy resulted rom intraelite competition,not interclass conficts.20 There is thus a large group o transitions orwhich these theories do not apply.

Second, even or transitions rom below, their predictions are un-likely to hold. Contrary to what most scholars have argued, inequalityactually has two opposite, potentially osetting eects on democrati-zation.21 On the one hand, inequality makes democracy more costlyor the elites by increasing redistribution, thus diminishing the prob-

ability o democratization. On the other hand, inequality increases thepopulations demand or regime change by increasing potential gainsrom redistribution or expropriation, thus increasing the probability odemocratization.22 I, as argued by the existing theories, democracy isdemanded by the population but requires the elites to acquiesce, the neteect o inequality on democratization is ambiguous.23

This objection is most clearly applicable to the rst school o thought,which argues that inequality decreases the likelihood o democratiza-

tion. This theory takes into account only the rst mechanism, accord-ing to which the elites become more willing to democratize as inequal-ity decreases. However, since people living in these countries are lesslikely to demand democracy in the rst place, the elites also have ewerincentives to concede it. While equality increases the willingness o theelites to democratize, it also decreases the willingness o the popula-tion to demand it. Because these authors completely ignore the secondmechanism, their predictions are unlikely to hold.

In contrast to the rst school o thought, the second set o theories

takes both mechanisms into consideration. At low levels o inequality,the elites do not democratize precisely because the population has noincentive to stage a revolution. However, the main nding o Acemoglu

19 See Ziblatt 2006.20 Collier 1999; Llavador and Oxoby 2005.21 For urther analysis o these dynamics, see Houle 2009.22 The empirical literature nds a positive association between inequality and political instability

and civil unrest. See, or example, Venieris and Gupta 1986; Alesina and Perotti 1996. Papaionannouand Siourounis 2005 make an analogous point by suggesting that high inequality increases the

likelihood o democratization by spurring opposition (pp. 3031). In his criticism o Boix 2003,Ziblatt 2006 makes a similar point.

23 O course, within a single case, one o these mechanisms may dominate, such that inequalityaects democratization in that particular instance. However, existing theories provide no ground toexpect one mechanism to prevail over the other in general.


7/35

594 worldpolItIcs

and Robinsonthe inverted U-shaped relationship between inequal-ity and democratizationrests on two unnecessary constraints on thecost o repression.24 One is that, short o a credible threat o socialist

revolution, maintaining an autocracy is assumed to have no cost, be-cause the elites do not need to repress at all. This drives the nding thatequal dictatorships do not democratize. While the cost o repressionis assumed to be zero in equal autocracies, the cost o redistribution isalways more than zero, although very low.

The problem is that in reality, even in very equal autocracies, main-taining the regime always requires some repression.25 What distin-guishes one dictatorship rom another is not whether or not it repress-es, but the levelo repression it imposes.26 Contrary to what Acemoglu

and Robinson assume, the choice o the population is not limited toeither staging a socialist revolution or not contesting at all.27 It canalso generate varying levels o social unrest that, by being costly to theelites to repress, push them to democratize.28 In other words, insteado mobilizing to install a socialist dictatorship, the population can pres-sure the elites to democratize by increasing the cost o maintaining theregime through, sometimes limited, contestation. Since there is alwayssome amount o social unrest and since scarce resources must be spent

repressing and limiting it, maintaining autocracy always requires somecost. Although equal autocracies require less repression than unequalones, the elites are also less willing to bear its costs, since democratiza-tion implies less redistribution than in countries that are more unequal.This leads to the same inconclusive predictions as above.

The other constraint is that the cost o repressing a revolt is assumedto be independent o its intensity. This assumption leads to the in-erence that very unequal countries do not democratize. According toAcemoglu and Robinson, the cost o redistribution rises with inequal-

ity, such that at a certain point it surpasses the cost o repression.29 In

24 Acemoglu and Robinson 2006.25 Wintrope 1998 demonstrates that even a perectly benevolent dictator who wants to maximize

the welare o his/her population relies on some (low) level o repression. One example is Singapore,which, despite being egalitarian, has experienced some social unrest since its independence (Smith2008).

26 As shown by Wintrope 1998, repression is best conceived as continuous rather than binary.27 Acemoglu and Robinson 2006.28 These costs include the direct cost o repression (or example, police orces) and its human

cost. There are also diverse economic losses related to social unrest and repression, such as capital

fight, reduced growth, decreasedfdI

infows, international sanctions and reduced oreign aid (Wood2000; Butkiewicz and Yanikkaya 2005; Campos and Nugent 2003; Meernik, Krueger, and Poe 1998).Repression is also likely to lead to divisions within the governing coalition, threatening the regimerom within. Finally, to limit contestation, a dictator may also have to engage in patronage, which iscostly (see Wintrope 1998).

29 Acemoglu and Robinson 2006.


8/35


reality, however, not all revolts are equally costly to repress.30 Wheninequality is high, the population also has stronger incentives to mobi-lize, thus increasing the required level o repression. In countries that

are only moderately unequal, a relatively small subset o the populationis likely to mobilize, and those that do sacrice little in the way o re-sources. Thereore, one cannot say, a priori, as assumed by Acemogluand Robinson, that at high levels o inequality the cost o repression isalways less than the cost o redistribution.31

The third issue is that both sets o theories ignore collective actionproblems. Mobilizing the population to oppose an autocracy posesserious challenges, because revolutions are public goods that cannotbe denied to nonparticipants.32 Moreover, each individual has a very

small impact on the likelihood o a revolution being successul, againdecreasing the expected benets o participation. These problems aremagnied by the act that participants ace large costs, possibly death.The population is thus unlikely to organize at all. The problem is thatthe existing theories expect inequality to aect democratization by de-termining the likelihood o the population rising up against the regime.But i the masses are unable to mobilize, the elites have no incentiveto respond to changes in inequality by adopting democracy. There-

ore, even i the earlier theories were right about the basic relation-ship between inequality and democratization although it has beencalled into question abovetheir explanatory power would be greatlyreduced.

InequalItyand consolIdatIon

The authors o both schools o thought discussed above agree that in-equality harms consolidation. They argue that the elites are more likelyto stage coups in unequal democracies because these are believed to

redistribute more, thus making democracy more costly or the elites.One striking observation is that most stable democracies that are poorturn out to be very equal. Some examples include India, Costa Rica,Uruguay, Jamaica, Mauritius, Papua New Guinea, and Mongolia sincethe all o the communist regime. Most East European countries are

30 Strictly speaking, according to Acemoglu and Robinson 2006, the cost o repression doesdepend on inequality, but notbecause inequality aects the intensity o the revolt. Instead, instabilityis assumed to destroy a xed portion o the total economy, implying that repression is more costly to

the elites when they own a larger share o the income. The problem is that the rate at which the cost orepression increases is implicitly assumed to be lesser than the rate at which the cost o redistributionrises with inequality. This last assumption is the key to driving the results, because it implies that athigh levels o inequality the cost o redistribution surpasses that o repression.

31 Acemoglu and Robinson 2006.32 See Tullock 1971.


9/35

596 worldpolItIcs

also examples o stable equal democracies. Greskovits argues that oneo the reasons democracies are more stable in Eastern Europe thanin Latin America is that the countries o the ormer are much more

equal.33

Moreover, Karl explains the relative stability o democracy inCosta Rica and Uruguay by the act that they are among the most equalLatin American countries. She urther argues that [i]nequalitys per-nicious undermining o democratic aspirations, institutions, and rulesis the greatestthreat acing democracy in the Americas today.34

Do theories linking inequality to consolidation suer rom the sameproblems as those linking inequality to democratization? No. The dis-tinction lies in the act that these two types o transitions tend to ollowdierent patterns. Whereas transitions rom autocracy to democracy

can ollow diverse paths, those rom democracy to autocracy almostalways take a similar one. As discussed above, democratization can bedriven either rom above or rom below. Some are the results o intrael-ite competition, others o pressure by the population, and still others othe direct seizure o power by the population. Democratic breakdowns,or their part, are almost always driven rom above. As argued by Hun-tington, With only one or two possible exceptions, democratic sys-tems have not been ended by popular vote or popular revolt.35 Coups

and rebellions against democracies are usually waged by the elites and/or the military, not by the population.36

Moreover, democratization (rom below) involves more groups oactors than democratic breakdowns. On the one hand, the theoreti-cal literature about the relationship between inequality and democra-tization has typically conceived democracy as being demanded by thepopulation but ultimately conceded by the elites. According to thesetheories, democratization is consensual, in the sense that it usually re-quires the agreement o both the population and the elites.37 In other

words, democratization rom below is a process that involves both so-cial classes. On the other hand, transitions rom democracy to dictator-ship do not require the explicit agreement o the population. Coups are

33 Greskovits 1997.34 Karl 2000, 156, emphasis added.35 Huntington 1996, 9.36 Here, it is not claimed that all democratic breakdowns are alike. In reality, transitions rom

democracy to autocracy have taken many orms (or example, see Linz and Stepan 1978). For example,they may result rom military coups or executive coups. The argument simply says that they tend to be

driven rom above, whereas democratization can be driven either rom above or rom below.37 In these theories, revolutions are ollowed by other (socialist) dictatorships, not democracies. In

reality, in some rare cases democratization has been imposed by the population without the agreemento the elites (or example, Nicaragua). However, this observation does not contradict the argument that

whereas democratization rom below is usually consensual, democratic breakdown rarely is.


10/35


oten successul even when not supported by the population.38 In thewords o Acemoglu and Robinson, [T]he move rom democracy todictatorship is almost never consensual.39 There have been instances

where coups against democracies have been prevented by popular mo-bilization. However, mobilization requires the population to solve itscollective action problem, which, as argued above, is unlikely. There-ore, while transitions rom autocracy to democracy can take diversepaths some o which involve many sets o actors, those rom democracyto dictatorship almost always ollow the same path, in which only theelites play a major role.40

The asymmetry between the two transition processes has key im-plications or the relationship between inequality and consolidation.

First, as democratic breakdowns ollow a single path, theories tryingto explain themcontrary to those concerned with democratizationcan be applied to almost all cases. Further, whereas inequality has twoopposite eects on democratization, it aects consolidation only nega-tively. The two eects o inequality on democratization come rom theact that democracy is demanded by the population but, in the end, isconceded by the elites. In these cases, inequality decreases the willing-ness o the elites to democratize but increases the incentives o the

population to contest the regime. By contrast, democratic breakdownsresult rom the direct seizure o power by the elites. Because the agree-ment o the population is not required, the eect o inequality on itswillingness to concede dictatorship has little impact. Inequality mainlyaects democratic breakdowns by increasing the cost o redistributionto the elites. Thereore, one should expect that when inequality in-creases the elites are more likely to wage coups against democracies, asargued by previous theories.

Finally, collective action problems do not signicantly reduce the

capacity o the elites to mobilize.41 Since the elites orm a much small-er group than the population, those involved in coups are more likelyto receive selective benets. For example, while those participating incoups oten obtain oces in the new regime, those who participate

38 This claim does not disagree with the act that, in many instances, the support o the populationacilitated the overthrow o democracy (or example, see Valenzuela 1978). It says only that democraticbreakdown is not a process in which the population concedes political power to the elites, but one in

which the latter seizes it (with or without opposition) rom the ormer.39 Acemoglu and Robinson 2006, 225.

40 It is important to note that these assumptions are shared by Boix 2003 and Acemoglu andRobinson 2006. Here, it is simply shown that, given their assumptions on transitions rom democracyto dictatorship, the objections presented in the previous section do not apply to their theories oconsolidation.

41 See Weede and Muller 1998.


11/35

598 worldpolItIcs

in revolutions rarely receive such benets. In addition, participants incoups are much more likely to aect its success than those participatingin revolts, again reducing collective action problems. An army ocer

has, or example, more infuence on the outcome o a coup than a singlepeasant on that o a revolution. Thus, the magnitude o the relationshipbetween inequality and consolidation, uncovered by the theoretical lit-erature, should not be signicantly reduced by collective action prob-lems.42 A combination o these arguments suggests that theories aboutthe eects o inequality on democratization should nd little empiricalsupport, whereas those about consolidation are likely to hold.

prevIous

lIterature

The empirical results on the relationship between inequality and de-mocracy are mixed. Table 1 summarizes the main large-N studies.Some authors nd strong evidence o a negative linear relationship,43while others nd no relationship.44 One o the major problems withmuch o this literature is the measure o inequality used. Most recentauthors rely on the Gini coecients o Deininger and Squire or theWorld Bank.45 Unortunately, these data come rom dierent sources

that may not be comparable.46

They contain observations based on ex-penditure and income, on net and gross income, and on household andper capita surveys. These dierent data sources may signicantly aectthe inequality measure. For instance, because o income redistribution,the observations based on gross income should systematically indicatemore inequality than those based on net income.47

The data sets used by previous authors also contain only a ewobservations and cover only a small number o countries. Boix, whouses the data o Deininger and Squire, has only 587 observations on

all countries or the period 1950 to 1990.48 The problem is that the

42 One could even argue that collective action problems increase the capacity o the elites to stagecoups against democracies, since they decrease the likelihood that the population will be able tomobilize to oppose coups.

43 See, or example, Muller 1988, 1995; Boix and Stokes 2003; Boix 2003.44 See, or example, Bollen and Jackman 1985; Midlarsky 1992.45 Deininger and Squire 1996.46 See Galbraith and Kum 2004; Atkinson and Brandolini 2001. Boix 2003 includes some

adjustments suggested by Deininger and Squire 1996 to account or these dierent data sources, orexample, by adding Gini points in cases where inequality is underestimated. However, Atkinson and

Brandolini 2001 argue that these adjustments do not eliminate the bias.47 It makes little sense to compare inequality observations based on net and gross income whenstudying democracy, since the causal mechanism relating inequality to democracy is preciselyredistribution.

48 Boix 2003; Deininger and Squire 1996. Boix increases the number o observations to about 1000by using ve-year averages.


12/35

table 1

empIrIcal studIesonthe relatIonshIpbetween

InequalItyand democracy

Studies Inequality Data Method Eect oInequality Limitations

Bollen andJackman

(1985)

income quintiles(World Bank,

N60)

weighted2SLS

none data quality and quantity no distinction

democratization/consolidation

no nonlinear testcross-sectional

Muller (1988) income Gini

income quintiles

(World Bank,N33)

bivariate reg. none (dem.)

negative

(cons.)

data quality and quantity

one control (econ. dev.)


Midlarsky(1992)

land Gini(N72)

agric. density

(N97)income quintiles

(World Bank,N55)

ols positive (land)none

(income)

data quality and quantityno distinction

democratization/consolidation

no nonlinear testpotential reverse causation

cross-sectional

Muller (1995) income Gini

income quintiles(World Bank,

N64)

ols negative data quality and quantity

no distinctiondemocratization/consolidation


Burkhart

(1997)

income quintiles

(World Bank,N224)

2sls inverted U data quality and quantity

no distinctiondemocratization/

consolidation

Barro (1999) income Giniincome quintiles(D&S, N303)

seeminglyunrelated

reg. (sur)

negative(weak)

data quality and quantityno distinction

democratization/

consolidationno nonlinear test

Boix (2003)

andBoix and

Stokes(2003)

income Ginib

(D&S, N1042)% amily arms

dynamic

probit

negative data quality and quantity

no nonlinear test

Epstein et al.(2004)a

inant mortality Markovtransition

Tobit

negative(Markov)

inverted U

(tobit)

data quality


13/35

600 worldpolItIcs

sample o available observations is strongly biased toward rich, demo-cratic countries that have the capacity and willingness to collect suchdata.49 Other authorswho do not use the data set o Deininger and

Squirenevertheless also rely on data sets with only a small numbero observations.50 For example, among cross-country studies, Mullercovers only 64 countries.51 By contrast, the data set used here includes116 countries.

Authors testing or an inverted U-shaped relationship also ndconficting results.52 Acemoglu and Robinson do not run any large-Ntest but look at only our case studies.53 Burkhart nds an inverted U-shaped relationship.54 However, he looks only at the level o democracy

measured with Freedom House scores and not at the probability o

49 While the average gdp per capita or the observations included in Deininger and Squire 1996is 9,260, it is only 6,342 in the ull population. By contrast, in the data set used here, the average gdpper capita is 7,099. Similarly, whereas the average polity score in the whole population is -0.74, it is3.61 or the observations in the data set o Deininger and Squire 1996. In my data set, the averagepolity score is 1.03.

50 Deininger and Squire 1996.51 Muller 1995.52 See, or example, Epstein et al. 2004; Papaioannou and Siourounis 2005. In an earlier version

(2004), cited by Acemoglu and Robinson (2006, 193), Papaioannou and Siourounis nd an invertedU-shaped relationship. However, once they include new observations, Papaioannou and Siourounis2005 are unable to nd any relationship.

53 Acemoglu and Robinson 2006.54 Burkhart 1997.

Papaioannouand

Siourounis(2005)a

income Gini(World Bank,

N1570)c

probit none data quality and quantity

Ansell and

Samuels(2008)a

income Gini

(B&M,N4728)c

% amily arms

dynamicprobit

positive(income)

negative

(land)

data quality

aUnpublished. Two o these papers have been published (Epstein et al. 2006; Papaioannou and

Siourounis 2008), but the published versions do not include the analysis on inequality.bN infated by taking ve-year averages.cN infated through linear interpolation. D&S denotes Deininger and Squire 1996. B&M denotes

Bourguignon and Morrisson 2002.

Studies Inequality Data Method Eect oInequality Limitations

table 1, cont.


14/35


regime transition.55 As a result, he is unable to dierentiate between theeects o inequality on democratization and their eects on consolida-tion. Further, his analysis contains data or only 56 countries, a set o

observations that may not be representative o the ull population. Forexample, Burkhart has ew observations on sub-Saharan Arica and,especially, the Middle East.56 The inclusion o more countries romthese regions may have aected the estimated relationship, since theytend to have both intermediate levels o inequality and authoritarianregimes.

Using tobit models, Epstein et al. nd an inverted U-shaped rela-tionship between inequality and the overall level o democracy, mea-sured with polity scores.57 However, when they distinguish between

democratization and consolidation, the nonlinearity vanishes. Instead,inequality decreases the probability o both democratization and consoli-dation. They measure inequality as the residual o the regression o inantmortality on variables unrelated to inequality.58 The problem is that eveni inant mortality is likely to be aected by inequality, it would make abetter measure o poverty; indeed, many poor and equal countries, likeIndia, have high inant mortality rates.59 Using this measure would leadus to think that inequality in these countries is actually high.

data

dependent varIable

Regime types are taken rom the Przeworski et al. extended database,which covers most countries rom 1950 to 2002.60 There are two typeso regimes: democracies and autocracies. To be dened as democratic,a country must satisy our conditions: (1) the chie executive must be

elected either directly or indirectly by the population; (2) the legisla-ture must be elected directly by the population; (3) there must be morethan one party; and (4) there must have been at least one alternation in

55 Burkhart 1997.56 Burkhart 1997.57 Epstein et al. 2004.58 These are access to sae water, access to health care, health expenditure per capita, total health

expenditure, calories per capita per day, a standard o living index, reports o amine, people aectedby drought, people aected by earthquakes, people aected by foods, people aected by unnamedstorms, and people aected by named storms.

59

Child mortality is nearly twice as large in India as in Brazil (World Bank data). However, Indiais relatively egalitarian, while Brazil is among the most unequal countries. In addition, inant mortalityis highest in sub-Saharan Arica, whereas Latin America is by ar the most unequal region.

60 Przeworski et al. 2000.


15/35

602 worldpolItIcs

power due to elections.61 Any regime that does not satisy one o theseconditions is authoritarian.

Independent varIable

As noted above, the most important obstacle to studying inequality isthe lack o reliable data. Problems arise both rom the low number oobservations and rom the poor quality o those that are available. Thisarticle measures inequality with the capital share o the value added inthe industrial sector, assembled by Ortega and Rodriguez.62 It mea-sures the proportion o the value added accruing to the capital owners.Dunning, Acemoglu and Robinson, and Przeworski et al. have also

recently used capital shares to measure inequality.63

According to Dun-ning, capital shares represent the best available cross-national indica-tor o private inequality.64 Low capital shares are associated with lowinequality, because a great portion o the value added in productionis accruing to the labor class, as opposed to the capital owners. Thedatabase has about 3500 observations covering 116 countries between1960 and 2000 and was constructed rom data collected by the UnitedNations Industrial Development Organization (unIdo).

Using capital shares to measure inequality has a number o theoreti-cal and empirical advantages. First, it is consistent with the theoreticalliterature that ocuses on intergroup inequality rather than on overallinequality. Most authors believe that it is only inequality across socialclasses that aects the regime type. For instance, Boix, and Acemo-glu and Robinson look at inequality between the poor and the elites.65Capital shares measure the relative income o the elites.66

A second advantage is that capital shares are assembled by a singlesourcethe United Nationsthat uses the same denitions and method

61 The last rule is applied retrospectively. For example, in Japan the ldp (Liberal DemocraticParty) won all elections until 1993, when it lost and voluntarily relinquished power. Because the ldprespected the electoral results, Japan is coded as democratic during the ull period. Malaysia held threeelections between 1957 and 1969. The incumbent party won the rst two but not the third. Even so,it reused to cede power. Malaysia and similar cases are autocracies during the whole period. Finally,there are instances, such as Botswana, where the incumbent party never lost elections. As one cannotknow whether it would give up power, Przeworski et al. 2000 assume that these countries are notdemocratic.

62 Ortega and Rodriguez 2006. Capital shares are dened as one minus the labor shares, whichmeasure the ratio o compensation o employees to the value added in production.

63 Dunning 2008; Acemoglu and Robinson 2006; Przeworski et al. 2000.

64 Dunning 2008, 143.65 Boix 2003; Acemoglu and Robinson 2006.66 For example, Acemoglu and Robinson 2006 dene the income o the population and the elites (or

rich) respectively as yp = (1-)y/(1-) and yr=y/, where y is the average income, the relative size o theelites, and the share o the income accruing to the elites. The capital shares thus directly capture .


16/35


or all countries. Thus, cross-country comparisons are meaningul. Thedata set o Ortega and Rodriguez also covers a ar larger proportion othe country-years than those employed by previous authors.67 For exam-

ple, while the data set o Deininger and Squire contains only 11 percento the possible observations, that o Ortega and Rodriguez contains morethan 67 percent.68 Since many observations are still missing, the robust-ness o the results is tested by imputing the missing values.69

control varIables

Diverse economic variables have been shown to aect democracy.Modernization theorists have argued that countries become morelikely to install and sustain democracy as they develop.70 People may

be more willing to demand political rights once their basic needs aresatised. Wealth may also be related to inequality, or instance, throughthe Kuznets curve. Moreover, many scholars argue that economic per-ormances infuence the stability o political regimes;71 or example, adrastic decrease in wealth may destabilize autocracies.72 Growth mayalso infuence inequality, because economic crisis or booms tend to a-ect diverse segments o the population dierently. The structure o theeconomy also aects democracy. In particular, countries that rely heav-

ily on natural resources are less likely to be democratic, notably becausethe elites are more vulnerable to taxation.73 Additionally, the revenuesemanating rom natural resources are usually controlled by the state,which can use them to prevent democratization. The analysis thus in-cludes gdp per capita, gdp per capita growth, and a dummy variable orlarge oil exporters.74

In addition, the social and cultural context can infuence democ-racy. Thus, or example, Islam is thought to be particularly harmul,and Protestantism helpul.75 Religion may also infuence the tolerance

o the population toward inequality.76 Included are variables measur-ing the percentages o the population that are Muslim, Catholic, andProtestant. Moreover, some scholars suggest that divided societies are

67 Ortega and Rodriguez 2006.68 Deininger and Squire 1996; Ortega and Rodriguez 2006.69 The multiple imputation model is described in the appendix.70 See, or example, Lipset 1959.71 See, or example, Gasiorowski 1995.72 See Haggard and Kauman 1995.73

See Boix 2003; Ross 2001.74 It takes the value one i the average ratio o uel exports to total exports in 198486 is greaterthan 50 percent, and zero otherwise.

75 See Huntington 1991; Midlarsky 1998.76 See Milanovic, Gradstein, and Ying 2001.


17/35

604 worldpolItIcs

less likely to establish and maintain democratic institutions.77 For ex-ample, an incumbent may be less willing to leave oce i his or heropponent belongs to another ethnic or religious group. Measures o

ethnic and religious ractionalization are thus added to the analysis.These indicate the probability that two individuals selected randomlyare rom dierent ethnic or religious groups.

Further, many political actors aect the regime type. A dummy vari-able is included here or ormer British colonies, which are said to haveinherited institutions particularly conducive to democracy.78 FollowingPrzeworski et al., another dummy variable is added or countries thatdid not exist in 1945.79 This captures, or example, the possibility thatdemocracies established by colonizers in new countries are less stable

because these countries lack the prerequisites or democracy.Moreover, I control or the number o transitions rom democracy to

dictatorship that a country has experienced. Previous studies generallynd that this variable increases the likelihood o transition.80 In act,countries that have been victims o many coups in the past are morelikely to experience coups in uture.81 Also, the interaction between theelites and the population is likely to be dierent in autocracies thathave experienced democracy than in those that did not. For example,

the population may nd it easier to mobilize and challenge the regimein the ormer.Some scholars also argue that presidential democracies are more

ragile than other types o democracies.82 Presidential elections are o-ten described as zero-sum games, where the losers have little incen-tive to accept electoral results. Because the president remains in oceor a xed term, it is dicult to depose an incompetent or unpopulargovernment without destabilizing the regime.83 Included is a dummyvariable indicating whether the regime is presidential. Finally, it has

recently been suggested that international political context aects thelikelihood o a country being democratic.84 The analysis thus controlsor the proportion o democracies in the world. All the control variablesare taken rom the extended data set o Przeworski et al.85

77 See, or example, Dahl 2000.78 See, or example, La Porta et al. 1998.79 See Przeworski et al. 2000.80 See, or example, Przeworski et al. 2000; Epstein et al. 2006.81

Londregan and Poole 1990.82 See, or example, Linz 1990.83 For a counterargument, see Cheibub 2006.84 See, or example, Gleditsch and Ward 2006.85 Przeworski et al. 2000.


18/35


descrIptIve statIstIcs

Beore undertaking the statistical analysis, I take a preliminary look at

the data. Democratic countries are much more equal than autocracies.86

This observation is consistent with several arguments linking inequal-ity and democracy: inequality may harm democratization, it may in-hibit consolidation o already established democracies, or it may itselbe aected by the political regime. It is argued here that the observedassociation can be accounted or mainly by the second o these pos-sibilities.

Table 2 presents the probabilities o regime change at low, interme-diate, and high capital share values, using the nonimputed data set.87 In

each regime type the observations are divided into three groups (tiers)containing the same number o country-years. The probability o tran-sition and the number o transitions within each group are reported.

The rst row looks at the probability o dictatorships transitioningtoward democracy. According to the rst group o authors (or exam-ple, Boix), greater inequalitythat is, larger capital sharesshould beassociated with smaller probabilities o democratization.88 As shown bythe rst row o Table 2, this hypothesis is not supported by the data.

I anything, inequality osters democratization. Very unequal countriesare more than three times as likely as very equal ones to democratize.Many authors view communist and noncommunist dictatorships as

dierent regime types.89 The ormer are created by the masses aterrevolutions to redistribute assets, while the latter are established andmaintained by the elites to prevent redistribution. Communist coun-tries are thus excluded in row 2. The results are unchanged. Thereore,the hypothesis that transitions toward democracy are more likely inegalitarian societies nds no evidence.

The hypothesis o Acemoglu and Robinson is not supported either.90As shown in row 1, there is no clear nonmonotonic relationship. In-equality is always associated with a higher probability o democratiza-tion. These results are unchanged when, in row 2, communist countriesare omitted. Thus, there is no evidence or either o the two main theo-ries linking inequality to democratization.

86 The mean capital share is 0.6765 in dictatorships and 0.6093 in democracies.87

The results reported below are unchanged when the dataset that includes imputed observationsis used instead (not reported).88 Boix 2003.89 See, or example, Boix 2003; Acemoglu and Robinson 2006.90 Acemoglu and Robinson 2006.


19/35

606 worldpolItIcs

Inequality, however, has a strong positive eect on the probability oa democracy backsliding to dictatorship. As illustrated in row 3 o Ta-ble 2, democracies that have high capital shares are much more likely tobreak down. Those with capital shares in the high tier are ve times aslikely to collapse as are those in the middle tier. Moreover, among thelow tier, only one democracyNiger in 1996has ever broken down.

One potential problem is that the eect may be driven by rich devel-oped democracies that are both stable and equal. Row 4 thus excludesWestern countries and Japan. The results are unchanged. Democracieswith high capital shares are still more than ve times as likely to breakdown as those with low capital shares. Thereore, consistent with theargument presented here, inequality does harm consolidation.

empIrIcal results

This section employs dynamic probit models to show that inequalityharms consolidation but does not aect democratization. This methodhas been used by other authors studying regime transitions.91 It esti-mates the probability o countries with a certain regime (in the currentperiod) transitioning to a new regime in the next period. This modelenables one to distinguish between the eects o dierent independentvariables on democratization and consolidation and to obtain dierentestimates or each transition pattern. For example, inequality is allowed

to have dierent eects on transitions rom democracy to dictatorshipand rom dictatorship to democracy. Tables 3 and 4 present respectively

91 See, or example, Dunning 2008; Przeworski et al. 2000; Boix and Stokes 2003; Boix 2003.

Capital Share Tiers#

Transitions N Low Middle High

1. dict. to dem. .017 (10) .024 (14) .0524 (31) 55 17612.a dict. to dem. .0181 (10) .0236 (13) .0524 (30) 53 1678

3. dem. to dict. .0018 (1) .009 (5) .0383 (22) 28 16884.b dem. to dict. .0097 (3) .026 (8) .0503 (16) 27 934

aExcludes communist countries.bExcludes Western countries and Japan. Number o transitions in parentheses.

table 2probabIlItyofregIme transItIonpercapItal share tIers


20/35


the impact o each independent variable on the probability o democ-ratization and on the likelihood o stable democracy.92

Table 3 reports the impact o capital shares on the probability o

transition rom dictatorship to democracy in both linear and nonlinearmodels. Positive coecients indicate that the associated independentvariables increase the probability o transition to democracy. Model 1tests the hypothesis o a negative monotonic relationship, advanced no-tably by Boix.93 It shows that lower capital shares are actually associatedwith smaller probability o democratization, though the relationship isnot signicant. Thus, contrary to what has been argued, inequality doesnot harm democratization but instead has a weak positive eect.

Model 2 o Table 3 estimates the nonlinear model by adding capital

share squared. The predictions o Acemoglu and Robinson, accordingto which the relationship is inverted U-shaped, would be supported ithe coecient on capital share is positive and the one on capital sharesquared is negative.94 As shown in model 2, both coecients turn outto have the wrong sign, although none is statistically signicant. Fig-ure 1 shows the predicted probabilities o democratization estimated inmodel 2, when other variables are at their mean or median. The rela-tionship between inequality and democratization is clearly not inverted

U-shaped. I anything, it is U-shaped.95

Next, models 3 and 4 rerun, respectively, models 1 and 2 with re-gion and decade dummy variables.96 Controlling or these variables isimportant, since inequality and democracy vary substantially acrossregions and time periods. Latin America is the most unequal region,ollowed by sub-Saharan Arica and the Middle East. Asian, Western,and East European countries are much more equal. The omission oregions and time periods could thus bias the results. For example, LatinAmerican countries are unequal and oten politically unstable. By ail-

ing to control or regions, we may simply capture the eect o beingrom Latin America. Another potential problem is that, as discussed

92 The results reported in Tables 3 and 4 are estimated by the same regressions. Dynamic probitmodels estimate the likelihood o transitions to and rom democracy at the same time. The results arereported separately to acilitate interpretation. The original regression outputs are available rom theauthor upon request.

93 Boix 2003.94 Acemoglu and Robinson 2006.95 However, there are ew observations at both extremities o the distribution plotted in Figure

1, such that the relationship is largely fat or weakly increasing. In act, more than 85 percent o

the observations have capital shares between 0.45 and 0.8. The generalized additive model (gam)estimated below also shows that the relationship is not U-shaped.

96 The region dummies are Western countries (Western Europe, North America, New Zealand,and Australia), Asia, sub-Saharan Arica, Eastern Europe, Latin America, and the Middle East. Asiaand the 1990s dummy are excluded during the estimation.


21/35

table3

dynamIcprobItana

lysIsoftheprobabIlItyo

ftransItIonfromd

Ictat

orshIptodemocracya

Baseline

Dec.&

Reg.

Excl.Comm

.

Imputed

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Cap.s

hare

.855

(.6584)

4.0

347

(6.4

495)

.9217

(.7248)

3.5

631

(6.7

459)

.5537

(.7006)

6.8

561

(6.5

543)

.5289

(.6024)

5.1

235

(6.0

093)

Cap.s

hare

squared

3.6

36

(4.7

785)

3.3

445

(4.9

848)

5.4

94

(4.8

456)

4.2

201

(4.7

446)

Loggdppc

.1

711

(.2454)

.1

538

(.2466)

.5

555

(.3023)*

.5

672

(.3054)*

.2

297

(.2573)

.2

217

(.2588)

.0969

(.2106)

.1115

(.2118)

Growth

.0

143

(.0052)***

.0

143

(.0053)***

.0

125

(.0057)**

.0

124

(.0057)**

.0

119

(.0064)*

.0

116

(.0065)*

.0

116

(.0043)***

.0

116

(.0044)***

Oil

.4

174

(.3202)

.4

701

(.3299)

.2

177

(.3457)

.2

718

(.3563)

.3

037

(.3204)

.3

804

(.3311)

.4

313

(.308)

.4

905

(.3168)

Muslim

.0

033

(.0032)

.0

029

(.0032)

.0

028

(.0036)

.0

027

(.0037)

.0

076

(.0035)**

.0

073

(.0035)**

.0

034

(.0029)

.0

031

(.0029)

Catholic

.0

004

(.003)

3.4

9e06

(.003)

.0

05

(.0049)

.0

05

(.005)

.0

04

(.0032)

.0

037

(.0032)

.0

007

(.0027)

.0

003

(.0027)

Protestant

.0016

(.006)

.0014

(.0059)

.0086

(.0068)

.0086

(.0068)

.0

016

(.0065)

.0

02

(.0065)

.0

001

(.0055)

.0

002

(.0054)

Ethnicfrac

.

.0

007

(.0031)

.0

006

(.0031)

.0

025

(.0035)

.0

025

(.0035)

.0

002

(.0032)

.0

002

(.0033)

.0

009

(.0028)

.0

007

(.0028)

Religiousfrac.

.0058

(.0053)

.0053

(.0053)

.0103

(.006)*

.01

(.006)*

.0113

(.0059)*

.0106

(.006)*

.0054

(.0042)

.0048

(.0043)

Britishcolony

.0208

(.1733)

.0257

(.1735)

.0

572

(.1931)

.0

592

(.1934)

.0143

(.1738)

.024

(.1739)

.0564

(.1609)

.0592

(.1611)


22/35

table3

dynamIcprobItana

lysIsoftheprobabIlItyo

ftransItIonfromd

Ictat

orshIptodemocracya

Newcountry

.3

377

(.2006)*

.3

228

(.2006)

.1

591

(.2561)

.1

321

(.2577)

.2

649

(.213)

.2

555

(.2109)

.2

297

(.1753)

.2

128

(.1758)

Pasttransitions

.2273

(.0724)***

.2174

(.0734)***

.1941

(.0759)**

.1851

(.0769)**

.2497

(.074)***

.2457

(.0749)***

.2605

(.0608)***

.2523

(.0615)***

%dem.inworld

2.5

164

(.6223)***

2.4

358

(.6325)***

1.6

99

(1.6

856)

1.6

553

(1.6

855)

2.6

557

(.6566)***

2.5

471

(.6673)***

2.3

359

(.5656)***

2.2

659

(.5721)***

Log-Lik.

310.2

1

309.8

8

302.4

7

302.0

2

297.6

8

297

384.5

383.7

9

N

3439

3439

3439

3439

3355

3355

4029

4029

***p

inequality and democracy ：why inequality harms consolidation but does not affect democratization

Documents