1

Inequality and Regime Change:

Democratic Transitions and the Stability of Democratic Rule

Stephan Haggard, University of California, San Diego

and

Robert R. Kaufman, Rutgers University

Abstract

Recent work by Carles Boix and Daron Acemoglu and James Robinson has focused on the role of

inequality and distributive conflict in transitions to and from democratic rule. We assess these claims through

causal process observation, using an original qualitative data set on democratic transitions and reversions

during the third wave period from 1980 to 2000. We show that the incidence of distributive conflict, a key

causal mechanism in these theories, is in fact quite low. It is present in just over half of all transition cases

and less than a third of all reversions.

Our thanks to Carles Boix, T.J. Cheng, Ruth Collier, Javier Corrales, Ellen Commisso, Sharon Crasnow, Anna

Grzymala-Busse, Allan Hicken, Christian Houle, James Long, Irfan Noorudin, Grigore Pop-Eleches, Andrew

Schrank, and Nic VanDewalle for comments on earlier drafts, including on the construction of the dataset.

We also thank Vincent Greco, Terence Teo, and Steve Weymouth for research assistance.

2

3

Are inequality and distributive conflicts a driving force in the transition to democratic rule? Are

unequal democracies more likely to revert to authoritarianism? These questions have a long pedigree in in the

analysis of the transition to democratic rule in Europe (Lipset 1960; Marshall 1965; Moore 1966), and have

been raised again in newer comparative historical work on democratization (Rueschemeyer, Stephens and

Stephens 1992; Collier 1999). More recently, however, an influential line of theory has attempted to ground

the politics of inequality on rationalist assumptions about citizens preferences over institutions (Acemoglu

and Robinson 2000, 2001, 2006; Boix 2003, 2008; Przeworski 2009). These distributive conflict approaches

conceptualize authoritarian rule as an institutional means through which unequal class or group relations are

sustained by limiting the franchise and the ability of social groups to organize. The rise and fall of democratic

rule thus reflect deeper conflicts between elites and masses over the distribution of wealth and income.

Despite its logic, there are a number of theoretical and empirical reasons to question the expectations

of these new distributive conflict models. Socio-economic inequality plays a central role in these models, but

has cross-cutting effects. The more unequal a society, the greater the incentives for disadvantaged groups to

press for more open and competitive politics. Yet the wider the income disparities in society, the more elites

have to fear from the transition to democratic rule and the greater the incentives to repress challenges from

below. Given this potential indeterminacy, theoretical models have hinged on a variety of other parameters,

such as the cost of repression or the mobility of assets. Even with these refinements, empirical attempts to

demonstrate the relationship between inequality and regime type have yielded only mixed results. In cross-

section, there is a relationship between income distribution and the level of democracy: ceteris paribus, more

equal societies are more democratic. But the causal relationship between inequality and either transitions to

democratic rule or its stability are much less robust.

We focus on transitions to democratic rule and reversions from it during the third wave period,

from 1980-2000. We find surprisingly limited evidence in support of the general claim that redistributive

conflicts between elites and masses account for either transitions to democracy or reversions from it. Using

an extremely generous definition of distributive conflict transitions, we find some that conform very

loosely to the causal mechanisms specified in the model. Yet we also find a number of alternative pathways.

4

In a number of cases, external actors, typically donors, pressed political reforms on authoritarian regimes as a

condition for continued support and these pressures seemed decisive. Moreover, there are also domestic

political conditions under which incumbents may choose to relinquish office in the absence of strong

challenges from below. Elite incumbents may be challenged by elite out-groups or defectors from the ruling

coalition who believe that their chances for political and material gain are better served by widening the

political space than they are by the authoritarian status quo. Alternatively, elite incumbents may rationally

choose to cede power in the absence of mass pressure because they believe they can control the design of

institutions in the new democratic system to protect both their incumbency and their material interests at

lower cost than through outright repression.

The majority of reversions similarly occurred in ways that were not characterized by the elite-mass

dynamics postulated in the theory. In several cases, incumbent democratic government were overthrown not

by elites, but by authoritarian populist leaders promising more extensive redistribution. Even more commonly,

reversions were driven by conflicts that either cut across class lines or stemmed from purely intra-elite

conflicts in which factions of the military staged coups against incumbent office-holders.

In contrast to purely quantitative tests of the relation between inequality and regime change, we

combine a qualitative data set of causal process observationsa within case analysis--with some simple

statistical analyses of the entire population of regime changes. The purpose of this approach is to test for the

presence or absence of the causal mechanisms stipulated by the theoretical models: the extent to which

transitions and reversions were not only associated with differing levels of inequality but were in fact triggered

by distributive conflicts between elites and masses.

Our approach differs from other forms of causal process observation (Collier, Brady and Searight

2010), however, in that it seeks to explicitly complement quantitative designs by examining all discrete

country-years that have been coded as transitions or reversions in two prominent data sets: Polity IV and the

dichotomous coding scheme developed by Przeworski et al (2000) and extended by Cheibub and Ghandi

(2004). Critics of medium-N designs have noted a tradeoff between depth and breadth. The tracing of

causal processes in such designs lacks the detail available in individual case studies or smaller-N designs, while

5

lacking the precision of well-specified econometric models. Yet we argue that these problems are endemic to

both small-n and large-n methods, and that that there are advantages to choosing a design that avoids both

the problems of case selection in small-N designs (Gerring 2007a, 86-150) and the lack of attention to causal

mechanisms and institutional detail that often accompanies reduced-form econometric specifications.

.Indeed, one of the interesting findings of our causal process observations is to call into question the validity

of the very codings of regime type that have been deployed in most cross-national panel designs.

We are thus motivated by methodological as well as substantive concerns. Despite the hortatory calls

for combining quantitative and qualitative research, there are to our knowledge few designs in the

democratization literature that systematically evaluate whether the causal processes stipulated in the

underlying formal modelswhich typically involve complex sequences of decisions and actions on the part

of incumbents and oppositionsare in fact present in the cases. We thus seek to demonstrate the value of a

particular mixed-method design as a way both of testing formal models and assessing the results of cross-

national panel designs. We believe this method can be used much more systematically than it currently is,

particularly when testing theories of relatively rare events such as democratic transitions and reversions, civil

wars, genocides and famines where the number of country-years in the panel may be large but the number of

cases to be explained is limited.1

We begin in the first section by reviewing distributive conflict models of regime change, focusing on

the contributions by Boix (2003, 2008), Acemoglu and Robinson (2000, 2001, 2003, 2006) and Przeworski

(2009). Section 2 discusses methodological issues. The remainder of the paper is structured around a

consideration of transitions to democracy (section 3) and reversions to authoritarian rule (section 4). Our

causal process observations show not only that transitions occur across cases with very different levels of

inequalityas the null findings in econometric models already attest--but that a large number of democratic

transitions and reversions occur in the absence of significant redistributive conflict altogether.

The returns from this exercise are several. First, the findings cast doubt on some of the core causal

mechanisms at work in the underlying model, including the presumed relationship between inequality and

1 . In the civil war literature, see Fearon and Laitin n.d and Sambanis 2003.

6

particular types of elite and mass behavior. Second, the exercise raises important questions about the validity

of reduced-form panel designs, including with respect to the coding of regime type itself. Finally, and more

positively, it suggests a fruitful method for combining quantitative and qualitative methods that focuses

attention on alternative transition paths rather than the partial-equilibrium treatment effects of favored

variables.

I. Theory

Adam Przeworski (2009, 291) poses the puzzle of democratic transitions in the clearest terms: Why

would people who monopolize political power ever decide to put their interests or values at risk by sharing it

with others? Specifically, why would those who hold political rights in the form of suffrage decide to extend

these rights to anyone else? The seminal work of Meltzer and Richard (1981, hereafter M-R) provides the

point of departure for all current distributive conflict models of regime change.2 The M-R model posits that

the distribution of productivity and income is skewed to the right, with most citizens falling at the lower and

middle range of the distribution, and a smaller tail constituting the rich. Those with incomes above the mean

lose from redistribution and should thus oppose it; those below the mean gain and therefore should favor it.

In more unequal countries, the poor have more to gain from redistribution; those countries should therefore

have more redistribution. In democracies where parties are assumed to appeal to the median voter -- the

wider the divergence between the median and mean income, the more is to be gained from redistribution, the

higher the level of tax-cum-transfer the median voter will prefer, and the larger the size of government. In

the distributive conflict theories of regime change, most notably in the work of Boix (2003) and Acemoglu

and Robinson (2000, 2001, 2007), these expectations are modified and expanded to endogenize the very

existence of democratic governments.

Carles Boixs (2003) Democracy and Redistribution defines a right-wing authoritarian regime as one in

which the political exclusion of the poor sustains existing economic inequalities. According to Boix (2003: 37)

a more unequal distribution of wealth increases the redistributive demands of the population[However]

2 See also Romer (1975) and Roberts (1977).

7

As the potential level of transfers becomes larger, the authoritarian inclinations of the wealthy increase and

the probabilities of democratization and democratic stability decline steadily. The translation of these

demands into a change in institutions hinges, however, on the balance of power between the wealthy and the

poor. Boix offers an informational model in which regime changes are triggered by exogenous shocks that

weaken the elite or reveal its weakness (28-30). A nececssary (although not sufficient) mechanism driving

regime change is pressure from below. As the least well off overcome their collective action problems, that

is, as they mobilize and organize in unions and political parties, the repression cost incurred by the wealthy

rise, forcing elites to make institutional compromises (Boix 2003: 13).

Boix also emphasizes the role played by capital mobility in mitigating this relationship (see also

Freeman and Quinn 2010). High levels of capital mobility enhance the bargaining power of elites. Fixed

assets, by contrast, limit the options of the wealthy and make them vulnerable to democratic redistribution

and thus more resistant to it. Given the decision of the poor to mobilize, the incentives of upper-class

incumbents to repress is a function of the level of inequality and mobility of assets. Transitions are most likely

when inequality is low, asset mobility is high and elites have less to lose from competitive politics. Elites have

stronger incentives to repress as inequality increases and when assets are fixed.

Although broadly similar in spirit, Daron Acemoglu and James Robinsons (2006) Economic Origins of

Dictatorship and Democracy (hereafter A & R) introduces several innovations. A & R concur with Boix that

regime type is a function of the balance of power between high and low income groups. Although elites

monopolize de jure power, masses potentially wield de facto power through their capacity to mobilize.

Because they constitute the majority, masses can sometimes challenge the system, create significant social

unrest and turbulence, or even pose a serious revolutionary threat (25). High inequality increases the

incentives for authoritarian elites to repress these political demands for redistribution.

To this observation, Acemoglu and Robinson add an important point about credible commitments.

When elites are confronted by mobilization from below, they can make short-run economic concessions to

diffuse the threat. But politically and economically excluded groups are aware that elites can renege on these

8

concessions when pressures from below subside. Since there is a cost to subsequently reversing democracy

once a transition has occurred, democratic institutions provide a means for elites to credibly commit to a

more equal distribution of resources not only in the present but into the future as well.

A & R concur with Boix that although inequality increases the incentive for excluded groups to press

for democracy, it also increases elite incentives to repress. High inequality is inauspicious for democracy.

However A & R argue that democratization is also unlikely to occur in authoritarian governments with low

levels of inequality since the demand for it is also attenuated; despite political restrictions, excluded groups

nonetheless share in societal income. They conclude that the relationship between inequality and democratic

transitions should exhibit an inverted-U pattern, with transitions to democratic rule most likely to occur at

intermediate levels of inequality.

It is important to emphasize, however, that the theory is not simply a structural one but operates

through incentives for collective action on the part of the masses and repression on the part of elites. At

middle levels of inequality, grievances are sufficient to motivate the disenfranchised to mobilize, but not

threatening enough to invite repression and thus preclude democratic class compromises (see also Burkhart

1997; Epstein et. al. 2006).

The arguments of Boix and A & R are cast in terms of democratic transitions, but both extend their

arguments to a consideration of the stability of democratic rule and reversion to autocracy as well. For Boix,

there is a direct linear relation between the degree of inequality and the likelihood of democratic breakdowns.

Higher inequality generates costly redistributive pressures on the part of lower class groups, which in turn

drive elites to seek to reimpose authoritarian rule. Although A & R posit an inverted-U shaped relation

between inequality and democratic transitions, they agree that countries that do manage to democratize at

high levels of inequality do not consolidate because coups are attractive (Acemoglu and Robinson 2006:

38).

2. Method: Causal Process Observations

9

At first glance, these theories appear amenable to relatively straightforward tests. Is the level of

inequality associated with transitions to and from democratic rule or not? As we will see, empirical tests are

complicated by the fact that different measures of inequality are capturing different socio-economic cleavages

and the quality of the data is notoriously poor. We also find that measures of democracy commonly used in

panel designs leave much to be desired as well.

In the following sections on democratic transitions and authoritarian reversions, we review this

literature in more detail. But the problems of testing these theories are not limited to the constraints posed by

the data: they are also related to the reduced-form nature of most cross-national panel designs, which omit

the intervening causal processes by focusing on the relationship between levels of inequality and regime

change. The empirical question is not only whether inequality is associated with regime change, however, but

whether its effects operate through the particular causal mechanisms postulated in the theory. In principle,

multi-stage models can be constructed that work from structural causes through intervening behaviors to

institutional effects (King et al. 1994: 85-87). Some critics of the mechanisms approach have argued that

mechanisms may be nothing more than such chains of intervening variables (Beck 2006, 2010; Gerring

2007b, 2010; Hafner-Burton and Ron 2009).

But causal process observation permits a different type of empirical strategy than the identification of

an average treatment effect in an econometric specification.3 By causal process observation, we mean the

reconstruction not only of a set of antecedent conditions but of a subsequent sequence of actor decisions,

ultimately strategic in form, that yield the postulated outcome. In the theoretical work we address here, these

mechanisms run from inequality to the mobilization of distributive grievances by the poor or more

commonly by coalitions of low- and middle-income groups; to elite calculations about the costs of

repressing these challenges or offering political concessions; to the iterated strategic response of the masses to

3 The concept of causal process observation (Collier, Brady and Seawright 2010) grew out of an earlier stream of methodological work on process tracing initiated by Alexander George (George and McKeown 1985; Bennett and George 2005) and subsequently joined by work on the empirical testing of formal models, including through analytic narratives (Bates et. al. 1998). A related strand of work is associated with the mechanism approach to causation (Gerring 2007b, 2010; Faletti and Lynch 2010; Hedstrom and Ylikoski 2010).

10

those elite decisions (see particularly Boix 2003, 27-36 and Acemoglu and Robinson 2006, 181-220 for

explication of the basic models).

The method of causal process observation has several advantages that can enrich the testing of

formal theories. In a quantitative model, the effects of either structural variables, such as inequality, or

behavioral ones such as protest are estimated across a heterogeneous set of cases, some of which transition as

a result of the stipulated causal mechanism and some which do not. The focus on average treatment effects

masks the heterogeneity of transition paths; the variable in question is either significant or not.

By contrast, causal process observations do not ask whether the variable in question is significant,

but whether the transition path conforms with the sequence of moves stipulated in the theoretical model. We

pay particular attention to the key political decisions that result in or constitute regime change. For

democratic transitions, these include decisions on the part of authoritarian leaders to make political

concessions or withdraw altogether. For reversions, these include actions by challengers within or outside of

the government that result in the overthrow of democratic rule.

Selection on the dependent variable is not only justified to conduct these tests, but is a central feature

of the theory-testing approach. The objective is not to test whether inequality or mass mobilization has the

stipulated effect or even whether the particular case in question is accurately predicted, both of which a fully-

specified empirical model can do perfectly well. Rather, each case is coded as to whether the stipulated causal

mechanisms, consisting of a particular sequence of moves, are present. If regime change is not accompanied

by the causal processes stipulated by the model, it would be disconfirmatory of the theory even if inequality was

statistically significant in a reduced-form quantitative analysis.

In contrast to the more common practice of purposeful (Gerring 2006, 2007a, b) or random (Fearon

and Laitin 2011) selection of cases for more intensive, process-tracing analysis, our approach is to select all

transition and reversion cases in the relevant sample period (1980-2000 in this case). The cases that we select

are coded on the basis of the continuous Polity IV metricusing different cutoffs--and the Przeworski et al

(2000) and Cheibub and Gandhi (hereafter, PCG) dichotomous measure.4 The full dataset is appended.

4 See also Cheibub, Gandhi and Vreeland 2010.

11

The method lends itself particularly well to the testing of game-theoretical models while getting

around an important problem in standard quantitative panel designs. In such designs, each country year is

coded as a transition or non-transition year (the dependent variables). The causal covariates are similarly

either contemporaneous or antecedent with some lag structure. But the causal sequences leading to the actor

choices associated with transitions and reversions may be more compressed or extended, and thus not well

captured by the artifact of the country-year coding constraint typical of the panel design. As we will see, many

cases that are coded as transitions prove to be dubious when a more extended but variable temporal context

is taken into account, a point emphasized more generally in the work of Pierson (2004).

In each of the remaining two sections on democratic transitions and reversions, we begin with a

review of the quantitative findings on the relationship between inequality and regime change and then present

both aggregate and select case-study findings from the causal process observations in our dataset. As we will

show, the support for the distributive conflict model of regime change is surprisingly weak, even under highly

generous coding rules; when these rules are tightened, the evidence is weaker still.

3. Transitions to Democratic Rule

A & R (2006) do not present systematic empirical evidence in support of their claims.5 Much of the

book is taken up with a discussion of the underlying intuition of the theory (pp. 1-47; 80-87); and the

presentation of a family of formal models of democratic and non-democratic regimes (pp. 89-172) and regime

change (pp. 173-320). A & R do present scatterplots showing a positive relationship between equality and the

level of democracy across a global sample of countries (pp. 58-61) and provide short case studies of Great

Britain, Argentina, South Africa and Singapore (pp. 1-14). But these correlations and cases are illustrative at

best.

5 Earlier work (2000, 2001) was motivated by experiences in nineteenth century Europe and early twentieth century Latin America. But in those articles, as well as in the later book, the formal theory is cast in general terms, without specifying scope conditions that might apply to third wave transitions. In the book, moreover, the illustrations from South Africa and Singapore rely on much more recent developments.

12

In his analysis of democratic transitions over the very long run (1850-1980), Boix (2003) finds that

the distribution of land, proxied by the share of family farms, has an effect on the transition to democratic

rule. More unequal societies are both less likely to transit and less stable when they do (pp. 90-97). Boix also

explores a highly uneven panel of countries for the 1950-90 period (only 587 observations), including

developed ones (pp. 71-88). Using a Gini index as his measure of inequality, Boix finds some evidence that

the probability of a democratic transition is decreasing in the level of inequality, but the findings are not

altogether robust (see for example Model 2A, p. 79).

More recently, a number of other quantitative studies have taken up the challenge raised by Boix and

A&R, but with very mixed results. Like Boix, Ansell and Samuels (2010) consider both long-historical and

postwar samples (1850-1993; 1955-2004). They find that land concentration makes democratization less

likely, but that increases in income inequality make it more likely. They argue that increasing income inequality

reflects not only an elite-mass dynamic but the emergence of a new capitalist class that challenges landed

elites, a dynamic consistent with Boixs (2003, 47-59) and A & Rs (2006, 266-286) extension of their models

into three-class variants.

The limited number of other tests in the literature generally fail to find a relationship between

inequality and democratic transitions. A cross-section design by Dutt and Mitra (2008) finds a relationship

between inequality measured by the Gini coefficient and political instability, but fails to find a relationship

between inequality and transitions to democratic rule, measured either by Polity or Freedom House measures.

Christian Houle (2009) creates a data set using an alternative measure of inequality: capitals share of income

in the manufacturing sector. Using the dichotomous coding scheme developed by Adam Przeworski and

associates (Przeworski et. al. 2000; Cheibub and Gandhi 2004), Houle shows that inequality bears no

systematic relationship to democratic transitions over the 1960-2000 period but is a significant predictor of

reversions to authoritarian rule, a finding to which we return in the next section. In a wide-ranging study of

the determinants of democratization, Teorell (2010, 60) also fails to find a relationship between a Gini

coefficient and democratic transitions.

13

Distributive Conflict and Non-Distributive Conflict Transitions

In sum, the qualitative and quantitative work on inequality and regime change is highly inconclusive

at best, and even more limited for the recent third wave transitions. What can causal process observations

add to this research? On the one hand, showing that distributive conflicts drive transitions might cast doubt

on any null statistical findings and suggest that pursuit of a better specification of the quantitative model or

more appropriate measures for inequality may be warranted. If regime change is not driven by distributive

conflicts in the manner posited by the theory, the finding could be considered disconfirmatory. Even if

inequality were statistically significant in the quantitative analysis, the effects would work through causal channels not

posited by the game-theoretic models. .

To undertake such causal process observations, we need to interpret the underlying causal

mechanisms at work in the theory, which in A and R are contained in a complex family of formal models.

Abstracting from this complexity, several core conditions appear central to the models. First, elites must

confront political-cum-distributive pressure from below, or a clear and present danger of it; in the absence

of such pressures or an imminent threat of them, it is not clear why elites would be motivated to cede power

at all, as Przeworskis trenchant question suggests.6 Second, there must be some evidenceminimally in the

temporal sequence of eventsthat the repression of these challenges appears too costly and that elites make

institutional compromises as a result. We therefore code distributive conflict transitions as ones in which:

The mobilization of redistributive grievances on the part of economically disadvantaged groups or

representatives of such groups (parties, unions, NGOs) posed a threat to the incumbency of ruling

elites;

And the rising costs of repressing these demands appear to have motivated elites to political

compromise or exit in favor of democratic challengers, typically indicated by a clear temporal

sequence (mass mobilization followed by authoritarian withdrawal).

6 A potential threat is inadequate unless clearly specified because it runs the risk of tautology; that authoritarian regimes are always vulnerable to pressure from below. We take up this question in more detail below in the discussion of coding.

14

In coding the cases, we were deliberately permissive. Our coding allows us to take into consideration a

variety of distributive conflicts that may not be captured by any single inequality measure, from urban class

conflicts to ethnic and regional ones. But they must be fought out around distinctive and identifiable

inequalities. Several other features of our coding also give the benefit of the doubt to the theory. The

economically-disadvantaged or the organizations representing them need not be the only ones mobilized in

opposition to the existing regime. And although mass mobilization must partly reflect demands for

redistribution, it can be motivated by other grievances as well. Finally, although we are skeptical about

potential threats in the absence of actual mobilization as we have noted above, we do take such threats into

account where there has been a recent history of mass mobilization demanding democratic reforms.7 In

arriving at our judgments, we have consulted with regional experts or with individuals knowledgeable about

specific cases. In all instances, however, we gave the benefit of the doubt to distributive conflict

interpretations. All cases in which such threats from below did not occur at all or appeared to play only a

marginal causal role are coded as non-distributive transitions. As noted above, these transitions take several

different forms. Most may involve effort on the part of elites to co-opt and control broader segments of the

population, and thus may have some distributive implications ex post. But where these alternative causal

mechanisms are in evidence, democratization does not come in direct response to pressures from below.

Rather, it arises as result of competition with other elites or as a calculated effort to fend off long-term threats

to the dominance of the ruling class as a whole. We return to a more extended discussion of these

alternatives below.

Inequality and the Incidence of Distributive Conflict Transitions

Table 1 shows the distributive and non-distributive transitions, using the definition of transitions in

the data sets constructed by Przeworski et al. (2000) and extended by Cheibub and Ghandi (2004) which we

refer to hereafter as PCG; in the text, we also report tests based on Polity transition codings. The cases are

7 . It is conceivable, of course, that elites might look even farther down the game tree and base their decision on the potential of mass threats relatively far in the future. But preemptive reforms of this sort follow a different logic in which elites weigh the costs of repression against the likelihood that they can control the course of the transition over the long-term.

15

arrayed according to three measures of inequality: Christian Houles (2009) measure of capitals share of

income in the manufacturing sector (capshare), a Gini coefficient from the University of Texas Inequality

Projects Estimated Household Income Inequality (EHII) dataset (2008) and the Vanhanen (2003) measure of

land inequality. We use one standard deviation from the mean to distinguish high and low from

medium levels of inequality. Although transitions occur at all levels of inequality, a somewhat higher

proportion of the distributive transitions occur at medium levels of inequality. However, these differences in

the incidence of cases are statistically significant only for Houles capshare measure (2 =19.08; p = .0001).

More problematic for both Boix and A & R, we see both a substantial share of all distributive conflict

transitions at high levels of inequality (between 20 and 31 percent depending on the measure) and a

substantial incidence of distributive conflict transitions among the high-inequality cases (50 percent in the

land, close to 60 percent in the capshare, and about 70 percent in the Gini). As in Houle and much of the

other quantitative literature, existing measures of inequality seem to have no effect on the likelihood of a

democratic transition. Nor is there is support for even a modified version of the theory that posits a link

between inequality and the type of democratic transition.

16

Table 1: Distributive and Non-Distributive Transitions by Level of Inequality, 1980-2000* Level of

Inequality Inequality Measures

Capital Share of Income in Manufacturing

Sector (capshare) Gini coefficient (Texas Inequality dataset) Share of Family Farms (Vanhanen)

Distributive Non-Distributive Distributive Non-Distributive Distributive Non-Distributive

High ( > 1 std. deviation)

Bolivia (1982) Burundi (1993) Indonesia (1999) Nigeria (1999) Peru (1980) Sierra Leone (1996) Thailand (1992)

Chile (1990) Ghana (1993) Moldova (1996) Sierra Leone (1998) Thailand (1983)

Bolivia (1982) Burundi (1993) Congo (1992) Kenya (1991) Lesotho (1993) Malawi (1994) Nepal (1991) Romania (1990) Sierra Leone (1996) Zambia (1991)

Ghana (1993) Guatemala (1986) Mongolia (1992) Sierra Leone (1998)

Albania (1992) Bolivia (1982) Bulgaria (1990) Latvia (1991) Peru (1980) Romania (1990) Russia (1992) South Africa (1994)

Chile (1990) Guatemala (1986) Honduras (1982) Hungary (1990) Moldova (1996) Mongolia (1992) Nicaragua (1984) Panama (1989)

% of distributive and non-distributive conflict cases

58.3% 41.7% 71.4% 28.6% 50% 50%

Medium

Albania (1992) Argentina (1983) Bangladesh (1991) Benin (1991) Bulgaria (1990) Cote dIvoire (2000) El Salvador (1984) Kenya (1998) Latvia (1991) Lesotho (1993) Madagascar (1993) Malawi (1994) Nepal (1991) The Philippines (1986) Poland (1989) Romania (1990) Russia (1992) South Korea (1988) Sri Lanka (1989) Sudan (1986) Uruguay (1985) Zambia (1991)

Croatia (1991) Guatemala (1986) Hungary (1990) Mexico (2000) Mongolia (1992) Nicaragua (1984) Pakistan (1988) Panama (1989) Senegal (2000) Turkey (1983)

Argentina (1983) Benin (1991) Cote dIvoire (2000) El Salvador (1984) Indonesia (1999) Peru (1980) The Philippines (1986) South Africa (1994) Sri Lanka (1989) Thailand (1992)

Central African Republic (1993) Chile (1990) Honduras (1982) Pakistan (1988) Panama (1989) Senegal (2000) Thailand (1983) Turkey (1983) Uganda (1980)

Argentina (1983) Benin (1991) Burundi (1993) Central African Republic (1993) Congo (1992) Cote dIvoire (2000) El Salvador (1984) Kenya (1998) Malawi (1994) The Philippines (1986) Uruguay (1985) Zambia (1991)

Pakistan (1988) Senegal (2000)

17

% of Distributive and non-distributive conflict cases

68.8% 31.3% 52.6% 47.4% 85.7% 14.3%

Low (< 1 std. deviation

Congo (1992) Niger (1993) Niger (2000) South Africa (1994)

Central African Republic (1993) Honduras (1982) Macedonia (1991) Uganda (1980)

Albania (1992) Bangladesh (1991) Bulgaria (1990) Croatia (1991) Latvia (1991) Madagascar (1993) Nigeria (1999) Poland (1989) Russia (1992) South Korea (1988) Uruguay (1985)

Hungary (1990) Macedonia (1991) Mexico (2000) Moldova (1996) Nicaragua (1984) Taiwan (1996)

Bangladesh (1991) Indonesia (1999) Lesotho (1993) Madagascar (1993) Nepal (1991) Nigeria (1999) Poland (1989) Sierra Leone (1996) South Korea (1988) Sri Lanka (1989) Thailand (1992)

Croatia (1991) Ghana (1993) Macedonia (1991) Mexico (2000) Sierra Leone (1998) Taiwan (1996) Thailand (1983) Turkey (1983) Uganda (1980)

% of Distributive and non-distributive conflict cases

50% 50% 64.7% 35.3% 55.0% 45.0%

Missing Data Mali (1992) Suriname (1988)

Cape Verde (1991) Suriname (1991) Taiwan (1996)

Mali (1992) Niger (1993) Niger (2000) Sudan (1986) Suriname (1988)

Cape Verde (1991) Suriname (1991)

Mali (1992) Niger (1993) Niger (2000) Sudan (1986) Suriname (1988)

Cape Verde (1991) Suriname (1991)

Note: Countries in italics are cases in the PCG dataset but omitted from Houle 2009 because of missing data.

Sources: Capshare: Houle (2009); Gini (EHII 200 2008); Land Vanhanen 2003). Transitions: Przeworski et al (2003) Cheibub and Ghandi (2004); Polity IV Project: Political Regime Characteristics and Transitions, 1800-2002 (2002); by Monty G. Marshall and Keith Jaggars.

18

Table 2 offers additional insight into the causal role of distributive conflict. Columns 2 and 3 divide

the PCG transitions into distributive and non-distributive types; columns 4 and 5 replicate the exercise for

Polity transitions. We also identify the non-overlapping cases. The last column shows the average Polity score

from the time of the transition through either the end of the sample period or until an outright reversion to

authoritarian rule.

In the first place, the information contained in the table raises serious questions about the validity of

the coding of democratic transitions in these two major data sets, and the inferences that have been drawn in

the quantitative work that employs them. As might be expected, there are a number of cases on which the

judgment of the two datasets diverge, in part over questions of timing. But the overlap between them is only

45.5 percent. Moreover, fourteen of the 57 PCG transitions had Polity scores of less than 6 following their

transitions, and another only just achieved that threshold. Overall, the average Polity scores for the two data

sets came to only 5.9.

A closer look raises more substantive questions about the nature of many of the regime changes and

about the redistributive logic postulated by Boix and A&R, even in cases where the two coding schemes

overlap. Insiders and elites repressed opposition and/or exercised disproportionate control over them in Cote

dIvoire, Croatia, Niger, and Thailand (1982). Transitions in Guatemala (1986) and Honduras (1982)

empowered nominally democratic governments that actually intensified repression of social movements with

redistributive objectives, and death squads continued to terrorize the opposition in El Salvador following the

transition in 1984. In at least four cases, the military and/or incumbent elites continued to exercise

disproportionate influence over the allocation of resources following the transition including Ghana under

Rawlings, Kenya under Moi, Malawi, where an insider won the transitional election, and Romania. In all of

these cases, the cases appear to conform more closely to what Levitsky and Way (2010) call competitive

authoritarianism than to democracy. In addition, a number of nominally democratic governments were

unable to gain adequate control over population and territory and reverted within just a few years of the

transition (Sudan, Sierra Leone, Niger, Congo, and Burundi, and Uganda); we return to these weak

19

democracies in more detail in our discussion of reversions, but these cases raise important questions about

whether they should have been coded as transitions in the first place.

Table 2. Distributive and Non-Distributive Transitions

1980-2000

Country/Year PCG Transitions Polity Transitions Polity score

Distributive Non-

Distributive Distributive

Non-Distributive

Albania 1992 X Not a Polity transition 5 Argentina 1983 X X 8 Bangladesh 1991 X X 6 Benin 1991 X X 6 Bolivia 1982 X X 8 Brazil 1985 Not a PCG transition X 7 Bulgaria 1990 X X 8 Burundi 1993 X Not a Polity transition 0 Cape Verde 1991 X Not a Polity transition - Central African Republic 1993

X Not a Polity transition 5

Chile 1989 (Polity)/1990 (PCG)

X X 8

Congo 1992 X Not a Polity transition 5 Cote dIvoire 2000 X Not a Polity transition 4 Croatia 1991 X Not a Polity transition -3 Croatia 2000 Not a PCG transition X 8 El Salvador 1984 X X 6 Estonia 1991 Not a PCG transition X 6 Fiji 1999 Not a PCG transition X 6 Ghana 1993 X Not a Polity transition -1 Guatemala 1986 X Not a Polity transition 3 Guatemala 1996 Not a PCG transition X 8 Honduras 1982 X X 6 Honduras 1989 Not a PCG transition X 6 Hungary 1990 X X 10 Indonesia 1999 X X 6 Kenya 1998 X Not a Polity transition -2 Latvia 1991 X X 8 Lesotho 1993 X X 8 Lithuania 1991 Not a PCG transition X 10 Macedonia 1991 X X 6 Madagascar 1992 (Polity)/1993 (PCG)

X X 9

Malawi 1994 X X 6 Mali 1992 X X 7 Mexico 1997 Not a PCG transition X 6 Mexico 2000 X Not a Polity transition 8

20

Moldova 1993 Not a PCG transition X 7 Moldova 1996 X Not a Polity transition 7 Mongolia 1992 X X 9 Nepal 1991 X Not a Polity transition 5 Nepal 1999 Not a PCG transition X 6 Nicaragua 1984 X Not a Polity transition -1 Nicaragua 1990 Not a PCG transition X 6 Niger 1993 X X 8 Niger 2000 X Not a Polity transition 5 Nigeria 1999 X Not a Polity transition 4 Pakistan 1988 X X 8 Panama 1989 X X 8 Paraguay 1992 Not a PCG transition X 7 Peru 1980 X X 7 The Philippines 1987 (Polity)/1986 (PCG)

X X 1

Poland 1991 (Polity)/1989 (PCG)

X X 5

Romania 1990 X Not a Polity transition 5 Romania 1996 Not a PCG transition X 8 Russia 1992 X Not a Polity transition 5 Russia 2000 Not a PCG transition X 6 Senegal 2000 X X 8 Sierra Leone 1996 X Not a Polity transition 4 Sierra Leone 1998 X Not a Polity transition 0 South Africa 1994 X X 9 South Korea 1988 X X 6 Sri Lanka 1989 X Not a Polity transition 5 Sudan 1986 X X 7 Suriname 1988 X Not a Polity transition - Suriname 1991 X Not a Polity transition - Taiwan 1992 Not a PCG transition X 7 Taiwan 1996 X Not a Polity transition 8 Thailand 1983 X Not a Polity transition 2 Thailand 1992 X X 9 Turkey 1983 X X 7 Uganda 1980 X X 3 Uruguay 1985 X X 9 Yugoslavia 2000 Not a PCG transition X 7 Zambia 1991 X X 6

N/% 35/61.4% 22/38.6% 26/54.2% 22/46.8% 5.9

Note: Polity Scores are the average of all years following the transition until either the end of the series or a reversion.

Sources: Przeworski (2003); Cheibub and Gandhi (2004); Polity IV (2002).

In principle, coding decisions such as these can be corrected through more refined data sets, and

there are a number that seek to accomplish this for specific regions (e.g. Mainwaring et al. 2001 for Latin

21

America). Robustness checks that test against alternative versions of the DV can also correct for random

errors in any single data set. To an important extent, however, the difficulty is inherent in quantitative panel

designs that require country-year codings for complex, temporal processes that involve many discrete

institutional and political changes that may unfold at different speeds. Because we seek to engage the

quantitative analysis that deploys such data, we do not discard or reclassify any of the cases specified as

transitions, but we return to the implications of these coding deficiencies in our discussion of reversions in

the next section.

What about the theoretical expectations with respect to the role of distributive conflict in democratic

transitions? We find that distributive conflict played a causal role in propelling transitions in about 64 percent

of the PCG transition cases, and this falls to 54 percent of the Polity transitions. These are substantial, but

by no means overwhelming shares of the cases.

But even these findings needs to be tempered by the generosity of our coding rules. Pressure from

below did play an unambiguously significanteven primaryrole in a number of middle-income countries

with well-organized and militant urban working class movements, including Poland, Argentina, and South

Korea. In Bolivia and South Africa, social movements of indigenous peoples and the black majority

respectively played a similar role. But the role of mass pressures was much more limited in many other cases

coded as distributive conflict transitions, either because redistributive demands appeared secondary to wider

political and economic disaffection with authoritarian incumbents or because the ability to map from the

income distribution to political behavior in the way postulated by the theory is much more questionable.

Of particular significance in this regard are the codings of many of the African distributive

conflict transitions. During the severe recessions and fiscal crises of the l980s and early 1990s, protests over

back wages or salary adjustments on the part of civil service and state-owned enterprise unions played an

important role in the democratization process. University students also frequently engaged in mass

mobilization against incumbents. But these groups were relatively well-off and represented only a very small

proportion of the population; the bottom of the income distribution is concentrated overwhelmingly in the

rural sector and to a lesser extent among urban informal workers (Bratton and van de Walle 1997).

22

Moreover, the demands of organized urban groups for wages, benefits and prerogatives were probably even

regressive.

Niger provides an example. In 1990 the military strongman, General Ali Saibou, agreed to convene a

national Conference which then assumed de facto power of a transitional government, and held free elections

in March 1993. His decision to yield authority was a response to severe cross-pressures between creditors

who demanded a tough adjustment program and opposition from the Union of Nigerien Workers (USTN)

and the Union of Nigerian Scholars (USN), which opposed these programs. Distributive conflicts played a

major role in this process, and we coded Niger as a distributive conflict case. However, the conflicts

leading to Saibous decision to relinquish power did not engage the poor, in the M-R sense of those falling

below the median income. To the contrary, the USTN membership was drawn almost exclusively from

Nigers 39,000 civil servants, and the student organization represented only about 6 percent of the countrys

school-age population (Gervais 1997: 93). As Gervais (1997: 105) writes about the economic crisis at the

time, the political stakes raised by adjustment policies tended to compromise the benefits of the organized

groups of the modern sector as much as the privileges of the traditional political class. These pressures

drove the transition process, but do not map to the underlying Meltzer-Richard model in which the interests

of the poor or even middle classes are pitted against the rich.

In addition to Niger, similar questions can be raised about the nature of mass protests in Mali,

Malawi, Benin, the Republic of the Congo, Lesotho, Cote dIvoire, and Kenya, all of which were coded as

distributive conflict cases. Protest was as in Niger typically limited to civil servants, students, and the

Church and/or was repressed or managed by insiders (Cote dIvoire), the military (Lesotho), or by the

incumbent himself (Kenya). Even though all of these casesalong with Niger--meet our coding rules

because of the presence of mobilization from below that has at least some impact on the democratic

concessions made by incumbents, they are clearly borderline in terms of the involvement of those who could

be considered poor. If we were to recode them as non-distributive conflict transitions, the percentages of

distributive transitions would fall to 47.4 percent in the PCG transitions and to 43.8 percent in the Polity.

23

A second group of distributive conflict cases that are subject to question are a number of the post-

communist transitions in which nationalist challenges to Soviet control or influence played a central role.

Mass protests in the Baltic countries focused on independence from the Soviet Union. In Poland, nationalism

as well as economic grievances drove the Solidarity protests. We are more confident in our coding of these

cases, particularly in the Baltics, where broad nationalist movements also targeted the prerogatives of

substantial Russian minorities. Nevertheless, recoding these cases as non-distributive--or at least counter to

the conflict theories considered here--would further lower the proportion of distributive conflicts: to only

43.9 percent of the PCG transitions and a mere 35.4 percent of Polity transitions.

Even with the expansive coding of distributive conflict transitions, we found a large share of cases

36 percent using the PCG measure, 46 percent of Polity casesin which distributive conflict played only a

marginal role in the transition process. Why, in the absence of significant pressure from below, would elites

withdraw or make institutional compromises that risk the redistribution of assets and income not only in the

present but into the indefinite future? As others have argued (Huntington 1991; Collier 1999), there are a

variety of routes from closed political systems to democracy.

One factor that was underplayed in the early democratization literature (ODonnell, Schmitter, and

Whitehead 1986) seems particularly relevant for understanding the third wave transitions is the importance

of international pressure (Whitehead 1996; Drake 1998). A United States military invasion was decisive in

Panama in 1989, and external military and/or political intervention played a major role in Guatemala (1986)

and El Salvador (1984) even though we were constrained to code them as distributive conflict cases because

of the presence of pressures from below. External economic pressures from donors were of determining

significance in three of the low-income African casesCape Verde, the Central African Republic and

Malawiand played a significant role in a number of others (such as Niger) that we coded as distributive

conflict cases.

But even if we set aside the role of international factors, threats from below are by no means the only

domestic pressures that can cause elites to acquiesce to democratizing institutional changes. A common

cause of transition in the non-distributive conflict cases stems from rivalries and competition among the

24

political, military, and economic elites that constitute the authoritarian coalition. The 1983 transition in

Thailand provides an example. The key decision was a compromise between military and party leaders that

allowed key military prerogatives to lapse, including the right of non-elected military officers to hold reserved

legislative seats and cabinet positions and provided the appointed Senate with a veto. The main cleavages

were between the military and the emergent political parties, which were increasingly divided between

Bankok-based business interests and an influx of new patronage politicans from the provinces seeking pork.

By that time, the rural insurgence of the early 1970s had died out and the military had acted to blunt any

leftist activity in the cities. The core issues were questions of military and bureaucratic prerogatives and the

power of the legislature. Accommodation, in fact, was possible because of the fundamentally conservative

nature of the party politicians.

In Moldova -- to take a somewhat different example -- the first competitive election in 1995 resulted

from a stalemate between two leaders of the ruling party: president Snegur and prime minister Luscinci.

Neither could prevent the other from competing for office, both were constrained to accept the risk of a

relatively competitive election and neither had the institutional resources to guarantee a favorable result

(Levitsky and Way 2010). In several competitive authoritarian settings, including Senegal, declining

patronage led to defections from the incumbent ruling party that undercut the regimes capacity to prevail in

elections (Galvan 2001). We also find circumstances in which factions of the military preferred the risks of

democracy to continued support for the government in power. As in Thailand, the transition in Pakistan in

1988 was driven by rivalries within the military establishment that permitted open or tacit alliances with

democratic oppositions.

Elite challenges from outside the regime can also spur democratic transitions (Slater and Smith 2010).

For example, in Mexico the ruling PRI was challenged primarily by business elites and an opposition party

(PAN) that wanted less rather than more redistribution. Popular protest over alleged fraud in local elections

strengthened the bargaining leverage of the PAN in its negotiations with the ruling party, but the political left

played only a marginal role in pushing the regime out of power. In several other cases, concessions to elite

rather than mass challenges appear as important as distributive conflicts pitting rich against poor, including

25

the militarys acquiescence in the assumption of power by parliamentary politicians in Pakistan, the Thai

militarys accommodation of emerging political-economic elites from the Northern part of the country and

arguably the KMTs accommodation of native Taiwanese elites.

Even when elites remain relatively unified, they may still acquiesce toor even leaddemocratic

reform if they believe they can retain leverage over the political process and reduce the long-term threats to

their continuing power at lower cost than under authoritarian rule. Incumbent elites can do this through the

design of political institutions that give them effective vetoes or through the organization of political parties

that exploit other cleavages to dampen distributive conflicts. In Turkey, the military had come to power

stating that its intervention would be of limited duration, and in 1983 it transferred power to a new civilian

government. But it made these concessions after crushing violent left and right factions that had been a

feature of Turkish politics in the late 1970s. In the aftermath of the coup, the government undertook a wide-

ranging purge of the government, arrested all major political leaders, brutally repressed unions, and suspended

all societal organizations. As it gradually reopened the political space in early 1983, moreover, it vetoed most

of the new parties that had formed around established politicians. Although the military elite was surprised

by the victory of the one opposition party it had allowed to function, the military retained veto power over

the new government and there is no indication that threats of mass mobilization influenced the decision to

allow the election results to stand.

We see similar processes of reform in Chile, where outgoing governments built in quite specific

mechanisms through which the military would continue to exercise oversight and supporters of the outgoing

government would be over-represented. (Haggard and Kaufman 1995). These included the establishment of

national security councils with a veto role for the military establishment, constitutional and judicial guarantees

limiting the authority of incoming governments, and the allocation of Senate seats to be filled by the head of

the outgoing regime. In dominant party regimes, elites have also sometimes permitted transitions when they

believe organizational and resource advantages would allow them to compete effectively in a more open

political environment. In Kenya, Mexico and Taiwan, incumbents ceded power gradually while competing

26

aggressively and successfully in the newly-liberalized environment. Several communist transitions, including

Hungary and Mongolia, also fit this pattern.

Note that each of the alternative domestic causal mechanisms we have sketchedintra-elite conflict

or defection and elites changing institutions confident of their post-transition chancesmay in fact be related

precisely to the weakness of immediate threats from below. Where such threats are limited, elites are more

likely to control the transition. Societies in which the poor are not mobilized through programmatic parties,

unions or other organizations may be especially prone to vote buying, patronage, and other forms of

clientelistic control that would guarantee elite control of politics even in nominally democratic settings

(Kitschlet and Wilkenson 2006); we return to this point in more detail in the conclusion.

3. Reversion: the Collapse of Democratic Rule

Although A & R and Boix diverge in their predictions about transitions to democracy, they agree that

when democracies do emerge at high levels of inequality, they are more likely to revert to authoritarian rule.

As A & R put it succinctly, in democracy, the elites are unhappy because of the high degree of redistribution

and, in consequence, may undertake coups against the democratic regime (2006: 222). This view comports

with an earlier generation of theory on bureaucratic authoritarian installations in the Southern Cone

(ODonnell 1973; for critiques, see Collier 1979, Linz and Stepan 1978; Valenzuela 1978): Brazil (1964),

Argentina (1966, and 1976), Chile and Uruguay (both 1973) with extensions to other regions as well (for

example, Im 1987 on Korea).

Unlike the quantitative evidence on transitions, there is somewhat more cross-national evidence to

support the view that inequality is incompatible with democratic stability (Dutt and Mitra 2008, Reenock et al.

2007, Houle 2009). Given these positive findings, we replicated Houles models to orient our causal process

observations. In keeping with our focus on third wave transitions, we divided the full 1960-2000 sample into

two halves on the belief that different international and social dynamics might be operating in the two periods

(Table 3). As noted in the introduction, we expected that in the contemporary period, redistributive pressures

might play a more limited role than it did in the heyday of Cold War politics when great power rivalries often

reinforced cleavages along left-right lines.

27

In the 1960-1979 period, inequality does appear to be associated with an increased likelihood of

authoritarian reversals. These results are consistent with the fact that thirteen of the 20 reversions during this

period were in Latin America, where deep divisions between anti-communist elites, and militant unions and

labor-based parties were amplified by Cold War politics. Cases that fit this model include the well-known

right-wing coups in Brazil (1964), Argentina (1966 and 1976), Chile (1973), and Uruguay (1973). Along with

the European transitions of the late 19th and early 20th century, these experiences were important in

motivating the broader theoretical work of Boix (2003) and the early models developed by Acemoglu and

Robinson (2000, 2001).

But in the tests on the 1980-2000 period, reversions occurred more frequently in very poor societies

that were often divided along non-economic lines and where low-income groups faced severe barriers to

collective action. Eight of the fifteen were in low-income African countries. Reversions in this period were

significantly influenced by low per capita GDP and economic growth, but not as in the earlier period by

inequality. When distributional issues did come into play, they appeared to involve ethnic divisions to a

greater extent than during the earlier period; ethnolinguistic fractionalization (ELF) (as well as the share of

Muslims in the population) reached standard thresholds of significance. As we will show, these findings

comport with our causal process observations, in which we find that opposition to democracy often reflected

broad, cross-class dissatisfaction with the general inability of incumbent democratic governments to deal

effectively with economic crisis.

28

Table 3. Reversions: Developing Countries, 1960-1979 and 1980-2000

1980-2000 1960-1979

Cap. Share -3.215 -7.867***

2.016 2.488

Log GDP 1.936** 1.242

0.965 0.924

Growth 0.065** 0.005

0.026 0.012

Oil 1.230 0.193

0.904 0.589

Muslim -0.012* 0.004

0.007 0.008

Catholic 0.012 -0.001

0.011 0.013

Protestant 0.010 -0.003

0.023 0.016

Ethnic frac. 0.019* -0.013

0.011 0.009

Religious frac. -0.008 0.008

0.021 0.015

British colony 0.476 0.636

0.600 0.728

New country 0.242 0.243

0.825 0.585

Past transitions -0.109 0.011

0.187 0.196

% dem. in world 4.607*** 1.507***

1.695 3.185

Log-Lik. -208.534 -140.433 N 1821 1451

Source: All data are from Houle (2009)

Causal Process Observations: Reversions

In our causal process observations of the reversion cases, we undertake the same exercise that we did

with respect to the transition cases, considering whether there were political pressures for redistribution and

whether it drove the breakdown of democratic rule. First, as in the case of transitions, we identify a category

of elite reaction reversions that conform with the distributive conflict model: elites undermine democracy

29

either by a) seeking to oust incumbent governments that rely on the political support of lower class or

excluded groups and are actively committed to the redistribution of assets and income; or b) seeking to

imposing restraints on political competition in order to prevent coalitions with explicitly redistributive aims

from taking office. In these cases, distributive conflicts are in evidence and elites are acting against

governments, parties and organized social forces that are actively committed to greater redistribution through

the democratic process.

We also identify a second type of distributive conflict reversion in which the incumbent democratic

government is overthrown by authoritarian populist leaders. These types of reversion do not comport fully

with the theory, since populist dictators should not be expected to be able to commit credibly to long-term

redistributionist policies. Still, it is a form of reversion which does clearly involve redistributive conflict and is

given some attention in Boix (2006: 18, 214-219). Whereas in elite reaction reversions, challengers to

democratic rule appeal to elite interests and target the masses for repression, in what we call populist

reversions, authoritarian challengers appeal to the masses and target the elite.

Finally, non-distributive reversions are unambiguously instances of the null hypothesis. In some

cases, as noted in the introduction, support for a reversion cuts across class lines: authoritarian challengers

exploit wide disaffection with the performance of democratic incumbents and invoke broad issues that cut

across distributive cleavages, such as economic performance, corruption, and political scandals. In some

cases, reversions may also be caused by purely intra-elite conflicts. The militaryor factions within itmight

stage a coup against incumbent office-holders or competing economic elites might mobilize military, militia

or other armed forces against democratic rule.

Table 4 reports the incidence of distributive conflict and non-distributive conflict reversions by level

of inequality; the distributive conflict column aggregates both elite and populist reversions. Table 5 shows the

types of reversion, Polity scores of the deposed regimes, and economic circumstances surrounding the

change. Although there is little overlap between the ranking of cases on the three measures of inequality,

reversions do cluster at the middle and high levels of inequality; relatively few took place at the lowest levels.

However, we find only a minority of cases that conform with the populist conflict model, and an even smaller

30

minority that conform with the elite model of reversion. In the dataset used to replicate Houles regressions

for the 1980-2000 period, only two casesBolivia and Burundiare coded as elite driven, and two others

(Ghana 1981 and Ecuador 2000) are populist. In the full PCG data set displayed in Table 6including cases

omitted by Houle because of lack of data--28 percent of the cases are elite reversions, another 17 percent

were populist, and the majority (55 percent) were non-distributive.

Of those cases coded as reversions using the standard Polity measure (not shown here), only five of

the fifteen cases were coded as elite-driven cases conforming with the distributive conflict model: Armenia,

Fiji (twice, in 1987 and 2000), Turkey, and Zambia. Four other cases were populist (Ghana, Ecuador, Haiti,

and Suriname (1980). However, the majority of reversions in both democracy data sets were ones in which

distributive conflicts of the sort postulated by the theory play little if any role in the collapse of democratic

rule.

31

Table 4. Elite and Non-Elite Reversions by Level of Inequality, 1980-2000*

Level of Inequality

Inequality Measures

Capital Share of Income in

Manufacturing Sector (capshare) Gini coefficient (Texas Inequality

dataset) Share of Family Farms (Vanhanen)

Distributive

Elite/Populist Non-Distributive

Distributive Elite/Populist

Non-Distributive Distributive

Elite/Populist Non-Distributive

High ( > 1 std. deviation)

Bolivia(1980) E Burundi(1996)E Ghana (1981) P

Nigeria (1983) Peru (1990) Sierra Leone (1997) ] Thailand (1991)

Bolivia(1980)E Burundi(1996)E Ghana(1982)E

Congo (1997) Guatemala (1982) [ Sierra Leone (1997)

Bolivia (1980)E

Guatemala (1982) Peru (1990)

Medium Turkey(1980)E Ecuador(2000)P

Guatemala (1982)] Pakistan (1999) Sudan (1989)

Turkey(1980)E Ecuador(2000)P

Pakistan (1999)] Peru (1990) Thailand (1991) Uganda (1985)

Burundi(1996)E Ecuador(2000)P

Congo (1997) Pakistan (1999)

Low (< 1 std. deviation)

Congo (1997) Niger (1996) Uganda (1985)

Nigeria (1983)

Turkey (1980)E Ghana(1981)P

Nigeria (1983) Sierra Leone (1997)] Thailand (1991) Uganda (1985)

Missing Data

Suriname (1980) [P

Comoros (1995)]Suriname (1990)

Suriname (1980) [P]

Comoros (1995)] Niger (1996) Sudan (1989)] Suriname (1990)]

Suriname (1980) [P]

Comoros (1995) Niger (1996) Sudan (1989) Suriname (1990)

Sources: Capshare: Houle (2009); Gini (EHII 200 2008); Land Vanhanen 2003). Transitions: Przeworski et al (2003) Cheibub and Ghandi (2004);

Polity IV Project: Political Regime Characteristics and Transitions, 1800-2002 (2002); by Monty G. Marshall and Keith Jaggars.

Notes. E, elite reversion; P, populist reversion. Countries in italics are cases in the PCG dataset but omitted from Houle 2009 because of missing data.

32

Table 5. Distributive and Non-Distributive Reversions,

Polity Scores, Prior Coups, per capita GDP and GDP Growth, 1980-2000 Country/Year Distributive

Reversions Non- Distributive Reversions

Polity Score

Prior Coups GDP/ capita ($)

Growth

Armenia (1995)* E 7 0 497 5.9

Bolivia (1980) E -4 3 1070 -1.4 Burundi (1996) E 0 2 113 -8.0 Comoros (1995) X 4 1 386 3.6 Congo (1997) X 5 0 104 -5.6 Ecuador (2000) P 9 0 1295 2.8 Fiji (1987)* E 6 0 2100 -1.7 Fiji (2000)* E na Gambia (1994) X 8 1 321 0.2

Ghana (1981) P 6 4 224 -3.5 Guatemala (1982) X -5 1 1556 -3.5 Haiti (1999)* P 7 1 428 2.7 Honduras (1985)* X 6 2 1041 4.2

Niger (1996) X 8 1 168 3.4 Nigeria (1983) X 7 2 319 -5.3 Pakistan (1999) X 7 0 526 3.7 Peru (1990) X 7 0 1657 -5.1 Sierra Leone (1997) X 4 3 168 -16.7 Sudan (1989) X 7 2 282 8.9 Suriname (1980) P NA NA 2536 -5.3 Suriname (1990) X na Thailand (1991) X 3 1 1500 8.6 Turkey (1980) E 9 1 2427 -2.4 Uganda (1985) X 3 2 170 -3.3 Zambia (1996)* E 6 0 322 6.9

% or average 28%/17% 55% 5 1.23 835 -0.5 Notes and sources. E, elite reversion; P, populist reversion. Polity scores are the countrys score the year preceding the reversion (Marshall, Gurr & Jaggers 2010). Prior coups is the number of coups (excluding attempted coups or plots) in the ten years prior to the reversion (McGowan 2007, Marshall and Marshall 2010), GDP/capita and Growth refer to the values of these variables in the year of the reversion (World Bank 2010). Countries in italics are in the PCG dataset but omitted from Houle 2009 and the foregoing regressions because of lack of data. (2) Countries marked with an asterisk are Polity cases but not in PCG.

To elaborate the implications of these findings, we focus on the high-inequality cases as using

Houles capshare measure, the measure we used in the regressions above. According to the theory, these

would be the ones most likely to revert as a result of elite reactions to distributive demands from below.

Given the low correlation between measures of inequality, alternative measures would show a somewhat

33

different set of high-inequality cases. However, as can be seen from Table 4 no measure of inequality

generates an incidence of reversions that conforms with theoretical expectations; selection of different cases

from the set of causal process observations would yield similar results.

Distributive Conflict I: Elite Reactions in Bolivia and Burundi

Bolivia and Burundi are the only high inequality cases to revert to authoritarian rule through the

causal process stipulated by the theory. In Bolivia, a right-wing military faction led by General Luis Garcia

Meza deposed acting President Lidia Gueiler on July 17, 1980 following the victory of leftist Hernan Siles in

an election held earlier that year. The coup occurred in the context of severe, ongoing conflicts between

militant miners unions and more conservative political and economic forces after the breakdown of the long-

standing Banzer dictatorship in 1978. The Meza dictatorship was in turn ousted only two years later by

working-class protests that forced new elections. Significantly, severe distributive conflicts continued to

threaten the stability of the new democratic regime and ended only in 1985, when the elected government

severely repressed union opposition and implemented an aggressive structural adjustment program.

In Burundi, inequality is by no means correctly captured by the capshare measure; much more

significant are the deep ethnic cleavages that divide the country. A Tutsi minority (about 15 percent of the

population) had long dominated the military, civil service, and the economy; Hutus constituted a large and

clearly less well-off majority, producing a highly fraught political environment that periodically erupted in

violence (Lemarchand 1996). The deposed democratic government was led by moderate Hutu politicians but

was extremely fragile. Melchior Ndadaye, the first democratic president in our sample period, died in an

unsuccessful coup attempt in 1994 and his successor, Cyprien Ntaryamira, was killed in a suspicious plane

crash in the same year. After a massacre of over three hundred Tutsis by radical Hutu rebels in 1996, a

military coup by Pierre Beyoya, who had previously held power from 1987 to 1992, restored the Tutsis to

power.

Although this account conforms with the distributive conflict model, it must be qualified by the fact

that ongoing ethnic conflicts made it difficult to establish a political equilibrium of any kind. Between 1966

34

and 1996, the country experienced no fewer than 11 coups and attempted coups (McGowan 2007) and the

elected Hutu government was insecure from its inception. As with a number of other reversions, the coding

of a prior democratic transition--in this case in 1993is highly dubious.

Distributive Conflict II: Populist Reversion in Ghana

Jerry Rawlings coup in Ghana constitutes a clear example of a populist reversion, although once in

office his military government later shifted sharply to the right. In 1981, Rawlings overthrew the feckless

constitutional government of Hilla Limann with the backing of militant student organizations, unions, and

left social movements. By the time of the coup, the economy had deteriorated badly, and the Limann

government found itself faced with strikes and confrontations with workers over back pay and a tough

austerity program. Upon seizing power, Rawlings actively solicited the support of these forces by placing

representatives of radical left organizations on the militarys Provisional National Defense Council and

creating a raft of populist consultative organizations (Graham 1985; Hutchful 1997). Rawlings populism only

aggravated Ghanas economic problems, and the military regime ultimately reversed course entirely and

vigorously embraced the Washington consensus. But the initial overthrow of the democratic regime clearly

appealed to, and mobilized support from, populist and leftist social forces.

The Null Cases: Non-distributive Reversions

The other high-inequality reversions are Peru, Nigeria, and Thailand; Sierra Leone is a high-inequality

case on the capshare measure but is not included in the Houle analysis.8 Some of these involved broad

appeals that cut across class lines, while others resulted primarily from conflicts within the elite itself. But in

none of the cases were redistributive cleavages central to the reversion and in several the specific political

pressures stipulated by the theoryredistributive democratic governments or social movementsseem

altogether absent.

8 Sierra Leone was riven with violent ethnic and geographic divisions, but these did not map to cleavages between elites and marginalized masses and we coded itimperfectlyas a cross-class reversion.

35

Peru. Alberto Fujimoris decision to close congress and rule by decree in April 1992 drew support

from a broad cross-section of Peruvian society. Military backing was, of course, essential and was motivated

in part by the desire for a free hand to confront a growing terrorist insurgency, the Shining Path, that had

pretenses of representing disadvantaged peasants in some highland areas of the country. But in other

important ways, the case does not correspond with the theory. First, the self-coup initially met opposition

from international and some local business sectorsin short, from economic elites--who were concerned

that an outright dictatorship would be destabilizing. Although these sectors eventually warmed to the regime

after Fujimori agreed to a faade of constitutionalism, they were by no means drivers or even supporters of

the coup.

At the same time, Fujimori enjoyed surprisingly wide popular support, which materialized most

immediately in a referendum on Fujimoris proposal for a new constitution aimed at legitimating his hold on

power. The unions and the political left did oppose the coup, but their organizations had been decimated by

the hyperinflation and economic collapse of the late1980s and they themselves enjoyed little popular support.

The large majority of the Peruvian poor were attracted by a leader who promised to deal with the economic

crisis and the insurgency with a strong hand. One 1992 survey showed that almost 76 percent of low-income

people supported Fujimoris plan for constitutional reform (Rubio 1992, 7, cited in Weyland 1996, fn 16.),

and in fact they gave Fujimori an overwhelming victory in the vote that followed. While undertaking

economic reforms, Fujimori also strengthened his electoral base through the expansion of clientelistic anti-

poverty programs as the economy revived (Weyland 1996). In the late 1990s, as the economy once again

slowed and corruption scandals surfaced, Fujimoris popularity waned and he was eventually forced to

withdraw from power. Until that time, however, his government rested on a surprisingly broad cross-class

coalition.

Nigeria. As in Peru, the 1983 coup in Nigeria occurred in the context of severe economic

deterioration and a widespread loss of public confidence in the willingness or capacity of a highly corrupt

government to cope with it. But there are no indications that the takeover was motivated by class or ethnic

36

demands on the state, nor by significant involvement of civil society one way or the other. The most

consequential divisions were within the elites, most notably, the military, clientelistic politicians, and the

business class that had formed the core of the ruling class during the oil boom of the 1970s. When oil

revenues collapsed, the ruling coalition fragmented under the competing claims for patronage. The inability

of the hegemonic party to reconcile these conflicting interests, argues Augustine Udo (1985: 337), came to a

head in a blatantly corrupt election in 1983, which exposed unprecedented corruption, intimidation, and

flagrant abuse of electoral privilege by all parties. The leader of the coup, Major General Muhaamadu

Buhari, waslike his predecessorstied closely to the Muslim north and had held a high position within the

deposed government. There is little evidence that factional rivalries within the military were connected with

broader conflicts that could be modeled in elite-mass terms, whether engaging class, ethnic or regional

interests.

Thailand 1991. In the other high-inequality reversions, redistributive conflicts were even less in

evidence. The 1991 coup in Thailand was undertaken by a military faction that bridled under both the existing

military leadership and the efforts of the elected Assembly to exercise greater control over military spending

and prerogatives (Baker and Phongpaichit 2002). Elected officials were concerned, among other things, with

channeling patronage resources to disadvantaged parts of the country, but they were linked closely to up-

country business interests. Although the distribution of income had deteriorated in Thailand during the

economic reforms of the 1980s, left parties remained confined to the fringes of political life and a long-

standing rural insurgency had long since petered out. The coup had the effect of galvanizing mass

opposition, including groups explicitly representing the poor, and this subsequently played a role in the

transition back to democratic rule. But there is no evidence that the coup either responded to popular

pressures for redistribution or reflected populist-authoritarian dissatisfaction with democracys failure to

redress redistributive grievances.

Low and Medium Inequality Cases. Although we do not elaborate here on the reversions in the

medium- and low-inequality cases, redistributive conflicts did not play a significant role (Table 4). In Ecuador,

popular pressures from the powerful indigenous movement propelled a short-lived populist reversion. In the

37

remainder of the cases, however, distributive conflicts were absent and reversions appear to be driven either

by the exploitation of broad cross-class support or by intra-elite conflicts. As Table 4 also makes clear, this

finding is true regardless of the measure of inequality used.

The Stability of Democratic Rule: The Weak Democracy Syndrome

We do not seek to elaborate an alternative theory of democratic instability during the third wave,

but our analysis suggests a weak democracy syndrome that comports with a growing body of literature on

democratic vulnerability (Diamond 2008; Levitsky and Way 2010). A first point, however, concerns the

coding choices required to create large-N panel data sets. Table 5 shows that eight of the 25 cases coded as

reversions in the PCG data set did not rise above the standard Polity cutoff score of 6 in the year preceding

their collapse (Bolivia, Burundi, Comoros, Congo, Guatemala, Sierra Leone, Thailand, and Uganda). Another

five cases (Ghana, Nigeria, Pakistan, Peru, and Sudan) barely make that threshold with scores of either 6 or 7.

The average Polity score for all of the reversion countries in the year preceding the collapse of democratic

rule is only 5, and a difference-of-means test shows that this is significantly different from the non-reversion

cases at the at the .01 level. Reversions are occurring against democracies that are marginally democratic at

best.

Yet the weakness of the distributive conflict theory of regime change is not simply an artifact of the

coding rules; the causal mechanisms stipulated in the theory do not appear to operate either. Electoral

competition in Thailand, Nigeria, Pakistan, Honduras, Ecuador, Ghana, and Guatemala was dominated by

patronage parties with close ties to economic elites or the military establishment, and with the political playing

field tilted heavily toward incumbents. In none of these cases do we see a significant presence of parties,

interest groups or social movements representing the interests of the poor that could serve as the trigger for

elite intervention.

Rather, conflicts within the political elitebetween ins and outswas more likely to pose a challenge

to democratic rule, with the military playing a pivotal role. In the eleven cases in which distributive conflicts

were implicated (the elite reaction and populist reversions) the military could plausibly be seen as an agent of

either elites (elite reaction reversions) or excluded social forces (populist reversions). But in many of the other

38

fourteen cases, the military entered politics largely on its own behalf. This was more likely to occur where

prior military intervention had established a precedent. Cross-national quantitative work on both Latin

American and Africa finds that the likelihood of a military coup is strongly affected by the previous history of

coups (Collier and Hoeffler 2005; Lahoucq and Perez-Linan 2009). Table 5 shows that fourteen of twenty-

five reversions came in countries that had already experienced at least one prior coup, and in eight of these

cases, the military was a repeat offender.

Our replication of the Houle model in Table 3 also shows that both the level of income and

economic growth are consequential for the stability of democratic rule during the 1980-2000 period. As

Londregan and Poole (1990) and Przeworski et al. (2000) have shown convincingly, the probability that

democratic governments will survive is strongly affected by level of development; the reversion cases have

low average income and the African reversion cases include some of the poorest countries in the world.

Among the non-African cases only Thailand, Ecuador, and Peru are middle-income countries. Average GDP

per capita for the reversion cases at the time of the collapse of democratic rule was only $835, way below the

thresholds for consolidated democracies.

The relationship between short-run economic performance and reversions has also been explored in

some detail (Haggard and Kaufman 1995; Gasiorowsksi 1995; Teorell 2010; Kricheli and Livne 2011).

Przeworski et al. (2000) show that the odds of democratic survival decrease substantially after three

consecutive years of negative economic growth. Most of the reversion cases were experiencing zero or

negative growth, and a number were in the midst of full-blown economic crises (Table 5). On average, the

economies of the reversion countries declined by 0.5 percent in the year of