Human Rights Violations and Competitive Elections in Dictatorships

Human Rights Violations and Competitive Elections in Dictatorships Jessica Maves The Pennsylvania State University Department of Political Science jessica.maves@psu.edu Seiki Tanaka Syracuse University Maxwell School of IR stanaka@maxwell.syr.edu April 11, 2013

1 Introduction It has been increasingly established in the comparative politics literature that dictators hold elections in order to gain information about their popular support and to appease and coopt a threatening domestic opposition (see, e.g. Cox 2009; Gandhi & Przeworski 2006). However, literature on human rights violations provides mixed notions as to whether we should expect more or less repression in these situations of elections under threat from opposition groups. Some suggest that opposition parties result in an increased likelihood of human rights violations (Vreeland 2008; Conrad 2011) while others contend that democratic institutions like elections and especially those that involve competition decrease the willingness of leaders to torture (Conrad & Moore 2010). This paper is an effort to better understand under what conditions human rights violations will be more or less likely around elections in dictatorships. We are interested not only in the comparison between competitive and non-competitive elections, but also how uncertainty over the electoral process and election outcomes shape the likelihood of repressive actions by the regime. We begin by presenting a brief overview of the recent quantitative literature on human rights violations, and build on this tradition to forward our own expectations about such violations in competitive versus non-competitive election environments. We go on to present additional hypotheses conditioning our expectations about competition, by dynamics of uncertainty surrounding the election event itself. Our initial efforts at multivariate empirical tests of these hypothesized associations support most of our expectations. We conclude with a discussion of the findings and plans for future work on this topic. 2 Background The quantitative and theoretical literature on human rights violations has exploded considerably in recent years, both in terms of quality and quantity. Of considerable interest to 1

us is the burgeoning work being done on how characteristics of political institutions, across all regime types and within autocracies in particular, shapes expectations and outcomes of human rights violations. Vreeland (2008) begins to get at an aspect most relevant to our study: repression in the face of political competition and organized opposition political parties. He finds that dictators are more likely to sign the Convention Against Torture (henceforth CAT) as a cheap concession when faced with challenges from one or more opposition parties, but that these regimes are also more likely to engage in torture than those regimes that lack opposition parties. Conrad (2011) also argues that opposition parties provide an incentive to torture, but that this should only hold for those countries with ineffective domestic judiciaries a common trait of many dictatorships. Conrad and Ritter (2012) contend that politically insecure leaders (defined by their time in office, previous trends in leadership change, and economic performance) repress in order to contain dissent. Although they focus on non-institutionalized forms of challenges to the state, such as protests, we choose to examine more closely the relationship between human rights abuses and institutionalized forms of challenges to the state, namely elections. In particular, we are interested to see if competitive elections serve as a sufficiently unsettling threat to the stability and security of the incumbent s time in office. Conrad and DeMeritt (2013) consider how the regime s sources of support to stay in office matter for its willingness to engage in repression. They expect that states reliant on citizens for revenue are less likely to repress than those states with unearned economic revenues, because leaders of the former type of state need voters to have a favorable view of the incumbent when going to the ballot box. The authors demonstrate that states with considerable unearned economic revenue are more likely to torture, but that democratic institutions effectively mediate this relationship. Conrad and Moore (2010) find, similarly, that democratic institutions that provide cit- 2

izens with a chance to vote for political officials have a moderating effect on the likelihood that states engage in torture. This finding, however, is conditional on the absence of armed challenges to the state if the regime is threatened by violent dissidents, torture will continue regardless of what domestic political institutions are in place to give citizens a voice. 3 Theory This work suggests that regimes beholden to the public for security in office should be less likely to torture, but that on some level the natural inclination of leaders would be to engage in repressive behavior. In particular, the extant literature suggests that dictators will engage in human rights violations when faced with opposition political parties. Conrad and Moore (2010) find that, even in the presence of democratic institutions, leaders will torture when faced with a violent challenge, but we aim to examine whether non-democratic leaders will also choose to torture disproportionately when under threat from a less direct form of challenge competitive elections. The electoral institution exists in a different environment, or means something different, in dictatorships as compared to democracies. Dictators often use elections not to fairly decide who will hold power and to facilitate alternation between parties (as is the case for democracies), but instead elections are held in order to gain information about their popular support (Cox 2009) or to appease the opposition and afford it some level of participation in a controlled environment, especially one in which the opposition can be coopted and eventually have shared interests with the ruling party (Gandhi & Przeworski 2006; Magaloni 2006). As a result, to limit the potential for confounding variation, we focus our study on non-democratic states. This allows us to examine with greater assurance the differences imposed on the likelihood of human rights violations by environments of competitive, noncompetitive, and the complete absence of elections. Based on existing work on torture and opposition parties (namely, Vreeland 2008; Con- 3

rad 2012; and Conrad and Ritter 2013) we anticipate that dictators will be more inclined to torture when these opposition parties are participating for positions of power via a competitive election, as compared to a non-competitive election. 1 However, it is important to note that these authors do not consider the way in which election periods might be different from non-election periods instead, they simply argue that torture will be more likely in the presence of opposition parties. Figure 1 shows varying frequencies of torture in country-years where there are competitive elections as compared to years in which there are no competitive elections (top row). The bottom row of Figure 1 breaks down the distribution of torture frequencies in non-competitive election years to differentiate between years in which there was a non-competitive election and years where no election occurred at all. Importantly, we see that torture rates in non-election years are, across the board, lower than in years where there is either type of election. Thus, election periods appear to bring about a heightened likelihood of torture in dictatorships. Figure 1 about here Despite the lack of apparently obvious difference between rates of torture in competitive versus non-competitive election years as depicted in Figure 1, we expect that the narrative told in the extant literature about the threat from opposition parties increasing propensity for torture should still hold, when we control for other potentially confounding factors. Dictators will feel under particular threat as a result of the multiparty nature of the election, and therefore will be inclined to go to great lengths in order to deter the opposition from mounting a serious challenge to the incumbent. Importantly, this is somewhat in contrast to the expectations of Conrad and Moore (2010) who suggest that, in general, situations in which citizens are given a voice (e.g. through elections where they have a choice of candidates), the regime should be less likely to torture. We anticipate the opposite with respect to dictators facing competitive elections. 1 We follow Cederman et al. (2013) by defining competitive elections as those where opposition was allowed, more than one party was legal, and there was a choice of candidates on the ballot three variables drawn from the National Elections Across Democracy and Autocracy (NELDA) dataset (Hyde and Marinov 2012). 4

Hypothesis 1 Human rights violations will be more likely around competitive elections, as compared to non-competitive elections. We are especially interested to examine whether the introduction of genuine competition to the electoral environment makes torture more likely, as compared to electoral systems that are non-competitive, or non-existent. We expect that uncertainty surrounding the preferences and behavior of the opposition in competition, as well as the outcome more generally, should lead ruling elites to distrust and feel threatened by the electoral system when competition is first introduced. Instead, we expect that dictators will prefer to take matters into their own hands by repressing the opposition candidates and supporters. This is much in line with what Dahl (1971) suggests might happen when elites are faced with the introduction of competitive institutions, rather than first tolerating the existence of institutions followed several iterations later with the allowance for opposition groups to participate. We conduct a few bivariate examinations to see whether the introduction of competitive elections produces different repression outcomes in situations where there is a history of non-competitive elections as compared to those environments where elections never took place to begin with. Dahl s theory might also have some lessons for what we should expect after competitive elections are introduced. Once political actors have had some experience with competition in this environment, and they have a sense of their opponents preferences and behaviors, they should be less inclined to commit human rights violations. As such, we should see a decline in the frequency or degree with which regimes engage in repression after the first competitive elections are held. Returning to the notion of elite mistrust of the opposition and institutions in nondemocratic (and previously non-competitive, in particular) regimes, we expect that uncertainty surrounding the election process is what drives the spike in the likelihood of human rights violations around elections. We propose two hypotheses, which consider alternative (though potentially complimentary) types of uncertainty about the election process: uncer- 5

tainty over the process and uncertainty over the outcome. Dictators who have not previously had experience with elections and are thrust immediately into the world of competitive elections (examples) should be less inclined to trust that this institution will facilitate their maintenance of office, as compared to those leaders who have previously held (and won office through) non-competitive elections. Uncertainty over how the process will work will predispose dictators who are completely new to the world of elections to opt to manage threats from the opposition by way of human rights violations. Dictators would rather take matters into their own hands and repress the opposition in hopes that it will deter any meaningful challenges during the election (or after the event as well) than sit back and hope that this new form of competition will allow the status quo to persist. Thus, we expect the following, our Uncertainty Over Process hypothesis: Hypothesis 2 Human rights violations will be more likely in dictatorships that hold competitive elections but have not previously held an election, as compared to dictatorships that have past experience with non-competitive elections. Somewhat similarly, we also expect that human rights violations will be more likely in the presence of uncertainty over how the election will turn out. If dictators are unsure as to whether they will actually win, they will be inclined to target opposition figures and supporters in hopes that such repressive actions will compel the opposition targets to cease and desist in their activities against the regime around an election. While this logic should apply to competitive elections primarily, it is possible that noncompetitive elections also experience uncertainty. Recall that the criteria for competitive elections involves having multiple legal parties, including an opposition party, as well as a choice of candidates on the ballot. Non-competitive elections can involve a choice of different candidates within a single party (this occurs in some Communist states, for example). Therefore, the leadership may feel under threat from within its own party and choose to engage in repression of those supporting co-party opponents even though the election is technically non-competitive. 6

The following is our Uncertainty Over Outcome hypothesis: Hypothesis 3 Human rights violations will be most likely in dictatorships that hold competitive elections shrouded in uncertainty over the probable outcome. Hypothesis 4 Dictatorships that hold non-competitive elections with uncertainty over the outcome will be less likely to engage in human rights violations than those regimes facing competitive elections with uncertain outcomes, but more likely to torture than those regimes facing elections with more certain outcomes. 4 Research Design We are interested in the likelihood of human rights violations surrounding elections in dictatorships, and how these violations are conditioned by the type of election being held. The unit of analysis in this study is the election (following Hafner-Burton et al. 2013), and it includes election events from the NELDA dataset covering 1960-2006. For reasons described previously, we focus our analyses on election events in dictatorships. We conceptualize two types of human rights violations, both of which are binary. This allows us to use a maximum likelihood estimator, and our model of choice is the logistic regression. The first dependent variable of interest to us is whether the regime engages in torture, according to the CIRI dataset (Cingranelli & Richards 2010). This variable is coded 1 if the CIRI dataset codes the state as engaging in widespread torture in a given year, and 0 otherwise. These data only begin in 1981, so our temporal domain for these tests is left-censored accordingly. The other version of human rights violations we consider as a dependent variable concerns election-specific human rights violations. Drawing from variables in the NELDA dataset, we construct a variable electoral violations, which is coded 1 if the regime harassed members of the political opposition (NELDA15) or if it engaged in significant violence involving civilian deaths immediately before, during, or after the election (NELDA33). For example, 7

the Khmer Rouge killed several members of rival political parties ahead of 1993 elections in Cambodia. Surrounding the 1974 Guatemalan elections, the government and its allied forces violently targeted the opposition party National Opposition Front, and also shut down radio stations and occupied the party headquarters of the Christian Democrats. Electoral violations is coded 0 if neither of these variables from NELDA are coded in the affirmative. The primary independent variables concern whether the election was competitive. Again, we follow Cederman et al. (2013) in our definition of competitive elections: coded 1 if opposition was allowed to participate (NELDA3), more than one party was legal (NELDA4), and there was a choice of candidates on the ballot (NELDA5). The variable is coded 0 if one or more of these conditions is not met. In order to test our hypotheses about the two types of uncertainty, we create two variables to interact with competitive elections. The first, election history, is a binary indicator of whether the country held a non-competitive election in the past 10 years. This variable is intended to capture uncertainty over the process of elections. The second variable, certain outcome, is again drawn from the NELDA dataset (NELDA12) and is coded 1 if an incumbent is overtly confident, according to public statements, of victory in the polls. This variable represents uncertainty over the outcome of elections. According to the information on elections drawn from NELDA, within our dataset 126 countries held some sort of election. In 67 of these states, the first election was competitive, whereas 35 states were holding non-competitive elections prior to their first competitive election. 24 states have never held a competitive election. We also consider whether the election, if competitive, is the first competitive election in a country s recent history, and differentiate this event from second-and-later competitive elections. This variable is coded 1 if a competitive election is the first such event in at least 10 years, and 0 otherwise. Second competitive election actually captures all competitive elections after the first such event, as described above. This variable is coded 0 if the election is the first competitive election or is non-competitive, and 1 otherwise. 8

We also control for a handful of covariates commonly associated with political violence. We account for the degree to which a dictatorship is dictatorial by including their Polity score (Marshall & Jaggers 2002). We also include logged values of a country s population, as well as its per capita GDP (Penn World Tables v6.3). 5 Findings The first set of results we will discuss deal with the effect of competitive elections on torture and election-specific human rights violations, without conditionality on types of uncertainty. Table 1 presents results for both of the dependent variables using two different specifications, one concerned with competitive elections in general and another differentiating between first and subsequent competitive elections. Table 1 about here Models 1 and 2 address how competitive elections affect the likelihood of torture in particular, as defined by CIRI s widespread torture category. In both models, we observe a marked increase in the probability of observing torture around competitive elections, as compared to non-competitive elections (the reference category). We do not observe a considerable difference between the first and subsequent competitive elections with respect to the likelihood of torture, although the coefficient for first competitive election is slightly larger suggestive of a comparatively higher propensity for torture around the first competitive election as compared to subsequent competitive election periods as well as non-competitive elections. Models 3 and 4 consider the likelihood of election-specific human rights violations in the presence of competitive elections as compared to non-competitive elections. Again, we find that competitive elections are more likely to correspond to increased levels of human rights abuses, this time in the form of the harassment and killing of civilians. Model 4 demonstrates that competitive elections in dictatorships lead to higher likelihoods of human rights viola- 9

tions not only around the first such event but also in subsequent contests, although, again, the coefficient is slightly larger in the context of the initial competitive election. In sum, across Models 1 through 4, we find support for our first hypothesis: among dictatorships, human rights violations (specifically in the form of torture, here) are more likely during competitive elections as compared to non-competitive elections. Hypotheses 2 through 4 suggest a conditional relationship; namely, that uncertainty makes dictators particularly prone to human rights violations in the presence of competition. Table 2 presents results for models where the dependent variable is torture, specifically, whereas Table 3 focuses on the same set of models that instead employ election violations as the dependent variable. Table 2 about here Model 4 addresses how competitive elections in the presence of uncertainty over election outcomes shape the dictator s propensity to engage in widespread torture. Here, we see that when the dictator is uncertain over his chances of retaining office (certain outcome=0), he is more likely to engage in torture, as compared to dictators engaged in non-competitive elections or those more certain of their fate. This is in keeping with Hypothesis 3. However, we do not observe statistical significance on the other constituent term certain outcome nor on the interaction term, so we cannot confidently conclude that dictators holding noncompetitive elections with uncertain outcomes are more likely to torture than those facing competitive elections with certainty (Hypothesis 4). A model considering outcome certainty for first versus subsequent competitive elections would not converge, so we cannot determine how the nature of the newness of competitive elections affects the likelihood of torture, given election (un)certainty. Models 5 and 6 consider our expectations about how the likelihood of torture is influenced by uncertainty over the election process, for competitive elections in general as well as first versus subsequent competitive elections, respectively. The results are quite similar to the findings for uncertainty over election outcomes. Widespread torture is more likely 10

surrounding competitive elections in general (regardless of whether it is the first or subsequent such events), but only in countries that have not held any sort of election in the last 10 years. In those regimes where their first election is competitive, and the ruling elite is not yet comfortable with the nature of vying for power, torture is more likely than in countries that held, or continue to hold, non-competitive elections. This is in keeping with our second hypothesis. Table 3 about here Models 7 through 10 focus on the likelihood of election-specific human rights violations in the presence or absence of uncertainty during competitive and non-competitive elections in dictatorships. First, looking at Model 7 which considers competitive versus non-competitive elections in general, competitive elections in the absence of certainty over election outcomes make harassment and severe violence against civilians more likely than in non-competitive election environments. This is in keeping with Hypothesis 3. However, and interestingly, dictators are less likely to engage in these election-specific human rights violations during competitive elections in which they express certainty over the outcome, as compared to non-competitive elections. This is largely in line with our expectation that dictators facing non-competitive elections will be more likely to torture than dictators who are more assured of victory (Hypothesis 4). Model 8 looks at competitive elections in general, and again finds that dictators whose first (recent) election is competitive are more likely to harass or kill civilians than those dictators facing non-competitive elections, or with a history of non-competitive elections in the last 10 years. This supports our expectation of Hypothesis 2, that uncertainty over the election process compels dictators to engage in human rights violations when faced with competitive elections. Models 9 and 10 investigate these same relationships, but build on them by considering how the first competitive election varies with respect to the likelihood of election-specific human rights violations as compared to subsequent such elections. Yet again, we observe 11

that competitive elections are more prone to violence than non-competitive elections in the presence of both uncertainty over the process and the outcome (looking at the constituent term of first and second competitive election). And, again, the coefficient for the former is greater in magnitude than the latter, suggesting that dictators facing their first competitive election are especially likely to commit election-specific human rights violations. Electionspecific human rights violations are more likely given competitive elections where the outcome is not obvious, as well as when there is a lack of familiarity with the election process by way of previous non-competitive elections. Thus, Hypotheses 2 and 3 receive support for electionspecific human rights violations as well. Looking at the interaction terms in Model 9, concerning uncertainty over the outcome of an election, we see that dictators who are more certain of the outcome of a competitive election are less likely to harm civilians, as compared to dictators holding non-competitive elections. This provides additional support for Hypothesis 4, with respect to election-specific human rights violations. 6 Conclusion This paper represents an effort to better understand what types of election environments in dictatorships are especially prone to human rights abuses. We demonstrate that competitive elections are potentially dangerous events, particularly when they take place in the absence of a tradition of non-competitive elections (so the dictator does not trust the institution of elections) and when they occur in the presence of uncertainty over the outcome. These findings provide important practical implications for policymakers hoping to stave off election violence and human rights violations. Practitioners would be well-advised to focus their efforts to protect civilians from repression in those dictatorships that are holding elections namely competitive ones for the first time, and where it looks as though the incumbent might lose. Our findings indicate that these are situations where civilians are 12

most at risk for abuses by the government. It is entirely likely that dictators design electoral systems in keeping with their willingness and ability to torture. Dictators who face a high cost for engaging in human rights violations (because they are members of CAT, experience pressure from other international actors/agencies, or heavily rely on popular support to stay in office) may choose to hold non-competitive elections, or avoid holding elections in general, because they do not want to torture. After all, our initial results here suggest that competitive elections are especially predisposed to encouraging human rights violations as compared to these other electoral options. As a result, in the future we will include a two-stage model that takes into account the first step in this process: the dictator s decision to hold a specific type of election. Initial attempts using a Heckman selection model did not converge, but we will keep at this effort to address the possibility of endogeneity between institutional design and the decision to torture. 13

References Cingranelli, D.L., and D.L. Richards. N.d. The Cingranelli-Richards (CIRI) Human Rights Dataset.. Conrad, Courtenay R. 2011. Constrained Concessions: Beneficent Dictatorial Responses to the Domestic Political Opposition. International Studies Quarterly 55: 1 21. Conrad, Courtenay R., and Emily Hencken Ritter. N.d. Treaties, Tenure, and Torture: The Conflicting Domestic Effects of International Law. Journal of Politics. Forthcoming. Conrad, Courtenay R., and Jacqueline H.R. DeMeritt. 2013. Constrained by the Bank and the Ballot: Unearned Revenue, Democracy, and State Incentives to Repress. Journal of Peace Research 50(1): 105 119. Conrad, Courtenay R., and Will H. Moore. 2010. What Stops the Torture? American Journal of Political Science 54(2): 459 476. Cox, Gary W. 2009. Authoritarian Elections and Leadership Succession, 1975-2004. Presented at the Annual Meeting of the American Political Science Association, Toronto. Gandhi, Jennifer, and Adam Przeworski. 2006. Cooperation, Cooptation, and Rebellion under Dicatorships. Economics & Politics 18(1): 1 26. Hafner-Burton, Emilie M., Susan D. Hyde, and Ryan S. Jablonski. 2013. When Do Governments Resort to Election Violence? British Journal of Political Science p. Forthcoming. Hyde, Susan D., and Nikolay Marinov. 2012. Which Elections Can Be Lost? Political Analysis 20(2): 191 201. Magaloni, Beatriz. 2006. Voting for Autocracy: Hegemonic Party Survival and its Demise in Mexico. Cambridge University Press. 14

Marshall, Monty, and Keith Jaggers. 2002. Polity IV Project: Political Regime Characteristics and Transitions, 1800-2002.. Vreeland, James Raymond. 2008. The Effect of Political Regime on Civil War: Unpacking Anocracy. Journal of Conflict Resolution 52(3): 401 425.???? 15

Figure 1: Comparison of Torture Rates Across Election Environments 16

Table 1: Random-effects logit models, no interaction Model 1 Model 2 Model 3 Model 4 Dependent Variable (Torture) (Torture) (Election Violations) (Election Violations) Competitive election 2.743 1.575 (0.910) (0.272) First competitive election 2.835 1.748 (1.065) (0.306) Second competitive election 2.698 1.428 (0.945) (0.295) Polity -0.149-0.149-0.057-0.053 (0.080) (0.080) (0.028) (0.029) ln(population) 0.700 0.705 0.130 0.136 (0.381) (0.384) (0.107) (0.108) ln(gdp per capita) -0.421-0.418 0.012 0.025 (0.414) (0.415) (0.169) (0.169) Constant -0.753-0.814-3.072-3.205 (4.750) (4.779) (1.651) (1.658) Log likelihood -83.00-82.99-391.8-391.0 No. of observations 379 379 657 657 Notes: Significance levels (two-tailed): : 95% : 99% : 99.9%. Coefficients with robust standard errors in parentheses. 17

Table 2: Random-effects logit models, interactions (DV=Torture) Model 4 Model 5 Model 6 Competitive election 2.896 2.539 (0.944) (1.209) First competitive election Second competitive election 2.659 (1.209) 2.496 (1.209) Certain outcome 21.480 (28.668) Competitive x Outcome -22.029 (28.668) First competitive x Outcome Second competitive x Outcome Election history -0.508-0.508 (1.022) (1.022) Competitive x History 0.193 (1.484) First competitive x History 0.162 (1.991) Second competitive x History 0.180 (1.570) Polity -0.148-0.160-0.160 (0.078) (0.084) (0.086) ln(population) 0.700 0.734 0.739 (0.365) (0.397) (0.401) ln(gdp per capita) -0.544-0.420-0.417 (0.417) (0.421) (0.423) Constant 0.085-0.709-0.792 (4.594) (4.898) (4.937) Log likelihood -82.13-82.82-82.80 No. of observations 379 379 379 Notes: Significance levels (two-tailed): : 95% : 99% : 99.9%. Coefficients with robust standard errors in parentheses. 18

Table 3: Random-effects logit models, interactions (DV=Election Violations) Model 7 Model 8 Model 9 Model 10 Competitive election 1.708 1.094 (0.290) (0.426) First competitive election 2.236 1.293 (0.351) (0.468) Second competitive election 1.459 0.947 Certain outcome 1.277 1.379 (0.850) (0.857) (0.317) (0.442) Competitive x Outcome -1.550 (0.907) First competitive x Outcome -2.965 (1.025) Second competitive x Outcome -0.908 (0.932) Election history -0.596-0.694 Competitive x History 0.610 (0.493) (0.391) (0.391) First competitive x History 0.546 (0.601) Second competitive x History 0.630 (0.534) Polity -0.058-0.062-0.065-0.059 (0.029) (0.030) (0.030) (0.030) ln(population) 0.143 0.138 0.152 0.144 (0.110) (0.108) (0.114) (0.108) ln(gdp per capita) -0.017 0.015-0.001 0.028 (0.173) (0.169) (0.179) (0.169) Constant -3.030-2.676-3.287-2,813 (4.594) (1.666) (1.747) (1.673) Log likelihood -390.4-390.2-384.2-389.4 No. of observations 657 657 657 657 Notes: Significance levels (two-tailed): : 95% : 99% : 99.9%. Coefficients with robust standard errors in parentheses. 19