Abstract
While research into why repression/human rights violation goes up or down has thrived over the past 50 years, essentially no effort has been made to examine what stops this behavior once under way – especially activity that is large-scale as well as violent. To address this topic, we put forward the idea of a repressive spell (similar to that in the study of war, civil war, and terrorism) and a new theoretical framework that conceptualizes repression as a sticky process that is unlikely to terminate unless it is disturbed in some manner. Such an orientation is important because it leads us to conclude that disturbance is more likely to happen under situations of democratization compared to any of the factors typically highlighted in the literature and relevant policy community. Investigating a new database regarding 239 large-scale repression spells from 1976 to 2006, we find that democratization is associated with spell-termination and there is little systematic pacifying influence from anything else. Additionally, we find that nonviolent movements for change principally drive democratization but that these movements have little direct impact on state repression spells in and of themselves.
Since World War II, it has been a global priority to stop ongoing, large-scale state repressive activity – hereafter LSSR (i.e. government sponsored, violent action that is both widespread and systematic). 1 With this objective in mind, policymakers, NGOs, and activists have suggested and implemented a number of policies, including military intervention, economic sanctions, naming/shaming, international law, preferential trade agreements, political democratization, and nonviolent direct action – all are presumed to diminish state repression because they increase the costs of such activity. What (if anything) works? What stops LSSR? Despite the significant amount of interest in this question, we still know very little about it (e.g. Bellamy, 2015; de Waal & Conley-Zilkic, 2006). The primary reason for this limitation is that almost all empirically rigorous work has been focused on understanding what accounts for variation within levels of repression and what impact diverse cost-producing factors have had on this phenomenon. While related and providing important insights, this variation work is largely focused on something quite different from termination in terms of theory, data structure, and appropriate estimation. 2
To shed light on this important topic, the current article draws upon prior research but shifts the focus from movements in repression levels to termination of repressive spells. 3 Here, we argue that not all costs are comparable to political authorities in terms of their centrality to state repression and that political actors already involved in relevant activity are more likely to stay with enacted behavior unless something major changes in their environment. 4 We maintain that among existing policies put forward to increase costs and decrease repression, the controversial and often maligned policy of political democratization is the factor most likely to stop an LSSR already underway, as it involves the most significant cost that could be imposed – increasing accountability by involving more individuals and institutions in the political process and providing authorities with an opening/opportunity to step away from the highly controversial and ineffective coercive/forceful policy.
In an effort to assess our argument, we investigate a new dataset of 239 high-repression spells from 1976 to 2006. Results disclose that democratization is the only factor that consistently stops LSSR. In contrast, and equally as important as what we find support for, statistical investigations reveal that factors believed to impact repression in the literature as well as within diverse policy circles have essentially no direct impact on spell termination. 5 Pushing deeper into the causal chain in order to better understand what stops LSSR, we also examine what influences democratic transition. Here, we find that specific nonviolent campaigns are linked to a greater likelihood of democratization, suggesting that this political change has an important indirect role to play in stopping LSSR while having little direct impact on spell termination viewed on its own.
With these findings, our work calls for a re-evaluation of the policies currently put forward to stop large-scale state repression, advancing an approach that is more oriented toward disrupting/perturbing political institutions as well as those individuals/groups who have a vested interest in repressive policies. Additionally, our work calls for a re-evaluation of how repression is conceptualized and studied. In part, the shift from variation to spell termination marks an important pivot for research that not only better reflects the nature of how large-scale repressive behavior is enacted but also helps to more accurately identify the factors that are associated with driving it/bringing it to an end. Such a shift has the added benefit of bringing the systematic study of state repression more closely in line with work conducted on other forms of large-scale political violence (such as interstate and civil war); this research has long employed the conception of spells and termination with the distinct theoretical insights, units of analysis, and estimated strategies that accompany them (e.g. Hegre, 2004).
Repression and its variation
Over time, a large literature has emerged which seeks to understand what explains variation in state repressive action, that is, what moves repressive behavior up or down some scale (e.g. Henderson, 1991; Poe & Tate, 1994; Davenport, 1995, 2007a; Zanger, 2000; Davenport & Armstrong, 2004; Hafner-Burton, 2005; Krain, 2005; Peksen, 2009; Simmons, 2009; Wood, 2008; Kathman & Wood, 2011; DeMeritt, 2012; Murdie & Davis, 2012; Spilker & Bohmelt, 2012; Nordås & Davenport, 2013; Hill & Jones, 2014; Jones & Lupu, 2018). This work has been useful in advancing understanding of the relevant phenomenon. Nonetheless, the research here has developed some characteristic features that have directed and consequently limited scholarly effort.
For example, the primary focus of variation work has been in identifying explanations and variables that if changed would increase the ‘operational’ (e.g. salaries, training, and weapons) and/or ‘political’ costs (e.g. removal from office or damaged reputation) of repressive action. In these contexts of raised cost, the use of coercion and force is expected to be reduced because political authorities and their agents are expected to limit any losses that might be incurred from engaging in relevant behavior. Toward this end, research has consistently highlighted a set of factors/variables that are believed to have a negative impact: for example, military intervention (e.g. Krain, 2005; Kathman & Wood, 2011), economic sanctions (e.g. Wood, 2008; Peksen, 2009), naming/shaming (e.g. Hafner-Burton, 2005; Murdie & Davis, 2012), international law and courts (e.g. Simmons, 2009; Appel, 2018), preferential trade agreements (e.g. Hafner-Burton, 2005; Spilker & Boehmelt, 2012), democratization (e.g. Davenport, 1996; Zanger, 2000), and, more recently, civil resistance (e.g. Chenoweth, Perkoski & Kang, 2017). Accordingly, these variables also reflect the leading policies put forward to address the topic advocated by governments, NGOs, and INGOs the world over.
Although the idea of costs and the diverse measures used to operationalize them are useful, there are some limitations with this work as it relates to the objectives of terminating LSSRs. At the most basic level, while understanding what reduces repressive action across a range of values is relevant to the topic of termination, this is not the same thing as ‘ending’ a sustained spell/campaign/episode of repressive action (such as terminating interstate or civil war). A major part of the difference here is conceptual. One implicit point made in existing work is that political authorities make repeated and detailed evaluations and decisions regarding whether or not and/or how to apply repression and do so at regular intervals (e.g. every year). This, however, overlooks the difficulty of sustaining repressive action which, in turn, tends to underestimate the difficulty in shifting relevant practices over time. As portrayed, variation research maintains that decisionmakers are always open to changes in their political, economic, and demographic environment. It is possible though that decisionmakers are unlikely to change after a policy has been developed/implemented and that they would require something quite dramatic in the environment to shift in order to prompt alterations in calculations, perceptions, and behavior. In addition to this assumption, variation research also treats all levels of state repression as being equally interesting, assuming (generally implicitly) that the various steps up/down the relevant scale are influenced in the same manner by explanatory factors.
Such positions are problematic for two reasons. First, state repression that is extremely violent, widespread, and systematic (e.g. genocide) has consistently been viewed as being quite different from government coercion/forceful behavior that is limited in terms of scope and lethality (e.g. protest policing). Indeed, these differences have prompted distinct levels of attention from international legal bodies, human rights NGOs, and media, distinct laws (e.g. genocide and crimes against humanity), and distinct fields of study (e.g. genocide and mass atrocity). In line with this, it seems reasonable that LSSR would be influenced by different explanatory factors than less violent, smaller-scale as well as less frequent behavior, and that it would respond to different costs.
Second, particular levels of state repression might involve very different amounts of inertia and stickiness than others (e.g. Hafner-Burton, 2005). For example, it is likely the case that levels of repression that involve a larger number of actions, actors, targets, and degree of coercion/force violence are potentially less likely to be stopped. To assume that different levels of repressive action are comparably influenced by the same independent variables (a position currently maintained) is likely to underestimate the difficultly with which LSSR can be terminated once underway. This realization prompts us to reflect on how best to conceive and study the topic.
Repression as a spell
In an attempt to address the issues raised above, we maintain that violent, widespread, systematic state repressive action is essentially the end product of (1) political authorities who design the basic contours of the relevant policy, (2) the thoughts and discussions of coercion- and force-wielding institutions who work out the more specific details of the repressive effort, and (3) the actions of the relevant government agents who engage in repressive behavior and the norms of practice that develop after they have begun. Here, potentially thousands (or tens of thousands) of individuals and actions come to be involved. 6 This view is different from one where the decision calculus and repressive behavior are viewed as events or as an evaluation of distinct attributes regarding diverse activities (e.g. Wood & Gibney, 2010). In both of these circumstances, we maintain that the process of repression is somewhat obscured and simplified.
Drawing upon these insights, we wish to argue that repressive behavior in general – but especially the violent, widespread, and systematic behavior being referenced above – involves a great number of actors, actions and phases before governments use this behavior in the first place, and once set on the relevant path there is a lot involved with getting governments to stop. This should actually come as no surprise to researchers of state repression. Prior research has consistently referred to this phenomenon with a variety of labels that suggests such a framing (e.g. bureaucratic inertia [Gurr 1986a], the ‘law of the instrument’ [Gurr, 1986b], the weight of the past [Davenport, 1996], and ‘stickiness’ [Hafner-Burton, 2005]). 7 Related but in a slightly different vein, most scholars have found that lagged repression is a statistically significant influence on current state repression – in fact, when included within estimated models results usually disclose that this is the most significant variable of all those considered. 8 While discussed, however, what has been missing from previous empirical literature is the explanation of what lags and stickiness mean for how we think about, study, and understand state repressive behavior.
On this point, we argue that explicit attention to the factors noted above is important because it significantly deviates from prior work which normally conceives of the process involved as being one that can be easily altered by only introducing just the ‘right’ type of cost/variable/policy. This is not likely with our conception. Indeed, if the repressive process is difficult to shift once underway, then the cost-producing variable required to perturb the relevant process would need to be quite powerful. Without significant change, the repressive process would not be shifted in any manner and repression (i.e. the spell) would continue.
What (if anything) imposes a sufficiently significant cost on political systems engaged with repression that might shift behavior when applied? Given the situation described above, we argue that democratization holds the greatest promise for stopping an ongoing spell of LSSR because of its ability to fundamentally alter the government engaged in relevant behavior. Generally, this is because democratization impacts both political authorities and coercive/force-wielding agents. To explain this relationship further, we discuss in the next section what is involved with democracy and democratization, as well as what is involved with state repression.
Concepts and connections
In our research, ‘democracy’ refers to a particular form of institutional arrangement where some constituency (i.e. group of citizens) selects another group of individuals (i.e. politicians/representatives) to make decisions concerning the general direction of the relevant political unit (e.g. Cheibub, Gandhi & Vreeland, 2010; Coppedge, 2013: chapter 2). Democracy thus represents a particular set of relationships and practices where democratization signals a shift from a position of lesser toward a position of greater political democracy. Now, exactly how leaders are chosen or which actors are highlighted varies. Most centrally connected with the concept employed within this article and the literature writ large, scholars focus on the existence/quality of elections, what percentage of the population has an opportunity to participate, and whether political parties contesting the election represent the population. 9
In contrast to democracy and democratization, with its focus on influencing individuals at the pinnacle of the political hierarchy, repressive action is focused on influencing individuals at the nadir. Specifically, repression refers to situations when security forces engage in a particular type of coercion and force against those under the jurisdiction of the relevant political authority in an effort to control the target’s behavior and/or beliefs. This includes arrests, beatings, banning, torture, disappearances, and killing.
It is often assumed that those in charge of implementing repression are automatically beholden/accountable to political authorities – thereby linking the leaders of the political system to the behavior of the security apparatus, but this is not always the case. Indeed, non-responsiveness and self-serving interests are the central claims of one of the more critical and newer theories used to explain state repression (e.g. Demeritt, 2012; Mitchell, 2004; Tyson, 2018).
We find it useful to distinguish the upwardly focused institution of democracy from the downwardly focused policy/behavior of repression because it is the ability of the former to disrupt the latter that determines how likely it is that large-scale repressive action will stop after being initiated. For example, with new leadership, newly incorporated individuals/institutions, and the introduction of diverse mechanisms of accountability (relative to the situation under authoritarianism), we expect democratization to lead to transformations in how coercive/forceful power is used against those under its jurisdiction. Accordingly, this results in changes with the policies that leaders wish to pursue, how they allocate resources, where they derive support, and how they train, arm, and issue commands related to repression.
Large-scale repressive policies are specifically worthy of attention here because they can deleteriously impact the lives of many within the citizenry – intentionally or unintentionally. Moreover, repressive behavior has been found to have quite varied impacts on behavioral challenges (referred to as ‘backlash’) which could threaten the new political authorities with diminished support or actual removal. This sensitivity to backlash is further enhanced because of the increased involvement and importance given to citizens, as well as various institutions within the newly established democratic political system. Not only might the policymaking aspect of the repressive process be altered following democratization, but the willingness of security force agents to comply with coercive/forceful directives associated with the old (as well as the new) regime might be reduced. Such a position emerges because relevant agents would acknowledge that support for potentially unpopular policies might lead to unfavorable evaluations of their institution from the newly empowered political actors in the democracy, prompting diminished financial/political support as well as potential resistance/rebellion.
Now, note that we are not saying that political leaders and agents of repression become more pacific and peaceful after democratizing, which is how the subject is normally framed within the literature. Rather, we are saying two things.
First, given a shift in the political terrain, which enhances the various ways that political leaders could be negatively impacted by citizens, government personnel take advantage of the opportunity provided to them by the change in regime type in order to create an identity that is distinct from the group of individuals that were recently displaced. Differentiation through policy change is one of the more effective ways to accomplish this. Such an argument is consistent with some research that advances the claim that ‘[c]ontentious politics lies at the heart of political development because it powerfully and persistently shapes how political alliances and rivalries will be defined. At its core, politics is not about individuals pursuing material gains, or internally cohesive economic classes pursuing shared wealth. It is about factions striving and struggling for political power, advantage and survival’ (Slater, 2010: 276). Interestingly, as the imposition of repressive behavior is believed to assist one cohort to struggle, survive, and thrive, the withdrawal of repression is believed to do the same. 10
Second, again different from prior research, we are not maintaining that political authorities step away from repressive action in the wake of democratization because they believe that a decent part of the population might not view such behavior favorably. Rather, we wish to acknowledge the general instability, weakness, and inefficiency involved with repression (e.g. its mixed impact on political contention [Davenport, 2007a]) and that a new democratic government would be eager to move away from such a policy as a way to facilitate its establishment and longevity in office, relying on the principles of deliberation, tolerance, and channeling grievances into the political system through the vote. It is thus not a commitment to democracy per se but the availability of an alternative mechanism of influence/control that explains the movement away from repression.
Our approach is distinct from existing literature because once the LSSR has been initiated, we do not expect most variables and the diverse policies associated with them to have much of an impact on government coercion and force. This is explained by the fact that these variables/policies do not profoundly disrupt the decisionmaking and implementation processes involved. Indeed, in our conception, the members of the particular clique/cohort associated with the initial decision are more likely to close ranks and limit criticism than they are to open themselves up to a change of policy. Democratization perturbs because it disrupts the inner workings of the coalition of policymakers and security force agents that coalesces around large-scale applications of repressive action. 11
Now, we should be clear: maximalist conceptions of democracy/democratization tend to conflate structural discussions about how leaders are selected with a particular constellation of preferred policy outputs like reduced state repression (e.g. Hill & Jones, 2014). Our approach to the topic is comparably more minimalist, and in line with diverse theorists, policymakers, activists, and citizens, we view democracy/democratization as simply and straightforwardly a mechanism of leadership selection. Such a position we view as reasonable because it allows us to gauge the impact of specific institutions/practices on specific government behavior, avoiding tautology. Such a position reflects the longstanding opinion of theorists, activists, and citizens from around the world who view democracy as a (and perhaps ‘the’) solution to the problem of state repressive action. Such an opinion is also consistent with the extensive quantitative literature which consistently has produced results that democracy decreases repression. Clearly, though, this impact has not been found to be as simple as is often thought. Indeed, some nuanced and sophisticated research reveals support for only certain levels of democracy decreasing repressive behavior (e.g. Bueno de Mesquita et al., 2005; Davenport, 2007b), a few pieces show that movements toward democracy reduce repression (e.g. Davenport, 1999; Zanger, 2000), and some work discloses that full democracies can avoid repression only when behavioral challenges are not present (e.g. Davenport, 2007b; Conrad & Moore, 2010).
Research design
Conceptually, a spell of large-scale state repression is defined as the duration of time when government repression is above a certain level of severity/lethality and scope (i.e. size of the targeted population). The threshold used to determine a spell was derived from the Political Terror Scale (hereafter PTS) – the most frequently used indicator of repressive behavior in existing scholarship, and the most relevant to our discussion as it simultaneously addresses the lethality of behavior enacted and the scope of the population being targeted (Wood & Gibney, 2010).
Operationally, the PTS measure is coded on an ordinal 1 to 5 scale. Scores of 1 and 2 correspond to countries where political torture, beatings, and murder are extremely rare and isolated. In contrast, a significant change in repression/human rights violation is noted above level 2, which corresponds to the interest of citizens, lawyers, theorists, activists, and advocates alike. Indeed, with a score of 3, there is unlimited detention for political views, extensive political imprisonment, execution, and brutality. A score of 4 means that murders, disappearances, and torture have expanded to large numbers of people for political reasons, and, finally, a score of 5 represents an environment where state terror of the worst kind has expanded to the whole population, not just those engaged in political activities.
Based on these definitions and given our focus, the threshold to determine a spell is designed as a PTS score of 3 and above. 12 This LSSR continues until the PTS drops below 3 for at least two consecutive years. The coding rule is consistent with prior work that looks at state repression in the worst situations (e.g. Hafner-Burton, 2005).
Primary response variable
Our dependent variable is the duration of an LSSR. Adopting the coding rule noted above, the dataset contains a total of 239 repression spells that produce a total of 2,527 observations (i.e. spell years) for the period between 1976 and 2006 (both completed by the last year and those still underway). A complete list can be found in the Online appendix. 13 In these data, the average duration of a spell is 11 years, while the minimum length is one year and the maximum is 32. Derived from our interest in LSSR termination, we focus on the probability (or hazard) that repression ends in a year, given that it has survived up to that time. 14
Explanatory variables of interest
Consistent with existing research, we include numerous explanatory variables/policies that are believed to impose costs on the repressive process.
Military interventions
To measure this variable, we use Kisangani & Pickering’s (2009) military interventions (IMI) dataset. Drawing on previous work, we focus on three types: pro-government, anti-government, and neutral interventions. We code a binary variable for each intervention type and lag these variables by one year in all models. In the data, there are 59 pro-government interventions (24%), 73 antigovernment interventions (30% of all spells), and 55 neutral interventions that occur during spells (23%).
Economic sanctions
To code this variable, we use monthly data from the Threat and Impositions of Sanctions (TIES) project (Morgan, Krustev & Bapat, 2006). As designed, we only include sanctions that are targeted at human rights violations and/or modifying contentious behavior taking place within nation-states. We create a binary variable, lagged one year, that equals 1 when a third party enacted economic sanctions on a government involved in a repression spell, and 0 otherwise. In the data, 32 spells involve economic sanctions (13%).
Naming and shaming
We use data from Lebovic & Voeten (2006) on resolutions by the United Nations Commission on Human Rights (UNHCR). As conceived, a binary variable equals 1 if the Commission condemned a state for engaging in state repression and 0 otherwise. This variable is lagged one year in the models presented below with the UNHCR adopting resolutions in 63 spells (26%).
International law
We include two treaty-based variables: the International Covenant on Civil and Political Rights (ICCPR) and the Convention Against Torture (CAT). Both measures are lagged one year and equal 1 if the state in question ratified the relevant treaty, and 0 otherwise (Hill & Jones, 2014). In our data, 116 (48%) and 163 (68%) spells have governments that have ratified the CAT and ICCPR, respectively.
Preferential trade agreements
For this variable, we include a measure that equals 1 if the state in question is a member of at least one PTA in a given year, and 0 otherwise (Spilker & Bohmelt, 2012). This variable is lagged one year in all the models presented below. In our data, we have 71 spells with at least one PTA (29%).
Foreign imposed regime change (FIRC)
In an effort to pay some attention to the different ways that countries undergo change, we use data from Downes & Monten (2013). Here, the authors define a FIRC as when a state forcibly removes the leader of another state and then makes concrete efforts to promote democracy in the targeted state over the next five years. Similar to our other measures, we employ a binary indicator that equals 1 if a FIRC occurs, and 0 otherwise. This variable is coded a 1 in 23 observations in the data (.5%).
Democratization
To measure this variable, we use an indicator of democracy from Przeworski et al. (2000) and most recently updated by Cheibub, Gandhi & Vreeland (2010). Within these data, a regime is classified as democratic if it meets four conditions: (1) the population or a popularly elected body select the executive; (2) the legislature is popularly elected; (3) at least two parties participate in the elections; and (4) there is alternation (i.e. an incumbent executive is replaced by new executive via elections) and the electoral rules must be identical to those that brought the incumbent to power in the first place. Following Cheibub, Gandhi & Vreeland (2010), the democratization variable equals 1 the year the regime changes from a dictatorship to a democracy plus five additional years as long as the state remains a democracy. There are 50 cases of democratization in our sample from 1976 to 2006 (20%). 15
Nonviolence
Drawing on newer interests, we also include a measure from the Nonviolent and Violent Campaigns and Outcomes (NAVCO) data project (Chenoweth & Stephen, 2011). This measure is a binary variable that equals 1 when a campaign primarily using nonviolent methods is ongoing, and 0 otherwise. 16 Consistent with our interests, we only include campaigns that target regime change or significant institutional reform (i.e. improve repression/human rights practices). There are 104 campaign-years in our sample (4%).
Control variables
Following standard practice, we employ several control variables in our study (e.g. Poe & Tate, 1994; Davenport, 1995, 2007b; Hill & Jones, 2014). It should be noted that these variables have never been employed to analyze repression spell termination because the topic has not been examined. It is possible therefore that the effects differ between units of analysis.
First, we include a measure of Economic development operationalized as GDP per capita (based on purchasing power parity) lagged one year (World Bank, 2018). In line with existing research, we expect higher values to be associated with shorter repressive spells as well-resourced governments should be quicker and more efficient at finishing what they set out to do.
We include a dummy variable for whether the state in question is involved in an Ongoing civil war. The variable uses the UCDP conflict data lagged one year and equals 1 when a state is in a war (based on the 25 battle-death threshold) and 0 otherwise (Themnér & Wallensteen, 2012). We anticipate that conflict will be associated with longer LSSR spells as behavioral challenges increase government efforts to suppress them and deal with the population. Related to this is a variable accounting for Civil war termination. Given the strong relationship between behavioral challenges and repression identified in the literature, this provides an important alternative explanation for spell termination; when behavioral challenges stop, repression should stop as well.
In line with the domestic democratic peace proposition (e.g. Davenport, 2007b), we include a measure for the Level of democracy as described by Cheibub, Gandhi & Vreeland (2010). We expect democracies will have shorter spells (if any at all) as these political systems have alternative means for resolving conflicts, as well as greater concerns for popular accountability. Relatedly, recent literature suggests that domestic rule of law as an agent of political democracy and in particular an Independent judiciary constrains governments. We therefore include the Linzer & Staton (2015) latent measure of judicial independence, lagged one year. 17
Finally, two demographic variables are also considered; we include the Logged population of the state (lagged one year) taken from the World Bank Development Indicators (World Bank, 2018). The expectation is that logged population should prolong state repression as it represents a general demographic threat. Second, drawing on Nordås & Davenport (2013), we include a measure of Youth bulges from Urdal (2006: 608) defined as ‘large cohorts in the ages 15–24 relative to the total adult population’. We expect large youth bulges to prolong repressive spells as they magnify the degree of potential and actual behavioral challenge. 18
Model specification
Different from the repressive variation approach but consistent with the logic of spell termination found in interstate and civil wars, we employ the semi-parametric Cox proportional hazards estimator with robust standard errors, clustered on the country to test our argument. 19 To facilitate the interpretation of the findings, we present our results in hazard ratios (exponentiated coefficients). Here, values above 1 increase the probability of the hazard occurring (i.e. spell termination), while values below 1 decrease the likelihood of a repressive spell ending. 20
To conduct our analysis, we estimate five equations: one baseline model that only includes the control variables and four models that account for different specifications. We use a p-value of .05 or lower as the standard to assess the statistical significance of the independent variables.
Empirical findings
Termination of high-repression spells
Based on Cox regression. Robust standard errors in parentheses, clustered on country. * p < .05.
In Table I, Model B, we explore the impact of diverse variables on the termination of LSSR that are believed to have an impact on state repression specifically because of their impact on costs, as well as the fact that previous studies have found an effect. The statistical results suggest that most of these variables have little systematic impact on the termination of large-scale state repressive behavior. For example, military intervention, naming/shaming, the international law variables, PTAs, and nonviolent direct action all fail to reach standard levels of significance. Economic sanctions prolong rather than shorten spells, which is consistent with some findings in the existing literature (Wood, 2008; Peksen, 2009) but generally against most expectations. Additionally, we find that FIRCs are associated with shorter spells. While this result is interesting with potentially important policy implications, it is worth noting that there are only two cases in the data where a FIRC led to spell termination (i.e. Grenada and Panama). Consequently, the finding should be interpreted with caution.
In Table I, Model C, we present the model with democratization, our variable of theoretical interest. As expected by our argument, the findings indicate that democratization increases the likelihood that an LSSR spell terminates. The hazard ratio is 2.112 and significant (p < .01), indicating that a repression spell is 112% more likely to end when a state has democratized within the last five years.
In Table I, Columns D and E, we present the full specification with the control variables along with the relevant factors generally found in a repressive model. Note that we present two models – one with and one without the UN naming/shaming variable due to missing data with this variable that reduces the total number of observations by almost 600. In Table I, Column D, we present the model with the naming/shaming variable included. The results are largely consistent with the specifications from above, although in this model independent judiciary and economic sanctions are no longer statistically significant. More importantly, the hazard ratio for democratization is greater than 1 and significant, indicating (again) that democratization is associated with shorter spells. Further, consistent with the results above, the variables that are most highlighted in discussions regarding the raising of costs with state repression have essentially no impact on the duration of repressive spells.
In Table I, Column E, we present the full model without the UN naming/shaming variable. Here, once more we find that democratization is statistically significant and associated with reducing LSSR spell duration. The control variables are largely consistent with the previous models. In line with the baseline model, independent judiciary returns to statistical significance, although sanctions remain insignificant.
As revealed, the empirical results are generally in line with our theoretical expectation that it takes a fundamental shift/perturbance to the political apparatus in order to stop governments from continuing to violate human rights on a large scale. While important and providing an important clue to how repression ends, we recognize that democratization is very difficult to achieve as a policy in its own right. Consequently, this makes changes in democracy worthy of attention so that the process by which LSSR spells are ended can be even clearer. To begin a discussion on this issue, we briefly examine the determinants of democratic transitions among highly repressive, autocratic regimes.
Clearly, such an investigation is necessary. Scholars generally disagree on the causes of democratization and existing research has found very few (if any) consistent predictors (e.g. Gassebner, Lamla & Vreeland, 2013; Ulfelder, 2013). Further, we are interested in a different set of cases than most work on the subject. For example, our interest is limited to highly repressive autocrats, while the existing literature studies democratization across all types of non-democratic regimes. Despite these challenges, we draw on the extant literature and our interest in spell termination to model democratic transitions in LSSR. In particular, we assess the determinants of democratic transitions using the same set of variables from the spell termination model plus three additional variables based on the democratization literature.
First, we add a democratic diffusion variable to the model, measured as the proportion of neighboring states that have democratized in the previous two years using the Cheibub, Gandhi & Vreeland (2010) data. Several reasons exist for this inclusion. There is the demonstration effect where actors learn from and emulate neighboring countries that have undergone successful transitions (e.g. Gleditsch & Ward 2006). Scholars also suggest that neighboring democracies provide more resources to democratic movements, while others argue the costs of ruling for autocrats increase as democracy becomes more common among neighbors (e.g. Celestino & Gleditsch, 2013).
Second, we add the number of previous democratic transitions in the country to the model, again using the Cheibub, Gandhi & Vreeland (2010) measure. Similar to the demonstration effect, domestic actors will be more confident that they can successfully democratize when the state has a history of doing so, leading domestic actors to be more likely to push for such transitions in the current period (e.g. Gassebner, Lamla & Vreeland, 2013).
Third, we include a variable that accounts for rainfall shocks in the model. The variable measures the annual deviation in precipitation from the ten-year moving average at the country-year level (Hendrix & Salehyan, 2012). 21 We take the absolute value of this variable because we expect both positive and negative shocks to be linked with democratization. This is done because several scholars find that climate shocks such as droughts or floods can lead to a window of opportunity for democratic transitions due to their impact on economic growth (Aidt & Leon, 2016). That is, climate shocks can lead to domestic dissent following economic downturns, which in turn, can lead some autocratic regimes to make democratic concessions to the opposition, leading to transitions.
In line with the existing literature on democratic transitions, we recode the response variable such that it equals 1 if the state democratized in the current year, and 0 otherwise. 22 Drawing on the same research, we also use dynamic probit to estimate our models in which we present the results for the sample estimated on repressive, autocratic regimes (Boix, 2003). This allows us to calculate the likelihood that, in a repressive spell, a government that started the year as an autocratic regime transitions to a democracy by the end of the year.
From these analyses, the three new variables are all statistically significant and in the expected direction (Table II). Most interestingly, we find that civil resistance is associated with a higher probability of democratization. These results are consistent with several other studies that find a relationship between nonviolent movements and democratization (Bratton & van de Walle, 1997; Chenoweth & Stephan, 2011; Celestino & Gledistch, 2013). Along with our perturbance argument, scholars suggest that civil resistance is successful because it leads to mass participation, destabilizing and undermining autocratic rulers and ultimately causing them to make concessions/movements towards democracy to offset the increased pressure on them (Chenoweth & Stephan, 2011; Celestino & Gledistch, 2013). 23 In contrast, we find little support for factors that generally dominate discussions about stopping repression, such as military interventions.
Determinants of democratization
Robust standard errors in parentheses, clustered on country. * p < .05.
Considering endogeneity
Related to the above discussion, some readers may be concerned that our results concerning democratization suffer from selection effects and that these findings are spurious. To address this, we use instrumental variable (IV) regression to account for the omitted variable problem that is the source of the potential endogeneity issue here. IV regression requires that a variable meet two assumptions 24 to be considered a valid instrument: (1) it must be correlated with democratization but (2) it must be uncorrelated with the error term in the primary model (and subsequently spell termination) conditional on the independent variables. While recognizing how difficult it is to find a valid instrument, we employ the rainfall shocks variable from above as an instrument. 25 As noted above, existing literature finds that climate shocks are linked with democratic transitions, satisfying the strength of instrument criteria. At the same time, this variable meets the exclusion restriction assumption because we see no theoretical reason to believe that rainfall is correlated with the error term (and thus spell termination) except through the independent variables.
As can be seen in Table III, the results suggest that democratization is associated with a greater likelihood of termination, even when accounting for the potential endogeneity issue. While we recognize the challenges that come with estimating IV regression and we should therefore interpret the results with some caution, they do suggest that the primary finding that democratization is associated with a greater likelihood of spell termination is robust to concerns related to selection effects.
Conclusion
Our research began with an interest in understanding what stops ongoing large-scale state repression (or LSSR). Despite immense humanitarian, political, academic, and popular interest in the topic, as well as similar research in the areas of terrorism, interstate and civil war, spells of the kind we are discussing have not previously been explored. To investigate this line of inquiry, we developed a theoretical argument where it was maintained that, once initiated, LSSR is not likely to change unless the decisionmakers and implementers of repressive behavior were affected by some factor that significantly raised costs, such as democratization. Without a cost that perturbed the repressive process, it is simply not expected that a spell would end.
Termination of high-repression spells, IV regression
Standard errors in parentheses. * p < .05. Time variables omitted.
In contrast to much research investigating variation in repression (i.e. movement up and down some scale), we find that many factors believed to impose a cost on the repressive process have little systematic impact on spells of large-scale state repression. This non-result may be surprising to many – especially since this deviates from much of the literature, popular perception, and common policy prescriptions put forward to address the problem. We find the result unsurprising, however, given: (1) the sticky nature of the repressive process once underway and (2) the general inadequacy of many policies to perturb those involved in state repression. From these findings, two broad implications emerge.
Implications for researchers
Influenced by the current study, scholars interested in understanding why/when LSSR ends should prioritize including democratization in their standard repertoire of solutions and independent variables. These findings reinforce the work of the ‘domestic democratic peace’ where the level of democracy and, to a much lesser extent, certain increases in the level of democracy have served as a (or ‘the’) pacifying influence on state repression. This effort should be expanded. Our research also suggests that we should not dismiss conventional approaches to internationally oriented humanitarianism, but we should think more creatively about exactly which, how, and when they matter. For example, they may interactively have an influence or perform in some delayed fashion. Additionally, although we began an assessment of selection effects above, it is clear that there should be more attention given to the issue of why large-scale state repression starts in the first place and which countries select into this group.
Implications for policymakers, advocates, and activists
It is further revealed from our work that those interested in stopping state repression should focus their efforts on facilitating democratization. Now, this is not to say that such an outcome is easy. Rather, it is to say that there are some rigorous empirical grounds for such a pursuit and that such a finding should be conducted as the merits of democratic promotion are debated. How does one facilitate democratization? We find that facilitating democratically oriented social movements might work but there may be other answers/pathways worthy of exploration. Addressing this topic is likely among the most meaningful subjects and controversies in political science (e.g. Møller & Skaaning, 2012). We hope that this article has facilitated and inspired the continued pursuit of this topic, as we find it linked to one of the largest concerns for humankind – ending large-scale state repression.
Footnotes
Replication data
Acknowledgements
We thank participants at the University of Michigan’s Conflict and Peace and Research and Development group, Michigan State University’s Workshop on Conflict, the International Relations Faculty Colloquium at Princeton University, the Political Science Department at Arizona State University, and Georgetown University’s International Theory and Research seminar for helpful comments and feedback.