Democracy and police violence

Abstract

Are police agencies less likely to use torture in democracies than in non-democracies? Much empirical research has shown that democracies are less likely to engage in torture in general, but most of this research does not distinguish among victim types or state agencies. Using the Ill-Treatment and Torture (ITT) Data, we focus on police agencies and evaluate whether they are less likely to use torture against (separately) political dissidents, criminals, and marginalized communities. Using logistic regressions with random effects, we find that the well-established and relatively high level of democracies’ respect for the rights of dissidents extends to police agencies as well. However, we find weaker statistical evidence that police agencies in democracies are less likely to use torture against criminals, and no evidence that they are less likely to torture marginalized communities. Our results suggest that one of the most robust empirical facts in the literature must be qualified. The protection from violence offered by democratic institutions does not seem to generalize beyond violence directly related to political competition and dissent.

Introduction

In February of 2015 the British newspaper The Guardian reported that the Chicago (USA) police department was using a warehouse on the west side of the city known as Homan Square as a black site to detain and interrogate criminal suspects. People detained at the facility, as well as attorneys who represent them, claim that police at Homan Square keep no written or digital record of who is taken inside, hold suspects without formal charges for long periods of time, and routinely deny detainees access to attorneys. Former detainees have alleged physical abuse including beating, prolonged shackling, and sexual assault, all of which amount to torture and cruel treatment under domestic and international law. At least one detainee has died in the facility. ¹ The existence of Homan Square is in tension with the picture painted by quantitative research on state violence. Practices like disappearance and torture are known to be more common in autocracies than they are in democracies. In fact, this is one of the most robust empirical facts in the literature (Davenport, 2007a,b; Hill and Jones, 2014; Poe and Tate, 1994; Poe et al., 1999). Nevertheless, police violence of this severity does not make the USA unique among democracies. In the Philippines, President Rodrigo Duterte’s continuing drug war has also resulted in gross, violent abuses of individual rights. Since coming to power in the summer of 2016, Duterte has implemented anti-drug policies that have resulted in thousands of extralegal killings by the national police. ² Duterte enjoys high public approval ratings according to the latest polls, though the recent killing of a high school student has caused some backlash against these policies. ³ As a final example consider Brazil, where police violence increased after democratic reforms in the 1980s (Caldeira and Holston, 1999). Brazil has recently received international attention for an escalating campaign of police violence in Rio de Janeiro during the lead-up to the 2016 Summer Olympics, which included torture and extralegal executions. Abuses in Rio were so common during this period that two human rights organizations labeled the city’s police force “the most violent in the world.” ⁴

While it is a fact that democracies use less violence, on average, than autocracies, it is also a fact that state agents in democracies use torture. In democracies, torture and other forms of violence can become widespread because of concerns about national security that lead the military and/or intelligence agencies to use torture during the interrogation of dissidents with (perceived) ties to violent groups, and to use other violent tactics like disappearance and extralegal execution (Conrad and Moore, 2010; Conrad et al., 2017; Conrad et al., 2018; Davenport, 2007b; Davenport et al., 2007; Rejali, 2007). Torture can also become endemic in democracies due to police violence like that described above, which does not target citizens due to their political activities and affiliations. This is another reason that police violence in the USA, the Philippines, and Brazil does not fit comfortably with much cross-national research. Much of this research theorizes exclusively about the repression of dissent, but not all state violence targets groups who organize to oppose the government. Much of it targets criminal suspects, immigrants, the homeless, and other groups who are not perceived as challenges to the government’s authority. The annual reports used to create indicators of state violence reflect this fact. Many of the abuses discussed in these reports are committed by the police and target vulnerable social groups beyond dissidents. However, the most commonly used indicators do not distinguish between violence that targets dissent and violence used for other purposes. It is possible that democracy’s apparent pacifying effects merely reflect the fact that more democratic countries are less likely to use violence in response to dissent, and that this relationship does not extend to violence used for other reasons. It is not possible to determine whether this is the case without distinguishing between violence that targets dissidents and violence against other groups.

There is a growing body of quantitative research that examines torture specifically and the extent to which democratic institutions reduce its occurrence (Conrad and Moore, 2010; Conradet al., 2017; Conrad et al., 2018; Moore and Armstrong, 2007). The recently released Ill-Treatment and Torture (ITT) Data (Conrad et al., 2013, 2014) can be disaggregated by both victim type (dissident, criminal, etc.) and agency responsible for custody (military, police, etc). We take advantage of this useful feature to explore the relationship between police violence and two institutions often associated with democracy: political competition, typically conceived as the essence of democracy (Schumpeter, 1950; Dahl, 1971), and judicial independence, which is viewed as an effective constraint on state abuse of individual rights (North and Weingast, 1989; Powell and Staton, 2009). We use several indicators of competition and judicial independence, and examine separately violence that targets dissidents, criminal suspects, and members of marginalized social groups. Our results suggest that competition and judicial constraints are associated with less police torture that targets dissidents. We find some evidence that these institutions discourage the torture of criminals, but no convincing evidence that either is associated with a lower probability that police torture members of marginalized groups.

Theoretical motivation

Explanations for state violence tend to focus on a particular kind of violence: the repression of dissent (Haschke, 2011). ⁵ Repression is usually defined as actual or threatened coercion intended to discourage or punish attempts to challenge or influence the government (Bissell et al., 1978; Davenport, 2007a; Goldstein, 1978; Stohl and Lopez, 1984; Tilly, 1978). This definition includes coercion that targets people who explicitly challenge the authority or policies of the government or are perceived to oppose its authority/policies. Potential victims of such violence include but are not limited to (suspected) members and supporters of particular political parties or other organized groups, people engaging in protests and demonstrations against the government, and (suspected) members and supporters of armed rebel groups. One reason studies consistently find that democracies are less violent than autocracies is that the concept of democracy, defined as genuine political competition, overlaps with repressive violence. According to this definition of democracy, using violence to target peaceful political opponents is sufficient to classify a state as non-democratic. That is, states must be considered less democratic if they target members and supporters of peaceful political groups with violence. Indices of democracy that measure political competition thus partly measure uses of violence against the government’s opponents, with those that use more violence being classified as less democratic. Results from studies analyzing the relationship between political competition and state violence reflect this fact, at least to some extent (Hill Jr, 2016).

Suppressing challenges to political authority is one of several reasons states use violence against people who live in their territorial jurisdiction. States also target people who are not perceived to be challenging their authority, but who are (suspected of) violating laws or social customs. Bissell et al. (1978) use the term “oppression” to refer to coercion unrelated to political challenges, and we adopt that label. This category of violence includes acts that target suspected criminals or people viewed as threats to social order simply by virtue of their placement in a particular social category. The connection between democracy and oppressive violence is more ambiguous. Violence that is not related to political challenges is easier to keep conceptually distinct from democracy than violence that targets peaceful opposition groups. Our main theoretical claim is that it is necessary to distinguish between repressive and oppressive violence when analyzing the relationship between violence and political competition, given the conceptual overlap between competition and repression.

Does political competition reduce oppressive violence? There are good reasons to doubt it does. The main reason competition is supposed to reduce state violence is that it gives the public the opportunity to remove abusive leaders from office (Poe and Tate, 1994). While this would seem to apply to oppressive violence, Moore (2010) argues that competition is a weaker constraint on state violence than institutions designed to guard the rights of the minority, like constitutional protections and courts of law. He draws on Walzer’s (1973) “dirty hands” argument, in which the public holds leaders accountable for their personal security and expects them to use abusive tactics like torture when they are deemed necessary, but to also publicly deny the use of such tactics. Moore argues that leaders may even be able to publicly acknowledge using torture and avoid electoral punishment. The fact that torture typically targets small, marginalized groups who are viewed by the general public as threatening (Rejali, 2007) is further reason to believe that leaders need not fear removal even if its use becomes publicly known (Conrad et al., 2018). Indeed, oppressive violence is used frequently in some democracies. In his study of torture in democracies, Rejali (2007) discusses police torture in the criminal justice system, arguing that it results from reliance on confessions rather than physical evidence in criminal trials and legal rules that allow police to prolong pre-arrest detention. He also discusses violence that targets people who are not necessarily suspected of any criminal violation, but have transgressed customs concerning social hierarchies based on class, ethnicity, or some other social distinction. Rejali (2007) refers to this as violence in pursuit of “civic discipline.” In the same vein as Walzer (1973), he describes an informal social arrangement between citizens and state agents where citizens demand “social order” and turn a blind eye to torture if they believe it accomplishes that goal.

Apart from the accountability created by competition, scholars have appealed to governmental checks on executive power, which often exist in democracies, as potential constraints on violent abuse (Davenport, 2007b). Chief among these is an independent/powerful judiciary, which in theory serves to prevent various forms of government predation, including excessive extraction/taxation and violations of civil and political rights (North and Weingast, 1989). Recent empirical research has established a strong connection between judicial power and state violence, including torture (Hill and Jones, 2014; Keith, 2002; Keith et al., 2009; Mitchell et al., 2013; Powell and Staton, 2009). Unlike the effect of competition, the negative effect of judicial power on violence cannot be attributed to conceptual overlap with repression. Effective courts of law can prevent violent abuse because citizens are more likely to allege violations when they expect that courts will actually address them. Judicial institutions can therefore increase state agents’ expectation of civil or criminal penalties for torture (Powell and Staton, 2009). ⁶ In contrast with political competition, the punishment meted out by courts does not depend as directly on public opinion regarding the desirability or effectiveness of torture, and so may be a more effective constraint (Moore, 2010).

Our purpose is to establish a set of baseline empirical facts rather than develop a theory of oppressive violence. In the remainder of the paper we examine the relationship between domestic institutions and police violence. On the institutional side, we examine separately political competition and judicial constraints. On the state violence side, we distinguish between repressive and oppressive violence. Many existing analyses separate the various domestic institutions associated with democracy, but very few distinguish between violence that targets dissidents and violence that does not. We are interested in knowing whether previous findings concerning competition/judicial power and state violence hold for indicators of oppressive violence.

Data and models

To examine torture committed by specific government actors against specific types of victims, we use the ITT country-year data (Conrad et al., 2013), which contains indicators created from content analyses of Amnesty International’s annual reports, press releases, and Action Alerts. The data distinguish between both the state agency accused of abuse and the identity of the victim. Agencies include the police, the military, the paramilitary, prison guards, intelligence services, and immigration detention centers. We use only indicators for which the state agency is designated as “police.” ⁷ Victim types examined here include criminal, dissident, and marginalized individuals, which includes religious and ethnic groups, youths, the elderly, and immigrants (Conrad et al., 2013: 203). ⁸ This allows us to examine police violence against distinct victim types. We focus exclusively on the police because this agency is the most useful for comparing results for repressive and oppressive violence. This is because the distribution of victim types is more even for the police than for other government agencies. According to the ITT data, the police are the most likely agency to be accused of torturing dissidents and marginalized groups, and are the second most likely (after prison staff) to be accused of torturing criminals. Intelligence services and the military are unlikely to have access to criminal suspects, resulting in few allegations of torture targeting this group, while prison staff are relatively unlikely to be accused of torturing members of marginalized groups.

ITT’s torture variables are ordinal scales ranging from 0 to 5, with higher values indicating allegations of more systematic/widespread torture. It should be noted that ITT’s country-year data record allegations of torture that occur throughout the entire country. This results in a preponderance of zeroes (no alleged torture) and few observations in higher categories. Because of this, we transform these ordinal measures into dichotomous ones. ⁹ In the online appendix we show results for ordered response and linear models. Our substantive results do not change.

In the first set of models, we examine the effect of competition on police torture against political dissidents. Violence against dissident groups is commonly the focus of human rights studies, and in these models we seek to establish whether the negative association between competition and the repression of dissidents extends to police torture that targets other groups. In the second set of models, we examine whether judicial constraints have an effect on police torture. We use three indicators of competition in our analyses, and three indicators of judicial independence/effectiveness. Using several variables helps to protect against one dataset driving our results, and using two categories of measures—political competition and executive constraints—allows us to analyze two distinct features of democracy.

For the competition measures, we first use the Polyarchy variable from the Varieties of Democracy (V Dem) data. Polyarchy is a continuous variable ranging from 0 to 1 and is calculated from a weighted average of measures of freedom of expression, freedom of association, fair elections, an elected executive, and suffrage, and from an interaction between all of these variables (Coppedge et al., 2017). Next, we use Parcomp from the Polity IV data, which is an ordinal measure ranging from 0 to 5, with 5 indicating that changes in governmental policies can be relatively safely pursued. Finally, we use the dichotomous ACLP variable (Alvarez et al., 1996) as recoded and extended by Cheibub et al. (2010). The authors create a dichotomous indicator for whether a country is democratic based on political contestation for executive and legislative offices. Countries that have two or more parties, an elected executive, and an elected legislature (without reverting to non-competitive political practices) are classified as democracies.

For the executive constraints measures, we first use the V Dem measure of judicial constraints, which is a continuous 0-1 scale that is calculated via Bayesian factor analysis of five indicators: executive respect of the constitution, compliance with the judiciary, compliance with the high court, the independence of the high court, and the independence of lower courts (Coppedge et al., 2017). Second, we use the CIRI indicator for judicial independence, which is an ordinal variable ranging from 0 to 2, with higher values indicating a more independent judiciary (Cingranelli et al., 2014). Finally, we use the Polity IV measure of executive constraints, an ordinal measure ranging 1 to 7, with higher levels indicating more institutional constraints on executive decision making, including an effective high court (Marshall and Jaggers, 2009).

We use logit models with random effects for country to account for the panel structure of our data. Including these unit-specific effects accounts for unspecified heterogeneity across countries and allows us to draw more reliable inferences. The data cover 147 countries for the years 1995 to 2005 (inclusive). We conduct separate analyses for the three types of victim groups: marginalized communities, dissidents, and criminals. Separate models are also used for each indicator of competition and judicial constraints, for a total of 18 models.

In addition to measures of democracy, in each model we include (the natural logs of) GDP/capita and population size from the World Bank. We also include ITT’s measure of restricted access, as suggested in the user’s guide (Conrad and Moore, 2011: 14). This dichotomous variable, Restricted Access, is coded 1 for years in which Amnesty International reports that it, or another INGO, had trouble obtaining access to a detention center.

Results

Coefficient estimates

Figure 1 shows the results from our models examining the effects of political competition on police torture. ¹⁰ Point estimates of the coefficients are displayed with associated 90% confidence intervals. ¹¹ Coefficient values are displayed along the horizontal axis.

Figure 1.

Political competition coefficients from models of police torture, by victim type.

Our results suggest a negative association between democracy and the torture of dissidents. All three measures of democracy are associated with a statistically significant decrease in the likelihood that police agencies use torture to any extent. The results of these models suggest that the relationship between democracy and the torture of political dissidents extends to police agencies.

When we look at other victim types, however, we find weaker evidence that democracies have higher respect than autocracies, on average, for the human rights of these victim types. Although our analyses of police–criminal torture shows a negative relationship between democracy and torture, the estimates are significant at the $\alpha =0.10$ level in one of the three models. Polyarchy has a negative and significant relationship with the probability of torture by police agencies, while Parcomp and ACLP show negative but insignificant relationships with police torture against criminals.

With respect to the police–marginalized communities analyses, none of the measures of democracy have a statistically significant association with torture by police agencies, all of the estimates are close to zero, and one is actually positive. Given the usual finding that democracies are less likely to engage in torture than non-democracies, these null results are interesting. When we look beyond political dissidents, we find weaker evidence that police agencies in democracies are less likely to use torture.

Figure 2 shows the results from our models examining the effect of executive constraints on democracies. As with political competition, all three measures of judicial constraints show a negative and significant relationship with police torture of political dissidents.

Figure 2.

Executive Constraint coefficients from models of police torture, by victim type.

In the police–criminal models, we find mixed results. Judicial constraints and executive constraints have a negative and significant relationship with the likelihood that police agencies torture, while CIRI is insignificant. Across all six models of police–criminal torture, half of the measures—parcomp, ACLP, and CIRI—produce no significant effect on the likelihood of police torture.

Finally, in the police–marginalized communities models, we again cannot reject the null hypothesis that democracy has no effect on the use of torture by police. Surprisingly, all three models return positive coefficients, although judicial constraints and executive constraints are close to zero. These results mimic the results of the police–marginalized communities models with the competition measures; across all six police–marginalized models, we fail to find statistical significance with respect to democracy and the use of torture.

Further evaluation of null results

In order to more thoroughly evaluate our null results, we perform the confidence interval version of the two one-sided test (TOST) proposed by Rainey (2014). Rainey argues that demonstrating a negligible effect requires more than statistically insignificant results. Rather, researchers should demonstrate that the estimated sampling distribution of their quantity of interest does not contain values that would constitute a meaningful effect. This requires that we choose a threshold for a “meaningful effect.” Since the ITT data have not been extensively analyzed, it is unclear how to define a meaningful effect, so we proceed as follows: for each model that produced a null result we simulate 10 000 draws from a multivariate normal distribution using the estimated coefficients and covariance matrices. We calculate a baseline probability by setting the values of the control variables to their means (the natural logs of GDP per capita and population) or modes (restricted access), and setting each competition or judicial constraints variable at a typical value. ¹² This resulted in 10 000 predicted probabilities for each model, and we used the mean probability as the baseline. We then changed the value of each variable of interest to its maximum and calculated the mean change in the probability from the baseline, with a 90% confidence interval. The lower bound of each confidence interval serves as the maximum plausible reduction in the probability of torture that results from changing each indicator from a typical to a high value.

Tables 1 and 2 show, for each model, the baseline probability of torture, the maximum plausible reduction, and this reduction expressed as a percentage of the baseline. We calculate percentage reductions since the baseline probability is different in each case, and in some cases is low enough to preclude a large absolute reduction. The ITT country-year data measures allegations of torture that implicate state agents across the entire country, which are rare, so for all models the baseline probability is low. The larger the absolute values in columns 2 and 3 of each table, the larger the range of threshold values for which we can not rule out a meaningful effect. Thus the evidence for a negligible effect is stronger for marginalized groups than for criminals.

Table 1.

TOST for Competition Variables.

Competition variable	Baseline probability	Max. reduction	% reduction
Criminals
Parcomp	0.067	–0.022	33
ACLP	0.097	–0.080	82
Marginalized groups
Polyarchy	0.031	–0.013	42
Parcomp	0.028	–0.009	32
ACLP	0.033	–0.025	76

Table 2.

TOST for Judicial Variables.

Judiciary variable	Baseline probability	Max. reduction	% reduction
Criminals
CIRI	0.077	–0.038	49
Marginalized groups
CIRI	0.030	–0.004	13
Xconst	0.031	–0.007	23
Jud. Con.	0.031	–0.011	35

Conclusion

There is a broad consensus among political scientists that democratic political competition and governmental constraints on the executive reduce state violence. The analyses on which this consensus is based do not typically distinguish between violence that targets opposition groups and violence that does not. This makes it impossible to determine whether the pacifying effects of these institutions apply equally to the distinct groups that are the most likely targets of violence. Further, since democratic competition necessarily entails a limited amount of violence against peaceful dissidents, it is imperative to draw a distinction between repressive and oppressive violence when conducting such an analysis. We address this issue by employing the recently released ITT country-year data (Conrad et al., 2013), which allows analysts to disaggregate allegations of state violence by the identity of the victim and the agency accused of abuse. With respect to violence perpetrated by the police, we find strong evidence that competition and constraints are negatively associated with torture that targets dissidents. This is consistent with the body of research on which we build, and is unsurprising. We find only mixed evidence that these institutions help reduce torture that targets criminals. Finally, we find no compelling evidence that competition or judicial constraints decrease the use of torture against marginalized groups. In the case of judicial institutions, this is somewhat surprising, as we expected these institutions would prove effective at deterring oppressive violence. Our results suggest that one of the most robust findings in the literature must be qualified: some vulnerable groups do not enjoy the protection created by political competition and strong courts of law. If recent episodes of police violence in democracies are aberrant, it is only because police violence does not typically rise to the level seen in these cases, not because such violence is occurring in democracies.