In Darfur approximately 2 million people have been displaced and at least 180,000 have died, reportedly due to the actions of Sudanese government-backed Janjaweed militia. In the Ivory Coast, Laurent Gbagbo relied on the violence of the Young Patriots to stay in power. In the former Yugoslavia pro-government militias like the Red Berets or Arkan’s Tigers influenced the character of the conflict. In Afghanistan, the coalition that defeated the Taliban included Afghan militias, and the Awakening Groups were used to counter the insurgency in Iraq. While the importance of these pro-government armed groups is recognized in particular contexts, little is known about their general impact on the respect for human rights.

How does the presence of certain actors affect the risk of physical harm to civilians? Does the type of actor increase the risk of repression beyond structural factors that are linked to a deterioration of human rights? Militias are generally linked to state failure and violence (for example, Bates 2008; Mason and Krane 1989; Reno 2002), and the case literature highlights the extreme impact that these groups can have on human rights in specific countries (for example, Álvarez 2006; Campbell and Brenner 2002; Kirschke 2000). From 1982 to 2007, governments in 61 countries had links to an informal armed group within their own country (Carey, Mitchell, and Lowe 2013).

Given the prevalence of these groups, we seek to explain whether and how they affect respect for human rights.¹ It is plausible that the presence of these informal armed groups raise control and accountability problems that will contribute to higher levels of repression. We use the definition of a pro-government militia (PGM) from Carey et al. (2013:250), who define it “as a group that (1) is identified as pro-government or sponsored by the government (national or subnational), (2) is identified as not being part of the regular security forces, (3) is armed, and (4) has some level of organization.” These groups can be differentiated by their acknowledged proximity to the state. Some groups have a recognized and semiofficial link to the state, as with village defense committees in India for example. With others the link is looser and more informal, as with the pro-government militias in Darfur. The nature of the group’s relationship to the government has theoretical importance. It affects the nature of the agency problem and has consequences for the level of violence. We develop a theoretical argument that helps us understand why these groups increase the risk of harm for civilians and test it empirically using evidence from global data that span over 25 years. Our argument modifies and extends conventional principal-agent logic. The principal-agent approach draws attention to how the principal controls a self-interested agent in possession of superior information. The familiar control problems attributable to delegation are likely to be particularly severe when governments use informal armed groups that operate outside the official chain of command. Pro-government militias bring private interests to the security task, which are more difficult to monitor and check. The use of these often ill-trained and poorly monitored groups is likely to produce more opportunistic violence and contribute to higher levels of human rights violations. In addition to these simple agency problems, governments may take advantage of these informal armed groups in order to shift responsibility for the violence they commit (Fiorina 1985; Mitchell 2004, 2012). Assuming governments take advantage of the presence of these groups to shift responsibility for violence, then human rights violations are likely to increase. But not all rights are as easily violated by individual agents. The capability of individual agents to commit violations varies, as does the benefit the agent may extract from the violation. We expect that “agent-centered” violations, where the agent has the capacity to influence violations and where they might bring some personal benefit to the perpetrator, are most likely to be affected by the presence of PGMs (Bohara, Mitchell, Nepal, and Raheem 2008; Butler, Gluch, and Mitchell 2007). We test these ideas empirically on a global dataset for the years 1982 to 2007. The empirical results provide support for our theoretical arguments, highlighting the importance of the role of agents in understanding why human rights are violated.

In the following section we build upon existing studies that link militias with increased violence and the human rights literature that emphasizes the importance of accountability for protecting these rights. Based on these insights, we develop the theoretical argument to explain why pro-government militias are likely to increase the risk of human rights violations. We seek to examine how human rights are affected when these groups are present. While not our focus here, if these groups have important human rights consequences, then empirical investigation of the incentives for governments to use these groups has added significance.

REPRESSION AND THE ROLE OF ACCOUNTABILITY

Existing research has linked the variation in repression to institutional, economic, and demographic structures, to the level of threat posed by the opposition, and, through democracy, to accountability and the likelihood of withdrawal of public support (Cingranelli and Richards 1999; Davenport 2007; Davenport and Armstrong 2004; Gartner and Regan 1996; Landman 2005; Poe and Tate 1994). Bueno de Mesquita, Downs, Smith, and Cherif (2005:439) emphasize accountability in explaining the greater respect for human rights shown by democracies: “accountability appears to be the critical feature that makes full-fledged democracies respect human rights; limited accountability generally retards improvements in human rights.” Cingranelli and Filippov (2010) examine how accountability is “individualized” and how electoral rules affect politicians’ incentives to monitor human rights protection. Conrad and Moore (2010) show that democratic institutions, which demand accountability, decrease the likelihood of torture when the threat from violent dissent disappears. While governments in democracies face internal accountability mechanisms, less democratic regimes attract monitoring and pressure from international institutions and NGOs (Kirschke 2000; Landman 2005; Risse, Ropp, and Sikkink 1999; Simmons 2009).² If governments face the prospect of accountability internally and externally from sanctions from the international community, they may choose to avoid repression. DeMeritt (2014) suggests that principals are less likely to order killings and agents are less likely to carry out these orders when international actors intervene and might hold agents and principals to account for their actions.

This insightful analysis directs attention to a decisive moment when the international community can intervene to incentivize agents “to shirk the order to kill” (DeMeritt 2014:9). The implication of this research is that accountability leads governments to forego repression. Yet rather than forego repression, an alternative for government officials is to seek to evade accountability for this violence. Evasion of accountability can take a variety of forms, including interference with monitoring activities of NGOs and the media, the use of alternative types of repression, such as disappearance or encounter killings, delegating repression to other states, as with the policy of rendition, and to other actors, which is the focus of our study. Governments have the opportunity to reduce accountability for human rights violations when informal armed groups carry out security tasks. Questions of how the agent influences the level and type of human rights violations have received less attention (Ahram 2011a, 2011b; Lyall 2010), partly because of the absence of cross-national data on agents.³

Case and comparative studies have linked militias with increased violence toward civilians. Clayton and Thomson (2014) show that the use of militias by coalition forces in Iraq increase Iraqi civilian deaths. Focusing on Arkan’s Tigers in Serbia and the Rwandan Interahamwe, Álvarez (2006) stresses the importance of plausible deniability in explaining the use of paramilitary groups in the context of genocide. Kirschke (2000) highlights the usefulness of these groups in implementing repression when countries are pressured by the international community to democratize. Staniland (2012) discusses the violent counterterror activities of former rebels operating on the government side in Kashmir. Analyzing the development and demise of mercenaries and pirates, Thomson (1994:42) argues that the authorizing states are unable “to resist the temptation to allow or even authorize nonstate violence while they denied responsibility and accountability for its consequences.” She refers to the Nicaraguan contras and says “exploiting nonstate violence remains a powerful temptation for state rulers” though “contemporary state leaders must do this in secret” (Thomson 1994:152). Only a case treatment and lengthy legal proceedings (for example, Milosevic ICTY) can establish responsibility for the actions of even quite visible militia groups (for example, Arkan’s Tigers) and whether it is a simple agency problem, a case of can’t control, or a case of won’t control (Mitchell 2004). But systematic analysis can show whether the presence of pro-government militias is linked with an increase in repression beyond the factors that are generally associated with an increase in human rights violations in the quantitative literature. In the following section, we extend the principal-agent model to explain why PGMs are likely to increase the risk of harm to civilians.

AGENCY PROBLEMS AND THE RISK OF REPRESSION

Delegation is advantageous to a principal who lacks the time or skill for a task. In political science, the principal-agent model has been applied to areas such as the relationship between elected officials and the bureaucracy (for example, Wood and Waterman 1991), to accountability in democracies (Strøm, Müller, and Bergman 2006), to international organizations (Hawkins, Lake, Nielson, and Tierney 2006), to rebel recruitment (Gates 2002), humanitarian interventions (Kuperman 2008; Rauchhaus 2009), and to the sponsoring of foreign rebel groups (Bapat 2012; Salehyan 2010). Our focus is the implementation of repression (see also Cingranelli and Filippov 2010; DeMeritt 2014). Across these applications, the theory balances the conventional benefit of delegation, not having to do the task oneself, against the difficulties and problems associated with the relationship between the principal and the agents. Problems for the principal arise from the information asymmetry and goal variance that characterize the relationship with the agent (Miller 2005; Mitnick 1980; Waterman, Rouse, and Wright 2004). The agent has private information not shared with the principal and possibly different goals to those of the principal. Agents may shirk rather than work, or they may secure some other private benefit that was not part of the principal’s intention (Brehm and Gates 1997). As Laffont and Martimort (2002:2) point out, if there was no information asymmetry, “the principal could propose a contract that perfectly controls the agent.”

We expect that agency problems pose a heightened risk for the security and safety of citizens. The conventional responses to private goods-seeking agents are selection, monitoring, and holding agents accountable (Banks and Weingast 1992; Policzer 2004). To minimize costly monitoring, attention is paid to screening agents, training, compensation, and to professional values (Arrow 1985; Morrow 2007). With respect to militias, these mechanisms are likely to be absent or implemented less carefully (Álvarez 2006:5) than they are in the formal security sector. The informal organizations are likely to recruit former rebels, strategic extremists, or those interested in private gain, such as criminals, football hooligans (Arkan’s Tigers), or others with their own motivations for committing violence. For example, the Janjaweed militia in Darfur is described as a “mix of bandits and camel herders” (Rice 2006). The Chimeras in Haiti associated with President Aristide were recruited from street gangs (Adams 2002). In Turkey, “the Turkish security apparatus developed a relationship with criminals, with official sanction. The criminals were reportedly assigned to perform killings and to carry out other attacks or what were seen as counterterrorist operations” (Kinzer 1996). Even assuming that the leaders of these groups aim to inculcate a code of conduct, these informal organizations are likely to be less equipped for that task than state agencies. Members of informal armed groups are less likely to have internalized systems of ethics, less likely to value their positions, and are less likely to have Arrow’s “reputational enforcements” (Arrow 1985). In short, there is both case evidence and a strong and widely accepted theoretical argument drawn from the analysis of delegation for the expectation that the use of these groups will increase the likelihood of some types of human rights violations.

Pro-government militia groups vary with respect to how formally they are linked to the official security apparatus. While governments will deny any links to some groups for as long as possible, such as in the case of the Spanish Anti-Terrorist Liberation Group GAL in the 1980s, in other cases this link is more formal and institutionalized, as with the Village Defence Committees in India. While both types of militias are more likely than regular security forces to experience the agency problems outlined earlier, armed groups that are separate from the regular security forces but have formalized and official links to governments are better equipped to provide “reputational enforcements” than informal militias. Opportunistic behavior is more likely in informal groups. We expect that informal groups with no formal or official link to the government will have greater recruitment and operational discretion and less monitoring—and consequently a higher risk of agency problems and therefore more substantially affect human rights conditions.

TURNING AGENCY PROBLEMS INTO AGENCY ADVANTAGES

The thrust of principal-agent logic is to identify the control problems confronting principals. It suggests you cannot trust the agent. But this logic also offers opportunities to unscrupulous principals as well as to agents. The principal may knowingly recruit those with a reputation for violence (for example, criminals) and then refuse to control these agents—rather than actually lose control over them (Mitchell 2004; Poth and Selck 2009).⁴ There may be circumstances when the principal derives a strategic benefit from what would appear to be the hidden actions of the agent, where the principal can exploit the distance to the agent. In his discussion of Schelling’s work on nuclear deterrence, Miller discovers a policy area where an agent who has opposing goals to the principal is a bargaining asset (Miller 2005; Schelling 1960). For nuclear deterrence, Schelling sees the benefit of an agent (Curtis “Boom-boom” LeMay) who, contrary to the principal (President Kennedy), prefers “mutual annihilation to acquiescence” and who is “out of the principal’s control” (Miller 2005:219). In other policy areas, political scientists highlight the political, blame-shifting benefits of delegation. Fiorina observed a political incentive to delegate: “By charging an agency with implementation (…) legislators not only save themselves the time and trouble of making specific decisions, they also avoid or at least disguise their responsibility for the consequences of the decisions” (Fiorina 1985:187). Miller identifies this as a modification of the traditional principal-agent model, in which only the agent can exhibit moral hazard. Miller argues that under certain circumstances the moral hazard argument can also apply to the principal. Agencies can be created to provide the principal with the opportunity to avoid responsibility, as was done in the United States in 1988 and 1990, when Congress delegated authority to a base-closing commission in order to avoid responsibility for this unpopular action (Miller 2005). The recognition of the political advantages of delegation is an important extension of the principal-agent argument beyond its more conventional economic and administrative implications of saving the principal time and trouble.

In the context of repression, the possibility to shift responsibility for violent acts becomes particularly important. Governments perceive the use of violence as a tool to stay in power and to fight internal threats (for example, Poe 2004). At the same time, governments that are seen to violate basic human rights of their citizens may expect to be punished for this by their domestic but also by the international community—for example, in the form of naming and shaming, withheld foreign aid, reduced trade relationships, and exclusion from international organizations. Militias, and in particular informal militias rather than semiofficial militias, put “daylight” (Fiorina 1985:187) between the government and the implementation of the policy of repression. Beyond Fiorina’s shifting responsibility to another agency, we argue that “legislators,” when they use militias to apply repression, have the opportunity to not only shift responsibility to another agency but also onto the private interests or goals of the militia members, recognizing that they might have personal motivations for committing violence, and to claim a “runaway bureaucracy.” In this context, claims of information asymmetry and goal variance become an asset rather than a problem, as leaders can address threats with violence while shifting the blame to the agents.⁵ Similarly, it is a move beyond Schelling’s President Kennedy and the “runaway general” example. In this example, both principal and agent were motivated by varying conceptions of strategic interest. Militias present a moral hazard for the principal in offering privately motivated violence at reduced risk and from which a principal may desire to derive strategic benefit. As a tool to evade accountability, militias provide a response to threats a government perceives itself to be under, yet these informal armed groups are not part of the state bureaucracy, and their strategically useful violence is attributable to goal variance resulting from the militia members’ private motives rather than to the government itself. We expect that the ability to evade accountability as a result of a government collaborating with militias increases the risk to civilians; governments are assumed to be more likely to use a violent strategy if they expect to avoid being held responsible for it. If challenged by other state or nonstate actors, they can claim a simple agency problem and lack of control, passing the costs to the “out of control” or “bad apple” agents. If this argument holds, we expect to find higher levels of violations where these groups operate.

The risk that militias pose to civilians is therefore not only due to simple agency loss. Governments have the temptation to exploit the appearance of information asymmetry: the problem of artificial information asymmetry (Mitchell 2004). The agents follow private goals, such as violence for its own sake or revenge, while the principal may use militias in pursuit of strategic goals, such as obtaining information, eliminating opposition groups, or precipitating population flight (ethnic cleansing). In short, we expect that the more “daylight” there is between the principal and the agent, the more difficult the control problems will be and the more likely governments are to shift responsibility to these groups.

The implication of our argument is that in addition to human rights violations initiated by governments, and those individually initiated by out-of-control agents, there are jointly initiated and jointly beneficial violations where the government refuses rather than loses control of the agent. The risk of such additional violations is greater with informal militias where the agent is more loosely connected to the principal. This modification recognizes that the goals of principal and agent may vary. The former may have a strategic interest in violence and the latter a private enjoyment of violence. But although these goals are different, they are not necessarily conflicting or opposing, as the conventional principal-agent model suggests. In the area of repression, the private interest of the agent in violating a prisoner’s or civilian’s rights may coincide, not conflict, with the principal’s strategic interests, perhaps in extracting information or displacing unwanted populations or even in motivating or rewarding the agents (Mitchell 2012).⁶ A government will tolerate these informal armed groups as long as the violence they commit is consistent with strategic goals and as long as the government is not held accountable for them. Accountability is evaded by the militia’s organizational separation from the regular security forces and the claim of an agency problem.⁷ In short, where militias operate, there are likely to be higher levels of human rights violations. This argument is summarized in the following hypothesis:

We expect that not all types of human rights violations are equally affected by the presence of pro-government militias. As argued earlier, some violations are more “agent-centered” than others. To be agent-centered, the violation must be within an agent’s control and something from which an agent might extract individualized benefits. Torture or extrajudicial killing may satisfy some immediate desire for revenge in a way that imprisonment does not. Imprisonment may require the cooperation of other actors, such as the courts or other official government agents. In contrast, torture may be initiated by the agents themselves, by orders from principals, or by some combination of the two. We expect that torture, killings, and disappearances increase when informal militias are present. We expect that political imprisonment is likely to be unaffected by informal militias.

OPERATIONALIZATION AND RESEARCH DESIGN

Our key variable, Informal pro-government militias (PGMs), is coded 1 for all country-years with at least one informal pro-government militia and 0 otherwise. In cases where we do not have specific information on termination dates, we code the last year in which PGM activities were reported as termination. Coding the time between the first and last recorded year of activity compared to coding PGMs as present for only those years for which we have specific sources has the advantage of minimizing bias toward confirming our theoretical argument as we include years of inactivity or unreported PGM activity. Using this adjusted measure of PGMs is a tougher test for our theory than correlating only country-years with documented PGM activities with human rights. Fifteen percent of our observations are coded as having informal pro-government militias. We also control for the presence of semiofficial PGMs, which have “a recognized legal or semiofficial status, in contrast to the loose affiliation of informal PGMs” (Carey et al. 2013:251). Semiofficial PGMs were present in 22% of observations. In comparison to informal PGMs, we expect that semiofficial PGMs are less likely to increase the risk of agent-centered violations, because those groups are more closely monitored by governments and provide fewer opportunities for plausible deniability for violent acts they might commit.⁸

Operationalizing the Dependent Variable: Respect for Human Rights

To test our argument that informal PGMs increase the risk of violations, we measure human rights with the 9-point cumulative Physical Integrity Rights (CIRI) Index (Cingranelli and Richards 2010). This CIRI Index ranges from 0 (no respect for physical integrity rights) to 8 (full respect for these rights) and is based on human rights reports by Amnesty International and US State Department Country Reports. The advantage of the CIRI data is that they allow us to disaggregate types of human rights violations and that these data explicitly include violations by both the formal security sector and “private groups” acting on the government’s behalf.⁹ Based on Hypothesis 1, we expect that our variable Informal pro-government militia has a negative impact on the CIRI Index.

Hypothesis 2 holds that informal PGMs have a particularly damaging effect on human rights where violating those rights might bring personal benefits to their members beyond any expected strategic benefits and where the violation is within the agent’s individual capability to commit. To evaluate this argument we divide the Physical Integrity Index into two human rights measures. We differentiate violations that are more easily within the scope of individual agents and that may yield some personal benefit from those violations that likely require the cooperation of other government agencies. As agent-centered violations we identify torture, disappearances, and killings, but exclude political imprisonment.¹⁰ By adding the individual CIRI scales for torture, disappearances, and killings, each ranging from 0 (no respect for these rights) to 2 (full respect for these rights), we create the variable Agent-centered violations, which ranges from 0 (no respect for torture, disappearances, and killings) to 6 (full respect of these three human rights). We also use the CIRI variable Political imprisonment, which ranges from 0 (no respect of the right to be free from political imprisonment) to 2 (full respect of this human rights). Based on Hypothesis 2, we expect that informal PGMs have a negative impact on Agent-centered violations. According to Hypothesis 3, we expect PGMs to not have a significant effect on Political imprisonment because it is a less agent-centered violation.

Figure 1 shows the presence of informal pro-government militias at different levels of agent-centered violations, where higher values indicate a higher respect for the right not to be tortured, killed, and disappeared. Each bar indicates the percentage of observations with PGMs for each level of agent-centered violations. The bar in the category “0” stands for no respect of these rights, category “6” for full respect. Figure 1 points to a clear trend: the lower the respect for human rights, the more common were pro-government militias. For example, PGMs were present in 0.4% of cases that were coded as having full respect of human rights (value 6 on the human rights measure), but in 70% of those cases that were coded as having no respect (value 0 on the agent-centered violations). This figure provides initial support for our theoretical arguments that pro-government militias are associated with a higher risk of repression.

The Impact of Pro-Government Militias on Human Rights Violations

All authors

Neil J. Mitchell, Sabine C. Carey & Christopher K. Butler

https://doi.org/10.1080/03050629.2014.932783

Published online:

08 October 2014

FIGURE 1 Informal pro-government militias and agent-centered violations, 1982–2007.

Display full size

FIGURE 1 Informal pro-government militias and agent-centered violations, 1982–2007.

Control Variables

To test the impact of pro-government militias on human rights, we add the PGM measure described earlier to a standard model of human rights violations, which contains the following control variables: Armed conflict, Democracy, Level of economic development, Population size, and Lagged repression (for example, Cingranelli and Filippov 2010; Davenport and Armstrong 2004; Landman 2005). With armed conflict governments face large-scale and violent opposition, to which they often respond with violations of the right to physical integrity (for example, Davenport 2007; Harff 2003; Krain 1997; Poe, Tate, and Camp Keith 1999). We operationalize Armed conflict with the Uppsala/PRIO Armed Conflict Dataset, using the 25 battle-death threshold to identify such types of conflict (ACD; Gleditsch, Wallensteen, Eriksson, Sollenberg, and Strand 2002). We expect Armed conflict to have a negative impact on respect for human rights.

Controlling for armed conflict enables us to test the argument that PGMs have an effect on repression independent of the impact of armed conflict. One might argue that a link between repression and PGMs is spurious, due to the impact of civil war on human rights. It is reasonable to assume that PGMs are more common during civil wars, and civil wars are one of the key determinants of repression. While 50% of observations with an armed conflict also have informal PGMs, 45% of observations with these groups occur outside of armed conflict.

Our second control variable captures institutionalized democracies, another key determinant of levels of human rights violations. The variable Democracy is based on the Polity IV scale (Jaggers and Gurr 1995), where all country-years with the value of 8 or higher are coded as 1 and 0 otherwise (Bueno de Mesquita et al. 2005; Davenport and Armstrong 2004). In Figure 2 we show how informal PGMs are distributed across different degrees of democracy, using the Polity2 scale. For each level of the Polity2 scale, which ranges from –10 (fully authoritarian) to +10 (fully democratic), we plot the percentages of observations with informal PGMs. The graph shows that there are no PGMs in the most autocratic regimes, and only six (0.8%) cases of fully institutionalized democracy with informal militias.¹¹ Such groups are most commonly found in regimes with both democratic and autocratic characteristics; 46% of observations at Polity2 = 0 are coded as having informal PGMs. We control for economic development, which has been shown to reduce the risk of repression (for example, Landman 2005; Zanger 2000), and for population size, which is associated with an increase in violations (for example, Henderson 1991). Economic development is measured with the log of real GDP per capita and Population size with the log of population size, both from Penn World Tables. Based on past research we expect a positive coefficient for GDP per capita and a negative coefficient for population size. Finally, we account for past human rights violations, as research has shown that repression tends to continue (for example, Carey 2010; Poe et al. 1999).¹²

The Impact of Pro-Government Militias on Human Rights Violations

All authors

Neil J. Mitchell, Sabine C. Carey & Christopher K. Butler

https://doi.org/10.1080/03050629.2014.932783

Published online:

08 October 2014

FIGURE 2 Informal pro-government militias and democracy, 1982–2007.

Display full size

FIGURE 2 Informal pro-government militias and democracy, 1982–2007.

THE IMPACT OF PGMs ON HUMAN RIGHTS

The results of our ordered logit analysis using the full sample are shown in Table 1. In Model 1 the dependent variable is the 9-point CIRI Index of physical integrity, in Model 2 it is agent-centered violations (torture, disappearances, killings) on a 7-point scale, and in Model 3 political imprisonment on a 3-point scale. The higher the value in these measures, the higher the respect for the respective rights.

All control variables are highly statistically significant as in earlier research, apart from economic development in the model explaining political imprisonment (Model 3). Informal PGMs have a highly statistically significant negative impact on the respect for physical integrity rights (Model 1), supporting H1. As proposed in H2, these groups decrease the respect for freedom from torture, killing, and disappearance (Model 2). Finally, Model 3 supports H3, which argued that informal PGMs do not have a discernible impact on political imprisonment. Comparing the impact of informal and semiofficial PGMs on human rights violations across these three models, we can see that groups with a recognized link to governments do not increase the violation of physical integrity rights in general or of agent-centered violations more specifically. However, these semiofficial militias reduce the right to be free from political imprisonment. The respective variable in Model 3 is statistically significant at p < .05 with a negative coefficient. This suggests that for violations where at least some collaboration of official government agents is required, such as the courts or prisons, the informal armed actors have no discernible impact on the extent of these violations. Armed groups that have a recognized status within the state but are outside of the regular security apparatus do increase the risk of these types of violations.

We perform several robustness checks of these analyses. First, we replace the binary indicator of PGMs with count variables for the number of informal (ranging from 0 to 20 groups) and semiofficial PGMs (ranging from 0 to eight groups). While most results remain substantively the same, among the militia variables only the number of informal PGMs has a statistically significant and negative impact on physical integrity. This suggests that for the violation of agent-centered rights it is more important whether there are any, or at least one, militia and not how many there are. Second, to test whether characteristics related to the professionalization of the military explains PGMs as well as human rights violations, we control for military coup d’etats, using data from Powell and Thyne (2011), and logged military personnel per capita, take from the COW data. The variable Coup d’etat does not have a statistically significant impact on our human rights measures in any of the three models. Military personnel per capita is statistically significant (p < .001) and negative only in Model 3 analyzing political imprisonment. Again, the impact of militias and support for the central theoretical arguments remain substantively unchanged, noting that our measure for semiofficial PGMs drops below the p < .05 statistical significance threshold in the model of political imprisonment. Third, to test whether our results might be driven by armed conflict cases, we restrict our sample to nonconflict cases. The results remain substantively the same. Informal PGMs continue to have a statistically significant and negative impact on physical integrity rights and agent-centered violations even outside of the context of armed conflict. Fourth, we repeat Models 1–3 for democracies (with polity2 > 7) and nondemocracies separately. In the sample of democracies, we cannot distinguish from zero the impact of informal PGMs on the three measures of human rights, which could be because informal PGMs are present in only 8% of observations in this sample. For nondemocracies, the results resemble the ones reported in Table 1.¹³

In the second step, we use coarsened exact matching (CEM), using Informal PGMs as the treatment. The goal is to create two samples, one with and one without the treatment, that are nearly identical on the covariates in order to single out the effect of informal PGMs on human rights. The matching substantially reduced the multivariate and the univariate distances between the control and treatment group in both matching exercises. The two groups are virtually identical with respect to armed conflict and democracy, the two most consistently important covariates with human rights violations, while some small differences with respect to economic development and population size remains.¹⁴ In our final step, we include the weights that have been produced by CEM in the ordered logit analysis. The results are reported in Table 2.

The sample size and number of countries are substantially reduced due to the matching. All models are statistically significant with a pseudo R² of 0.23 or higher and with highly statistically significant Wald χ² statistic. Most control variables show the expected impact. Armed conflict is negative and statistically significant in all three models, reducing the respect of all the type of human rights tested in these analyses. Democracy is statistically significant in Models 4 and 6, increasing the respect for physical integrity rights and freedom from political imprisonment. For agent-centered violations in Model 5, the democracy variable just fails to reach statistical significance at p < .05 but is also positive as in the other models. Population size is consistently statistically significant and negative. As in earlier research, the respect for human rights declines as population size increases. Economic development fails to reach conventional levels of statistical significance in all models using the matched sample. Finally, the lagged indicators for human rights show that higher respect in the past leads to higher respect in the future.

The results presented in Table 2 using the matched sample support our results obtained from the full sample shown in Table 1. Informal PGMs have a statistically significant and negative impact on physical integrity rights and on agent-centered violations but have no discernable impact on the right to be free from political imprisonment. Semiofficial PGMs fail to reach meaningful levels of statistical significance in all three models. In Model 4, exponentiating the coefficient for Informal PGMs we arrive at the odds ratio of 0.666, which indicates that the respect for physical integrity rights deteriorates by about 33% when militias are present. This finding is consistent with our theoretical argument (H1). In Models 5 and 6 we test the impact of informal PGMs on different types of human rights. As a further test for our argument, Model 5 tests the impact of Informal PGMs on Agent-centered violations. We expect that the respect for these agent-centered rights (torture, extrajudicial killings, and disappearance) is most likely to deteriorate in the presence of these groups. Again, PGMs are highly statistically significant in this model and carry a negative coefficient, supporting H2. With an odds ratio of 0.591, the substantive impact of informal PGMs is even larger on these agent-centered violations, with a decline of about 41%, than on physical integrity rights more generally. Pro-government militias almost halve the respect for the right to freedom from torture, killings, and disappearances. Calculating the predicted probabilities of no violations for a nondemocratic country that respects these rights at t–1, with average population size and GDP, and no armed conflict, we find that a country without informal PGMs has a probability of 0.24 (95% confidence interval ranging from 0.15 and 0.33) of not experiencing any of these human violations. This probability declines to 0.16 (95% confidence interval ranging from 0.08 and 0.23) for a country with informal PGMs.

Model 6 uses Political imprisonment as dependent variable. We expect that informal PGMs do not affect the respect for this human right, as the agents are less able to imprison opponents without cooperation from the formal state apparatus. Informal PGMs fail to reach conventional levels of statistical significance, lending support to H3.¹⁵

DISCUSSION

Accountability is central to our understanding of human rights violations. Earlier research links democratic accountability to lower levels of violations. In this study we shift the focus to how the possibility to evade accountability, both on the part of the individual agent and on the part of the government, affects the respect for human rights. Using the analysis of delegation, we draw out the implications of particular agents for the respect for human rights. We argue that once governments cooperate with informal armed groups, respect for human rights will suffer. Both goal variance and information asymmetry, the central features of the agency problem, are likely to be more accentuated with armed groups that are outside of the formal chain of command. Recruitment, training, codes of conduct, and monitoring standards are likely to be lower than in the formal security sector. In addition to the temptation of hidden action for individual agents and simple control problems leading to abuse and killing, principals have the temptation to opportunistically use agents that are, or appear to be, difficult to control in order to evade accountability to state and nonstate actors and international organizations. The presence of these groups permits governments to shift responsibility for the consequences of the use of violence. Our argument that where these groups are present, levels of human rights violations are likely to be higher, is supported by the empirical analysis. This research contributes theoretically with a modification of standard principal-agent theory and empirically with incorporating the organization of the security sector and government militias into the explanation of human rights violations. Including relevant agents in addition to structural features such as democracy, levels of conflict, or economic conditions, is an important step forward in understanding the nature of and variation in human rights violations.

Further research on the creation of these groups will provide additional observable implications for the theoretical argument. While the focus of this analysis has been the consequences of these groups for the protection of human rights, the causes of pro-government militias is an important research puzzle in its own right. Evading accountability is not the only motivation for choosing to delegate security tasks to these groups. For example, for governments in particularly weak states, militias may be a low-cost and possibly more reliable alternative to coup-prone regular forces.¹⁶ At the same time, these states have limited capacity to exercise control over these irregular forces. While our focus is on the domestic impact of militias, another useful avenue for future research is to investigate how neighborhood and transborder factors influence the use and impact of these irregular groups (for example, Bapat 2012).

We argue that the nature of violence as well as the extent of peace—in terms of respect for human rights—is influenced by the organization of the security sector. A parallel line of research examines the organization of forces on the rebel side and its implications for violence (Cunningham, Bakke, and Seymour 2012). Empirically, we tested the theoretical argument with a measure of whether the security sector included a militia. Our findings suggest that the existence of informal pro-government militias does indeed put civilians at greater risk of harm. This relationship is most pronounced for those categories of violations that can be identified as more agent-centered violations. We show that when these groups are present, with or without armed conflict, the likelihood of violations increases. Our focus on the choice and contribution of the agents of repression shifts policy attention from factors such as regime type and the economic environment to the organization of violence. It highlights the responsibilities of governments in the choice of its agents, as well as recruitment procedures, training, monitoring, and the importance of the mechanisms familiar to the analysis of delegation. It raises the issue of the monopoly of the use of violence and the implications for the well-being of civilians. Attention to the choice of agent makes the policy problem one of policy and of governance rather than of less-tractable sociological or economic transformations.

ACKNOWLEDGMENT

Data and online appendix are available at http://dvn.iq.harvard.edu/dvn/dv/internationalinteractions. Please direct all correspondence to the authors.

FUNDING

This research was supported by the Economic and Social Research Council (ESRC), United Kingdom, RES-062-23-0363, and by the Centre for the Study of Civil War (CSCW) at Peace Research Institute Oslo (PRIO).