Human Rights

ButlerGluchandMitchell-SecurityForcesandSexualViolence.pdf

669

Introduction

This article examines the issue of delegation in the implementation of violence. Earlier research in the conflict and human rights areas has tended to examine structural or institutional factors. These factors include ethnic fragmentation, economic develop- ment, warfare, geographical terrain, or the

presence or absence of democracy. Those who approach political violence as a choice focus more on the strategic interactions, for example the level of threat presented by the opposition, and less on the motivation of those who carry out the violence. The purpose of this article is to show that violence, and notably sexual violence, may be its own reward for agents; that agents may misuse the power and authority attached to their public office to extract additional benefits; and that, in order to understand the use of violence and repression by the state, the interaction

Security Forces and Sexual Violence: A Cross- National Analysis of a Principal–Agent Argument*

CHRISTOPHER K. BUTLER, TALI GLUCH

Department of Political Science, University of New Mexico

NEIL J . MITCHELL

Department of Politics and International Relations, University of Aberdeen

This article investigates sexual violence committed by government security forces. It focuses on the issue of delegation. It uses principal–agent logic to understand sexual violence committed by these forces and to set up a cross-national empirical analysis. The article provides an approach to measuring the amount of agent discretion in a political system and an analysis of new cross-national data on the incidence of rape and sexual violence committed by police and security forces. It is argued that sexual violence is a category of human rights violation that is more likely to be attributable to the selfish motivations of agents, and it is an act that is likely to be hidden. This article identifies the conditions under which these acts are most likely to occur as conditions of conflict, system-wide slack bureaucratic control, and where there are constraints on information and organization. The theoretical argument makes sense of earlier findings in the human rights literature, such as the importance of democracy and conflict, while directing attention to motivations, accountability, and, for the first time in systematic analyses of human rights, to the particular issue of sexual violence. The use of cross-national data on the incidence of sexual violence for the year 2003 provides preliminary support for the theoretical argument. After controlling for factors likely to increase the incidence of sexual violence, such as military size and ethnic fraction- alization, it is found that, where agents are more accountable and subject to tighter control, sexual vio- lence is less likely. Finally, the article points to the importance of additional data collection over time, consistent with other human rights and conflict datasets, and draws out some policy implications follow- ing from the theoretical argument and analysis.

*Thanks to the editor and the reviewers for their helpful com- ments and suggestions. The data used in this article were ana- lyzed using Stata 9.2 and can be found at http://www.prio. no/jpr/datasets. Correspondence: [email protected].

between principal and agent is a critical relationship.1 The article argues that agents seeking private goods influence the choice, level, and type of violence, and that the way principals and agents interact is central to our understanding of political violence, including the perpetration of sexual violence by govern- ment or security forces.

While the leader is formally in charge, principal–agent logic holds that the agents’ superior information and selfish natures loosen control. This logic draws attention to the potential for conflicting goals between principals and agents, and, although the prin- cipal is supposed to be in control, it draws attention to the advantages that the agent enjoys over the principal (Arrow, 1985). Entrusting the task to others creates an oppor- tunity for the agents’ own selfish and hidden actions – it creates a moral hazard. The selfish agent prefers shirking to working for example (Brehm & Gates, 1997). Additionally, the prin- cipal is likely to encounter adverse selection in recruiting agents where individuals volunteer for service with characteristics and tastes potentially harmful to the principal’s mission, comparable to an insurance company receiv- ing applications from bad risks. In the rela- tionship with the self-interested agent, the principal is likely to be at an information dis- advantage about the policy area and about the extent of the agent’s selfish character and actions. The size of the information disad- vantage affects the ability of the principal to control the agent. Economists and political scientists have applied this logic and the assumptions of superior information and selfish actions on the part of agents to the behavior of bureaucracies (Banks & Weingast, 1992; Meier, 1993), including the use of unnecessary force by police officers in

the United States (Brehm & Gates, 1997). With a comparative research design, this article aims to show that the issue of delega- tion is equally important to our understand- ing bureaucracies that supply violence and the choice among types of repression imposed by government forces. What should be clear and what gives impetus for this article is that not all security forces are alike, nor all con- flicts, as Wood (2004, 2006) makes clear in her pioneering comparative work on sexual violence. They differ in the quality of recruits selected, in the degree of discipline and control exercised over agents, and in the frequency of sexual violence.

If this logic applies as it does in other policy areas, some of these agents will misuse their operational knowledge, their uniform, and government resources for their own private benefit. Specifically, what are the likely private goods of the agents? What evi- dence is there that the independent selfish activity of agents is contributing importantly to levels of repression? How do we use this logic to combat violence? This article pro- vides evidence of the agents’ impact on the level of violence by developing an approach to collecting and analysing cross-national data on the incidence of sexual violence com- mitted by government forces. There is an accumulation of research on the use of sexual violence by government and security forces and on the broader issue of women and the military and women’s rights (Apodaca, 2000; Arat, 2002; Beevor, 2002; Brownmiller, 1975; Chang, 1997; Enloe, 1983, 2000; Goldstein, 2001; Henderson, 2004; Niarchos, 1995; Walzer, 1977; Wood, 2004). There is an empirical analysis of government responses to sexual violence (see Weldon, 2002), but, to our knowledge, there are no large-N cross- national empirical analyses of sexual violence committed by governments’ security forces. Finally, the article points to policy implica- tions and priorities consistent with this theoretical approach.

journal o f PE AC E RE S E A RC H volume 44 / number 6 / november 2007670

1 See, for example, Bawn (1995) and Huber & McCarty (2004) for formal arguments, and Brehm & Gates (1997) and Waterman, Rouse & Wright (2004) for empirical analysis of the principal–agent relationship. See Mitchell (2004) for this theoretical argument applied to human rights violations in three historical cases.

The Agent’s Motive

The concepts of principal and agent identify the key policy participants. These concepts imply goal variance, where agents have goals independent of those of the principal, and they imply information asymmetry, where agents have an information advantage over the principal. Goal variance and information asymmetry may lead to the delivery of a policy in a way or in an amount different from what is ordered by the leaders. So, what are the selfish motivations of agents in security forces?

They may use violence to satisfy their desire for revenge or to extract loot, or they seek gratification in the violence itself. In the final days of the British Mandate, British sol- diers left four Jewish Haganah soldiers to be killed by Arabs, seeking revenge for the killing of some of their own (Morris, 1999: 197). Torture is reportedly common in Indian prisons, in part because it is used to extract payments from prisoners or their rel- atives (Rothman & Neier, 1991). Yet, the most agent-centered of these acts is rape and sexual violence, a form of torture in interna- tional human rights law. It is committed by security and police forces in both peacetime and during conflict. Sexual violence is the most likely category of violent behavior to indicate a principal–agent problem, as it is more immediately attributable to selfish rather than strategic motivations, although the latter are not precluded.

Identifying the motivation as agent- centered or selfish does not mean that it is simple. The literature suggests that the self- ishness of the agents may have a variety of dimensions. Brownmiller (1975: 256), for whom ‘all rape is an exercise in power’, pro- vides evidence from the Vietnam War and elsewhere that suggests the ‘ordinariness’ of rape, that getting along with peers was a strong motivation, noting the ‘fear of ridicule’ if you did not go along (Brownmiller, 1975: 102). Browning (1998: 168) identifies similar

psychological pressures in his examination of the actions of a German police battalion in Poland. He describes a ‘nucleus of enthusias- tic killers’. Most of the rest went along, also fearing abuse about their manhood and bravery if they did not participate (Browning, 1998: 66). Whether seeking sex, power, peer ingratiation, or avoiding ridicule, the agent perceives some personal gain and exercises the power vested in him by the state in order to realize this gain. The victim’s reluctance to report, a reluctance that is specifically and acutely associated with sexual violence, is a factor that increases the chances of this type of hidden action by agents. Because of the general public sensitivity to the use of political violence and human rights violations, and the disincli- nation of leaders or principals to audit security agencies, problems of information asymmetry may be particularly severe in this policy area.

These private motives are robust to the passage of time and to cultural and ideolog- ical variance. In World War II, not just German soldiers but also Red Army soldiers (Beevor, 2002) and Japanese soldiers (Chang, 1997) routinely committed sexual violence. There are also reports of sexual vio- lence by Allied soldiers (Silone, 1961: 189; Walzer, 1977: 133) and, to take a recent example, by US soldiers in the conflict in Iraq. Five soldiers from the United States 101st Airborne Division are accused of raping an Iraqi girl aged 14 and then killing her, her parents, and her six-year-old sister in March 2006. The soldiers are being prosecuted and their actions ‘embarrassed the American military’ and complicated relations with the Iraqi authorities (Semple, 2006). While this case caught the headlines, a recent United Nations report (2006a: 45) has identified violence against women as a widespread problem: ‘Violence against women has been reported from conflict or post-conflict situations in many countries or areas including Afghanistan, Burundi, Chad, Colombia, Côte d’Ivoire, Democratic

Chri s topher K. But l er e t a l . SE C U R I T Y FO RC E S A N D SE X UA L VI O L E N C E 671

Republic of the Congo, Liberia, Peru, Rwanda, Sierra Leone, Chechnya/Russian Federation, Darfur, Sudan, northern Uganda, and the former Yugoslavia.’ The report con- centrates on conflict or post-conflict condi- tions, yet sexual violence by security forces is not limited to combatants in the conflict.

In the Democratic Republic of the Congo, UN peacekeepers from Pakistan, Uruguay, Morocco, Tunisia, South Africa, and Nepal are alleged to have committed rape, sexual exploitation, and abuse of minors (Lynch, 2004). These activities are in addition to the already very high levels of sexual violence per- petrated by combatants in the civil war in that country. In addressing the problem, the United Nations has investigated 296 peacekeepers (United Nations, 2006b). In Somalia in 1993–94, it was Belgian and Italian peacekeep- ers (Amnesty International, 2004: 55) that were reported to have committed sexual violence. At the United States Air Force Academy, the Pentagon’s inspector general found that, over a ten-year period, sexual assault was reported over 150 times and that ‘the overall root cause of sexual assault problem at the Air Force Academy was the failure of successive chains of command over the past ten years to acknowl- edge the severity of the problem’ (Shanker, 2004). Acknowledging the severity of the problem is the first step to establishing control.

This admittedly selective presentation of evidence suggests that the agents’ hidden actions vary in extent and severity, but have been a problem for security forces over time and across cultures. The challenge is how to begin to investigate the agents’ contribution to violence more systematically. Security forces will differ in the degree of control and discipline, and the empirical task of this article is to examine how variation in control, or the degree of slack given agents, influences the incidence of sexual violence. For this analysis, we assume governments lose control as opposed to employing the strategic cyni- cism of refusing to control.

Measuring and Modeling the Agents’ Use of Sexual Violence

Data on sexual violence and rape are notori- ously hard to collect. The perpetrator gener- ally hides from accountability, and the victim hides from the double harm of social humili- ation. Unfortunately, the existing measures of human rights violations are too aggregate to be useful for our analysis.2 Even Henderson’s (2004) exploratory research aggregates multi- ple dimensions (harassment, provision for women’s hygiene, virginity exams, hostage taking, absence of female guards, threats of rape, rape, forced prostitution) to one scale. Our approach is to adopt the methods used to construct the Political Terror Scale (Gibney & Dalton, 1996; Hathaway, 2002; Lopez & Stohl, 1992; Poe & Tate, 1994) with a spe- cific focus on the use of sexual violence com- mitted by security forces. The term ‘security forces’ refers to both militaries and govern- ment-supported militia, as well as police and prison guards.

Rape and sexual violence are defined in international law as follows:

While rape has been defined in certain national jurisdictions as non-consensual inter- course, variations on the act of rape may include acts which involve the insertion of objects and/or the use of bodily orifices not considered to be intrinsically sexual.… The Chamber defines rape as a physical invasion of a sexual nature, committed on a person under circumstances which are coercive. Sexual vio- lence, which includes rape, is considered to be any act of a sexual nature which is committed on a person under circumstances which are coercive. (International Tribunal for Rwanda, quoted in Askin, 1999: 109)

Both Amnesty International and the US Department of State annual country reports provide discussion of rape and sexual violence, which includes sexual assault and abuse, by government agents. These are the standard

journal o f PE AC E RE S E A RC H volume 44 / number 6 / november 2007672

2 See McCormick & Mitchell (1997) for the argument for disaggregating measures of human rights violations.

sources for human rights violations data that have been used in earlier research, and, at least since the end of the Cold War, they are quite consistent (Poe, Carey & Vazquez, 2001). As the US Department of State reports provide more detailed and comprehensive coverage, we relied on these reports.3 We collected data for 163 countries for 2003 and developed a five- point ordinal scale to measure the level of rape and sexual violence by government security forces, ranging from ‘0’, indicating the lowest incidence to ‘4’, indicating the highest inci- dence. A ‘0’ was assigned to countries where there was no mention of sexual violence by government forces. This scale represents an ordinal scale where ‘0’ represents the value below ‘1’, but not necessarily that there were no actual cases of sexual violence. The Department of State reports had substantially fewer zero values than the Amnesty International reports. As with ordinary crime data, we anticipate that the most likely error with these data is that they undercount the frequency of sexual violence and rape.4 Tables AI and AII in the appendix provide a description of the scale and the score coded for each country.

Our argument is that the incidence of sexual violence by government forces is driven by out-of-control agents. It is an undeniably challenging task to measure the degree of slack given agents in bureaucracies. We develop

three ‘agency variables’ that are expected to lower levels of sexual violence if the theoreti- cal focus on principal–agent logic is to be sus- tained empirically. First, where organizations can form to monitor human rights violations, where information critical of government officials compensates for information asym- metries, and where elected officials may be held accountable for the activity of their agents, then we expect there to be a lower like- lihood of sexual violence by security forces. The formation of advocacy and interest groups is stimulated by public policies, and these organizations then assume the costs of auditing and monitoring public policies. These organizations form more or less easily across different policy areas and political systems, and, as Banks & Weingast (1992) observe, the cost of policy monitoring and auditing varies. It is plausible that the costs of establishing even private auditing mechanisms over agencies engaged in human rights vio- lations will tend to be higher. Those that care most about the use of violence may encounter an intimidating environment and severe prob- lems in collective action and gathering and reporting information about repression. In turn, repressive agencies can claim to be working for the most vital interest of the state and safeguarding national security and sur- vival, which makes those who raise questions concerning these agencies’ activities highly vulnerable to counter-questions about their loyalties.

Self-regulating mechanisms for auditing the bureaucracies of repression are most likely to develop where there are rights to free organiz- ation and association and free and critical media, which provides a further agency-based theoretical foundation for the finding of a cor- relation between lower human rights violations and democracies in the quantitative analysis of human rights violations (Beer & Mitchell, 2004; Davenport & Armstrong, 2004; Poe & Tate, 1994; Zakaria, 2003). Repression gener- ates at least latent interest communities and

Chri s topher K. But l er e t a l . SE C U R I T Y FO RC E S A N D SE X UA L VI O L E N C E 673

3 In order to collect the information for the Department of State, embassies monitor NGOs, media, government officials, jurists, armed forces sources, academics, and other sources. Within the Department of State, the Bureau of Democracy, Human Rights, and Labor then corroborates information with UN representatives, NGOs, and others. Using this source, the United States is not included in the analysis. When we compared the Department of State reports with the Amnesty International country reports for 2003, we found that the widest discrepancies between the two sources were for Myanmar and Colombia. 4 As we are analyzing just 2003 data, we examined that year in relation to other years on the Political Terror Scale (Gibney & Dalton, 1996) to provide some context. The Department of State PTS scores for 2003 were lower than 2002 but higher than the historical average (1980–2004). The correlation between the PTS and the measure of sexual violence was 0.63 for 2003 – strong enough to increase confidence in the sexual violence index but not so strong as to diminish its uniqueness.

networks. But we know, from the example of the mothers of the disappeared during the dirty war in Argentina, the extraordinary courage it takes to organize and become active (see Brysk, 1994). In democracies, these latent interest communities, by using democratic rights and ideas of democratic legitimacy, have a better chance of achieving organizational form. An example of this sort of organization would be B’Tselem, a major human rights organization that monitors the Israeli–Palestinian conflict. In contrast, a Russian court shut down the Russian–Chechen Friendship Society, a human rights monitoring organization (Amnesty Inter- national, 2006). As a result of the higher infor- mation flows in democracy, the oversight and accountability of agents increases, and we expect that, where there is a high level of democracy, there will be a low level of sexual violence. This expectation conforms to the findings of more general analyses of human rights violations (Davenport & Armstrong, 2004; Hathaway, 2002; Henderson, 1991; Moore, 2000; Poe & Tate, 1994; Poe, Tate & Keith, 1999) and alternative theoretical argu- ments linking democracy to less repression, perhaps pointing to the effect of democratic institutions in lowering the level of opposition threat or in disseminating democratic values. Consistent with these studies, we use Polity IV regime scores for 2003 (Marshall & Jaggers, 2002) to measure the level of democracy in a political system. We also examine a categorical measure of democracy (Polity greater than 7) to address the more recent claim (Davenport & Armstrong, 2004) that only ‘full’ democra- cies have a beneficial impact on human rights violations.

Second, higher levels of economic development, measured by GNI per capita (in thousands) for 2003 (World Bank, 2004), permit better compensation of public officials and – in the absence of reliable comparative data on wages and salaries – this measure has been used as the best available measure of public sector wages (Montinola & Jackman,

2002: 154). As Montinola & Jackman state (2002: 150): ‘If public officials are well paid, they will value their positions more highly and will have fewer incentives to jeopardize those positions.’ We extend this argument to officials in security forces and expect that better pay results in better control. Not only will those in well-paid positions value the positions more highly, but better pay repre- sents a means to address the problem of adverse selection and minimize the recruit- ment of bad risks.

As with democracy, there are other theor- etical arguments for a relationship between economic development and low levels of human rights violations generally. The exist- ence of these alternative theoretical arguments makes the empirical test of economic develop- ment and democracy critical to, but not deci- sive for, a principal–agent argument. If we found no relationship between these variables and the dependent variable, it would be evidence against our theoretical argument. Therefore, the variables are important to the analysis. Yet, in this rich theoretical context, finding a relationship does not help sort bet- ween the various theoretical interpretations. In order to capture more directly the amount of discipline and control in a political system, for our third agency variable we use the degree to which public officials are perceived to use their offices for private financial gain.

Financial corruption deals with the finan- cial rewards that public officials extract from performing or not performing their duties. High levels of overall financial corruption reflect a general absence of monitoring, accountability, and control in a political system and a high degree of agent discretion to pursue hidden actions. It is plausible that this high degree of agent discretion applies to security forces as it does to other public bureaucracies. Where there is systemic finan- cial corruption indicating slack bureaucratic control, members of security forces and police forces will perceive increased opportu-

journal o f PE AC E RE S E A RC H volume 44 / number 6 / november 2007674

Nanking emphasizes the importance placed by Japanese soldiers on the inferiority of the Chinese. Therefore, as a community becomes more fractionalized along ethnic lines, there is a greater likelihood of sexual violence. Finally, conflict likely increases the oppor- tunity for hidden actions and, therefore, the likelihood of sexual violence by agents. Under conflict conditions, agents may be in close contact with the civilian population and expect to hide their actions more easily and victims may be even more disadvantaged in informing on the perpetrators. Civil conflict and civil war may also increase recruitment problems and the risk of adverse selection. Generally, empirical studies have identified a positive relationship between civil war, mea- sured as at least 1,000 battle-deaths a year, and state repression (Burkhart, 2002; Davenport & Armstrong, 2004; Harrelson- Stephens & Callaway, 2003; Poe & Tate, 1994; Keith, 1999, 2002; Poe, Tate & Keith, 1999). Yet, recognizing that even lower levels of violence could prompt increased govern- ment use of violence (Gartner & Regan, 1996; Moore, 2000), we also examine the lower threshold of conflict as provided by the Uppsala/PRIO database to investigate the relationship between conflict and rape by government security forces.

Empirical Analysis

We begin the empirical analysis with the cor- relation matrix in Table I and draw attention to some important features.6 First, the highest absolute pair-wise correlation with the dependent variable is with the World Bank Control of Corruption score. This neg- ative correlation is consistent with our argument. In political systems where there is little control over the selfish or hidden

Chri s topher K. But l er e t a l . SE C U R I T Y FO RC E S A N D SE X UA L VI O L E N C E 675

5 The World Bank’s ‘Control of Corruption’ measure is one of six ‘governance research indicators’, available for 162 countries from 1996 to 2002. Although a similar measure of corruption is available from Transparency International for 2003, it is available for only 130 countries. In order to increase the number of observations, we chose to employ the World Bank’s corruption measure for 2002. The two measures are highly correlated at 0.97. Table AII in the appendix provides a listing of cases with their Control of Corruption score and our coding of sexual violence. Three countries do not have a Control of Corruption score but were coded for their level of sexual violence: Albania (0), Belize (1), and the Solomon Islands (2).

6 The case of Iraq was dropped from all analyses, owing to its status as an occupied country for much of 2003. The Department of State report describes rape as routine under the Saddam Hussein regime, but the discussion is not limited to the year 2003.

nities and a more favorable balance of incen- tives to pursue hidden actions. For some of these agents, the expectation of increased opportunities for hidden action will translate into an increased likelihood of committing sexual violence. Financial corruption is measured by the World Bank’s Control of Corruption index for 2002 (Kaufmann, Kraay & Mastruzzi, 2005) and is based on surveys of business executives and country analysts’ perceptions.5 Perceived financial corruption provides a general indicator of the degree of slack given to agents by principals, and this indicator is independent of the reported incidence of sexual violence by security forces. We therefore expect that political systems with higher levels of hidden activity in the form of financial corruption also will have a higher likelihood of sexual violence committed by security forces.

There are several other factors that are likely to increase the incidence of sexual vio- lence. These factors include the size of a country’s military forces (in millions of troops for 2001 – the most recently available; see Singer, Bremer & Stuckey, 1972), the degree of ethnic fractionalization (Fearon, 2003), and the level of conflict as coded by the Uppsala/PRIO data for 2003 (Gleditsch et al., 2002). As the size of military forces increases, it is plausible that the difficulties of control increase. It is also plausible that vio- lence, including sexual violence, is more likely when the victim is perceived as ethni- cally distinct from the perpetrator. Chang’s (1997) account of Japanese soldiers at

actions of public officials in financial matters, we also find those public officials employed in security forces more likely to commit sexual violence. Second, the other agency variables – GNI per capita and democracy – are also related to the sexual violence measure as expected. Third, it is important to note that there is a high corre- lation between Control of Corruption and GNI per capita, and both of these variables are quite highly correlated with the level of democracy (whether measured by the Polity index or the categorical variable for Polity greater than 7). We cannot disentangle the causality issue with the single-year cross- national data that we analyze here. However, we will demonstrate that each of these three variables reduces the level of sexual violence, consistent with the hypothesis that greater control over agents leads to better behavior by those agents.

The factors that are hypothesized to worsen the prevalence of sexual violence are the size of a country’s military forces, the degree of ethnic fractionalization, and the level of civil violence. The size of a country’s military is highly correlated with its total population (0.83) and moderately correlated with its geographic size (0.50). A country’s population and geographic size are more distant proxies for the number of govern- ment agents whose behavior is being exam- ined than is the size of its military. The degree of ethnic fractionalization may be a factor making sexual violence worse to the extent that government agents are more likely to abuse individuals from other ethnic groups. If this is the case, then a greater degree of ethnic fractionalization creates more opportunities for such behavior. Finally, we examine how the level of civil violence affects the prevalence of sexual vio- lence. Given an armed rebellion between the government and some rebel group (or groups), current levels of violence surround- ing this rebellion can loosen the constraints

journal o f PE AC E RE S E A RC H volume 44 / number 6 / november 2007676

Ta bl

e I.

C or

re la

ti on

M at

ri x

of V

ar ia

bl es

Se xu

M ili

ta ry

hn ic

on tr

ol o

f vi

ol en

ce pe

rs on

ne l

G eo

gr ap

hi c

fr ac

tio na

l- C

iv il

C or

ru pt

io n

G N

I pe

r Po

lit y

in de

x (2

00 1)

Po pu

la tio

n siz

e iz

at io

n co

nf lic

t C

iv il

w ar

(2 00

2) ca

pi ta

sc or

M ili

ta ry

p er

so nn

el in

m ill

io ns

(2 00

1) 0.

25 9

1. 00

0 Po

pu la

ti on

in th

ou sa

nd s

0. 22

5 0.

83 2

1. 00

0 G

eo gr

ap hi

c si

ze in

k m

2 0.

22 1

0. 50

4 0.

43 0

1. 00

0 E

th ni

c fr

ac ti

on al

iz at

io n

0. 19

6 −0

.1 74

−0 .0

10 −0

.0 01

1. 00

0 C

iv il

co nf

lic t

0. 23

0 0.

11 6

0. 03

5 0.

17 6

0. 12

1 1.

00 0

C iv

il w

ar 0.

21 4

0. 25

3 0.

43 0

0. 11

8 0.

14 3

−0 .0

35 1.

00 0

C on

tr ol

o f C

or ru

pt io

n (2

00 2)

−0 .4

31 −0

.0 79

−0 .0

59 0.

00 2

−0 .3

53 −0

.0 99

−0 .0

69 1.

00 0

G N

I pe

r ca

pi ta

in th

ou sa

nd s

−0 .3

73 −0

.0 65

−0 .0

66 0.

01 2

−0 .3

79 −0

.1 05

−0 .0

70 0.

85 6

1. 00

0 Po

lit y

sc or

e −0

.4 09

−0 .1

47 −0

.0 39

0. 03

4 −0

.1 90

0. 00

4 −0

.0 36

0. 43

5 0.

39 5

1. 00

0 H

ig h

de m

oc ra

cy −0

.2 53

−0 .0

29 0.

03 6

0. 02

3 −0

.2 95

−0 .0

06 0.

02 2

0. 53

7 0.

50 1

0. 74

government places on agents. Countering civil violence also increases the demand for recruits, resulting in the mobilization of poorer quality agents. Either of these effects of civil violence thus increases the likelihood of uncontrolled violence, including sexual violence. We examine ‘civil conflict’ (having 25 to 999 battle-dead in the current year) and ‘civil war’ (having more than 999 battle- dead in the current year) as separate dichoto- mous variables. Thus, the base category is the absence of either of these categories of violence.

The factors that are hypothesized to lower the prevalence of sexual violence are Control of Corruption, GNI per capita, and democracy. We argue that high levels of all three ‘agency’ variables place high constraints on government agents, while low levels of all three variables imply few constraints on government agents. To this end, we constructed two dummy vari- ables to capture these populations. CGDhigh represents those cases (N = 31) that are in the highest 33rd percentile on Control of Corruption (≥ 0.2142), GNI per capita (≥

Chri s topher K. But l er e t a l . SE C U R I T Y FO RC E S A N D SE X UA L VI O L E N C E 677

Table II. Combined Effects Reducing Sexual Violence

Model 1

Military personnel (2001) 0.70 0.056 (0.36)

Ethnic fractionalization 0.24 0.582 (0.44)

Civil conflict 0.85 0.012 (0.34)

Civil war 1.33 0.055 (0.70)

CGDhigh −1.83 0.000 (0.48)

CGDlow 0.72 0.015 (0.30)

Cut point 1 0.25 Cut point 2 0.67 Cut point 3 1.31 Cut point 4 2.87

Number of observations 145 LR χ2 (6) 58.31 Prob. > χ2 0.000 Pseudo R2 0.17 Log likelihood −146.85

Value entries are coefficients and p-values with standard errors in parentheses.

100

Lowest third on agency variables

Base Highest third on agency variables

P ro

ba bi

lit y

of o

ut co

m e

ca te

go rie

s (%

)

Low level of sexual violence High level of sexual violence

Figure 1. Demonstrating Combined Effects

3.508), and the Polity index (≥ 8). CGDlow represents those cases (N = 16) that are in the lowest 33rd percentile on Transparency (≤ –0.7300), GNI per capita (≤ 0.656), and the Polity index (≤ –1). The remaining 98 cases in the estimation have some other combination of the three variables that makes it difficult to say whether they are clearly at the high end or clearly at the low end.

Table II presents the ordinal probit estimation results using these two dummy variables. The results show that being at one extreme on all three agency variables clearly produces an anticipated extreme effect in the level of sexual violence. Of the other factors, consistent with expectations, civil conflict is positively and significantly related to the level of sexual violence. The number of military

personnel is just insignificant, as is the civil war variable. Ethnic fractionalization is not significantly related to the level of sexual vio- lence. Figure 1 presents some marginal effects regarding this estimation, setting military size and ethnic fractionalization at their means and the conflict variables to their modal cate- gory of zero. This figure shows more starkly that being at the low end of Control of Corruption, GNI per capita, and the Polity index makes it much more likely that a country will have a high level of sexual vio- lence (categories 2, 3, or 4), while being at the high end of these three agency variables makes it almost certain that a country will have a low level of sexual violence (categories 0 or 1).

Model 2 of Table III presents ordinal probit estimation results for only the factors hypoth-

journal o f PE AC E RE S E A RC H volume 44 / number 6 / november 2007678

Table III. Independent Effects Reducing Sexual Violence

Model 2 Model 3 Model 4 Model 5 Model 6

Military personnel (2001) 0.85 0.017 0.74 0.041 0.74 0.043 0.65 0.083 0.83 0.023 (0.36) (0.36) (0.37) (0.37) (0.37)

Ethnic fractionalization 1.01 0.011 0.42 0.325 0.36 0.411 0.67 0.113 0.70 0.099 (0.40) (0.43) (0.43) (0.42) (0.42)

Civil conflict 0.86 0.009 0.83 0.014 0.90 0.012 0.96 0.007 0.85 0.015 (0.33) (0.34) (0.36) (0.35) (0.35)

Civil war 1.27 0.065 1.32 0.062 1.53 0.043 1.57 0.083 1.51 0.089 (0.69) (0.71) (0.76) (0.90) (0.89)

Control of Corruption −0.73 0.000 (0.14)

GNI per capita −0.09 0.000 0.02

Polity score −0.07 0.000 (0.02)

High democracy −0.58 0.008 (0.22)

Cut point 1 0.82 0.68 0.11 0.43 0.45 Cut point 2 1.17 1.10 0.53 0.84 0.83 Cut point 3 1.74 1.74 1.15 1.49 1.44 Cut point 4 3.22 3.40 3.11 3.36 3.22

Number of observations 145 144 139 140 140 LR χ2 (.) 26.44 56.89 47.96 41.22 30.35 Prob > χ2 0.000 0.000 0.000 0.000 0.000 Pseudo R2 0.08 0.16 0.14 0.12 0.09 Log likelihood −162.79 −146.98 −143.22 −145.93 −151.37

esized to increase the level of sexual violence. The results show that the size of the military, the degree of ethnic fractionalization, and the presence of civil conflict are indeed positive predictors of the level of sexual violence when not including agency variables. The presence of a civil war just fails to achieve statistical significance, but this is hardly surprising as there are only three instances of this in our cross-national data for 2003.

Models 3 through 6 show the results of ordinal probit estimations examining one agency variable at a time. Each of these three agency variables acts as predicted, sig- nificantly reducing the level of sexual vio- lence after controlling for the positive predictors from Model 2. The one addi- tional consistent effect across these models (and Model 1) is that ethnic fractionaliza- tion remains insignificant.

The high pairwise correlations among the three agency variables become apparent when two or more are included in a given estimation. Table AIII in the appendix examines all com- binations of our agency variables that include the Control of Corruption score. Control of Corruption fails to achieve statistical signifi- cance in just one of the five models tested and performs better than economic development. GNI per capita is not significant in any of the models for which it is included. This may well be indicative of a larger problem of high cor- relations among the independent variables in the human rights literature. While level of economic development is invariably included in models of human rights violations, a number of studies report no statistically sig- nificant relationship between economic development and human rights violations.7

Similarly, while the Polity index measure is consistently negative and significant, the High

Democracy categorical variable does not reach significance.

Overall, the results of these analyses are supportive of the theoretical argument and the importance of including the contri- butions of agents in explanations of the frequency of sexual violence by government security forces. A combination of electoral accountability for those in charge of agents, compensation levels for agents, and the monitoring and transparency of agents’ actions is likely to provide leverage in address- ing the problems of slack within public bureaucracies and notably within security forces.

Discussion

Earlier research in the comparative analysis of human rights violations identifies the eco- nomic and political conditions associated with violations and the strategic motives that might lie behind such violations (Burkhart, 2002; Cingranelli & Richards, 1999; Gartner & Regan, 1996; Harrelson-Stephens & Callaway, 2003; Poe & Tate, 1994). But there has not been attention to those who carry out the violence. It has been assumed that violence has only strategic purposes (for a notable exception, see Kalyvas, 2006). The evidence presented in this article suggests that this assumption should itself be more rigorously investigated, and its likely over- simplification acknowledged, as we think about ways to combat sexual violence. Alternative measures of the control and disci- pline of security forces and more extended data on sexual violence will add confidence to these findings.

While we note their preliminary nature, the findings suggest that women are vulnerable during armed conflicts and wartime, but also more vulnerable in countries where there is a low level of control over public officials. Security forces are not all alike and differ in levels of disci- pline and esprit de corps. To measure variation in

Chri s topher K. But l er e t a l . SE C U R I T Y FO RC E S A N D SE X UA L VI O L E N C E 679

7 In Burkhart (2002), GNP per capita is significant in 8 of the 18 models tested. In Hathaway (2002), GNP per capita is significant in 1 of the 8 models tested. In Richards, Gelleny & Sacko (2001), it is not significant in the model tested.

the overall tightness of discipline of public officials across political systems, we used a cor- ruption measure, democracy, and level of eco- nomic development. Higher corruption levels likely raise expectations of opportunities for hidden action among public officials, including members of security forces. There is some normative ambivalence toward financial cor- ruption in the social science literature, and some have argued that it may correct for or ‘grease’ bureaucratic interference with growth and effi- ciency, and that there may be some tolerable level of corruption (see Acemoglu & Verdier, 2000; Bardhan, 1997: 1322). The finding that the level of financial corruption in a political system is robustly associated with the extensive- ness of sexual violence committed by policemen and soldiers and the logic underlying this rela- tionship suggest the prioritization of agencies for increased monitoring and ‘propitious recruit- ment’. The usefulness of this knowledge lies in shifting attention away from economic or cul- tural conditions that are not policy sensitive within a meaningful time frame.

Even warfare is not an equal hell, at least for women (Wood, 2004, 2006). A few con- flicts, Israel–Palestine for example, or the English Civil War (Mitchell, 2004), have a low frequency of sexual violence. Some con- flicts, on the other hand, have very high levels of sexual violence, the Russian Civil War, for example. We would note, in particular, that ethnic fractionalization is plausibly connected to high levels of sexual violence in some par- ticular conflicts. In Burma/Myanmar, the army is reported to use rape as a weapon of war against particular ethnic groups:

In May 2002, the SHRF and Shan Women’s Action Network (SWAN) alleged that the Burmese Army used rape as a systematic weapon of war against the ethnic populations in Shan State. The report described 173 inci- dents of rape or sexual violence against 625 women and girls committed by soldiers from 52 military battalions between 1992 and 2001. (Bureau of Democracy, Human Rights, and Labor, 2004: Burma)

Yet, when we examine ethnic fractionaliza- tion as a general factor in explaining levels of sexual violence, it is not significant when agency variables are included in the analysis. The policy implication is that ethnically divided societies are not irretrievably locked into violent patterns of behavior. Focusing on agents and the control issue shifts atten- tion from ethnic, cultural, or even conflict conditions to thinking about incentives and enforceable codes of conduct that are shaping agents’ decisions, and suggests that the capacity for violence – the definitional resource of government – requires leadership and management like any other resource.

Principal–agent logic usually implies two types of intervention to limit the hidden actions of agents. First, the government must reduce the amount of goal variance. The general analysis of bureaucracies suggests that adequate compensation for officials is an important tool in combating goal variance and the problem of corruption (we note that our analysis did not measure actual compen- sation rates for security forces cross-nation- ally, but used GNI per capita as a proxy). Formal and empirical analyses of corruption recognize the importance of agent compen- sation. Financial compensation is a direct response to bribery, the most general form of corruption, and is part of the response to the bureaucracy of repression’s proclivity for hidden violence. Better compensation increases the value of the position, lowers the incentive to extract payments from families and victims with the threat of violence, and improves recruitment.

The second type of intervention is to redress the information asymmetry through moni- toring. The particular difficulty with govern- ment use of violence is that in this area agents are particularly hard to monitor. In hard-to- monitor policy areas, formal models of out-of- control bureaucracies and corruption suggest, again, that paying higher wages is the best choice (Acemoglu & Verdier, 2000: 195). But, in the

journal o f PE AC E RE S E A RC H volume 44 / number 6 / november 2007680

use of violence, non-monetary rewards are at stake. In other words, there is no substitute for monitoring and accountability in this policy area. While the expectation is that under democracy, in addition to internal government systems of auditing and monitoring, other organizations and the media monitor public policy delivery, including the delivery of repres- sion. The increasing engagement of the interna- tional system, through NGO activity and through international law, is a positive develop- ment in overcoming the inadequacies of domestic monitoring and accountability and information asymmetries that increase the opportunities for hidden action (Keck & Sikkink, 1998). In our analysis, we used a measure of democracy as a measure for infor- mation flow and accountability. Yet, we would also note that an indicator of external moni- toring, specifically whether the government granted or restricted access to the International Committee of the Red Cross (ICRC) to monitor prison conditions, is consistent with the analysis – the worst offenders were more likely to deny access. Of the worst offenders (scores 2, 3, or 4), some 39% also denied access to the Red Cross. This percentage falls to 17% for the better performers (scores of 0 and 1), and the difference is large enough to achieve statisti- cal significance (χ2(1) = 9.14, p = 0.003).8

Finally, while again noting the impor- tance of additional empirical data, to the extent that we can attribute a substantial amount of sexual violence to principal–agent problems, then a third and ‘softer’ instru- ment (see Arrow, 1985: 50) follows from this theoretical argument. This instrument is to

attend to the selection and value commit- ments of agents. Only some of the agents seek violent gratification, and encouraging, recruiting, and training for the appropriate norms among the members of the security forces can isolate and contain these individ- uals and increase the likelihood of self-moni- toring among agents. Returning to the Iraq rape case, the main defendant illustrates the importance of selection. Recruitment pres- sure increases the risks of adverse selection and the odds of agents engaging in hidden actions.

Green’s case has helped to spur a closer look at the Army’s standards for recruitment and training. Green enlisted and passed basic train- ing at a time when the Army was under terrific pressure to bring in new soldiers and had relaxed its entry requirements. In 2005, about the time Green was accepted, the Army raised the limit on the so-called Category 4 recruits it would allow, the designation for soldiers with the lowest scores on its aptitude test. (Green’s score is not known.) The Army has also been handing out more waivers – includ- ing case-by-case exceptions for criminal offenses – which increased by 3 percent last year. (Childress & Hirsh, 2006)

The argument presented in this article points to the costs of adverse selection and compro- mising recruitment procedures. Discussing these dimensions of the control problem in the neutral language developed to under- stand issues that insurance companies face is analytically helpful, yet it must not inure us to the devastating consequences that the choice and control of agents have for indi- vidual lives.

Chri s topher K. But l er e t a l . SE C U R I T Y FO RC E S A N D SE X UA L VI O L E N C E 681

8 The State Department country reports note whether there are restrictions on the ICRC’s access to prisoners. If we include the access measure in the multivariate analysis in Model 1, the access measure is not significant, and the results for the vari- ables of theoretical interest are similar to those presented. The access variable does achieve significance in some of the other models where we examine the agency variables individually.

journal o f PE AC E RE S E A RC H volume 44 / number 6 / november 2007682

Appendix Table AI. Measuring Sexual Violence by Security Forces∗

Level Description

0 No mention of ‘rape’, ‘sexual assault’ or ‘sexual abuse’ by security forces. 1 Isolated reports of ‘rape’, ‘sexual assault’ or ‘sexual abuse’ by security forces. 2 ‘There were reports’, ‘there continued to be reports’ or ‘some’ reports of ‘rape’, ‘sexual assault’ or

‘sexual abuse’ by security forces, which occurred ‘sometimes’, or ‘remained a problem’. 3 There were ‘numerous reports’ of ‘rape’, ‘sexual assault’ or ‘sexual abuse’ by security forces, which

were ‘routine’, ‘common’, ‘widespread’, ‘systematic’, ‘reported repeatedly’, or ‘rape’, ‘sexual assault’, and ‘the threat of rape’ against detainees and their family members was used as a tool of torture ‘to extract information, to intimidate and to punish’.

4 ‘Rape’, ‘sexual assault’ or ‘sexual abuse’ by security forces was used as ‘a tool of war’ or ‘a systematic weapon of war’.

Source: US Department of State Country Reports. ∗The term ‘security forces’ refers to both militaries and government-supported militia, police, and prison guards. Inter-coder reliability: Two coders ranked each country. There were seven cases of disagreement resulting from their different interpretation of the language used in the reports that were ultimately resolved. Where the coders disagreed, it was by one scale point, and mostly between values of 2 and 3.

Table AII. Corruption and Sexual Violence Score by Country

Control of Sexual Control of Sexual Country Corruption violence Country Corruption violence

Afghanistan −1.35 3 Chile 1.55 0 Algeria −0.70 0 China −0.41 2 Angola −1.12 3 Colombia −0.47 1 Argentina −0.77 0 Comoros −0.73 0 Armenia −0.72 0 Congo, DRC −1.42 4 Australia 1.91 0 Congo, Republic of the −0.94 2 Austria 1.85 0 Costa Rica 0.88 0 Azerbaijan −1.07 3 Cote d’Ivoire −0.86 3 Bahrain 0.95 0 Croatia 0.23 0 Bangladesh −1.12 2 Cuba −0.13 2 Belarus −0.78 0 Cyprus 0.89 0 Belgium 1.57 0 Czech Republic 0.38 0 Benin −0.61 0 Denmark 2.26 0 Bhutan 0.91 0 Djibouti −0.73 1 Bolivia −0.82 3 Dominica 0.52 0 Bosnia and Herzegovina −0.60 0 Dominican Republic −0.39 2 Botswana 0.76 1 East Timor Leste −0.53 1 Brazil −0.05 3 Ecuador −1.02 0 Brunei 0.32 0 Egypt −0.29 3 Bulgaria −0.17 1 El Salvador −0.54 0 Burkina Faso −0.04 0 Equatorial Guinea −1.89 2 Burundi −1.02 3 Eritrea 0.04 2 Cambodia −0.90 0 Estonia 0.66 0 Cameroon −1.10 0 Ethiopia −0.35 2 Canada 2.03 1 Fiji 0.12 0 Central African Republic −1.02 3 Finland 2.39 0 Chad −1.02 2 France 1.45 0

(continued)

Table AII. (continued)

Control of Sexual Control of Sexual Country Corruption violence Country Corruption violence

Gabon −0.55 2 Myanmar (Burma) −1.37 4 Gambia −0.83 0 Namibia 0.21 0 Georgia −1.03 3 Nepal −0.30 1 Germany 1.82 0 Netherlands 2.15 0 Ghana −0.40 0 New Zealand 2.28 0 Greece 0.58 0 Nicaragua −0.44 1 Guatemala −0.71 0 Niger −1.10 0 Guinea −0.58 2 Nigeria −1.35 1 Guinea-Bissau −0.61 0 Norway 2.00 0 Guyana −0.50 0 Oman 1.03 0 Haiti −1.70 1 Pakistan −0.73 3 Honduras −0.78 0 Panama −0.24 1 Hungary 0.60 0 Papua New Guinea −0.90 1 Iceland 2.19 0 Paraguay −1.22 1 India −0.25 3 Peru −0.20 3 Indonesia −1.16 3 Philippines −0.52 2 Iran −0.38 0 Poland 0.39 0 Ireland 1.67 0 Portugal 1.33 0 Israel 1.08 0 Qatar 0.92 0 Italy 0.80 0 Romania −0.34 0 Jamaica −0.46 1 Russia −0.90 3 Japan 1.20 0 Rwanda −0.58 3 Jordan 0.00 0 Saudi Arabia 0.57 1 Kazakhstan −1.05 0 Senegal −0.17 0 Kenya −1.05 2 Serbia-Montenegro −0.80 0 Korea, North −1.18 2 Sierra Leone −0.82 2 Korea, South 0.33 0 Singapore 2.30 0 Kuwait 1.06 1 Slovakia 0.28 0 Kyrgyzstan −0.84 2 Slovenia 0.89 0 Laos −1.25 0 Somalia −1.19 2 Latvia 0.09 0 South Africa 0.36 2 Lebanon −0.34 0 Spain 1.46 0 Lesotho −0.28 0 Sri Lanka −0.14 2 Liberia −0.98 3 Sudan −1.09 3 Libya −0.82 0 Swaziland −0.26 1 Lithuania 0.25 0 Sweden 2.25 0 Macedonia −0.73 0 Switzerland 2.17 0 Madagascar 0.14 2 Syria −0.29 3 Malawi −0.91 0 Taiwan 0.81 0 Malaysia 0.38 0 Tajikistin −1.07 3 Mali −0.32 0 Tanzania −1.00 2 Mauritania 0.23 0 Thailand −0.15 0 Mauritius 0.53 0 Togo −0.68 0 Mexico −0.19 1 Trinidad and Tobago −0.04 0 Moldova −0.89 0 Tunisia 0.35 3 Mongolia −0.14 0 Turkey −0.38 3 Morocco −0.04 2 Turkmenistan −1.21 3 Mozambique −1.01 1 Uganda −0.92 2

(continued)

Chri s topher K. But l er e t a l . SE C U R I T Y FO RC E S A N D SE X UA L VI O L E N C E 683

References

Acemoglu, Daron & Thierry Verdier, 2000. ‘The Choice Between Market Failures and Corruption’, American Economic Review 90 (March): 194–211.

Amnesty International, 2004. ‘It’s in Our Hands: Stop the Violence Against Women’ (http://web. amnesty.org/actforwomen/reports-index-eng).

Amnesty International, 2006. ‘Russian Federation: Russian Chechen Friendship Society Closed Under New NGO Law’ (http://news.amnesty.org/index/ENGEUR46 0482006, last accessed 23 November 2006).

Apodaca, Clair, 2000. ‘The Effects of Foreign Aid on Women’s Economic and Social Human Rights’, Journal of Third World Studies 17(2): 205–219.

journal o f PE AC E RE S E A RC H volume 44 / number 6 / november 2007684

Table AII. (continued)

Control of Sexual Control of Sexual Country Corruption violence Country Corruption violence

Ukraine −0.96 0 Venezuela −0.94 0 United Arab Emirates 1.19 0 Vietnam −0.68 0 United Kingdom 1.97 0 Yemen −0.69 2 Uruguay 0.79 0 Zambia −0.97 1 Uzbekistan −1.03 2 Zimbabwe −1.17 3

Table AIII. Combined Effects Reducing Sexual Violence

Model A1 Model A2 Model A3 Model A4 Model A5

Military personnel (2001) 0.59 0.117 0.76 0.043 0.73 0.047 0.61 0.107 0.76 0.041 (0.38) (0.37) (0.37) (0.38) (0.37)

Ethnic fractionalization 0.33 0.462 0.36 0.422 0.27 0.544 0.29 0.507 0.34 0.438 (0.46) (0.45) (0.41) (0.44) (0.44)

Civil conflict 0.91 0.016 0.81 0.031 0.93 0.010 0.92 0.010 0.83 0.020 (0.38) (0.37) (0.36) (0.36) (0.36)

Civil war 1.56 0.092 1.40 0.123 1.59 0.041 1.61 0.082 1.47 0.109 (0.93) (0.91) (0.78) (0.93) (0.91)

Control of Corruption −0.42 0.081 −0.50 0.036 −0.54 0.022 −0.60 0.000 −0.67 0.000 (0.24) (0.24) (0.24) (0.15) (0.15)

GNI per capita −0.02 0.509 −0.02 0.479 −0.02 0.476 (0.03) (0.03) (0.03)

Polity score −0.05 0.003 −0.05 0.004 (0.02) (0.02)

High democracy −0.11 0.669 −0.12 0.617 (0.25) (0.25)

Cut point 1 0.26 0.42 0.45 0.42 0.57 Cut point 2 0.73 0.86 0.88 0.87 1.00 Cut point 3 1.42 1.51 1.52 1.58 1.67 Cut point 4 3.41 3.42 3.59 3.59 3.60

Number of observations 134 134 138 139 139 LR χ2(.) 57.24 48.72 53.17 59.40 51.45 Prob > χ2 0.000 0.000 0.000 0.000 0.000 Pseudo R2 0.18 0.15 0.16 0.18 0.16 Log likelihood −133.15 −137.42 −140.03 −136.26 −140.24

Arat, Zehra, 2002. ‘The Women’s Convention and State Reservations: The Lack of Compliance by Muslim States’, paper pre- sented at the 43rd Annual Convention of the International Studies Association, New Orleans, LA, 24–27 March.

Arrow, Kenneth J., 1985. ‘The Economics of Agency’, in John W. Pratt & Richard J. Zeckhauser, eds, Principals and Agents: The Structure of Business. Boston, MA: Harvard Business School Press (37–51).

Askin, Kelly D., 1999. ‘Sexual Violence in Decisions and Indictments of the Yugoslav and Rwandan Tribunals: Current Status’, American Journal of International Law 93 (January): 97–123.

Banks, Jeffrey S. & Barry R. Weingast, 1992. ‘The Political Control of Bureaucracies Under Asymmetric Information’, American Journal of Political Science 36 (May): 509–524.

Bardhan, Pranab, 1997. ‘Corruption and Development: A Review of Issues’, Journal of Economic Literature 35(3): 1320–1346.

Bawn, Kathleen, 1995. ‘Political Control Versus Expertise: Congressional Choices About Administrative Procedures’, American Political Science Review 89 (March): 62–73.

Beer, Caroline & Neil J. Mitchell, 2004. ‘Democracy and Human Rights in the Mexican States: Elections or Social Capital’, International Studies Quarterly 48(2): 293–312.

Beevor, Anthony, 2002. Berlin: The Downfall 1945. London: Viking.

Brehm, John & Scott Gates, 1997. Working, Shirking, and Sabotage: Bureaucratic Response to a Democratic Public. Ann Arbor, MI: University of Michigan Press.

Browning, Christopher R., 1998. Ordinary Men: Reserve Police Battalion 101 and the Final Solution in Poland. New York: HarperCollins.

Brownmiller, Susan, 1975. Against Our Will: Men, Women and Rape. London: Secker and Warburg.

Brysk, Alison, 1994. The Politics of Human Rights in Argentina: Protest, Change, and Democratization. Stanford, CA: Stanford University Press.

Bureau of Democracy, Human Rights, and Labor, United States Department of State, 2004. Country Reports on Human Rights Practices (http://www.state.gov/g/drl/hr/c1470.htm, last accessed 25 July 2006).

Burkhart, Ross E., 2002. ‘The Capitalist Political Economy and Human Rights: Cross-National Evidence’, Social Science Journal 39(2): 155–170.

Chang, Iris, 1997. The Rape of Nanking. London: Penguin.

Childress, Sarah & Michael Hirsh, 2006. ‘An Itchy Finger’, Newsweek, 7 August. (http://www. msnbc.msn.com/id/14097559/site/newsweek/, accessed 19 November 2006.)

Cingranelli, David L. & David L. Richards, 1999. ‘Respect for Human Rights After the End of the Cold War’, Journal of Peace Research 36(5): 511–534.

Davenport, Christian & David A. Armstrong II, 2004. ‘Democracy and the Violation of Human Rights: A Statistical Analysis from 1976 to 1996’, American Journal of Political Science 48 (July): 538–554.

Enloe, Cynthia H., 1983. Does Khaki Become You? The Militarisation of Women’s Lives. Boston, MA: South End.

Enloe, Cynthia H., 2000. Maneuvers: The International Politics of Militarizing Women’s Lives. Berkeley, CA: University of California Press.

Fearon, James D., 2003. ‘Ethnic and Cultural Diversity by Country’, Journal of Economic Growth 8(2): 195–222.

Gartner, Scott Sigmund & Patrick M. Regan, 1996. ‘Threat and Repression: The Non- Linear Relationship Between Government and Opposition Violence’, Journal of Peace Research 33(3): 273–287.

Gibney, Mark & Matthew Dalton, 1996. ‘The Political Terror Scale’, Policy Studies and Developing Nations 4(1): 73–84.

Gleditsch, Nils Petter; Peter Wallensteen, Mikael Eriksson, Margareta Sollenberg & Håvard Strand, 2002. ‘Armed Conflict 1946–2001: A New Dataset’, Journal of Peace Research 39(5): 615–637.

Goldstein, Joshua S., 2001. War and Gender: How Gender Shapes the War System and Vice Versa. Cambridge: Cambridge University Press.

Harrelson-Stephens, Julie & Rhonda Callaway, 2003. ‘Does Trade Openness Promote Security Rights in Developing Countries? Examining the Liberal Perspective’, International Interactions 29 (April–June): 143–158.

Chri s topher K. But l er e t a l . SE C U R I T Y FO RC E S A N D SE X UA L VI O L E N C E 685

Hathaway, Oona, 2002. ‘Do Human Rights Treaties Make A Difference?’, Yale Law Journal 111 (June): 1935–2042.

Henderson, Conway W., 1991. ‘Conditions Affecting the Use of Political Repression’, Journal of Conflict Resolution 35 (March): 120–142.

Henderson, Conway W., 2004. ‘The Political Repression of Women’, Human Rights Quarterly 26(4): 1028–1049.

Huber, John D. & Nolan McCarty, 2004. ‘Bureaucratic Capacity, Delegation, and Political Reform’, American Political Science Review 98(3): 481–494.

Kalyvas, Stathis, 2006. The Logic of Violence in Civil War. Cambridge: Cambridge University Press.

Kaufmann, Daniel; Aart Kraay & Massimo Mastruzzi, 2005. ‘Governance Matters IV: Governance Indicators for 1996–2004’, World Bank Policy Research Working Paper Series No. 3630, May (http://www.worldbank.org/ wbi/ governance/govdata/, last accessed 25 July 2006).

Keck, Margaret & Kathryn Sikkink, 1998. Activists Beyond Borders: Advocacy Networks in International Politics. Ithaca, NY: Cornell University Press.

Keith, Linda Camp, 1999. ‘The United Nations International Covenant on Civil and Political Rights: Does It Make a Difference in Human Rights Behavior’, Journal of Peace Research 36(1): 95–118.

Keith, Linda Camp, 2002. ‘Constitutional Provisions for Individual Human Rights (1977–1996): Are They More Than Mere “Window Dressing”?’, Political Research Quarterly 55(1): 111–143.

Lopez, George A. & Michael Stohl, 1992. ‘Problems of Concept and Measurement in the Study of Human Rights’, in T. B. Jabine & R. P. Claude, eds, Human Rights and Statistics: Getting the Record Straight. Philadelphia, PA: University of Pennsylvania (216–234).

Lynch, Colum, 2004. ‘U.N. Sexual Abuse Alleged in Congo: Peacekeepers Accused in Draft Report’, Washington Post, 16 December: A26.

McCormick, James M. & Neil J. Mitchell, 1997. ‘Human Rights Violations, Umbrella Concepts, and Empirical Analysis’, World Politics 49 (July): 510–525.

Marshall, Monty G. & Keith Jaggers, 2002. Polity IV Dataset. [Computer file; version p4v2002]

College Park, MD: Center for International Development and Conflict Management, University of Maryland.

Meier, Kenneth J., 1993. Politics and the Bureaucracy: Policymaking in the Fourth Branch of Government. Pacific Grove, CA: Brooks-Cole.

Mitchell, Neil, 2004. Agents of Atrocity: Leaders, Followers, and the Violation of Human Rights in Civil Wars. New York: Palgrave Macmillan.

Montinola, Gabriella R. & Robert W. Jackman, 2002. ‘Sources of Corruption: A Cross- National Study’, British Journal of Political Science 32 (January): 147–170.

Moore, Will H., 2000. ‘The Repression of Dissent: A Substitution Model of Government Coercion’, Journal of Conflict Resolution 44(1): 107–127.

Morris, Benny, 1999. Righteous Victims: A History of the Zionist–Arab Conflict, 1881–1999. New York: Knopf.

Niarchos, Catherine N., 1995. ‘Women, War, and Rape: Challenges Facing The International Tribunal for the Former Yugoslavia’, Human Rights Quarterly 17(4): 649–690.

Poe, Steven & Neil Tate, 1994. ‘Repression of Human Rights to Personal Integrity in the 1980s: A Global Analysis’, American Political Science Review 88(4): 853–872.

Poe, Steven; Sabine Carey & Tanya C. Vazquez, 2001. ‘How Are These Pictures Different? A Quantitative Comparison of the US State Department and Amnesty International Human Rights Reports, 1976–1995’, Human Rights Quarterly 23(3): 650–677.

Poe, Steven; Neil Tate & Linda Camp Keith, 1999. ‘Repression of the Human Right to Personal Integrity Revisited: A Global Cross-National Study Covering the Years 1976–1993’, International Studies Quarterly 43(2): 291–313.

Richards, David; Ronald D. Gelleny & David H. Sacko, 2001. ‘Money with a Mean Streak? Foreign Economic Penetration and Government Respect for Human Rights in Developing Countries’, International Studies Quarterly 45 (June): 219–239.

Rothman, David J. & Aryeh Neier, 1991. ‘India’s Awful Prisons’, New York Review of Books, 16 May.

Semple, Kirk, 2006. ‘Iraq Hearing on Rape and Murder Opens’, New York Times, 7 August: 9.

journal o f PE AC E RE S E A RC H volume 44 / number 6 / november 2007686

Shanker, Thom, 2004. ‘Inquiry Faults Commanders in Assaults on Cadets’, New York Times, 8 December: 18.

Silone, Ignazio, 1961. ‘Reflections on the Welfare State’, Dissent 8: 189.

Singer, J. David; Stuart Bremer & John Stuckey, 1972. ‘Capability Distribution, Uncertainty, and Major Power War, 1820–1965’, in Bruce Russett, ed., Peace, War and Numbers. Beverly Hills, CA: Sage (19–48).

Walzer, Michael, 1977. Just and Unjust Wars: A Moral Argument with Historical Illustrations. New York: Basic.

Waterman, Richard W.; Amelia A. Rouse & Robert L. Wright, 2004. Bureaucrats, Politics and the Environment. Pittsburgh, PA: University of Pittsburgh Press.

Weldon, S. Laurel, 2002. Protest, Policy, and the Problem of Violence Against Women: A Cross- National Comparison. Pittsburgh, PA: University of Pittsburgh Press.

Wood, Elisabeth J., 2004. ‘Sexual Violence During War: Explaining Variation’, paper pre- sented at the Annual Meeting of the American Political Science Association, Chicago, IL, 2–5 September.

Wood, Elisabeth J., 2006. ‘Vanation in Sexual Violence During War’, Politics and Society 34(3): 307–342.

World Bank, 2004. World Development Indicators (http://devdata.worldbank.org/ data-query/, last accessed 25 July 2006).

United Nations, 2006a. ‘In-Depth Study of All Forms of Violence Against Women’, Report of

the Secretary General. United Nations General Assembly A/61/122/Add.1 (6 July).

United Nations, 2006b. ‘Problem of Sexual Abuse by Peacekeepers’, United Nations Security Council SC/8649.

Zakaria, Fareed, 2003. The Future of Freedom: Illiberal Democracy at Home and Abroad. New York: Norton.

CHRISTOPHER K. BUTLER, b. 1972, PhD in Political Science (Michigan State University, 2000); Associate Professor, Department of Political Science, University of New Mexico (2001– ); current research interests: bargaining and conflict, fusing prospect theory and game theory.

TALI GLUCH, b. 1972, PhD student, Department of Political Science, University of New Mexico; current research interests: human rights and conflict.

NEIL J. MITCHELL, b. 1953, PhD in Political Science (Indiana University, 1983); Professor, Department of Politics and International Relations, University of Aberdeen (2005– ); current research interests: human rights and conflict, business and politics. Most recent book: Agents of Atrocity: Leaders, Follo- wers and Violation of Human Rights in Civil Wars (Palgrave, 2004).

Chri s topher K. But l er e t a l . SE C U R I T Y FO RC E S A N D SE X UA L VI O L E N C E 687