Corruption experiences and attitudes to political, interpersonal, and domestic violence

It is understood that corruption can change the incentives to engage in political violence. However, the scope for corruption to change attitudes toward the permissibility of violence has received less attention. Drawing on Moral Foundations Theory, we argue that experiences of corruption in the social environment are likely to shape individual attitudes toward violent behavior. Using Afrobarometer data, we document a statistically significant and sizable relationship between an individ-ual's experience of paying bribes and their attitudes to political, interpersonal, and domestic violence. These relationships are evident, and not significantly different, for men and women and are robust to the inclusion of variables capturing the local incidence of corruption, local norms regarding violence, and a proxy for the local incidence of violence with the community. Corruption is associated with permissive attitudes to violence even after controlling for the perceived legitimacy of the police and courts.


| INTRODUCTION
Corruption, commonly defined as the abuse of public power for private gain, can change the incentives that people face in relation to violence by lowering the costs and increasing the benefits. Police corruption can foster violence, particularly intimate partner violence, by deterring victims lacking in monetary or political resources from reporting crimes (Annan & Brier, 2010). In a corrupt society, control over the levers of the economy can be lucrative. For example, Baev (2018) points to corrupt elite competition for control of resources as being a key driver of Georgia's post USSR armed conflicts. Corruption also changes the trade-offs faced by potential revolutionaries by undermining economic growth (Mauro, 1995) and fostering poverty and inequality (Gupta et al., 2002). In transition economies, perceived corruption has also been shown to reduce evaluations of both local and national government performance (Moldogaziev & Liu, 2020) and even in a long standing democracy such as Sweden, perceptions of corruption undermine support for the democratic system (Linde & Erlingsson, 2013). As regard for incumbent governments or the system itself decreases, the external and internal costs faced by potential revolutionaries may also decrease. Corruption has also been shown to undermine subjective wellbeing in some contexts (Djankov et al., 2016;Gillanders, 2016;Tavits, 2008) which may lead to people feeling that they have less to lose. Indeed, the Arab Spring revolutions of 2011 can, in part, be traced to dissatisfaction with entrenched and pervasive corruption (Mansfield & Snyder, 2012).
While this compelling literature does point to a clear link between corruption and the economic, social, and institutional incentives to commit acts of violence, we know much less about how corruption changes attitudes to the permissibility violence. Social psychology points to mechanisms that can give rise to a relationship between corruption experiences to attitudes to violence. Drawing on Moral Foundations Theory, we present an argument that experiences of corruption in the social environment help to shape the moral foundations on which individuals base their attitudes and behavior. Specifically, we propose that corruption impacts the development of moral principles related to submission to authority, care and protection of others, attitudes toward abhorrent behavior or action and the coercion of less dominant individuals. In turn, these foundations provide a basis from which individual attitudes toward political, domestic and interpersonal violence develop.
This article builds on current understanding of the impact of corruption by proposing a theoretical link between corruption and the foundations of moral behavior and testing this new mechanism through which corruption can foster violence. Our study also goes beyond political violence, which, along with corruption enabled drugs violence, has been the focus of the extant literature. Using Afrobarometer data, we show that experiencing corruption is a statistically significant predictor of attitudes to violence and that the magnitude of the association is meaningful. Respondents are 7% more likely to view political violence as justifiable if they have been exposed to corrupt public officials. We show that attitudes to interpersonal violence and domestic violence are also significantly associated with exposure to corruption. Respondents who have experienced corruption are approximately 3% more likely to feel that such violence is acceptable. The fact that corruption influences attitudes to violence beyond the political domain suggests that this association is not a simple artifact of diminished state legitimacy or increased sense of grievance with government.
Our findings, particularly those relating to domestic violence, point to what is to the best of our knowledge a previously undocumented cost of corruption. Acts of petty corruption are associated with the bribe payers being more likely to view acts of violence against others, including wives and children, as acceptable in some circumstances. This finding adds to the weight of evidence that the burden of corruption tends to fall most heavily on already vulnerable groups such as victims of crime (Hunt, 2007), and infants (Azfar & Gurgur, 2008;Dincer & Teoman, 2019) and speaks to the important debate about how individually beneficial acts of corruption can still be harmful to others and on aggregate (Marquette & Peiffer, 2018;Persson et al., 2019).
Our findings are robust to the inclusion of competing explanations which are plausibly correlated with one's own likelihood of paying a bribe and attitudes to violence. Specifically, personal corruption experiences are still significant even when controlling for the local incidence of corruption, local violence norms, and a control for local violence. We also show that our results are robust to the inclusion of the perceived legitimacy of the courts and police. This suggests that our results are not simply capturing an effect of corruption on the willingness to comply with the law. While our findings support the direct relationship between corruption and attitudes to violence, contrary to previous research in other contexts (Blinkhorn et al., 2016), these relationships exist for both men and women with no significant gender difference.
In Section 2, we discuss the mechanisms through which corruption can plausibly influence attitudes to violence. Section 3 describes our data and approach, the results of which are presented in Section 4. Section 5 discusses the limitations of our study and the implications of our results for policy.

| THEORY
Violence can be defined as behavior that involves the use of physical force against oneself or another causing significant harm, injury or death (Ferguson, 2010). Although violence exists in many forms, our interest in this article is in understanding the correlates of political, interpersonal and domestic violence. Given the heavy costs and harm that is associated for unwilling victims with each of these forms of violence, they are typically viewed as an immoral behavior (Bratton, 2008). It has become clear in the extant literature that individual and collective levels of violence cannot be explained by a combination of motivation and opportunity but that factors influencing perception and ideas of violence influence the subsequent development of these immoral attitudes and behavior (Bhavnani & Backer, 2007). We argue that permissive attitudes toward violence are likely to be related to factors which influence the development of morality perceptions.
Morality refers to the social norms that guide behavior whereby certain actions are prohibited and others prescribed (Yoder & Decety, 2018). Moral Foundations Theory (Haidt, 2007;Haidt & Graham, 2007;Haidt & Joseph, 2004) proposes that moral perceptions and judgments are based on six key principles that humans are innately prepared to develop due to their alignment with recurrent social challenges (Haidt, 2012). These six moral principles or foundations are harm/care, fairness/reciprocity, ingroup/loyality, authority/respect, liberty/oppression and purity/sanctity. Of particular interest to our article are the principles of authority and respect which relates to submission to tradition and the legitimacy of authority; sanctity and purity which relates to an objection to disgusting behavior and degradation; care which relates to the protection of others and prevention of harm; and liberty which concerns the coercion or oppression of less powerful individuals.
Haidt and colleagues argue that although all humans are innately prepared to develop a moral position in relation to these principles, this position develops over time as a result of experiences of the social world (Haidt and Joseph, 2007). This idea of social environments and exposure as sharping morality is mirrored across many theories of moral development (Wheatley & Decety, 2015). The findings of Bucheli et al. (2019) that return migrants reduce violence by injecting social capital into their communities can be interpreted in this light.
Previous research has demonstrated repeatedly that experiences of corruption in the social environment are powerful in shaping attitudes and behavior. Indeed, research suggests that levels of corruption are linked to violence in the context of natural resources (Berman et al., 2017). We build on this literature to propose that experiences of corruption in the social environment are likely to influence attitudes to violence by challenging the moral foundations outlined by MFT.
In particular, the abuse of power inherent in the definition of corruption suggests that personal experiences of corruption are likely to undermine the authority foundation. Corruption is known to undermine the trust in, and the perceived legitimacy of, the state including politicians, officials, and the police (Anderson & Tverdova, 2003;Chang & Chu, 2006;Morris & Klesner, 2010;Seligson, 2002). The seminal study of Fisman and Miguel (2007) pointed to the importance of norms in determining corrupt behavior and provided an observational counterpart to the experimental finding of Abbink et al. (2002) that the threat of punishment can have a significant deterrent effect. The experimental work of Boly et al. (2019) points to a legitimacy effect whereby deterrent policies enacted by a corrupt policymaker have no effect on the embezzlement behavior of others. Importantly, Vecina and colleagues argue that a lack of deference to a legitimate authority is crucial in predicting violent behavior (Vecina et al., 2015). This may be influenced partially by the process of moral disengagement that allows people to engage in immoral actions which is commonly influenced by a displacement of responsibility to authority figures who have explicitly or implicitly condoned that action (Moore et al., 2012). In line with this, we hypothesize: Hypothesis 1. Experienced corruption is positively related to attitudes to political violence.
Although political violence has received considerable attention in the extant literature, our article also aims to examine the correlates of domestic and interpersonal forms of violence. Domestic violence refers to violence specifically directed toward an intimate partner (McClennan, 2005). In contrast, interpersonal violence is a term used to describe violence directed at another individually more generally. While the authority foundation is central to attitudes to political violence, in these more personal instances of violence, other moral foundations are likely to play a more central role. Specifically, experiences of corruption in the social environment are also likely to impact beliefs about the responsibilities of dominant individuals to protect others (care foundation) and to avoid disgusting actions (purity foundation) or coercion of less dominant individuals (liberty foundation).
Empirical research suggests that challenges to these foundations are likely to have important consequences for more personal forms of violence. Although previous work on domestic violence has not yet considered the role of corruption experiences, evidence suggests that moral foundations play an important role in predicting domestic violence. Vecina and Piñuela (2017) demonstrate that differences in moral foundations influence hostile sexism and suggest that a focus on moral foundations might be a useful mechanism for treating people who engage in domestic violence. Relatedly, the purity foundation which concerns beliefs about disgusting actions and degradation is closely linked with attitudes toward women and sexism (Sakalh-U gurlu & Glick, 2003). Furthermore, Gage (2005) reports that need for control and coercion is related to experiences of domestic violence. This need for control is likely to be underpinned by the liberty/oppression moral foundation particularly given the patriarchal power differentials that characterize many domestic violence cases (McClennan, 2005). As a result, we expect that as corruption undermines moral foundations related to care, purity, and liberty, it will influence attitudes toward domestic violence. Accordingly, we hypothesize: Hypothesis 2. Experienced corruption is positively related to attitudes to domestic violence.
The consequences we have considered thus far have related to attitudes toward the behavior of others, we propose that experienced corruption also has the potential to influence attitudes toward our own interpersonal violent behavior. One aggressive interpersonal behavior that is particularly susceptible to moral values and experiences of morality is revenge (Jackson et al., 2019). Measures of revenge include a judgment about the extent to which it is acceptable to enact revenge in response to violent crime or whether one should go to the police. In this instance, moral foundations related to authority, purity and care are all likely to be important in influencing attitudes. Specifically, experiences of corruption and abuse of power will shape individual perceptions of the need to submit to authority, avoid abhorrent behavior and protect others. Impact on the care/harm foundation of moral behavior is in line with social capital perspectives which suggest that corruption undermines people's willingness to cooperate and make themselves vulnerable to others (Bhavnani & Backer, 2007;Kelly et al., 2010). We expect the impact on these moral foundations will be associated with a more permissive attitude to interpersonal violence as a means of revenge and a decreased likelihood of turning to the authorities. In line with this, we hypothesize: Hypothesis 3. Experienced corruption is positively related to attitudes to revenge.
Although moral foundations are universally relevant to all humans, the development of moral positions on the basis of these foundations differs across individuals. Previous theoretical work on moral foundations theory argues that gender plays an important role in this process. In the empirical literature, Graham et al. (2011) demonstrate that women are more concerned with moral foundations related to harm, fairness and purity. More recent empirical work supports this distinction and demonstrates gender differences where women reported higher morality in the foundations of care/harm, fairness, sanctity and authority (Niazi et al., 2020). Gender differences in moral foundations are also supported more generally by the broader literature on moral dilemmas and ethical behavior. Women are thought to be more likely to value care/harm related considerations rather than justice related considerations when choosing action in the face of a moral dilemma (Rothbart et al., 1986). Meanwhile, men are less likely than women to judge specific business practices as being unethical or immoral (Franke et al., 1997). Similarly, neuroscience suggests significant gender differences in moral sensitivity and appraisals and call for more exploration of gender differences in moral behaviors in a field setting (Harenski et al., 2008).
Taken together, this evidence suggests that women are likely to hold higher initial positions on moral foundations, including authority, care and purity, and that those moral positions are less sensitive to change in response to external stimuli. Furthermore, women are generally less likely to hold permissive attitudes toward violence (Blinkhorn et al., 2016) and more likely to see interpersonal and particularly domestic violence as a personal threat (Jewkes et al., 2002). As a result, it is reasonable to expect that the impact of experienced corruption on attitudes toward violence will be weaker for women than for their male counterparts.
Hypothesis 4. Gender will moderate the relationship between experienced corruption and violence such that the effect will be weaker in women compared to men.

| DATA AND APPROACH
The Afrobarometer provides us with the opportunity to quantitatively test the hypotheses developed in Section 2. The Afrobarometer is a series of representative household surveys of political attitudes and economic conditions that has been carried out in an increasing number of African countries since the turn of the century. The data are collected from face to face surveys with respondents randomly chosen to ensure a nationally representative sample. The response rate to the survey averages in excess of 75%. In particular, we employ Round 2 of the Afrobarometer which was conducted in 2002 and 2003 in 16 countries in sub-Saharan Africa. We use this round for our analysis as in this round of the survey respondents were asked a series of questions relating to their attitudes to violence in three distinct arenas-political, interpersonal, and domestic.
Each of the three questions asks respondents to identify which statement they agree with from a pair capturing attitudes regarding the acceptability of violence in different contexts. Our measure of attitudes to political violence, political, takes a value of 1 if the respondent agrees (or very strongly agrees) with the statement that "In this country, it is sometimes necessary to use violence in support of a just cause" and zero if the respondent agrees (or very strongly agrees) with the statement that "The use of violence is never justified in [their country's] politics." Similarly, revenge, takes a value of 1 if the respondent agrees (or very strongly agrees) with the statement that "If you were the victim of a violent crime, you would find a way to take revenge yourself" and zero if the respondent agrees (or very strongly agrees) with the statement that "If you were a victim of a violent crime, you would go to the police for help." Finally, domestic, takes a value of 1 if the respondent agrees (or very strongly agrees) with the statement that "A married man has a right to beat his wife and children if they misbehave." and zero if the respondent agrees (or very strongly agrees) with the statement that "No-one has the right to use physical violence against anyone else." Table 1 displays summary statistics for our key variables and Table S1 in the online appendix provides full definitions. Table 1 shows that 22% of our sample hold the attitude that political violence is sometimes justifiable and that there is little difference between the attitudes of men (22%) and women (21%). Far fewer people (11%) express the view that taking revenge oneself is appropriate and once again there is little difference between the expressed attitudes of men (12%) and women (10%). However, the share of people who agree that there are circumstances in which a man can beat his wife and children is somewhat higher at 28%. There are also differences of opinion between male and female respondents with 31% the former agreeing versus 25% of the latter. Domestic violence is the domain in which violence has the most acceptance-by men and women.
It is possible that in some places, fear of revealing oneself to have committed a crime may lead people to hide their true views and so controlling for unobserved country effects in our models is important. Local norms may also be influential in this respect (and may also serve to change the mental cost of committing violence) and so we will control for the local average of each of these variables in robustness tests. We also cluster standard errors by local region. Table 2 presents the correlations between these three variables. These correlations are all rather low suggesting that people can and do hold different attitudes toward the acceptability of violence in different contexts, and that it makes sense to look at these as distinct outcomes.
Our main explanatory variable of interest captures whether the respondent has had to pay a bribe to public officials in the past year. The Afrobarometer asks a series of questions of the form "In the past year, how often (if ever) have you had to pay a bribe, give a gift, or do a favour to government officials in order to …." We create a dummy variable, bribe, which takes a value of one if the respondent reports having had to pay a bribe (with any frequency) in any of the following contexts; avoiding problems with the police; seeking to obtain documents or permits, a school placement, or a household service (such as piped water or electricity); or "anything else." bribe therefore captures a wide range of interactions between citizens and the state. 1 Table 1 shows that 25% of our sample have experienced bribery in the past year. Men are more likely to report experiencing this sort of corruption (29%) than women (22%). Much as in the case of attitudes to violence, in certain environments people may misrepresent their experiences for fear of exposure or being seen to break a norm. This is a valid concern and so it is worth noting that the bribery information in the Afrobarometer has been used in a variety of studies (Bratton & Mattes, 2003;Cho, 2012;Nunn & Wantchekon, 2011). Moreover, our inclusion of country fixed effects allows us to account for any systematic misreporting across countries.
While our main focus is on experienced corruption, we also control for individual perceptions of corruption. A perception of corruption could inform attitudes to violence independently of any actual experience and so we are interested in this relationship in its own right. By controlling for perceptions of corruption, which can differ from reality (Seligson, 2006), we can examine whether the experience predicts attitudes to violence over and above any effect of change in attitudes to the state's rulers. The survey asks "How many of the following people do you think are involved in corruption, or haven't you heard enough about them to say?" in relation to "The President and Officials in his Office," "Elected leaders, such as parliamentarians or local councillors," and "Government officials." Each of the questions can be answered "none," "some of them," "most of them" and "all" to which we assign the values from 0 to 3, respectively. Our measure of perceptions of corruption, perceptions, therefore takes values between 0 and 9 with larger numbers indicating a more pronounced sense that corruption is common in the country's leadership.
The final key control variable, history, is included to capture the respondent's history of violence within the family. Simon et al. (2001) find in their data from, the United States that recent victims of violence are more likely to display attitudinal acceptance of intimate partner violence. Moreover, people who have been exposed to violence within their family may be both more accepting of violence as a solution in general and more vulnerable to, or willing to engage in, corruption. The variable takes a value of one if the respondent answers "sometimes" "often" or "always" to the question "In your experience, how often do violent conflicts arise between people … [w]ithin your own family?" and zero if they answer "never" or "rarely." Approximately one fifth of our sample can be categorized as having a history of violence according to this criterion.
When studying attitudes to domestic violence we also employ an additional control variable, misogyny, which takes a value of one if the respondent agrees (or very strongly agrees) with the statement that "Women have always been subject to traditional laws and customs, and should remain so." and zero if the respondent agrees (or very strongly agrees) with the statement that "In our country, women should have equal rights and receive the same treatment as men do." Table 1 shows that such views are common in our sample with 33% of males and 24% of females agreeing with the first statement.
We estimate probit models in which the dependent variable is one of our three measures of attitudes toward violence. We report the marginal effects obtained from these probit models and focus on the relationship between bribe and each of the violence variables. As noted above, we include country dummies and cluster the standard errors by groups defined by the survey respondent's region. We control for age, gender, a dummy capturing an urban location, broad profession indicators, and an index of lived poverty that captures how often the respondent has had to go without food, water, medical care, electricity, fuel for cooking, and a cash income. Larger values of this last variable indicate greater levels of deprivation. We also control for education as more educated people may have different views of violence and may also be less willing to reveal permissive attitudes to violence, particularly domestic violence. Table 3 presents our main results in testing Hypotheses 1, 2, and 3. We begin in Column 1 by examining attitudes to political violence. People who have experienced bribery are 7% more likely to express the view that violence can be justified for political ends. For comparisons sake, women are 1% less likely to hold this view. Corruption perceptions are also a significant predictor, with every unit increase on our 0-9 perceptions index associated with a 1% increase in political. We can also see that history has a significant and sizable marginal effect on political. People with a history of violence in their families are 6% more likely to accept the validity of political violence. The comparable sizes of the bribe and history marginal effects reinforces the conclusion that corruption experiences play a large role in shaping attitudes to political violence.

| Main results
The second column acts as a robustness check of this result by employing a dummy variable which equals one if the individual admits to having used violence or that they would do so if they had the chance and zero if they claim that they would never do this. Both corruption variables and history are significant predictors of this self-reported willingness to engage in violence. This finding that corruption influences peoples' willingness to admit to engaging in violence for political ends adds weight to our conclusion that, on average, corruption experiences change people's attitudes toward violence. An additional explanation for this particular result is that, as noted in Section 1 above, corruption may lower the expected cost or increase the expected benefit from engaging in violence. Column 3 finds a similar relationship for the variable capturing attitudes to revenge, though the estimated marginal effect of bribe is smaller at 0.03. Once again, perceptions, history, and gender are strong predictors of one's attitude to violence with the results pointing toward both an experience of corruption and a perception of corruption increasing the likelihood that people will think it acceptable to take the law into their own hands.
The final columns of Table 3 examine attitudes toward domestic violence. Column 4 finds that those who have paid a bribe are more likely to think it is acceptable for a man to beat his wife and children under certain circumstances. Also, a greater perception of corruption is associated with holding this view regarding domestic violence. Not surprisingly, women are significantly and substantially less likely to agree with the acceptability of violence. The final column includes the additional control variable, misogyny. Holding this view is associated with a 20% increase in the likelihood of viewing domestic violence as acceptable. However, our conclusions regarding corruption are not changed by the inclusion of this variable.
Recall that (with the exception of Column 2) we are not looking at the incidence of violence which could be influenced by corruption changing the likelihood or severity of punishment. Rather we are looking at attitudes to violence. The results with respect to political and revenge can be understood in principle through the lens of disgust with the state and its agents. People who have been victims of malfeasance by state officials may view the state as less legitimate and its mechanisms for peaceful change as broken or rigged. Likewise, corruption may delegitimize law enforcement in the eyes bribe payers, leading them to view revenge actions as justifiable. These stories are less easy to tell when it comes to domestic. We interpret this to mean that corruption changes attitudes to violence over and above any change in attitudes regarding the legitimacy or efficacy of the state. We explore the legitimacy mechanism further below.
In terms of control variables, our lived poverty index is not a significant predictor of attitudes to violence. Nor is urban status except for the case of domestic which concludes that urban dwellers are less likely to view domestic violence as justifiable. More educated people are less likely to find domestic violence excusable though the pattern is less clear cut when it comes to political and revenge. Interestingly, while older people are less likely to view political violence or revenge as acceptable, older people are more likely to view domestic violence as justifiable. However when we control for misogyny, age is not a significant predictor of attitudes to domestic violence.

| Do men and women respond differently to experiencing corruption?
We have established that there is a robust correlation between corruption experiences and perceptions and attitudes to violence. We have also seen that women are less likely to view violence as acceptable. Next we test Hypothesis 4 and investigate the possibility of gender differences in response to corruption. Table 4 splits our sample by gender and estimates the baseline models from Table 3 for each group. bribe is a significant variable in each of our specifications and the estimated marginal effects for men and women are very similar. More conclusively, the interaction term between corruption experiences and gender that is included in T A B L E 4 Do men and women respond differently?
(1) Note: Probit marginal effects reported. The corresponding standard errors are clustered by region and reported in parentheses. All models include controls for age, education, poverty, and urban status. *Significance at the 10% level. **Significance at the 5% level. ***Significance at the 1% level. columns 3, 6, and 9 is not significant in any instance. We therefore conclude that, contrary to Hypothesis 4, there are no significant gender differences in the extent to which corruption experiences influence attitudes to violence. Table 5 considers potential confounding factors related to local conditions and norms. The first of these is how corrupt the respondent's locality is. Living in a corrupt area could influence both one's own chances of being asked for a bribe and one's willingness to do so. Likewise, those more deeply embedded in corrupt networks may find corruption less objectionable (Chang & Kerr, 2017) and therefore an experience of corruption may do little to undermine their moral foundations. Columns 1-3 of Table 5 address this concern by including the local incidence of corruption as an additional control variable. This variable, meanbribe, is simply the average of bribe for all those in the same region as the respondent. While the regions in the survey need not correspond with actual administrative divisions, they do allow us to capture some information as to how prevalent bribery is in the respondent's broad geographical region. While the local incidence of corruption is a significant predictor of attitudes to political and domestic violence, the inclusion of this variable does not alter our conclusions in relation to the importance of one's own experience. We next examine the role of local norms regarding the acceptability of violence. People living in areas in which attitudes to violence are more permissive may face a lower barrier or mental cost to holding and expressing such views themselves (Clark et al., 2018). Corruption could also change the local norms over time and so it is important to see if individual corruption experiences correlate with violence attitudes holding local norms constant. To this end, we average each of our outcome variables over the respondent's region. This results in three variables, politicalnorm, revengenorm, and, domesticnorm, which capture the share of people in the respondent's region who express the view that political violence, revenge, and domestic violence are acceptable. Columns 4-6 of Table 5 show that while local norms are a significant and strong predictor of individual attitudes, our results are robust to this exercise. Individual experiences of corruption are a statistically significant predictor of each of our outcomes.

| Local conditions and norms
The final local variable we consider is the extent of violence in the community. As noted above, corruption could lead to increased conflict and this in turn could influence attitudes to violence. To rule out this mechanism as being the driver of our results showing a correlation between one's experience of corruption and attitude to violence, columns 7-9 of Table 5 include a variable, violentcommunity, which captures the share of people in the respondents region who express the view that violent conflicts arise between people within the community where they live "sometimes," "often," or "always." While this variable is a significant predictor of political, it does not predict revenge or domestic. In all of our models, bribe remains a statistically significant and meaningful correlate of attitudes to violence.

| Legitimacy of law enforcement and crime
As noted above, several studies have pointed to the deleterious effect that corruption can have on trust in the state in general and on law enforcement institutions in particular. The perceived legitimacy of actors and institutions has also been shown to be important in terms of compliance with and efficacy of policy (Boly et al., 2019). Furthermore, Kirwin and Cho (2009)  Note: Probit marginal effects reported. The corresponding standard errors are clustered by region and reported in parentheses. All models include controls for age, education, poverty, and urban status. *Significance at the 10% level. **Significance at the 5% level. ***Significance at the 1% level.
T A B L E 6 Robustness-legitimacy and crime (1) (3) Note: Probit marginal effects reported. The corresponding standard errors are clustered by region and reported in parentheses. All models include controls for age, gender, education, poverty, and urban status. *Significance at the 10% level. **Significance at the 5% level. ***Significance at the 1% level.
that perceptions of state legitimacy predict attitudes to political violence using Afrobarometer data. To account for this potential effect of corruption on the perceived legitimacy of law enforcement, we generated variables reflecting respondent's attitudes to the police and the courts. The first of these, policelegit, takes a value of one if the respondent agrees or strongly agrees with the statement that "The police always have the right to make people obey the law" and zero otherwise. Similarly, courtslegit takes a value of one if the respondent agrees or strongly agrees with the statement that "The courts have the right to make decisions that people always have to abide by" and zero otherwise. The first two columns of Table 6 demonstrate that these attitudes toward the police and courts are indeed shaped in part by one's experiences of corruption with victims of corruption significantly and meaningfully less likely to view these institutions as legitimate. The local incidence of bribery, meanbribe, is not a significant predictor. We also control for the respondent's history of experiencing other types of crime (being attacked or being the victim of theft) directly or in their family. This too strongly predicts holding negative views as to the legitimacy of the police and courts.
These results reinforce the need to control for legitimacy in our analysis of the links between corruption experience and attitudes to violence as it may be the case that corruption operates on such attitudes through an effect on legitimacy. Moreover, Jackson et al. (2013) find that attitudes to violence among young males in London are associated with the perceived legitimacy of the police. The remaining columns of Table 6 do find that perceived legitimacy informs attitudes to violence in the cases of political and revenge but bribe remains a meaningful and statistically significant predictor. Over and above the potential effect on legitimacy, corruption is a significant correlate of attitudes to violence. Being a victim, or at least party, to corruption is also different to being exposed to other criminal activity as we find little evidence that crime predicts attitudes to violence, though we do find an association with attitudes to political violence in line with Kirwin and Cho (2009).

| Other rounds of Afrobarometer
Rounds three (2005-2006, 18 countries) and five (2011)(2012)(2013)34 countries) of the Afrobarometer allow us to examine whether our relationship of interest is evident in more recent data that also covers substantially more countries in the case of Round 5. Both of these later rounds contain questions that can be used to create an outcome variable that captures attitudes to political violence. 2 Round 5 also asks a question that allows us to examine attitudes to revenge. 3 While neither round has information that can be used to study attitudes to domestic violence and the contexts in which corruption is asked about can vary from round to round, Table S2 in the online appendix demonstrates that our conclusions regarding political and revenge are evident in these larger samples from later years. We conclude that the relationship between an experience of corruption and attitudes to violence is not an artifact of either the given time period or the set of countries covered in Round 2 of the Afrobarometer.

| CONCLUSIONS
Using several rounds of representative household survey data for sub-Saharan African countries, this article has shown that men and women that have experienced petty corruption are more likely to view violence as justifiable. Our results, particularly those in relation to domestic violence, uncover previously hidden costs of corruption that are not an artifact of the local incidence of corruption or violence, the prevailing norms regarding violence or erosion of the perceived legitimacy enjoyed by law enforcement. Drawing on Moral Foundations Theory we predicted that experiences of corruption undermine the moral foundations that provide a basis for our attitudes and behavior. Our findings indicate that this new theoretical mechanism is supported by direct relationships between corruption and attitudes to political, domestic and interpersonal violence. Contrary to our predictions, we did not find gender differences in these effects. The application of Moral Foundation Theory to the study of corruption opens up new avenues of research into attitudinal and behavioral responses to corruption and mis-governance.
While the Afrobarometer affords us a first opportunity to examine these previously unexplored links, it is important to note the limitations of our study. We lack a panel dimension that would allow us to control for individual fixed effects and it is conceivable that certain unobservable characteristics could lead to some people being both more likely to hold the view that violence is acceptable and being more likely to be asked for or willing to pay a bribe. Therefore, we refrain from making causal claims or statements. While the Afrobarometer is a representative survey, it must also be noted that it is representative only for the African countries covered. As always, one must be aware of the limits of any single study in terms of generalizability. Future work should explore the issue in other parts of the world, ideally with bespoke panel data. Field experiments could also test the effect of anti-corruption interventions on attitudes to and the incidence of violence.
Even with these limitations, the immense social and personal costs of violence means that our results still offer valuable insights to policymakers and civil society actors as they identify a set of previously undocumented indirect costs of corruption. To the extent that people incur disutility from carrying out acts that they view as hard to justify, our results could also be viewed as pointing to a mechanism linking corruption to violent outcomes via a reduction this psychological cost. Indeed, Shakya et al. (2017) demonstrate that changes in the acceptance of intimate partner violence predicts its incidence. Existing studies identify a substantial mental health consequences of exposure to intimate partner violence (Beydoun et al., 2012). Thus, this study also adds to the existing literature by further reinforcing that fact that while acts of corruption may benefit the individual there are spill-over effects that impose real costs (Persson et al., 2019) and that that the burden of corruption tends to fall disproportionately on already vulnerable groups (Azfar & Gurgur, 2008;Dincer & Teoman, 2019;Hunt, 2007).
We contend that this cost of corruption should be taken into account in policymaking. Our findings add further weight to the case for devoting substantial resources to anti-corruption efforts and they support the targeting of resources at particular groups. Gains and Lowndes (2018) point to the importance of institutions and state agents at the local level in the fight against gender based violence. We agree with this and further suggest that governmental and non-governmental agencies wishing to curb domestic violence have common cause with anti-corruption advocates and practitioners.

ACKNOWLEDGMENTS
We are grateful to seminar participants at the UCD Geary Institute and the DCU School of Law and Government, two anonymous referees, and the editor for helpful comments and suggestions.

CONFLICT OF INTEREST
We declare that there are no financial/personal interests or beliefs that could affect our objectivity.

DATA AVAILABILITY STATEMENT
This article uses data that are publicly available from afrobarometer.org. A file containing the Stata commands used in this paper is available on request from the corresponding author.
ORCID Robert Gillanders https://orcid.org/0000-0001-9462-0005 Lisa van der Werff https://orcid.org/0000-0003-4529-4690 ENDNOTES 1 It is not only the context of police corruption that drives our results. If we estimate our baseline models and include each context one at a time, we find that all are associated with attitudes to political violence, all bar the "anything else" context are associated with attitudes to revenge, and the contexts of documents and permits, household services and "anything else" are associated with attitudes to domestic violence. 2 In round three the survey asks "Which of the following statements is closest to your view? A: The use of violence is never justified in [your country's] politics. B: In this country, it is sometimes necessary to use violence in support of a just cause." In round 5 the survey asks "Which of the following statements is closest to your view? Statement 1: The use of violence is never justified in [your country's] politics today. Statement 2: In this country, it is sometimes necessary to use violence in support of a just cause." In both cases respondents are asked to indicate their views on a scale. We use this information in the same way as we did when using the round 2 data. 3 The survey asks "If you were a victim of crime in this country, who, if anyone, would you go to first for assistance?" We consider those who answer they "would personally take revenge" or "would join with others to take revenge" to hold the view that revenge is acceptable.