Policy preferences in a post-war environment

Abstract

Can reminders of violence committed in the past influence citizens’ policy preferences in the present? Prior work has found that under the threat of violence individuals prioritize safety and adopt policy views aimed at reducing the threat. Elites can then strategically employ concerns over personal safety and security to shape the public’s preferences. I contribute to this literature by conducting an exploratory study of whether invocations of violence committed in the past shape preferences in the long-term, years after the actual violence has ended. To do so, I fielded an experiment on a large (N = 1125) and nationally representative sample of respondents in Bosnia, the site of a major ethnic civil war in 1992–1995. I did not find evidence that reminders of wartime violence in and of themselves affect policy preferences. Ultimately, this study represents a first cut at a neglected question in the literature and has implications that could motivate future research on the relationship between violent conflict and policy preferences.

Various studies have found that the threat of politically motivated violence can directly affect citizens’ policy preferences. When confronted with such threats, individuals attach greater importance to safety and adopt policy views aimed at reducing the threat (Davis and Silver, 2004; Merolla and Zechmeister, 2009). Other work similarly shows that the threat of violence can produce or activate authoritarian attitudes (Feldman and Stenner, 1997; Hetherington and Suhay, 2011), induce hawkishness on foreign policy (Gadarian, 2010; Gershkoff and Kushner, 2005), and increase ethnocentrism (Kam and Kinder, 2007), contributing to the vote shares of hawkish, uncompromising parties and candidates (Getmansky and Zeitzoff, 2014; Oates et al., 2009). In turn, politicians can strategically employ the public’s concerns over perceived safety threats in order to shape citizens’ policy preferences or candidate evaluations (Gadarian, 2014; Gershkoff and Kushner, 2005), and their ability to do so can undermine political rights and civil liberties (Merolla and Zechmeister, 2009).

Prior research has focused on ongoing or recent threats. Does this dynamic also apply to violence that occurred in the past? While we may naturally expect memories of temporally distant violence to fade, recent studies show that violence committed in the often distant past can cast a long shadow on citizens’ social attitudes and voting behavior (Lupu and Peisakhin, 2017; Rozenas et al., 2017). Politicians, too, have exploited past conflicts for strategic purposes. Serbia’s former president Slobodan Milošević often diverted attention from his inability to improve citizens’ living conditions by referencing nationalist conflicts that occurred centuries earlier (Bieber, 2002). Russian president Vladimir Putin has often invoked World War II in order to mobilize the public against domestic and foreign adversaries (Edele, 2017). The Chinese Communist Party has similarly used memories of World War II-era Japanese atrocities in order to foster nationalism and sustain one-party rule (Coble, 2011).

However, are invocations of violence that was committed long ago actually effective at shaping what citizens care about? Can reminders of past violence keep the safety threat alive and thereby affect policy preferences decades later? In order to address these questions, I conducted an exploratory study of the relationship between temporally distant violence and current policy preferences. This study involved an experiment on a large (N = 1125) and nationally representative sample of respondents in the post-war country of Bosnia. I found that respondents who were primed on past violence did not consider safety-related policy issues any more important than respondents in the control condition or those who were primed on their ethnic identity. Moreover, this finding applied across (a) all of Bosnia’s major ethnic groups and (b) a host of moderating variables. This study represents a first cut at a neglected question in the literature on the long-term consequences of war. The findings serve to motivate future research on the relationship between violent conflict and preferences. I elaborate on the implications of these findings and potential avenues for future research in the conclusion.

Case selection and research design

I picked Bosnia as the test case for several reasons. First, the Bosnian War (1992–1995) ended more than 20 years ago, allowing me to study how reminders of past violence affect current preferences. Second, the violence was severe, widespread, and involved all three of the country’s major ethnic groups (Bosniaks, Croats, and Serbs), making the war easy to recall even years later. Finally, recent survey work indicates that concerns about future violence continue to permeate Bosnian society (Office of the UN Resident Coordinator, 2013, 2015). Therefore, given how the war continues to be socially relevant, if past violence can be invoked, engender a sense of safety threat in the present, and ultimately influence what kinds of issues citizens prioritize (safety vs. non-safety issues), this should be especially detectable in a setting like Bosnia. I discuss the case in more detail in section 1 of the Supplementary Appendix (SA).

The survey experiment was conducted in November and December of 2016 on a nationally representative sample of respondents. The fieldwork was administered by the survey firm Prism Research, whose enumerators conducted tablet-assisted, in-person interviews with 1125 adult Bosnian citizens. The full survey covered a number of topics. For this study, I focus on the part that examines policy preferences. Information concerning recruitment procedures and sampling locations can be found in SA2. Summary statistics for the sample are presented in SA3.

Assignment to experimental condition

The survey started with a basic set of demographic questions. After answering these, respondents were randomly assigned to one of three experimental conditions. The first condition was a control and respondents assigned to it immediately proceeded to the rest of the survey.

Treatment 1 (which I labeled the Violence Prime) was designed to invoke a sense of threat that was grounded in the events of the Bosnian War. Respondents in this condition were asked to reflect on the violence that was committed against their ethnic group by an ethnic out-group. While the alliance structure shifted several times during the war, significant violence occurred across all three ethnic dyads. Therefore, asking respondents about either of the two major out-groups sufficed for this study. I opted to ask Bosniaks about the violence committed against them by Serbs, Croats were asked about Bosniak violence, and Serbs were asked about Croat violence.

The design of this prime was loosely based on prior studies on the controlled recollection of past violence in experimental settings (Callen et al., 2014; Lerner et al., 2003). Note that while these studies have primarily focused on how inducing certain emotions (e.g., anger, fear, sadness) shapes preferences, I was instead interested in the simpler question of whether recalling past violence committed in the context of a civil conflict affects the importance individuals attach to safety-related policy issues in the present. Therefore, I remained agnostic about the potential emotions that reminders of past violence might induce and the precise psychological mechanisms that might explain how much importance respondents attach to certain policy areas. Indeed, a review of respondents’ answers to the Violence Prime question suggests that a number of different emotions were generated, including anger, fear, and sadness. Whether the effect on policy preferences is heterogeneous depending on what emotions are induced remains a task for future work. ¹

Treatment 2 (which I labeled the Identity Prime) asked respondents how they celebrated a recent ethno-religious holiday associated with their ethnic group. Because ethnicity and religion are overlapping cleavages in Bosnia, Bosniaks were asked about Bajram, Croats about the Catholic Christmas, and Serbs about the Orthodox Christmas. I included the Identity Prime because the Violence Prime may have increased the salience of identity among the respondents. It would have done so in a setting where some public policy issues are highly “ethnicized.” Therefore, should I have found that the Violence Prime had a significant effect on policy preferences, it might have been due to a heightened sense of threat (which is what I was interested in) or the increased salience of identity. By including a condition that (a) invokes identity but (b) does not mention past violence, I was better able to determine whether any treatment effects were due to threat perceptions or identity salience. After receiving either prime, respondents in both treatment conditions continued with the rest of the survey. ² The exact wording for both primes and a discussion of their effectiveness are presented in SA4. ³

Exercise measuring policy preferences

Post-treatment, respondents received a list of four public policy issues (child care, education, infrastructure, and policing) and were asked to rank them in order of importance. ⁴ I picked a diverse set of issues that citizens are likely to care about given their impact on people’s everyday lives. This made it more likely that respondents would carefully consider the tradeoffs involved in focusing on one issue over another.

Note that one issue related to safety (policing) while the others did not. Admittedly, the role of the police is broad and when asked about policing some respondents may think about activities that are not clearly connected to personal safety (i.e., non-violent crime). However, compared with the other policy areas included in this exercise, a prioritization of policing is the most obvious way of addressing threats to safety, including the threat posed by inter-group violence. In fact, I used policing as the safety-related issue due to the nature of the violence that occurred during the Bosnian War. Because it was primarily a civil war, examining views about foreign and military policy (the issues that are often examined in other studies) would not have been appropriate for this setting. Instead, policing is a policy area clearly linked to personal safety that might be threatened by fellow citizens, and is therefore more directly connected to the nature of the conflict in Bosnia. Indeed, several case studies on Bosnia have described how in the post-war period both policymakers and citizens have regarded policing as an important factor in reducing concerns about renewed inter-group violence (Celador, 2005; Stefanovic and Loizides, 2017).

One potential issue when examining policy preferences in a post-war setting is that institutions often lack the citizens’ confidence (Grosjean, 2014). In the case of policing, this is important because even if recalling past violence generates a sense of safety threat, this may not translate into prioritization of policing due to a widely held view that the police is an ineffective institution. However, recent survey data from Bosnia indicates that this is unlikely to be the case. In a 2015 survey commissioned by the Office of the UN Resident Coordinator, Bosnian citizens expressed more confidence in the police than in any other domestic or international institution. ⁵ This provides an additional reason why asking about policing was preferable to potential alternatives. If recalling past violence did not induce respondents to prioritize policing over other policy issues, then we can be more confident that this was not due to a lack of public faith in the police’s ability to provide adequate security. ⁶

Results

The results are presented in Figure 1. Each panel corresponds to a different pairwise comparison of experimental conditions: (a) control/Violence Prime, (b) control/Identity Prime, and (c) Identity Prime/Violence Prime. The independent variable is always binary and identifies the conditions that are being compared (Baseline = 0, Treatment = 1). I plotted the coefficient estimates for a number of model types, including ordinary least squares (OLS), ordered probit, and probit regression, and the dependent variable always measures how important respondents think policing is as a policy issue. For the OLS and ordered probit models, I employed a 4-point scale, from “least important” (1) to “most important” (4), indicating how important respondents believe policing to be. For the probit models, the dependent variable is binary and assumes a value of 1 if the respondent ranked policing as the most important policy issue (0 otherwise). Because of how the dependent variables are scaled, positive (negative) estimates indicate higher (lower) issue importance for policing. I include 95% confidence intervals for all estimates. ⁷

Figure 1.

Each panel corresponds to a pairwise comparison of experimental conditions. The independent variable is always binary and identifies the conditions being compared (Baseline = 0, Treatment = 1). The dependent variable always measures how important respondents think policing is as a policy issue, with higher values indicating greater importance. The figure displays coefficient estimates with 95% confidence intervals.

As the results show, the treatment effect for the Violence Prime was always insignificant. This was the case regardless of model type (OLS, ordered probit, or probit) or comparison condition (control or Identity Prime). In sum, I did not find evidence that recalling past violence induces citizens to attach greater importance to safety-related policy issues.

However, does this pattern hold across ethnic groups? After all, because the wording for both primes needed to correspond to respondent ethnicity, there are within-condition differences in what holiday or ethnic dyad the primes invoke. Additionally, the survey was designed to be nationally representative, and therefore, significant variation exists in the number of respondents associated with each ethnic group (54% Bosniak, 11% Croat, 35% Serb). Given this variation, is this finding driven by a particular ethnic group?

In order to explore whether differences across ethnicity exist, I partitioned the data by respondent ethnicity and replicated the main analysis for each ethnic group. The results are presented in Figure SA8.1 and indicate that little changes from the main analysis. Across every model for each ethnic group, the differences between experimental conditions in how important policing is as a policy issue were insignificant.

I further explored whether this relationship is moderated by proximity to actual violence by examining whether the treatment effects vary across (a) levels of local violence severity or (b) respondent age. To do so, I present results in Figure SA8.2 from models where I interact a binary treatment indicator with the wartime casualty rate of the respondent’s municipality of residence. I also do the same with respondent age in Figures SA8.3 and SA8.4. The results showed that the treatment effect for the Violence Prime did not vary significantly across levels of violence severity or respondent age.

Finally, in Bosnia, notable changes have occurred in the composition of local police forces over the years. While still ongoing work, making the ethnic makeup of local forces reflect the ethnic composition of localities has been an important policy priority (Coliver, 1999; Doyle, 2007). Therefore, individuals living in highly homogeneous areas (where effectively the entire police force is composed of co-ethnics) may be more trusting of the police than those living in ethnically diverse areas (where some of the police is composed of out-group members). For that reason, I examined whether the treatment effects varied across various measures of local ethnic diversity and present the results in Figures SA8.5 through SA8.10. The results once again showed that the Violence Prime did not have a significant treatment effect on policy preferences, regardless of the level of ethnic diversity.

Important to note is that the null effects I present in Figure 1 are not an artifact of insufficient sample size. With 1125 respondents, the sample is adequately large to unearth meaningful treatment effects if they in fact exist. A different concern may be that the Bosnian War is so salient among citizens that every respondent is already aware of the violence that occurred, and the null findings are therefore due to ceiling effects. Another potential issue could be that the Violence Prime does not reflect how past violence is actually invoked by elites and is therefore too weak or unrealistic to shift preferences. However, other parts of the survey suggest that ceiling effects and/or a weak treatment are unlikely to be responsible for the null findings. As noted earlier, the full survey covered several topics, including attitudes toward out-groups, political engagement, and party (ethnic and multi-ethnic) affinity. For these other outcomes, I did observe statistically significant and substantively meaningful treatment effects for the Violence Prime. Therefore, while I cannot definitely rule out these alternative explanations, the fact that I detected significant treatment effects for a number of outcomes that were examined on the same survey provides some reassurance that ceiling effects or a weak treatment do not explain the null findings in this study.

To summarize, I did not find evidence that potential safety threats grounded in past violence induce citizens to prioritize safety-related issues over ones not directly associated with safety. This is in contrast to other studies that find heightened safety concerns do lead individuals to attach more importance to some issues than to others. However, these studies have been conducted in settings where the violence is either ongoing or recent. This study was instead set in a different context (i.e., years after the violence has ended), and presents suggestive evidence that the link between safety concerns and policy preferences may not hold in the long-term.

Conclusion

This study represents a first cut at examining the relationship between temporally distant violence and current policy preferences. It presents preliminary evidence that the ability of elites to manipulate the salience of past violence to shape public attitudes may be limited. However, we should not conclude from this study that elites cannot under any set of circumstances invoke past violence to alter preferences. Rather, the findings I present suggest that recalling past violence in and of itself is insufficient to shape public policy preferences. Therefore, future work should delve deeper into this question and examine when the invocation of past violence is effective at shaping preferences and when it is not.

For instance, what role do emotions play? The wording of the treatment I employed in this study was admittedly emotionless and did not attempt to induce a specific kind of emotional response from the respondents. Indeed, a review of the respondents’ answers to the Violence Prime question indicates that it stirred up a number of emotions. However, it might be the case that invocations of violence that induce one kind of emotion are effective at shaping preferences while invocations that induce a different emotion are not.

Additionally, future studies should also consider the messenger rather than simply the message contained in these invocations. Political elites are a heterogeneous collection of individuals, and rhetorical ability, charisma, and status may all be important to understanding when references to the past are effective (or ineffective) at shaping public attitudes. Ultimately, the purpose of this study was to discern whether memories of past violence have the ability to shape current public policy preferences, but questions remain open about what conditions facilitate or hinder their ability to do so. This remains a task for future research.

Supplemental Material

Supplementary_Appendix – Supplemental material for Policy preferences in a post-war environment

Supplemental material, Supplementary_Appendix for Policy preferences in a post-war environment by Dino Hadzic in Research & Politics