Abstract
Research on civil conflict focuses primarily on identifying underlying and proximate causes while leaving many questions of subsequent social consequences unanswered. Few studies have systematically examined how these conflicts affect public opinion, especially tolerance attitudes. Additionally, cross-national comparisons reveal significant differences in political tolerance levels but few explanations accounting for this variation. In this study, I bring together these disparate literatures and demonstrate the negative, independent effects of civil conflict on political tolerance levels across thirty-two countries. Examining data from the 1995–97 World Values Survey using several statistical techniques to ameliorate problems with endogeneity and multilevel data, I find that civil conflict dampens the public’s willingness to extend basic civil liberties to nonconformist groups. By assessing the extent of domestic intolerance generated by various forms of civil conflict, this study makes important contributions to existing literatures and, more importantly, identify another obstacle to sustained peace in postconflict societies.
Death, destruction, and disorder are some of the immediate consequences of civil conflict on domestic populations. These conflicts often cripple countries while further catalyzing varying degrees of hostility throughout society. Yet, these are only the costs imposed during the conflict. Although we have a relatively firm understanding of the immediate costs of civil conflict and their effects on domestic populations, the long-term societal consequences remain unclear. Questions regarding what happens after the shooting stops are important to understanding why some societies succeed in recovering from civil conflict while others fail. In this article, I focus on one of these questions by examining how civil conflict affects divisions within a society through the prism of individual political tolerance.
The basic argument forwarded here is that civil conflict dampens political tolerance, which is an individual’s willingness to extend basic civil liberties to those whose ideas they strongly oppose or dislike (Sullivan, Piereson, and Marcus 1982). Political intolerance can generate long-standing societal problems down the road by exacerbating societal divisions. The results in this study suggest that the effect of civil conflict on tolerance may linger long after the violence ends. In societies attempting to craft postconflict political arrangements, this lingering intolerance may hamper reconciliation, and ultimately, long-standing peace.
This study makes several important contributions to our understanding of civil conflict and their consequences. First, it offers an important substantive insight into the growing literature on the social consequences of civil conflict. Previous studies examining these consequences focused primarily on the material/tangible costs associated with these conflicts or the success of postconflict settlements. With the exception of terrorism studies, less attention has been given to assessing the long-term attitudinal and behavioral consequences associated with this type of violence. Although these issues may not be perceived as critical as rectifying the economic consequences of the conflict postconflict reconstruction planning, long-term societal fractures exacerbated by domestic intolerance can certainly jeopardize future state stability. Intolerance is especially debilitating in fractured societies with weak institutional protections, which are exactly the type of postconflict states prone to conflict recurrence that Collier et al. (2003) notably identified as “conflict traps.” Understanding these attitudinal consequences may assist policy makers in finding ways to moderate societal divisions and future violence.
This study also offers an important methodological contribution to researchers interested in examining the relationship between civil conflict and political behavior. While much of the extant literature relies on individual-level or group-level hypotheses to explain civil conflict-related phenomena (e.g., conflict onset and postconflict outcomes), their empirical analyses almost exclusively use state-level indicators. Although these studies generate significant explanatory results across countries, they generally ignore the multilevel nature of the data inherent in the theoretical descriptions and, as a result, are exposed to a number of methodological biases, including ecological inference. Similarly, the tolerance literature has largely overlooked the effects of state-level factors on individual tolerance attitudes. Here I rectify these shortcomings by properly estimating how state-level civil conflict affects individual-level behavior. This approach capitalizes on the strength of the individual-level survey data while also harnessing the explanatory power of a cross-national design.
The key contribution of this study, however, is establishing the causal direction of the relationship between civil conflict and domestic intolerance. Indeed, this general association is often assumed because threat and intolerance have long been closely linked to one another, both theoretically and empirically (see Sullivan, Piereson, and Marcus 1982; Marcus et al. 1995; Sniderman et al. 2000; Shamir and Sagiv-Schifter 2006; Hutchison and Gibler 2007). Yet, the causal arrow has remained unclear because it has not been conclusively determined whether intolerance is driving conflict or conflict is furthering intolerance across society. This complicates efforts to assess the independent effects of civil conflict on tolerance attitudes. Here, I offer the first empirical attempt to systematically examine this relationship using cross-national survey data in a multilevel design while also addressing problems of endogeneity.
Using World Values Survey data from thirty-two countries, I rely on different statistical techniques designed to limit problems associated with endogeneity and multilevel data. To circumvent endogeneity problems and determine the causal direction between civil conflict and tolerance, I use an instrumental variable (IV) technique to purge the endogenous component from suspect variables. To account for the multilevel nature of the data, I also estimate these relationships using Hierarchical Linear Modeling (HLM) techniques. Correcting for these biases, I am able to gauge the independent effect of civil conflict on individual tolerance attitudes. Overall, I find lower aggregate tolerance levels in those states that recently experienced a civil war, internal armed conflicts, or fatal terrorist attacks. Furthermore, I show how much of a substantive effect each type of civil conflict has on individual tolerance.
In the sections that follow, I briefly review the literature on the causes and consequences of civil conflict highlighting the need for this type of approach to better understand long-term societal consequences. Drawing from the insights made in the political tolerance and group conflict research traditions, I generate expectations on how civil conflict generates domestic intolerance toward unpopular groups and then evaluate them using survey data from thirty-two countries. After analyzing the empirical evidence, I discuss the implications of these findings for both the civil conflict and the political tolerance literatures.
Civil Conflicts: Causes, Correlates, and Consequences
Reacting to the outbreak of well-publicized civil conflicts in the former Yugoslavia, Somalia, Rwanda, and Chechnya during the 1990s, researchers began taking a serious look at the roots of civil conflict and violence. These studies focused primarily on civil war, although other forms of internal violence such as insurgency and terrorism, also garnered attention. Given the steep rise in the occurrence and deadliness of internal conflicts since World War II, the attention is certainly warranted. Indeed, images from places like Darfur and the Congo continue to remind us of the terrible human costs associated with these conflicts. In their seminal study on civil war, Fearon and Laitin (2003, 75) note that the international system witnessed over 127 civil wars between 1946 and 1999 that resulted in over 16 million deaths.
Although a great deal is known about civil conflict, the debate over its proximate causes between grievance-based theories (Gurr 1971; Muller and Seligson 1987) and greed-based theories of civil conflict (de Soysa 2002; Collier and Hoeffler 2004; Ross 2004; Lujala, Gleditsch, and Gilmore 2005) continues to persist. One common thread between these sides, however, is their reliance on individual-based theories of agency in contributing to civil conflicts. Yet, most of the empirical analyses used to support these arguments do not explain civil conflict-related phenomena using individual-based factors. Rather, they focus on macro-level indicators to support what are actually micro-level explanations and ignore the multilevel nature of the data (Sambanis 2004). Given these methodological issues, more attention must be paid to properly specifying the empirical models to account for the separate macro- and micro-level components. Although Sambanis (2004) advocates for qualitative approaches to supplement models of civil conflict, multilevel statistical techniques are another way to generate theoretically appropriate results to help answer lingering questions relating to these events.
While the exact causes of civil conflict remain indefinite, their devastating consequences are not ambiguous. Given the levels of destruction and societal disruption caused by civil conflicts, the concentration in the extant literature on the material consequences stemming from this violence is not surprising (see Flores and Nooruddin 2009). Due to this focus, our understanding of the social consequences of civil conflict remains underdeveloped. While some research has examined related topics in areas such as public health (Ghobarah, Huth, and Russett. 2003; Montalvo and Reynal-Querol 2007), education (Lai and Thyne 2007), refugees (Salehyan and Gleditsch 2006), and social welfare spending (de Soysa and Neumayer 2008), scant attention is devoted to identifying changes in individual attitudes and behavior. Instead, the theorized effects on domestic attitudes are often assumed or discussed anecdotally and the handful of studies on this topic tend to focus more on theory building (Kalyvas 2008; Wood 2008) and single country analysis (Dyrstad et al. 2011) rather than systematic analysis across countries. That said, the work of Kalyvas (2008) and Wood (2008) offer critical insights into this complex relationship by describing the processes through which disruptive civil conflicts reconfigure individual affiliations, identities, social networks, and political attitudes.
Although restricted to the study of only one type of civil conflict, the extensive terrorism literature offers a critical mass of cross-national studies that systematically examine the influence of civil conflict on attitudes and behavior using individual-level data. This research suggests that terrorism exerts a strong influence on individuals’ political preferences and attitudes (see Davis and Silver 2004; Merolla and Zechmeister 2009). Recent work links terrorist attacks to changes in individual attitudes and orientations by noting increases in ethnocentrism (Huddy et al. 2005; Kinder and Kam 2009), authoritarianism (Hetherington and Weiler 2009), and prejudice (Merolla and Zechmeister 2009). Berrebi and Klor (2008) argue that the terrorist attacks during second Intifada affected Israeli political behavior. They link significant increases in the voting share for right-wing groups to those regions most affected by the violence. Finally, several single-country studies separately find evidence of increased support for accommodating policies to terror group demands in the United States (Karol and Miguel 2007), Israel (Gould and Klor 2010), Turkey (Kibris 2011), and Spain (Bali 2007; Montalvo 2011), thereby, demonstrating that terrorism can alter individual political preferences.
Overall, however, the terrorism literature suggests that domestic reaction does vary systematically across countries by noting the clear differences between the United States and other societies, such as Israel, Northern Ireland, and Spain (Spilerman and Stecklov 2009). Much of this variation is attributed to categorical differences in the type of attacks—single, massive attacks versus chronic campaigns—experienced by individuals in those societies (Spilerman and Stecklov 2009; Waxman 2011). This implies that differences in the scope and magnitude of the violence are important mitigating factors shaping the extent to which civil conflict affects attitudes and behavior. It also suggests that studies need to expand the analysis to evaluate whether the various forms of civil conflict, distinguished by their magnitude of violence, prompt different societal reactions.
Determining the independent effect of civil conflict on political tolerance attitudes may also shed some new light on other postconflict problems. As noted above, many studies have noted the “conflict trap” phenomenon in which states that experienced a civil conflict are more likely to experience a future conflict (Collier et al. 2003; Collier, Hoeffler, and Soderbom 2008; Elbadawi, Hegre, and Milante 2008; Walter 2004). While some of the factors associated with these “conflict traps” have been identified, certain domestic attitudes and behavior may contribute to these outcomes. 1 Widespread political intolerance is likely to exacerbate these difficulties because of its negative political repercussions, including repression (Gibson 2008) and inhibiting democratization and democratic consolidation (Dahl 1971; Inglehart 1997).
Davenport (2007) notes that strong societal democratic values inhibit state repression while Gibson (1998) contends that political intolerance may lead to the suppression of rights and political involvement of minority groups. Postconflict societies suffering from widespread intolerance directed toward the “losers” of the previous conflict and social out-groups may be more likely to turn to repression as a policy tool or seek a renewal of violence, especially given weak institutional controls typical in these environments. As Gibson and Gouws (2003, 25) observe, “[w]here institutional guarantees are weak or ineffective, the primary inhibiting factor for violent conflict must be political tolerance.” Consequently, efforts at reconciliation and amicable settlements may be undermined by mass domestic intolerance, thereby, indirectly increasing the prospects of future conflict. Understanding how civil conflict affects tolerance should help in creating more effective postconflict policies designed to ameliorate societal tensions created by the violence.
The Dampening Effect of Civil Conflicts on Political Tolerance Levels
Political tolerance is commonly defined as the willingness to allow groups whose principles or views are disliked to exercise basic civil liberties (Sullivan, Piereson, and Marcus 1982). Previous explanations of tolerance have largely concentrated on individual-level factors, such as socioeconomic characteristics (Stouffer 1955; Nunn, Crockett, and Williams 1978), political attitudes (Sullivan, Piereson, and Marcus 1982; Gibson 1992, 1998), or psychological attributes (Sullivan, Piereson, and Marcus 1982; Sniderman et al. 2000) to identify which individuals are more likely to tolerate unpopular groups.
Questions regarding the macro-level sources of tolerance have been largely left unanswered due to the reliance on single-country surveys (see Stouffer 1955; Sullivan, Piereson, and Marcus 1982; Shamir 1991; Gibson 1992, 1998; Marcus et al. 1995; Rohrschneider 1996; Mondak and Sanders 2003), which does not allow for systematic analysis of cross-national differences. Consequently, our understanding of what macro-level factors shape individual-level tolerance judgments is limited. The challenge lies in identifying those contextual factors reasonably expected to affect domestic tolerance attitudes from a myriad of possible state-level and international factors.
Studies relying on cross-national survey data reveal wide variation in tolerance levels across countries (Duch and Gibson 1992; Sullivan et al. 1993; Peffley and Rohrschneider 2003; Weldon 2006; Hutchison and Gibler 2007). A few studies have used state-level factors to explain these cross-national differences by concentrating on the influence of political institutions (Peffley and Rohrschneider 2003; Weldon 2006) but largely overlook the effects of other state-level factors. Aside from the arrangement of a country’s political institutions, the factors most likely to influence tolerance attitudes are various threats to society.
Researchers have consistently found threat to be the strongest predictor of individual tolerance (Sullivan, Piereson, and Marcus 1982; Marcus et al. 1995; Gibson and Gouws 2003; Davis and Silver 2004; Hutchison and Gibler 2007). Furthermore, this relationship holds whether threat is perceived or objectively real. In fact, support for this link dates back to Stouffer’s (1955) original study on tolerance in the United States when he observed unexpectedly low levels of political tolerance toward communists during the 1950s. This study highlighted the ability of specific groups to engender an intolerant response because of the threat they represented (Sullivan, Piereson, and Marcus. 1982). More recent literature, however, predominately examines the influence of individual threat perception. At the individual level, sociotropic threat perception, defined as threat to society as a whole, has the strongest negative effect on political tolerance. Gibson (2006, 25) asserts that “intolerance increases not necessarily when people feel their own security is at risk, but rather when they perceive a threat to the larger system of which they are a part.” Empirically, individuals respond far more negatively to those groups perceived to represent a realistic danger to the state (Marcus et al. 1995; Gibson 2006).
A small number of studies have expanded the scope of the threat–tolerance relationship by connecting states’ objective threat environment with changes in tolerance attitudes (Shamir 1991; Shamir and Sagiv-Schifter 2006; Hutchison and Gibler 2007). In each study, the threats targeted the state itself and posed a significant danger to each respective society. Intuitively, these findings make sense especially when considered in conjunction with studies showing that individuals respond more strongly to sociotropic threat than egocentric threat (Gibson 2006). An individual’s environment should have as much influence over their perception of threat as demographic or psychological characteristics. In short, context matters. Given the relationship between threat and tolerance, the general expectation here is that high objective threat levels will contribute to lower overall tolerance levels.
Further support for this claim comes from the social psychology literature on group dynamics. In their groundbreaking works, Coser (1956) and Simmel (1955) argued that groups respond with hostility to internal “renegadism,” particularly those threatening to group unity. In some instances, internal threats are perceived as more threatening than external threats from out-groups (Simmel 1955). Coser (1956, 103) described the typical group reaction to internal threats, “the perception of this inside ‘danger’ on the part of the remaining group members makes for their ‘pulling together.’” Internal threats, particularly those posing a danger to social order, foster internal cohesion of the larger group (in this case, citizens within the state). As a result, individuals increasingly value conformity and security over the civil liberties and rights of nonconformist groups regardless of whether those freedoms are institutionalized. Threats taking the form of conflict and violence generate conditions of enforced conformity throughout society (Coser 1956; Simmel 1955; also see Simon and Klandermans 2001). As Simmel (1955, 47) summarizes, “Groups in any sort of war situations are not tolerant. They cannot afford individual deviations from the unity of the coordinating principle beyond a definitely limited degree.” Under these conditions of salient threat, individuals will choose security over civil liberties and become less tolerant toward nonconformist groups who threaten national unity (Davis and Silver 2004; Shamir and Sagiv-Schifter 2006).
Identifying the type of internal threats considered salient by the domestic population and, therefore, more likely to be associated with political intolerance is the key to understanding this dynamic. Relative salience often determines whether the threat carries the necessary weight to moderate individual attitudes (Hutchison and Gibler 2007; Hutchison 2011a, 2011b; Gibler 2012; Gibler, Hutchison, and Miller 2012). To this end, I identify insurgency-based violence as one of the strongest internal threats facing states and a threat most likely to be considered as salient to the general public. In the civil conflict literature, insurgency is associated with multiple manifestations of internal violence affecting societies, ranging from civil wars to political assassinations (Fearon and Laitin 2003; Iqbal and Zorn 2006). Aside from their inherent danger to the existing social order, insurgencies, particularly those organized around ethnic divisions, engender such hostile domestic environments that even the ideas of dialogue or tolerance are unthinkable (Posen 1993; Kaufmann 1996). Under these conditions, individual social identities are organized by different principles as the public divides between those identifying with the state, those with the insurgency, and those who remain neutral (Horowitz 1985; Kalyvas 2008). 2 Group identities and intolerance become interrelated in deeply fractionalized societies as individuals often take on “us” versus “them” mentality toward nonconformist groups (see Gibson and Gouws 2003).
After splitting from the established social order, “breakaway” groups are typically perceived as “defectors” to the newly self-identified group and may trigger even stronger hostility than “normal” external threats (Simmel 1955; Coser 1956; Kalyvas 2008). Insurgencies not only fit the role of “defector” but also represent a strong sociotropic threat. Average citizens, under duress from the conflict, should be more likely to harbor intolerant views, especially toward all groups they perceive as a threat to the larger social order.
Hypothesis 1: Individuals in states threatened by insurgency-based violence are less likely to be politically tolerant than individuals in states facing little to no threat from insurgency-based violence.
Evidence supporting this contention would not only highlight an important consequence of civil conflict but also provide an important explanation for some of the systematic variation in tolerance across states. In the analyses that follow, I examine the effects of the most common manifestations of insurgency-based violence—civil war, lower-intensity armed conflicts, and terrorism—on individual political tolerance.
Endogeneity and Other Methodological Challenges
Isolating the independent effect of civil conflict on individual tolerance attitudes is the primary methodological challenge of this study because of the possibility that the relationship is endogenous. The argument of reverse causality is relatively simple to make in this case. Civil conflict and intolerance are correlated because intolerant societies beget internal threats. Mass intolerance not only makes individuals reluctant to extend civil liberties to disliked groups, but unwilling to coexist peacefully among them, thereby, increasing the likelihood of civil conflict. 3
Failing to address the possibility of endogeneity could lead to biased parameter estimates of the effects of civil conflict on individual political tolerance attitudes. Therefore, I need to purge the endogenous components from the civil conflict measures to more accurately assess the independent effect of these variables on individual tolerance in the subsequent analyses. Unfortunately, the two optimal solutions to deal with this problem—time-series analysis and/or experimental designs—are unavailable because of a lack of appropriate (i.e., panel) cross-national data (Gabel and Scheve 2007). Given these restrictions, I address this issue by adopting an IV approach. Specifically, I use two-stage least squares estimation with instrumental variables (IV-2SLS). Statistically, the IV-2SLS approach can address problems associated with reverse causality by purging the endogenous component from the suspect regressors. This statistical technique allows me to estimate the models using unbiased estimators resulting in a more accurate assessment of the relationships under examination (Baum, Schaffer, and Stillman 2007).
Although adopting the IV-2SLS approach is a straightforward method to address endogeneity, the method is not commonly used because of difficulties in satisfying the demanding specification requirements for valid instruments. This technique requires that valid instruments be correlated with the endogenous civil conflict variables but remain otherwise uncorrelated with political tolerance in general, except through its effect on civil conflict. Once identified, the IV estimation will purge the endogenous component from the regressors and generate unbiased estimations (see Gabel and Scheve 2007; Best and Krueger 2011).
To identify valid instruments, I draw from Fearon and Laitin’s (2003) seminal study. They find a strong correlation between rugged terrain and the likelihood of violent internal conflict onset. Using a country’s percentage of mountainous terrain as a proxy for rugged terrain, they demonstrate that more mountainous countries are more likely to experience violent internal conflict. The reason for this association is that rugged terrain allows the root cause of civil conflict—insurgency—to thrive because the ability of the government to stifle these groups is severely hindered in this type of territory. Mountainous terrain significantly curtails maneuverability of organized armed forces as well as offering numerous places to elude government pursuit. Here I use Fearon and Laitin’s (2003) measure of rugged terrain, which is simply a country’s percentage of mountainous terrain. 4
As a geographic variable, rugged terrain makes a good candidate for being an exogenous instrument because of a low likelihood of reverse causation. 5 However, as Baum, Schaffer, and Stillman (2007) note, endogenous regressors require multiple instruments for more efficient estimation and to reduce the risk of spuriousness. To this end, I include two additional instruments, the percentage of military personnel within the total population 6 and the percentage change in military expenditures in the five-year period prior to the survey. 7 By indicating government mobilization to threat, these two variables are associated with the onset of civil conflict. Of course, it remains possible that an intolerant society with a tendency toward conflict could affect each of these variables, thereby, potentially signifying reverse causality. To address this concern, I assess the validity of these instruments with several diagnostic tests. As I detail in the Results section, these tests clearly suggest no evidence of any independent connection, including the potential of reverse causality, between the instruments and the dependent variable.
Regrettably, this technique constrains the ability to properly account for the multilevel nature of the data because the IV-2SLS technique does not properly separate level-1 data and level-2 data. State-level data are disaggregated to the individual level which, in a normal regression, creates two problems. As Luke (2004, 7) states, “individuals belonging to the same context will presumably have correlated errors . . . by ignoring context, the model assumes that the regression coefficients apply equally to contexts.” One way to partially address the multilevel nature of the combined data set is to cluster the standard errors of the estimators by country. Baum, Schaffer, and Stillman (2007, 471) note, however, that the IV-2SLS default estimator assumes that the errors are “independently and identically distributed (i.i.d.).” In instances where clustering is used, they recommend adopting the Continuous Updating Estimator (CUE) because it “generates coefficient estimates that are efficient in the presence of the corresponding deviations from i.i.d. disturbances” (Baum, Schaffer, and Stillman 2007, 478). 8 Therefore, the CUE estimator is used to improve the models’ efficiency in the IV-2SLS analyses.
Given my other concern regarding the multilevel nature of the data, I also use HLM and nonlinear modeling to estimate the relationship between civil conflict and intolerance. These tests serve as critical robustness checks for the reported IV-2SLS results. 9 Using multilevel estimation techniques in a cross-national study is important because it avoids some of the inference problems inherent to the other alternative techniques. 10 Furthermore, replicating the substantive findings using multiple statistical techniques bolsters confidence in the overall model specifications while also capitalizing on the different strengths of each approach.
Data and Measurement
To assess the effects of civil conflict on individual tolerance attitudes, I draw on the individual-level data from the 1995–97 World Values Survey project (European Values Study Group and World Values Survey Association 2011). The WVS is one of the few cross-national surveys that measures within-state political tolerance levels using the “content-controlled” battery of tolerance questions originally developed by Sullivan, Piereson, and Marcus (1982). 11 The sample I derive from this wave of surveys range from the most developed democracies to the least developed authoritarian regimes and includes countries from several different regions, including Europe, North America, Latin America, Central and Southeast Asia, and Africa.
Dependent Variable
As mentioned above, the WVS uses the content-controlled survey instrument to measure individual tolerance (Sullivan, Piereson, and Marcus 1982). This instrument requires respondents to select their least-liked group from a list of unpopular groups. 12 The respondents are then asked whether they think that their least-liked group should be allowed to publicly demonstrate or hold political office. 13 I measure an individual’s tolerance level based on answers to these two questions. 14 Tolerant individuals would allow their least-liked group to hold office and/or publicly demonstrate. Affirmative responses to either question are coded as 1, while negative answers are coded as 0. I then use these scores to create an ordinal scale for each respondent ranging from 0 (least tolerant) to 2 (most tolerant).
Table 1 illustrates the wide variation in tolerance levels across countries. Tolerant responses to both questions were aggregated into a percentage of tolerant individuals for each country. In this sample of countries, the percentages range from the highest at 26.6 percent (New Zealand) to the lowest at 1.39 percent (Azerbaijan). A cursory examination of the tolerance distribution suggests some support for my hypothesis as many of the least tolerant states had experienced high objective threat levels prior to the survey.
Citizen Tolerance by Country.
Source: 1995–1997 World Values Survey.
Note: Values within parentheses denote the survey sample size for each individual country.
aThis index represents the percentage of respondents that provided a tolerant response to both “allow to demonstrate” and “allow to hold office” questions.
Individual-Level and State-Level Control Variables
This study focuses primarily on explaining cross-national variation of tolerance levels using state-level variables. However, not including powerful individual-level predictors of tolerance in the analyses would risk serious omitted variable bias. I use a relatively standard individual-level model of tolerance similar to Peffley and Rohrschneider (2003) and Hutchison and Gibler (2007). The subsequent models include measures for democratic ideals, democratic activism, political interest, free speech priority, conformity, media awareness, as well as measures of education, age, gender, and political ideology. These variables account for an individual’s political orientation, personality, political behavior, and socioeconomic characteristics. For the state-level control variables, I include measures for economic development, democratic longevity, and ethnic fractionalization. “Variable Descriptions and Supplemental Models”, which can be found at http://www.uri.edu/artsci/psc/hutchison.html.
Macro-Level Independent Variables
To measure states’ internal threat environment, I use a series of civil conflict variables: civil war, internal armed conflict, and fatal terrorist attacks. Each indicator offers a different measure of internal violence that varies in overall scope and intensity. As violent manifestations of unrest within the state, these variables indicate the presence and activity of direct, salient threats the domestic population. I also use a continuous measure of battle deaths resulting from internal armed conflicts to assess how fatalities affect tolerance. Finally, it is worth noting that all of these variables are lagged to ensure that the models reported below capture tolerance levels after the incidents of civil conflict.
Civil War
To assess the effect of a prior civil war, I use multiple indicators drawn from commonly used data sets in the civil war literature. Although similar in most respects, the three data sources classify civil wars using somewhat different criteria. Given the macro-level sample size, these seemingly slight differences may have a substantial effect on the analyses. 15 All three measures indicate whether a civil war occurred within five years prior to the survey for each country. I use this coding criterion to ensure significant variation within this sample because of their relative infrequency and the short temporal period covered by 1995–97 World Values Survey.
The three civil war indicators are taken from the Correlates of War (COW) Intrastate data set (Sarkees 2000), Fearon and Laitin's (2003) civil war data, and the Armed Conflict data set (UCDP/PRIO) collected by the Uppsala Conflict Data Program and the Centre for the Study of Civil Wars at the International Peace Research Institute, Oslo (Gleditsch et al. 2002). In the COW Intrastate War data set, a civil war is defined as involving at least one nonstate group fighting against a state in a militarized conflict resulting in at least 1,000 battle deaths (Sarkees 2000). Fearon and Laitin (2003) use almost identical criteria as the COW data set but differ by further specifying that the conflict average 100 deaths per year and both sides experience at least 100 deaths in the conflict. 16 Additionally, Fearon and Laitin used different criteria to note the start year of a conflict. As a result, their civil war data set is substantively different than the COW data.
The UCDP/PRIO Armed Conflict data set contains information on both large-scale civil wars and smaller-scale incidents of violence below the 1,000 battle-death threshold. In this data set, an internal armed conflict is a conflict involving a group and a state in which at least twenty-five fatalities occur over the duration. To generate a civil war indicator similar to the COW and Fearon and Laitin measures, I first limit the incidents to only “internal” armed conflicts occurring in the five-year period prior to the survey. I then calculate the total battle deaths for an entire conflict using the cumulative intensity indicator, which is derived from Lacina and Gleditsch’s (2005) estimates. If a conflict experienced a cumulative death count more than 1,000, then it is coded as a civil war.
Internal Armed Conflicts
As noted earlier, the UCDP/PRIO Armed Conflict data set includes measures of lower intensity armed civil conflicts (Gleditsch et al. 2002). I use this measure as a proxy for the presence of active insurgency groups within the state. The minimum threshold for an event to be recorded is twenty-five fatalities in any given year. Therefore, I consider this indicator as a measure of states’ lower intensity civil conflict level. The variable is the number of separate internal armed conflict events experienced in the five-year period prior to the survey. 17 The resultant measure ranges from 0 to 29 in this sample. 18
Internal Armed Conflict Fatalities
To measure the severity level of previous internal armed conflicts, I use Lacina and Gleditsch’s (2005) battle deaths data. Lacina and Gleditsch generate three estimates (high, low, and best) of the annual battle deaths for each conflict in the UCDP/PRIO Armed Conflict data set. For each conflict incident in this sample, I use the “best” fatalities estimate. In those cases where the best estimate is missing, I use the difference between the high and low estimates and divide it into half. 19 To fit with the previous civil conflict specifications, I sum the total battle deaths from internal armed conflicts in the five-year period prior to the survey. In this sample, this variable ranges from 0 to 51,248.
Terrorism
To generate an indicator of terrorist activity, I rely on data from the Global Terrorism Database (GTD; National Consortium for the Study of Terrorism and Responses to Terrorism 2010). Although other terrorism data sets are available, the GTD uses several criteria to code terrorist events that are more appropriate for this study. First, unlike the commonly used ITERATE terrorism data set, the GTD excludes acts of state terrorism and requires that the incidents be carried out by subnational actors. Second, according to the GTD codebook (2010, 5), the terrorist action also needs to “be outside the context of legitimate warfare activities.” These criteria are important to properly specify an internal threat to the state rather than those originating from abroad and better differentiate terrorism from civil wars and internal armed conflicts.
To best capture incidents most salient to the public, I focus on terrorist attacks that resulted in at least one fatality. By excluding the nonfatal attacks, I avoid attributing effects to events that are less likely to register with the public. Rare, massive, September 11th–type events notwithstanding, terrorist attacks tend to be lower intensity manifestations of civil conflict, at least with respect to fatalities. Given the more ephemeral nature of terrorist attacks compared to civil wars, I use a shorter lag for this variable because I want ensure these events have retained their salience to the domestic population. In this sample, this variable indicates the number of fatal terrorist attacks that occurred within the two-year period prior to the survey and ranges from 0 to 127.
Empirical Models and Results
In the first-stage models, estimation of the endogenous variables is conducted using the previously identified instruments discussed previously. Recall that if the instruments are significantly correlated with the endogenous internal threat variables but not with the error terms of the second-stage models, then the instruments are appropriate for the analysis. I report the three first-stage diagnostic tests which assess the validity of the instruments in Table 2.
Effect of Civil Conflict on Political Tolerance across Thirty-two Countries.
Source: 1995–1997 World Values Survey.
Note: First-stage diagnostics were estimated using ivreg2 models in Stata 10.1. Excluded instruments: Rugged Terrain, military personnel, change in military expenditures.
*p ≤ .10. **p < .05. ***p < .01.
First, the excluded instruments must account for a significant amount of the variance in the first-stage models of civil conflict. The two-stage model is only more appropriate than a standard approach when the excluded instruments effectively predict civil conflict, which I evaluate using the Stock–Yogo weak identification test (Stock and Yogo 2005). Given that I specify clustering in my models, a Wald F statistic derived from the Kleibergen–Paap rk statistic is used to test for weak instruments. 20 To verify that these instruments are not weak, the Kleibergen–Paap F statistic must exceed the Stock–Yogo critical value of 6.46, which is used for cases of three instruments and one endogenous regressor (Stock and Yogo 2005). 21 As the first-stage diagnostics in Table 2 show, the Kleibergen–Paap F statistic surpasses the critical value for each endogenous regressor, thereby, rejecting the null hypothesis that the instruments are weak. The second diagnostic is Shea’s partial R 2 for the excluded instruments. This statistic simply reports the percentage of variance of the endogenous regressor explained by the excluded instruments. Once again, in each case, the reasonably high partial R 2 reported here indicates that the instruments are appropriate predictors of the endogenous regressors.
The final, and perhaps most important, diagnostic used to evaluate the validity of these instruments is the Hansen’s J statistic. This statistic indicates whether the instruments independently predict the second-stage models’ error (see Baum, Schaffer, and Stillman 2007; Best and Krueger 2011). As I noted earlier, the excluded instruments must not predict individual political tolerance after accounting for the other independent variables in the model. In Table 2, I report the p-values of the Hansen’s J statistic for each model. Given that none of these are close to achieving statistical significance, this diagnostic indicates that the excluded instruments and the second-stage error terms are orthogonal. Considering the breadth and overall strength of these diagnostics as well as the inherent exogenous qualities of the instruments, I am highly confident in the appropriateness of using these instruments for the following analyses.
Table 2 also reports the second stage results of the IV estimation. In the second stage, the endogenous components have been purged so the independent effects of civil conflict on individual tolerance attitudes can now be estimated. As anticipated, the effect is overwhelmingly negative. In models 1, 2, and 3, I estimate only the influence of civil war on an individual’s willingness to extend basic civil liberties to their least-liked group using the different indicators. In line with my overall expectation, I find civil war has a negative effect on political tolerance across all three models. Additionally, I find no substantive difference between the three different civil war indicators suggesting that the overall independent effect of civil war on political tolerance is robust. Overall, models 1, 2, and 3 confirm my general expectation that civil wars have a negative, independent effect on tolerance levels, a relationship that I confirm across the thirty-two countries in the sample.
In terms of the control variables, these models are relatively consistent with previous research on political tolerance, particularly the individual-level factors. I find democratic activism, support for free speech, and education are positively related to tolerance while conformity, age, and females are related to intolerance. Although the political interest variable does not quite meet statistical significance in the IV-2SLS models, it does in later HLM estimations presented in Table 3. My expectation is that individuals who consume more media would be less tolerant because their increased media exposure would increase recognition of the threat to society posed by civil conflict (see Gartner, Segura, and Wilkening 1997; Karol and Miguel 2007). Furthermore, these individuals would more susceptible to government framing or depiction of the divisive civil conflict. However, I find only limited support for this contention as the effect of media awareness on tolerance is inconsistent in the IV-2SLS models and absent in the HLM analyses. The effects of the state-level controls are also inconsistent across models and estimation techniques. For example, I observe a positive relationship between democratic longevity and political tolerance in the IV-2SLS models, a finding consistent with Peffley and Rohrschneider (2003). Yet, I find no such effect in the HLM models, a finding consistent with Hutchison and Gibler (2007). Overall, however, the inconsistency of the macro-level control variables is somewhat surprising.
Effect of Civil Conflict on Political Tolerance across Thirty-two Countries.
Source: 1995–1997 World Values Survey.
Note: HLM = hierarchical linear modeling. Entries are restricted maximum likelihood coefficients and standard errors estimated with HLM 6.02. The robust standard errors are listed in parentheses.
*p < .10. **p < .05. ***p < .01.
In models 4 and 5, I evaluate the effects of lower intensity civil conflict measured here by both the number of internal armed conflicts and number of fatalities. Since these variables generally measure smaller scale conflict events than outright civil war, these models assess the overall influence of intermediate threats on individual tolerance attitudes. The results indicate that individuals in states that experience higher levels of violent insurgency activity are less likely to tolerate nonconformist groups. In Figure 1, I plot how internal armed conflicts affect individual tolerance. It shows that incremental increases in internal armed conflict events are associated with steep declines in individual tolerance before flattening at the highest levels of conflict. These findings once again support the assertion that violent internal threats independently dampen domestic tolerance levels and are consistent with the overall expectations and previous findings on other forms of salient threat (see Shamir and Sagiv-Schifter 2006; Hutchison and Gibler 2007).
Effect of internal armed conflicts on political tolerance.
In model 6, I examine how more immediate manifestations of civil conflict, fatal terrorist attacks, and influence individual tolerance across countries. Like the previous civil conflict indicators, I also find terrorist attacks to have a negative, independent effect on individual tolerance. However, given that this effect just satisfies the minimum threshold of statistical significance, this relationship may not be as robust as the previous models. Although the parameter estimates indicate that individual tolerance declines as terrorist attacks increase, a closer examination reveals that relationship is not as steeply linear as internal armed conflicts. Figure 2 shows that a small increase in terrorist activity is associated with only a minor decrease in tolerance. Only after experiencing a moderate increase in terrorist activity do tolerance levels begin to sharply decline. This suggests that domestic populations may come to view terrorist activity as a salient threat only after multiple attacks. This finding challenges some of the conventional thinking about the relationship between terrorism and tolerance and suggests that national context has a notable effect on this relationship (see Spilerman and Stecklov 2009). In certain societies, such as Israel or Northern Ireland, sporadic terrorist attacks against the population are more commonplace than in a country like the United States. This finding suggests that domestic populations may have thresholds of what constitutes as a salient internal threat shaped by their previous experience. In Israel, for example, a small number of attacks may hardly register with a public inured by its experience with chronic terrorism (Waxman 2011).
Effect of fatal terrorist attacks on political tolerance.
The previous IV-2SLS estimations show a clear and rather significant endogenous component in the relationship between civil conflict and political tolerance. Unfortunately, this technique cannot fully deal with the multilevel nature of the data. Therefore, I reestimate these models using HLM but include civil conflict variables now purged of their endogenous component. Although this solution is not optimal because I do not correct the standard errors of the purged estimators, the trade-off is that I can now appropriately account for the multilevel nature of the data (Luke 2004). It also evaluates whether the negative relationship between civil conflict and political tolerance levels is robust to changes in estimation technique.
In Table 3, I report the previous model specifications reestimated using the HLM statistical technique. Similar to the IV-2SLS models, the individual-level predictors such as democratic activism, support for free speech, and education are positively associated with political tolerance and conformity, gender, and age have negative effects. Political interest, however, is now statistically significant across all of the models, corresponding with previous findings (Peffley and Rohrschneider 2003; Hutchison and Gibler 2007). As discussed previously, the media awareness is no longer significant in any model.
Despite the change in estimation technique, the civil war variables, sans the endogenous component, are strongly and negatively associated with political tolerance in models 1-2, 2-2, and 3-2. Although some of the parameter estimates in the HLM model are different from those reported in the IV-2SLS models, the same coefficients remain statistically significant and in the expected direction further suggesting that these relationships are fairly robust. In models 4-2 and 5-2, I continue to find a negative relationship between internal armed conflict and political tolerance. Thus, despite accounting for the multilevel nature of the data, these models are consistent with the previous findings. 22
While the model looking at the effects of civil war and internal armed conflict on tolerance compare favorably with the previous models, model 6-2 reveals a far stronger relationship between fatal terrorist attacks and tolerance than in the previous IV-2SLS estimation. These results add confidence to some of the inferences that I draw from the earlier models. Further, it suggests that the relationship between fatal terrorist attacks and political tolerance is generalizable to a large cross section of countries while also supporting previous single-country studies that observe strong correlations between terrorism and intolerance (see Davis and Silver 2004; Huddy et al. 2005; Shamir and Sagiv-Schifter 2006; Merolla and Zechmeister 2009).
To gauge the extent of the substantive impact of the key variables on individual tolerance, I generate and compare the marginal effects for the civil conflict and significant individual-level variables in Table 4. 23 Holding all of the other variables at their mean, I estimate how changes in the independent variables affect the base probability of individual political tolerance. Recall that these are only rough approximations, given that I cannot fully account for the multilevel nature of the data in these analyses. However, I am confident that these results are reasonably accurate, given the similarity in parameter estimates between the multilevel and individual-level models. As Table 4 reveals, individuals in countries that experienced a civil war are between 12.6 percent and 14.2 percent less likely to tolerate their least-liked group depending on which civil war indicator is used. Internal armed conflicts and conflict fatalities exert a strong negative substantive effect on individual tolerance likelihood. I observe that when either internal armed conflicts or conflict fatalities increases from the minimum to the maximum, individuals are about 21 percent less likely to tolerate their least-liked group. Likewise, fatal terrorist attacks have similarly strong substantive effect on individual tolerance. If number of fatal terrorist attacks increases from the minimum to the maximum number, the probability that an individual would tolerate their least-liked group declines by 15.8 percent. Furthermore, Table 4 shows that the substantive effects of civil conflict are stronger than most of the individual-level tolerance predictors.
Marginal Effects of Key Predictors of Political Tolerance.
Note: All predicted probabilities were generated using Clarify software (Tomz, Wittenberg, and King 2003).
I must emphasize that the WVS tolerance measure used in this study does not ask respondents to provide their judgments toward groups specifically responsible for the internal violence in the state. Rather, these analyses examine whether civil conflict has a dampening effect on tolerance toward nonconformist groups in general. Lacking a direct measure should bias my results against observing a correlation between these internal threats and political tolerance. Thus, the fact that I find extensive support for my hypothesis suggests that the independent relationship between civil conflict and political tolerance may be stronger than reported here.
Overall, I find strong support for my general contention that civil conflict, as a salient internal threat to society, fosters individual intolerance. Additionally, these results are consistent with the overall expectations of the early social psychology literature on the effects of group conflict. Finally, they offer further evidence of a link between objective threat levels and political tolerance levels (Davis and Silver 2004; Shamir and Sagiv-Schifter 2006; Hutchison and Gibler 2007).
Conclusion
Civil conflict shapes individual tolerance decisions as citizens collectively choose conformity over civil liberties. Unfortunately, this is another sad legacy of civil conflict as societies are plagued with elevated levels of intolerance that linger on after the shooting stops. At the onset of this article, I outlined four issues I sought to address with the analyses. First, the extant civil conflict research is largely silent in evaluating what effects different types of civil conflict have upon individual attitudes, particularly on the issue of tolerance. Second, empirical analyses largely ignore the multilevel nature of the leading theoretical explanations of civil conflict. Third, while a relationship between civil conflict and intolerance has often been assumed, it had yet to be empirically tested in a systematic fashion using cross-national data. And finally, the causal direction of this relationship has been unclear due to concerns over endogeneity.
Throughout the course of this study, I addressed each of these issues through a variety of different statistical techniques and analyses. In constructing a multilevel model spanning thirty-two countries and over 20,000 respondents, I avoid the common problem of testing individual-level theories using only state-level data. The results presented here show that state-level conflict influence individual attitudes on tolerance. Individuals in states that recently experienced civil conflict are less likely to tolerate their least-liked group than are individuals in other states. While this finding is consistent with the conventional wisdom, these analyses offer empirical clarification on the casual direction of the conflict–tolerance relationship.
This article makes several important contributions to the civil conflict and political tolerance literatures. With respect to the civil conflict literature, the findings highlight another social consequence of civil conflict on the afflicted domestic populations and offer valuable policy implications for policy makers concerned with democratization. Aside from diminishing the general benefits of political tolerance, civil conflicts also decrease the likelihood of democratization and democratic consolidation. Granted, states experiencing high levels of internal strife are already less likely to democratize or consolidate democracy, but the increased divisiveness and acrimony toward nonconformist groups caused by internal threats certainly does not improve future prospects. As an “endorphin of democracy,” political tolerance represents not only a normative good but also a crucial democratic value that facilitates healthy civic culture (Gibson and Gouws 2003). Without healthy levels of tolerance and other democratic values, states run the risk of fostering repression and other abuses (Davenport 2007).
Furthermore, this study reinforces the need for future cross-national research on political tolerance that takes seriously the effects of macro-level factors on overall tolerance levels. As these results plainly demonstrate, individual attitudes on civil liberties, particularly toward nonconformist groups, are unmistakably shaped by state-level factors. While individual characteristics are critical determinants of political tolerance, I show that several state-level factors also have profound effects on these attitudes. All of this suggests that future studies must continue to explore the linkages between state-level factors and individual-level tolerance attitudes.
Finally, in demonstrating the causal direction between civil conflict and political intolerance as well as the strength of this independent effect, these findings suggest that postconflict societies face serious attitudinal constraints from their respective domestic populations. These constraints most likely hamper postconflict rehabilitation and reconstruction and offers a possible clue to unlocking the puzzle of why some societies engage in repeated civil conflicts while others settle into a longer-lasting peace. Research on “conflict traps” demonstrates that the policies and strategies implemented after civil conflicts strongly affects the likelihood of success in avoiding a recurrence of violence (Collier, Hoeffler, and Soderbom 2008). Typical postconflict solutions often fail to prioritize repairing damage to the social fabric of the state because of a narrow focus on restoring economic and political stability (Elbadawi, Hegre, and Milante 2008). Yet, intolerance influences how societal groups and individuals interact with one another affecting issues germane to postconflict societies, such as economic development and political institutions. Systematic changes in individual attitudes constrain the range of viable political alternatives in postconflict societies. Thus, for example, while UN mandates calling democratic elections after a civil war may be normatively appealing, mass intolerance may render that solution politically unpalatable to targeted minority groups in the short-term and jeopardize postconflict stability. Indeed, recently Flores and Nooruddin (2012) show how destabilizing early elections after civil wars are to their respective societies by increasing the likelihood of future conflict. Policy makers interested in healing divisions and reducing hostility must take seriously the need to account for and reduce intolerance in a postconflict society.
Footnotes
Author’s Note
I would like to thank Mark Peffley, Rich Fording, Matt Gabel, David Wildasin, Brian Krueger, Kristin Johnson, Toby Rider, Doug Gibler, Michal Shamir, the anonymous reviewers, and the editors for their helpful comments and suggestions. Of course, I am responsible for any errors that may remain. Replication materials and the web appendix can be found at 
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
Notes
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
