Group Membership and Individual Behavior: Microevidence

The general model of social identity we employ takes seriously a body of empirical work, mostly in economics and social psychology, which examines how group membership affects individual behavior (see Sambanis et al. Reference Sambanis, Schulhofer-Wohl and Shayo2012 and Shayo Reference Shayo2009 for reviews). This work documents a widespread tendency for ingroup bias (preferential treatment of members of one's group); higher levels of cooperation with ingroup than with outgroup members; and conformity to ingroup norms. This type of behavior corresponds to what we call social identification. However, identification is not automatic even when it comes to ethnic groups. Consider giving in dictator games. While ingroup bias has been shown to follow even an arbitrary assignment to groups (Chen and Li Reference Chen and Li2009), Habyarimana et al. (Reference Habyarimana, Humphreys, Posner and Weinstein2007) who study ethnic groups in Uganda, observe ingroup favoritism only under conditions of mutual knowledge of co-ethnicity. Fong and Luttmer's (Reference Fong and Luttmer2009) study of giving to Hurricane Katrina victims is illuminating in this respect. While on average they find no significant evidence for racial bias in giving, individuals (both whites and blacks) who “feel close” to their racial group bias their giving in favor of their racial group. Accordingly, our model seeks to go beyond demographics denoting group memberships and to incorporate the factors that affect the extent to which individuals identify with their groups.

What factors affect identification? This question has been studied intensively using the Minimal Group Paradigm (Tajfel et al. Reference Tajfel, Billig, Bundy and Flament1971)—experiments in which subjects are assigned to groups and then make anonymous allocation decisions between an ingroup and an outgroup member. The categorization into groups in these experiments often consists of highlighting a common (usually trivial) trait of the ingroup while contrasting it with the corresponding trait of an outgroup. Consequently, participants tend to indicate that they are more similar to their anonymous ingroup members than to the outgroup members. This very weak treatment is enough to lead subjects to systematically favor their ingroup members (see Brewer Reference Brewer1979 and Bourhis and Gagnon Reference Bourhis, Gagnon, Brown and Gaertner2001 for reviews). A large number of studies further vary group status. A meta analysis of 92 experimental studies of status effects (including 145 independent samples, employing both allocation decisions and reported attitudes as dependent variables) concludes that high-status group members favor their ingroup over the outgroup significantly more than do low-status group members (Bettencourt et al. Reference Bettencourt, Charlton, Dorr and Hume2001). Finally, factors that induce members of different groups to recategorize themselves as members of the same, more inclusive group can reduce intergroup bias (see Dovidio et al. Reference Dovidio, Gaertner, Shnabel, Saguy, Johnson, Stürmer and Snyder2009 for a review).

In our model, we assume that identification can be affected by the intensity of interethnic conflict. Empirical support for this assumption is found by Shayo and Zussman (Reference Shayo and Zussman2011) who study allocation decisions in naturally occurring data. The analysis is based on judicial decisions in Israeli small claims courts (where the assignment of a case to an Arab or Jewish judge is effectively random), during the period of the second Palestinian Intifada. While there is overall evidence for judicial ingroup bias, this bias is strongly associated with conflict intensity in the vicinity of the court in the year preceding the trial. Both Arab and Jewish judges display more bias towards litigants from their ethnic group as the intensity of (ethnically based) terrorism around the court increases. A complementary observation comes from Bosnia-Herzegovina during the postconflict state-building period. Alexander and Christia (Reference Alexander and Christia2011) examine behavior in a public goods game (where the dominant strategy for a purely selfish individual is to contribute nothing). Comparing students from recently integrated (Croat-Bosniak) and segregated schools, they find that while for students from segregated schools contributions are significantly higher in ethnically homogeneous groups than in mixed groups, no such difference is observed among the students from integrated schools.

Ethnicity, Identity, and Conflict

The formal depiction of identities in our model is quite abstract, but it is consistent with an understanding of social identities as labels that people can adopt subject to some constraints: they must share common attributes with other members of a social (identity) group and others must recognize them as plausible members of that group (Laitin Reference Laitin1998, 17). For ethnic identities, the literature usually assumes that these attributes are descent-based and relatively visible, therefore hard to change in the short run (Chandra Reference Chandra2006). Nonetheless, the salience of these attributes can change even in the short run (see Horowitz Reference Horowitz1985). Consistent with that understanding of ethnic identity, we assume that the number of social groups is fixed, as are the boundaries of those groups (the groups are characterized by a set of attributes that are exogenously assigned to each individual). However, categorization need not imply identification: an individual may possess attributes that place him in more than one social group, but in a given context, he may identify with only a subset of those groups.Footnote ⁴ Our depiction of identification as caring about a group's status and about one's similarity to the group is reminiscent of the commonly made assumption in political economy applications, that there is a commonality of tastes within a given ethnic group. But in contrast to standard political economy approaches, we do not assume that individuals automatically identify with any group they belong to and seek to endogenize this process.Footnote ⁵

In accordance with the behavioral evidence mentioned in the previous subsection we assume that, other things being equal, individuals will tend to identify more with higher-status groups. But status is not the only factor determining identification. In some cases, we may observe strong identification with low-status ethnic groups if perceived distance from the nation is large.Footnote ⁶ Our assumption that ethnic distance makes identification with the nation less likely (other things equal) is consistent with considerable empirical evidence. Across many countries, minorities tend to identify less with the nation than do majorities (Smith Reference Smith1986). This is also evident in empirical studies showing that members of a majority subgroup within a superordinate category are more likely than minority subgroup members to perceive their subgroup as representing the norms and values of the superordinate category (Lipponen, Helkama, and Juslin Reference Lipponen, Helkama and Juslin2003; Mummendey and Wenzel Reference Mummendey and Wenzel1999). Staerkle et al. (Reference Staerklé, Sidanius, Green and Molina2010) provide evidence from 20 countries to support this claim, as well as the more general point that subgroup identification comes at the expense of identification with the superordinate group. Manning and Roy (Reference Manning and Roy2010) study the extent to which people living in Britain think of themselves as British or if there is another identity that competes with their national (British) identity. They find that members of nonwhite ethnic groups are much less likely to think of themselves as British than are whites. But distance is not the only factor. Immigrants from poorer and less democratic (i.e., lower status) countries assimilate faster into a British identity than do immigrants from richer countries. Laitin (Reference Laitin1998) finds evidence that Russian minorities in the near-abroad shortly after the Soviet Union's collapse were more open to assimilation to the majority culture of the titular nation if the state they lived in was richer (more in Latvia and Estonia than in Kazakhstan). Yet social distance from the majority population also seemed to matter (more assimilation in Ukraine than in Kazakhstan). These empirical results highlight the importance of taking into account both distance and status effects, as well as the basic tradeoff between ethnic and national identities.

While distance and status affect identification, in the context of ethnic conflicts these factors cannot be taken as exogenously given. Conflict destroys and reallocates national resources between groups, which can directly affect both national and group status. Furthermore, an intense conflict fought along ethnic lines is likely to make ethnic attributes more salient to individuals. Indeed, Appadurai (Reference Appadurai2006) analyzes the genocide in Rwanda as a form of “community building”—a strategy designed to strengthen the identity of the perpetrators’ group; and Kaufman (Reference Kaufmann1996) goes so far as to argue that ethnic war causes permanent mistrust among ethnic groups.Footnote ⁷ While Manning and Roy (Reference Manning and Roy2010) find that in general whites are more likely than nonwhites to identify as British, national identification appears to be sensitive to patterns of conflict: Catholics from Northern Ireland rarely think of themselves as British while a majority of Protestants do.

Our model captures key insights of Horowitz's (Reference Horowitz1985) classic work Ethnic Groups in Conflict. Although Horowitz does not develop a formal model, he traces out carefully the implications of social identity theory for ethnic conflict. We share with Horowitz assumptions about the importance of status-seeking, group-derived self-esteem, and cognitive-psychological mechanisms in ethnic identification. Our arguments stand in contrast to purely instrumental theories of identity change, particularly as they apply to the relationship between conflict and ethnic identification (e.g., Kalyvas Reference Kalyvas2008). While we do not dispute the empirical relevance of minimum winning coalition arguments about how ethnicity can structure sometimes violent competition over resources (Bates Reference Bates, Rothchild and Olorunsola1983, Caselli and Coleman Reference Caselli and Coleman II2013, Fearon Reference Fearon1999, Posner 2004), we do not think that they capture all aspects of the process of social identification even where instrumental motives can be shown to be relevant.Footnote ⁸ Some empirical cases that are commonly interpreted as supportive of instrumental theories of ethnic mobilization are also consistent with the psychological mechanism that we propose. For example, Eifert, Miguel, and Posner (Reference Eifert, Miguel and Posner2010) use Afrobarometer survey data to show that electoral competition increases ethnic identification in ten African countries. However, as the authors note, the data do not allow them to attribute this pattern to mobilization of ethnic identities by elites in pursuit of electoral advantage. In fact, the results are also consistent with a noninstrumental, social identification mechanism: ethnicity might become more salient during periods when political events are likely to affect the relative status of ethnic groups. Yet in this and many other studies, the psychological mechanism is simply assumed away.

Within political science and sociology, scholarly debates between “primordialists,” who view ethnicities as fixed, acquired by birth, and “instrumentalists,” who argue that individuals can choose their ethnic identity according to their self-interest, have shifted the focus away from the basic fact that ethnicity is both malleable and strong. Donald Horowitz (Reference Horowitz2001, 52) best captures this view of ethnicity as “inspissate: a thickened version of a substance, the origins of which are invariably much thinner, a substance that a profound change of context may thin out yet again. A view of ethnicity as a strong affiliation is not incompatible with variation in the boundaries of groups and in their political saliency.” Pure instrumentalists emphasize individual calculus in an identity marketplace in which ethnic entrepreneurs can create and sell new identity categories to willing buyers. The usual implication here is that ethnicity does not matter; material interests do. Constructivist scholars share with instrumentalists an understanding of ethnic identity as flexible and changing, but unlike instrumentalists they do not focus on identity change as an individual choice as much as the result of a shifting social context, which shapes identity alternatives for individuals. Both approaches reflect the new consensus: “construction and choice, rather than blood and inheritance, is now the standard story line about identities” (Laitin Reference Laitin1998, 12). Our approach captures some key features of the new consensus, such as the instrumentalists’ assumption that the material benefits of social identity matter: in our model, individuals want to identify with high-status groups. However, unlike purely instrumental accounts, we show how ethnic attributes constrain identity choice and model a social identification process in which individuals also take into account how closely their attributes match the average attributes of the group they would like to identify with.

A related perspective on identity formation and change is Laitin's (Reference Laitin1995; Reference Laitin1998) tipping model in which “a critical mass of individuals [must] assimilate successfully before it is rational for any individual to do so” (Laitin Reference Laitin1995, 54). In that model, assimilation resembles a coordination problem. The problem is more severe for extremely marginalized minorities (e.g., Gypsies in Spain; Untouchables in India; Jews in medieval Europe). Laitin (Reference Laitin1995) argues that marginal groups choose not to assimilate because of economic benefits that only accrue to them by virtue of their ethnic minority status. Laitin's informal theory explains how this process can generate multiple equilibria: if the tipping point is not reached, ethnic group boundaries remain intact; if the balance tips toward assimilation, then the minority group can disappear. The logic of this argument is powerful, though one can conceive situations in which different incentives prevail and instead of a tipping model we observe competitive assimilation if, for example, there are first mover advantages to switching identities and early switchers benefit materially from erecting a wall to block further assimilation once they themselves have switched. Where we differ from Laitin is that we view identity as not just a focal point around which people coordinate. Rather, the desire for higher status in our model is counterbalanced by an individual's ethnic distance from the group that he would prefer to identify with. Multiple equilibria can arise endogenously as the result of uncoordinated action through a mechanism that puts ethnic attributes and psychological attachment at the forefront while also allowing for instrumental motives rather than the other way around.

In Laitin's theory, a social hierarchy generates disincentives for identity change through policing of boundaries or social stigma attached to those who switch and social (religious, linguistic) distance also influences decisions to assimilate, as his analysis of Russians living in the near abroad has shown. Minority group elites do much of the work, driven by how the assimilation process affects their own material interests. Elite preferences are exogenous and the theory concentrates on the elites’ material interests over emotional attachment to the group. This is common to most other models of ethnicity and conflict, where elite preferences drive outcomes. In de Figueiredo and Weingast's (Reference De Figueiredo, Weingast, Walter and Snyder1999) model of ethnic war, ethnic “entrepreneurs” use violence to create fear among their co-ethnics so that they can manipulate them and mobilize them in a fight that serves the elites’ material interests and political aspirations. This is perhaps the dominant narrative of violent ethnic conflict—from war to riots.Footnote ⁹ Fear of a hostile other, combined with strategic vulnerability, is also what drives Posen's (Reference Posen and Brown1993) ethnic security dilemma, according to which any action that a vulnerable ethnic group might take to improve its security in an environment of emerging anarchy is likely to generate a preemptive strike by members of the outgroup as it increases fear. Such violence generates incentives for members of the outgroup to work together so as to make their group more defensible. In effect, violence confirms individuals’ worse fears and leads to ethnic polarization (Kaufmann Reference Kaufmann1996).

All of these accounts must assume that individuals are easily manipulated by elites. While we acknowledge that fear might push people to make the “wrong” decisions and bargaining failure between elites is often an important cause of conflict, our focus here is on understanding processes operating at the individual level, so we want to relax the assumption that elites are rational and maximize their selfish interests, whereas the masses are susceptible to biases.Footnote ¹⁰ We therefore choose not to rely on an assumed asymmetry between elites and masses and rather try to map out a set of conditions that would constrain the elites’ ability to sway the public. While elites may well understand the way conflict can affect identification patterns and try to use violence strategically, what our analysis suggests is that the unravelling of a peaceful equilibrium can occur even without such an understanding. This opens the way for models of conflict in which elites are fully endogenous and rise to protect the group's interests when those are threatened.

Finally, by modeling an endogenous process of social identification, we can suggest a new mechanism for how extremists can escalate conflicts, dragging moderates into the fight. This mechanism is distinct from coordination models of revolution (e.g., Kuran Reference Kuran1989) as well as models in which vanguard violence provides information about the parties’ preferences or relative strength or the extent of public opposition to a regime, thereby shaping the public's estimates of the probability of success of a revolution (Baliga and Sjöström Reference Baliga and Sjöström2012; Bueno de Mesquita Reference Bueno de Mesquita2010, Reference Bueno de Mesquita2011; Lohmann Reference Lohmann1993, Reference Lohmann1994). While these informational mechanisms are important and can illuminate how some conflicts erupt, they are less relevant as explanations of conflict intensity or duration since uncertainty about type or preferences should be resolved with fighting. In the section on intragroup cleavages we examine the possibility that an ethnic group gets divided into extremists and moderates, but the extremists do not possess any private information that they may signal to the rest of the population. Nor does violence itself convey information to group members who would like to join the fighting if they only knew that a sufficient number of people would also join. Rather, violence changes the actual social environment individuals operate in, and thereby affects their preferences. Furthermore, the extremists themselves may be motivated by more than just material gains. Indeed, it is hard to make sense of nationalist leaders such as Milosevic or Tudjman, or terrorist vanguard groups such as Hezbollah or Hamas, and to understand their impact, without reference to their ethnic identification. Thus in our model the vanguard does not necessarily act purely instrumentally and violence cannot effectively mobilize co-ethnics if they do not have any emotional attachment to their ethnic group.Footnote ¹¹

While our model departs in important ways from conventional ways to analyze conflict, we want to emphasize that information, fear, elite manipulation, and strategic or instrumental motives are all important explanations for ethnic mobilization and conflict. Our contribution is to highlight an additional, psychological, explanation that has been neglected thus far, especially in formal equilibrium models. We do this by means of a simple model that combines behavioral regularities at the individual level—based on microlevel research in psychology and economics—with the tools of equilibrium analysis.

Conflict

Consider a country inhabited by a set N of individuals, half of whom are members of group A and the other half are members of group B.Footnote ¹² Later on, we allow for subgroups with potentially different interests, but for now think of the groups as homogeneous.

The two groups are contesting a resource of value V. This resource is divided between the two groups according to a standard contest success function and can support consumption of group-specific (club) goods. V can be actual “lootable” resources, e.g., natural resources that are not effectively controlled by the state and can be captured by the use of force. But V can also be thought of as those policies or government resources that can, given the institutional setting, be used to exclusively benefit a specific group and whose distribution can be influenced by the use of force.Footnote ¹³ Finally, while the state is not modeled as a separate actor, one could think of this setting as describing group competition over capture of the state (or competition in a near-anarchic, weak state environment). One could also consider the state as being dominated by one of the groups.

Let y_i > 0 be individual i’s exogenous resource endowment and denote by f_i individual i’s contribution to his group's struggle or his “fighting effort.” Individual i’s fighting effort is constrained to f_i ∈ [0, y_i ]. This restriction will not be particularly interesting in the baseline model, but will play an important role once we consider intragroup cleavages. Denote the aggregate effort of each side by F_J = ∑ _{i ∈ J} f_i for J = A, B and let F = F_A + F_B . Similarly, denote Y = Y_A + Y_B where Y_J = ∑ _{i ∈ J} y_i for J = A, B. We assume Y_J ≥ V/2 to ensure that a group's fighting effort is not simply determined by its resource constraint. Material payoffs when F > 0 are given by

(1) \begin{eqnarray} \pi _{i}(f_{i},f_{-i})=y_{i}-f_{i}+\frac{F_{J}}{F}V\text{\ \ \ \ \ \ for \} i\in J,\text{}J\in \lbrace A,B\rbrace. \nonumber\\ \end{eqnarray}

F provides a straightforward measure of the intensity of the conflict. It should be interpreted broadly as the value of any resources devoted to capturing the contested resource—from demonstrations and strikes to armed combat. “Fighting” in this model should thus be thought of as encompassing both violent and nonviolent conflict, with violent conflict being more intense.

The model is solved by assuming that individuals decide simultaneously on their fighting effort, taking as given the decisions of other individuals in the country.Footnote ¹⁵ This provides a simple framework for thinking about conflict as a phase of social life in which members of each group shift some of their productive capacities to appropriative struggle against the other group and resources are destroyed. We do not consider in this article the question of why conflict arises in the first place (Fearon Reference Fearon1995; see Jackson and Morelli Reference Jackson and Morelli2011 for a recent review). Rather, we assume that this can occur for a number of reasons, such as informational asymmetries or credible commitment problems. Our analysis allows, however, for some variation in the severity of the bargaining failure, which can be captured by variations in V, namely how large are the resources that are in fact up for grabs.

Absent any social identity considerations, one can verify that any pure strategy Nash equilibrium has F_A = F_B = V/4 and the intensity of conflict is F = V/2.Footnote ¹⁶

Social Identity

The model of social identity has three building blocks. First, a specification of the social groups or categories that exist in a given society. Second, the perceived distance between each individual and the typical member of his group. And third, the relative status of each group.

Social Groups . A society is characterized by a given set of social groups G. Social groups are categories that individuals learn to recognize as a result of being socialized in a given context. In principle, any combination of attributes could result in a social group, but in a given political context only some of these are likely to be relevant. We do not model the cultural or sociological process by which these categories evolved. Rather, we focus on the process of identification with a given set of social groups. For the issue at hand, a potentially important source of identification in addition to the two ethnic groups A and B is the country or the nation as a whole, N. The two groups together make up the nation, but not all countries will have a strong national identity. The set of social groups is thus G = {A, B, N}. Denote by G_i the set of social groups to which agent i belongs. Thus, G_i = {A, N} for i ∈ A and G_i = {B, N} for i ∈ B.

Perceived Distance . Each individual is characterized by a vector of attributes (or qualities). Assume, for now, that there is no within-group heterogeneity. Thus, there is a set of attributes shared by all residents of the country and a set of group-specific attributes shared by all the members of each ethnic group. For simplicity we can write all the country-wide or group-specific attributes as a single binary variable:

\begin{eqnarray*} q_{i}^{n}=\left\lbrace \begin{array}{l@{}l}1 & \text{if}i\in N \\ 0 & \text{otherwise} \end{array} \right. \ \ \text{and}\ \ \ q_{i}^{e}=\left\lbrace \begin{array}{l@{}l}1 & \text{if}i\in A \\ 0 & \text{if}i\in B \end{array} \right.{\!\!}. \end{eqnarray*}

(2) \begin{eqnarray} d_{iJ}^{2}=w_{n}\big(q_{i}^{n}-q_{J}^{n}\big)^{2}+w_{e}\big(q_{i}^{e}-q_{J}^{e}\big)^{2}, \text{\ \}J\in \lbrace A,B,N\rbrace, \nonumber\\ \end{eqnarray}

The concept of perceived distance and the associated attention weights originate in the study of categorization in cognitive psychology (see Nosofsky Reference Nosofsky1986). The idea there is that a stimulus i is less likely to be categorized as a member of category J the larger the perceived distance between them. Following Turner et al. (Reference Turner, Hogg, Oakes, Reicher and Wetherell1987) we adopt this idea to the process of categorizing oneself into a group. The weights capture the way attention is divided between the different dimensions involved in this categorization process. Importantly, which dimensions are more salient may depend heavily on the context in which judgements are made.Footnote ¹⁷ As we shall see, this helps us analyze a central insight of the political science literature on ethnicity and nationalism, which is that not all social groupings are pertinent at any given moment in time.Footnote ¹⁸ It should be clarified, however, that “salience” in our model means that certain attributes (e.g., the attributes that distinguish ethnic groups) are cognitively more important—it does not mean that people will necessarily identify ethnically in equilibrium, which is what salience often implies in other comparative politics studies of ethnicity and conflict.

Notice that since we are representing all common national (ethnic) attributes by a single variable, these weights capture not just the attention paid to one particular attribute, such as speaking the common national language, but the overall salience of all such common national (ethnic) attributes.

As we are interested in ethnic groups, for the most part we take the attributes themselves as fixed.Footnote ¹⁹ However, we allow their salience to change. In particular, the relative salience of ethnicity may increase with the intensity of conflict between the two ethnic groups. To keep the model simple, we assume a linear relation.

Assumption 1 . The salience of ethnic attributes weakly increases with the intensity of interethnic conflict. Specifically, w_e = η₀ + η₁ F where η₀ ∈ [0, 1], η₁ ≥ 0, and η₀ + η₁ Y ≤ 1.

The parameter η₀ is the relative salience of ethnic attributes (and 1 − η₀ is the relative salience of common national attributes) under zero conflict. It captures the various factors—other than the intensity of conflict itself—which affect the attention paid to ethnic versus national attributes. The parameter η₁ ≥ 0 captures the sensitivity of the salience of the ethnic cleavage to the intensity of conflict.Footnote ²⁰ Finally, to keep attention weights between 0 and 1, η₀ and η₁ are restricted so w_e ≤ 1 even at the maximal possible level of fighting.

Group Status . Social identification involves more than perceptions of similarity to other group members. It also includes an affective component that relates to the “value” or status of the group. And the evaluation of groups is often determined through social comparisons to other groups along valued dimensions of comparisons (Tajfel and Turner Reference Tajfel and Turner1986). Let −J be the reference group of group J and assume that one of the dimensions for evaluating groups is their material payoffs. Specifically, let Π _J measure group J’s total resources:

\begin{eqnarray*} \Pi _{J} &=&Y_{J}-F_{J}+\frac{F_{J}}{F}V\quad \text{for}J=A,B, \\ \Pi _{N} &=&Y-F+V. \end{eqnarray*}

(3) \begin{equation} S_{J}=\sigma _{J}+\Pi _{J}-\Pi _{-J},\text{\ \}J\in \lbrace A,B,N\rbrace, \end{equation}

Social Identification . We now come to our definition of social identification. We say that individual i identifies with social group J if (a) he prefers outcomes where group J’s status is high to ones where it is low; and (b) he prefers outcomes where his perceived distance from group J is low to ones where it is high. In other words, social identification is taken to mean caring about one's group and seeking to resemble other members of that group. The latter may be interpreted as a cognitive cost of categorizing oneself as a member of a group that is different from oneself, as the cost of being less well accepted by other group members the more different one is from that group, or as intragroup difference aversion.

The utility function of an agent i that identifies with group J takes the form

(4) \begin{equation} U_{i}(f_{i},f_{-i};\, J)=\pi _{i}-\beta d_{iJ}^{2}+\gamma S_{J}, \end{equation}

Three points concerning this formulation are worth highlighting. First, the specific group with which an individual identifies is going to be determined in equilibrium. In this sense (and despite the linear specification), d_iJ will mediate the extent to which individual i cares about group J’s status, since (other things being equal) high perceived distance from group J will make identification with J less likely. And similarly, S_J is going to mediate the desire to reduce the perceived distance from group J. But this is an equilibrium phenomenon: individuals may well see their commitment to a particular group as fixed, even if over time it is fluid. We return to this point when discussing our equilibrium concept. Second, we use the concept of utility as it is used in standard economics, namely as an ordinal index that describes how the individual ranks alternatives (i.e., whether alternative X is preferred to alternative Y or alternative Y is preferred to alternative X). Different identities mean different rankings of alternatives: identifying ethnically and identifying with the nation will imply different desired levels of fighting. Our focus is on which of these rankings prevails. While an individual might feel some attachment to both his nation and his ethnic group, what concerns us here is what he chooses to do, whether or not he supports intense fighting by his ethnic group. Third, since identities are defined by preferences, they can be inferred empirically by revealed preference: not from what people say but from observing what they choose to do (Klor and Shayo Reference Klor and Shayo2010).

Equilibrium

We employ a variant of the concept of Social Identity Equilibrium developed in Shayo (Reference Shayo2009). This concept takes into account the empirical observation that individuals are more likely to identify with a given group the more similar they perceive themselves to be to that group, and the higher the status of that group. Formally, an equilibrium is a profile of fighting efforts (f_i ) _{i ∈ N} and a profile of social identities (g_i ) _{i ∈ N} such that for all i ∈ N we have f_i ∈ [0, y_i ], g_i ∈ G_i and

\begin{eqnarray*} (i)\text{\}U_{i}(f_{i},f_{-i}g_{i}) \ge &U_{i}(f_{i}^{\prime},f_{-i}g_{i})\text{\ for all}f_{i}^{\prime}\in [0,y_{i}], \\ (ii)\text{\}U_{i}(f_{i},f_{-i};g_{i}) \ge &U_{i}(f_{i},f_{-i}g_{i}^{\prime})\text{\ for all}g_{i}^{\prime}\in G_{i}. \end{eqnarray*}

The first condition has to do with choice of actions (fighting efforts) under a given pattern of social identities. It is the standard Nash condition. The second condition is about the process determining the pattern of social identities. Formally, it requires that each individual's social identity be “optimal” given the social environment, i.e., given the status and perceived distances resulting from the pattern of fighting. We stress, however, that while we are using the language and tools of optimization, condition (ii) is not meant to describe some deliberative process in which individuals “choose” their social identities optimally. Rather, condition (ii) is a way for the modeler to capture the forces that have been found to shape social identification. Thus, our solution concept only requires that actions be optimal given current identities and that identities be optimal given current actions.Footnote ²² In particular, when an individual chooses his fighting effort, he takes his identity (preferences) as given. If he identifies with, say, his ethnic group then he behaves accordingly and does not consider the possibility that changes in the environment can lead him to change his identity. This is another way of saying that an individual is committed to his identity.

Social Identity Equilibria

In equilibrium, an individual i identifies with the nation only if

\begin{eqnarray*} U_{i}(f_{i},f_{-i};N)\ge U_{i}(f_{i},f_{-i};J)\quad\text{for}i\in J,\text{} J=A,B. \end{eqnarray*}

(8) \begin{equation} \gamma (S_{N}-S_{J})\ge \beta (d_{iN}^{2}-d_{iJ}^{2})\quad \text{for}i\in J, \text{}J=A,B. \end{equation}

Both sides of inequality (8) can be affected by the intensity of conflict F. This is illustrated in Figure 3 for a benchmark case, in which the two ethnic groups are identical in resources and in any other exogenous determinants of status (Y_A = Y_B ; σ _A = σ _B ). The solid upward sloping line shows the right-hand side of (8). To the extent that the intensity of fighting between the groups affects the salience of ethnic attributes (η₁ > 0), it increases the perceived distance from the nation relative to distance from own group. The dashed line represents what the left-hand side of (8) looks like if F_A = F_B = F/2. It is downward sloping since fighting destroys national resources, thereby reducing national status. The actual γ(S_N − S_J ) may diverge from this line if F_A ≠F_B . For example, if F_B > F_A , then group B is winning the contest, which in general implies S_B > S_A .Footnote ²⁸ The four dots in the figure depict γ(S_N − S_A ) and γ(S_N − S_B ) for the particular fighting configurations illustrated in Figure 2, resulting in total fighting intensities F^NN , F^NB , and F^AB (the F^AN case is analytically the same as the F^NB case). Notice that, generically, the two sides of (8) are not equal (in other words, one identity is strictly preferred to the other). Since the groups are homogeneous, this means that in equilibrium all members of a given ethnic group hold the same social identity. This will not necessarily be the case once we allow attributes to vary within ethnic groups.

FIGURE 3. Social Identity Equilibria (symmetric groups)

Notes: The figure depicts the difference in national and group status (dots) and the difference in distance to the nation and to own group (upward sloping line), for the possible configurations of fighting illustrated in Figure 2.

Recall that in a Social Identity Equilibrium, two conditions must hold. First, each individual's contribution to the fighting effort must be optimal given his social identity and others’ fighting efforts. Second, social identities must be optimal given the configuration of fighting. For national identity to be optimal for members of ethnic group J, it must be the case that γ(S_N − S_J ) ≥ β(d² _iN − d² _iJ ) for i ∈ J. In the situation depicted in Figure 3, this is true for both groups when F = F^NN , but only for group A when F = F^NB and for neither when F = F^AB . Conversely, for ethnic identity to be optimal for members of J, it must be the case that γ(S_N − S_J ) ≤ β(d ² _iN − d² _iJ ) for i ∈ J. In the figure, this holds for both groups when F = F^AB , but only for group B when F = F^NB and for neither when F = F^NN . The figure thus depicts a case of multiple equilibria. If F_A = F_B = F^AB /2 (hence F = F^AB ), then it is optimal for members of both groups to identify ethnically, and if everyone identifies ethnically then a profile of fighting efforts such that F_A = F_B = F^AB /2 represents individually optimal fighting efforts. But if F_A = F_B = F^NN /2 then it is optimal for all members of society to identify with the nation which in turn sustains F_A = F_B = F^NN /2. Finally, a situation in which A members identify with the nation, B members identify ethnically and F = F^NB is also an equilibrium.

The precise condition for the existence of multiple equilibria is given in equation (15) in the Appendix. Essentially, when status and distance parameters are at intermediate levels, both a high-intensity conflict and a low intensity conflict can be an equilibrium. If, for example, some historical conditions result in peaceful relations between the groups, this can lead to both a relatively high status of the nation and to the blurring of intranational cleavages. Thus, national identification and low conflict (F = F^NN ) is an equilibrium. In other words, relatively peaceful intergroup relations can persist, even when the original conditions which prevented high intensity conflict are gone. But this same country can also find itself in a situation where an intense conflict both degrades national status and accentuates ethnic differences, to the point where ethnic identification on both sides (and continued conflict) is an equilibrium. Conflict—and peace—can be self-reinforcing.

Proposition 1 (Multiplicity) . There exist conditions such that both an ethnic identification equilibrium with high-intensity conflict ( F = F^AB ), and a national identification equilibrium with low-intensity conflict ( F = F^NN ) exist.

As mentioned in the Introduction, one of the most persistent findings in the study of ethnic conflicts is that countries that are fairly similar in terms of their ethnic diversity, geography, and even economic endowments and political institutions, can find themselves in very different equilibria: some with years of persistent ethnic conflict, others with generally peaceful relations between a country's different ethnic groups. As we argue in the next section, Proposition 1 may help us understand this fact.

But while historical contingencies can matter and can persist, the model points to several other factors that facilitate or inhibit high levels of ethnic conflict (an important additional factor, namely the fighting capacity of radicals, will be examined when we analyze intragroup cleavages). Consider first exogenous variations in national status, due either to dimensions of comparison other than material payoffs (captured by σ _N ) or to the material payoffs of reference groups (captured by Π_−N ). In Figure 3, these are reflected in the intercept of the γ(S_N − S_J ) line. If, for example, national status is sufficiently high that both groups would identify with the nation even at F^NB and F^AB , then in equilibrium everyone must identify with the nation and the unique level of conflict is low at F^NN . Similarly, but perhaps more interestingly from a policy perspective, if national status is sufficiently low (and group status sufficiently high) such that individuals would identify with their ethnic group even when there is little fighting and F = F^NN , then such low level of fighting cannot be an equilibrium. Even if a halt to the violence could somehow be reached, absent other changes (e.g., significant nation building), the groups will end up at either the intermediate F^NB or the full-blown F^AB levels of conflict. This has implications for strategies of international peace building that we return to briefly in the Conclusion.

Proposition 2 (National status) . Ethnic identification coupled with high-intensity conflict is more likely, and national identification coupled with low-intensity conflict is less likely, the weaker are exogenous sources of national status (lower σ _N , higher Π_−N ).

A complementary observation pertains to perceived distance. Suppose that even in the absence of conflict, the attributes that distinguish members of the two ethnic groups are very salient relative to common attributes (namely, η₀ is high). Graphically, this would mean the intercept of the β(d² _iN − d² _iJ ) line is “high.”Assuming national status is not very high relative to group status, this can eliminate F^NN as an equilibrium, while making F^AB an equilibrium.

Proposition 3 (Salience of ethnic attributes) . Ethnic identification coupled with high-intensity conflict is more likely, and national identification coupled with low-intensity conflict is less likely, the higher is η₀ .

The model makes an important point with respect to the contestability of national resources. Specifically, consider the effect of reducing the share of national resources that can be captured to benefit one group over the other, that is, reducing V from, say, V^hi to V^low while keeping total national resources (Y + V) constant. For example, better political and legal institutions are generally believed to limit both the share of national resources that are potentially expropriable through fighting and the extent to which they can be used to favor one group over the other (see Besley and Persson Reference Besley and Persson2011). As in other models of conflict, if there is less to fight over, the level of fighting under any given profile of identification goes down. In Figure 3, F^NN , F^NB , and F^AB all shift to the left. Notice however, that if this reduction in violence is large enough, it can lead to a shift in the identification profile. Specifically, it may be the case that while ethnic identification is an equilibrium under V^hi , it is no longer an equilibrium under V^low . This means there will be an additional reduction of conflict intensity: not just from F^AB (V^hi ) to F^AB (V^low ), but to the (potentially much lower) level F^NN (V^low ).

Proposition 4 (Contestable resources) . For a given level of national resources (Y + V), ethnic identification is more likely, and national identification is less likely, the higher the share of national resources that are contestable.

Figure 4 provides a graphical summary of Propositions 1–4 for the case of symmetric groups (Y_A = Y_B ; σ _A = σ _B ). On the horizontal axis we have V, the value of the contestable resources. On the vertical axis we have the exogenous sources of the difference between national and group status, S_N − S_J . The latter include total national resources absent any fighting (Y + V), resources of the nation's reference group (Π_−N ), and any other determinants of national status (σ _N ) and ethnic group status (σ _J , J = A, B). Finally, the intercept of the two upward sloping lines (labeled NN and EE) is given by $\frac{\beta}{4\gamma}\eta _{0}$ (recall that η₀ is the relative salience of ethnic attributes absent any fighting). The region below and to the right of the EE line shows the combinations of parameter values for which an all-ethnic identification equilibrium exists. The region above and to the left of the NN line shows the parameter values for which an equilibrium with a common national identity exists. The intersection of the two regions is where both types of equilibria exist.

FIGURE 4. National and Ethnic Equilibria as a Function of Contestable Resources, Status and Distance Parameters

Notes: Regions in the parameter space in which national identity and ethnic identity equilibria exists, for the case of symmetric groups (σ _A = σ _B = σ _J ; Y_A = Y_B ).

For national identity to be an equilibrium, national status needs to be sufficiently high and the relative salience of ethnic attributes needs to be sufficiently low. The reverse is true for ethnic identity. If V = 0 then social identities are completely determined by the exogenous parameters affecting national and group status and perceived distances. As V increases (more resources to fight over), these parameters continue to be important, but status and distance are also affected by the level of violence and there is more scope for vicious and virtuous cycles (multiple equilibria). Notice, however, that V appears on both axes: while higher V leads to an increase in the intensity of conflict under any given profile of identities, it can also boost national status. Thus, an increase in V with Y held constant does not necessarily lead to an ethnic equilibrium.Footnote ²⁹ But if more of the national resources are contestable—if V increases while Y + V is fixed—this can move the nation to a region where having a common national identity is no longer an equilibrium.

So far we focused on symmetric all-ethnic or all-national equilibria, but the model also allows for the possibility of asymmetric equilibria, where members of one group identify with the nation and members of the other group identify ethnically. Consider the case where the two groups vary in their status (for reasons other than their fighting efforts and the resulting division of the contested resources). For concreteness, suppose that relative to the benchmark case of Figure 3, σ _A decreases and σ _B increases. The relatively low status of group A means S_N − S_A shifts up for any given configuration of fighting. If this shift is large enough, then ethnic identification of A members is no longer possible in equilibrium, even if F = F^AB . At the same time, the increase in B’s status shifts S_N − S_B down, possibly eliminating the F^NN equilibrium (ethnic identification is more attractive for B members even if there is little conflict). This can result in a unique equilibrium level of conflict, where the high status group (B) identifies ethnically and devotes substantial resources to contesting national resources, while the low status group (A) identifies with the nation and holds back its fighting efforts.

Proposition 5 (Asymmetric conflict) . Let J and J′ be the two ethnic groups. If σ _J is sufficiently low and σ_J′ is sufficiently high, then in equilibrium J members identify with the nation, J′ members identify with their ethnic group, and F_J′ > F_J .

A final point to note is that differences in group resources (Y_A and Y_B ) essentially combine the effects in Propositions 3 and 5. Starting from a situation with equal endowments, an increase in |Y_A − Y_B | would increase both perceived distances between the groups, and status differences.Footnote ³⁰ It would thus tend to reduce the likelihood of a peaceful, common-identity equilibrium, but also increase the likelihood of an asymmetric conflict.

Endogenous Attributes

Our model takes ethnic attributes as exogenously fixed. This conforms to the way they are defined in the literature: descent based, fixed in the short run. However, even when group membership is fixed by birth, some attributes that characterize ethnic groups can be changed. A common way by which one expresses one's identity is, for example, to wear the clothes that characterize the group one identifies with. In Appendix A we therefore analyze the case where the ethnic attribute q^e _i is completely flexible and freely chosen by individual i.Footnote ³¹ Specifically, we consider the case where each individual chooses not only a fighting effort f_i , but also an attribute q^e _i ∈ [0, 1]. This does not change the results in Propositions 1–5, except that Proposition 3 needs to be modified (as explained shortly). The main implications for the preceding analysis can be summarized in three points.

1. 1. If an all-ethnic equilibrium exists under fixed attributes, then an all-ethnic equilibrium also exists under endogenous attributes.

2. 2. If an all-national equilibrium exists under fixed attributes, then an all-national equilibrium also exists under endogenous attributes. Furthermore, there exist conditions such that an all-national equilibrium exists under endogenous attributes but not under fixed attributes.

3. 3. An asymmetric equilibrium may not exist under endogenous attributes even if such an equilibrium exists under fixed attributes.

Roughly speaking, if individuals can change their attributes, this makes a national identity (low conflict) equilibrium easier to sustain.Footnote ³² The reason is that national identification leads to the adoption of common attributes throughout the nation, thereby reducing perceived distances across ethnic groups. (The NN line in Figure 4 shifts down.) Furthermore, if everyone identifies with the nation, then even if there is some conflict over resources across subnational groups, this conflict has less of an effect on perceived distances, as differences are smaller to begin with. (The slope of the NN line flattens.) The same is true with respect to other factors that affect the salience of ethnic attributes (captured by η₀): while such factors continue to be important under ethnic identity, they are no longer important under national identity. This is the modification of Proposition 3: while an ethnic identity equilibrium is more likely to exist the higher is η₀, a national identity equilibrium is no less likely to exist.

At the same time, if an ethnic identity (high conflict) equilibrium exists under fixed attributes, such an equilibrium can also be sustained when ethnic attributes are flexible and costless to change (the EE line in Figure 4 does not move). The reason is that if everyone identifies ethnically, the two groups can settle on the two extremes of the attribute interval (with all members of a given ethnic group choosing the same attribute), thereby keeping the differences across groups just as high as in the fixed attribute case.Footnote ³³ However, some asymmetric equilibria, where, for example, A members identify with the nation and B members identify ethnically, may not carry over to the endogenous attributes case. National identification would lead A members to adopt the attributes of the B members, thereby shrinking perceived distance from the nation also among these B members, possibly ruling out their holding an ethnic rather than a national identity.

Yugoslavia

In 1985, the Serbian Academy of Sciences published a memorandum condemning Tito and the Yugoslav Communist Party state for three decades of “anti-Serb policies” that left Serbia poorer than other Republics, and denouncing “genocidal” anti-Serb policies by Albanians in Kosovo. The next year, a prominent Serb politician, Slobodan Milosevic, became head of the Serbian Communist Party with an agenda to restore the Serbs’ lost status in the federation. To that end in 1989, on the 600th anniversary of the Battle of Kosovo, Milosevic visited the Kosovo province to deliver a speech rife with symbolism. Kosovo is an Albanian-majority region, claimed as the spiritual home of the Serbs. Milosevic spoke about restoring the dignity of Serbia and fostering intra-Serbian unity in the face of new challenges that he paralleled to the challenges (slavery) created by the Ottoman Muslim victory in the battle of 1389.

That speech marked the beginning of a process of ethnic mobilization and conflict escalation. Yugoslavia was an uneasy amalgam of mostly territorially concentrated ethnic groups, which Communism had pulled together into a single nation. The collapse of Communism was followed by manifest ethnic identification among large segments of the population and fueled aspirations to self-determination in Yugoslavia's constituent Republics as well as smaller regions. The elections of 1990 “proved an ethnic census” (Mann Reference Mann2005, 367). Nationalist parties (expressions of ethnic identification, in terms of our model) gained ground. Leaders like Tudjman in Croatia, Izetbegovic in Bosnia, and Seselj and Draskovic in Serbia had “moral authority” due to their imprisonment and torture for inciting ethnic conflict under Communism (Mann Reference Mann2005). Serbian reformers like Stambolic were forced off the stage by the public, which rallied around nationalists like Milosevic (Petersen Reference Petersen2002). For the most part, Serbian voters agreed with the memorandum of 1985, which had been signed by 216 Belgrade intellectuals and reflected the ideas of Serbia's ethnic core, which included the Serbian Orthodox Church, academics, intellectuals, and ordinary people without a political agenda (Milisevic Reference Milisevic2004, 119–130; Petersen Reference Petersen2002, 222–3). Milosevic stoked the flames of ethnonationalism by dissolving the Kosovo assembly, revoking the province's autonomy. In response, ethnic Albanian legislators declared Kosovo a Republic. The League of Communists split along ethnic lines. Newly elected nationalist leaders claimed the right to a nation state, even in Bosnia, where there was no ethnic majority.

In Croatia, the situation was similar. Tudjman, with greater nationalist credentials than Milosevic, gained two thirds of parliamentary seats while calling the Serbs “Turks” (Cohen Reference Cohen1995, 211) and drawing support from Croats who felt ethnically victimized and resented outgroup privileges (Mann Reference Mann2005, 375; Siber Reference Siber, Seroka and Pavlovic1993, 152–3). One after the other of Yugoslavia's ethnic republics seceded, starting with Slovenia (the richest and most ethnically homogenous one). Croatia followed suit in 1991 and the Serbs in Croatia's Krajina region (having lived there since the 17th century), agitated for a separatist bid of their own, demanding to be unified with Serbia. Krajina Serbs were part of the Serbian ethnic core and were the first to mobilize for violence. Local Serb militias were formed with help from Belgrade ultranationalists like Vojislav Seselj, a former political science professor, “Chetnick Vojvoda,” and President of the Serbian Radical Party.Footnote ⁵⁵ Paramilitaries and volunteers from Serbia joined, some of them organized by the Serb security police (SDB).Footnote ⁵⁶ Krajina Serbs set up roadblocks to prevent Croatian state officials from interfering in a planned referendum on independence. Milosevic supported the Krajina Serbs’ right to independence and argued for the unification of all Serb territories. Seselj pursued his long-standing goal of forming a homogenous “Greater Serbia” by stoking Serb fears of victimization by fascist Croat hordes and ultimately recruiting and deploying thousands of “Seseljevci” to fight for the separation of Serbs from other Yugoslavs.Footnote ⁵⁷ Tudjman spurred the Seseljevci with his words and deeds, helping to polarize the Serbs (Petersen Reference Petersen2002). In Krajina, Serb militants were able to play into Serbs’ fears of Croatian aggression by pointing to Croatian provocative actions, such as the adoption, by Croatia, in July 1990, of the sahovnica, the traditional red and white checkerboard symbol that was identified with the Ustasha state during World War II; or the downgrading of Serbs in Croatia from a “constituent nation” to a “national minority.”Footnote ⁵⁸ Croatian militias and local Serbs engaged in guerrilla battles, leading to engagements by Yugoslav army units.

The Croatian conflict spilled over to neighboring Bosnia, where a Muslim majority voted for independence, and the Serbs dissented. As soon as the votes were counted, Serbs set up roadblocks around major cities (which were heavily populated by Muslims), cutting them off from the mostly Serb-controlled countryside. Serbs began to leave the cities with the help of Serb-controlled Yugoslav National Army units. A Bosnian Serb parliament was set up and Serb paramilitary units formed, targeting Muslims. The violence in both Croatia and Bosnia ended in 1995 with heavy international diplomatic involvement and peace enforcement. What started as militia violence by a core group of extremists ended up as a multiyear war that killed more than 100,000 people and displaced around 2.5 million.Footnote ⁵⁹

Yugoslavia in Light of the Model . This pattern of conflict escalation in Yugoslavia is rarely explained without reference to the “actions of recently empowered and unpoliced thugs” who were able to “whip up a fair amount of hatred” (Mueller Reference Mueller2000, 47) and many see it as the result of elite manipulation of the masses (De Figueiredo and Weingast Reference De Figueiredo, Weingast, Walter and Snyder1999). But this was not an elite conspiracy. Ethnicity had been made salient by the collapse of Communism, the economic crisis, and the political power grab that followed, as well as by recalling memories of violence at the hands of the Croat Ustasha, the Serb Chetnicks, and Sandzak SS regiments during World War II (Glenny Reference Glenny1992). The Yugoslav case had several of the risk factors for conflict escalation that we model: declining international status; high value of contestable resources (the state itself was up for grabs and could be exploited for the benefit of the group, excluding non-co-ethnics); and high salience of ethnic attributes. Moreover, the trajectory of the conflict was shaped by radicals with sufficient resources to fight hard; and these radicals were not simply opportunists. Many—if not most—identified ethnically and shared a commitment to enhance the relative status of their group.

In the academic literature as well as in press reports on Yugoslavia, Milosevic is widely seen as “the paradigmatic case [of] a nationalist of convenience, rather than conviction” (Brubaker Reference Brubaker and Hall1998, 289).Footnote ⁶⁰ This has supported the view of Yugoslavia's violent unravelling as the result of elite manipulation. But Milosevic rode a wave of nationalist mobilization and he spoke directly to the “core constituencies of Greater Serb nationalism—Serb refugees, Serbs in threatened areas, especially Kosovo, and some rural Serbs who expected more economic development” (Mann Reference Mann2005, 370). His opposition was actually even more nationalist than him, with Vuk Draskovik calling for “unity of all Serb lands” with grassroots support from Orthodox cultural-educational societies (Mann Reference Mann2005, 372). Even as we acknowledge the fact that elites can be strategic and manipulative, we must also acknowledge that “the elite manipulation thesis fails to specify the particular conditions that make key segments of the Yugoslav population especially responsive to elite manipulation. . . [it] overpredicts the severity and violence of ethnic conflict” (Brubaker Reference Brubaker and Hall1998, 291).

Accounts differ as to the degree of ideological commitment and ethnic identification of Serb paramilitaries.Footnote ⁶¹ Some, like Seselj, had a long history of ethnic identification and fought to reclaim the Serb's lost status. Others were more opportunistic and might have been motivated by the desire for personal profit. But while the violence did involve elements of criminality,Footnote ⁶² this was not a war caused by “a few bad guys.”Footnote ⁶³ The violence was set in motion by politicians who identified with, drew support from, or acted on behalf of an ethnic core. The identity of the early movers is consistent with the predictions of our model. They were dominated by ultranationalists, who made use of people with proclivities or specialization in violence to pursue their aims. Zeljko Raznatovic (Arkan), who is usually described as an opportunistic thug, actually organized soccer fans for Red Star Belgrade into an ethnonationalist group that provided a pro-Milosevic militia and had worked for the Serbian Secret Service (SDB) in the past (Mann Reference Mann2005, 404). Milisevic (Reference Milisevic2004) argues that the perception that the paramilitaries were out of control criminals was deliberately cultivated by Serbian leaders so as to more plausibly deny their involvement in the war. The White Eagles, the paramilitary group attached to the Serbian Radical Party, was led by Mirko Jović, who called for “a Christian, Orthodox Serbia with no Muslims and no unbelievers” (Velikonja Reference Velikonja1992, 268). Vuk Draskovic, leader of Serbian Renewal Movement, organized the Serbian Guard; Seselj, leader of ultranationalist Serbian Radical Party, formed the Chetniks. In Croatia, the ultranationalist Croatia Party of the Right, led by Dobrosav Paraga, organized the Croatian Defense League (HOS), followed by the Zebras and Black Legions, two other militias formed by right-wing groups (Sikavica Reference Sikavica, Udovicki and Ridgeway1997, 141). These groups identified ethnically and their leadership drew strength from the popular appeal of their message. Thousands of volunteers joined the fight: “these men from Serbia [who] decided to cross the rivers Sava and Drina to join the war. . . were the core constituencies of support for pro-nationalist and pro-war politics” (Milisevic Reference Milisevic2004, 17).

As the model would predict, violence perpetrated by the ethnic core intensified ethnic identification among moderates. Although many—perhaps most—people in Yugoslavia identified ethnically prior to the war (most selected only one ethnic category in the census and fewer than 10% identified as “Yugoslavs”), it was not until the outbreak of violence that the masses became radicalized. There is evidence of substantial interethnic cooperation and exogamy in Yugoslavia (Petrovic Reference Petrovic1986, Bromlei and Kashuba Reference Bromlei and Kashuba1982), but in the 1991 census—the year of Croatia's declared independence—the share of “Yugoslavs” in the population dropped to its lowest level in decades. In Croatia it declined from 8.2 to 2.2% (Botev Reference Botev1994; Burg and Berbaum Reference Burg and Berbaum1989; Woodward Reference Woodward1995). The lack of strong ethnic identification capable of supporting violence prior to the start of the violence was evident in the mass-level defections and draft dodging that occurred in the Serb Army (Milisevic Reference Milisevic2004, 16). In Croatia in 1990, Tudjman's party (HDZ) received 1.2 million votes, whereas the reformed communists and allies who preferred a stronger Yugoslav federation received almost as many votes (1 million) and most Croatian Serbs did not vote for Serb nationalists, but rather supported the Communists or centrist parties. A year later, almost all Croats voted for independence and almost all Serbs boycotted the vote and favored independence for Serb Krajina (Obserschall Reference Oberschall, Chirot and Seligman2001).

The conflict over Croatian independence was quick to turn violent (the first incident in Plitvice occurred at the end of March 1991) and violence escalated quickly. The indiscriminate retaliation by Croats helped mobilized moderate Croatian Serbs. While at first the violence was the work of a select few, later on there was widespread support in retaliatory mass killings (Sudetic Reference Sudetic1998, 350–2). “Most of those indicted by the ICTY committed atrocities in their own home areas [and] locals were the most radicalized” (Mann Reference Mann2005, 418–420). Killers and torturers knew their victims.Footnote ⁶⁴ Former army officers and police were over-represented among violence perpetrators and they identified more ethnically than the rest of the population.Footnote ⁶⁵ The Serbian Volunteers Guard and other militias cultivated the ethnicization of violence by adopting religious rituals as part of the enlistment procedure. Soldiers had to be baptized and prayed together (Milisevic Reference Milisevic2004, 73–4) and the Guard offered an attractive combination of religion and “national mythology” that drew many volunteers who responded to reports of war atrocities against Serbs and joined to defend Serbia and Yugoslavia (Milisevic Reference Milisevic2004, 74; 85; 97).

Eventually, “elites, militants, and ordinary people were all involved” (Mann Reference Mann2005, 360). A Croatian Catholic priest who witnessed atrocities in his village testified before the International Criminal Tribunal that “There had been no tensions before. . . Orthodox children would attend the Sunday School. . . [but] when the war started. . . the local Serbs became the main killers” (Bottica 1992, 253). This pattern, which identifies a sharp increase in ethnic identification sparked by violence, was apparent to other observers at the time. A UN report noted that “in places where the local Serb population was initially fairly friendly, once Arkan's thugs arrived the situation changed” and Muslims were ostracized and treated with hostility.Footnote ⁶⁶ This change in behavior is consistent with a shift in preferences, although other explanations, such as intimidation, are also possible. The pattern of violence-induced polarization describes not only the escalation of the Croatian war of 1991, but also subsequent Croat-Serb violence in Croatia and Bosnia, as well as Croat-Muslim violence. An influential anthropological study of Visnjica, a mixed Muslim/Croat village in Bosnia, clearly illustrates how the war “start[ed] out as a war waged by outsiders [and] developed into one where neighbor was pitted against neighbor after the familiar person next door had been made into a depersonalized alien, a member of the enemy ranks” (Bringa Reference Bringa1995, xvi). In Croatia, interethnic relations were peaceful until the spring of 1991, when the actions of Serb paramilitary groups caused a rapid polarization and ethnic mobilization for war.Footnote ⁶⁷ Extremism was cultivated by the state, including the Serb radio, which served as the primary media outlet for most of the rural population in Serbia and Croatia.Footnote ⁶⁸

Ethnic polarization in Serbia was widespread and persisted through the 1990s, as evidenced in the large-scale participation of Serbs in the cleansing of the Albanian population from Kosovo in 1999 and in the apparently little remorse they displayed after the fact (Petersen Reference Petersen2002, 242–3). Even after Dayton, ethnic identification persisted: police and paramilitary units were ethnically homogeneous; expulsions of Muslims continued two years after the accords; Muslims have been reluctant to return to their villages; and Croatian initiatives for repatriation have preferred monoethnic cantons (Petersen Reference Petersen2002, 234–5).

The Yugoslavia case demonstrates the key mechanism of the model in that a subset of the ethnic group, who have the will and resources to fight hard on behalf of their ethnic group, can destabilize a relatively peaceful equilibrium. Conflict escalation was made possible by the fact that extremists had sufficient resources to fight.Footnote ⁶⁹ By most accounts, most citizens of Yugoslavia did not support the dissolution of the federation before 1990. Ethnic identities did not dominate other social and political cleavages (Petersen Reference Petersen2002, 208–210). Class, political party, and urban/rural divides constituted important social identities in prewar Yugoslavia that mattered just as much—if not more—than ethnic identities. This all changed with the outbreak of ethnic violence. Not only did violence reduce the salience of a common Yugoslav identity but, consistent with the model, it was also effective in reducing the salience of intragroup cleavages. The socially and politically important differences between rural and urban Serbs ceased to matter (Bennett Reference Bennett1995) as did a number of previously salient political cleavages within each group, and the violence had the effect of “switching off” a host of alternative social identities (Yugoslav, Reformist, Pluralist, Liberal) that had been resurgent with the collapse of Communism (Gagnon Reference Gagnon2004). As stipulated in the model, ethnic violence accentuated intergroup divisions while narrowing intragroup distances, leading to increased ethnic identification and escalating conflict.Footnote ⁷⁰

Finally, the case of Yugoslavia also suggests that the strategic use of violence (to the extent that core group violence was strategic) is more likely to be effective in societies with pre-existing ethnic divisions. While we do not model elites, we can help elucidate the mechanisms which allow them to be effective in mobilizing moderates (noncore group members). But, as in our description of Yugoslavia, the vanguard does not necessarily act purely instrumentally and violence cannot effectively mobilize co-ethnics if they do not have any emotional attachment to their ethnic group. Rather, core members are often motivated by ethnic differences or ideological fervor and have an interest in minimizing intragroup divisions.