Self-determination movements are the most frequent

Transcription

1 Building Reputation: Why Governments Fight Some Separatists but Not Others Barbara F. Walter University of California, San Diego This article attempts to show that future players and future stakes two factors generally ignored by political scientists strongly influence government decisions to cooperate or fight at least against ethnic minorities seeking self-determination. Data on all separatist movements between 1956 and 2002 reveals that governments are significantly less likely to accommodate one challenge if the number of ethnic groups in a country and the combined value of the land that may come under dispute in the future is high. Governments that refused to accommodate one challenger were also significantly less likely to face a second or third challenge down the road. This provides some of the first systematic evidence that governments invest in reputation building a least in the domain of domestic ethnic relations. Self-determination movements are the most frequent source of violent conflict in the international system today (Marshall and Gurr 2003). Between 1956 and 2002, 146 ethnic groups in 78 countries demanded greater territorial autonomy or independence from their central government. In the vast majority of these cases the government responded by refusing to compromise on any issue related to territory, even if they faced armed rebellion as a result. Governments in Russia, Iran, and Myanmar for example, have fought lengthy and costly wars rather than offer a compromise settlement to the Chechens, Kurds, or Karens. And in Indonesia, the chief of the armed forces, General Sutarto, has promised to fight rebels in Aceh until his troops last drop of blood (New York Times 5/5/03, 6). The fact that governments are so unyielding toward separatists, however, does not mean that peaceful accommodation never occurs. Leaders in Canada, India, and the former Czechoslovakia have all offered varying degrees of sovereignty to ethnic minorities within their borders. This variation is puzzling and raises the following question. Why are governments sometimes willing to accommodate ethnic minorities seeking greater self-rule, and other times not? The two explanations frequently offered to explain why governments make concessions in some cases but not others focus on the value of the stakes or the relative capabilities of the disputants (Bartkus 1999; Diehl 1999; Goertz and Diehl 1992; Huth 1996; Toft 2003). According to this view, governments are more likely to fight for territory that is economically, strategically, or psychologically valuable, and give up territory that is not. They are also more likely to fight if the risks and costs of confronting a particular opponent are low. The nature of the immediate stakes and the relative capabilities of the disputants, therefore, determine how a government is likely to respond. Building on work by Walter (2003), I argue that governments are much more forward-looking than current theories allow. Governments weigh their immediate interests and capabilities when determining whether to grant concessions, but they also carefully calculate the effect this behavior may have on future challenges and future losses. If a government believes it could face multiple additional challenges over numerous pieces of territory, it has greater incentives to invest in building a reputation for toughness than if it knew it would face only one challenge, or relatively few challenges. The risks and costs of future confrontations, therefore, should factor into a government s decision to compromise or fight at least in cases where it expects a series of similar challengers making similar demands over time. What follows is divided into four sections. The first section walks the reader through the logic of this Barbara F. Walter is associate professor of international relations and Pacific studies, Mail Code 0519, University of California, San Diego, La Jolla, CA (bfwalter@ucsd.edu). I am grateful to Rui de Figueiredo Jr., Hein Goemans, Zoltan Hajnal, Andrew Kydd, Bob Powell, Kenneth Schultz, and participants at seminars at Yale, Columbia, University of Southern California, Dartmouth, and UCSD for very helpful comments and suggestions. This project was funded through the generous support of the National Science Foundation Award # American Journal of Political Science, Vol. 50, No. 2, April 2006, Pp C 2006, Midwest Political Science Association ISSN

2 314 BARBARA F. WALTER reputation building argument, and presents two hypotheses for testing. The theory I am presenting is not new. In fact, it builds directly on the well-known chain store paradox first described in the economics literature by Selten (1978) and then modified by Kreps and Wilson (1982), and Milgrom and Roberts (1982). What is new is that it applies this theory to a particularly pressing problem in political science and finds that it explains a lot about government behavior at least towards minority ethnic groups seeking self-determination. The second section presents two alternative explanations for government behavior that governments care more about their present interests and capabilities than they do about future challenges. The third section tests the competing theories against a dataset of all self-determination movements initiated between 1956 and 2002 and discusses the findings. The results show that governments are significantly more likely to fight against a particular separatist group if the number of future challengers and the potential long-term losses from future challenges are high. By contrast, almost none of the factors associated with the present value of the land, or the present capabilities of the disputants had any affect on government behavior. The final section discusses the implications this research may have for our understanding of conflict and cooperation more generally, as well as the ongoing debate on reputation building. The Logic of Reputation Building The Importance of Future Challengers and Future Stakes This article demonstrates that it may be rational to go to war if doing so allows governments to build a reputation that will deter additional challengers in the future. 1 The mechanism I m looking at is private information and the incentives governments have to misrepresent this information when many future rivals exist, and the argument goes as follows. Assume that a country contains only one minority group and that this group has to decide whether to challenge the government for greater autonomy. If the group challenges, the government then has to decide whether to accommodate these demands or fight. If the government accommodates, it gets peace in return for 1 This argument is derived from work on entry-deterrence and advertising in the economics literature. See especially Spence (1973), Schmalensee (1981, 1983), Rosenthal (1981), Kreps and Wilson (1982), Milgrom and Roberts (1982), Dixit (1982), and Kennan and Wilson (1993). There is also a long literature on reputation and deterrence in political science. See especially Schelling (1966), Alt, Calvert, and Humes (1988), Huth (1988), Nalebuff (1991), Hopf (1994), Mercer (1996), Huth (1997), and Toft (2001, 2003). some territorial concessions. If the government chooses to fight, it has a chance to maintain full control over its territory but it pays the costs of fighting. 2 When governments face only one challenger the logic of how to respond is fairly straightforward. Since there is little value in developing a reputation for toughness and paying the costs of war, governments have strong incentives to accommodate this challenger. The situation is more complicated, however, when a government believes it could face a series of potential challengers over time. Under these conditions, a government has to consider that the game could be repeated as many times as there are separatist groups and that the government s behavior in the first period could affect decisions by other separatists later on. Subsequent groups, for their part, do not know what type of government they face. They could face a government that is conciliatory and thus willing to part with some territory in return for domestic peace, as the Swiss were willing to do with the canton of Jura in Or they could face a government that is more extreme unwilling to compromise at any cost as the Indonesian government was with East Timor under President Suharto. This uncertainty fundamentally affects the game. 3 Potential challengers must decide whether to challenge based largely on a government s past behavior. 4 If the government they face has been resolute and unaccommodating in the past, potential challengers can expect the government to fight if they choose to challenge. If, on the other hand, the government has granted concessions in return for peace, they know they face a conciliatory government and the incentives to challenge go up. This means that conciliatory governments have strong incentives to strategically misrepresent their willingness to fight in order to deter these additional challenges and avoid the higher cumulative costs of future disputes. This reasoning gives us our first hypothesis for testing. A government s decision to accommodate demands for self-determination will be negatively related to the number of challengers it expects to face in the future. 2 Fighting denotes anything from repression to outright war. 3 If the parties were operating under complete information, the separatists would always challenge an uncommitted government and an uncommitted government would always accommodate. Separatists would also always refuse to challenge a committed government. The result in all cases would be peace. 4 This account considers only one of the many factors that potential separatist groups consider before deciding whether to challenge. For a more complete account of the logic driving ethnic groups to rebel, see Gurr (1970), Tilly (1987), Smith (2001), Fearon and Laitin (2003), and Walter (2006).

3 BUILDING REPUTATION 315 The number of potential challengers, however, should not be the only factor affecting government decisions to accommodate or fight. The logic of the reputation theory implies that a government will not only care about the number of additional challengers, but also about the cumulative value of land that may come under dispute in the future, and the relative strength of these additional groups. 5 Deterrence makes no sense if a government does not wish to retain the citizens who may seek self-determination or the land they may seek to acquire. The decision to accommodate one challenger, therefore, should also be strongly influenced by the combined value of the stakes a government believes it may lose in the future. This modification yields a second hypothesis for testing. A government s decision to grant territorial autonomy or independence will be inversely related to the cumulative value of all the land within its borders that could come under dispute. Existing Explanations The Importance of Present Stakes and Present Capabilities Most work on territorial disputes points to the economic, strategic, or psychological value of land to explain why governments choose to part with some pieces of land but not others. 6 Disputed territory, for example, could contain oil, minerals, or agriculturally rich regions that are important revenue sources for the central government. It could include strategic features that are crucial for maintaining the security of the state, such as mountain ranges, seacoasts, or other geographic features (Holsti 1991; Luard 1986; Mackinder 1919; Richardson 1960; Touval 1972; Vanzo 1999). A piece of territory could also have 5 The logic of the reputation model also implies that governments should care about the relative strength of future challengers when determining how to respond to a given challenge. A government that expects to face a series of weak opponents should be less concerned about building reputation than a government that expects to face a series of more powerful and costly opponents. This prediction was impossible to test given data limitations and will have to be left to future research to answer. Recent work by Triesman (2004), however, has argued that governments with limited resources (weak governments) may have strong incentives to appease early challengers in order to conserve sufficient resources to deter others. This supports the notion that the relative strength of all challengers a government believes it may encounter factors into reputation building calculations. 6 Gilpin (1981), Holsti (1991), Goertz and Diehl (1992), Coakely (1993), and Diehl (1996) have all argued that interests are a predominant factor in territorial disputes between states. While each of these works has focused on interstate conflicts, the same arguments are often heard to explain territorial disputes within states. strong symbolic or psychological value if it contains sites, landmarks, and buildings that form the basis of a country s identity or represent a country s historical homeland (Coakley 1993; Newman 1999; and Toft 2003). Theories that focus on the interests at stake, therefore, offer three additional hypotheses for testing. The higher the economic value of the disputed territory, the less likely a government is to accommodate a challenge. The higher the strategic value of the disputed land, the less likely a government is to accommodate. And the greater the psychological value of the land, the less likely accommodation. The value of the land currently under dispute, however, may not be the only factor affecting government responses to territorial challenges. Recent work by Press (2005) on interstate crises has argued that the prevailing balance of power should be more important than reputation in how governments respond to challenges. If this argument were applied to self-determination disputes, it would mean that governments with greater capabilities relative to a separatist group should be less likely to make concessions regardless of what they may expect to happen in the future. One final hypothesis, therefore, can be drawn from theories that focus on relative capabilities. 7 The stronger the government and the weaker the challenger, the less likely a government is to accommodate. Research Design My first goal is to establish whether governments are at all concerned about future challengers and the value of future lands when deciding to make concessions, or if they are really more interested in current interests and capabilities. To determine this, I collected data on every self-determination movement included in the Center of International Development and Conflict Management s (CIDCM) data on self-determination movements, phase II. 8 Nonviolent challenges, as well as violent challenges, were included in the dataset in order to avoid selecting 7 Realists have long argued that the military balance of power between combatants is likely to determine whether combatants choose a violent or peaceful means to resolve their differences. A sample of work in the realist tradition includes Morgenthau (1978), Waltz (1979), and Grieco (1988). 8 CIDCM defines a self-determination movement as any attempt launched by a territorially concentrated ethnic group for autonomy or independence from the central government using political or military means and includes all movements initiated between 1956 and Readers should note that CIDCM codes only those ethnic groups listed in the Minorities at Risk dataset and thus may exclude some ethnic groups in some countries. For greater elaboration of the coding rules see Marshall and Gurr (2003).

4 316 BARBARA F. WALTER only those cases where the government and the ethnic group failed to resolve their dispute without first engaging in armed conflict. This avoids biasing the results in favor of only the most difficult cases to resolve. 9 Each self-determination movement represents one case in the analysis. The study is limited to cases where self-determination movements have been launched and does not extend to ethnic minorities that have not chosen to activate similar demands. This decision was made for theoretical and empirical reasons. Theoretically, there is reason to believe that the decision to formally seek self-determination is quite different from demands made in the normal give and take of domestic politics. Governments, therefore, are likely to respond differently to a formal challenge than to the frequent lobbying on the part of internal interest groups for larger slices of the political, social, or economic pie. This appears to be confirmed in empirical studies of ethnic mobilization. Research by Gurr and Moore (1997), Cetinyan (2002), and Fearon and Laitin (2003) have found that factors such as past repression, demographic distress, a country s per capita GDP, a prior history of rebellion, and state weakness appear to affect the decision by ethnic groups to formally rebel against the government. Each of these factors is in large part unrelated to and very different from the factors that lead groups to compete in the domestic political arena. 10 Still, readers should be aware of a potential drawback of limiting the study in this way. It is possible that excluding cases where groups never articulated demands for self-determination systematically biases the types of cases included in the study. Exceptionally weak groups, for example, might never make any demands on a government, and exceptionally strong governments might never hear of them. Conversely, especially weak governments might give in before any formal demands are made, and extremely strong groups might obtain concessions without making a formal request. If this is true, then at least some of the causal factors associated with a group s decision to seek self-determination are also associated with a government s response, which may bias the estimates in the current analysis. 9 In addition, to ensure that no single country (or a small number of countries) with many potential challengers was driving the results, I sequentially deleted countries from the analysis. In alternate analysis, standard errors for potential nonindependence within countries were corrected using the cluster option in Stata. Neither test revealed any substantive effects on the results. 10 Importantly, none of the empirical or theoretical research looking at group-level mobilization cites the number of other potential groups in a country or the net value of their land as relevant to decisions about group mobilization and group tactics. In order to further assess this potential bias, I compared challengers in the current dataset to all nonchallenging ethnic groups listed in the MAR dataset (221 ethnic groups). I found that challengers were more concentrated geographically, had resided in the region longer, had more autonomy to begin with, and lived in areas with slightly higher economic value than nonchallengers. There were, however, no differences between challengers and nonchallengers on any of the measures related to the reputation model. Thus, while estimates of the effects of land value and relative capabilities could be biased by selecting only challengers, this selection should have less effect on assessments of the reputation model. The Dependent Variables Two dependent variables are included in this study. First, to determine which factors are likely to affect government accommodation, I created an ordered categorical dependent variable called Accommodation, which included four outcomes: no accommodation; some accommodation but not over territory (hereafter referred to as reform ); accommodation offering some form of territorial autonomy; and accommodation granting full independence. 11 Each self-determination movement was coded with the highest level of accommodation ever offered by the government. Of the 146 self-determination challenges included in the dataset, 86 (59%) were not granted any accommodation. Of the 60 groups that were granted some accommodation, 30% obtained reform, 63% were given some form of territorial autonomy, and less than 1% were granted full independence. Summary statistics are included in the appendix. Second, to see if government behavior had the desired effect of deterring additional challengers, I created a second dependent variable called Subsequent Challenge which was coded based on the number of subsequent selfdetermination challenges a government faced after the self-determination movement under observation. Details about the coding of each of these variables is included in the appendix. The Independent Variables The first set of explanatory variables was drawn from existing accounts that cite the economic, strategic, and 11 Although these four categories are clearly distinct on a theoretical level, I reran all of the subsequent analysis with one or more categories of accommodation collapsed to ensure that each category was in fact empirically distinct. The results indicate that none of the alternate codings led to a significantly better fit than the four category variables used here.

5 BUILDING REPUTATION 317 psychological value of the land under dispute as the critical factors influencing government behavior. The main variable, Economic Value, was measured using an additive 31-point scale indicating how many marketable resources were known to exist on a given piece of land. This 31-point scale was derived from the combined number of resources listed on both U.S. Geological Survey and CIA maps. 12 Strategic Value was measured using a 6-point additive scale indicating whether a given piece of territory included: (1) a sea outlet, (2) a shipping lane, (3) a military base, (4) an international border, (5) an attack route, and/or (6) a mountain range. 13 Psychological or symbolic value was measured using two crude proxies. The first, Length of Residence, was the length of time the challenging group had resided on a piece of territory. A government s attachment to a piece of land may decrease the longer a minority group resides on a piece of territory. The second, Autonomy, was a dummy variable indicating whether the group in question had been historically autonomous from the central government at any point prior to the conflict. Governments were expected to be less psychologically attached to territory that had once been under another group s control. Two additional measures were included to help assess current interests. The challenging group s population as a proportion of the national population, Proportion of Population, was included in the main model to take into account the value of group size to the government. In alternate specifications I also included the amount of land occupied by the group as a proportion of a country s total land mass (Proportion of Territory). The second set of hypotheses focused on the relative capabilities of both the government and the current challenger. Since measures for the military strength of individual groups in society are generally not available, I assessed Relative Capabilities by examining several different indicators of strength on each side. Two main indicators were used to measure the strength of the targeted government. The first was the average number of military personnel as a proportion of the total population during the du- 12 While a dollar value might have been preferable, regional estimates of the monetary value of different resources are rarely available and thus could not be included. In an attempt to include a measure that might more closely approximate the monetary value of the resources, in alternate tests I included a dummy variable indicating whether a given region included either oil or natural gas. To assess the relative value of the land compared to rest of the country, in alternate specifications I also included a measure of the proportion of total resources a piece of land contained. The substantive findings did not change with this latter measure. 13 In alternate tests I substituted a dummy variable indicating whether a given territory had access to the sea arguably the most important strategic resource. The change had no effect on the results. ration of a challenge. A government with more soldiers under arms throughout a dispute was assumed to be less willing to accommodate separatists than one with fewer soldiers. In alternate tests, I substituted a measure of the average annual military expenditures of the government as a percent of the average GNP of a country during the duration of a challenge. 14 Countries with relatively small defense expenditures were expected to be more likely to accommodate challengers than countries with large defense expenditures. 15 The second measure, government instability, was a dummy variable indicating whether a country in question experienced rapid regime change at any point during the duration of a challenge. 16 Governments experiencing rapid regime change were expected to be weak relative to those that had not. Finally, a measure of average gross domestic product per capita over the course of a dispute was also tested in alternate specifications. Four indicators were used to measure the strength of the separatist group. The first neighboring ethnic groups indicates whether the group seeking selfdetermination was part of a larger ethnic group that extended beyond that country s borders. All else equal, groups with ethnic brethren in neighboring states were viewed as relatively more powerful than groups with no such outside support (see especially Cetinyan 2002; Saideman 1997). The second measure focused on group concentration. Groups that were more highly concentrated geographically were assumed to be better able to overcome difficult collective action problems and mobilize more effective resistance against the government (Toft 2001, 2003). The third measure indicated whether a country contained mountainous terrain, which was assumed to be advantageous to the rebel side (Fearon and Laitin 2003). The last measure was the presence of a diaspora community in the United States. Ethnic groups with a large proportion of co-ethnics living in the United States 14 Neither military personnel nor expenditures is a perfectly reliable measure of government strength. Weak governments, for example, may feel particularly insecure and invest in large armies as a result. A large number of soldiers and a large military budget, therefore, would better measure state weakness than strength. It is also possible that a spurious relationship exists between large militaries and government accommodation if hardline governments (ones that do not to accommodate) also choose to have a bigger military, but the decision to accommodate and to have a large army are unrelated. Readers should be aware of the crudeness of this measure, but should also realize that it is the best we have to date. 15 I also looked at arms imports as a possible measure of government strength, but problems with missing data and concerns about accurate reporting led me to exclude this measure from the final analysis. It was not, however, significantly related to government accommodation in any test that was undertaken. 16 Countries that experienced a change of three or more points in any year on Polity IV s combined measure of democracy/autocracy were coded 1. All others were coded 0.

6 318 BARBARA F. WALTER were assumed to have access to more remittance money than those that did not (Collier and Hoeffler 2001). By contrast, the reputation model argued that the decision to accommodate would be less focused on the land currently under dispute and the relative capabilities of the current combatants, and more on future challengers and expected future costs. Hypothesis one predicted that the greater the number of potential separatist challengers, the less willing a government would be to accommodate a given challenge. Governments would look down the road, determine how many groups could potentially make demands and respond accordingly. 17 Three proxies were used to measure the number of potential separatist challengers. The first was a measure of the total number of ethnic groups in each country as identified by the Encyclopedia Britannica. This measure included all ethnic, linguistic, or religious groups, as well as all foreign national groups in each country. 18 Since all groups are not equally likely to seek selfdetermination and since governments are likely to be particularly attentive to those parties that are most likely to challenge, I included two more nuanced measures of potential separatists. The first is a measure of the number of Ethnopolitical Groups in a given country. 19 This measure 17 This hypothesis assumes that governments can determine who the future challengers are likely to be, the extent of their claims, and the resources they are likely to have at their disposal. Although the government can never know exactly how many groups will seek selfdetermination, or the exact future value of territory, governments are likely to have some knowledge about potential troublemakers. Since governments generally control a confined geographical area, with a relatively fixed population, assessing potential challengers should not be too difficult. By examining several different estimates of the number of potential separatist challengers, I hope to show that governments can and do distinguish between different groups to identify the most likely challengers. 18 The advantage of using a measure that includes all ethnic groups regardless of their level of mobilization or grievance is that it ensures that any relationship found between government accommodation and the number of groups is not a function of government behavior itself. This is less the case when a measure of concentrated or ethnopolitical groups from the MAR dataset is used. Governments that are unwilling to accommodate or cooperate with any group under any conditions may, by their very behavior, increase the number of mobilized or dissatisfied groups in their country. This would make it appear as if a higher number of ethnopolitical groups caused governments to refuse accommodation when in fact causation was reversed. 19 Note that this measure, as coded by the Minorities at Risk (MAR) dataset, is not trouble free. Laitin and Fearon (2001), for example, point out that MAR was not systematic in its sample selection procedure. Many African countries with large numbers of ethnic groups, for example, are only coded for one or two groups in MAR, while other more well-known countries are coded for significantly more. Still, this measure should give a rough estimate of whether the number of aggrieved groups in a country is more likely to affect government behavior than simply the number of ethnic groups. counts only those groups that were systematically discriminated against, and/or mobilized in defense of their interests and thus may better identify those groups most likely to capture the government s attention as a potential future threat. 20 The second measure, the number of Concentrated Groups, includes only those ethnopolitical groups that are geographically concentrated in a single region. Both measures are taken from the Minorities at Risk (MAR) data project. The reputation model also predicted that the decision to accommodate one challenger would also be strongly influenced by the combined value of the land future challengers might demand. To assess the combined value of land, the previous measures of economic, strategic, and psychological value were once again used, only this time the measures incorporated the value of all land occupied by all ethnopolitical groups in a given country. 21 Finally, as a measure of the potential human costs of future challengers, the Combined Population of all disaffected groups in the country as a proportion of the total population was also included in the analysis. 22 Two additional variables have been linked to government behavior and are included as controls. A range of studies has demonstrated that regime type affects various decisions to fight or cooperate. Although to date no relationship has been found between democracy and the decision by groups to rebel (at least independent of income), there is reason to believe that regime type might be important in the ways in which governments respond to minority demands for self-determination. 23 Democratic governments are likely to be more sensitive to 20 To test for the possibility that the relationship between government concessions and the number of potential challengers is nonlinear, various nonlinear models and thresholds were tested but all provided a significantly worse or at best similar fit to the linear model presented here. It did not appear that there was a critical threshold beyond which increasing the number of groups no longer affected government actions. Nor did it seem that the impact of the additional potential challengers decreased as the total number increased. 21 Note that data were collected on each ethnopolitical group and not on each ethnic group. This decision was made because subnational data on ethnic groups are generally not available, while similar data on ethnopolitical groups are. 22 In alternate tests, I included a measure of the proportion of the total land occupied by ethnopolitical groups who could challenge the government. Since this measure was highly correlated with the Combined Population measure, it was dropped from the final model. These alternate tests suggest, however, that Combined Land (like combined population) may be related to government accommodation. Further tests need to be undertaken to clarify this relationship. 23 Studies by Muller and Weede (1990), Hegre et al. (2001), Fearon and Laitin (2003), and Walter (2006), however, have all found a significant relationship between anocracies and the outbreak of civil conflict.

7 BUILDING REPUTATION 319 the rights of individuals seeking self-determination and have a greater range of possible compromise solutions to offer these groups than nondemocracies (Doyle 1986; Goemans 2000; Morgan and Campbell 1991). A variable, Democracy, was therefore included to take into account this potential effect. 24 I also included measures of federalism, proportional representation, and anocracy in alternate tests to determine which aspect of democracy may more heavily influence a government s decision to make concessions. Finally, a measure of the Duration of a conflict was also included since it is possible that longer disputes are more costly to governments and therefore more likely to end in some form of accommodation. 25 Finally, it is worth noting that in alternate analysis, I included dummy variables for different regions of the world to ensure that none of the significant variables were proxying for some unmeasured region-specific factor. None of the region dummies were significant and their inclusion or exclusion did not affect the overall conclusions. The exact measurement and coding of all of these variables, as well as details on data sources, are available in the appendix. Results and Interpretation The results of the ordered probit regression with Accommodation as the dependent variable are presented in Table 1. What you see in the first column is that three of the factors associated with the reputation theory are significantly related to government accommodation while none of the factors associated with present interests or capabilities are similarly important. In line with the reputation model, the greater the number of ethnic groups in a country, the greater the combined strategic value of future lands, and the greater the potential future loss of population the less likely a government is to accommodate any given challenge. 26 This strongly suggests that in the case 24 To measure democracy, I used the average level of democracy during a conflict as indicated by the Polity IV data set. In alternate tests I also used the highest level of democracy attained during the dispute assuming that governments would be most likely to accommodate at this point. Both measures led to similar results. 25 A dummy variable indicating whether the conflict was violent was also included to see if the use of force by a separatist group increased the likelihood that a government would accommodate their demands. It was insignificant in all specifications and was dropped from the final model. 26 Note that future economic value was neither significant (even when all other measure of future value are dropped) nor signed in the direction predicted by the reputation theory. In his study of self-determination challenges, governments are influenced more by the risks and costs of future challenges than by the costs of current ones. Reputation-building is not just a significant factor in government decision making, it is also a substantively important one. Converting the ordered probit coefficients in Table 1 into predicted probabilities reveals that expectations about what might happen in the future greatly affects government behavior. 27 Governments in countries with a relatively large number of ethnic groups (11) are 30% more likely to refuse any form of accommodation than those that faced a relatively small number of groups (3). A large number of ethnic groups also significantly reduced the odds of a government offering reform or territorial autonomy. Governments facing a high number of ethnic groups were 6% less likely to offer reform and 24% less likely to offer some form of increased territorial autonomy than those facing a low number. 28 The combined strategic value of the land had an even larger substantive effect on government behavior. When the strategic value of all future lands was high, governments were 72% more likely to refuse any form of accommodation than if the combined value was low. Similarly, when the combined proportion of the population in these areas included the entire population of the country, governments were 59% more likely not to accommodate than if potential challengers represented only a tiny fraction of the population. 29 Taken together, these three factors (the number of ethnic groups, combined strategic value, and the size of future populations) greatly determined government actions. Governments that preside over countries with many ethnic groups that occupy particularly strategic land and represent a large proportion of the population are 94% less likely to accommodate any group of separatists than are governments that preside over fewer ethnic groups that control less valuable of interstate territorial conflicts, Huth (1996) also found a positive although insignificant relationship between economic stakes and compromise settlement. He hypothesized that this was because economic resources can be divided more easily, leading to a higher number of settlements. 27 In each case, predicted probabilities are calculated by varying the measure of interest from the 10 th to the 90 th percentile, while holding all other variables at their mean values (or modal values for categorical variables). See King, Tomz, and Wittenberg (1999). 28 Bivariate results confirm this relationship. Overall, 40% of the governments in countries with fewer than three ethnic groups offered territorial accommodation, while only 10% of the government in countries with more than eight ethnic groups did so. 29 In some countries such as Yugoslavia or Iraq all ethnic groups express grievances and are, therefore, potential challengers.

8 320 BARBARA F. WALTER TABLE 1 Ordered Probit Analysis of Factors Affecting Government Responses to Self-Determination Demands Level of Government Accommodation Concentrated Disaffected All Groups Groups Groups REPUTATION MODEL Number of Potential Future Challengers Number of Ethnic Groups.102 (.046) Number of Concentrated Groups.265(.090) Number of Disaffected Groups.268(.075) Value of All Land Occupied by Potential Challengers Combined Economic Value.046 (.031).080 (.034).060 (.032) Combined Strategic Value.313 (.136).360 (.142).252 (.143) Combined Psychological Value.277 (.239).341 (.243).370 (.263) Proportion of Population (All Groups) 2.03 (1.03) 1.41 (1.01) 1.49 (1.10) VALUE OF LAND CURRENTLY UNDER DISPUTE Economic Value Strategic Value.036 (.085).103 (.091).101 (.092) Psychological Value:.049 (.194).091 (.195).038 (.194) Length of Residence.015 (.408).056 (.419).044 (.458) History of Autonomy.392 (.296).544 (.306).575 (.315) Proportion of Population (Current Group) 1.37 (2.03).483 (2.01).303 (2.10) RELATIVE CAPABILITIES: CURRENT DISPUTANTS Government Military Personnel 13.6 (28.6) 5.28 (27.32) 9.36 (28.9) Government Instability.408 (.358).516 (.354).362 (.382) Group Concentration.023 (.209).167 (.221).066 (.212) Neighboring Ethnic Groups.144 (.136).233 (.143).301 (.150) CONTROLS Average Level of Democracy.105 (.027).099 (.027).097 (.028) Duration of Conflict.002 (.009).006 (.009).007 (.009) Constant (1.40) 1.31 (1.43) 1.99 (1.59) Constant (1.40).764 (1.43) 1.42 (1.59) Constant (1.41) 1.66 (1.45).970 (1.58) Pseudo R N p <.10, p <.05, p <.01 land, and represent only a small proportion of a country s population. 30 In sharp contrast, none of the factors related to the land presently under dispute are significantly related to government accommodation. 31 Governments are no 30 Note that data on the size of ethnic groups may also be used to measure the strength of these groups. 31 Even if all of the measures of the combined value of land occupied by ethnopolitical groups is dropped from the model, none of the more or less likely to accommodate demands for selfdetermination if the land in question has high economic value or is strategically important. 32 Table 1 also indicates that the psychological value of a given piece of land, at measures related to the value of the land currently under dispute approached significance. 32 Logged measures of economic and strategic value were both not significant when introduced into the model.

9 BUILDING REPUTATION 321 least as measured by available indicators, does not appear to play any role in a government s decision to compromise. Again, readers should be somewhat careful about interpreting these results. The logic of the reputation argument implies that governments should not ignore the present value of stakes, only that they should assess the value of present stakes compared to future stakes. Governments, therefore, should be willing to fight for particularly important pieces of land, such as those that contain large oil reserves, even if there are few reputational gains to be made for doing so. When more nuanced measures of economic value are included in the analysis (results presented in Table 2), the significance of present value becomes more apparent. Finally, there is no significant relationship between any of the measures of relative strength during the period of conflict and the level of accommodation in that conflict. 33 Balance of capabilities arguments predict that stronger governments will be less willing to accommodate challengers than weaker governments. Yet neither military size nor government instability had a significant influence on government decisions to accommodate. Governments with small militaries or those that experienced rapid institutional change were no more likely to accommodate than those that maintained a relatively large number of soldiers under arms or were institutionally stable. Finally, the strength of the challenger group also appears to be unrelated to how a government behaves. Governments were no more or less likely to accommodate separatist groups that extended into neighboring states or were highly concentrated in a particular region. 34 The fact that so many measures of future challengers and stakes are significant suggests that previous studies have overlooked an important strategic element in government decisions to fight or accommodate. Although these results should be considered with some caution due both to the imprecise nature of the indicators and the potential bias inherent in selecting only cases where a formal challenge has occurred, they are still striking. Govern- 33 Goertz and Diehl (1992) had a similar finding on their work on interstate territorial disputes. They found no strong relationship between the balance of military capabilities and the use of armed force in achieving territorial changes from 1914 to However, in separate studies, Mandel (1980); and Kacowicz (1994) found that armed conflict over territory was more likely between states of roughly equal military capabilities. For a discussion of these differences see Huth (1996, 13 14). 34 This mirrored two of the findings in Huth s 1996 study of interstate territorial disputes. He found that the presence of minority groups along the border with ethnic ties to the general population of the challenger had no strong effect on the likelihood of a compromise settlement. He also found that military strength of the challenger was not a necessary condition for the challenger to remain unyielding in its negotiating position over disputed territory. ments appear to consistently factor in the value of future stakes as well as the number and strength of future players when determining how to respond to a given challenge. Present stakes and power may still matter, but various features associated with future confrontations appear to matter even more. One factor outside of reputation also affected government behavior. More democratic regimes were significantly more likely to accommodate separatist demands than less democratic regimes. In fact, converting the ordered probit coefficients into predicted probabilities revealed that highly democratic regimes were 61% less likely than highly autocratic regimes to refuse to agree to some type of accommodation. Democracies were also more willing to offer both reform and increased territorial autonomy when confronted with demands for self-determination. There are a number of possible explanations for this, to be examined later in the article. Robustness In testing the reputation model, I attempted to include as many valid measures of alternative explanations as possible but the inclusion of all of these variables in the regression in Table 1 could be misleading if collinearity is a problem. To address this potential issue, I repeated the analysis in Table 1 with a range of different specifications. In a series of tests, I repeated the regression dropping all of the measures associated with the reputation model, including only one measure for each underlying construct (e.g., including only economic value and dropping other measures of land value), and dropping all insignificant measures. The substantive conclusions were the same. Bivariate correlations also demonstrate very similar patterns to those illustrated in Table Since there is also some concern that looking only at groups that have actively challenged a government might lead to bias, I repeated the analysis in Table 1 including all ethnic groups included in the Minorities at Risk (MAR) dataset not just those that actively challenged the government. 36 The results confirm the central role that repu- 35 The only signs of collinearity are between combined economic value and combined strategic value. The two measures are correlated at.77 and when combined strategic value is dropped the coefficient on combined economic value becomes positive and almost significant suggesting once again that future stakes weigh heavily on government decision makers. 36 Due to data limitations, the regression model for this alternate test only included the number of ethnic groups and not the combined population size or the combined strategic, economic, and psychological value of land occupied by potential challengers.

10 322 BARBARA F. WALTER TABLE 2 Ordered Probit Analysis of Factors Affecting Government Responses to Self-Determination Demands Level of Government Accommodation Model 1 Model 2 Model 3 REPUTATION MODEL Number of Potential Future Challengers Number of Ethnic Groups.083 (.046).087 (.052).095 (.046) Value of All Land Occupied by Potential Challengers Combined Economic Value.030 (.030).028 (.027).049 (.034) Combined Strategic Value.301 (.138).295 (.146).276 (.148) Combined Psychological Value.364 (.249).420 (.268).116 (.255) Proportion of Population (All Groups) (.940) 2.92 (1.23) 2.37 (.996) VALUE OF LAND CURRENTLY UNDER DISPUTE Economic Value Presence of Oil/Gas Reserves.032 (.086).038 (.096) Strategic Value.970 (.420) Psychological Value:.093 (.191).050 (.218).018 (.198) Length of Residence.084 (.403).052 (.428).028 (.501) History of Autonomy.606 (.317).533 (.332).123 (.438) Proportion of Population (Current Group).147 (1.99) (2.37).958 (1.95) RELATIVE CAPABILITIES: CURRENT DISPUTANTS Government Military Personnel (Per Capita) 8.05 (28.87) (34.60) Government Military Expenditure (Per Capita) 1.20 (2.32) Government Instability.651 (.425) Average GDP 2.16 (3.33) Mountainous Terrain.150 (.165) Group Concentration.124 (.209).046 (.272).026 (.214) Neighboring Ethnic Groups.072 (.133).121 (.147).147 (.143) CONTROLS Average Level of Democracy.094 (.031).088 (.029) Proportional Representation System.627 (.387) State Centralization//Federalism.422 (.248) Anocracy.630 (.411) Duration of Conflict.010 (.009).002 (.011) Constant (1.39) 1.75 (1.46) 2.31 (1.76) Constant (1.39) 1.21 (1.45) 1.76 (1.76) Constant (1.44) 1.21 (1.48).434 (1.73) Pseudo R N p <.10, p <.05, p <.01 tation plays. Even when all MAR groups are included, accommodation is significantly less likely in countries with a large number of potential challengers. Finally, I assessed an alternate model that looked at the duration of time before a government accommodated rather than at how much accommodation the government offered. Although it is difficult to code duration since accommodation can occur in different amounts at different times, I was able to conduct analysis using the date of the last accommodation and a simple dummy variable indicating whether the government offered territorial accommodation or not. This weibull hazard model

11 BUILDING REPUTATION 323 reaffirmed the role reputation building plays in government decision making. Territorial accommodation took significantly longer in cases where a government could potentially face more challengers and lose more valuable territory. All of these tests help to increase confidence that the conclusions in Table 1 are robust (analysis not shown). Which Groups Matter The results so far support the reputation theory but tell us nothing about the types of groups that are most likely to threaten a government. Governments, for example, know that some ethnic groups are far more likely to seek self-determination than others. Groups that are more geographically concentrated, are mobilized to pursue particular interests, or have grievances against the state have greater motivation and opportunity to seek selfdetermination. Government behavior, therefore, should be more closely linked to these more threatening candidates than to the larger population of ethnic groups used in the previous analysis. I tested this proposition by substituting into the analysis two different measures of the number of potential challengers in a country. 37 The second regression in Table 1 shows the effect of the number of concentrated ethnopolitical groups on government behavior, while the third regression shows the effect of the number of ethnopolitical groups. Not only are both significantly related to government behavior, but the predicted probabilities make clear that governments are particularly concerned with these more troublesome groups. Governments in countries with large numbers of concentrated groups were 70% less likely to offer any form of accommodation than governments in countries with the fewest concentrated groups. Similarly, going from a country with the largest number of ethnopolitical groups to one with the fewest increased the probability of accommodation by 77%. Each of these effects is almost twice as large as the effects we saw when we looked at the more general measure of the number of ethnic groups. This suggests that governments are particularly influenced by groups that show real signs of future action. Alternate Specifications A number of the measures included in the previous analysis are quite rough. In what follows, I test a series of alternate measures in order to corroborate these findings and to delve more deeply into the exact nature of a number of the relationships. 37 Due to collinearity between the three measures of potential challengers, including all three in the same model would be suspect. One of the key findings to emerge out of the Table 1 is the significant relationship between democracy and government accommodation. In all three models, more democratic countries are significantly more likely to accommodate demands by separatists than less democratic countries. But what is it about democracy that leads to accommodation? To assess the effect of different institutional features on accommodation, model 1 in Table 2 includes measures for Proportional Representation, Federalism, and a dummy variable singling out Anocracies. The results of this more in-depth analysis of democracy are interesting but far from conclusive. Each of the three measures is positively and significantly or nearly significantly related to government accommodation, suggesting that all three factors may play a role. Governments may be more willing to accommodate demands because they already have a system in place that grants regional autonomy (e.g., federalism), because minority groups are influential in government decisions (e.g., proportional representation), or because they can neither fully incorporate nor fully repress groups (e.g., anocracies). However, issues of collinearity make each of these conclusions questionable. It is also possible that federal reforms occur after territorial autonomy has been granted to a group and not the other way around. Future research will be required to distinguish which, if any, of these alternatives is at play. In the second model of Table 2, I attempt to assess more fully the role current value has on government behavior. Since counting the number of resources contained in region is an imprecise measure of the economic value of the land, I substituted a measure that singled out regions with significant oil or gas reserves. This measure might better assess the economic importance of a region in at least some countries. The results in Table 2 indicate that governments are, in fact, less willing to give up autonomy over a piece of land that contains either of these important resources. Finally, in the third model of Table 2, I include three new measures of government strength not tested in Table 1. Since the number of military personnel does not always translate into government military strength, I substituted a measure of annual military expenditures as a proportion of the total GDP. Model 3 also includes a measure of mountainous terrain, a geographic feature that was assumed to reduce a government s ability to defeat rebel forces. Lastly, I include a measure of the average annual gross domestic product of the country to reflect the fact that government strength is at times tied to its economic resources. In the end none of these alternate measures of government capabilities is significantly related to government accommodation, while the earlier findings regard-

12 324 BARBARA F. WALTER ing the importance of reputation building remained robust. Does Reputation Building Work? But does a reputation-building strategy work? If it is true that government behavior is motivated by a desire to deter additional challenges, then it should also be true that refusing to accommodate one challenger has the desired deterrent effect. If it does not, then investing in reputation would have little merit, and studies questioning the efficacy of reputation building would likely be correct. A closer look at the data suggests that a government strategy of reputation building in the face of a multiethnic population is effective. Governments that refused to accommodate or offered only modest political reforms when they were first challenged faced a subsequent challenge for self-determination only 27% of the time (15 of 55 cases). 38 By contrast, governments that granted territorial autonomy or independence to their first challenger faced a subsequent challenge 59% of the time (13 of 22). This pattern was evident in Indonesia in the aftermath of Habibie s decision to grant independence to East Timor. In the words of an Indonesian diplomat, this decision sparked a scramble among secessionists from Aceh and to a certain extent even in Irian Jaya of wanting similar referendums held in their provinces with similar objectives (New Straits Times Press 5/29/03, 2). Greater degrees of accommodation also appear related to the number of subsequent challengers. Countries that offered no territorial autonomy faced, on average, well under one subsequent challenge (.78 challenges), whereas countries that offered territorial autonomy faced, on average, slightly over one subsequent challenge (1.05 challenges). And countries that gave their first challenger full independence faced, on average, two additional challenges (2.0 challenges). Preliminary multivariate analysis presented in Table A1 in the appendix suggests that this relationship holds even after controlling for the number of potential subsequent challenges, the number of years since the first challenge, and the value of the land other groups might claim. Although this analysis does not assess the full range of motivations driving potential challengers and more work needs to be done, these preliminary results support the idea that reputation building is an effective means of deterring future challenges. 38 Although this was not the very first challenge that some of the governments had encountered in their history, it still represented an early challenge relative to all others included in the dataset. Ideally, one would want to include the universe of all self-determination movements that had ever been launched in a given country. Limitations on time as well as on data made this impossible. Conclusion This article attempted to show that future players and future stakes two factors generally ignored by political scientists strongly influence government decisions to cooperate or fight at least when self-determination is at stake. Applying insights from the work on entrydeterrence in economics, this article argued that governments of multiethnic countries have strong incentives to fight against early separatists in order to deter additional separatists later on. Fighting a war against the first separatist challenger is the trade-off governments consciously make in order to influence how others are likely to play the game. Data on all self-determination movements between 1956 and 2002 strongly supported this theory. Not only did the number of future challengers significantly affect a government s decision to accommodate a challenge, but governments were more likely to invest in reputation when the combined value of future stakes was high. This investment also appeared to pay off. Governments that refused to accommodate one challenger were significantly less likely to face a challenge from a second or third group down the road. It appears that governments use war not only to influence the behavior of one particular opponent, but other opponents as well. Despite these strong results, this study leaves a number of important questions unanswered. This article, for example, says almost nothing about the strategic behavior of challengers. If it is true that governments make decisions based on how they believe ethnic groups will respond, then it must also be true that ethnic groups are equally strategic in their behavior toward their governments. Potential separatists, for example, should wait until after they observe the government granting concessions to another group before they themselves challenge. 39 They should also coordinate their challenges with other ethnic groups in order to present a unified front against the government. A united front would reduce the incentives a government had to invest in reputation building (there are fewer additional challengers to deter), and make conciliation more likely. 40 Finally, this article says nothing 39 Work by Fearon (1994) gives us insight as to why some ethnic groups might choose to challenge the government before any others. Fearon argued that ex ante observable variables should be taken into account by rational potential challengers in their decision to challenge or not. Since ethnic groups understand that governments have greater incentives to fight early challengers, especially if there are many additional challengers waiting in the wings, only the most resolved challengers will move early. 40 Failures by potential challengers to combine their forces may be due to straightforward collective action and coordination problems, an argument that is developed further in Walter (2004).

13 BUILDING REPUTATION 325 about the effects of a leader s time in office on his or her decision to invest in reputation. Leaders who operate under term limits, who know they are likely to be in power for a short time, or are nearing the end of their tenure should have fewer reasons to build a reputation for toughness and more incentives to grant concessions. I have not addressed these questions in this particular article due to time and space constraints. Readers interested in these additional implications of the reputation argument can find more detailed analyses of each issue in a related book manuscript by the author. Even though additional work needs to be done, the research to date contributes to our understanding of conflict and cooperation in three important ways. The first has to do with the importance of including future parties and future stakes in our models of bargaining and war. Some of the most interesting work being done in international relations today sees the onset, duration, and resolution of war as part of an ongoing negotiation process, where war is simply a continuation of bargaining using different means (Filson and Werner 2002; Powell 2002; Smith and Stam 2001). But almost all of this literature focuses only on the two players directly involved in a conflict and not on other parties that almost certainly observe this behavior and adjust their strategies accordingly. What this research shows is that we need to move away from static two player models, toward more dynamic models that take into account future players and future stakes. Second, the theory and results presented in this article should apply equally well to any situation where one party expects a series of similar challenges from similar parties over time. Governments, for example, with a history of territorial disputes with neighboring countries, should be particularly concerned with reputation building. The same can be said about government responses to terrorist demands. The tough stance of the United States toward kidnapper in Iraq, for example, is designed to deter additional kidnappings from occurring. Finally, one might also apply this logic to the problem of conflict diffusion or contagion. Each of these situations exhibit the same underlying strategic problem: in an uncertain world where present behavior often provides important information about future behavior, how do you prevent challenges from multiplying? Finally, this study contributes to debates on reputation and deterrence. Reputation arguments have lost some influence in the international relations literature because existing empirical studies do not appear to support the notion that reputation building works. 41 For example, in a study of six Third World countries, Hopf (1994) 41 See especially Huth and Russett (1984, 1988), Hopf (1994), Mercer (1996), and Press (2001). found that Moscow did not infer anything about America s likely behavior in Europe based on how it had behaved in more peripheral regions. The problem with these studies is that they have examined the effects of reputation building across dissimilar contexts where different stakes were under dispute. Fighting to create a reputation for resolve might not have a significant effect when you look across different issue areas, geographic regions, or if you include a diverse set of players. But it does appear to have a strong effect when you look at similar players, fighting for similar stakes, against the same opponent over time. This article, therefore, provides some of the first systematic evidence that in a particular domain government responses to self-determination movements governments invest very heavily in reputation, and this strategy appears to work. Appendix Coding Rules and Sources Dependent Variables Accommodation. Coded (0) no accommodation, (1) reform not over territory, (2) some form of territorial autonomy, (3) independence. A dispute was coded as having ended in no accommodation if the government was unwilling to make any concessions. If the government was willing to offer some concessions such as greater participation in government, or greater cultural autonomy in response to a challenge, the case was coded as reform. A dispute was coded as having ended in territorial autonomy if the government was willing to grant the challenger greater political or economic autonomy over their own region. Finally, a dispute was coded as having ended in independence if the government granted full independence to the challenging party. If a government experienced multiple negotiations with the same challenger over the course of a dispute, the case was coded based on the highest level of accommodation offered. Source: Keesing s Contemporary Archives, group profiles provided by the Minorities at Risk data project (see bsos.umd.edu/cidcm/mar/list.html), and individual case histories. Subsequent Challenge. Coded based on the number of subsequent self-determination challenges a government faced after the case under observation had begun. Source: CIDCM case list of violent and nonviolent self-determination movements between 1946 and 2002, phase II study.

14 326 BARBARA F. WALTER Independent Variables Hypotheses Derived from the Reputation Theory Number of Ethnic Groups. The total number of ethnic groups in a country, as listed in the People and Ethnicity section of the Encyclopedia Britannica Country Essays from Encyclopedia Britannica Online. See website: Number of Concentrated Groups. The total number of ethnopolitical groups that were either a majority in one region,orconcentratedinoneregion.basedonthemar coding of group spatial concentration. Source: Minorities at Risk (MAR) dataset, version IV, variable GROUPCON. Number of Ethnopolitical Groups. A variable indicating the total number of ethnopolitical groups in a given country. Each group had to meet one of two criteria. A group had to be disadvantaged in comparison to other groups in their society, usually because of discriminatory practices, or a group had to be politically organized to promote or defend their collective interests. Source: Minorities at Risk (MAR) dataset, version IV. Combined Economic Value, Combined Strategic Value, Combined Autonomy. Composite measures that add up the economic and strategic value of all territories occupied by all ethnopolitical groups in the country, as well as the number of ethnopolitical groups that had been historically autonomous from the central government at any point in time. Coding for each individual measure from which these composite measures are created and its source is noted below. Combined Population Proportion. A variable indicating the proportion of the total national population represented by all ethnopolitical groups in a country. The measure was the average population for the years 1990, 1995, and 1998, the only years for which MAR data were available, divided by the average total population in those years. MAR IV, variable GPRO. Hypotheses Derived from Theories Based on Present Interests Economic Value. A 31-point scale indicating how many marketable resources were known to exist on a given piece of land. These resources included oil/petroleum, natural gas, coal, iron ore, steel, aluminum, titanium, bauxite, salt, sulfur, tin, nickel, chromium, cobalt, tungsten, phosphate, gold, silver, copper, zinc, lead, magnesium, uranium, diamonds, water, wheat, timber, fishing, tourism, commercial nuclear plants, or important manufacturing plants. Source: U.S. Geological Survey maps, the CIA Factbook, CIA maps, and group profiles provided by the Minorities at Risk data project. Population Proportion. The proportion of the total national population represented by the group launching the challenge under observation. The measure was the average population for the group in the year 1990, 1995, and 1998, the only years for which MAR data were available divided by the average national population in those years. MAR IV, variable GPRO; data on total population were obtained from the World Bank (WDI) and Penn World Tables. Proportion of Territory. The percent of the total land mass (in square kilometers, 1995) that an ethnopolitical group occupied. Source: For identification of the region occupied by a group: Fearon and Laitin dataset, and MAR group profile. For size of region occupied by the group: The Columbia Gazetteer, and the Food and Agriculture Organization of the United Nations. When I could not identify a case with either of these sources, I searched for alternate sources through the Perry-Castaneda Library Map Collection, University of Texas. For total land mass of a given country: World Bank Indicators. Strategic Value. Scale indicating whether territory included (1) a sea outlet, (2) a shipping lane, (3) a military base, (4) an international border, (5) an attack route, and/or (6) a mountain range. Source: U.S. Geological Survey maps, the CIA Fact book and related maps, and groups profiles provided by MAR. Length of Residence. Coded 1 if immigration occurred since 1945, 2 = in the nineteenth or early twentieth century, or 3 = pre Source: MAR VI data set, variable TRADITN2. Autonomy. A dummy variable coded 1 = the group was historically autonomous at some point in time, 0 = group was never historically autonomous. Source: MAR IV data set, variable AUTON. Hypothesis Derived from Theories Based on the Balance of Capabilities Military Personnel/Population. Average annual number of military personnel during the duration of the selfdetermination movement under observation as a percent of the total population, divided by the average total population for a country during the duration of the movement. Source: J. David Singer and Melvin Small s Material Capabilities, dataset, which collected its information from The Military Balance which is published annually by the International Institute for Strategic Studies. Information for was added using IISS. Data on total population were obtained from the World Bank (WDI) and Penn World Tables.

15 BUILDING REPUTATION 327 Military Expenditures/Average GDP. Average annual military expenditures in thousands of 1996 US dollars during the self-determination movement under observation as a proportion of average GDP during that same time period. Source: same as military personnel, see above. Government Instability. A dummy variable indicating whether the country in question had experienced a three or more change in its Polity IV combined democracy/autocracy score in any year during the course of the self-determination movement under observation. Source: Polity IV, variable POLITY. Gross Domestic Product/Per Capita. Measures the average per capita GDP of a country over the duration of a challenge. Source: Penn World Data. Neighboring Ethnic Groups. The total number of adjoining countries where segments of the ethnopolitical group involved in the dispute also resided. Source: MAR data set, variable: NUMSEGX. Group Concentration. The geographical concentration of a group ranging from widely dispersed to concentrated in one region. Source: MAR data set, variable GROUPCON. Mountainous Terrain. The proportion of the country that is mountainous according to the codings of geographer A.J. Gerard. Source: Fearon and Laitin (2003) dataset on civil wars. Diaspora Community in the United States. Relative size of the group s population among the U.S. foreignborn population. Source: U.S. Bureau of the Census ( World Bank. Controls Democracy. A democracy-autocracy measure derived from a government s autocratic features and its democratic features. The incumbent government s autocracy score was subtracted from its democracy score to produce a net democracy number that ranges in value from very autocratic ( 10) to very democratic (+10). Based on the average democracy score for a country for the duration of a challenge. Source: Polity IV data set, Marshall and Jaggers (2003), variable: POLITY. Proportional Representation. A dummy variable coded 1 if a political system was based on proportional representation, 0 if not at any point during the duration of the dispute. Source: World Bank (Keefer Dataset) for years ; Stateman s Yearbook for years Degree of Centralization/Federalism. From the Polity III data set that indicates the degree to which political authority is centralized (variable: CENT). Coded 1 for unitary states (centralized) where only moderate decision-making authority is vested in local or regional governments; 2 for an intermediate category; 3 for a federal state (decentralized) where local and/or regional governments have substantial decision-making authority. This variable was coded based on the lowest centralization score during the duration of the dispute. Source: Jaggers and Gurr (1996). Anocracy. A dummy variable indicating whether a government scored between 5 and 5 on Polity IV s democracy/autocracy measure at any point during a challenge. Source: Polity IV, Marshall and Jaggers (2003), variable POLITY. Duration/Years. Number of years from the year a challenge was initiated to the year it officially ended. Source: CIDCM; and Group profiles provided by MAR. Violence. A dummy variable coded 1 if a given group scored a 4, 5, 6, or 7 on MAR s REBEL scale. Coded 0 if it scored < 4. Source: MAR data set, variable: REBEL45X REBEL98X. TABLE A1 Ordered Probit Analysis of Factors Affecting the Likelihood of a Subsequent Territorial Challenge Standard Independent Variables Coefficient Error Reputation Model Accommodation Factors Related to the Value of Future Land Combined Economic Value Combined Strategic Value Combined Proportion of the Population Combined Autonomy Opportunity Number of Ethnopolitical Groups Years Since First Conflict Log of Country Population Controls Violence Duration Mountainous Terrain Constant Pseudo R N 59 p <.10, p <.05, p <.01

16 328 BARBARA F. WALTER Summary Statistics Variables Mean Std. Deviation Minimum Maximum ACCOMODATION REPUTATION MODEL Number of Potential Challengers Number of Ethnic Groups Number of Concentrated Groups Number of Disaffected Groups Value of All Land Held by Potential Challengers/Strength of Future Challengers Combined Economic Value Combined Strategic Value Combined Psychological Value Proportion of Population (All Groups) VALUE OF LAND CURRENTLY UNDER DISPUTE Economic Value Presence of Oil/Gas Reserves Strategic Value Psychological Value: Length of Residence History of Autonomy Proportion of Population (Current Group Only) BALANCE OF POWER: CURRENT COMBATANTS Government Military Personnel Government Military Expenditures Government Instability Average GDP 2.55e e e+09 Mountainous Terrain Geographic Concentration Neighboring Ethnic Groups CONTROLS Average Level of Democracy Proportional Representation System State Centralization/Federalism Anocracy Duration of Conflict SUBSEQUENT CHALLENGE Subsequent Challenge References Alt, James E., Randall L. Calvert, and Brian Humes Reputation and Hegemonic Stability: A Game-Theoretic Analysis. American Political Science Review 82(2): Bartkus, Viva O The Dynamic of Secession. Cambridge: Cambridge University Press. Cetinyan, Rupen Ethnic Bargaining in the Shadow of Third-Party Intervention. International Organization 56(3): Coakley, John, ed The Territorial Management of Ethnic Conflict. London: Frank Cass. Collier, Paul, and Anke Hoeffler Greed and Grievance in Civil War. Research Working Paper 2355, World Bank, Washington. Diehl, Paul F., Territorial Dimensions of International Conflict: An Introduction. Conflict Management and Peace Science 15(1):1 10. Diehl, Paul F., ed A Road Map to War: Territorial Dimensions of International Conflict. Nashville: Vanderbilt University Press.

17 BUILDING REPUTATION 329 Dixit, Avinash Recent Developments in Oligopoly Theory. American Economic Review 72(2): Doyle, Michael Empires. Ithaca: Cornell University Press. Doyle, Michael Liberalism and World Politics. American Political Science Review 80(4): Fearon, James D Signaling Versus the Balance of Power and Interests: An Empirical Test of a Crisis Bargaining Model. Journal of Conflict Resolution 38(2): Fearon, James D., and David D. Laitin Ethnicity, Insurgency, and Civil War. American Political Science Review 97(1): Filson, Darren, and Suzanne Werner A Bargaining Model of War and Peace: Anticipating the Onset, Duration, and Outcome of War. American Journal of Political Science 46(4): Gilpin, Robert War and Change in World Politics. New York: Cambridge University Press. Goemans, Henk Erich Domestic Politics and the Causes of War Termination: The Fate of Leaders and the First World War. Princeton: Princeton University Press. Goertz, Gary, and Paul Diehl Territorial Changes and International Conflict. London: Routledge. Grieco, Joseph Anarchy and the Limits of Cooperation. International Organization 42(3): Gurr,TedRobert.1970.Why Men Rebel. Princeton: Princeton University Press. Gurr, Ted Robert, and Will H. Moore Ethnopolitical Rebellion: A Cross-Sectional Analysis of the 1980s with Risk Assessments for the 1990s. American Journal of Political Science 41(4): Hegre, Havard, Tanja Ellingsen, Scott Gates, and Nils Petter Gleditsch Toward a Democratic Civil Peace? Democracy, Political Change and Civil War, American Political Science Review 95(March): Holsti, Kalevi J Peace and War: Armed Conflicts and International Order, New York: Cambridge University Press. Hopf, Ted Peripheral Visions: Deterrence Theory and American Foreign Policy in the Third World, Ann Arbor: University of Michigan Press. Huth, Paul K Extended Deterrence and the Prevention of War. New Haven: Yale University Press. Huth, Paul K Standing Your Ground: Territorial Disputes, Ann Arbor: University of Michigan Press. Huth, Paul K Enduring Rivalries and Territorial Disputes, Conflict Management and Peace Science 15(1):7 42. Huth, Paul K Reputations and Deterrence: A Theoretical and Empirical Assessment. Security Studies 7(1): Huth, Paul, and Bruce Russett What Makes Deterrence Work? World Politics 35(4): Huth, Paul, and Bruce Russett Extended Deterrence and the Outbreak of War. American Political Science Review 82(2): Kacowicz, Arie The Problem of Peaceful Territorial Chance. International Studies Quarterly 38(2): Kennan, John, and Robert Wilson Bargaining with Private Information. Journal of Economic Literature 31(1): King, Gary, Michael Tomz, and Jason Wittenberg Making the Most of Statistical Analyses: Improving Interpretation and Presentation. edu/preprints.shtml. Kreps, David M., and Robert Wilson Reputation and Imperfect Information. Journal of Economic Theory 27: Laitin, David, and James Fearon Minorities at Risk Database and Explaining Ethnic Violence. NSF grant proposal, posted on apsanet.org. Luard, Evan War in International Society. London: I.B. Tauris and Company. Mackinder, Halford Democratic Ideals and Reality. New York: Henry Holt. Mandel, Robert Roots of the Modern Interstate Border Dispute. Journal of Conflict Resolution 24(3): Marshall, Monty G., and Ted Robert Gurr Peace and Conflict, 2003: A Global Survey of Armed Conflicts, Self- Determination Movements, and Democracy. University of Maryland, College Park: Center for Int l Development and Conflict Management. Mercer, Jonathan Reputation and International Politics. Ithaca: Cornell University Press. Milgrom, Paul, and John Roberts Predation, Reputation, and Entry Deterrence. Journal of Economic Theory 27(August): Morgan, T. Clifton, and Sally Campbell Domestic Structure, Decisional Constraints, and War: So Why Kant Democracies Fight? Journal of Conflict Resolution 35(2): Morgenthau, Hans Politics Among Nations. NewYork: Alfred A. Knopf. Muller, Edward N., and Erich Weede Cross-National Variation in Political Violence: A Rational Action Approach. Journal of Conflict Resolution 34(4): Nalebuff, Barry Rational Deterrence in an Imperfect World. World Politics 43(3): Newman, David Real Spaces, Symbolic Spaces: Interrelated Notions of Territory in the Arab-Israeli Conflict, in Diehl ed A Road Map to War: Territorial Dimensions of International Conflict. Nashville: Vanderbilt University Press. Powell, Robert Bargaining Theory and International Conflict. Annual Review of Political Science 5:1 30. Press, Daryl G Power, Reputation, and Assessments of Credibility During the Cuban Missile Crisis. Presented at the 97th Annual Meeting of the American Political Science Association, August, Boston. Press, Daryl G Calculating Credibility. Ithaca: Cornell University Press. Richardson, Lewis Fry Statistics of Deadly Quarrels. Chicago: Quadrangle Publications. Rosenthal, Robert W Games of Perfect Information, Predatory Pricing and the Chain-Store Paradox. Journal of Economic Theory 25(August):

18 330 BARBARA F. WALTER Saideman, Stephen M Explaining the International Relations of Secessionist Conflicts: Vulnerability vs. Ethnic Ties. International Organization 51(4): Schelling, Thomas Arms and Influence. New Haven: Yale University Press. Selten, Richard The Chain-Store Paradox. Theory and Decision 9(2): Schmalensee, Richard Economies of Scale and Barriers to Entry. Journal of Political Economy 89(6): Schmalensee, Richard Advertising and Entry Deterrence: An Exploratory Model. Journal of Political Economy 91(4): Smith, Alastair, and Allan C. Stam Bargaining and the Nature of War, Unpublished manuscript, New York University. Smith, Anthony Nationalism. Cambridge: Blackwell Publishers. Spence, Michael Job Market Signaling. Quarterly Journal of Economics 87(August): Tilly, Charles From Modernization to Revolution. Reading: Addison-Wesley. Toft, Monica Duffy A Theory of Territory, Indivisibility, and Ethnic Violence. Unpublished paper, JFK School of Government, Harvard University. Toft, Monica Duffy The Geography of Ethnic Conflict: Identity, Interests, and the Indivisibility of Territory. Princeton: Princeton University Press. Triesman, Daniel Rational Appeasement. International Organization 58(2): Touval, Saadia The Boundary Politics of Independent Africa. Cambridge: Harvard University Press. Vanzo, John P Border Configuration and Conflict: Geographical Compactness as a Territorial Ambition of States. In A Road Map to War: Territorial Dimensions of International Conflict, Paul F. Diehl ed. Nashville: Vanderbilt University Press, pp Walter, Barbara F Explaining the Intractability of Territorial Conflict. International Studies Review 5(4): Walter, Barbara F War as a Reputation Problem. Unpublished Manuscript. University of California, San Diego. Walter, Barbara F Information, Uncertainty and the Decision to Secede. International Organization 60(1). Waltz, Kenneth Theory of International Politics. Reading, MA: Addison-Wesley Publishing Co.