Incomplete Democratization and the Outbreak of Military Disputes

Transcription

1 International Studies Quarterly ~2002! 46, Incomplete Democratization and the Outbreak of Military Disputes Edward D. Mansfield University of Pennsylvania Jack Snyder Columbia University Whereas most research on the democratic peace has focused on relations within pairs of states, research on the relationship between democratization and armed conflict has centered primarily on the behavior of individual states. Moreover, the existing literature has placed primary emphasis on explaining the effects of democratization on war, rather than military disputes more generally. In this article, we find that certain types of democratic transitions markedly increase the risk of such disputes within dyads, even when economic and political relations between states are taken into account. Particularly prone to violence are dyads in which either state undergoes an incomplete democratic transition; that is, a shift from an autocratic to a partially democratic ~or anocratic! regime that stalls prior to the establishment of consolidated democratic institutions. Because mature democracies have rarely fought wars against each other, many political leaders and scholars urge the spread of democracy as a means to promote peace ~Doyle, 1986; Russett, 1993; Clinton, 1994!. But while a world of mature, consolidated democracies might indeed be more peaceful, a set of recent studies has argued that countries undergoing democratic transitions are substantially more likely to become involved in wars than stable regimes of any type, whether democratic or autocratic ~Mansfield and Snyder, 1995a, 1995b, 2002!. Much research on the democratic peace has focused on relations within pairs ~or dyads! of fully democratic states. Whereas many studies have concluded that mature democracies rarely fight each other, the bulk of the evidence indicates that they are about as likely as autocracies to fight states that are not democratic. In contrast, research on democratization and war has centered mainly on the behavior of individual states, rather than pairs of states. Much of this work indicates that democratization has a strong monadic effect, increasing the likelihood of hostilities with regimes of any type. This difference between the dyadic democratic peace and the monadic impact of democratization raises a conceptual question about how to join the two sets of findings. It also highlights Authors note: For helpful comments, we thank Erik Gartzke, Bruce Russett, and the anonymous reviewers. For outstanding research assistance, we thank Patrick McDonald and Jon Pevehouse. We gratefully acknowledge the financial support provided by the Belfer Center at Harvard University, the Harry Frank Guggenheim Foundation, the United States Institute of Peace, the Mershon Center at Ohio State University, and the Hoover Institution at Stanford University, where Mansfield was a National Fellow during International Studies Association. Published by Blackwell Publishing, 350 Main Street, Malden, MA 02148, USA, and 108 Cowley Road, Oxford OX4 1JF, UK.

2 530 Incomplete Democratization and Military Disputes the importance of controlling for dyadic attributes when analyzing the monadic influence of democratization on military conflict. In addressing these issues, we extend one of the few studies of democratization and conflict based on a dyadic research design. In that study, Oneal and Russett ~1997! concluded that democratic transitions have little bearing on military disputes. Using exactly the same data that Oneal and Russett analyzed, but conducting a set of tests that corresponds more closely to the logic of existing theories linking democratization to hostilities, we arrive at a different conclusion. The likelihood of interstate violence rises markedly if either state in a dyad undergoes an incomplete democratic transition; that is, a shift from an autocratic to a partially democratic ~or anocratic! regime that stalls prior to the establishment of consolidated democratic institutions. Besides challenging Oneal and Russett s conclusions, our findings break new ground in three ways. First, based on a more refined research design than that found in earlier work on the topic, this article aims to identify the conditions under which democratization stimulates military disputes. To this end, we distinguish between incomplete democratic transitions and transitions that quickly culminate in a fully coherent democracy. We argue that the former changes are much more perilous than the latter. Second, the dyadic research design used here allows us to take into account interstate conditions emphasized in various theories of international conflict. Realists, for example, might suspect that fluctuations in the relative military strength or the alliance relations of a democratizing state encourage aggression for reasons that are incidental to regime type; but we show that the relationship between incomplete democratization and armed conflict holds even when controlling for these factors. Likewise, we show that factors stressed in liberal theories including trade relations between states, their foreign policy preferences, and their static regime characteristics at a given point in time do not account for the belligerence of democratizing countries. Yet while we address the influence of these dyadic variables, we also focus on the monadic effects of democratic transitions, since the basic logic of our argument about the aggressive foreign policies stemming from partial democratization is monadic in nature. Third, previous research drawing a connection between democratization and international conflict has focused primarily on explaining the onset of war, whereas much of the research on the peace between mature democracies has examined both wars and less violent military disputes. Our results break new ground in showing that incomplete democratic transitions heighten the prospects of various types of military disputes, including those short of war. Moreover, they correspond closely to the findings of recent research on the outbreak of war that addresses individual states rather than dyads ~Mansfield and Snyder, 2002!. Overall, then, our analysis underscores the conclusion of earlier studies that certain types of democratization are a potent source of military conflict, and it focuses attention on incomplete democratic transitions as the main danger. Why Democratization Causes International Conflict The peace among mature democracies rests on the presence of strong institutions that regulate mass political participation. These institutions guarantee that the officials making foreign policy will be accountable to the median voter, who bears the costs and risks of military conflict. The democratic peace depends on the establishment of stable institutions guaranteeing the rule of law, civil rights, a free and effective press, and representative government. Even those observers who stress the importance of liberal norms and liberal identity in creating the democratic peace usually acknowledge that these norms can function only in an effectively institutionalized democratic setting ~Owen, 1997:155!. Moreover, they frequently define democracy in institutional terms ~Russett, 1993!.

3 Edward Mansfield and Jack Snyder 531 When demands for broader political participation arise in a nondemocratic state, embryonic institutions needed for effective democracy may already exist. For example, democratic transitions in nineteenth-century Britain and contemporary South Africa were facilitated by the prior establishment of many of the necessary institutions for representative government among the dominant classes. In other cases such as the post-communist Czech Republic a highly literate, politically sophisticated, large, urban middle class may have the skills needed to create such institutions quickly. If so, full democracy may be consolidated relatively smoothly and the benefits of the democratic peace achieved rapidly. More commonly, however, demands for broadening political participation arise in authoritarian settings that lack the institutional prerequisites for effective democracy. The rule of law is poorly formed, state officials are corrupt, elections can be rigged, militaries or warlords may threaten to overturn electoral outcomes, and journalistic media are unprofessional and depend on the state or economic elites. As Huntington ~1968! has pointed out, when rising levels of mass political participation outstrip the ability of weak institutions to manage political activity, the result is not democracy but a free-for-all in which each social grouping may threaten to take unilateral action to protect its interests: workers may strike, ethnic groups may riot, and militaries may make coups. Governing in this kind of setting requires finding an ideological basis for popular political support in the absence of the legitimacy that mature democracies derive from institutionally established democratic procedures. In this sense, ideology must fill the gap between high levels of participation and weak political institutions. Only a well-institutionalized democracy can guarantee rule by the people, but a populist ideology can promise the plausible-sounding substitute of rule for the people. This ideology may appear in several forms, but most frequently it takes the form of some variety of nationalism ~Snyder, 2000!. Nationalism is a statebuilding ideology that holds out the promise of either gaining a state for a people that lacks one or strengthening the capacity of an existing state to better serve the distinctive aims of a people. As such, it can close the gap between popular demands and weak institutions. Nationalism, however, does not necessarily promise to do so by enacting a system of civil rights and strict accountability to the median voter. On the contrary, many forms of nationalism base their appeals on the political exclusion of ethnic or ideological enemies of the nation, and in this way set up the possibility of abridging rights of political participation. According to this ideology, governing for the people may require placing limits on government by the people. The precise content of nationalism depends on the nature of both the postauthoritarian political setting and the enemies to be excluded. For example, French revolutionary nationalism branded aristocratic opponents of the revolution as enemies in league with foreign foes; counter-revolutionary German nationalism appealed to the middle class against antinational revolutionary workers; and ethnic Serbian nationalism based its exclusions and enmities on ethnic grounds. Religious fundamentalism which Juergensmeyer ~1993! argues is a form of religious nationalism may serve similarly as a populist, but exclusionary, state-building strategy through its antisecularist ideology. Elsewhere, as in Argentina, military populism has played an analogous role ~Levy and Vakili, 1992!. Facing the institutional challenges of a post-authoritarian environment, elites of all sorts may perceive an incentive to turn to populist or nationalist ideologies. These ideologies may appeal to rising elites representing formerly excluded social groups, such as ethnic minorities, as well as to elites who are threatened by the collapse of the prior regime and are trying to generate a basis of mass political support. Nationalism offers the latter elites a way to evade strict democratic accountability, while giving the appearance of being responsive to popular

4 532 Incomplete Democratization and Military Disputes interests. Nationalism may also serve as a useful vehicle for old elites who have a parochial interest in military, imperial, or protectionist enterprises that lay at the heart of the former authoritarian state and its ruling coalition. This kind of nationalist politics in a weakly institutionalized post-authoritarian setting may lead to international military disputes through several related mechanisms. One is nationalist outbidding: both old and new elites may bid for popular favor by advancing bold proposals to deal forcefully with threats to the nation, claiming their domestic political opponents will not vigorously defend the national interest. Another mechanism is blowback from nationalist ideology: nationalists may find themselves trapped by rhetoric that emphasizes combating threats to the national interest because both the politicians and their supporters have internalized this worldview. A final mechanism is logrolling. Frequently, nationalist coalitions include military, imperial, or protectionist groups that are held together by logrolling. Such tactics increase the prospects of embroiling the state in foreign disputes ~Mansfield and Snyder, 1995a, 1995b!. In a mature democracy, accountability to the median voter and a free press limit the ability of politicians and interest groups to successfully use these ruling strategies. In the early stages of a transition from autocracy toward democracy, however, weak participatory institutions create both the motive and the opportunity for political strategies that heighten the prospect of interstate conflict. In this article, we distinguish between two phases in the process of democratization: the transition from autocracy toward a partially democratic regime, and the shift to a fully institutionalized democracy. We expect the likelihood of hostilities to be especially pronounced in the first phase, during which old elites threatened by the transition are often still powerful and the institutions needed to regulate mass political participation tend to be very weak. Like in prior research on the initial stages of democratic transitions, we include in this category cases in which elites conclude bargains involving limited political liberalization ~O Donnell and Schmitter, 1986! and those in which most elites consider voting to be only a temporary expedient ~Przeworski, 1991:52!. As we explain in more detail below, regime transitions are coded using the Polity III data compiled by Jaggers and Gurr ~1995!. Among the many states ~and pairs of states! beginning transitions toward democracy during the era since World War II, and that experienced military disputes soon thereafter, are Greece and Turkey, Ecuador and Peru, Argentina and several neighbors at different times, revolutionary Iran, and India and Pakistan. Many of these are cases of countries that experienced an incomplete transition; that is, one that stalled prior to the establishment of coherent democratic institutions. In such instances, the rhetoric of popular sovereignty is frequently grandiloquent, but the power of voters to control government policy is weak. When military, communist, colonial, dynastic, or other authoritarian regimes break down and mass politics enters its beginning stages, democratic procedures are likely to be intermittent, manipulated, and animated by nationalist or other populist ideologies that give rise to international frictions. Unless the state has the rare luck to enjoy fairly strong political institutions at the outset of the transition from autocracy, the turbulence of this first step on the road to democracy is hard to avoid ~Mansfield and Snyder, 2002!. The second distinct phase of democratization occurs when the regime moves to adopt a system of unfettered political competition and full governmental accountability to a broad electorate. On the one hand, this phase may create incentives for elites who fear the consequences of democratic consolidation to play the nationalist card in public debate or to gamble for resurrection in a foreign crisis. On the other hand, democratic institutions are often strong enough to frustrate such maneuvers at this more advanced stage of the transition. In general, we find that the latter tendency predominates; therefore the transition to complete democracy does not systematically increase the likelihood of military disputes.

5 Edward Mansfield and Jack Snyder 533 In a few noteworthy cases, however, a more dangerous dynamic develops. Based on the measures used here, the two most recent regimes to initiate a war in the wake of a transition to coherent democracy were Turkey in the 1974 Cyprus War and Pakistan in the 1999 Kargil War. 1 Both of these countries have been marked by a history of alternating between military regimes and multiparty electoral politics. Over time they have developed many of the outward trappings of democracy, yet the ever-present threat of military intervention prevents democracy from becoming consolidated. In this situation, military elites have an incentive to show that they rule on behalf of the popular will, whereas civilians have an incentive to show that they stand firm on behalf of national security concerns. As a result, populist nationalist politics has flourished in both states, leading each of them to become embroiled in military rivalries with neighboring countries. We expect that transitions to coherent democracy will be dangerous primarily for states with these characteristics. States undergoing incomplete democratic transitions are likely to become drawn into conflicts with regimes of all types: democracies and autocracies as well as other transitional states. The effect of incomplete democratization on conflict arises from the monadic properties of the transitional state, not the dyadic properties of country-pairs. This view is consistent with research on the democratic peace, which not only shows that mature democracies are more peaceful toward each other, but also documents a broad range of monadic findings about the distinctive foreign policies of mature democracies. Democracies choose their wars more wisely ~Reiter and Stam, 1998!, tend to win them and suffer fewer casualties ~Lake, 1992; Siverson, 1995; Bennett and Stam, 1998!, rarely fight preventive wars ~Schweller, 1992!, and are more adept at signaling the credibility of their commitments ~Fearon, 1994!. As a whole, it seems plausible that these monadic effects are caused by a mutually reinforcing set of institutional, informational, and normative characteristics distinctive to mature democracies, such as accountability to cost-conscious voters, greater transparency of facts and preferences in policy debates, and respect for the civil liberties that make democracy possible. In relations between mature democracies, these characteristics of each party interact in ways that make war very unlikely ~Russett and Oneal, 2001:47 79!. The dyadic properties of such relationships may include the effects of common democratic norms and identities on the legitimacy of conflict ~Owen, 1994!, as well as the greater efficiency of interdemocratic bargaining and dispute resolution ~Dixon, 1994; Schultz, 1999!. It is likely that these dyadic properties emerge in large part because democracies are already different in their strategic propensities at the monadic level. Consequently, we see no conceptual mismatch between our monadic argument and the dyadic and monadic democratic peace literatures. The Research Design Recently, a number of efforts have been made to test the preceding argument. Some studies have concluded that, throughout the nineteenth and twentieth centuries, states conducting democratic transitions have been more likely to become embroiled in hostilities than other states, including those conducting autocratic transitions ~Mansfield and Snyder, 1995a, 1995b, 1996, 1997, 2002!. Other studies, however, have challenged these conclusions ~Enterline, 1996; Oneal and Russett, 1997; Thompson and Tucker, 1997; Maoz, 1998; Ward and Gleditsch, 1998; Gleditsch and Ward, 2000; Russett and Oneal, 2001!. Virtually all of these analyses have been cast at the level of the individual state. The purpose of this article is to respond to the repeated calls for an assessment of 1 Note that the Kargil War is not included in our data set, since it lies outside the time period we analyze. However, see the discussion of this incident by Russett and Oneal ~2001:48!.

6 534 Incomplete Democratization and Military Disputes the influence of democratization on hostilities between pairs of countries ~e.g., Mansfield and Snyder, 1995a; Oneal and Russett, 1997; Maoz, 1998!. In one of the very few empirical studies of this topic pitched at the dyadic level of analysis, Oneal and Russett ~1997! examined what are commonly referred to as politically relevant dyads that is, pairs of states that either were geographically contiguous or included a major power during the period from 1950 to They focused on evaluating the influence of democracy and economic interdependence on militarized interstate disputes ~MIDs!, events where one state threatens to use, displays, or actually uses military force against another state ~Gochman and Maoz, 1984:587; Jones, Bremer, and Singer, 1996!. However, they also conducted a set of analyses indicating that democratic transitions have little effect on disputes. Here, we extend Oneal and Russett s study. We start with their data set, covering politically relevant dyads between 1950 and 1985, as well as their statistical model. But whereas they centered on explaining whether country-pairs are embroiled in a MID in a given year, regardless of whether the dispute began in that year, existing arguments about the effects of democratization pertain to the conditions precipitating the outbreak of hostilities ~e.g., Mansfield and Snyder, 1995a, 1995b!. As such, we place primary emphasis on the onset of MIDs, although we briefly examine whether democratic transitions influence the existence of MIDs as well. Furthermore, Oneal and Russett s analysis focused on two aspects of democratization, but neither of them is emphasized in our argument. First, they analyzed the effects on disputes of the total amount of movement in a democratic direction made by the two countries in each pair taken as a whole both within and across regime types. This factor, however, is not central to our theory, since we do not address the implications of increased political openness within a given regime type ~e.g., an already consolidated democracy that further improves its civil rights or an autocracy that becomes a little less autocratic without becoming an anocracy!. Second, they analyzed the most pronounced type of democratic transition, where an autocracy is replaced by a coherent democracy. But that focus is too narrow to adequately test our argument, which suggests that transitions from autocracy that fail to produce coherent democratic institutions are an especially potent impetus to conflict. Incomplete democratic transitions stimulate mass political mobilization and introduce new political groups onto the domestic stage without fully developed democratic institutions through which to channel their political participation. Imperfectly formed institutions, in turn, can facilitate elites ability to exploit their power in ways that promote a belligerent foreign policy. Hence, in the following analysis, we evaluate the effects of both transitions to a coherent democracy and transitions that become stalled prior to the formation of democratic institutions. A Statistical Model of Democratization and Military Disputes To begin, we estimate the following model: MID ij b 0 b 1 Complete Demtransition ij b 2 Incomplete Demtransition ij b 3 Complete Auttransition ij b 4 Incomplete Auttransition ij b 5 DEM L b 6 DEM H b 7 GROWTH L b 8 ALLIES ij b 9 CONTIG ij b 10 CAPRATIO ij b 11 DEPEND S b 12 DDEPEND H u ij 2 Note, however, that some empirical research on this topic has been cast at the regional level of analysis ~e.g., Enterline, 1998!. That research is closely linked to studies addressing the relationship between democratization and conflict at the dyadic level.

7 Edward Mansfield and Jack Snyder 535 The observed value of the dependent variable is dichotomous: it equals 1 if a military dispute breaks out between countries i and j in year t, and 0 otherwise. As mentioned earlier, we measure democratization using the Polity III data set ~ Jaggers and Gurr, 1995!, which includes annual information for each state on: ~1! the competitiveness of the process through which the state s chief executive is selected, ~2! the openness of this process, ~3! the extent to which institutional constraints exist on the chief executive s decision-making authority, ~4! the competitiveness of political participation within the country, and ~5! the degree to which binding rules govern political participation within it. Gurr, Jaggers, and Moore ~1989:36 39! combine these factors to create REG, a summary measure of regime type that takes on values ranging from 10 to 10. Jaggers and Gurr ~1995! define countries where REG. 6 as coherent democracies, those where REG, 6 as coherent autocracies, and all other countries as incoherent or anocracies. In addition to this measure of regime type, which we refer to below as the composite index, we focus on three of its components: the extent of the constraints placed on the chief executive, the openness of executive recruitment, and the competitiveness of political participation. We emphasize these factors and analyze them separately for a number of reasons. First, each factor is stressed in the argument presented above, but some of them have a disproportionately small influence on the value of the composite index ~Gleditsch and Ward, 1997!. Second, these factors are not closely related. On average, in our model, the bivariate correlation between the value of a given variable pertaining to regime change ~i.e., Complete Demtransition ij, Incomplete Demtransition ij, Complete Auttransition ij,orincomplete Auttransition ij! based on one of these three factors and the value of the same variable based on another factor is only about.22, indicating they are not tapping the same institutional characteristics. Third, prior research on the relationship between democratization and war has analyzed these factors separately ~Mansfield and Snyder, 1995a, 1995b, 2002; Thompson and Tucker, 1997; Ward and Gleditsch, 1998!. Finally, unlike these factors, the remaining variables that make up REG are coded in a way that makes distinguishing between democracies and autocracies quite difficult. For each measure of regime type the composite measure, the extent of the constraints placed on the chief executive, the openness of executive recruitment, and the competitiveness of political participation we code every state in the sample as democratic, autocratic, or anocratic in year t 1 and again in year t 6. 3 A state is considered to be democratizing if it experiences a change in regime type leading to a more democratic polity during this five-year interval, the length of which is chosen to maintain consistency with Oneal and Russett s study as well as with other research ~e.g., Mansfield and Snyder, 1995a, 2002; Thompson and Tucker, 1997!. We noted earlier that it is important to distinguish between the effects of democratic transitions that culminate in a coherent democ- 3 We use the guidelines laid out in our earlier work ~Mansfield and Snyder, 1995a! to code regime type using executive constraints, the openness of executive recruitment, and the competitiveness of political participation. The institutional constraints on the chief executive are measured using a seven-point scale in the Polity III data. We code regimes as autocratic if the chief executive has unlimited authority or if his authority is not unlimited but he faces less than limited constraints. We classify regimes as democratic if accountability groups have effective authority equal to or greater than the executive in most areas of activity ~Gurr, Jaggers, and Moore, 1989:16! or if a state is coded as lying between the just-mentioned regimes and those in which substantial constraints exist on executive authority. We classify all remaining regimes as anocratic. The openness of executive recruitment is measured using a four-point scale in the Polity III data set. We code closed regimes and those where executive recruitment is unregulated as autocratic; regimes where recruitment occurs via hereditary succession, executive or court selection, or electoral succession as anocratic; and open regimes as democratic. The competitiveness of political participation is measured using a five-point scale in the Polity III data. We code states characterized by suppressed competition as autocratic, those characterized by competitive competition as democratic, and all other regimes as anocratic.

8 536 Incomplete Democratization and Military Disputes racy and those that become stalled. To this end, we analyze two variables. Complete Demtransition ij equals 1 if state i or state j changes from either an autocracy or an anocracy to a democracy during the period from t 6tot 1, and 0 otherwise. Incomplete Demtransition ij equals 1 if i or j changes from an autocracy to an anocracy during this period, and 0 otherwise. Various studies have concluded that, in addition to democratic transitions, regime change in an autocratic direction may foster international conflict ~Mansfield and Snyder, 1995a; Enterline, 1996; Maoz, 1998; Ward and Gleditsch, 1998!. We therefore include Complete Auttransition ij and Incomplete Auttransition ij in the model. The former variable measures transitions yielding a coherent autocracy: it equals 1 if state i or state j changes from either a democracy or an anocracy to an autocracy during the period from t 6tot 1, and 0 otherwise. The latter variable measures incomplete autocratic transitions, taking on the value of 1 if either state shifts from a democracy to an anocracy during this period, and 0 otherwise. In the following analysis, four sets of estimates are generated. To derive the first set, the composite index is used to measure Complete Demtransition ij, Incomplete Demtransition ij, Complete Auttransition ij, and Incomplete Auttransition ij. To derive the next three sets, these variables are measured using executive constraints, the openness of executive recruitment, and the competitiveness of political participation, respectively. The final eight independent variables in the model and the data used to measure them are taken directly from Oneal and Russett s study. Besides our interest in replicating that study, it is important to include these variables because they are emphasized in various leading theories of international conflict and we need to ensure that they do not account for any observed relationship between democratization and military disputes. DEM L and DEM H measure the regime types of i and j. For each pair of countries, DEM L is the smaller value of REG ~i.e., the value for the less democratic state! and DEM H is the larger value ~i.e., the value for the more democratic state! in t. Oneal and Russett expect the coefficient of DEM L to be negative, arguing that theories about the democratic peace predict a reduction in the likelihood of political conflict as the least democratic state in a dyad grows more democratic ~see also Bueno de Mesquita and Lalman, 1992!. GROWTH L is a measure of economic growth, calculated as the percentage change in per capita gross domestic product ~GDP! over the three-year interval prior to t for the country in each dyad that experiences the smallest change. ALLIES ij equals 1 if i and j are political-military allies or if both are allied with the United States in t, and 0 otherwise. To distinguish between the likelihood of military disputes for dyads that are geographically contiguous and for those that include a major power the two types of country-pairs composing Oneal and Russett s sample CONTIG ij is included. It equals 1 if i and j are contiguous, and 0 if these countries are not contiguous and either one of them is a major power ~i.e., China, France, the Soviet Union, the United Kingdom, or the U.S.!. CAPRA- TIO ij measures the distribution of capabilities between i and j in t, where each state s capabilities are assessed by averaging its share of the international system s total population, urban population, military expenditures, military personnel, iron and steel production, and energy consumption. CAPRATIO ij is the ratio of the share of these capabilities controlled by the larger state to the share controlled by the smaller state. Recall that one purpose of Oneal and Russett s study was to analyze the links between interdependence and conflict. To this end, Oneal and Russett defined two variables. They computed the sum of exports and imports between i and j in year t 1. Then this sum was divided by i s GDP in t 1 and by j s GDP in t 1. The smaller ratio of bilateral trade to GDP in each dyad is

9 Edward Mansfield and Jack Snyder 537 DEPEND S. 4 Further, DDEPEND H is the change in DEPEND from t 4tot 1 for the state having the higher ratio of trade to GDP in t 1. 5 Finally, u ij is a stochastic error term. Since the observed value of the dependent variable is dichotomous, logistic regression is used to estimate our model. We follow Beck, Katz, and Tucker in basing tests of statistical significance on Huber standard errors that are clustered by dyad and including in the model a spline function of the number of years that have elapsed since a MID last broke out between i and j to account for any temporal dependence in the data ~Beck and Tucker, 1996; Beck and Katz, 1997!. 6 Note, however, that, to conserve space, the estimates of this function are not reported in the following tables. Statistical Results We start by considering the results based on the composite index, since it has been widely used in research on the democratic peace, including Oneal and Russett s study. As shown in the first column of Table 1, the estimate of Incomplete Demtransition ij is positive and statistically significant, indicating that incomplete democratic transitions are especially likely to precipitate interstate violence. 7 Moreover, the effects of such transitions are quantitatively large as well as statistically strong. The predicted probability of a military dispute is more than 50 4 A recent study has argued that it is preferable to include separate measures of bilateral trade flows and GDP in analyses of the relationship between commerce and conflict rather than combining these factors via the ratio of trade to GDP ~Mansfield and Pevehouse, 2000!. This same study found that the effects of trade flows were contingent on whether trade partners belong to the same preferential trading arrangement ~PTA!. More specifically, heightened trade is more likely to inhibit hostilities between PTA members than between other states. Nonetheless, given the purpose of this article, it is useful to avoid making any unnecessary change to Oneal and Russett s model. Moreover, the following results are quite similar to the results that are derived when we include separate variables for trade flows and GDP, and when we include PTA membership and the interaction between this factor and both trade flows and GDP. 5 All of the data used in our analysis except data on regime change and the similarity of foreign policy preferences are taken directly from Oneal and Russett ~1997!. They obtained data on MIDs from Jones, Bremer, and Singer ~1996! and data on GROWTH L from Summers and Heston ~1991!. They took data on ALLIES ij, CAPRATIO ij, and the major powers from the Correlates of War ~COW! Project ~Singer and Small, 1968, 1993, 1994; Oren 1990!. Data used to generate DEPEND S and DDEPEND H were obtained from the International Monetary Fund, Direction of Foreign Trade ~Washington, D.C.: International Monetary Fund, 1993! and from Summers and Heston ~1988, 1991!. Note that where GROWTH L or DDEPEND H cannot be measured over three-year intervals due to missing data, either two-year or one-year intervals are used instead. 6 In recent research, Russett and Oneal ~2001! have argued in favor of using a general estimating equation ~GEE! approach rather than Beck, Katz, and Tucker s technique for addressing autocorrelation in cross-section time-series data with a binary dependent variable. Following Russett and Oneal, we also estimated the model using a GEE specification and modeling the temporal dependence as a first-order autoregressive process. The results are very similar to those presented below. In addition to addressing temporal dependence in the data, it is important to ensure that our results are not undermined by multicollinearity. An analysis of bivariate correlations between the independent variables and an analysis of the condition number for our data ~Belsley, Kuh, and Welsch, 1980; Greene, 1993: ! indicated that the only cases in which multicollinearity might be a problem involve the spline function. However, further analysis indicated that the sign, size, and significance of the variables pertaining to regime type and regime change are quite robust with respect to whether or not the spline function is included in our model. Hence, there is no reason to suspect that our results are threatened by multicollinearity. 7 Throughout this article, one-tailed tests of statistical significance are conducted for the estimates of Complete Demtransition ij and Incomplete Demtransition ij, since our argument is that the sign of each of these coefficients should be positive. It should be noted, however, that of the 21 instances in which the estimate of either Complete Demtransition ij or Incomplete Demtransition ij is statistically significant at the.05 level or better in Tables 1, 2, 3, 4, or 5, there are only two cases in which an estimate would not be significant at the.05 level if a two-tailed test rather than a one-tailed test was conducted. Although our theory makes no prediction about the signs of Complete Auttransition ij and Incomplete Auttransition ij, we also conduct one-tailed tests of their parameter estimates. We do so since certain observers have argued that autocratization also promotes conflict ~Enterline, 1996; Maoz, 1998; Ward and Gleditsch, 1998! and to give the fairest treatment possible to the argument that regime change in general rather than democratization in particular stimulates hostilities. Since our theory does not lead to predictions about the remaining coefficients, two-tailed tests of their estimates are conducted.

10 538 Incomplete Democratization and Military Disputes Table 1. Effects of Democratization and Autocratization on Military Disputes, Based on the Composite Index of Regime Type, Variable Base Model DISPUTE Regime Change Preference Similarity from t 5tot Tau-Beta S Intercept 2.041** 0.543** 1.980** 2.023** 1.901** ~0.270! ~0.206! ~0.266! ~0.241! ~0.239! Complete Demtransition ij ~0.235! ~0.256! ~0.220! ~0.244! ~0.237! Incomplete Demtransition ij 0.449** 0.425** 0.408* 0.390* 0.466** ~0.177! ~0.178! ~0.192! ~0.189! ~0.173! Complete Auttransition ij * * ~0.161! ~0.127! ~0.157! ~0.163! ~0.166! Incomplete Auttransition ij ~0.267! ~0.277! ~0.259! ~0.324! ~0.299! DEM L 0.044** 0.044** 0.044** ~0.014! ~0.013! ~0.014! ~0.014! ~0.014! DEM H 0.034** 0.020* 0.034** 0.021* ~0.012! ~0.010! ~0.012! ~0.010! ~0.011! GROWTH L 0.036** 0.037** 0.034** 0.040** 0.042** ~0.013! ~0.012! ~0.013! ~0.015! ~0.014! ALLIES ij 0.338* 0.384** 0.378* ~0.158! ~0.140! ~0.160! ~0.188! ~0.178! CONTIG ij 1.138** 0.665** 1.162** 1.341** 1.362** ~0.190! ~0.162! ~0.185! ~0.193! ~0.205! CAPRATIO ij * * * * * ~0.0008! ~0.0007! ~0.0008! ~0.0006! ~0.0006! DEPEND S ~14.271! ~15.326! ~14.342! ~10.873! ~11.359! DDEPEND H ~4.413! ~3.915! ~4.907! ~5.341! ~4.884! UN ij 0.916** 0.660* ~0.241! ~0.271! Log Likelihood N Note: Entries are logistic regression estimates, with robust standard errors in parentheses. Each model is estimated after including a natural spline function with three knots. ** p.01; * p.05. One-tailed tests are conducted for the estimates of Complete Demtransition ij, Incomplete Demtransition ij, Complete Auttransition ij, and Incomplete Auttransition ij. Two-tailed tests are conducted for all of the remaining estimates. percent greater if a dyad includes a country shifting from autocracy to anocracy than if neither state is experiencing a regime change, holding constant the remaining variables in the model. In contrast, the estimates of Complete Demtransition ij, Complete Auttransition ij, and Incomplete Auttransition ij are relatively small and none of them is statistically significant. The results based on executive constraints also furnish evidence that incomplete democratic transitions stimulate belligerence, since the estimate of Incomplete Demtransition ij in the first column of Table 2 is positive and statistically significant. Like when the composite index is analyzed, the effects of incomplete democratic transitions are quite large when regime change is measured using the constraints placed on the chief executive. The predicted probability of a dispute is roughly 45 percent higher if a country-pair includes a state engaged in a regime change toward democracy that stalls prior to the establishment of coherent democratic institutions than if neither state makes a regime change. Equally, there continues to be no evidence that any other type of regime change has a significant influence on the outbreak of military disputes. Our findings derived using the openness of executive recruitment are somewhat different, although they too indicate the dangers of democratization. As shown in the first column of Table 3, transitions to a coherent democracy pro-

11 Edward Mansfield and Jack Snyder 539 Table 2. Effects of Democratization and Autocratization on Military Disputes, Based on Constraints on the Chief Executive, Variable Base Model DISPUTE Regime Change Preference Similarity from t 5tot Tau-Beta S Intercept 2.056** 0.516* 2.046** 2.073** 1.954** ~0.264! ~0.206! ~0.258! ~0.233! ~0.232! Complete Demtransition ij ~0.286! ~0.326! ~0.253! ~0.304! ~0.290! Incomplete Demtransition ij 0.384** 0.262* 0.318* 0.422** 0.472** ~0.135! ~0.127! ~0.148! ~0.148! ~0.136! Complete Auttransition ij ~0.159! ~0.135! ~0.160! ~0.168! ~0.163! Incomplete Auttransition ij ~0.246! ~0.237! ~0.259! ~0.273! ~0.274! DEM L 0.047** 0.048** 0.048** * ~0.014! ~0.013! ~0.014! ~0.014! ~0.014! DEM H 0.033** ** 0.021* ~0.012! ~0.010! ~0.012! ~0.010! ~0.011! GROWTH L 0.038** 0.040** 0.038** 0.041** 0.042** ~0.013! ~0.012! ~0.013! ~0.015! ~0.014! ALLIES ij 0.328* 0.376** 0.352* ~0.160! ~0.142! ~0.163! ~0.191! ~0.182! CONTIG ij 1.145** 0.658** 1.153** 1.347** 1.375** ~0.187! ~0.161! ~0.188! ~0.192! ~0.202! CAPRATIO ij * * * * * ~0.0008! ~0.0007! ~0.0009! ~0.0006! ~0.0007! DEPEND S ~13.797! ~14.892! ~13.988! ~10.763! ~11.220! DDEPEND H ~4.417! ~3.966! ~4.530! ~5.374! ~4.918! UN ij 0.859** 0.648* ~0.230! ~0.270! Log Likelihood N Note: Entries are logistic regression estimates, with robust standard errors in parentheses. Each model is estimated after including a natural spline function with three knots. ** p.01; * p.05. One-tailed tests are conducted for the estimates of Complete Demtransition ij, Incomplete Demtransition ij, Complete Auttransition ij, and Incomplete Auttransition ij. Two-tailed tests are conducted for all of the remaining estimates. mote disputes, since the estimate of Complete Demtransition ij is positive and statistically significant. Moreover, dyads in which a country is experiencing this type of transition are almost twice as likely to become embroiled in hostilities as dyads in which neither state is undergoing a regime change, holding constant the remaining variables in the model. Conversely, Incomplete Demtransition ij, Complete Auttransition ij, and Incomplete Auttransition ij have little bearing on the outbreak of conflict, based on the results in Table 3. Finally, regime change is weakly related to hostilities when the focus is on the competitiveness of political participation. While the estimates of Complete Demtransition ij, Incomplete Demtransition ij, and Complete Auttransition ij in the first column of Table 4 are positive and relatively large, neither they nor the estimate of Incomplete Auttransition ij is statistically significant. That transitions to a coherent democracy rather than incomplete transitions tend to promote military disputes when we focus on the openness of executive recruitment stems from the way this variable is scaled in the Polity data ~Gurr, Jaggers, and Moore, 1989; Jaggers and Gurr, 1995!. Regimes receiving the highest score for the openness of executive recruitment include not only multiparty democracies but also some authoritarian systems, such as single-party regimes in which all the politically active population has an opportunity, in principle, to

12 540 Incomplete Democratization and Military Disputes Table 3. Effects of Democratization and Autocratization on Military Disputes, Based on the Openness of Executive Recruitment, Variable Base Model DISPUTE Regime Change Preference Similarity from t 5tot Tau-Beta S Intercept 2.057** 0.569** 2.043** 2.077** 1.956** ~0.267! ~0.206! ~0.266! ~0.238! ~0.235! Complete Demtransition ij 0.660** 0.753** 0.506** 0.691** 0.750** ~0.188! ~0.168! ~0.199! ~0.196! ~0.182! Incomplete Demtransition ij ~1.019! ~1.161! ~0.527! ~0.961! ~0.984! Complete Auttransition ij ~0.199! ~0.177! ~0.199! ~0.200! ~0.200! Incomplete Auttransition ij a ~0.946! ~1.008! ~0.913! ~0.954! DEM L 0.048** 0.051** 0.048** * ~0.014! ~0.013! ~0.014! ~0.014! ~0.014! DEM H 0.033** ** ~0.012! ~0.010! ~0.012! ~0.011! ~0.011! GROWTH L 0.031** 0.031** 0.034** 0.034* 0.035* ~0.013! ~0.012! ~0.013! ~0.015! ~0.014! ALLIES ij 0.352* 0.413** 0.373* ~0.161! ~0.143! ~0.165! ~0.192! ~0.183! CONTIG ij 1.135** 0.650** 1.148** 1.338** 1.371** ~0.189! ~0.162! ~0.189! ~0.195! ~0.207! CAPRATIO ij * * * * * ~0.0008! ~0.0007! ~0.0008! ~0.0006! ~0.0006! DEPEND S ~13.639! ~14.598! ~13.969! ~10.561! ~11.033! DDEPEND H ~4.385! ~3.915! ~4.482! ~5.340! ~4.874! UN ij 0.878** 0.667* ~0.235! ~0.275! Log Likelihood N a In estimating this model, Incomplete Auttransition ij was dropped due to collinearity. Note: Entries are logistic regression estimates, with robust standard errors in parentheses. Each model is estimated after including a natural spline function with three knots. ** p.01; * p.05. One-tailed tests are conducted for the estimates of Complete Demtransition ij, Incomplete Demtransition ij, Complete Auttransition ij, and Incomplete Auttransition ij. Two-tailed tests are conducted for all of the remaining estimates. attain the position through a regularized process ~Gurr, Jaggers, and Moore, 1989:11!. This coding rule has the effect of including as maximally open a number of states that are anocracies based on most of the other institutional dimensions analyzed here. As such, our findings based on the openness of executive recruitment are quite consistent with our other results, since there are various cases in which states making a complete democratic transition based on this measure make an incomplete democratic transition based on the remaining measures. Taken as a whole, these findings provide considerable evidence that partial democratization fosters antagonism. 8 Further, contrary to the conclusions of some earlier studies ~e.g., Thompson and Tucker, 1997; Maoz, 1998!, they do not 8 Our results show that dyads composed of a state undergoing an incomplete democratic transition are more likely to experience a military dispute than other dyads, but they do not establish that the democratizing state tends to be the more belligerent party in the dyad or the initiator of conflict. Data on the initiators of military disputes are of dubious quality, limiting our ability to address this issue. However, data on the initiators of wars are of better quality, and in a recent study we find that democratizing states are more likely to initiate wars than other countries and that they are not especially likely to be the targets of attack ~Mansfield and Snyder, 2002!.

13 Edward Mansfield and Jack Snyder 541 Table 4. Effects of Democratization and Autocratization on Military Disputes, Based on the Competitiveness of Political Participation, Variable Base Model DISPUTE Regime Change Preference Similarity from t 5tot Tau-Beta S Intercept 1.990** 0.505* 1.997** 1.994** 1.860** ~0.266! ~0.204! ~0.265! ~0.238! ~0.235! Complete Demtransition ij ~0.293! ~0.319! ~0.309! ~0.321! ~0.312! Incomplete Demtransition ij * 0.368* ~0.189! ~0.190! ~0.178! ~0.184! ~0.168! Complete Auttransition ij * * 0.330* ~0.157! ~0.130! ~0.151! ~0.163! ~0.163! Incomplete Auttransition ij ~0.257! ~0.278! ~0.223! ~0.307! ~0.299! DEM L 0.043** 0.044** 0.045** ~0.014! ~0.013! ~0.014! ~0.014! ~0.014! DEM H 0.031* ** ~0.012! ~0.010! ~0.012! ~0.010! ~0.011! GROWTH L 0.035** 0.038** 0.038** 0.036* 0.039** ~0.013! ~0.012! ~0.013! ~0.015! ~0.014! ALLIES ij 0.381* 0.418** 0.384* ~0.163! ~0.145! ~0.165! ~0.194! ~0.185! CONTIG ij 1.137** 0.663** 1.134** 1.339** 1.366** ~0.194! ~0.165! ~0.193! ~0.198! ~0.210! CAPRATIO ij * * * ~0.0009! ~0.0007! ~0.0008! ~0.0007! ~0.0007! DEPEND S ~14.178! ~15.161! ~14.243! ~10.645! ~11.191! DDEPEND H ~4.417! ~3.970! ~4.515! ~5.313! ~4.873! UN ij 0.904** 0.679* ~0.235! ~0.271! Log Likelihood N Note: Entries are logistic regression estimates, with robust standard errors in parentheses. Each model is estimated after including a natural spline function with three knots. ** p.01; * p.05. One-tailed tests are conducted for the estimates of Complete Demtransition ij, Incomplete Demtransition ij, Complete Auttransition ij, and Incomplete Auttransition ij. Two-tailed tests are conducted for all of the remaining estimates. reflect any general tendency for regime change to stimulate belligerence, since there is no indication that transitions in an autocratic direction influence the outbreak of military disputes. Moreover, additional tests indicate that, based on the openness of executive recruitment, the difference between the estimates of Complete Demtransition ij and Complete Auttransition ij is statistically significant. So too is the difference between the estimates of Incomplete Demtransition ij and Incomplete Auttransition ij based on the composite index, the constraints on the chief executive, and the competitiveness of political participation, respectively. That autocratization has a much less pronounced effect on conflict than democratization is entirely consistent with our argument: when popular participation in politics is curtailed, elites gain less from using nationalist rhetoric to maintain their rule and are less likely to become trapped in nationalist bidding wars. 9 9 Our statistical analysis does not directly measure nationalism or test its effects, and the dearth of systematic data on nationalism renders it impossible to do so here. Elsewhere, however, we have presented narrative case studies showing that nationalism is a key reason why democratization tends to precipitate conflict ~Mansfield and Snyder, 1995a, 2002; Snyder, 2000; see also Adamson, 2001!.