Abstract
Recent scholarship suggests that democracies tend to fight shorter conflicts that can be easily won. This is most likely due to the accountability incentives that constrain democratic leaders. Fearing removal from office, democratic leaders will try to choose short conflicts against weaker opponents. The authors question this argument by presenting an alternative explanation for the connection between democracy and shorter disputes and victories. Building on prior works that have identified a territorial peace, this article argues that democracies often have few territorial issues over which to contend. In fact, rarely do democracies have territorial disputes with their neighbors. Thus, democracies have less difficult issues to resolve, and this makes conflict escalation less likely against neighbors. Without neighbors ready to attack the homeland, states at territorial peace can more easily choose favorable conflicts to escalate. This logic applies to all states at territorial peace, of which democratic states are just a subset. Analyses of directed-dispute dyads between 1816 and 2001 provide confirmation for our argument. Regime type does not predict conflict selection or victory once controls are added for issue salience.
Most theoretical challenges to the democratic peace focus on the core empirical regularity that democracies do not fight each other. Ignored in these challenges are the many corollary findings and arguments associated with democracies in conflict. For example, democracies are thought to select well their potential rivals (Bueno de Mesquita and Siverson 1995; Bueno de Mesquita et al. 1999, 2003), which enables democratic leaders to emerge victorious in conflict (Lake 1992; Reiter and Stam 1998, 2003) and to fight shorter disputes and wars (Bennett and Stam 1998). The unified models that incorporate these empirical patterns remain derivative of regime-based explanations (Bueno de Mesquita et al. 2003; Russett and Oneal 2001). We change this by extending a theory of territorial peace to the democratic peace findings associated with conflict selection.
We argue that states must often settle territorial issues with their neighbors in order to democratize. Political power centralizes when the state is under constant territorial threat, and only by removing these threats are political systems able to liberalize. The political systems that develop from territorial peace have also, as a consequence of their development path, selected out of their sample of disputes the most dangerous, difficult-to-resolve issues facing the state. This implies that democratic leaders do not necessarily fare better than other leaders when choosing potential targets. Rather, democratic leaders have the luxury of being insulated from direct attacks by neighbors which enhances their ability to choose whether and when to escalate conflict against the (often noncontiguous) targets of their choice.
Our key expectation is that the issues confronting democratic leaders are substantively different from the issues that face the leaders of other regime types. Put simply, democratic leaders seldom have disagreements over homeland territories, which are disputes that are difficult to win, difficult to resolve, and last longer than disputes of other types. This suggests that the correlation of democracy with short, victorious conflicts is the result of state development paths and the removal of territorial issues; once these are accounted for, regime type has little independent effect on the conflict-selection process.
By extending the territorial peace argument to explain additional empirical regularities associated with regime type, we present one of the first theoretical challenges to the larger body of democratic peace scholarship that has developed since the core, peaceful dyads finding. More than this, however, we believe our approach presents an important methodological advance as well since we model spatial and temporal dependencies in the extant conflict data that have often been ignored. International conflict is not randomly distributed across states or across time (see, e.g., Goertz and Diehl 1992). We argue that state development paths are an important predictor of when and where conflicts occur, and the evolution of states greatly affects the conclusions of almost any study of the correlates of war.
We begin our argument in the next section by briefly outlining the studies that link conflict selection to the incentives facing democratic leaders. We then contrast these selection effects arguments with the expectations of our model, which is based on state development paths. Our research design outlines a simple model of conflict selection, using mostly interaction terms to specify key differences between our model and regime-based selection. We then present the results from the tests of our argument and conclude with a discussion of the implications of our study.
Democracy and Conflict Selection Patterns
The propensity for democracies to avoid war with one another has been one of the most robust empirical findings in the entire discipline of political science. Indeed, the connection between democracies and the avoidance of war is so strong that it might be the closest thing we have to a law in international relations (Levy 1988). The democratic peace research program—the catalog of statistical tests that provide wide empirical support for the democracy-peace connection—has given way to additional empirical regularities regarding the behavior of democratic leaders within the international system. Here, we focus on the growing literature that contends democratic leaders carefully select their opponents so that the fights will be shorter and their chances of victory will be maximized.
Though democracies have been demonstrated to be generally more pacific than their nondemocratic counterparts (Benoit 1996; MacMillan 2003; Rousseau et al. 1996; Rummel 1983), once in war, democracies generally emerge victorious more often than nondemocratic states (Lake 1992; Reiter and Stam 1998, 2003). The theories that explain this empirical regularity typically fall into two competing explanations: war fighting and selection effects (Reiter and Stam 1998, 2003).
As Reiter and Stam (2002) describe, the war-fighting argument rests on the principle that military soldiers in democracies fight better than their counterparts in authoritarian regimes. The emphasis on individualism and popular consent in democracies develops soldiers that are more loyal and trusting of their government, better equipped to demonstrate initiative in battle, and have a higher morale than soldiers deployed by other regimes. A competitive civilian authority also reinforces merit-based advancement within the military itself, and this translates into better war-fighting performance overall.
Perhaps a better argument for conflict differences across regime types turns on the likelihood of selection effects in the escalation choices of leaders. Democracies are only likely to enter wars when the probability of winning the war is high; after all, democratic publics turn their leaders out of office following foreign policy disasters. Both Bueno de Mesquita and Siverson (1995) and Bueno de Mesquita et al. (1999, 2003) confirm that democratic leaders suffer more severe punishments for losing wars. Similarly, Reiter and Stam (2002) find that democracies win over 90 percent of the wars they initiate and over 60 percent of the wars in which they are targeted. While the difference across participant status slightly contradicts the claim of democratic superiority in war fighting, democracies do win their conflicts, and this does not seem to be a chance occurrence.
Most criticism of the selection argument has focused on the particular cases that do not conform to the expectations of the theory, like the poor decision-making process that went into launching the extended US war against Iraq in 2003 (Cramer 2007). Downes (2009) offers a more systematic critique of Reiter and Stam’s evidence, suggesting their analyses are not robust to slight changes in the treatment of cases and method. Further, Downes’ in-depth study of Lyndon Johnson’s decision to escalate the Vietnam War demonstrates that domestic politics may have variegated effects on wars of choice. For Johnson, escalation of a war with a slim chance of victory was preferable to losing the ability to push forward a reform agenda at home. Thus, at least in this case, a democratic leader understood failure was likely but still escalated the conflict; moreover, the democratic leader escalated because his domestic political situation made losing the war preferable to withdrawal.
Two notable criticisms of the accountability mechanism provide additional evidence that electoral punishment may not constrain democratic leaders. Desch (2008) argues that, though painful to the leader’s pride, electoral punishment for lost wars in democracies hardly outweighs the consequences of leader replacements in nondemocracies, events that often include death or exile. This argument is supported by Chiozza and Goemans’s (2004, Table 2) findings associating lost wars and crises with leader removal for autocracies and mixed regimes, but not for parliamentary or presidential democracies.
A direct extension of the selection effects argument is the expectation that democracies will try to fight wars that are expected to be short. Quick victories are likely to be correlated with electoral success while longer conflicts drain society of personnel, materiel, and patience with the leadership. Bennett and Stam (1996, and later 1998) provided two of the first studies to demonstrate this relationship and found that the war-fighting advantage for democratic initiators is mostly short term. Within the first eighteen months of combat operations, democratic initiators enjoy a significant advantage over their autocratic counterparts. The predicted probability of war victory for democratic initiators is 49 percent if the war lasts for one year and 6 percent if the war lasts for five years or more (Bennett and Stam 1998).
Since initiator advantages dwindle as wars persist, recent scholarship has stressed that the decision to initiate conflict should be modeled as a function of both expected duration and expected likelihood of victory. Slantchev (2004), for example, argues that conflicts start when differences persist in expectations over outcomes, but battlefield results provide information updates that force eventual convergence on a likely winner. The information gained from battle proves much more valuable to leaders than ante bellum judgments of capabilities. This argument would suggest that the process of choosing easy opponents may be difficult for most leaders ex ante.
Slantchev (2004) also finds that conflict initiators are more likely to lose when the conflict issue is more salient to them. This follows the selection effect logic since leaders would tend to only initiate conflicts over low salience issues when the chances of victory are overwhelming. Important for our argument is the fact that salience is operationalized, in part, by disaggregating the issue descriptions provided by Holsti (1991), which include territory as a key component. Thus, Slantchev (2004) affirms that the success rate for initiators drops when the conflict involves territorial issues. Also important for our argument, the variable describing regime type is statistically insignificant when estimated jointly with issue salience. 1
We have identified two ancillary empirical regularities associated with the broader democratic peace: democracies tend to fight shorter wars and democracies tend to win these conflicts. Each of these findings has predominantly been explained as outgrowths of the incentives facing democratic leaders. Fearing electoral challenges following failures, democratic leaders select well their conflicts and only escalate issues that are easily won. In the next section, we argue that these empirical regularities can be better understood as products of state evolution. States tend to democratize and survive qua democracies in relatively safe geographic environments, and, because of this tendency of democracies to be at territorial peace, the issues that face democratic leaders seldom provoke long wars or conflicts that are difficult to win. Peace with neighbors also allows democratic leaders (or any other leader in similar circumstances) to effectively choose when to engage and when to escalate against other states in the system.
Territorial Peace and State Development
Over a century ago, Otto Hintze suggested that “self-government” and democratic authority structures had their roots in insulation from conflict ([1906] 1975). Island states like Great Britain and the United States, as well as nations like Switzerland that are protected from invasion by rugged or mountainous terrain, developed decentralized governmental and military structures quite different from continental states like Germany and France. For continental states, the persistent threat of invasion forced the construction of large land armies to protect against external threat, which in turn produced more autocratic authority structures. The land armies necessary to defend and potentially to occupy territory demanded higher levels of centralization, extraction, and coercion, which stood in stark contrast to the relatively decentralized militia structures found in insular states. War making and the exigencies of survival thus conditioned the internal organization of the state.
The idea of a relationship between conflict behavior and the structure of the state has been well studied in the hundred years that followed Hintze. A number of scholars have told similar theoretical stories (Thompson 1996; Desch 1996; Mann 1988; Rasler and Thompson 1989; Tilly 1990), but empirical support for the proposition has been mixed at best. Direct tests of the relationship between participation in conflict and regime type produce a variety of inconclusive or even conflicting results (James, Solberg, and Wolfson 1999; Mitchell, Gates, and Hegre 1999; Oneal and Russett 2000; Przeworski 1988; Rasler 1986; Ray 1995; Reiter 2001; Reuveny and Li 2003; Thompson 1996). Where these studies generally focus on direct participation in conflict, however, they miss the nuance of the early argument; a persistent sense of threat to a state’s territorial integrity, a basic exposure to the hazard of territorial conflict, should drive the construction of the military and political organizations that lead to centralized or autocratic structures of authority. If these developments precede participation in conflict, then most tests of the peace-to-democracy thesis, which usually assesses the effects of conflict participation only, have been biased against finding an effect for “peace” on democracy. Similarly, variations in the level of threat exposure across the globe can replace Hintze’s deterministic reliance on geography; peaceful territorial regions can mimic real islands as both provide borders isolated from threat.
We focus on territorial issues because their salience as an issue type has been well demonstrated by numerous studies (see, e.g., Gibler 1996; Hensel 1996; Huth 1998; Senese and Vasquez 2008; Vasquez 1993, 1995, 2001, 2009; Vasquez and Henehan 2001). These issues are closely associated with the realpolitik bargaining practices that provoke disputes (Vasquez 1993), and the ensuing disputes tend to result in higher fatalities, indicative of the salience of that issue to public and elites alike (Senese 1996). Difficult to resolve, crises over territory are likely to recur (Hensel 1998) and are significantly more likely to result in war (Hensel 1996; Senese and Vasquez 2003, 2005).
Ongoing territorial issues are especially likely to alter the development paths of states. We argue that threats to the homeland by revisionist neighbors, because of their high salience and the nature of the threat, are most likely to force the creation of large standing armies to defend targeted lands from neighbors. The economic nature of large land armies reinforces the trend toward greater concentration of power in the hands of the elite within society. Standing armies require high levels of taxation as well as a broad centralization of authority—to acquire, arm, equip, feed, and otherwise maintain the troops. High taxation and centralization both contribute to a widened gap between the fortunes of the elites and the poor as compared to the status quo. High levels of military spending and frequent conflict also depress domestic consumption and economic growth. This makes the costs of adopting democracy and conceding to the poor’s redistributive demands far higher than the costs of using the army to pursue a strategy of exclusion and suppress competing social groups.
The process of political power centralization takes hold as territorial threat affects the relative bargaining power of the leader and opposition. Leaders are in a privileged domestic political position when the state is confronted with external threats. Many studies have documented rally effects among threatened publics (Lai and Reiter 2005; Parker 1995; Mueller 1994; Zaller 1993), but, perhaps more importantly, most political oppositions are loathe to challenge the leader during times of threat and uncertainty. Leaders are then able to take advantage of pliant oppositions and publics and can move to centralize their authority by removing potential veto players from the regime (Gibler 2010). The presence of a large, standing army only reinforces the political power of the leader, as the costs of repressing dissent decrease markedly when the ruling regime has access to such forces (Boix 2003).
When states are able to remove dangerous issues of territory from their agenda—when borders with neighbors have been settled—there is little need for the existence of large, standing armies. Over time, reliance on the standing army for repression becomes more difficult as the public and domestic opposition forces no longer equate state patriotism with leader support. Without an excuse to centralize power, the leader must bargain politically with opposition forces, and the factors that encourage democratization can take hold. 2
We do not assume that democratization will always follow territorial peace. Instead, we argue that the political centralization necessary to fend off direct homeland threats, including the presence of a large, standing army, severely hampers the influence of those conditions that favor democratization in the state. In other words, if we assume wealth and a middle class are correlated with democracy, as is often thought (Bernhard, Nordstrom, and Reenock 2001; Boix and Stokes 2003; Burkhart and Lewis-Beck 1994; Lipset 1959, 1994; Moore 1966; Przeworski et al. 1996), then the effects of wealth and a middle class will only take hold in states that have been absent from conflict for some time. Meanwhile, states at territorial peace that lack the causes of democratization may remain autocratic for some time.
Gibler (2007, 2012) provides the strongest evidence to date in support of the argument connecting external threats to conflict and regime type (but see also Thompson 1996; Rasler and Thompson 1989, 2004). By adding geographic controls for international borders that are more likely to be stable, the pacific influence of joint democracy disappears in a model predicting dyadic militarized disputes, 1946 to 1999. International borders that follow strong salients like mountainous terrain, colonial borders, and borders that clearly divide ethnic groups, are less likely to experience territorial threats and disputes, are more likely to lead to joint democracy in the dyad, and are more likely to be peaceful generally. 3
Implications for Conflict Selection
Our model of state development has clear expectations for the variation in the occurrence and types of conflict involving states at territorial peace. We would of course expect that states at territorial peace would not be likely to fight their neighbors, especially over territorial issues. After all, resolving territorial issues was necessary to promote decentralized political power within the state. Absent threats from neighbors, states at territorial peace are more likely to be at peace generally, and any disputes that do occur are likely to be against states that are not contiguous to homeland territories.
Territorial peace also brings a change in the nature of the issues that confront the state. No longer concerned with survival in the international system, most disputes become matters of choice for the regime. Absent are the direct threats to homeland territory, which tend to be highly contentious and difficult to resolve. Instead, the territorial issues that concern the regime most often involve colonial territories and other imperialist claims, which are less salient domestically. Even less consequential are the questions of policy and regime status involving far-flung states. 4
Our argument provides expectations for the conflict patterns of all states at territorial peace. Since territorial peace is connected to the development of democracy in the state, we further expect that democracies are more likely than other regime types to be at peace with their neighbors. This of course has implications for the many arguments connecting leader incentives to certain types of dispute behavior. Indeed, we would expect that democratic leaders will behave no differently than leaders of other regimes when confronting similar issues. Territorial peace just increases the likelihood that the most dangerous issues facing democracies have already been resolved. 5
Conflict initiation
We test our theory against a baseline model of expectations derived from the democratic peace literature. According to most theories focusing on regime type differences, conflict selection by democracies should turn on the strategic incentives facing democratic leaders. Democracies should rarely initiate disputes, but the disputes that democracies do initiate will be against much weaker opponents. But if our argument is correct, then state development paths should affect the distribution of issues facing democracies in identifiable ways. Our theoretical assumption is that neighboring conflicts lead to centralization and hamper the processes that foster democratization. States need to remove dangerous territorial issues from their agenda in order to decentralize political power and become democratic; the removal of these issues also fosters territorial peace with the state’s neighbors. Democracies would therefore be less likely than other types of states to have disputes against their neighbors.
6
Notice, however, that no expectations are made that regime type controls conflict behavior with states that are noncontiguous. We argue that democracies are as likely as any other type of regime to have disagreements and eventual disputes with other states in the system. Indeed, we expect that
Hypothesis 1: Democracies are unlikely to initiate disputes against their contiguous neighbors.
Hypothesis 2: Democracies are as likely as other regime types to initiate disputes against noncontiguous states in the international system.
Victory in conflict
According to the selection-effects argument, democratic leaders have greater incentives to maximize their chances of winning a dispute. This constrains democracies to only fight against much weaker than average opponents and produces the empirical regularity that democracies are more likely than other regime types to win the conflicts they initiate. Our argument also implies a selection effect, but, instead of capability distributions affecting victory, we believe states at territorial peace have a greater ability to choose their potential targets. We expect that disputes against contiguous neighbors are less likely to be conflicts of choice. If a neighbor begins arming, forms an aggressive alliance, or otherwise seeks revision of an issue, targeted leaders would have a difficult time ignoring such a threat. However, states divorced from threats by neighbors have greater freedom to choose those conflicts they wish to escalate. All leaders would probably wish to win the disputes they initiate; the difference is that democracies are more insulated from threats and have a greater ability to choose their potential targets in order to do so.
Hypothesis 3: Leaders are less likely to win disputes initiated against contiguous targets.
Hypothesis 4: Democratic leaders are no more likely than the leaders of other regime types to win disputes initiated against contiguous states.
Conflict duration
The final set of expectations we test concerns the argument that democratic conflict selection greatly influences the length of conflict. According to much of the extant literature, democratic leaders are turned out of office if conflicts linger and casualties mount. Therefore, the conflicts democracies initiate should be against much weaker opponents, who will have difficulty mounting a prolonged defense. Short, quick wars are favored by democratic leaders, while nondemocratic leaders face no similar pressures. We once again agree that there is a selection effect in the data, but the sample is nonrandom by issue type, not by variation in the incentives facing leaders. We expect shorter disputes over conflicts of choice, which tend to be against noncontiguous states, concerning issues other than territory. If our argument is correct, democracies should be no different from other regime types when forced to fight against their neighbors, especially over homeland territories.
Hypothesis 5: Disputes between contiguous states and disputes over territorial issues should last longer than other types of disputes.
Hypothesis 6: Democratic leaders are no more likely than the leaders of other regime types to fight shorter disputes initiated against contiguous states.
Modeling Choices and Variable Definition
We test our argument using three separate dependent variables that are all based on the Correlates of War Militarized Interstate Dispute (MID) data set (Jones, Bremer, and Singer 1996). We use Maoz’s (2005) dyadic-dispute data from 1816 to 2001 as our sample, which produces over 1.3 million directed dyad years. This large sample is restricted to disputes only when examining conflict victory and duration.
While much of the conflict selection literature has been restricted to wars, we focus on MIDs for the simple reason that leaders do not know, ex ante, which disputes other leaders may escalate to the point of war. Our analyses therefore examine the information available to leaders at the start of a dispute to determine whether democratic leaders are indeed especially adept at selecting conflicts that will be easily or quickly won. If the selection effects argument is correct, then we would be analyzing a heavily biased sample if we included only wars as observations. Indeed, analyzing only wars throws out the bulk of the data for which the theory of democratic selection is applicable. 7
This point is also important when considering a key difference between the MID and Interstate War data sets from the Correlates of War Project regarding how conflict initiations are coded. An initiator is defined with the start of actual combat in the war data set, but this is not the case for the MID data since a show of force or even a threat can begin a dispute (Sarkees and Wayman 2010, 58). Thus, for example, the Assam War of 1962 (War #160) was initiated and won by China, but the same conflict in the MID data (MID #199) was coded as initiated by India with its incursions deep into Chinese territory just a week prior to the coded start of the war (Sarkees and Wayman 2010, 154). Reiter and Stam (2002, 56) code the war as a successful initiation by a nondemocracy (China) against a democracy (India). However, if we use the MID data, the same data point is coded as an unsuccessful initiation by a democracy. Perhaps, one could argue that the war actually began with China’s escalation of the conflict in 1962, but this argument would miss the fact that India gravely miscalculated both the likelihood of war and the likelihood of success against a powerful opponent. Of course, neither miscalculation is consistent with most theories of democratic selection. India lost a war that developed from an issue it pressed.
The second reason why the dispute data is especially important for our tests is that there are likely to be few observed differences between our territorial peace expectations and the theories of democratic war selection when the tested sample includes wars only. Democracies are believed to choose their conflicts wisely, avoiding wars that are likely to cause heavy casualties, last a long time, or be difficult to win. In contrast, we believe all leaders would like to choose such easy wins but only states at territorial peace are actually able to select carefully their conflicts, and, since democracies are more likely than other regime types to be at territorial peace (Gibler 2012), the correlation between democracy and conflict selection makes sense. Because of this observational equivalence, we test the ability of all states to select their conflicts at earlier stages. If our argument that democracies are selected out of the difficult territorial issues that cause so many conflicts, then this would be best observed in the dispute data.
Some observations from our sample are clearly consistent with this argument. For instance, the sample of all MIDs includes twenty-six cases in which a democracy (Polity 2 score of 6 or greater) initiated a dispute that reached a level of fatalities equal to war (6 on the fatality scale for the dispute). Of those cases, only six were wars fought by neighbors over territory. The large majority of cases—77 percent of the total—involved other types of issues, which, according to our theory, are more prone to conflict selection by leaders. Our argument would also suggest that the difference in conflict type explains the propensity for democratic victories in war, and there is again initial evidence to support this conclusion. Democratic initiators in our sample did indeed win eighteen of twenty-six wars (or almost 70 percent), but that rate is deceptive. Democracies won only three of the six war initiations that were fought over territorial issues with a neighboring state; this is the same victory rate as flipping a coin. In other words, once the conflict issue is taken into consideration, the democratic advantage disappears.
Finally, we also wish to assure that our results are not determined entirely by sample selection, so we also reestimated each analysis using a selected sample of disputes that involved at least one fatality. We present these analyses in the supplemental appendix that supports this article. 8 In all cases, the additional analyses are consistent with our argument that issue selection outweighs regime type as a factor explaining conflict initiation, conflict victory, and conflict duration.
Dependent Variables
Our first dependent variable is a dummy variable that is positive for the presence of an MID initiation by State A of a directed dyad. We define initiation in the dyad as any case of new dispute originating in a given directed dyad year in which State A is listed as revisionist in the dyad. Of course, this definition of initiation does not necessarily imply that State A fired the first shot of the conflict; the first mover is impossible to identify in most cases. However, we believe that the revisionist coding is most consistent with the logic of the selection arguments. Those leaders who wish to revise the status quo are most likely to choose the conflicts with which they become involved.
Our second dependent variable uses the outcome variable from the MID data set to determine all cases of MID victory by state A (a value of 1 on the outcome variable). All other types of outcomes are coded as 0. Positive coefficients suggest relationships that increase the likelihood of dispute victories by state A over state B. However, since the 0 category includes several possible values (yield, compromise, stalemate, release, etc.), negative coefficients only identify relationships that make victory for state A less likely; negative coefficients should not be interpreted as victories for state B in the dyad.
Our third dependent variable is the length of an MID, measured in days. We code dispute length using the start and end dates of the initiator, which we identify as above. In approximately forty cases, the start day or end day was missing; we substituted values for these days that would make each missing day dispute as short as possible. Thus, if the dispute began on the nineteenth of the month and the end date was missing, we coded the dispute as ending on the twentieth. Nevertheless, treating these cases as missing values and omitting the disputes from the analyses does not change the substance of any of the results we report.
We present analyses using only directed dispute years for the second (dispute victory) and third (dispute length) dependent variables, which greatly reduces our sample size for these three sets of analyses. While there are good theoretical reasons to expect a correlation of errors between MID selection and MID outcome, victory, or duration (Heckman 1979; Reed 2000; Sartori 2003), we do not find this to be the case empirically. Nevertheless, we also estimated outcome models that jointly estimate sample selection with the victory and duration models and find consistent results across all models. 9
Independent Variables
We define democracy as 6 or above on the Polity IV scale (Marshall and Jaggers 2002) and include in all analyses two separate dummy variables for the presence (or absence) of democracy in each state of the dyad. When relevant, we also include the interaction of these two dummy variables that is positive for joint democracy in the dyad. Experimentations with different cutoff levels for democracy (5 or 7) do not substantively change the results we report, in any of the models. 10
Our control variables are, for the most part, common indicators used in many large-N studies of international conflict. These include the presence of contiguity, which is defined as direct land contiguity on the Correlates of War scale (Stinnett et al. 2002), the presence of any alliance in the dyad (Gibler and Sarkees 2004), and the capability distribution in the dyad, measured as the Composite Index of National Capabilities (CINC) score of state A divided by the CINC score of state B (Singer, Bremer, and Stuckey 1972). We also control for the presence of the hegemon in the dyad, which we code as Great Britain from 1816 until 1945 and the United States from 1946 to 2001. The hegemon of course has a tendency to involve itself in multiple conflicts across the globe. For the samples that include only directed dyad dispute years, we include a variable that identifies territorial disputes, which are central to our study; these cases are defined as all disputes in which the revisionist state (state A) is attempting to change the status quo.
Our argument and the democratic peace arguments on conflict behavior rely mostly on the interaction of several of the conflict predictors. For example, to test the argument that democracies are better than other regime types at selecting their conflicts, we include the interaction of a democratic initiator (the state A democracy dummy) and the presence of a very weak potential target (state B). We code weak targets as any states with capabilities that are one-quarter the size of the potential initiator. 11 Similarly, we would expect that democracies are part of a larger group of states that select noncontiguous conflicts, and we construct an interaction of democratic initiator and contiguity to test this.
Finally, since our first model estimation is a binary, cross-sectional time series of cases, we correct for temporal dependence in the data by adding a variable for the number of peace years since the last dispute in the dyad (Beck, Katz, and Tucker 1998). We also include cubic splines of this variable in the estimation but omit their coefficients in the presentation; these estimates are available from the authors.
Conflict Selection, Success, and Duration
Conflict Initiation
Our first set of tests assesses the ability of democracies to choose their dispute rivals. As our discussion above notes, increased audience costs among democratic governments should place strong incentives on choosing rivals wisely. Democracies are therefore likely to fight weaker opponents that can be beaten quickly since longer wars risk calls for leader replacement. If our argument is correct, however, the development of democracies begins with a process that also removes dangerous border issues from their conflict agendas. This establishes a stable peace with democratic neighbors and makes it less likely that democracies will initiate conflicts against contiguous states (Hypothesis 1). Thus, any conflict selection made by democratic leaders occurs amid an already constrained sample of potential rivals, and, if regime indeed has little effect on conflict selection, then there should be few regime-type differences over initiations against noncontiguous states (Hypothesis 2). 12
Table 2 presents results from four separate estimates of the likelihood of initiating a MID in a directed-dyad sample that includes the years 1816 to 2001. We vary the estimation across models to test each conflict selection explanation. We begin with a base model that includes all controls and three variables for regime type. In this model, democracies are more likely to be targeted in disputes, but, if both states are democratic, conflict initiation is unlikely. This is consistent with previous findings from the democratic peace literature.
Dispute Initiation in Directed-Dyads, 1816 to 2001.
Note. Logistic regression estimates of directed dyads; robust standard errors in parentheses. *p < .05. **p < 0.01. ***p < 0.001. aPreponderance is coded as one state possessing 80 percent of capabilities in the dyad; the table only reports confidence intervals for changes that are statistically significant, p < .10.
Victory for Side A in Directed Dispute Dyads, 1816 to 2001.
Note. aPreponderance is coded as one state possessing 80% of capabilities in the dyad; the table only reports confidence intervals for changes that are statistically significant p < .10.
The control variables also behave as expected. Contiguous states, dyads at parity, and the presence of the hegemon in the dyad all predict an increased risk of dispute initiation while lengthy peace spells (omitted from table) decrease the likelihood of conflict initiation. Only the presence of an alliance produces an unexpected result in the baseline model. Allied dyads are likely to have conflicts, even after controlling for contiguity. This leads us to believe that alliances may still be serving as a proxy for increased interactions in the dyad. Overall, the estimates for the control variables change little in either substantive effect or significance across the four models.
With model 2, we begin the test of the democratic selection explanation. If correct, democratic leaders should choose weaker rivals when initiating their disputes, but we do not find this to be the case. Using a continuous measure of capability scores (the CINC scores of state A divided by the scores of state B), we find no statistically significant effect for either the base term of capabilities or the interaction of democracy and capability ratio. The substantive effects analysis at the bottom of Table 1 confirms this finding. There is literally no change in the probability of dispute initiation by state A when we vary democracy and/or capability ratio. 13
Model 3 changes the interaction term to assess the explanation linking conflict selection to state development path. If states become democratic in part because of secured borders, then they should have little reason to fight their neighbors. To test this explanation, we added a dummy variable that is positive for all contiguous dyads in which the potential initiator is a democracy. In the estimation, the interaction variable is negative and statistically significant (p
We find strong confirmatory evidence for our expectations. The conflicts initiated by democratic leaders tend not to be contiguous to the democracy, which is consistent with the expectations of a territorial peace and confirm Hypothesis 1. Further, democracies are similar to other states in targeting noncontiguous rivals, which is consistent with the expectation of Hypothesis 2. These results for MID initiation affirm a strategic link in the distribution of the conflict data but provide little evidence that democracies are unlike other types of regimes when targeting weaker rivals. Democracies are less likely to have conflicts with other democracies. Democracies are also likely to be targeted in disputes. But nothing suggests that democratic leaders behave differently than other leaders when choosing their foes. We do, however, find support for the contention that democracies are among a class of states that are unlikely to target their contiguous neighbors. This suggests that the selection effects previously demonstrated for democratic governments may largely be determined by the ability of democratic leaders to select easier victories abroad.
Victory in Conflict
The democratic selection models contend that elected leaders select their wars well, their soldiers fight exceptionally well, and/or their governments and economies fight especially hard when involved in conflict. These explanations are all based upon the assumption that democracies are more likely than other regime types to be victorious in their conflicts. Our argument considers selection to again be the product of state development paths. Removal of dangerous territorial issues allows certain states, and most democracies, the ability to choose which rivals to engage. Thus, conflicts initiated by states with stable borders are more likely to be wars of choice, and wars of choice most often occur when the initiator considers victory to be the likely outcome. In terms of our hypotheses, we expect that contiguous conflicts will be more difficult to win (Hypothesis 3) and that democracies have no advantage when fighting these conflicts (Hypothesis 4).
We test these two logics with estimation procedures similar to those presented in the MID initiation models above. However, this time we restrict the sample to all directed dyads during the initial year of a dispute. This is of course a selected sample of cases, but we also now have a good understanding of the types of dyads involved in this sample. From Table 1, disputes tend to be initiated in contiguous, allied dyads that have had recent conflicts in the past. The presence of the hegemon in the dyad also increases dyadic dispute proneness. Strategic selection is present in the dyad since weaker states and democracies tend to be targeted, while states contiguous to a democracy are not.
The first model presents a baseline estimation of MID victory. According to the analysis, initiating states tend to fare better if they possess greater capabilities than their opponents and if their opponent is not a democracy. However, note again that this second result does not mean that democratic targets are winning these disputes since there are many possible outcomes to a dispute, including compromise, yields, and stalemates. Democracies do not win the disputes they initiate since the standard error for this dummy variable (0.253) is actually larger than its coefficient (0.219). Democracies do not win their conflicts in any of our models.
Initial support for our argument can be found in the first model as we find that contiguous disputes are less likely to be won by the initiator (confirming Hypothesis 3). We argued earlier that contiguous disputes are less likely to be conflicts of choice, and, if leaders are less able to choose when to engage in these disputes, then selection based on chances of victory becomes difficult. Conversely, since territorial issues are often so difficult to resolve, it may seem surprising that territorial disputes are more likely to be won. However, this result simply suggests that leaders only initiate territorial disputes, especially noncontiguous territorial disputes, when their chances of victory are greatest. It may also imply that not all territorial disputes are especially dangerous (e.g., border clashes and fishing rights). Leaders are probably more willing to engage in other, less dangerous types of disputes without careful planning for possible outcomes.
Model 2 adds the democratic selection variable. According to the logic of leader choice, democratic leaders have incentives to choose rivals that will be defeated easily. However, again, neither the dummy variable for democratic initiator nor the interaction term for democratic initiator and weak target is statistically significant at any conventional level. This is actually true for both models that use this interaction term. As the predicted probabilities at the bottom of Table 2 demonstrate, capability ratio changes have no substantive effect on the probability of victory, while the interaction term has only a modest effect of a 1.30 percent increased chance of victory, even when there is a massive change in the capability share of the initiator. It would seem that victories in disputes are only marginally determined by the capabilities of the actors involved, which is consistent with Slantchev’s (2004) expectations on the quality of information prior to conflict.
Model 3 presents a model that analyzes our argument on conflict selection. Democracies that fight conflicts against their neighbors most often lose their disputes. This is true for all states, actually, since contiguity remains negative and statistically significant in the model (p < .001); conflicts initiated against contiguous states are 51.65 percent less likely to be won, which is again consistent with Hypothesis 3. But what is important for our argument is that democracies behave no differently from other types of states when fighting their neighbors. Regime type makes almost no difference in predicting victory. Instead, the results suggest that the ability to choose conflict matters only for leaders who can choose among contiguous rivals.
The substantive effects do well at establishing this pattern. Among noncontiguous states, democracies actually tend to do less well than other types of states in choosing conflicts that are winnable. Democracies hold a 7.3 percent chance of winning these types of disputes, but this compares poorly to the average dyad’s 6.8 percent chance of victory in noncontiguous disputes. Thus, the only reason democracies are often naively correlated with conflict victory is that they are more likely to be isolated from the effects of contiguous rivalries. Contiguity controls the likelihood of victory, not ex ante capability differences. 15
Conflict Duration
Our final dependent variable is the length of conflict, measured in days between the initiator’s start and end dates. Observations of the measure range from 1 to 4,779 days, and we use ordinary least squares regression to estimate the effects of the covariates on dispute length. We report these results in Table 3.
Length of Dispute for Side A in Directed Dispute Dyads, 1816 to 2001.
Note. aPreponderance is coded as one state possessing 80 percent of capabilities in the dyad; the table only reports confidence intervals for changes that are statistically significant, p < .10.
According to the arguments associated with the broader democratic peace, we should find that democracies are more likely to fight disputes that are shorter in length. However, the baseline model for this dependent variable shows no statistically significant effect for democratic initiators. Ignoring statistical significance for a moment, the sign of democracy is negative as expected, but its substantive effect is quite small (at 3.1 days) compared to the other predictors. The only regime effects in this model are for democratic targets, which do fight shorter disputes on average, but this does not provide strong support for the contention that democratic leaders choose their conflicts especially well.
The control variables in the model confirm expectations. Territorial disputes are more difficult to manage and last longer on average. The presence of an alliance decreases the length of a dispute by approximately 81 days, which may imply that alliances can serve as a tool for conflict resolution. Contiguous disputes also tend to last longer than do other disputes, confirming Hypothesis 5, and we believe this supports the contention that disputes against neighbors are less likely to be conflicts of choice, started by the initiator.
Model 2 adds the capability-based selection variable to the baseline model. The addition of the interaction between democratic initiator and capability ratio has no effect, however. The base component of capability ratio still continues to have no effect in predicting dispute length. Substantively, even if the variables were statistically significant, the overall effects of major capability changes would remain quite small, at less than 0.18 percent, which makes capability differences an especially poor ex ante predictor of conflict duration.
Model 3 adds the selection variable based on contiguity, and we again find substantial differences between contiguous and noncontiguous conflict, but not according to regime type. Democratic initiations against noncontiguous states last 147 days on average, while democratic initiations in contiguous dyads last more than 211 days. These rates compare to a baseline of 154 days for the average dispute. In fact, the presence of democracy in an initiator actually extends the length of a dispute—by 10 days against noncontiguous rivals (137.76 days for all noncontiguous dyads vs. 147.68 days for noncontiguous dyads with democratic initiators) and by almost 40 days for contiguous disputes (172.92 days vs. 211.42 days when democracies initiate).
Together, these results suggest strongly that rival selection in disputes is controlled more by the location of conflict rather than by the regime type of the initiator. Contiguous disputes last longer than noncontiguous disputes (confirming Hypothesis 5), and, if democracies have regime-based incentives to choose shorter conflicts, their leaders are apparently do not respond to these incentives well. The presence of a democratic initiator actually prolongs the length of a dispute, once contiguity is controlled (confirming Hypothesis 6).
These results suggest that the apparent advantage of democracy in fighting shorter conflicts actually may result from inattention to dispute location. Democracies are more likely to fight noncontiguous disputes, which are shorter on average, and the prevalence of these disputes in their samples of conflicts biases the average conflict length downward. The selection process that controls dispute location is responsible for democracies’ shorter conflicts.
Conclusion
We used this article to assess several theories of democratic conflict behavior; specifically, that democratic leaders strategically select their potential targets so that democracies fight shorter conflicts which they are more likely to win or to negotiate a compromise. We find little support for democratic selection arguments, however. Democracies do not fight shorter disputes and are not more likely to win their disputes, once controls are included for the presence of contiguity. Instead, we find support for a theory of dispute selection that is based on the development paths of states. States that are democratic have fewer territorial conflicts with neighbors, and this implies that most disputed issues involving democracies are of lower salience. Insulated from direct threats, democratic leaders are able to choose when to escalate issues to conflict and against which targets. This leads to shorter conflicts that are more easily won.
Our theory and findings provide one of the first challenges to the larger democratic peace project. While much attention has been paid to the observation that democracies do not fight each other, few if any of these challenges have been extended to the many corollary conflict findings based on the incentives faced by democratic leaders. Here, however, we have provided a unified theory of territorial peace that explains both the observation of peace between democracies and the variation in dispute selection, duration, and chances for victory. In the future, we hope to extend our theory to other correlates of peace and war.
We have tried to highlight the advantages of paying close attention to state development paths in order to understand consequent conflict behavior. The conflict literature tends to treat all dyadic relationships as equal, missing the fact that rivalries and friendships change over time. Also missing are the effects that conflict has on the development of the state. Democracies and peace cluster in space and time for a reason, and this affects the types of disputes various regimes will likely face.
Footnotes
Acknowledgment
Gibler would like to thank Dean Robert Olin and the College of Arts and Sciences at the University of Alabama for sponsoring the conference. Gibler also thanks the National Science Foundation for partial support during the completion of this project (Grant # SES-0923406).
Authors’ Note
Article originally prepared for the University of Alabama mini-conference on Territory, Rivalry, and Domestic Politics.
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.
Notes
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
