Who’s careful: Regime type and target selection

Abstract

Democratic leaders are commonly thought to be more likely than autocrats to select into conflicts where the ex ante probability of victory is high. We construct a novel empirical test of this notion by comparing democracies to various types of autocracies and determining which states are most likely to initiate disputes against relatively strong or weak opponents. Contrary to common belief, we find that democracies are not more selective than most forms of autocracy. Only military regimes demonstrate unique patterns of target selection, with these states being particularly likely to initiate disputes against relatively strong opponents. Our findings suggest a need to further scrutinize the conventional wisdom on democratic target selection and to disaggregate autocratic regimes into more refined categories when doing so.

Keywords

Autocracy democracy international conflict military regime relative capability target selection

Introduction

Scholars have offered a variety of explanations for the fact that democracies rarely go to war with one another and tend to be victorious in the conflicts they are involved in. Among these are arguments that democracies are better able to signal resolve than other states (Fearon, 1994, 1997; Schultz, 2001) and fight more effectively once hostilities are underway (Bueno de Mesquita et al., 2003, 2004; Lake, 1992; Reiter and Stam, 1998, 2002). One of the most fundamental explanations, however, is “democratic selection,” or the notion that democracies are uniquely cautious and only select into conflicts when the ex ante probability of victory is high (Bueno de Mesquita and Lalman, 1992; Bueno de Mesquita et al., 2003, 2004; Filson and Werner, 2004; Reiter and Stam, 1998, 2002).

Despite its widespread influence, recent research has challenged the theoretical foundations that underlie this body of work. A key assumption behind democratic selection is that democratic leaders are more accountable for conflict outcomes than their autocratic counterparts, with unfruitful foreign policy actions leading to electoral punishment in these regimes. A comprehensive evaluation of leadership tenure, however, calls this into question, revealing that the tenure of democratic leaders is largely unaffected by unfavorable outcomes in military crises and war (Chiozza and Goemans, 2004). Furthermore, a growing body of research finds that domestic audiences can hold leaders accountable in some autocracies, and that autocratic leaders often risk severe punishment when removed from office (Debs and Goemans, 2010; Weeks, 2012, 2014). If this is the case, autocrats might also have strong incentives to behave cautiously. The conclusions of these studies warrant a re-examination of the democratic selection argument to determine whether democracies are, indeed, selecting easy targets and winnable conflicts and, if so, whether this care differs meaningfully from that of different types of autocracy.

In this study, we conduct a novel empirical test of the democratic selection argument that builds off previous analyses by addressing several issues endemic to the target selection literature. Among them is the fact that scholars have often tested this argument using one-dimensional measures of democracy and autocracy. While this provides some intuition on whether increases or decreases in the aggregate level of democracy or autocracy are associated with different patterns of target selection, it masks what variation exists between democracies and different types of autocratic regimes. To remedy this, we employ a range of commonly used regime classification typologies to determine whether democracies are really more cautious than all types of autocracies, or merely a subset thereof.

Another core issue is that existing studies often test target selection by examining which states experience the most favorable outcomes in the disputes or wars they initiate. This approach fails to distinguish between target selection and other factors that might explain the relationship between regime type and conflict outcome, such as signaling capacity and war-fighting ability. Instead, we identify relative capabilities as a criterion by which states identify “easy” targets and winnable wars, and examine how this affects the decision to initiate international conflicts across regime types. This allows us to examine target selection without conflating these ex ante processes with those that arise after a conflict has begun.

Our findings indicate that the empirical patterns and theoretical mechanisms of democratic selection are not as simple as previously believed. Contrary to conventional wisdom, democracies are not more likely than most forms of autocracy to select into disputes against relatively weak opponents. If democracies are selecting into conflicts that they are likely to win, they appear to be doing so through mechanisms that are more nuanced than those commonly identified in existing empirical work. Additionally, we find that only military autocracies are unique in their target selection practices, with these states particularly likely to initiate conflicts against relatively strong opponents. These results prompt a need to further scrutinize the conventional wisdom regarding target selection behaviors across democratic and autocratic regimes.

Target selection and democracy

Democratic selection has become a fundamental concept in the growing literature assessing domestic sources of international conflict. While scholars have noted that states generally initiate disputes in which they expect to be victorious (e.g. Gartner and Siverson, 1996; Gelpi and Grieco, 2001; Rousseau et al., 1996), democratic regimes are often thought to be most cautious in selecting of easy targets and winnable wars (e.g. Bueno de Mesquita and Lalman, 1992; Bueno de Mesquita et al., 2003, 2004; Filson and Werner, 2004; Reiter and Stam, 1998, 2002). Two of the most prominent theories constructed around this assumption are the institutional explanation of democratic peace offered by selectorate theory (Bueno de Mesquita et al., 1999, 2003, 2004), and Reiter and Stam’s (1998, 2002) explanation for why democracies tend to be victorious in war.

Selectorate theory utilizes assessments of leadership survival to explain why different regimes behave differently in interstate conflicts (Bueno de Mesquita et al., 2003). Leadership survival is said to be determined by the relative size of the selectorate within a state, defined as those individuals with a say in choosing the state executive, and the winning coalition, defined as the subset of the selectorate whose support is necessary for a leader’s continued survival (Bueno de Mesquita et al., 2003). In democracies, the winning coalition is relatively large and political leaders have a better chance at satisfying their large body of supporters by investing in public, rather than private, goods. As a result, democratic leaders rely heavily on the allocation of public goods to remain in office. Autocratic leaders, by contrast, are dependent on smaller coalitions and have an advantage in providing private goods to winning coalition members as a means of retaining office. In this way, the institutional make-up of a regime determines the type of policies that political leaders are likely to pursue.

When addressing interstate disputes, selectorate theory begins by assuming that conflict outcomes are public goods¹ and that the citizenry of any regime generally has a preference for foreign policy victories and incentive to punish foreign policy defeats. In small-coalition autocracies, leaders can offset these defeats through the allocation of private goods to winning coalition members. Put differently, the small coalition size of these regimes insulates leaders from the cost of defeat. In democracies, however, winning coalitions are too large for the widespread allocation of private goods to be a tenable strategy. Instead, democratic leaders avoid defeat by expressing caution when selecting into interstate conflicts and vigorously pursuing victory once a conflict has begun. Because of this, these leaders are unlikely to become involved in conflicts where they are weak relative to their opponents, and fight harder in ongoing conflicts by putting more resources into the war effort.²

Reiter and Stam (1998, 2002) develop similar expectations in explaining why democracies have a favorable record in war. Here, the pattern of cautious target selection among democracies is driven by two mechanisms. First, building upon Bueno de Mesquita and Siverson (2003) and Downs and Rocke (1994), the authors again point to the fact that democracies tend to hold regular and competitive elections, and domestic audiences are capable of punishing leaders who engage in reckless or bellicose foreign policy practices. Given the specter of electoral defeat, democratic leaders display more risk-averse tendencies than their non-democratic counterparts, initiating disputes only when the ex ante probability of victory is quite high.

Second, in addition to their fear of losing office after defeat, Reiter and Stam (2002) argue that democracies are able to select into winnable conflicts due to their ability to generate accurate ex ante predictions concerning the likelihood of victory. Unlike many autocracies, democratic regimes feature the presence of opposition parties and impose few restrictions on press and media services. These institutions have incentive to identify poor policies and critique risk projections put forth by the executives, potentially curbing the adoption of reckless foreign policy practices. When states lack these institutions, as is the case in many autocracies, the formulation of foreign policy is not subjected to a thorough vetting process. Instead, the cronyism and nepotism rampant in many autocratic institutions insulates leaders from the type of critical evaluation that might prevent quagmires aboard. Whether democracies are able to attain information that is not readily available to autocracies or whether autocracies are simply more likely to distort common knowledge is not explicitly articulated by Reiter and Stam; the net result, however, is again that democracies should be better able to identify conflicts in which they are likely to prevail.

While they differ in their specific mechanics, both selectorate theory and the work of Reiter and Stam (1998, 2002) share the assumption that democratic leaders are more sensitive to foreign policy failures than autocratic leaders and the expectation that democracies will be uniquely cautious:

Democratic Selection Hypothesis: Democratic leaders will be more cautious than their autocratic counterparts, initiating interstate conflicts only when the ex ante probability of victory is relatively high.

This expectation has received a wealth of empirical support in the conflict literature. Bueno de Mesquita et al. (2004), for example, find that the balance of military capabilities has a stronger bearing on the decision to initiate a war for large winning coalition regimes (i.e. democracies) than it does for more autocratic regimes. Reiter and Stam (1998) infer evidence of cautious selection in finding that democracies are particularly likely to win the wars they initiate, a result upheld using alternate data sources and modeling techniques (Clark and Reed, 2003; Rioux, 1998). Indeed, even when evaluated against competing theories centering on the superior military capacity of democracies (Lake, 1992), or the influence of capitalism and common policy interests among democracies (Faber and Gowa, 1995; Gartzke, 2007), the Democratic Selection Hypothesis is considered to be a robust explanation for why democracies succeed in the wars they initiate (Anderson and Souva, 2010; Gelpi and Griesdorf, 2001). Given this body of support, the democratic selection logic has become one of the most widely supported explanations behind the frequency of democratic victory.

Despite this widespread influence, the notion that democratic regimes are inherently cautious has been the subject of considerable debate. The qualitative evidence supporting the Democratic Selection Hypothesis appears to be particularly indeterminate, with prominent historical cases appearing to contradict the conventional wisdom on both democratic selection and victory. The US war in Vietnam is often referenced as a prominent case in which democratic leaders knowingly entered into a war where the prospects of victory were relatively low (Downes, 2009). Similarly, Desch (2002, 2008) conducts an examination of several historical cases of democracies at war and rejects the notion that democratic institutions have a strong bearing on the decision to initiate conflicts or the manner in which they are fought.³

Beyond this evidence, several empirical studies have questioned the extent to which the effect of democracy on conflict outcomes is substantively large or generalizable. Often, democracy exerts only a weak impact on conflict outcomes when compared to other factors, such as the quality of human capital, the balance of military capabilities, and the number of coalition partners in a multilateral war (Desch, 2002; Graham et al., forthcoming; Henderson and Bayer, 2013). Moreover, Henderson and Bayer (2013) find that only Western democracies enjoy a significant advantage in interstate conflicts, suggesting that traditional links between democracy and conflict outcomes may not be universal.⁴

Each of these studies undercuts some of the fundamental assumptions that link democracies with favorable conflict outcomes. Nevertheless, in focusing on victory in war, this work does not directly test whether democracies are cautiously selecting their adversaries at the outset of these conflicts. We begin this assessment by examining recent work on autocratic regimes to determine the extent to which various institutional constraints perpetuate cautious or reckless target selection.

Target selection among autocracies

A growing body of work on autocratic institutions suggests that democracies may not be unique in terms of their vulnerability to conflict outcomes or their caution when selecting into disputes. Work on leadership survival, for example, demonstrates that while democratic leaders have a high probability of being removed from office in general, the risk of removal is unassociated with negative conflict outcomes (Chiozza and Goemans, 2004). Considerable variation also exists in terms of how autocratic leaders are removed from office. In some autocratic regimes, transitions of power are peaceful. In others, however, removal from office often means death or exile for standing leaders (Debs and Goemans, 2010; Goemans, 2008). While this research does not address the issue of target selection specifically, it nevertheless calls into question why democracies should be uniquely cautious if they are not uniquely accountable.

In addition to studies on leader tenure and post-exit fate, there is reason to believe that some autocrats are driven by political constraints similar to those in democracies. Weeks (2014), for example, argues that audiences in party-based autocracies are just as likely as those in democracies to sanction their leaders for poor performance in international affairs. Party-based regimes and democracies are therefore expected to behave similarly in international conflicts. To this end, party-based autocracies have been found to initiate conflicts at a similar rate to democracies (Weeks, 2012), appear to be equally adept in signaling resolve in these disputes (Weeks, 2008), and win a similar proportion of the conflicts they initiate (Weeks, 2014: ch. 3). This latter result has led to the conclusion that leaders in democracies and party-based autocracies act with similar degrees of caution internationally.

While the consolidation of power in a dominant political party may induce the cautious formulation of foreign policy, scholars are more divided in identifying the specific institutional features that make some autocracies especially prone to initiating high-risk conflicts. The role of militarism in conflict processes has been quite divisive. Much of the work on civil–military relations suggests that military regimes and individuals with a military background may be conservative in their decisions to use military force abroad (Andreski, 1980; Betts, 1991; Huntington, 1957; Van Evera, 1999). A common assumption in this body of work is that individuals with military training will be more attuned to the costs of military operations, while their civilian counterparts may be overly optimistic about both the costs and efficacy of military force. This belief has been buttressed by survey evidence that military officers are less likely than civilian elites to support interventionist policies over issues like human rights violations and state-building (Feaver and Gelpi, 2004: ch. 2).

More recent empirical work has challenged traditional notions of military conservatism. Horowitz and Stam (2014), for example, find that state leaders with military training and no combat experience have a relatively high propensity to initiate interstate disputes, while leaders that are veterans or have no military experience are indistinguishable from one another. At the state level, there is a considerable body of evidence that military regimes are more likely to initiate militarized conflicts than are non-military regimes (Lai and Slater, 2006; Sechser, 2004; Weeks, 2012). Some attribute this relationship to the fact that military regimes are typically bereft of the representative institutions necessary to co-opt domestic opposition forces, which, in turn, strengthens incentives to initiate diversionary conflicts as a means of quelling domestic unrest (Lai and Slater, 2006). Others attribute this to the training and parochial interests of the military, claiming that military leaders are more likely to view the use of military force as a legitimate and effective way to resolve international disputes and less likely to recognize diplomatic alternatives to dispute resolution (Weeks, 2012, 2014). Regardless of what the specific link is, each of these studies share a common assumption that the militarization of international disputes is seen as less costly for military leaders than it is for their civilian counterparts.

Military leaders may also be less cautious in selecting targets than civilian leaders and regimes. If military leaders are, indeed, less likely to seriously consider diplomatic strategies, they may more quickly adopt relatively high-risk preventive or pre-emptive strategies. Weeks (2014: ch. 5), for example, attributes the Argentine junta’s ill-fated decision to initiate the Falklands War to an exaggeration of the threat posed by the UK and the desire to secure first-strike advantage. Similar incentives appear to have been at work in Germany and Austria-Hungary in the decades preceding the First World War, with military elites frequently advocating preventive war, only to be restrained by their more cautious civilian counterparts (Mulligan, 2014). In addition to this anecdotal evidence, military regimes have been found to perform relatively poorly in the disputes and wars that they do initiate, which may be consistent with poor target selection (Weeks, 2014: ch. 3).

In addition to the strength of party and military institutions, personalism, or the degree to which an autocrat has removed institutional constraints on executive power, has been explored as an explanation for why regimes may be either cautious or reckless. As previously stated, much of the existing literature operates under the assumption that personalist leaders are more insulated from the costs of initiating and losing conflicts than the leaders of any other type of regime (Bueno de Mesquita et al., 1999, 2003, 2004; Weeks, 2008, 2012, 2014). Evidence of this dynamic is purportedly found in the history of leaders like Saddam Hussein, who was able to stave off multiple international defeats without severe ramifications from domestic opposition forces (see Reiter and Stam, 2002). Perhaps due to this insulation, personalist leaders initiate conflicts at a higher rate than other regimes and have a poor record in both disputes and war (Reiter and Stam, 2003; Weeks, 2012, 2014).

Despite these findings, some have argued that other forms of autocracy might still be more reckless than personalist regimes. Reiter and Stam (2002: 24) argue and find evidence for the contention that extremely dictatorial regimes will actually be more cautious in entering into disputes than less centralized autocracies because the latter will be prone to overexpansion and logrolling coalitions that produce suboptimal foreign policy strategies. Further, personalist leaders are killed or forced into exile more often than other autocrats (Escribà-Folch, 2013). While this may drive leaders to bargaining tactics that ultimately result in war (Debs and Goemans, 2010), it is also reasonable that the prospect of a violent end might cause personalist leaders to express some degree of caution.

In sum, relative to the literature on democracy, there is far less agreement on which configuration of autocratic institutions makes states cautious or reckless. We are agnostic as to which of these autocratic institutions will be most strongly linked with either caution or recklessness. Instead, we outline this body of work to highlight the critical importance of examining variation among different types of autocracies when assessing whether democracies are, indeed, unique in their selection behavior.

Testing democratic selection

The literature on democratic target selection has also been hampered by several methodological difficulties. Among these, three are particularly acute. The first is how to go about inferring evidence of the selection process or the means by which state leaders determine which potential opponents are “easy targets.” Oftentimes, scholars utilize conflict outcomes or information revealed ex post toward this end. This approach is problematic because it inhibits the ability to adjudicate between ex ante processes and those that take place once a conflict is underway.

More specifically, much of the empirical literature on target selection finds that democracies are very likely to win the conflicts they initiate, which is interpreted as evidence that these regimes are choosing their opponents cautiously (Gelpi and Griesdorf, 2001; Reiter and Stam, 1998; Weeks, 2014). Careful target selection is, however, just one of several potential explanations for this finding. The literature on audience costs, for example, argues that democracies (and some forms of autocracy) can issue more credible threats and signals of resolve than other states (Fearon, 1997; Weeks, 2008). If this is the case, democracies may succeed in the majority of the disputes they initiate not because of careful selection, but because they are better able to send credible signals of resolve, which leads their opponents to back down or make policy concessions. Similarly, democracies are commonly thought to fight harder once fighting has begun (Reiter and Stam, 1998, 2002; Bueno de Mesquita et al., 2003, 2004). The motivation to fight hard may be highest when democratic leaders start conflicts since domestic audiences will see these leaders as being culpable for these conflicts and more likely to punish poor performance as a consequence (Croco, 2011). Put differently, democracies might generally fight harder than autocracies, but especially so in conflicts they initiate. Beyond the capacity to signal resolve and fight effectively, democracies tend to fight in larger international coalitions, which itself improves the prospects for victory (Choi, 2003, 2004, 2012; Graham et al., forthcoming).⁵ Cautious target selection is therefore just one of many explanations that potentially explain the phenomenon of democratic victory.

Reiter and Stam (2002) construct a more nuanced test of target selection. Here, the authors estimate the probability of losing a war for a sample of states in crises and test whether democracies become more likely to initiate militarized disputes when their likelihood of winning increases. They do this by interacting the estimated probability that a state will lose a given conflict by its democracy score and find that non-democracies are more likely to initiate conflicts with lower likelihoods of victory than are democracies (Reiter and Stam, 2002: 50). These estimated probabilities are generated by regressing an indicator of MID (Militarized Interstate Dispute) outcome on factors such as military strategy, capabilities, regime type, and terrain. While many of these factors are observable ex ante, their effect on dispute outcome is estimated using information on dispute outcomes, which is only revealed ex post.⁶ This approach attempts to mimic a leader’s forecasting assessment before initiating a conflict, but any information after war initiation was unavailable at the moment of decision-making. Due to this, the probability measures are endogenous to a leader’s decision to initiate a militarized dispute in the first place. Moreover, the estimated probabilities are already affected by a leader’s ex ante selection, which can also be tampered by unobservable factors. For this reason, the uncertainty around the point estimates of the probabilities of victory should be appropriately captured in the equation of conflict initiation.⁷ For example, this uncertainty itself might affect a leader’s decision to initiate a conflict, and, more importantly, the magnitude of uncertainty might vary systematically between democracies and non-democracies if democracies are better at distinguishing between winnable wars and disadvantageous wars. In sum, examining whether democracies experience favorable outcomes in the wars they initiate provides, at best, indirect evidence for whether or not careful target selection is actually occurring.

A second empirical issue concerns the units of analysis typically employed by scholars testing for cautious target selection. The concept of democratic target selection is centered on the manner in which states decide to transition from a period of peace to one of militarized conflict. Several existing studies, however, omit observations of peace and test the Democratic Selection Hypothesis solely by examining the characteristics of states involved in disputes or war (Bueno de Mesquita et al., 2004; Clark and Reed, 2003; Reiter and Stam, 1998, 2002). By ignoring the larger pool of potential challengers that choose not to initiate disputes, these approaches are prone to selection bias.

The problem with omitting peaceful dyads is illustrated in Figure 1, which displays the overall distribution of potential challengers’ relative capabilities⁸ among all directed dyads displayed across Polity scores⁹ using a fractional polynomial approximation. This distribution is given for both the total population of directed dyads in the interstate system and the subsample of directed dyads where states are engaged in MIDs. When only examining the sample of MID participant dyads, democratic challengers typically hold a power advantage against their opponents. This might, at first, appear to be evidence of careful target selection since democracies initiate disputes against relatively weak targets. Examining non-conflictual dyads, however, reveals that democracies also hold a relative power advantage against potential targets in this population as well. Figure 1 also demonstrates that the relative capabilities of all regime types increase when moving from the total population of dyads to MID participant dyads, and that this increase does not necessarily appear to be larger for democracies than it does for less democratic regimes. This finding indicates that states tend to select relatively weaker targets, but that this tendency does not appear to be exceptionally large for democracies. The fact that democracies often fight relatively weak opponents is therefore insufficient evidence for distinctively cautious target selection behaviors of democracies.

Figure 1.

Challenger’s relative capabilities across polity scores in directed dyads.

A related problem occurs when testing target selection by examining whether disputes escalate to the level of war. This approach conflates the decision to initiate a dispute, which is made by a single state, with the process of escalation, which is the product of decisions made by two or more states. For example, the observation that democracies tend to have a favorable balance of capabilities in war may be caused either by democracies initiating disputes against relatively weak opponents or because weak opponents are more likely to resist against strong democracies. It might also be that democracies are more likely to experience bargaining failures with weak states once a conflict has become militarized. For these reasons, examining conflict escalation into war alone tells us little about whether democracies are cautiously selecting into the disputes that precede war.

Third, and finally, much of the existing literature fails to explore variation in target selection among different types of autocratic regimes. Most often, scholars either assess target selection by comparing democratic regimes to a residual category of all non-democracies, or utilizing the Polity scale or measures of winning coalition size in testing for a linear association between democracy and the propensity to select targets carefully. This approach is problematic because it may mask variation in patterns of target selection among autocracies by placing them all in a single residual category or assuming that their selection patterns vary as a linear function of the level of democracy or autocracy. More recent work is beginning to address the question of autocratic target selection, but is doing so by examining conflict outcomes, which does not allow one to adjudicate between selection, signaling, and war-fighting mechanisms (Weeks, 2014: ch. 3).

While various studies have addressed some of these problems in isolation, none that we are aware of has addressed each of these issues. This is an important omission in the empirical literature of democratic selection and one that impedes the ability to make inferences about whether democracies are indeed uniquely cautious when entering into interstate conflicts.

Research design

We construct a rigorous test of democratic target selection by addressing each of the shortcomings outlined in the previous section. As target selection refers to the process by which a state chooses to initiate disputes against potential targets, we begin by establishing the directed dyad year as our unit of analysis. Using this framework, our data contain two observations for all pairings of countries in the international system in a given year, one where each state serves as the potential challenger and the other as the potential target. These data span the 1946 to 2001 period.

In order to evaluate the onset, and not escalation or conclusion, of interstate conflicts, our dependent variable is MID Initiation, which records whether the challenger (State A) initiated a MID against the target (State B) in a given year (Ghosn et al., 2004).¹⁰ MID Initiation is a dichotomous variable that is coded as 1 if a state threatens, displays, or uses military force against a target state within a given year, and 0 otherwise.¹¹ The initiator of a MID is simply the first state recorded to have taken military action and not necessarily the state responsible for instigating hostilities. Nevertheless, this operationalization allows us to determine which regimes are most prone to militarized interactions as a means of managing or resolving international issues.

Testing the Democratic Selection Hypothesis also requires that we identify a means by which potential challenger states differentiate between “easy” and “difficult” targets. We contend that evidence of selection can best be observed based on the characteristics or perceptions of potential target states that are observable to the challenger a priori. Following Bueno de Mesquita et al. (2004), we use national capabilities as one such characteristic, and expect that cautious leaders will be more likely to initiate military conflicts against relatively weak opponents. Our first primary explanatory variable is therefore Target Relative Capabilities, which is obtained by dividing the target’s Composite Index of National Material Capabilities (CINC) score by the sum of the challenger’s and target’s CINC score (Singer et al., 1972). This approximates the balance of military power between two states in any given year.

While this measure certainly does not capture all characteristics of a target relevant in interstate disputes, it is nevertheless among the most apparent factors related to target strength that are observable ex ante. Alternative approaches often require modeling choices that are tenuous or difficult to implement in the dyad-year framework. Bueno de Mesquita et al. (2004), for example, use the dispute participant as their unit of analysis and measure relative capabilities using the summed capabilities scores for all participants in each side of an observed dispute. In the dyad-year setting, however, this would require that a researcher guess whether and which third parties would become involved if a dyad at peace were to experience a conflict. As there is often considerable uncertainty in this regard, this approach would be questionable in a dyad-year setting. Similar issues are encountered if one were to attempt to assign battlefield strategies to dyads at peace, as Reiter and Stam (2002) did in their analysis. In short, we use dyadic relative capabilities because this provides for a clear, interpretable test of the Democratic Selection Hypothesis that can be implemented for both dyads in conflict and at peace.

Our second primary explanatory variable is regime type, for which we utilize several different classification schemes in order to capture the differences both between democracies and non-democracies, and among different types of autocratic regimes. First, we utilize a state’s Polity score to determine whether that state can be classified as democratic or non-democratic (Marshall and Jaggers, 2009). We operationalize Democracy as a dichotomous indicator that is equal to 1 if a state has a Polity score higher than 6. All states with Polity scores at or beneath this cutoff are classified as non-democracies.

Second, we identify regime type using the Cheibub, Gandhi, and Vreeland (2009) (hereafter CGV) classification scheme. Unlike Polity, CGV code democracy in a dichotomous fashion. Specifically, the Democracy indicator obtained via CGV is equal to 1 if: the chief executive is either popularly elected or chosen by a body that was popularly elected; the legislature is popularly elected; there are multiple parties competing in elections; and alternation in the leadership occurs, that is, multiple parties have peacefully assumed power under the same set of electoral rules (Cheibub et al., 2009: 69). Three additional dichotomous measures are also obtained from CGV’s classification of autocratic regimes including Civilian, Military, and Monarchy. The CGV coding criteria are simple. Monarchies are regimes where the head of government bears the title of king, and where rules of succession are based on heredity. If the head of government is a current or past member of the armed forces, the regime is classified as a Military autocracy, if not, the regime is classified as a Civilian autocracy. Note that the CGV classification of autocracies is based on characteristics of the regime leader alone and does not account for factors such as personalism or autocratic consolidation.

Third, we employ the Geddes, Wright, and Frantz (2014) (hereafter GWF) classification of regime type. The GWF classification is dedicated primarily to differentiating between autocracies based on the characteristics of a leadership group within a given regime.¹² From this typology, we utilize mutually exclusive dichotomous measures of Personalist, Single Party, and Military regimes. Personalist is an indicator of whether an autocrat derives power from a small group of political elites with few institutional checks constraining the leader’s authority. Single Party is a dichotomous measure of whether power in a state is concentrated through the infrastructure of a single, dominant political party. Military identifies whether political power is concentrated primarily within military institutions. Finally, the GWF indicator of Democracy is used, which is coded using similar coding rules to CGV with respect to free and fair elections, but without the alternation requirement. Note that the primary difference between the CGV and GWF typology is that the latter takes personalism into account. For example, when using the CGV typology, the military regime indicator can either pertain to regimes where a single military leader has consolidated power, or regimes characterized by the collegial rule of several military officers. In the GWF typology, only the latter constitutes a military regime.

Last, we employ the Weeks (2012) classification of autocratic regimes. Here, regimes are stratified along two dimensions: personalism and the political power of military institutions. From this, another mutually exclusive typology is generated. Machine identifies non-personalist regimes where the leader is constrained by a civilian audience of political elites. Junta also identifies non-personalist regimes, but here the constraining audience is made up of military officials. Boss refers to personalist regimes led by an individual with a civilian background. Strongman codes personalist regimes with a military leader whose small inner circle is comprised mainly of military advisors.

We include several control variables. Capabilities measures the absolute CINC score of a given state; this measure is included for both the potential challenger and target states. We include these measures of absolute capabilities because they are likely associated with both the relative capability score within a dyad and the likelihood of conflict initiation. We expect that stronger states may be better able to project their force and engage in militarized conflicts abroad. Contiguity is included to adjust for the geographic proximity between two states (Stinnett et al., 2002). We expect that contiguous states may be more likely to engage in conflict with one another either due to border disputes or their frequent interactions. Alliance Similarity is measured using Signorino and Ritter’s (1999) S score under the presumption that states with similar alliance portfolios may be less likely to initiate conflict against one another. Trade Dependence measures the degree to which two states are economically dependent on one another (Gleditsch, 2002).¹³ This measure is included because high economic interdependence might reduce states’ resolve for using military force (Russett and Oneal, 2001). Finally, to account for temporal non-independence, we include Peace Years, which records the number of years since the last militarized dispute for a directed dyad, along with its squared and cubed terms (Carter and Signorino, 2010).

Model specification and expectations

Given the dichotomous nature of our dependent variable, MID Initiation, we estimate a series of logit regression models. For illustrative purposes, the general form of the model specification is given in Equation 1:

Equation 1 : M I D_{i j} = f (\begin{array}{l} β_{0} + β_{1} T a r g e t R e l a t i v e C a p a b i l i t i e s_{i j} + β_{2} D e m o c r a c y_{i j} \\ + β_{3} D e m o c r a c y \times T a r g e t R e l a t i v e C a p a b i l i t i e s_{i j} \\ + β_{4} A u t o c r a c y_{i j} + β_{5} A u t o c r a c y \times T a r g e t R e l a t i v e C a p a b i l i t i e s_{i j} \\ + β_{6} C o n t r o l s_{i j} + ϵ_{i j} \end{array})

Note that the measurement of democracy and autocracy will differ depending on which of the four typologies is used, as will the reference category and the number of autocracy indicators included in the model. Each of the regime-type indicators in the model is interacted with the target’s relative capabilities score. The Democratic Selection Hypothesis maintains that democratic leaders should be more cautious in initiating disputes than non-democratic leaders. In terms of Equation 1, this would mean $β_{1} + β_{3} < β_{1} + β_{5}$ , or that the (presumably negative) effect of target relative capabilities is stronger for democracies than it is for autocracies. The strongest support for the Democratic Selection Hypothesis would be obtained if target relative capabilities were found to have a stronger negative impact on dispute initiation for democracies relative to all forms of non-democracy.

Given the symmetric nature of interaction terms (Berry et al., 2012), the fact that the effect of capabilities on dispute initiation is expected to vary across regime type implies that the effect of regime type also varies with relative capabilities. To this end, we also examine the effect of regime type across the range of relative capabilities. Evidence of cautious target selection by democracies would be uncovered if the difference between democracies and other regime types is strongest when the potential target is relatively strong, with democracies unlikely to initiate disputes. With respect to the notation in Equation 1, this implies $β_{2} + β_{3} T a r g e t R e l a t i v e C a p a b i l i t i e s < β_{4} + β_{5} T a r g e t R e l a t i v e C a p a b i l i t i e s$ , especially when the value of Target Relative Capabilities is high. For the Democratic Selection Hypothesis to hold, we expect both that the effect of relative capabilities is stronger for democracies than autocracies, and that the effect of democracy is negative and strongest when the potential target is more powerful than the potential challenger.

Results and analysis

Table 1 contains the results from the models estimated to test whether democracies are more or less selective in initiating MIDs compared to various classifications of autocracy.¹⁴ Models 1–4 feature the results obtained when using the Polity, CGV, GWF, and Weeks typologies, respectively. While the coefficient estimates are informative in some capacity, these results are better presented graphically given the relatively large number of regime types used in the CGV, GWF, and Weeks typologies.

Table 1.

Effects of relative capabilities and regime type on initiation of MIDs, 1946–2001.

	(1)	(2)	(3)	(4)
	Polity	CGV	GWF	Weeks
Target Relative Capabilities	−0.381**	−1.004**	−0.268	−1.220***
	(0.158)	(0.453)	(0.463)	(0.211)
Democracy	−0.101	0.274	0.252	−0.696***
	(0.160)	(0.293)	(0.298)	(0.192)
Target Relative Capabilities × Democracy	−0.250	0.192	−0.154	0.709**
Target Relative Capabilities × Democracy	(0.245)	(0.477)	(0.500)	(0.295)
Military		0.394	0.338
		(0.292)	(0.343)
Target Relative Capabilities × Military		1.175**	0.904
Target Relative Capabilities × Military		(0.494)	(0.585)
Civilian		0.418
		(0.287)
Target Relative Capabilities × Civilian		0.204
Target Relative Capabilities × Civilian		(0.475)
Single Party			0.223
			(0.289)
Target Relative Capabilities × Single Party			0.188
Target Relative Capabilities × Single Party			(0.489)
Personalist Autocracy			1.025***
			(0.304)
Target Relative Capabilities × Personalist Autocracy			0.223
Target Relative Capabilities × Personalist Autocracy			(0.289)
Machine				−1.191***
				(0.261)
Target Relative Capabilities × Machine				1.518***
Target Relative Capabilities × Machine				(0.449)
Boss				−0.337
				(0.239)
Target Relative Capabilities × Boss				1.216***
Target Relative Capabilities × Boss				(0.422)
Junta				−0.798**
				(0.328)
Target Relative Capabilities × Junta				2.165***
Target Relative Capabilities × Junta				(0.472)
Strongman				−0.161
				(0.215)
Target Relative Capabilities × Strongman				1.251***(0.368)
Capabilities (Side A)	7.802***	7.677***	8.610***	9.976***
	(1.412)	(1.433)	(1.335)	(1.302)
Capabilities (Side B)	9.306***	10.124***	8.657***	9.051***
	(1.373)	(1.318)	(1.361)	(1.435)
Alliance Similarity	−0.780***	−0.776***	−0.819***	−0.856***
	(0.159)	(0.162)	(0.162)	(0.158)
Contiguity	3.680***	3.725***	3.604***	3.721***
	(0.127)	(0.127)	(0.129)	(0.130)
Trade Dependency	−4.915	−3.337	−4.017	−2.310
	(4.829)	(4.051)	(4.255)	(4.021)
Constant	−5.357***	−5.739***	−5.657***	−4.944***
	(0.177)	(0.308)	(0.309)	(0.193)

Observations	918,421	1,012,769	845,868	873,847

Note: *** p < 0.01; ** p < 0.05; * p < 0.1. Robust standard errors, clustered by dyad, are given in parentheses. Regime coefficients should be interpreted relative to the reference category, which is non-democracy in Model 1, monarchy in Models 2 and 3, and “other” non-democracies in Model 4. The peace-year polynomials are not reported in this table.

If the Democratic Selection Hypothesis is correct, democracies should be more sensitive to increases in the target’s relative capabilities than should autocracies. Figure 2 displays the difference in the effect of relative capabilities on dispute initiation between democracies and autocracies. These values are generated by first calculating the change in odds associated with a one standard deviation increase in target relative capabilities for democracies and each type of autocracy. We then calculate the difference in these values to determine if democracies are more strongly affected by changes in relative capabilities. Negative values indicate that democracies are less likely than a given type of autocracy to initiate a dispute when a target becomes more capable. Put differently, negative values indicate that a type of autocracy is more reckless than democracies when choosing targets of interstate conflicts. Positive values indicate the opposite, that a democracy is more likely to initiate a dispute when a target’s capabilities increase, suggesting that autocracies are more cautious than democracies.

Figure 2.

Comparing the effect of target’s relative capabilities across regime type.

Contrary to expectations, democracies are not more cautious than most forms of autocracy. Beginning with the Polity typology, we find that the difference in the effect of relative capabilities is not significantly different between democracies and non-democracies. In Figure 1, this is evidenced by the confidence intervals pertaining to non-democracies overlapping with zero.¹⁵ We therefore cannot conclude that democracies are more cautious than a residual category of all non-democracies.

Disaggregating autocracies reveals that democracies tend only to be more cautious than military regimes. Within the CGV typology, we find that the effect of capabilities is not significantly different between democracies and either monarchies or civilian regimes. Military regimes, however, are significantly more reckless than democracies, as evidenced by the negative estimate with confidence intervals that do not contain zero. More precisely, the odds that democracies will initiate a conflict are about 30% lower than they are for military regimes after a standard deviation increase in the target’s relative capabilities score. In other words, these military autocracies are significantly more likely to initiate disputes against relatively strong targets.

Similar patterns persist when using other classifications of regime type. Moving to GWF, democracies are not significantly more cautious than all non-military autocracies, including monarchies, as well as personalist and party-based regimes. This contradicts recent work suggesting that personalist regimes will be most reckless in their foreign policy practices due to the lack of institutional checks on the executive. Instead, military rule again emerges as the only form of governance that is significantly more reckless than democracies. Specifically, the odds that a democracy will initiate a dispute are 32% lower than they are for military regimes after an increase in relative capabilities. Moreover, because GWF does not classify highly personalist autocracies within the military regime category, these results indicate that collegial military rule, in particular, is linked with an increased propensity to initiate disputes against relatively strong opponents.

Unlike the CGV and GWF typology, the Weeks typology disaggregates party-based and military regimes into personalist and non-personalist categories, allowing us to more deeply examine the specific institutional characteristics that distinguish some regimes from others. In this framework, democracies are only more cautious than juntas and are not significantly different from autocracies classified as boss, machine, or strongman. The fact that juntas, not strongman, regimes are most reckless further underscores the fact that non-personalist military rule is most tightly linked with incautious target selection.

No matter which regime typology is used, the effect of relative capabilities is not stronger for democratic regimes than it is for autocracies as the Democratic Selection Hypothesis predicts. Instead, we consistently find that the effect of relative capabilities is indistinguishable between democracies and all forms of non-military autocracy. We therefore cannot conclude that democracies are more cautious than these regimes. Rather than democracies behaving most cautiously, these findings highlight the reckless behavior of military regimes. Further, we find that military regimes are also more reckless than monarchies and civilian autocracies in the CGV typology and personalist regimes in the GWF typology.

While we find little support for the Democratic Selection Hypothesis when examining the differential effect of relative capabilities across regime type, it is also useful to examine how the effect of democracy varies with respect to relative capabilities. Investigating this issue informs the target selection debate by allowing us to identify behaviors such as caution and opportunism more precisely. Highly cautious regimes, for example, would be very unlikely to initiate disputes when they are relatively weak, and would become significantly more similar to other regimes as their capabilities increase. Highly opportunistic regimes, on the other hand, would be relatively likely to initiate disputes when they are relatively powerful, and would become significantly more similar to other regimes as their capabilities decrease. It is also possible that some regimes meet both criteria, being very unlikely to initiate disputes when they are weak, and very likely to do so when they are strong.

Figure 3 reports the effect of changing between various regime types across the range of relative capabilities. As already established, the effect of relative capabilities is only significantly different between democracies and military autocracies. These are therefore the only candidates for regimes that are systematically more reckless or opportunistic than democracies. The effect of a change from democracy to military regimes in the GWF and CGV typologies, as well as juntas in the Weeks typology, is reported in panels (b), (e), and (k) of Figure 3, respectively. In each case, when a target is relatively weak, democracies are no more or less likely to initiate disputes than military autocracies. As the target’s strength increases, however, military regimes become increasingly more likely to initiate disputes. Put differently, by choosing relatively strong opponents, military regimes appear to be selecting into conflicts in which they are likely to lose. This is consistent with reckless, not opportunistic, behavior on the part of these military regimes.

Figure 3.

Effect of regime type across target’s relative capability score.

The level of relative capabilities does not significantly alter the relationship between democracy and the remaining categories of autocracy. In terms of the coefficients reported in Table 1, this is evidenced by the fact that the interaction between democracy and relative capabilities is either statistically insignificant or not significantly different from the interactions between relative capabilities and other types of autocracy. This is also illustrated by the shallow slopes of the lines in the remaining panels in Figure 3. We therefore cannot draw any conclusions about democracies being more or less cautious than these other regimes.

While perhaps no more cautious, we do find that democracies initiate conflicts at difference rates from other regimes. Panel (a), for example, displays the effect of changing from democracy to non-democracy using the dichotomous Polity indicator. When potential targets are relatively weak, the difference between democracies and non-democracies is statistically insignificant, as indicated by the zero line overlapping the 95% confidence intervals. As a target increases in strength, the difference between these regime types becomes slightly larger, until democracies become significantly less likely to initiate disputes. While the general trend of this relationship is consistent with the Democratic Selection Hypothesis, the difference between democracies and non-democracies is not significantly larger at high levels of relative capabilities. As a result, we cannot infer evidence of cautious selection on the part of democracies, only that non-democracies are somewhat more likely to initiate interstate disputes.

The effect of changing from a democracy to a civilian, party-based, or machine regime is reported in panels (c), (f), and (i), respectively. In each case, the difference between democracies and these regimes is statistically insignificant across the range of relative capabilities scores. Thus, in addition to being no more cautious than party-based or civilian autocracies, democracies are also no more or less likely to initiate disputes. This is consistent with expectations that the presence of strong party-based institutions leads autocracies to initiate disputes at a similar rate to democracies (Weeks, 2012). More broadly, this supports the contention that party institutions may constrain autocratic leaders in such a way that leads to relatively peaceful foreign policy practices.

The effect of changing from democracy to a GWF personalist regime is always positive and statistically significant, as displayed in panel (g). Panels (j) and (l) display the effect of changing to Weeks’ personalist categories of boss and strongman. While strongman regimes are always more likely to initiate disputes, bosses are indistinguishable from democracies when a target is very weak (with a relative capabilities score less than about .15) and become more likely to initiate disputes once a target is moderately strong. As before, however, this change with respect to relative capabilities is not significant, indicating neither caution nor recklessness. The overall difference between democracies and personalist regimes therefore appears to be that the latter initiate disputes more frequently, even if they do not select the targets of these disputes any differently from democratic regimes.

Finally, panels (d) and (h) report the results pertaining to monarchies in the CGV and GWF typologies. Interestingly, the difference between democracies and each of these types of monarchies is statistically insignificant. This is somewhat counterintuitive since the lack of political constraints in these regimes might otherwise be expected to produce a level of bellicosity similar to that observed among personalist leaders. Nevertheless, these results should be interpreted with caution given the relative paucity of monarchies in the modern international system.

In sum, while democracies are less likely to initiate disputes compared to various types of autocracy, they are only significantly more cautious than military regimes. The difference between these regimes is starkest when potential targets are strong, with military regimes increasingly likely to initiate conflicts, a pattern indicative of reckless target selection behavior. These results lend additional support to much of the literature which suggests that democracies do not always proceed with caution and that military regimes often become engaged in ill-fated international conflicts. We should perhaps consider historical events like the US decision to escalate in Vietnam as less aberrant or anomalous than the conventional wisdom would suggest. Moreover, cases such as the Argentine experience in the Falklands War and the Japanese engagement in the Second World War appear to be indicative/demonstrative of a broad pattern of reckless target selection from military regimes. More broadly these findings bolster calls in recent research to challenge the conventional linkages between democracy and victory in interstate disputes and war.

We perform a variety of robustness checks to ensure that our particular modeling or sample choices do not drive the main findings.¹⁶ Our conclusions are supported even after making several changes to our research design. For example, our findings are unaffected after we restrict our dependent variable to include only MIDs that feature the use of force. We account for the logic underpinning the loss of strength gradient by interacting distance with absolute and relative capabilities and find that our results hold. Following Henderson and Bayer (2013), we also disaggregate Western and non-Western democracies, but do not uncover significant differences in the degree of caution displayed by these two groups. Finally, we constructed a two-stage model to examine whether democracies are more or less cautious within disputes that escalate to the level of war. We continue to find that democracies are not significantly more or less selective than non-military autocracies, and that military regimes are significantly more reckless. In addition, changing the temporal dimension of the analysis or the size of the sample does not alter the main results either.¹⁷

Conclusion

This article constructs a novel test of the conventional wisdom that democracies are more cautious than other regimes when initiating international disputes. The results produced here offer little support for this conventional wisdom. Democracies appear to be no more or less sensitive to the relative balance of capabilities than most forms of autocracy. Non-military autocracies and democracies demonstrate strikingly similar degrees of caution in selecting targets of interstate disputes. Even personalist regimes, where leaders are dependent on a very small set of supporters, seem to express some degree of care when choosing targets. Only military regimes are found to be distinct in this regard, with relative capabilities having little impact on their decisions to initiate militarized disputes.

This analysis provides some direction for further theoretical and empirical examinations of target selection behaviors in international relations. In particular, we have highlighted the need to separate the analysis of target selection from that of processes that occur once a conflict is underway. Analyses of target selection must therefore focus on ex ante characteristics of potential targets and challengers. Furthermore, our findings demonstrate that considerable variation exists among autocratic regimes. Unlike the bulk of previous research, however, we find that this variation centers on the strength of military institutions more than a leader’s consolidation of political power.

Although we have found little evidence supporting democratic target selection on observable capabilities measures, it may yet be true that democracies are selecting targets along some other criteria yet to be identified, such as the target’s perceived resolve. That is, democracies may be particularly adept at identifying some set of characteristics of potential targets — weak political leadership and/or low military professionalism — that makes victory more likely. Our focus on capabilities is motivated by a desire to find ex ante factors that are or can be observed by leaders considering the initiation of militarized disputes.

Finally, while we limit our focus to the examination of different regime types, it may be that more specific forms of domestic political institutions inform target selection behaviors. Reiter and Stam (1998, 2002), for example, argue that it may not be democracy per se that allows democratic regimes to select themselves into winnable conflicts, but the presence of a free press and a depoliticized bureaucracy. Thus, while we cannot fully rule out careful target selection as a possible explanation for why democracies rarely go to war with one another and are often victorious in the conflicts they do engage in, our work suggests the need for additional research to examine this empirical pattern more fully.

Footnotes

Acknowledgements

We would like to thank Bruce Bueno de Mesquita, Nils Petter Gleditsch, and Jessica Weeks for their helpful comments on the earlier version of this article. We are also grateful to two anonymous referees and the EJIR editorial team.

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Notes

Author biographies

Daehee Bak is an Assistant Professor of Political Science at Texas Tech University, USA. He is interested in examining the linkage between leadership tenure and international conflict, especially focusing on leaders’ foreign policy incentives and domestic constraints across different regime types.

Michael R. Kenwick is a fifth year PhD candidate in the Department of Political Science at Pennsylvania State University, State College, PA, USA. His research spans the topics of militarism, interstate conflict, alliance politics, and military intervention in civil wars. His work has been published in The Journal of Politics, and Conflict Management and Peace Science.

Glenn Palmer is Professor of Political Science at Pennsylvania State University, USA, and Executive Director of the Peace Science Society (International). His research focuses on the domestic sources of foreign policy, particularly international conflict, and updating the Militarized Interstate Dispute data set. Palmer’s more recent work has appeared in Political Analysis, International Studies Quarterly, Conflict Management and Peace Science, and Foreign Policy Analysis.

References

Anderson

Souva

(2010) The accountability effects of political institutions and capitalism on interstate conflict. Journal of Conflict Resolution 54(4): 543–565.

Andreski

(1980) On the peaceful nature of military dictatorships. Journal of Strategic Studies 3(3): 3–10.

Bas

Signorino

Walker

(2008) Statistical backward induction: A simple method for estimating statistical strategic models. Political Analysis 16(1): 21–40.

Beardsley

(2010) Pain, pressure and political cover: Explaining mediation incidence. Journal of Peace Research 47(4): 1–12.

Bennett

Stam

(1996) The duration of interstate wars, 1816–1985. American Political Science Review 90(2): 239–257.

Berry

Golder

Milton

(2012) Improving tests of theories positing interaction. Journal of Politics 74: 653–671.

Betts

(1991) Soldiers, Statesmen, and Cold War Crises (2nd edn). New York, NY: Columbia University Press.

Bueno de Mesquita

Lalman

(1992) War and Reason: Domestic and International Imperatives. New Haven, CT: Yale University Press.

Bueno de Mesquita

Morrow

Siverson

. (1999) An institutional explanation of the democratic peace. American Political Science Review 93(4): 791–808.

10.

Bueno de Mesquita

Siverson

(2003) War and the Survival of Political Leaders: A Comparative Study of Regime Types and Political Accountability. American Political Science Review 89(December): 841–55.

11.

Bueno de Mesquita

Smith

Siverson

. (2003) The Logic of Political Survival. Cambridge, MA: MIT Press.

12.

Bueno de Mesquita

Morrow

Siverson

. (2004) Testing novel implications from the selectorate theory of war. World Politics 56(3): 363–388.

13.

Carter

Signorino

(2010) Back to the future: Modeling time dependence in binary data. Political Analysis 18(3): 271–292.

14.

Cheibub

Gandhi

Vreeland

(2009) Democracy and dictatorship revisited. Public Choice DOI: 10.1007/s11127–009–9491–2.

15.

Chiozza

Goemans

(2004) International conflict and the tenure of leaders: Is war still ex post inefficient? American Journal of Political Science 48(3): 604–619.

16.

Choi

(2003) The power of democratic cooperation. International Security 28(1): 142–153.

17.

Choi

(2004) Democratic synergy and victory in war, 1816–1992. International Studies Quarterly 48: 663–682.

18.

Choi

(2012) Fighting to the finish: Democracy and commitment in coalition war. Security Studies 21(4): 624–653.

19.

Clark

Reed

(2003) A unified model of war onset and outcome. Journal of Politics 65(1): 69–91.

20.

Croco

(2011) The decider’s dilemma: Leader culpability, war outcomes, and domestic punishment. American Political Science Review 105(3): 457–477.

21.

Debs

Goemans

(2010) Regime type, the fate of leaders, and war. American Political Science Review 104(3): 430–445.

22.

Desch

(2002) Democracy and victory: Why regime type hardly matters. International Security 27: 5–47.

23.

Desch

(2008) Power and Military Effectiveness. Baltimore, MD: Johns Hopkins University Press.

24.

Downes

(2009) How smart and tough are democracies? Reassessing theories of democratic victory in war. International Security 33(4): 9–51.

25.

Downs

Rocke

(1994) Conflict, agency, and gambling for resurrection: The principal–agent problem goes to war. American Journal of Political Science 38(May): 362–380.

26.

Escribà-Folch

(2013) Accountable for what? Regime type, performance, and the fate of outgoing dictators, 1946–2004. Democratization 20(1): 160–185.

27.

Faber

Gowa

(1995) Polities and peace. International Security 20(2): 123–147.

28.

Fearon

(1994) Domestic political audiences and the escalation of international disputes. American Political Science Review 88(September): 577–592.

29.

Fearon

(1997) Signaling foreign policy interests: Tying hands versus sinking costs. Journal of Conflict Resolution 41(1): 68–90.

30.

Feaver

Gelpi

(2004) Choosing Your Battles: American Civil–Military Relations and the Use of Force. Princeton, NJ: Princeton University Press.

31.

Filson

Werner

(2004) Bargaining and fighting: The impact of regime type on war onset, duration, and outcomes. American Journal of Political Science 48(2): 296–313.

32.

Gartner

Siverson

(1996) War expansion and war outcome. Journal of Conflict Resolution 40(1): 4–15.

33.

Gartzke

(2007) The capitalist peace. American Journal of Political Science 51(1): 166–191.

34.

Geddes

Wright

Frantz

(2014) Autocratic breakdown and regime transitions: A new data set. Perspectives on Politics 12(2): 313–331.

35.

Gelpi

Grieco

(2001) Attracting trouble: Democracy, leadership tenure, and the targeting of militarized challenges, 1918–1992. Journal of Conflict Resolution 45(6): 794–817.

36.

Gelpi

Griesdorf

(2001) Winners or losers? Democracies in international crisis, 1918–94. American Political Science Review 95(3): 633–646.

37.

Ghosn

Palmer

Bremer

(2004) The MID3 data set, 1993–2001: Procedures, coding rules, and description. Conflict Management and Peace Science 21: 133–154.

38.

Gleditsch

(2002) Expanded trade and GDP data. Journal of Conflict Resolution 46: 712–724.

39.

Goemans

(2008) Which Way Out? The Manner and Consequences of Losing Office. Journal of Conflict Resolution 53(6): 771–794.

40.

Graham

BAT

Gartzke

Fariss

(forthcoming) Regime type, coalition size, and victory. Political Science Research and Methods.

41.

Greene

(2000) Econometric Analysis (4th edn). Upper Saddle River, NJ: Prentice Hall.

42.

Henderson

Bayer

(2013) Wallets, ballots, or bullets: Does wealth, democracy, or military capabilities determine war outcomes? International Studies Quarterly 57(2): 303–317.

43.

Horowitz

Stam

(2014) How prior military experience influences the future militarized behavior of leaders. International Organization 68(3): 527–559.

44.

Huntington

(1957) The Soldier and the State: The Theory and Politics of Civil–Military Relations. Cambridge, MA: Harvard University Press.

45.

Lai

Slater

(2006) Institutions of the offensive: Domestic sources of dispute initiation in authoritarian regimes, 1950–1992. American Journal of Political Science 50(1): 113–126.

46.

Lake

(1992) Powerful pacifists: Democratic states and war. American Political Science Review 86(1): 24–38.

47.

Maddala

(1983) Limited-Dependent and Qualitative Variables in Econometrics. New York, NY: Cambridge University Press.

48.

Marshall

Jaggers

(2009) Polity IV project: Political regime characteristics and transitions, 1800–2007. Available at: http://www.systemicpeace.org/polity/polity4.htm

49.

Mulligan

(2014) Restraints on preventative war before 1914. In: Levy

Vasquez

(eds) The Outbreak of the First World War: Structure, Politics and Decision-Making. Cambridge: Cambridge University Press.

50.

Murphy

Topel

(1985) Estimation and inference in “two-step” econometric models. Journal of Business and Economic Statistics 3: 370–380.

51.

Newey

McFadden

(1994) Large sample estimation and hypothesis testing. In: Griliches

Intriligator

(eds) The Handbook of Econometrics, Volume 4. Amsterdam: Elsevier Science Publishers, pp. 2111–2245.

52.

Reiter

Stam

(1998) Democracy, war initiation, and victory. American Political Science Review 92(2): 377–389.

53.

Reiter

Stam

(2002) Democracies at War. Princeton, NJ: Princeton University Press.

54.

Reiter

Stam

(2003) Identifying the culprit: Democracy, dictatorship, and dispute initiation. American Political Science Review 97(2): 333–337.

55.

Rioux

(1998) A crisis-based evaluation of the democratic peace proposition. Canadian Journal of Political Science 31(2): 263–283.

56.

Rousseau

Gelpi

Reiter

. (1996) Assessing the dyadic nature of the democratic peace. American Political Science Review 90(3): 512–533.

57.

Russett

Oneal

(2001) Triangulating Peace: Democracy Interdependence and International Organizations. New York, NY: Norton and Company.

58.

Schultz

(2001) Democracy and Coercive Diplomacy. New York, NY: Cambridge University Press.

59.

Sechser

(2004) Are soldiers less war-prone than statesmen? Journal of Conflict Resolution 48(5): 746–774.

60.

Signorino

Ritter

(1999) Tau-b or not Tau-b: Measuring the similarity of foreign policy positions. International Studies Quarterly 43(1): 115–144.

61.

Singer

Bremer

Stuckey

(1972) Capability distribution, uncertainty, and major power war, 1820–1965. In: Russett

(eds) Peace, War, and Numbers. Beverly Hills, CA: Sage, pp. 19–48.

62.

Siverson

(1995) Democracies and war participation: In defense of institutional constraints. European Journal of International Relations 1(4): 481–489.

63.

Stinnett

Tir

Schafer

. (2002) The correlates of war project direct contiguity data, version 3. Conflict Management and Peace Science 19(2): 58–66.

64.

Valentino

Huth

Croco

(2010) Bear any burden? How democracies minimize the costs of war. Journal of Politics 72(2): 528–544.

65.

Van Evera

(1999) Causes of War: Power and the Roots of Conflict. Ithaca, NY: Cornell University Press.

66.

Weeks

(2008) Autocratic audience costs: Regime type and signaling resolve. International Organization 62(1): 35–64.

67.

Weeks

(2012) Strongmen and straw men: Authoritarian regimes and the initiation of international conflict. American Political Science Review 106(2): 326–347.

68.

Weeks

(2014) Dictators at War and Peace. Ithaca, NY: Cornell University Press.