Being Gulliver: Diversionary War, Political Capital, and U.S. Intervention in Militarized Disputes. 10,957 Words

Transcription

1 Being Gulliver: Diversionary War, Political Capital, and U.S. Intervention in Militarized Disputes 10,957 Words

2 2 Abstract How do public evaluations of recent international conflict performance affect leaders decisions to intervene in additional foreign military conflicts? We utilize Prospect Theory to construct a theoretical synthesis of two seemingly opposed approaches to this question: the political-capital and diversionary-war frameworks. We find substantial support for this Prospect-Theory based synthetic framework through an analysis of decisions of U.S. presidents to intervene in militarized interstate disputes between 1948 and Our analysis makes two important contributions to our understanding of military intervention. First, we show how Prospect Theory can help to identify the conditions under which high or low levels of domestic approval for recent performance of U.S. presidents in militarized disputes translates into a greater or lesser propensity of those presidents to join new conflicts. Second, we undertake a comparative assessment of the impact of international and domestic conditions on the decisions of U.S. presidents to intervene in militarized disputes.

3 3 Introduction Since the outset of the 20 th century U.S. policy makers, opinion leaders, and members of the public have debated whether to intervene in overseas military conflicts. The eventual collapse of American public support for the war in Vietnam in the late-1960s and the Iraq War in the 2000s had devastating effects on the Lyndon Johnson and George W Bush administrations. The centrality of maintaining public support in order sustain military interventions and continue to govern effectively suggests that leaders should try to anticipate public responses to military action. The continuing centrality of public opinion on this matter has been highlighted during the current military conflicts in Syria and Ukraine, where reporting on public attitudes toward intervention has been almost as prevalent as has attention to the strategic wisdom or likely success of U.S. intervention. But how precisely does the anticipation of domestic political support shape intervention in military conflicts? Much of the media coverage on intervention assumes, and some scholarly works suggest, that public approval affords leaders the political capital necessary to carry out risky and potentially costly military action. An implication of this model is that a lack of public support will constrain intervention. On the other hand, an extensive academic literature on diversionary war suggests exactly the opposite effect. That is, the diversionary perspective suggests that a lack of popularity may propel presidents to intervene in an effort to gamble for resurrection. We argue that insights from Prospect Theory (Kahneman and Tversky 1979; Levy 1992; Levy 1997; McDermott 2001; Mercer 2005) on risk propensity can unify these seemingly opposed mechanisms into a unified model of public approval and military intervention. Our theoretical model should apply to any state leader facing the question of intervention in the context of a domestic constituency. However, we argue that U.S. military intervention behavior since World War II represents an especially promising empirical testing ground for our argument. Our analysis of U.S. intervention in militarized disputes between 1948 and 2002 provides substantial support for our integrated model of domestic approval and military intervention. Our analysis makes two important theoretical and empirical contributions to our knowledge about domestic political conditions and foreign military conflict. First, we identify the international conditions under which domestic assessments of leadership performance play a

4 4 greater or lesser substantive role in shaping leaders decisions to use military force. Second, within those constraints we identify the conditions under which high or low levels of domestic approval may translate into a greater propensity to use military force. The Diversionary-War and Political-Capital Logics for the Use of Force The literature on diversionary war suggests that leaders who face political, social, or economic difficulties at home will seek to divert domestic attention away from those difficulties by engaging in military conflicts abroad (Levy 1988, 1989). Specifically, such leaders may use military force in order to provoke a rally round the flag effect as a way of gambling for resurrection (Downs and Rocke 1994; Goemans 2000; Goemans and Chiozza 2011). Formal representations of this problem suggest that a foreign victory allows leaders to demonstrate competence and thus regain the support of citizens (Richards, Morgan, Wilson, Schwebach, and Young 1993; Smith 1996). Empirical studies by Gelpi (1997), Davies (2002), and Pickering and Kisangani (2005) have found that democratic leaders in general are more generally susceptible to such diversionary uses of force. And more specifically, Ostrom and Job (1986), James and Oneal (1991), Hess and Orphanides (1995, 2001), and Fordham (1998a) have reported that diversionary-war dynamics help account for U.S. military behavior. Similarly, Clark (2003) finds that deterioration in U.S. economic conditions is associated with a heightened risk that U.S. presidents will employ force abroad, even after taking into account the efforts of potential foes to avoid being targeted (Smith 1996, 1998), and differences in the preferences of Republicans and Democrats for fighting inflation as opposed to promoting employment (Fordham 1998b). Morgan and Anderson (1999) find similar evidence in the United Kingdom. Yet, a substantial number of studies fail to find evidence of diversion. For example, Meernik and Waterman (1996) and Meernik (2000) find little systematic empirical support for the diversionary-war thesis in the U.S. case, and Mitchell and Moore (2002) find that domestic conditions, while a partial explanation for the U.S. use of force, take a back seat to international conditions in determining when such force is employed. In the same vein, with one important exception, to which we will return in a moment, Lian and Oneal (1993) do not find that U.S. presidents systematically enjoy significant improvements in public approval polls if they engage in external military conflicts. Looking beyond the U.S. case, Leeds and Davis (1997) find little evidence that democracies are prone to diversionary wars, and Goemans (2000) finds that, during

5 5 World War I, there was greater gambling for resurrection by Germany than by Britain and France. Indeed, in direct opposition to diversionary war theory, Oneal and Tir (2006) find that the United States seems to be more likely to use force when economic conditions are especially strong. 1 Thus, while the literature on diversionary incentives is well developed and substantial, the empirical evidence suggests that other mechanisms may be at work as well. In particular, some scholars have suggested that a diametrically opposed mechanism links domestic support to the use of force: political capital. That is, elected officials require a reservoir of public support in order to undertake costly or risky policy initiatives. If they enjoy that support, they may be more likely to employ force abroad, and when they lack that support they may be less willing to do so. For example, Ostrom and Job (1986: 549) suggest, that it can be expected that, all other things being equal, the president will act when he perceives he can afford to lose or when he possesses a "popularity buffer." Conversely, they suggest that [p]residents with relatively low levels of popular support, and in turn congressional support, will tend to become immobilized." 2 A variety of empirical studies lend credence to this political-capital framework. As noted above, Oneal and Tir (2006: 769), find that U.S. presidents are more likely to use military force abroad in a context in which there has been positive recent economic performance. They interpret this association as suggesting that presidents are more likely to use force when they are especially popular by virtue of that favorable domestic environment. Ostrom and Job (1986), Morgan and Bickers (1992), and Morgan and Anderson (1999) find similar evidence of a propensity by U.S. presidents to use military force when their domestic popularity is high. In the same vein, students of American politics have found that presidents with political capital (Light 1999) specifically, high levels of public approval, large margins of electoral victory, or high levels of party-based support in Congress have marked advantages in pressing their domestic policy agendas (also see Johnson and Roberts, 2004, 2005). 1 However, Oneal and Tir (2006) do find evidence of diversion among immature democracies. 2 Ostrom and Job (1986) also suggest that presidents may be more inclined to use force as their popularity declines. Combining their two arguments suggests a convex curvilinear relationship between domestic support and the use of force. Our combination of the diversionary and political capital arguments, on the other hand, suggests a concave curvilinear relationship. See also Pickering and Kisangani (2005:27).

6 6 A Unified Model of Public Support and Leadership Decisions to Use Force The diversionary-war and political-capital models produce highly dissimilar predictions about the conditions under which leaders might be especially prone to join a new foreign militarized dispute. Indeed, the two models would seem to be irreconcilable, and the theoretical tensions between these two models would seem to explain the mixed and sometimes contradictory empirical findings regarding public opinion and the use of force. We contend, however, that a third theoretical perspective can integrate these dissonant models of elite responses to public opinion and yield an internally consistent set of expectations about the impact of popular approval on leader s propensities to use force. Moreover, we contend that this unified theoretical model can bring the diverse and seemingly contradictory empirical findings into coherent focus as well. The key insight linking these apparently contradictory models is that both of these arguments conceptualize the relationship between public approval and the use of force in terms of leaders responses to decisions under risk. Both models view the decision to use force as risky because the public s reaction to conflict can be difficult anticipate and may vary widely from extremely supportive to extremely critical. The distinction between the political-capital and diversionary models, however, concerns their assumptions about leaders attitudes toward risk. The political-capital argument assumes that leaders will be risk averse and will be concerned about a potentially negative popular response. As a result, this model expects that leaders will only undertake the use of force when they have a sufficient buffer of public support that will allow them to weather a hostile public reaction. The diversionary argument, on the other hand, assumes that leaders will be risk acceptant and will be focused on the potential benefits of a public rally round the flag (Richards et. al 1996; Smith 1996). As a result, this model expects leaders will undertake the use of force in the hopes of a boost in support. But while the political-capital and diversionary models make broad uniform assumptions about leaders risk propensities, Prospect Theory provides us with a theoretically coherent and empirically validated mechanism for identifying the conditions under which we should expect leaders to be risk averse or risk acceptant (Kahneman and Tversky 1979; Levy 1992; Levy 1997; McDermott 2001; Mercer 2005). As a result, Prospect Theory can integrate these two contrasting mechanisms and provide a single coherent set of expectations about the relationship between popular approval and the use of military force. Specifically, Prospect Theory holds that

7 7 individuals will be risk averse when they view themselves as choosing within the domain of gains relative to some reference point. When they see themselves as choosing within the domain of losses relative to that reference point, on the other hand, leaders will become risk acceptant in order to avoid potential losses. A key difficulty in applying Prospect Theory to situations outside an experimentally controlled laboratory is the identification of the circumstances--or, more precisely, the reference point--in connection to which a leader in a democratic polity, such as a U.S. president, defines himself or herself as moving from a position of the domain of gains to the domain of losses (Levy 1997:100; Mercer 2005:3). There are many such reference points that might reasonably be expected to act as such a political reference point for democratically elected national leaders. For a U.S. president, for example, it might be the percentage of the popular vote he or she attained at election time; it might be a running average of the approval rating that president has had since election time; it might be a sense of how well some number of presidents have done in recent memory; or it might be a readily identified focal point such as 50% approval. The subjective construction of the domains of gains and losses implies that it will be impossible to identify a single threshold at which all leaders will switch domains. However, we can say that below some threshold all leaders will eventually view themselves as acting within the domain of losses. Moreover, we can say that leaders will become increasingly likely to view themselves in the domain of gains as their level of popularity increases. Consequently, we can use Prospect Theory to unify the political-capital and diversionary frameworks, because this theory implies that leaders will tend to be risk averse - and therefore be guided by the politicalcapital mechanism - when their approval rating is high. But leaders will be increasingly likely to be risk acceptant - and therefore guided by the diversionary mechanism - as their approval decreases. Rather than a strictly linear relationship between approval and the propensity to use force, this unified perspective would predict a U-shaped relationship, depicted by the black line in Figure 1, where the propensity to use force decreases as approval rating decline from high levels, but then eventually increase as approval ratings continue to drop and leaders become risk acceptant. Our central contribution to this analysis of public approval and leadership decisions to use force abroad is to suggest that there may be a concave curvilinear relationship between domestic approval and military intervention. However, we wish to emphasize that there are theoretical reasons to anticipate that this association may only operate under particular conditions relating to the type of overseas conflict in which the leader is considering intervention.

8 8 Specifically, the gambling for resurrection mechanism posited by the diversionary-war model requires that the resurrecting use of force be completed successfully and at low cost (Levy 1988). Thus, we may expect the impact of the diversionary mechanism to depend on the military capacity of the states already involved in the dispute. When the existing disputants are militarily weak, leaders may believe they can achieve success at low cost. However, when at least one of the disputants is militarily more substantial then success will seem more costly and risky, making the gamble for resurrection seem less attractive. The political-capital argument, on the other hand, does not seem as likely to be conditioned by the power of the states involved in the dispute. In fact, the logic of the political-capital argument specifically focuses on the ability to undertake costly or risky policy initiatives and so seems likely to hold regardless of the states engaged in the dispute in question. Different presidents may code disputants as weak or strong in different ways, and it is difficult, in a manner similar to defining gains/losses reference points ex ante, to know what is the threshold of power that produces a disputant-power characterization by U.S. presidents. We have, in our statistical modeling, employed various thresholds for identifying disputants as strong or weak. In general we find that our results hold across a variety of thresholds between large and small disputants. That is, while a concave curvilinear association between domestic approval and military intervention obtains with respect to conflicts that involve relatively weaker participants, a positive linear association--as depicted by the grey line in Figure 1--will operate with conflicts that are constituted by relatively more powerful participants. In sum, the predictions from our unified model of public approval and the use of force are depicted in Figure 1. For disputes involving relatively militarily weak states, we expect the kind of curvilinear relationship that is depicted by the black line in Figure 1. With respect to disputes involving at least one relatively strong state, on the other hand, we expect to see a more linear positive association as depicted by the grey line in Figure 1. Figure 1 About here Data and Methods Our argument about domestic support and the use of force should apply, at least in principle, to any state leader who is constrained by a domestic constituency. However, a number of logistical constraints make it difficult to assess this expectation in a variety of circumstances. First, one must be able to identify the constituency or selectorate (Bueno de Mesquita et. al.

9 9 1999) to which the leader is held accountable. In democratic states this process is fairly straightforward, but the relevant constituents in an autocratic regime may be difficult to identify. Second, one must be able to measure the leader s level of support within his or her constituency. Polling data generally provide a good measure of this support for democratically elected leaders, but states vary widely in the prevalence of polling data. Third, we must have a large number of opportunities to use force and a fairly large number of actual uses of force in order to estimate a reliable statistical model of a state s militarized conflict behavior. Finally, and as discussed above, we have theoretical reasons to distinguish between militarized disputes that involve weak states and those that involve at least one powerful state. Given these constraints, we suggest, in the first place, that the United States in the post- World War II era represents a highly appropriate testing ground for our model, for it meets each of our criteria for testing our argument, and it has the added advantage of being substantively important. As we noted above, the United States is featured frequently as a case study or motivating example in the literature on external intervention and electoral incentives for war. Moreover, numerous studies of domestic politics and the use of force focus specifically on the United States (for example, as noted above, Ostrom and Job 1986; Morgan and Bickers 1992; Morgan and Anderson 1999; and Oneal and Tir 2006). In addition, the availability and transparency of public opinion for the United States over this period is high. Finally, the United States is among the most conflict-prone states in the world. According to the Militarized Dispute Dataset, the United States has been involved in more than 5% of all the dispute-years across the entire international system. This rate of conflict involvement far outstrips even major power rivals such as the People s Republic of China and Russia / The Soviet Union. Only Israel and Jordan are involved in more dispute years than the United States. In principle, we also expect our model of domestic approval and the use of force to apply to various kinds of decisions to use force, including both the initiation of conflicts and the decision to join ongoing conflicts among other states. However, aggregation problems and the potential for simultaneity bias plague the impact of public approval on dispute initiation. Models of dispute initiation are generally constructed at the annual or dyad-year level because of data availability. Approval data, on the other hand, is available on a monthly or even weekly basis, and may vary quite substantially within a year. Moreover, while the onset of a military conflict might give us a specific date for measuring approval, it is not clear when one should measure approval in years when no conflict occurs. Decisions not to initiate a conflict are presumably occurring constantly at least by default and so aggregating over an entire year will necessarily include

10 10 levels of support that occurred after the decision not to use force. One could, of course, avoid problems of simultaneity bias by lagging approval by one year, but this procedure is also problematic because approval can change so dramatically over such a long period. Thus the average level of approval during the previous year may have little bearing the actual level of approval that a leader is facing at the time he or she considers the use of force. Decisions to join ongoing military conflicts, on the other hand, present substantially fewer measurement and methodological problems. The onset of the conflict provides a date for measuring approval accurately, and the conflict represents a clear opportunity to use military force. 3 Focusing on American dispute intervention still provides us with a substantial number of cases to evaluate. No other state approaches the rate of American intervention in military conflicts. For example, of the 1,779 MIDs that occurred between 1946 and 2010, there are 374 external interventions across 137 unique disputes: the United States accounted for 39 (11.2%) of those cases of intervention, while Britain, the Soviet Union, and France, accounted, respectively, for 24 (6.4%), 21 (5.6%), and 15 cases (4%). Moreover, the frequency of American intervention in militarized disputes combined with the frequent policy debates regarding American intervention around the world demonstrate that dispute intervention is a substantively important topic in its own right as well. As a result, in our empirical analysis below we examine the propensity of the United States to join ongoing militarized disputes. The universe of cases draws from the Militarized Interstate Disputes v4.01 dataset (Palmer et al. 2015), and consists of 1,370 disputes between 1948 and 2002 in which the United States was not an original participant that is, militarized disputes that the United States had the opportunity to join or not to join. 4 Our dependent variable 3 One implication of this research design is that we assume that the onset of militarized disputes between other states is not contingent on domestic approval of the president. While some literature suggests that leaders may account for presidential approval when choosing whether to challenge the United States directly (Brule et. al. 2010), we find it unlikely that leaders will consider presidential approval when targeting third-party states. 4 Restricting our analysis only on intervention in MIDs that escalate to the use of force (level 3) reduced our N by about 50% to 689 cases and yielded similar results. Restricting our analysis to MIDs that resulted in at least on fatality substantially reduced our sample size by about 75% to only 352 cases (261 small and 91 big MIDs). In this case our estimated coefficients for the

11 11 is simply coded as whether the United States is listed as a MID participant but not as an originator, regardless of which side they joined. 5 The United States decided to join 39 of these disputes, for an overall joining rate of 2.65%. As noted above, we believe it is important to distinguish between militarized disputes with weak participants and those with at least one powerful participant. 6 Figure 2 presents the distribution of the maximum CINC score for the participants in each of the 1,370 MIDs other than the United States. CINC scores measure the proportion of the total military capability in the world that each states possess during a particular year. First, it is worth noting that the distribution is highly skewed, with majority of the disputes only involving states with very low CINC scores. The distribution in Figure 2 presents several potentially salient divisions between what might be termed big MIDs (that is, MIDs with at least one powerful participant is involved) and small MIDs (those in which only weak states are participants). There are very few MIDs involving states with maximum CINC scores just under 0.05 (i.e. 5% of global military capability), and the density function drops all the way to zero at about before moving back up at somewhat higher levels. (As a point of reference, U.S. CINC scores range between about 0.15 and 0.3 over this period). Based on this distribution we categorized MIDs involving at least one state with a CINC score of greater than as a big MID, while disputes in which no state rises to the threshold are categorized as small. Any specific dividing line is, of course, inevitably arbitrary. However, we undertook a number of robustness tests that indicated that any dividing line between impact of domestic approval retained the same direction, but their statistical significance was reduced to the p<.15 and p<.3 levels respectively. 5 Note that the unit of observation is the dispute, not the dispute-year. For MIDs that spanned multiple years, the start date of the dispute is used as the reference point for all yearly variables. 6 An alternative specification would model U.S. intervention with a Boolean logit or probit estimator (Braumoeller 2003; Braumoeller and Carson 2011), which would simultaneously estimate multiple causal pathways to intervention. However, because U.S. intervention in MIDs is a relatively rare event, we did not have sufficient data to allow the estimator to converge. Thus we conduct separate analyses of big and small MIDs. Since both the diversionary and politicalcapital mechanisms may be working in either of these contexts, we continue to allow for the possibility of a curvilinear impact for foreign policy support in each of the analyses

12 12 approximately and yielded very similar results. The threshold divides our dataset into 1,017 small MIDs and 353 big MIDs. Overall, we find that the United States is somewhat more likely to intervene in big disputes relative to small ones. Specifically, the United States intervenes in 22 of the small MIDs (a 2.16% intervention rate), and 17 of the big MIDs (a 4.82% intervention rate). Figure 2 About Here Our primary independent variables are two distinct measures of public support. As is common in the literature on both political-capital and diversionary conflict, our first measure of support is the president s overall job approval rating. We rely on Gallup s data for approval because it is consistently available over time and allows us to account for house effects, etc. We calculate overall approval as the average monthly approval rating for the current president over the nine months prior to the start date of the MID. Robustness checks indicated that the results remain similar for any aggregation of foreign policy support ranging from six to twelve months. However, the average over the previous nine months provided the best fit to the data. 7 Our second measure of public support is an issue-specific indicator that focuses on popular evaluations of the president s handling of recent interstate conflicts. We believe that this issue-specific measure of approval may be more influential with regard to intervention decisions because it reflects the president s level of support on the policy issue at hand, the use of force. The literature on priming and Presidential approval, for example, notes that media attention shifts the weight that individuals place on issues when evaluating the President (Edwards, et. al. 1995; Miller and Krosnick 2000). Thus according to the political-capital model, Presidents with high ratings for their handling of interstate conflict may feel a stronger buffer of support and therefore 7 Some analyses of diversionary incentives focus on other measures of domestic dissatisfaction such as unemployment or inflation. These proxy measures are generally used where direct measures of approval are not available. Public approval represents the causal mechanism or intervening variable through which variables such as aggregate economic performance should shape presidential behavior. Thus we would not expect such antecedent variables to remain significant once we account for the impact of approval. In auxiliary analyses we included a lagged indicator of the misery index (a combination of inflation and unemployment). Its coefficient was not statistically significant, and our results regarding approval remained unchanged.

13 13 feel less constrained with regard to initiating a new conflict. With regard to the gambling for resurrection mechanism, on the other hand, public judgments of a leader s incompetence in foreign policy may cause to use military force in order to create an opportunity to reverse that judgment (Smith 1996). 8 Our measure of the public s perception of the president s performance in interstate conflict draws on polling data collected from the Roper Center for Public Opinion Research database. Specifically, we searched the database at the Roper Center for Public Opinion Research at the University of Connecticut Database for all survey questions that were fielded to a national random sample of adults relating to U.S. involvement in each MID in which the United States was involved between 1948 and After compiling all questions referencing U.S. participation in a militarized dispute, we selected only those that asked respondents for their approval of the mission involving the MID, their approval of the president s handling of the mission, or their judgments about the success of the mission. We excluded questions that asked for hypothetical judgments about future events or potential policy choices. For each of the remaining questions we divided the response categories into those that were supportive of the president, those that were opposed to the president, and those that were neutral (e.g. not sure, don t know, etc.). We dropped neutral responses and recorded the proportion of the positive or negative responses that were supportive of the president. 9 For each observation in our dataset, we measure presidential foreign policy approval as the average approval rating over all relevant polls questions regarding the handling of militarized disputes in the nine months prior to the start date of the MID. 10 Relevant polling questions are 8 This conceptualization of domestic support differs from many analyses of diversionary war because it does not imply that leaders are distracting constituents from the issues that created their disapproval of the leader. We feel it is appropriate to test this kind of measure of domestic approval because we are building a broader model of domestic support and the use of force than is implied by diversionary arguments. 9 Analyses that included neutral responses (effectively counting them as opposition to the President) yielded similar results, but dropping the neutral responses provided a slightly stronger fit. 10 Data on overall approval of the president s handling of foreign policy are not available with any regularity prior to the 1980s, and our measure focuses on approval of the president s handling of specific military conflicts rather than overall foreign policies.

14 14 those that satisfy two conditions: (1) they must assess the public s approval of the United States performance in a prior MID as described above; and (2) they must have been fielded during the current president s term in office. 11 For instance, a MID initiated in June 1993 would have a foreign policy approval rating of 45.6%, which is the average of 22 relevant polls between September 1992 and May An additional 16 polls are excluded from this average, because they were fielded during the elder Bush s presidency. As was the case with overall approval, we measure the president s average monthly approval for the handling of MIDs over the 9 months prior to the new MID. Aggregations ranging from 6 to 12 months yielded similar results. Overall measures of presidential approval are widely used and require no further description, but since our measure of approval of the President s handling of MIDs is original to this work, we provide some further description of these data. Figure 2 displays the levels of MID approval that we observe during the 9 months prior to each of the 1,061 MIDs for which we have relevant polling data. In addition to the histogram displaying how frequently we observe each level of approval, the figure also depicts a normal distribution for reference as well as a kernel density estimate. The median level of MID approval in the 9 months leading up to the onset of a MID between 3 rd parties is 52%, and just under 60% of the observations fall between 40% and 60% approval of the President s handling of ongoing MIDs. While these data do not conform precisely to a normal curve, the kernel density estimate indicates that the distribution is quite close to normal. The distribution of MID approval is also quite similar in the aggregate to the distribution of overall approval, which has a median of 55% with 60% of the observations between 45% and 65%. Perhaps somewhat surprisingly, however, the correlation between overall approval and approval of the President s handling of MIDs is only 0.26 (p<.01). Thus while these two measures of approval are clearly related to one another, they are both conceptually and empirically distinct, and our model should have little difficulty in distinguishing their effects. Figure 3 About Here To account for new presidents and other observations that lack relevant polling questions, we include in our models a dummy variable that indicates if there was any MID polling data. The 11 An important assumption of this coding scheme is that newly inaugurated presidents essentially have a blank slate in terms of foreign policy approval.

15 15 United States was involved in at least one MID during every year between 1948 and However, these crises were not all salient enough to generate polling questions from the media about the dispute. Thus we created a dummy variable identifying years in which the United States is involved in a MID that generates at least one polling question. Including this variable in our analysis accounts for the fact that we are measuring the impact of approval contingent on a question being fielded to a national sample of adults. 12 To allow for a curvilinear relationship and properly test our hypotheses, we also include a squared transformation of both of our measures of domestic support. We also attempt to control for the public s war weariness as a potential determinant of the president s decision to intervene in an ongoing MID (Levy and Morgan 1986). We measure the logged number of U.S. military casualties in the previous year, as well as the number of MIDs in which the United States participated in the previous year. Of course, military intervention is not entirely a function of domestic political calculations. For example, realist scholars have argued that states intervene to advance national interests and affect the balance of power in their favor. During the Cold War, for instance, the fear of open war leading to nuclear annihilation forced the United States and the Soviet Union to use interventions and proxy wars to contest their rivalry (Morgenthau 1967). Other scholars have argued for the importance of strategic alliances (Tillema 1994), geographic proximity (Pearson 1974), the distribution of power in the international system (Young 1968), and the existence of other potential interveners (Gent 2007). Alternatively, liberal perspectives have examined dyadic democracy (Kegley and Hermann 1995, 1996) and economic interdependence (Fordham 2007; Aydin 2008) as determinants of intervention, while constructivists have highlighted the role of norms like just war doctrine (Butler 2003). 12 During years in which there is no polling on US MIDs, we code support for the president s handling of MIDs as zero. This allows the impact of foreign policy support to drop out of the model when it is not measured while not simply dropping those cases from the analysis. The inclusion of the foreign policy poll dummy variables allows the model to reset the constant term for cases in which polling questions were asked about MIDs. This specification allows us to estimate the impact of foreign policy support on the 1,041 MIDs where that level of support was measured, while estimating the impact of the rest of the variables on the 1,371 MIDs in the entire dataset. Our results regarding the impact of support for the president s handling of MIDs are identical if we simply drop the cases in which no polling exists on US MIDs.

16 16 Thus, in addition to our domestic political variables, we also include a number of international-level variables that may influence U.S. intervention in militarized disputes. These variables describe the characteristics of the MID participants and their relationships with the United States. The number of MID participants ranges from 2 to 29, with the distribution strongly skewed to the lower end of this range. Interstate distance data was generated using the EUgene v3.204 software package (Bennett and Stam 2000) and is measured as the straight-line distance between national capitals. In our models, we use the logged distance between the United States and the nearest MID participant as a control variable. Bilateral trade data is drawn from the COW Trade Data Set v3.0 (Barbieri and Keshk 2012; Barbieri, Keshk, and Pollins 2009). We measure the United States trade dependency on a state as the sum of imports and exports as a proportion of U.S. gross domestic product (GDP). We include in our models the maximum value of U.S. trade dependency across all MID participants. Data on regime types is drawn from the Polity IV dataset (Marshall, Gurr, and Jaggers 2013). Regime type is measured using the polity2 variable, which is a 21-point scale from most autocratic to most democratic. Similar to U.S. trade dependence, we include in our models the maximum value of polity scores across all MID participants. State capabilities are measured using Composite Index of National Capabilities (CINC) scores from the COW National Material Capabilities v4.0 dataset (Singer 1987). We control for the maximum CINC score among all MID participants. Alliance data is drawn from COW Formal Alliance v4.1 dataset (Gibler 2009). We include in our models two dummy variables that indicate whether any MID participant has (1) a defensive pact or (2) another type of alliance with the United States. Results and Discussion Table 1 presents our analysis of U.S. intervention in small and big militarized disputes. Turning first to our analysis of small MIDs, we find strong support for the impact of popular support for the president s handling of current MIDs on his decision to intervene in a new MID. Moreover, as anticipated by our unified model, we find support for the operation of both the diversionary and political-capital mechanisms. Specifically, the coefficient for foreign policy support is negative while the coefficient for foreign policy support squared is positive, and both coefficients are statistically significant. This combination of coefficients indicates that at low levels of foreign policy support, we observe an inverse relationship between support and U.S.

17 17 intervention. That is, as support drops to low levels the probability of intervention increases. This result is consistent with the diversionary incentives discussed above. On the other hand, the positive coefficient for the squared term indicates that as foreign policy support increases, its impact on the likelihood if intervention begins to reverse. This result is consistent with the political-capital mechanism. Table 1 About Here Thus, Prospect Theory helps us identify the conditions under which the diversionary and political-capital mechanisms shape presidential interventions in small MIDs. Specifically, the concave curvilinear relationship that we observe between public approval of presidential handling of recent MIDs and the propensity to intervene in new MIDs indicates that presidents shift their behavior from risk-acceptant to risk-averse as their level of such approval increases. Interestingly, while approval of the president s handling of recent MIDs has a significant impact on American intervention, the coefficients for the president s overall approval rating and his approval rating squared do not approach statistical significance. These results support the potential importance of issue specific measures of approval and suggest that presidents may be aware of and may respond to public evaluations of their competence on specific issues. With regard to our control variables, we see that international-level variables in general do not affect U.S. decisions to intervene in small MIDs. Specifically, distance, U.S. trade dependence on disputants, the presence of U.S. casualties in the prior year, and ongoing U.S. participation in other MIDs have no significant impact on the decision to intervene. The maximum CINC score also has no significant effect, but it is important to note that this coefficient only captures variation in maximum CINC scores within the category of small disputes. The only international factors that do appear to influence U.S. decisions to intervene in small MIDs is the presence of a defense pact between the United States and one of the MID participants, and the regime type of the participants. As expected, the presence of a defense pact increases the probability of U.S. intervention. However, democracy appears to have the opposite effect: the United States is less likely to intervene in small MIDs that involve at least one relatively democratic country. This result is inconsistent with our expectations. It may be a

18 18 function of the Cold War period, when U.S. presidents were more willing to intervene on behalf of autocrats friendly to U.S. interests even at the expense of local democracy Our analysis of big MIDs, on the other hand, reveals a very different pattern of presidential behavior. First, the coefficient for public support for the president s handling of recent MIDs is positive and statistically significant, while the coefficient for support squared is negative but does not achieve statistical significance. This combination of results indicates that, U.S. public support for the president s handing of recent MIDs is positively associated with intervention in a new MID when the support for recent MID performance is at relatively low levels. That is, presidents are constrained from intervening in big (and potentially more costly and less successful) MIDs when their recent handling of military conflict is unpopular with the public. The negative coefficient on the squared support term, on the other hand, indicates that this constraining effect decreases as foreign policy support increases, which suggests that higher levels of support for past presidential MID performance may not translate into a higher U.S. propensity to intervene. However, this coefficient appears to be substantively smaller than squared term in the model of small MIDs, and it is not quite statistically significant. Overall, the relationship between foreign policy support and intervention appears to be roughly linear and risk-averse with regard to big MIDs. Notice that, once more, the coefficients for overall presidential approval and presidential approval squared are near zero and do not approach statistical significance. This result underlines the fact that our observed domestic political effects are a consequence of the president s handling of military conflict, and not simply an artifact of overall approval. Additionally, our analysis of U.S. intervention in big MIDs indicates that a number of international and strategic factors have a substantial impact on presidential calculations. Specifically, the maximum CINC score of the participants in the dispute has a strongly positive impact on the likelihood of intervention, and trade dependence on one of the participants in the MID also makes the United States significantly more likely to intervene. The coefficient for U.S. involvement in other ongoing MIDs is positive and statistically significant. We had expected that involvement in other MIDs might deter intervention in new ones, however, this result seems to indicate that U.S. involvement in MIDs begets additional involvement. This process could be due

19 19 to spillover or contagion effects of specific MIDs or could indicate the impact of broader systemic factors that are driving the United States to become involved in conflict. 13 Finally, U.S. alliance commitments continue to have a significant impact on intervention decisions in big MIDs, but the pattern of effects is more complicated. Specifically, the presence of a defense pact between the U.S. and one of the disputants has a negative impact on the probability of intervention, but the presence of other types of alliances ties (e.g. an entente or a non-aggression pact) increases the likelihood of intervention. The result concerning U.S. defense pacts is not simply a result of colinearity in the model between defense pacts and other alliances. For example, a bivariate analysis indicates no relationship between U.S. defense pacts and intervention in big MIDs. Thus it seems more likely that this result is due to the fact that powerful states account for these defense pacts when making their decisions to initiate MIDs (Fearon 1994; Smith 1996). Weaker alliance ties, however, signal some increased likelihood of U.S. intervention but in big MIDs but do not create such strong selection effects. The coefficients in Table 1 provide solid support for our unified models of the impact of public approval on presidential decisions to intervene in militarized disputes. Both the diversionary-war and political-capital mechanisms appear to influence presidential decisionmaking in different contexts, as anticipated by Prospect Theory. The analyses in Table 1 also provide support for the distinction we draw between big and small disputes, since the causes of intervention seem to vary across these categories. In particular, while intervention in small MIDs appears to be primarily a function of domestic politics, intervention in larger disputes appears to be driven by a mixture of domestic and international factors. However, we cannot fully evaluate the support for our hypotheses without investigating the predicted substantive effects. In particular, since foreign policy support appears to have curvilinear effects (at least with regard to small MIDs) we need to evaluate the net impact of foreign policy support on intervention behavior as it moves from low to high levels. Table 2 presents the estimated marginal impact of changes in public support for the president s handling of recent militarized disputes on the likelihood of U.S. intervention. Support for the president s handling of MIDs ranges from 21.6% to 85%, but not surprisingly the data 13 The most obvious candidate here would be the Cold War rivalry, but data limitations prevent us from estimating separate Cold War and post-cold War models.

20 20 become quite thin at the extremes of both tails of the distribution. To illustrate the curvilinearity of the estimated effects across a reasonable range of public approval for recent presidential MID performance on the likelihood of intervention in a new MID, we evaluate the impact of foreign policy support for recent MID performance as it ranges from 25% to 50%, and from 50% to 75%. The table presents the predicted change in the probability of U.S. intervention, the 95% confidence interval around that change, and the relative risk of U.S intervention. Relative risk of intervention is a useful metric in this context because intervention is a relatively rare event and it is important to view the changes in predicted probabilities within the context of the low base-rate of U.S. intervention. Relative risk is calculated as the probability of the event in the treated category divided by the probability in the control category. Thus, a relative risk score of 2.0 indicates that the probability of an event has doubled, while a relative risk score of 0.5 indicates that the probability of an event has been cut in half. Table 2 About Here The first half of Table 2 presents the impact of foreign policy support on intervention in small MIDs. Here we can see that a shift in support for the president s handling of MIDs from 25% to 50% decreases the probability of U.S. intervention from 10.2% to 1.2%. This steep drop in the probability of intervention is statistically significant at the 0.05 level. The relative risk score of 0.12 indicates that the risk of U.S. intervention in a small MID drops by 88% as the president s foreign policy support increases from 25% to 50%. This result is strongly consistent with the diversionary-war mechanism. Presidents with very low foreign policy approval scores are substantially more likely to gamble for resurrection by intervening in minor disputes. But, as anticipated by Prospect Theory, this risk acceptant behavior disappears as their foreign policy approval ratings approach moderate levels. As presidential foreign policy approval continues to increase, however, its impact on U.S. intervention in small MIDs reverses direction. Specifically, at approval of the president s handling of ongoing MIDs increases from 50% to 75%, the probability that the United States will intervene increases from 1.2% to 3.4%. This increase of 2.2% is not nearly as steep as the decline in the probability of U.S. intervention as approval moved from 25% to 50%, but the marginal impact remains substantial. For example, change in relative risk associated with this increase in foreign policy support is 2.8, which indicates that the president is nearly three times as likely to intervene in a small MID when his foreign policy support is 75% as opposed to 50%. As the

21 21 confidence interval in Table 2 indicates, this increase in propensity to intervene falls just short of significance at the 0.05 level. This result is consistent with the political-capital mechanism. That is, presidents feel emboldened to act when the public is strongly supportive of their handling of military conflicts, but they are constrained as public support drops to middling levels. The second half of Table 2 displays the estimated marginal effects of changes in foreign policy support on presidents propensity to intervene in big MIDs. While we observed that low levels of approval led to gambling for resurrection through intervention in small MIDs, we see the opposite mechanism at work with regard to big MIDs. Specifically, an increase in foreign policy support from 25% to 50% increases the probability that the United States will intervene in a big MID from 0.2% to 1.5%. This increase is statistically significant at the.05 level, and while a 1.3% increase might not appear large at first glance, this change represents more than a sevenfold increase in the probability of intervention. Once again, this result is strongly consistent with the political capital mechanism. Presidents with very low levels of support for their handling of militarized disputes are constrained from intervening in larger MIDs because of the potential costs of conflict and the greater opportunity for failure. Thus while a president with 25% foreign policy support has a 10.2% probability of intervening in a small MID, the probability that a president in the same circumstance would intervene in a big MID is fifty times lower at 0.2%. Finally, as foreign policy support continues to increase from 50% to 75%, the probability of U.S. intervention in big MIDs continues to increase, but at a more moderate pace. Specifically, the probability of intervention increases from 1.5% to 2.5%. However, the 95% confidence interval for this change ranges from -2.6% all the way to 9.0%, so the change in probability does not approach statistical significance. The change in relative risk is also much substantively much smaller. While the increase in foreign policy support from 50% to 75% increased the likelihood of intervention in small MIDs nearly three-fold, this same change in support only raises the risk of intervention by a factor of Figure 4 brings together the various marginal effects described in Table 2 by displaying the predicted probability of U.S. intervention in small and big MIDs as support from the president s handling of ongoing MIDs varies from 25% to 75%. The black line depicts the probability of intervention in small MIDs while the grey line depicts the probability of intervention in larger disputes. For small MIDs we see the combination of the diversionary and political capital mechanisms creating a curvilinear relationship between foreign policy support

22 22 and intervention. The impact of the diversionary mechanism is quite a bit larger, but both effects are substantial. Moreover, we can see that nearly all of the impact of foreign policy approval in these cases occurs when it drops below 40% or above 60%. Within that middle range the impact of public approval is essentially zero. Turning back to our description in Figure 3, this result indicates that approval of the President s handling of ongoing MIDs has a significant impact on intervention in small MIDs when that level of approval is in the lowest 20% or the highest 20% of the distribution, but has little impact for the middle 60% of cases. Figure 4 About Here The pattern for intervention in big MIDs is quite different. Here we see that most of the impact of foreign policy support occurs as it varies between 40% and 60%. Specifically, as support declines from 60% to 40% the probability of intervention in a big MID declines from nearly 4% to virtually zero. Further declines in support have no impact. Similarly, increases in foreign policy support above 60% do not raise the probability of intervention further. Instead, the likelihood of intervention declines modestly, although this decline is not statistically significant. Returning once again to Figure 3, these results indicate that MID approval has a substantial impact on intervention in big MIDs for the 60% of our observations with moderate levels of approval, but that variations in MID approval have little impact for the top and bottom 20% of the distribution. Table 3 displays the estimated marginal effects for the statistically significant international level variables. As noted above, many of the international factors in our model do not figure prominently in American decisions to intervene in small MIDs. We can see, however, that the presence of a state that shares a defense pact with the United States does have a substantial impact on U.S. intervention. Specifically, the presence of a defense pact increases the probability of U.S. intervention from 1.2% to 12.8%. This increase translates to more than a 10- fold increase in the relative risk of intervention. Thus a defense pact has a slightly more powerful impact on presidential decisions to intervene in small MIDs than the diversionary incentives outlined above. Presidents have also been substantially less likely to intervene in small MIDs that involve a democracy. Shifting the maximum democracy score of the states involved in the MID from -10 to 10 decreases the probability of intervening in a small MID from 6.1% to 0.7%. Table 3 About Here

23 23 A much wider variety of international factors influence U.S. intervention in big MIDs. First, the maximum CINC score of any of the states participating in the dispute has a substantial impact on U.S. intervention. Shifting the maximum CINC score from (the minimum value within the big MID category) to its maximum value of increases the likelihood of intervention from 2% to 36%. While the political capital effects outlined above are substantial, this result clearly indicates that international factors have the greatest impact on American decisions to intervene in big MIDs. Looking farther down the results in Table 3, we can see that U.S. trade dependence on the disputants also has a dramatic effect on the likelihood of intervention. Specifically, moving from the 5 th to the 95 th percentile ( to ) increases the probability of U.S. intervention in a big MID from 1.7% to 23.6%. This represents nearly a fifteen-fold increase in the relative risk of intervention, which rivals the impact of the maximum CINC score. Similarly, American involvement in other ongoing MIDs also has a profound impact on the likelihood of intervention in a new MID. Specifically, changing the number of ongoing MIDs for the United States from their minimum (0) to their maximum level (5), increased the probability of intervention from 0.5% to 20%. This effect suggests that U.S. participation in MIDs is strongly clustered temporally. This effect could be due to contagion or spillover effects between MIDs, or it could be due to systemic factors driving dispute involvement. Finally, the presence of alliance ties with the disputants had a significant if somewhat substantively more modest impact on U.S. intervention. The presence of a state with an entente, non-aggression pact, or neutrality agreement with the United States increased the probability of U.S. intervention from 2% to 9.4%. This change represents a nearly five-fold increase in the relative risk of intervention. As noted above, the presence of a U.S. defense pact with one of the disputants actually reduces the probability of U.S. intervention in a big MID. This result runs directly contrary to the effects we find regarding small MIDs, and may be due to the unwillingness of powerful states to initiate disputes against state that share a defense pact with the United States (Fearon 1994; Smith 1996). Conclusion

24 24 Our results underscore the importance of public opinion in shaping U.S. presidential decisions to intervene in militarized disputes. Perhaps just as importantly, they provide support for our use of Prospect Theory to create a theoretical and empirical synthesis of what at first seemed to be disparate arguments about the conditions that promote or inhibit U.S. military interventions in foreign militarized disputes. That existing theoretical literature has posited two distinct and contrasting mechanisms by which public opinion may influence the use of force by democratic states. First, the political-capital mechanism suggests that elected leaders will be more likely to engage in military conflict when public attitudes toward their handling of military conflict gives them the political capital necessary to use force and provides them with a cushion of support in the event that the conflict fares poorly. As popular support for the president s handling of military conflict declines, however, he will feel increasingly constrained from using force for fear of aggravating the public s negative judgments about his competence. Second, the diversionary-war mechanism suggests that leaders will be more likely to engage in conflict when public support for their handling of military conflicts is very weak. In these instances, elected leaders may gamble for resurrection by intervening in a conflict in the hopes of succeeding and reversing the public s negative perception. As public support for the president s handling of MIDs increases, however, the incentive to engage in diversion will wane. Much of the literature on public opinion and U.S. decisions to use force has generated mixed results, and we argue that this turbulence may be due to the contrasting diversionary and political-capital mechanisms. By framing both of these arguments in terms of their assumptions about the risk propensities of national leaders, however, we were able to use Prospect Theory to articulate the conditions under which leaders behavior should conform to each mechanism. Specifically, we argued that the political-capital model assumes risk averse leaders who will avoid gambling on conflict without a strong buffer of support, while the diversionary model assumes risk acceptant leaders who will jump at even the remote chance to recoup their losses. Prospect Theory leads us to expect that leaders will act according to the political-capital model when they enjoy relatively high levels of public approval, but according to the diversionary model when their approval levels are relatively lower. Consistent with this expectation, our results indicate that diversionary incentives have a powerful impact on presidential decisions to intervene in militarized disputes, but only with regard to very small disputes and only when approval of the president s handling of recent MIDs is relatively low. At the same time, we find that political-capital dynamics constrain U.S. presidents from intervening

25 25 in militarized disputes involving both small and large states. With regard to small MIDs, we find that relatively high levels of public support for the handling of recent MIDs may embolden presidents to intervene in such small MIDs. With regard to MIDs involving larger states, on the other hand, declines in public support appear to constrain presidents from intervening in new MIDs, especially at very low levels of public approval for such recent MID performance. Additionally, our analyses suggest the importance of issue-specific as opposed to overall measures of presidential approval. These results suggest that when presidents gamble for resurrection they do so by directly confronting (and attempting to reverse) public perceptions of their incompetence rather than deflecting attention from domestic to international issues. Moreover, the results suggest that political capital may be issue specific in terms of the buffer that it provides presidents to engage in risky or costly policies. Finally, while our theoretical focus has been primarily on the domestic sources of American military intervention, our analysis also sheds light on the importance of strategic international factors on the president s willingness to intervene. With regard to MIDs involving very weak states, we find that domestic political factors have primacy in determining American intervention. However, international strategic factors have substantial effects on American decisions to intervene in militarized conflicts with larger states.

26 26 References Aydin, Aysegul Choosing Sides: Economic Interdependence and Interstate Disputes. Journal of Politics 70 (4): Barbieri, Katherine, and Omar Keshk Correlates of War Project Trade Data Set Codebook, Version 3.0. Online: Barbieri, Katherine, Omar M. G. Keshk, and Brian Pollins Trading Data: Evaluating our Assumptions and Coding Rules. Conflict Management and Peace Science 26(5): Bennett, D. Scott, and Allan Stam EUgene: A Conceptual Manual. International Interactions 26: Brulé, D.J., Marshall, B.W. and Prins, B.C., Opportunities and Presidential Uses of Force A Selection Model of Crisis Decision-Making. Conflict Management and Peace Science, 27(5), pp Butler, Michael J U.S. Military Intervention in Crisis, : An Empirical Inquiry of Just War Theory. Journal of Conflict Resolution 47 (2): Chiozza, Giacomo, and H. E. Goemans Leaders and International Conflict. Cambridge: Cambridge University Press. Clark, David H Can Strategic Interaction Divert Diversionary Behavior? A Model of U.S. Conflict Propensity. Journal of Politics 65 (4): Davies, Graeme A. M Domestic Strife and the Initiation of International Conflicts: A Directed Dyad Analysis, Journal of Conflict Resolution 46 (5): Edwards III, G.C., Mitchell, W. and Welch, R., Explaining presidential approval: The significance of issue salience. American Journal of Political Science, pp Fearon, James "Domestic Political Audiences and the Escalation of International Disputes." American Political Science Review 88 (3):

27 27 Fordham, Benjamin. 1998a. A Political Economy Model of U.S. Uses of Force: Supply, Demand, and Endogenous Threat Perception. International Studies Quarterly 42 (3): Fordham, Benjamin. 1998b. Partisanship, Macroeconomic Policy, and U.S. Uses of Force, Journal of Conflict Resolution 42 (4): Fordham, Benjamin O "Revisionism Reconsidered: Exports and American Intervention in World War I." International Organization 61 (2): Gelpi, Christopher Democratic Diversions: Governmental Structure and the Externalization of Domestic Conflict. The Journal of Conflict Resolution 41 (2): Gibler, Douglas M International Military Alliances, CQ Press. Goemans, H.E War and Punishment. Princeton: Princeton University Press. Hess, Gregory D., and Athanasios Orphanides War Politics: An Economic, Rational Voter Framework. American Economic Review 85 (4): War and Democracy. Journal of Political Economy 109 (4): James, Patrick, and John R Oneal The Influence of Domestic and International Politics on the President s Use of Force. Journal of Conflict Resolution; 35 (2): Timothy R. Johnson, and Jason M. Roberts Presidential Capital and the Supreme Court Confirmation Process. Journal of Politics 66 (3): Pivotal Politics, Presidential Capital, and Supreme Court Nominations. Congress & the Presidency, 32:1: Kahneman, Daniel, and Amos Tversky Prospect Theory: An Analysis of Decision Under Risk. Econometrica 47 (2): Kegley, Charles W., and Margaret G. Hermann How Democracies Use Intervention: A Neglected Dimension in Studies of the Democrat ic Peace. Journal of Peace Research 33 (3): Kegley, Charles W., and Margaret G. Hermann Military Intervention and the Democratic Peace. International Interactions 21 (1): Leeds, Brett Ashley, and David R. Davis Domestic Political Vulnerability and International Disputes. Journal of Conflict Resolution 41 (6):

28 28 Levy, Jack S., and T. Clifton Morgan The War-Weariness Hypothesis: An Empirical Test." American Journal of Political Science 30 (1): Levy, Jack S "Domestic Politics and War." Journal of Interdisciplinary History 18 (4): "The Diversionary Theory of War: A Critique." In Handbook of War Studies, ed. Manus I. Midlarsky. Ann Arbor: University of Michigan Press Prospect Theory and International Relations: Theoretical Applications and Analytical Problems. Political Psychology 13 (2): Prospect Theory, Rational Choice, and International Relations. International Studies Quarterly 41 (1): Lian, Bradley, and John R. Oneal Presidents, the Use of Military Force, and Public Opinion. Journal of Conflict Resolution 37 (2): Light, Paul. C The President's Agenda: Domestic Policy Choice from Kennedy to Clinton. Baltimore: Johns Hopkins University Press. Marshall, Monty G., Ted R. Gurr, and Keith Jaggers Polity IV Project: Political Regime Characteristics and Transitions, : Dataset Users Manual. College Park, MD: Center for International Development and Conflict Management, University of Maryland. McDermott, Rose Risk-Taking in International Politics: Prospect Theory in American Foreign Policy. Ann Arbor: University of Michigan Press. Meernik, James Modeling International Crises and the Political Use of Military Force by the USA. Journal of Peace Research 37 (5): Meernik, James, and Peter Waterman The Myth of the Diversionary Use of Force by American Presidents. Political Research Quarterly 49: Mercer, Jonathan Prospect Theory and Political Science. Annual Review of Political Science (8): Miller, J.M. and Krosnick, J.A., News media impact on the ingredients of presidential evaluations: Politically knowledgeable citizens are guided by a trusted source. American Journal of Political Science, pp

29 29 Mitchell, Sara M., and Will H. Moore Presidential Uses of Force during the Cold War: Aggregation, Truncation, and Temporal Dynamics. American Journal of Political Science, 46 (2): Morgan, T. Clifton, and Kenneth Bickers Domestic Discontent and the External Use of Force. Journal of Conflict Resolution 36: Morgan, T. Clifton, and Christopher J. Anderson Domestic Support and Diversionary External Conflict in Great Britain, Journal of Politics 61: Morgenthau, Hans J To Intervene or Not to Intervene. Foreign Affairs 35.3: Oneal, John R., and Anna Lillian Bryan The Rally 'Round the Flag Effect in U.S. foreign Policy Crises, Political Behavior 17(4): Oneal, John R., and Jaroslav Tir Does the Diversionary Use of Force Threaten the Democratic Peace? Assessing the Effect of Economic Growth on Interstate Conflict, International Studies Quarterly 50 (4): Ostrom, Jr., Charles, and Brian L. Job The President and the Political Use of Force. American Political Science Review 80 (2): Palmer, Glenn, Vito D Orazio, Michael Kenwick, and Matthew Lane The MID4 Dataset, : Procedures, Coding Rules and Description. Conflict Management and Peace Science (forthcoming). Pearson, Frederic S "Geographic Proximity and Foreign Military Intervention: Journal of Conflict Resolution 18 (3): Pickering, Jeffrey War-Weariness and Cumulative Effects: Victors, Vanquished, and Subsequent Interstate Intervention. Journal of Peace Research 39 (3): Pickering, Jeffrey, and Emizet Kisangani Democracy and Diversionary Military Intervention: Reassessing Regime Type and the Diversionary Hypothesis. International Studies Quarterly 49 (1): Richards, Diana, T. Clifton Morgan, Rick K. Wilson, Valerie Schwebach, and Garry D. Young Good Times, Bad Times, and the Diversionary Use of Force: A Tale of Some Not-So-Free Agents. Journal of Conflict Resolution 37 (3):

30 30 Singer, J. David "Reconstructing the Correlates of War Dataset on Material Capabilities of States, " International Interactions 14: Smith, Alastair Diversionary Foreign Policy in Democratic Systems. International Studies Quarterly 40 (1): International Crises and Domestic Politics. American Political Science Review 92 (3): Tillema, Herbert K Cold War Alliances and Overt Military Interventions, International Interactions 20 (3): Werner, Suzanne Deterring Intervention: The Stakes of War and Third-Party Involvement. American Journal of Political Science 44 (4): Young, Oran R Intervention and International Systems. International Affairs 22 (2):

31 Figure 1 Predicted Effects of Diversionary and Political Capital Incentives on the Decision to Join Small and Large MIDs 31

32 Figure 2 32

33 Figure 3 33