Two Limitations

Before elaborating on why a leader's culpability will affect his or her war termination decision, we must first understand the utility of this new concept. In the Introduction I claimed that existing theories have a difficult time accounting for variation in leader behavior. This limitation stems from two common practices. First, much of the existing literature focuses on the differences between broad classes of leaders—namely, democratic leaders vs. nondemocratic leaders—which often has the effect of obscuring important distinctions within groups. Second, current theories do not offer a causal mechanism for the pressure to secure a victory to pass from one leader to the next. Because substantial variation exists both within regime types and even within leaders from the same state, overcoming these limitations is essential.

As to the first problem, regime-based theories assume leaders of the same regime type will have similar sets of incentives, especially in comparison to other types of political systems.Footnote ² Although recent research cautions scholars against making the crude assumption that all leaders of the same regime type can be treated as a homogenous group,Footnote ³ work in this vein still assumes that, on the average, broad differences exist between the different types of states. Democratic leaders under condition X may be less sensitive to public opinion than leaders under condition Y, but both leaders will be more sensitive than their nondemocratic counterparts. Indeed, several prominent theories of war termination and wartime behavior more generally turn on this assumption (e.g., Bennett and Stam Reference Bennett and Stam1998; Reiter and Stam Reference Reiter and Stam III2002). This insight accounts for a number of democratic leaders throughout history, but it lacks a causal story for leaders who defy public opinion during unpopular wars. Related theories based on the assumption of democratic sensitivity to casualties suffer from a similar problem. They can tell us why some democratic leaders end costly wars, but not why others continue to fight in the face of heavy losses.Footnote ⁴

The second limitation—the lack of a mechanism to explain how and when the pressure to win will pass from the first leader to a successor (hereafter, “replacement leader”)—is not often mentioned in the literature, because scholars rarely focus on wartime leadership changes.Footnote ⁵ The standard logic regarding who will feel the pressure to attain victory is that the leaders who are in charge when the war starts (hereafter, “first leaders”) will, but subsequent leaders will not. For example, implicit in Downs and Rocke (Reference Downs and Rocke1994) is the assumption that any leader who leader takes power during a war should be more amenable to a war-weary constituency's demands than a leader who started the conflict. Leaders of the latter type, the authors argue, are more likely to face punishment if they lose, and therefore have the incentive to “gamble for resurrection” by staying in the conflict. Stanley (Reference Stanley2009) and Stanley and Sawyer (Reference Stanley and Sawyer2009) agree with the notion that first leaders should be more likely to continue fighting, but for slightly different reasons. For them, first leaders are more likely than replacements to remain in a conflict because they are more prone to a variety of informational biases and are more likely to feel entrapped. Bringing a new leader to power, the authors argue, can help overcome these impasses by either introducing new and possibly more informed perspectives or reducing the political costs of admitting defeat that kept the first leader in the conflict.Footnote ⁶

Both theories almost certainly capture the incentives of several first leaders who have exhibited a strong determination to win. What they cannot explain, however, is the variation among the set of leaders that replace them.Footnote ⁷ Why do some replacement leaders continue to gamble, whereas others choose to cut their losses and end the war? Why are some replacement leaders able to slip the bonds of entrapment, whereas others seem to inherit them from their predecessors? Without a causal means by which the incentive to continue fighting might pass from leader to leader, these questions remain unanswered. This problem is not inconsequential: Of the 259 warring states in this study, 51 (or nearly 20%) have changed leaders at least once by the end of the war.Footnote ⁸ Indeed, because some states change leaders multiple times; across the entire data set there are 96 replacement leaders, accounting for more than a quarter of the observations. Assuming a priori that all 96 will have similar incentives not only oversimplifies the political dynamics of the leader changes but also, as I demonstrate later, masks important differences among these leaders.

The Solution

Recognizing the importance of a leader's culpability addresses both of these limitations. First, it identifies a way for otherwise similar leaders to have different incentives for continuing a war. Because a leader's culpability stems from the specific role he or she plays in the conflict, we need not assume all democratic leaders or all leaders who take over midwar will be subject to the same constraints. Second, it also provides a means for replacement leaders to inherit culpability for a war. Because a leader's culpability for a war is largely determined by his or her position in the government or political connections to previous executive, predicting which replacement leaders will have incentives to keep fighting is fairly straightforward. Additionally, as I mentioned previously, culpability also provides a way to predict which leaders citizens will want to remove. Currently, scholars expect domestic audience members to punish leaders who lose. The historical record, however, suggests that their logic is more complex: Some leaders who fail to secure victories are punished, but others are not. Once we recognize that citizens will only want to punish culpable leaders who lose, explaining variation in leaders’ fates becomes an easier task.

In the following sections, I explain which leaders will be subject to the pressures of culpability; provide a rationale for why culpable leaders will have reason to gamble, whereas nonculpable leaders will not; and explain why citizens will want to punish culpable leaders who lose and spare nonculpable leaders who do the same.Footnote ⁹ I also present hypotheses on the effect of a leader's culpability on war outcomes and the likelihood of domestic punishment.

Determining Culpability

Two types of leaders are subject to the pressures of culpability. As I mentioned earlier, the first type consists of leaders who were in power when the war began, or first leaders. “First leader,” importantly, is not synonymous with “initiator.” Instead, this group includes initiators, leaders who choose to join conflicts already in progress, and leaders whose states were attacked. The defining criterion is that the leader oversaw the state's transition from peace to war. Although initiators and third-party joiners could, potentially, deserve more culpability because they actively chose to involve their states in conflicts, some scholars have argued that targeted leaders will acquire a similar burden (Bueno de Mesquita and Siverson Reference Bueno de Mesquita and Siverson1995). The theoretical rationale behind this claim is that citizens will view an attack on their state as a failure on the part of the leader to either foresee the danger and pre-empt the attack, or otherwise placate the aggressor and avoid violence altogether. Given this, and the fact that very few wars are the result of true “surprise attacks,” in which the targeted leader had no inkling of the impending conflict, it is reasonable to think that citizens in a targeted state will not automatically forgive the leader for their state's involvement in a war.Footnote ¹⁰

The second type of leader whom citizens will hold culpable is a leader who comes to power during the war and who is from the same political faction as the first leader. The term “faction” is intentionally vague, to accommodate any regime type. Membership in a leader's faction implies close proximity to the decision-making process and a political relationship with other faction members, including the executive. Although these successor leaders do not have direct responsibility for the war, the clear ties that exist between the new leaders and their predecessors will lead their citizens to transfer the blame with minimal, if any, discounting.Footnote ¹¹ In nondemocratic states a transfer of culpability between a first leader and his or her replacement occurs if the new leader comes from the former leader's inner circle (e.g., the political elite in a one-party system, generals in a military junta, or the royal family in a monarchy). In democracies, culpability will transfer to a replacement leader if that leader is from the same political party or, in the case of a parliamentary system, the same party or cabinet as his or her culpable predecessor.Footnote ¹²

This assumption of a partisan cue draws on the large body of literature in American politics that demonstrates that citizens commonly rely on political heuristics such as party affiliation to make assumptions about and to draw connections between leaders.Footnote ¹³ There is also evidence that citizens are willing to act on these connections when administering punishment. Cotton (Reference Cotton1986, 625), for example, in a study directly relevant to the question at hand, provides evidence of democratic citizens’ willingness to pass blame along party lines in his study of U.S. wartime elections, where he finds that members of the president's party fare especially poorly. This leads him to conclude that the costs of war “can hurt the party associated with America's entry into the fighting.”Footnote ¹⁴

In general, citizens transfer culpability along factional lines because the first leader's fellow faction members are more likely than other actors to have played a role in or to have supported the war when it began. This is a safe assumption to make because executives should be reluctant to participate in wars unless they have, at minimum, support from their fellow faction members, regardless of their regime type. As Schultz (Reference Schultz2001b) noted, this strategic behavior is evident in democracies, where most of the wars executives choose to fight are supported (at least initially) by a large majority of the legislature. A similar story should also be true for nondemocratic leaders. Although these leaders face fewer constraints from the need for formal institutional approval, they should still be interested in the approval of their winning coalitions, whose support they need to remain in power. Consequently, we should not expect to observe many instances of leaders going to war without the support of those closest to them, regardless of regime.

Understanding Leader Behavior

Having established the determinants of culpability, I proceed to how culpability (or the lack thereof) affects a leader's war termination calculus. This helps to answer the related questions of why a culpable leader may continue to fight, even if victory is unlikely, and why a nonculpable leader may end a war, even when winning remains a possibility. As I argue in more detail subsequently, the primary factor in determining a leader's choice between continuation and termination of the conflict is the probability of punishment if he or she accepts any outcome less than a win. For culpable leaders this probability will almost always be very near one, whereas for nonculpable leaders it will most likely be zero.

The expectation that culpable leaders will face punishment if they do not secure a victory draws on the standard audience costs argument, which posits that citizens will punish a leader who fails to follow through on public statements successfully (Fearon Reference Fearon1994, Reference Fearon1997; Schultz Reference Schultz2001a, Reference Schultz2001b). The argument I present here builds on this logic in two important ways. First, it extends the domain of audience costs by suggesting that the actions a leader takes at the beginning of a war (either starting it or supporting it) will continue to affect the leader's behavior until either the conflict ends or he or she loses power. Extending the effect of audience costs in this way constitutes an innovation because both existing theories about and empirical tests of the argument are couched almost exclusively in the context of crises.Footnote ¹⁵ Crises and wars are different in important ways, but this does not mean that citizens will discount the leader's role in the decision that brought the state to war once the conflict escalates beyond the crisis. Indeed, as recent empirical work (e.g., Tomz Reference Tomz2007) suggests, the likelihood that a leader will face audience costs increases as the leader engages in more escalatory behavior. To account for this, I argue that citizens will want to make the actor(s) responsible for the state's participation in the war “pay” the audience costs in the event of a bad outcome, regardless of how long the conflict continues after the initial crisis.

The second improvement I make, which I discussed previously, is to provide a way to predict when this pressure to win will pass to replacement leaders. The current literature on crises overlooks this possibility, because most crises are too short to witness a change in leadership. The same does not hold, however, for wars, where leadership changes are not uncommon. The argument I present here fills this gap by positing that citizens will transfer the pressure to secure a victory from leader to leader along factional lines.

Once we realize that the pressure to deliver a promised outcome can extend beyond a crisis, and that this pressure can pass to subsequent leaders, we can explain why culpable leaders, regardless of when they came to power or their regime type, will have strong incentives to avoid settling without a win. As audience costs scholars would argue, the domestic opposition will leap at the chance to target a leader who ends a war he or she had a role in starting with anything less than a total victory.Footnote ¹⁶ If a first leader fails to win a war, domestic opponents will suggest the setback is indicative of a larger competence problem in an attempt to weaken the executive in other policy realms. Any culpable successor who presides over unfavorable results will face similar allegations. Affiliation with the first leader and earlier support of the conflict will complicate denying the opposition's claim that he or she is responsible for the loss.Footnote ¹⁷

Because culpable leaders who perform badly face a heightened risk of punishment, what types of behaviors should we expect from them? I predict that these leaders will continue fighting even if doing so carries a low chance of winning and the costs of staying in the war are high. Put differently, of leaders who fight wars, culpable leaders should have the higher probability of fighting until they have “lost the power to hurt” (Slantchev Reference Slantchev2003).Footnote ¹⁸ So long as winning the war (and, consequently, escaping punishment) remains a possibility, continuing to fight will always carry a higher expected utility for the leader than settling for less than victory in the current period. This is, of course, the logic Downs and Rocke (Reference Downs and Rocke1994) offered for why leaders who find themselves in losing battles should “gamble for resurrection.”Footnote ¹⁹ The argument I present here builds on this insight by offering a way to predict which leaders will take this gamble (culpable leaders), and which will admit defeat (nonculpable leaders). Because culpable leaders—regardless of when they come to power—will face punishment if they lose, they should always opt to keep fighting. We should also expect them to subject their states to higher costs (via riskier strategies, more aggressive tactics, troop increases, etc.) in an effort to increase their chances of winning. Although this approach may be costly for the state and anger the domestic audience further, the culpable leader has nothing to lose by trying, because the risk of punishment if he or she loses stays the same regardless of the additional resources invested in the war. The leader cannot, in other words, do worse by continuing to fight. Indeed, it is the only chance he or she has to do better.Footnote ²⁰

For nonculpable leaders, the opposite is true. These leaders can only do worse if they choose to stay in the conflict. They face little risk of punishment for backing down—even if this means accepting a loss—but they expose themselves to more risks if they continue. Put differently, we should expect nonculpable leaders to be risk-averse compared to culpable leaders. Recall that culpable leaders will embrace the uncertainty of the future, because maintaining the potential for victory is the only way they can justify keeping the state in the conflict and keep the possibility of escaping punishment open. A nonculpable leader, in contrast, should eschew courses of action with uncertain outcomes, especially because he or she can achieve a good outcome (for him or her self) with certainty by backing down.

Although some may argue that winning is better than losing, ceteris paribus, opting to stay in a war will be unattractive to nonculpable leaders for at least two reasons, even if winning is still possible. First, victory is unlikely to be certain. Even if the state has fair odds of winning, staying in a war necessarily involves a gamble that may not pay off, and this potentially leaves the leader worse off than if he or she had terminated the war.Footnote ²¹ Second, choosing to stay in the war has the possibility of saddling the leader with the burden of culpability. Because his or her ability to back down without negative repercussions is common knowledge, declining this option would signal to the domestic audience that the leader is willingly taking on responsibility for the war. Put differently, whereas citizens might expect a culpable replacement leader to feel compelled to keep fighting because of ties to his or her predecessor, they will likely perceive the same decision made by a nonculpable successor as an active choice precisely because he or she lacks such ties. Given that culpability increases the probability of punishment in the event of a loss, a nonculpable leader should have no incentive to risk this outcome by staying in the war.Footnote ²²

Nonculpable leaders should also have little reason to invest additional resources in the conflict. In addition to the risk of ultimately losing the conflict, continuing to pour resources into the war robs them of the chance to allocate those resources to other matters on their agendae. Moreover, unlike culpable leaders who having nothing to lose and everything to gain by taking risks in an effort to eke out a win, nonculpable leaders stand to lose substantially by engaging in such behavior for reasons discussed earlier. A nonculpable leader also faces the added risk of squandering goodwill with the domestic audience by subjecting them to additional costs, especially if they feel that victory is not possible.

Understanding Citizen Behavior

I now turn to the question of why citizens want to punish culpable leaders who perform poorly in wars, and spare nonculpable leaders who do the same. At least two reasons exist: incompetence and the state's international reputation. Culpable leaders are more susceptible than nonculpable leaders to being labeled “incompetent” because citizens can more easily link them to the decisions that led to poor outcomes. The first place a leader's incompetence can develop in the eyes of citizens is in the selection of the conflict. If a leader involves a state in a war in which the state is clearly outmatched or that goes poorly quickly, opponents of the leader could argue he or she lacks sound judgment in foreign policy. Citizens will have incentives to remove such a leader not only to prevent him or her from making the same mistake again, but also to deter future leaders from behaving similarly (Downs and Rocke Reference Downs and Rocke1994). Culpable leaders are more likely to face punishment for this reason than nonculpable leaders because they either oversaw the onset of the war or supported the decision to participate in the conflict.

The second way a culpable leader can appear incompetent in the eyes of citizens is in the way he or she prosecutes the war. If a culpable leader selects into a war the citizens initially viewed as a good bet, and goes on to lose (or simply fails to win), citizens will assume the leader must have done something to squander the state's chances of victory. Critically, the fact that the war seemed like a good idea at the start should not encourage the citizens (or the opposition) to be more forgiving if the war ends poorly. Indeed, it should inspire more indignation. If a state had, for instance, an estimated 95% chance of winning a war at the outset, yet it still loses, citizens will likely assume that a loss in the face of such overwhelmingly favorable odds must have resulted from the leader's severe mismanagement of the war (Gelpi and Greico Reference Gelpi and Grieco2001; Smith Reference Smith1998). A nonculpable leader, predictably, will have an easier time dodging such accusations of poor war management by credibly distancing himself or herself from the war's initial phases and/or earlier tactical decisions made by the previous leader(s).

The second reason citizens will be more willing to punish culpable leaders than nonculpable ones centers on how the leader is perceived by the international audience. As others have argued, domestic audience members have an incentive to remove executives when they feel the executives have damaged their international reputations (McGillivray and Smith Reference McGillivray and Smith2008). Although scholars typically discuss this concept in the context of international cooperation, where leaders damage their reputations by reneging on agreements, the concept is equally applicable to conflict situations (Tomz Reference Tomz2007). In a war, leaders can damage their reputations by failing to secure favorable outcomes for wars they had a hand in starting. A failure to win will call the leader's resolve into question in the eyes of other national leaders, potentially inviting more challenges in the future. Citizens, therefore, will want to remove such leaders from office and replace them with leaders whose international credibility is intact. This preference favors nonculpable leaders because of their ability to disassociate themselves both from the specific decision to go to war in the first place and from the previous leader(s) more generally. The new leader may prove to be a poor choice, potentially damaging his or her own reputation at a later date, but in the current time period, a leader without a record will be preferable to a leader with a tarnished one.

Hypotheses

The preceding logic implies several hypotheses regarding how a leader's culpability should affect both the outcomes he or she is likely to preside over and the probability that he or she will be punished if the state does not win. First, because culpable leaders face a heightened likelihood of domestic punishment if a war they are linked to ends poorly, they have stronger incentives than nonculpable leaders to secure the best outcome possible. This suggests the following hypothesis:

Of course, simply having stronger incentives to win does not mean that leader will always be able to do so. All culpable leaders should try to win, but a stochastic component determines which gambles will pay off. Some culpable leaders may succeed, whereas others may stay in too long, making their state worse off than if they had settled earlier (i.e., the aforementioned tendency for culpable leaders to fight until they “lose the power to hurt”). This suggests that culpable leaders should be associated with extreme outcomes, that is, both major victories and catastrophic defeats. Nonculpable leaders, on the other hand, should be more likely to preside over medial to poor outcomes. This suggests the following hypothesis:

Although culpable leaders across all states should feel more pressure to end the war on favorable terms than their nonculpable counterparts, variation in domestic institutions might create differences among culpable leaders. For instance, in a democracy, where institutions facilitate removing the leader (e.g., a viable opposition and a means for regular changes in power), the leaders may be more responsive to the pressures associated with culpability.Footnote ²³ This suggests the following hypotheses.

The culpable leader's incentive to fight until victory, of course, stems from anticipation of punishment by the domestic audience if he or she fails to win. As previous discussions argued, citizens have a multitude of reasons for wanting to punish culpable leaders who involve their states in wars that end poorly. A nonculpable leader, in contrast, faces a much smaller probability of domestic punishment, regardless of the ultimate terms of settlement. This suggests the following hypothesis:

Finally, as noted previously, because only some polities have institutions that enable the removal of leaders, the frequency of observed punishments might vary by regime type. Specifically, because domestic audiences can more easily remove their leaders through regular means in democracies, we might expect culpable democratic leaders to be punished more than their culpable nondemocratic counterparts.Footnote ²⁴ This suggests the following hypothesis:

Dependent Variables

War Outcome. Operationalizing the war's outcome required devising a coding method that would allow me to use every warring state in the data set as a separate observation. Scholars interested in modeling war outcomes typically drop all but one state from each war from their analyses, because including all of the states would introduce correlation among the error terms.Footnote ²⁸ To reduce this problem, and include as many states as possible, I designed a coding scheme that focuses on the degree to which a leader achieved the state's maximum war aimsFootnote ²⁹ instead of how he fared relative to his opponent. I used the maximum war aims for two related reasons. First, because war aims can change over the course of the war (Goemans Reference Goemans2000, Reiter Reference Reiter2009), relying on the original war aims would miss important changes in the state's goals. Second, using the maximum war aims is important because they provide the baseline against which citizens and, importantly, members of the domestic opposition will assess a leader's degree of accomplishment. Although a leader may want to revise the war aims downward to minimize the appearance of failure when predicting that he or she cannot win, the opposition will have strong incentives to remind citizens of the earlier higher aims.

Coding a war's outcome with respect to the state's maximum war aims produced the following six-point scale: 2: total victory (accomplish all war aims, no losses); 1: partial victory (achieve most, but not all war aims, no losses); 0: prewar status quo (accomplish no war aims, no losses); -1: partial loss (accomplish no war aims, make minor concessions, ally suffers major loss); -2: major loss (accomplish no war aims, make major concessions); -3: total loss (i.e., loss of sovereignty). The distribution for this variable was as follows: total loss (13%), major loss (6%), partial loss (22%), prewar status quo (12%), partial victory (7%), total victory (39%).Footnote ³⁰

In addition to offering a more nuanced conceptualization of the different ways in which wars can end, the coding scheme represents a theoretical advance by providing a more accurate measure of a leader's achievement.Footnote ³¹ It also allows assignment of different outcomes across members of large coalitions, allowing a more accurate depiction of how individual states fared. In WWII, for instance, although Belgium and the United Kingdom were both on the same side, the former was conquered early in the war and occupied by Nazi Germany, whereas the latter maintained its sovereignty. Given these very different war experiences, I code Belgium's final outcome as a “total loss” because the state's formal participation in the war ended with the Nazi invasion, whereas I code the United Kingdom's final outcome as a total victory.Footnote ³²

Domestic Punishment. Although the war termination literature offers several measures of leader punishment, testing my hypotheses required that I operationalize this variable in a different way. Existing designs, though both reasonable and intuitive, only incorporate one leader per war.Footnote ³³ This difference would be irrelevant if the leader who started the war were always the same as the leader who ended it, but this is not the case. Instead, using only one leader per war introduces the potential for error into the analyses, regardless of which leader the researcher chooses to study. For example, if the sample only includes leaders who were in power when the war began, testing hypotheses regarding culpability becomes impossible because the key independent variable is a constant. Examining the leaders who are in power when the war ends introduces a different yet equally pernicious problem. Under this approach, the data set contains none of the leaders who were removed from power during the war, and includes all the leaders who were put into office to replace them. This exclusion is consequential because it systematically eliminates from the sample all of the culpable leaders who were removed during the war because of poor performance. This research design also oversamples nonculpable leaders who may have been brought to office with a specific mandate to end the war, many of whom will be nonculpable.Footnote ³⁴

To address these concerns, I coded the dependent variable representing domestic punishment as a 1 if the leader was removed from power because of the war and a 0 otherwise.Footnote ³⁵ Leaders qualified for a coding of 1 if (a) the majority of sources agreed their removal was a clear result of the war and (b) the removal was the result of coordinated punishment by domestic actors or by the leader's own doing when domestic punishment was imminent.Footnote ³⁶ Using this coding rule, I determined that 61 leaders (around 17%) were punished because of a conflict.Footnote ³⁷ Of these punished leaders, 22 (36%) were from democracies, whereas 39 (64%) were from nondemocracies. Importantly, disagreement among the sources was rare; of the 355 leaders, the role the war played in their departure from office was unclear in only 10 cases (or about 3% of the time).Footnote ³⁸ Leaders received a coding of 0 if the majority of sources agreed they left office for reasons unrelated to a war, which includes a wide variety of scenarios. First, leaders can leave for institutional reasons that are separate from a war, such as having served the maximum number of terms in office (e.g., Gabriel Moreno of Ecuador, who left power in 1865), or fulfilling their positions as interim, regents, or other provisional leaders (e.g., Frank Forde of Australia, who took over as an interim after John Curtin died during World War II, or ‘Abd al-Ilah of Iraq, who was the regent for King Faisal II until the latter came of age in 1953). In these scenarios, the leadership change would have occurred even if the war had not taken place.Footnote ³⁹

Five other scenarios also arise independent of the leader's performance in the war, but are more stochastic in nature. These include natural death, leaving for another high-ranking political position, sickness, retirement, and assassination. Whereas the first is self-explanatory (e.g., Franklin Roosevelt's death from a stroke during World War II or Victor Emmanuel II of Italy dying of natural causes in 1878), the latter four are not as straightforward.Footnote ⁴⁰ Instances of leaders leaving an executive office to take another key position in government are rare (this occurs twice in the data set), but careful investigation of the circumstances surrounding their transition from one post to another demonstrated that they were not cases of leader punishment. I carefully considered cases in which the leader resigned for health reasons or retired, because leaders can offer both of these reasons in an attempt to provide themselves with a relatively graceful means of dodging political problems. For a leader to be classified as sick, he or she had to be bedridden (e.g., Woodrow Wilson or Vladimir Lenin, both incapacitated by strokes) or have a majority of sources state that the primary reason for his or her leaving office was illness (e.g., Tomáš Masaryk of Czechoslovakia, who left office in 1935). I classified a leader as “retiring” if a majority of sources agreed he or she left office by choice (e.g., James Polk of the United States, who left office in 1849, or Robert Menzies of Australia, who retired in 1963). Instances of assassination are interesting because although they appear to be prime examples of leader punishment in its most extreme form, a lethal shot from a lone gunman (e.g., the fates of Nasser al-Din Shah Qajar of Iran and Sidónio Pais of Portugal, who were slain by fanatics in 1896 and 1918, respectively) does not qualify as a coordinated political sanction from the domestic audience that is a result of the war and thus receives a coding of zero for this variable.Footnote ⁴¹

Finally, several others types of cases are coded as zeroes: Leaders who lost power because their states were absorbed into new political entities (e.g., the small European states that eventually were united under German control); leaders who remained in power for several years after the last war in the data set ended in 2000 (e.g., Qaboos bin Said Al Said of Oman, who participated in the first Gulf War in 1991 and is still in power as of 2010); leaders who were removed or who resigned over domestic matters (e.g., Richard Nixon's resignation in the wake of Watergate in 1974, or Saito Makoto of Japan, who resigned in 1934 after a financial scandal); or leaders who were overthrown by a foreign power during a war (e.g., Haile Selassie of Ethiopia, whom the Italians removed in 1936, or Manuel Bonilla of Honduras, who lost power at the hand of José Zelaya and the Nicaraguan army in 1907).Footnote ⁴²

Independent Variables

Leader Culpability. I considered a leader culpable who met one of two criteria: (1) being in power when the state's involvement in the war began, or (2) sharing a political connection with a culpable predecessor. A political connection exists if the new leader is from the same political party as the culpable leader, or was a member of the culpable leader's cabinet, military junta, or close circle of advisors or, in the case of a monarchy, was a member of the culpable leader's family.Footnote ⁴³ Nonculpable leaders, in contrast, are leaders who come to power during wars who lack such connections with the earlier, culpable leader(s).

Of the 355 leaders in the data set, 96 (27%) are new leaders who come to power during a war. Of these 96 replacement leaders, 38 (40%) are from democracies, whereas 58 (60%) are from nondemocracies; 45 (47%) are nonculpable, whereas 51 (53%) share a political connection with a culpable predecessor. Interestingly, the distribution of culpability among replacement leaders across regimes is not as lopsided as one might anticipate. Given the relative ease with which democratic domestic audiences can replace leaders with whom they are dissatisfied, we might expect most replacement leaders in democracies to be nonculpable, because most leader changes during unpopular wars would, presumably, involve replacing a culpable leader with a nonculpable one. In actuality, however, a majority of the replacement leaders in democracies are subject to the pressures of culpability (24 of 38, or 63%). In nondemocracies, 47% (or 27 out of 58) of replacement leaders are culpable. Table 1 provides a sense of how the replacement leaders are distributed across several different categories.

TABLE 1.

Descriptive Statistics

Regime Type. I determined a state's regime type using the Polity IV data set, drawing the information from the last year the leader was in power while the war was in progress (Marshall and Jaggers Reference Marshall and Jaggers2000).Footnote ⁴⁴ Following standard practice, I subtracted the “democracy” and “autocracy” scores for the state for each observation and created a scale in which -10 represents the most autocratic states and 10 represents the states that are most democratic. I considered states with a score of -10 to 5 as nondemocracies, and states that scored between 6 and 10 as democracies.

Voluntary Participation in the War. To control for the possibility that members of a domestic audience will hold leaders who initiated wars more responsible than leaders who were attacked, and to account for the fact that initiators are more likely to win, I created a variable that takes on a value of 1 if the leader either initiated or voluntarily joined the war (65% of the states) and 0 if the state was attacked (35%). I coded this variable using secondary sources that describe the initial military clashes that mark the beginning of the war.

Battlefield Situation. Because many leaders leave office midwar, a measure of how the state was faring at the end of the leader's tenure during the war is needed to determine if battlefield developments played a role in the leader's departure from office. To this end, I constructed a variable that captures how the state was progressing during the war. This variable took a value of -1 if the state had suffered a major setback (failure to repel a major offensive, loss of ground, etc.) during (roughly) the year preceding the leader change (36% of the observations); 0 if the state had made no progress without a setback (i.e., a stalemate) (44%); and 1 if the state had made progress (20%).Footnote ⁴⁵ Importantly, I use only major events because domestic audiences are more likely to notice these given their lack of access to high-quality, fine-grained information about war developments (Berinsky Reference Berinsky2009). I use this variable for observations in which the leader leaves before the war ends, and the value of the war outcome variable for all cases in which the leader is in charge when the war ends.

The British experience in the Crimean War provides a nice example. For the first leader, Aberdeen, I used the battlefield situation variable covering the time between the United Kingdom's declaration of war in March of 1854 and early January of 1855, when Aberdeen leaves office on January 31. For the second leader of the war, Palmerston, who presides over the end of the war in 1856, but does not leave office until 1858, I used the war outcome variable. Although this may strike some as an “apples and oranges” comparison, because a final outcome is both qualitatively different and more important in the long term than a snapshot of the state's progress at a particular point in the war, this method provides the closest representation of the information citizens would have had at the time of the leader change. During Aberdeen's wartime tenure, for instance, the United Kingdom had met with limited success and had suffered a major setback with the now infamous “Charge of the Light Brigade” at the Battle of Balaklava in October 1854, in which the British lost 40% of their forces on the scene. The United Kingdom was dealt another blow soon after on Mount Inkerman, a battle that would claim the lives of 23% of all their forces dedicated to the war (Clodfelter Reference Clodfelter1992, 297). The United Kingdom and its allies, however, ended up winning the war in 1856, when the Russians dropped their territorial ambitions. This reversal of fortunes illustrates why two observations based on different sets of information are necessary. The British public in early 1855 would have been evaluating Aberdeen based on what they knew at the time (i.e., the recent failures and lack of progress), and not on knowledge of the United Kingdom's eventual win. Given this, I code Aberdeen's “outcome” as a loss and Palmerston's (which was also the war's final outcome) as a total victory.

Control Variables

Adversary's War Aims. I include the adversary's war aims as a control for three reasons. First, as I noted earlier, any leader should have incentives to prevent an adversary with extremely aggressive war aims from winning, because a loss in this case could have serious consequences for any leader, regardless of culpability. Second, because leaders set their states’ war aims strategically, they are only likely to select high war aims if they feel they are likely to achieve them. Leaders facing adversaries with high war aims, therefore, are more likely to lose. Third, the adversary's war aims proxy for how much the targeted country's citizens are likely to suffer if their state loses the war. Citizens of the targeted state, therefore, will be more upset with a leader who is unable to prevent an aggressive adversary's victory than they would be with a loss to a less aggressive opponent. I based the variable on the following broad classification of possible war aims: 1: achieving the prewar status quo; 2: extracting a policy change from the targeted state; 3: taking a portion of an ally's homeland territory; 4. taking a portion of the targeted state's homeland territory; 5: regime change in the targeted state; 6: total conquest of the targeted state. I coded adversaries as having “high” war aims if they received a coding of 4 or higher on this scale. Roughly 42% of states (110) have high war aims.

Casualties. As with the adversary's war aims variable, I control for the number of military fatalities suffered by the state, because of the strong effect scholars assume they have on the probability of leader punishment (Gartner and Segura Reference Gartner and Segura1998; Goemans Reference Goemans2000; but see Gelpi, Feaver, and Reifler Reference Gelpi, Feaver and Reifler2005/06).Footnote ⁴⁶ In states with at least one wartime leadership transition, each leader's share of the casualties was weighted to reflect his or her personal share of the war (e.g., if a leader was in power for 15% of the war's duration, I allotted him or her 15% of the war's total casualties).Footnote ⁴⁷ I log the final, weighted variable to account for a skewed distribution.

Prewar Tenure. This variable counts the number of days from the day the leader took office to the day the war began.Footnote ⁴⁸ I include this variable because some have posited that leaders with longer prewar tenures will have a stronger hold on power and, consequently, should be more likely to retain office in the event of a loss, especially in nondemocratic polities (Bueno de Mesquita and Siverson Reference Bueno de Mesquita and Siverson1995). I logged the variable to account for a skewed distribution.

Relative Capabilities. This variable captures how strong an individual state is relative to its opponent(s), and is a necessary control for models predicting war outcome. I calculated this variable as the ratio of the state's capabilities over the sum of capabilities on the battlefield. I drew data on national capabilities from EUGene; I utilized the composite capability score in the final calculations (Bennett and Stam Reference Bennett and Stam2000).