Opting Out in 2012: Military Casualties, Vote Choice, and Voter Turnout in Obama s Bid for Reelection. Christopher Gelpi.

Transcription

1 Opting Out in 2012: Military Casualties, Vote Choice, and Voter Turnout in Obama s Bid for Reelection Christopher Gelpi Kristine Kay The Ohio State University Abstract: We investigate whether American casualties in Afghanistan and Iraq influenced voter behavior in the 2012 Presidential election. Our analysis of 2012 state- level voting returns suggests that military casualties from Iraq and Afghanistan did have a significant impact on voting behavior despite the striking lack of elite attention to this issue. Unlike Bush s reelection bid in 2004, we find that casualties did not influence vote choice as measured in the incumbent s portion of the two- party vote. This result is expected, since Obama and Romney took essentially identical stands on the Afghan War. However, we find that casualties did have a significant impact on voter turnout rates both among Democrats and the electorate at large. States with higher casualty burden s as a proportion of their population experienced significantly lower rates of voter turnout.

2 Scholars have known for some time that politics does not stop at the water s edge. Numerous studies demonstrate that popular attitudes toward foreign policy are often an important determinant of presidential approval and voting behavior (Aldrich et. al. 1989; Nincic and Hinkley 1991; Anand and Krosnick 2003; Aldrich et. al. 2006). In particular, casualties from military conflicts have been an especially important determinant of incumbent Presidents ability to attain reelection. Truman and Johnson, for example, abandoned their bids for reelection in response to popular dissatisfaction with casualties from unpopular wars. George W. Bush was able to retain office despite American casualties in Iraq, but only by campaigning strongly on the war and persuading enough Americans to continue supporting the war despite the casualties (Hillygus and Shields 2005; Gelpi et. al. 2007, 2009). Even despite Bush s strong campaign on Iraq, studies of aggregate voting returns suggest that casualties cost the President votes at least outside of his electoral strongholds in the South (Karol and Miguel 2007). As Barack Obama campaigned for reelection in the fall of 2012, the United States had been at war in Afghanistan for more than a decade. And while the war was begun under George Bush, Obama consistently supported the US mission in Afghanistan throughout the 2008 campaign. Shortly after taking office in 2009, Obama sharply escalated America s military presence in Afghanistan, and in doing so he took on the mantle of responsibility for the outcome of the war. Thus the very minor role that Afghanistan seemed to play in the campaign 2012 is rather puzzling in light of the historically important role that wartime casualties have played in shaping popular evaluations of sitting presidents. In this paper we investigate whether American casualties in Afghanistan and Iraq influenced voter behavior in 2012 despite the low profile of this issue in the stump speeches and media coverage of the fall campaign. Much of the scholarship on electoral behavior emphasizes the central role of partisan elite cues (Zaller 1992; Bartels 2002; Achen and Bartels 2006; Berinsky 2007) or the news media (Iyengar and Simon 1993; Allen et. al. 1994; Baum and Groeling 2010) in shaping foreign policy attitudes. Other scholarship on American public opinion and casualties, however, suggests that individuals may respond to information about battlefield events such as casualties even without (and perhaps even despite) cues from partisan elites or the media (Mueller 1973; Gartner and Segura 1998; Gelpi et. al 2005/2006, 2009; Gartner 2008; Gelpi 2010). Our analysis of 2012 state- level voting returns suggests that military casualties from Iraq and Afghanistan did have a significant impact on voting behavior despite the striking lack of elite attention to this issue. Unlike Bush s reelection bid in 2004, we find that casualties did not influence vote choice as measured in the incumbent s portion of the two- party vote. This result is not surprising, since Obama and Romney took essentially identical stands on the Afghan War. However, we find that casualties did have a significant impact on voter turnout rates both among Democrats and the electorate at large. States with higher casualty burden s as a proportion of their population experienced significantly lower rates of voter turnout.

3 Moreover, we find that the negative impact of casualties on voter turnout was strongly contingent on the size of the military presence within the state. Casualties had a strong negative impact in states where active duty personnel, reserves, and veterans comprised a relatively high proportion of the population, but they had no significant impact on turnout when the military presence in the state was relatively low. These results yield several important implications for our understanding of the impact of war on electoral behavior. First, we find that the public is able to formulate attitudes toward war and express those attitudes in voting behavior even when elites choose not to engage the public on the issue. This result is supportive of rational expectations cost/benefit models of public attitudes toward war (Gartner 2008; Gelpi 2010). Second, we find evidence that choosing not to vote can be a deliberate expression of policy preferences rather than an expression of indifference towards policy outcomes or the lack of ability or resources to express a preference between candidates. And finally, our results highlight another way in which the unequal sharing of the burdens of war drives a growing wedge into the civil- military divide in America (Kriner and Shen 2010). The constantly shrinking share of the population that remains connected to the US military continues to carry a disproportionate share of the human costs of war. Most Americans, on the other hand, are disconnected from these costs because they are not connected to the military community. In this context, the lack of elite attention to the costs of war seems to have driven many Americans in military communities out of the voting booth and out of the political process. Military Casualties and the Reelection of Incumbent Presidents Since the end of WWII the United States has engaged in four wars that resulted in more than 1,000 US military deaths: Korea, Vietnam, Afghanistan, and Iraq. The costliness of these conflicts make them obvious candidates to influence voting behavior in any context, but such costs should be especially important in shaping electoral behavior when an incumbent president is seeking reelection because they provide such a clear opportunity for retrospective voting (Fiorina 1978). Casualties even in substantial numbers will not always make a war unpopular with the American public (Larson 1996; Eichenberg 2005; Gelpi et. al. 2009). The sustained level of public support for America s engagement in WWII should make this point clear. However, large numbers of casualties in a war that is viewed as unsuccessful or ill advised will accelerate popular opposition to the war (Jentleson 1992; Eichenberg 2005; Gelpi et. al. 2009), and should spell trouble for a presidential reelection bid. Both Harry Truman and Lyndon Johnson abandoned their bids for reelection because the public response to casualties in what were widely viewed as unsuccessful wars in Korea and Vietnam. Conversely, both the Eisenhower and Nixon campaigns were successful in 1952 and 1968 in part because they promised to extract America from these costly and unsuccessful conflicts. Wartime casualties

4 also posed a significant obstacle for George Bush s reelection campaign in With regard to Afghanistan, Bush was fortunate that American casualties remained fairly light in the fall of 2004, and the mission was highly popular in at that time. Iraq, on the other hand, was a more difficult problem. The US had suffered 1,101 soldiers and contractors killed in action by Election Day in 2004 (Karol and Miguel 2007), making it the most costly American conflict since Vietnam. And while the Iraq war was initially quite popular with the public, the failure to find the promised weapons of mass destruction combined with a mounting insurgency against the continuing US presence had made the war highly controversial. By October of 2004, polls generally showed the American public almost evenly divided on their support of Bush s handling of the war and their willingness to continue fighting it. Not surprisingly, in this context Iraq became the central issue in George Bush s campaign for reelection. The Center for Media and Public Affairs found that Iraq was the most covered news story of 2004 on the evening broadcasts of the major TV news networks, and stories on Iraq took up 44% of the combined minutes of those broadcasts. Public evaluations of the most important issues in the campaign matched closely with this distribution of coverage. Gallup consistently found that American s most frequently named Iraq as the most important problem facing the nation in Numerous studies of the 2004 election found that Iraq had a profound influence on President Bush s bid for reelection (Hillygus and Shields 2005; Norpoth and Sidman 2007; Gelpi et. al. 2007; Karol and Miguel 2007). These studies often differed in their conclusions regarding the precise nature of the impact of Iraq on the election, but a broad consensus of studies found that Iraq was central in determining voter behavior. As Obama sought reelection in the midst of the longstanding war in Afghanistan, on the other hand, politicians, the media, and much of the public seemed to be ignoring the issue. The Tyndall Report found that the Afghan War was only the tenth most covered news story by the major TV news networks. Even within the realm of foreign policy, Afghanistan was third in volume coverage behind coverage of the civil war in Syria and the attack on the American consulate in Benghazi. Once again, the priorities of the American people seemed to match this pattern of media coverage. Afghanistan often did not even make the reported list of most important problems facing America in Responses for this issue remained in the single digits and generally hovered around 3%. This pattern of media neglect and popular inattention is surprising in light of the mounting costs of the Afghan War. Shortly after taking office Barack Obama made a highly publicized decision to escalate the war with a surge similar to the one executed by President Bush in Iraq in American casualties began to mount in Afghanistan in the wake of this escalation. Thus during Obama s first term in office the United States had suffered 1,780 soldiers killed in action (KIA) in Afghanistan and Iraq. The overwhelming majority of these losses were in Afghanistan, as the Obama moved to end the US combat mission in Iraq. Moreover, while even as late as 2009 the mission in Afghanistan had initially been popular with the American public,

5 by the fall of 2012 the mission had become deeply unpopular as Obama s surge of US forces into the conflict failed to show any demonstrable progress toward a successful result. Media attention to Afghanistan had become so scant by the fall of 2012 that polling data is actually quite thin on this issue leading up to the election, but a July 2012 Quinnipiac poll found that 60% of Americans thought the US should not be involved in Afghanistan while only 31% thought the US was doing the right thing. And an October 2012 poll by Pew Research found that 60% of Americans though the US should withdraw from Afghanistan as soon as possible, while only 35% thought that the US should stay until the situation had stabilized. This level of opposition to the war in Afghanistan was significantly higher than the opposition that Bush faced to Iraq in 2004, and is closer to the levels of public opposition to Iraq that we observed in 2006 after its severe collapse into civil war. Nonetheless, neither Obama nor Romney chose to make Afghanistan a major part of their campaign in The two candidates took nearly identical positions on the issue, and Romney even endorsed Obama s September 2014 deadline for withdrawing troops. Thus while both candidates spoke generally about withdrawing US forces, both were also committed to keeping US forces in a combat role in Afghanistan for at least two more years at the time of the 2012 election. This stance placed both candidates in opposition to the views of a strong and consistent majority of the American electorate. Vote Choice, Turnout, and the Impact of Afghanistan in 2012 How should we expect popular dissatisfaction with the human costs of the war in Afghanistan to influence the 2012 election in light of the way that both the candidates and the media treated this issue? A large body of scholarship would suggest that Afghanistan should not affect voter preferences between Obama and Romney. The lack of media attention and elite partisan cues on this issue should mean that attitudes toward Afghanistan will not be salient to voters as they enter the voting both, making it unlikely that they will rely on these attitudes when making their choice (Aldrich et. al. 1989; Zaller 1992; Berinsky 2007). Moreover, the nearly identical policy stances of Obama and Romney on Afghanistan should make it impossible for voters to use this issue to form a preference for one candidate over the other (Page and Brody 1972). It is important to remember, however, that electoral behavior includes not only the selection of one candidate over another, but also the decision to go to the voting both in the first place. That is, if individuals have a strongly felt policy preference that they feel is not addressed by any of the available candidates, they do have another alternative: opt out. Thus we cannot evaluate the impact of Afghanistan on the 2012 election without considering voter turnout. Since Downs (1957) established the economic model of voting, the decision to vote has been seen as a cost- benefit analysis. The costs generally entail the time and effort it takes to decide on a candidate and submit a ballot. These costs are lower if

6 the media and campaigns effectively communicate the importance of issues and the differences between candidates. Heuristics such as group cues can be extremely helpful to individuals in decreasing these costs. More abstractly, the benefits voters consider when deciding to vote as defined by Downs are streams of utility derived from government activity, which are in turn the benefits in a citizen s mind (Downs, 1957:36). Within this cost- benefits framework, a wide variety of observational and experimental work has been done. A considerable this research has centered on what makes some people vote while others do not, regardless of specific election variables. Work that has focused on habitual voters over time has tended to highlight the effects of education and income on decreasing the costs of voting (Wolfinger & Rosenstone, 1980). Party identification also increases the likelihood of voting and is a significant predictor of which candidate a voter will support (Campbell et al., 1960). However, many Americans are one- issue voters rather than strong partisans; the differences between candidates on a specific issues are more likely to predict changes in turnout across elections, given that the main parties in any contest are the same, but candidates positions on specific issues often change from one election to the next. Research on turnout in specific elections has focused instead on the effects of campaigns (Hillygus & Jackman, 2003) and media (Dalton et al., 1998; Gerber et al, 2011; Iyengar & Kinder, 1981; Henderson, 2013), either in providing information that decreases the costs of voting, issue- framing, or agenda- setting effects that determine on which issues voters will base their candidate preferences. However, most of these studies focus on the effects of seeing an ad or news story on attitudes towards a specific issue or candidate based on that information. The influence of issues that are overlooked by the media on voting is more difficult to discern and has largely been ignored. In terms of benefits, voters may be motivated to turnout by the belief that they are helping their preferred candidate to win, but as Aldrich explained (1993), voters know that their vote is unlikely to be influential given that it is only one among many. Instead, a sense of civic duty or social pressure likely contribute more to the decision to vote. Converse (1971) suggested that there are both internal motivations to vote, i.e. the learned need for citizen participation, and external stimulation, meaning the mobilizing effects of the excitement surrounding a presidential campaign. As field experiments have found, turnout is higher when voters feel social pressure to vote (Gerber, Green & Larimer, 2008). However, implicit in this concept of responsibility is a preference for one candidate over another; if we have a responsibility to participate in elections, it is because we have a responsibility to form impressions of candidates and voice our preference for one over the other. However, if voters do not see a significant difference between candidates, then they will have little internal motivation to vote, and may not be susceptible to external stimuli.

7 Zipp (1985) was the first to empirically analyze the effect of ideological distance, called alienation, and indifference, the inability to distinguish the candidates, on turnout in presidential elections. He found an inverse relationship between alienation and turnout, and to a lesser degree between alienation and turnout, concluding that when citizens feel they do not have an option of voting for a candidate who represents their ideological position, they are less likely to vote. Plane & Gershtenson (2004), found support for the same conclusion except that the effect of indifference was stronger than that of alienation. Adams et al. (2006) also found similar results, but noted that sometimes the effects of alienation and indifference cancel each other out in that the more similar two candidates are, the more likely they are to be near the median voter, and thus the more representative they are of the public s preferences overall. Leighley & Nagler (2013) found that overall, a standard deviation increase in indifference made respondents 5-10% less likely to vote, using ANES data on presidential turnout from They further found that lower income was associated with significantly more indifference. However, the above analyses utilize data from a single election at a time, and thus do not take into account the difference between habitual voting and the decision to vote in a specific election. They also focus on ideology broadly, or an aggregate of specific issue positions that tend not to change across elections. Finally, they depend on self- reported voting, which is known to be upwardly biased (Holbrook & Krosnick, 2010). Unlike some of the previous work on turnout, our dependent variable is the change in turnout from 2008 to 2012 measured in terms of actual voting returns which cannot have the upward bias of self- reported turnout. Moreover, instead of focusing on ideological distance, we investigate the effect of the specific issue of war and war casualties on both vote choice and turnout. Hypotheses on Afghanistan and the 2012 Election Elite cue models of electoral behavior (Zaller 1992; Bartels 2002; Achen and Bartels 2006; Berinsky 2007) would suggest that casualties from the war in Afghanistan will have little or no impact either on candidate preference or on voter turnout because the public will not draw upon these attitudes in making either of these decisions in the absence of cues from elites to guide them. A negligible impact on candidate preference is over determined, since the lack of policy distance between Obama and Romney on this issue makes it impossible for voters to use Afghanistan to form a candidate preference even if they wanted to do so. However, an elite cue approach would also expect that Afghanistan will have a negligible impact on voter turnout as well, since the lack of elite attention to Afghanistan should mean that voters will not rely on attitudes toward the war when deciding whether or not to turn out to vote. Much of the growing literature on public attitudes toward war, on the other hand, suggests that individuals update their attitudes in a rational expectations fashion

8 based on available information from the battlefield about the progress of the war. (Larson 2000; Gartner 2008; Gelpi 2010). One of the most salient and consistent pieces of information that individuals can obtain from the battlefield is information on casualties (Mueller 1973; Gartner and Segura 1998; Gartner 2008). Moreover, information on casualties is likely to remain relatively consistently available to the public, even if politicians and the media pay relatively little attention to the war. Because of a strong and widely held desire among political elites and the media to recognize the service and sacrifice of members of the military, national news outlets have continued to report at least episodically on events that result in the deaths of American military personnel. While the news media may not frequently remind the public of the total number of American deaths in Afghanistan as they did throughout much of the Iraq War, the strong norm of reporting the deaths of US military personnel ensures that the public continues to hear episodic reports of casualties as the war continues. Moreover, reporting military casualties is one of the few ways in which local news reporting regularly and reliably connects with foreign policy issues. Local news sources whether print or television regularly report on the deaths of soldiers in their home state. Thus a rational expectations approach to foreign policy attitudes suggests that the public will continue to update its attitudes toward Afghanistan as they receive information about battlefield events such as casualties. In addition, as the public becomes increasingly skeptical about the wisdom and likely success of the mission in Afghanistan, new information about casualties is likely to have a larger impact on opposition to the war. These two perspectives on attitude formation do not differ from one another with regard to the impact of casualties in Afghanistan on candidate preference in Both arguments predict no effect for casualties because the lack of candidate differentiation on this issue makes it impossible for individuals to use their views on Afghanistan to choose a candidate. Hypothesis 1: Casualties from the wars in Afghanistan and Iraq will have no significant impact on President Obama s margin of victory. The two perspectives differ sharply, however, on the impact of casualties on turnout. As noted above, an elite cue perspective suggests that casualties will have no significant impact on turnout because attitudes toward Afghanistan will not be cognitively accessible to voters. The rational expectations approach, on the other hand, suggests that casualties in Afghanistan should significantly reduce voter turnout. As casualties rise in a conflict that is widely viewed as unsuccessful and unwise, voters should have increasingly intense negative views on the war. Faced with a situation in which they cannot express their dissatisfaction through voting because the candidates are indistinguishable on the issue voters will express their dissatisfaction by opting out of the electoral process.

9 Hypothesis 2: Casualties from the wars in Afghanistan and Iraq will reduce voter turnout. Finally, we expect that the impact of casualties on voter turnout will depend on the intensity with which voters feel the costs of the human losses. While the burden of casualties in war has never been entirely evenly distributed across society, the recent wars in Iraq and Afghanistan have been marked by an especially unequal distribution. In 2012 less than 1% of American adults were serving in the military. While the military s active duty presence in American society was small in 2012, this pattern has been building for decades. The shift to an all- volunteer force in 1973 combined with the de- escalation of the Vietnam War substantially shrank the size of the US military in the early 1970 s. And the end of the Cold War and the Gulf War brought another substantial decline in the early 1990s. Thus the military has been producing a shrinking number of veterans for more than 40 years. In 2012 Gallup reported that a majority of American men over the age of 65 were military veterans, while only 12% of men age 25 to 34 were veterans. 1 Thus a shrinking slice of the American population is connected to the military either through current or previous service. The reliance on a smaller all- volunteer force has also led to greater geographic variation in terms of the military s presence in society. In many states, such as New York, New Jersey and California active duty personnel, reserves and veterans make up only about 5% of the state s population. In other states, such as Hawaii, the military personnel and veterans make up nearly 15% of the population. We expect the presence of military personnel and veterans to have an important impact on responses to casualties. Numerous studies suggest that individuals with strong ties to the military feel more strongly about the human costs of military operations (Feaver and Gelpi 2004; Gartner 2008b). Thus we expect that casualties will have a more negative impact in communities where a larger proportion of voters have a direct connection to the military. 2 Hypothesis 3: The negative impact of casualties on turnout will depend on the level of military presence in society. Data 1 This pattern is reversed among women, of course. But female vets remain a small percentage of the veteran population, so the overall pattern strongly reflects the experiences of men. 2 It is worth noting that this pattern could occur through a variety of causal mechanisms. It could be that military personnel and veterans themselves respond more strongly to casualties. Individuals who have family members or friends serving in the military may respond more strongly to casualties. Or it could be the case that all citizens of the community respond more strongly to casualties when the military is a salient institution in the local community.

10 We test these hypotheses with data on casualties and voting returns at the state level. Previous studies of the impact of casualties on aggregate voting behavior have found the most substantial effects at the state level rather than the county or media- market level (Karol and Miguel 2007). This pattern seems to be due to the fact that local media generally report deaths of any solider that is from the same state regardless of what county or media market the soldier was from. We collected data on the number of US military from each state personnel killed in action in Afghanistan and Iraq from January 2009 (when Obama took office) through October of These data were obtained from the Department of Defense. We cannot disentangle the impact of casualties from Iraq and Afghanistan because the two variables are correlated at Thus we measure the total number of casualties in each state from these two conflicts. However, this total is heavily driven by deaths in Afghanistan, which are an order of magnitude larger than the deaths in Iraq. Consistent with much of the literature on casualty tolerance, we take the log of casualties suffered in each state (Mueller, 1973). The log transformation is appropriate because we expect that casualties will have a declining marginal impact on popular attitudes as the public becomes desensitized to the costs of war and as the media devotes declining attention to casualties as they cease to be news. 3 Finally, we weight the log of casualties by the population of the state, since smaller communities will feel the burden of casualties more sharply with a lower number of deaths. Thus our final measure of the casualty burden for each state is the log of the number of soldiers killed in Afghanistan and Iraq during Obama s first term in office who from that state per 100,000 total population in the state. We also measure the percent of the population with connections to the military as the total number of active duty personnel, reserves, and veterans as a proportion of the state s total population. Figure 1 About Here Figure 1 describes the relationship between state casualty burden and the percent military population. Not surprisingly, these two variables are fairly strongly correlated (r=.57, p<.01). States with large military populations tend to be the states that suffer a higher proportion of casualties. The strong relationship between these two variables suggests that we should use some caution in generating counterfactual predicted effects from our model. For example, our model will generate a predicted vote margin and turnout rate for state with the casualty burden of Wyoming and the military population of New York, but Figure 1 clearly indicates that no such state existed in Thus we would be reaching well outside the bounds of our data to make such a prediction. But while caution is warranted in extrapolating our model carefully, the relationship between casualties and percent military is not so strong as to preclude further analysis. 3 However, our results remain consistent even if we do not log casualties.

11 Our remaining data were also collected from the US government sources. Specifically, we collected data on Obama s margin of victory (or defeat) as a percentage of the two- party vote in each state in 2008 and 2012 as well as data on turnout by Democratic voters and all voters in 2008 and We measure Democratic and total turnout respectively as the number of votes cast for Obama and the number of votes cast in total each divided by the voting age population. Since voting behavior is often driven by the state of the economy (Kramer 1971; Tufte 1975; Hibbs et. al. 1982), we also collected data on the rate of GDP growth in each state in 2012 as well as the change in GDP growth rate between 2008 and Not surprisingly, given his status as the first African- American president, race played an important role in both the 2008 and 2012 campaigns. Thus we collected data on the percentage of African- American and non- Hispanic white voters in each state. Finally, we coded variables that identify the home states of the Republican candidates in 2008 and In order to isolate the impact of casualties on vote choice and turnout, we focus on the change in vote margins and turnout rates between 2008 and This procedure allows us to control for a great variety of factors that determine voting and turnout decisions in each state and focus our analysis only on things that changed between 2008 and 2012 such as casualties. Thus we construct variables that capture the change in GDP growth, the percentage of black and white voters, and the change in the Republican ticket by subtracting the 2008 value from the 2012 values. In addition, we include Obama s margin and the state turnout rate in 2008 because we expect that these variables may regress to the mean. That is, if Obama receives an unusually high margin in 2008, we expect that the margin will decrease toward the mean value in We also include the GDP growth rate for 2012 because voters may focus heavily on recent economic conditions. Figure 2 About Here Figure 2 displays the relationship between the change in Obama s margin of victory between 2008 and 2012 and the change in overall turnout rate between these elections. The figure demonstrates that unlike casualties and military presence, margin of victory and turnout rates are quite separate concepts. The two variables are not correlated with one another except among the 4 states that are affected by the change in the Presidential ticket. Alaska and Arizona saw significant drops in turnout as well as increases in Obama s margin as McCain and Palin left the ticket. Utah and Wisconsin, on the other hand, experienced increases in turnout along with a decrease in Obama s margin in response to local support for Romney and Ryan. Empirical Results Since our dependent variables are continuous and roughly normally distributed, ordinary least squares regression is an appropriate tool for our statistical model.

12 The results of our analysis of the change in Obama s margin of victory are displayed in Table 1. Table 1 About here Consistent with hypothesis 1, we find that casualties have no significant impact on changes in Obama s margin of victory at the state level. The first analysis in Table 1 models the impact of casualties as a constant effect, while the analysis in the second column makes their impact contingent on the size of the state s military population. The impact of casualties in the first model is negative but does not achieve statistical significance. In the second model neither the casualties nor its interaction with military population approaches statistical significance. 4 But while casualties have no impact on Obama s vote margin, other variables in our model do a good job of capturing much of this variation. Obama s margin in 08 has a positive impact on the change in margin. This result indicates that instead of regressing to the mean, Obama continued to perform strongly in areas where he was strong in Areas where Obama was weaker in 2008, on the other hand, saw greater erosion. This pattern helps explain why Obama was able to perform so strongly in the Electoral College despite the much closer popular vote. Between 2008 and 2012 Obama lost votes in states where he was already likely to lose. Changes in the population of non- Hispanic white voters and African- American voters also had a strong impact on changes in Obama s margin between 08 and 12. Change in the percent of African- American voters ranges by about 0.08, so the coefficient for this variable suggests that moving from the minimum to the maximum on this variable increases Obama s margin by about 5%. The coefficient for changes in percent white appears substantively smaller, but this variable has greater variance. So moving form the minimum to the maximum on this variable reduces Obama s margin by about 4%. Finally, geographic changes in support networks of the Republican candidates also had a significant impact on Obama s margin of victory. One somewhat unusual result from the model is that GDP growth in 2012 had a negative and statistically significant impact on Obama s margin of victory. This result is directly opposite to the expectations of the economic voting literature. That is, Obama s vote margin declined by nearly 2% in states with the highest growth rates in We can only speculate about the causes of this anomalous result, however, we believe that states with higher GDP growth in 2012 may also have tended to have Republican governors who claim credit for the economic progress and campaign against Obama. 4 Auxiliary regressions indicated that multicolinearity is not a problem in our analyses. The auxiliary r- squared for the log of casualties per 100,000 population is only Residuals from both models of vote margin conform closely to a normal distribution, suggesting that our model specification is appropriate.

13 Table 2 displays our analyses of the impact of casualties on voter turnout both among Democrats and the public at large. The first two columns in Table 2 depict our models of turnout for Obama, while the third and fourth columns depict our models of turnout for both Democratic and Republican candidates. The first and third columns model a constant effect for casualties on turnout, while the second and fourth columns allow the impact of casualties to depend on the percent military population in the state. 5 Table 2 About Here Consistent with hypothesis 2, the first and third columns in Table 2 indicate that casualties have a negative and statistically significant impact both for Democratic turnout and total turnout. As Figure 1 indicates, the log of casualties per 100k population varies.014 to.36. Thus the models suggest that a shift in casualties from the 5 th to the 95 th percentile reduces Democratic turnout by about 2.75% and reduces overall turnout by 4.5%. These are substantial changes in turnout rates in response to casualties that as we shall see below - are comparable to the effects of changes in economic growth, and changes in the racial makeup of the electorate. This result is quite striking because of the minimal and tangential role that Afghanistan played in the elite debate surrounding the 2012 campaign. Neither the candidates nor the news media paid much attention to the Afghan War. Yet despite this inattention from elites, dissatisfaction with casualties from the wars was substantial enough to motivate large numbers of voters to opt out of the electoral process. Turning to the control variables in the model, we can see that economic growth had no significant impact on turnout rates for Obama, but GDP growth in 2012 did have a positive and significant impact on total voter turnout. A shift from the 5 th to the 95 th percentile in 2012 growth increased overall turnout by just over 3.5%. Changes in the African- American and non- Hispanic white populations continued to have an impact on turnout as well as Obama s vote margin. Varying the change in the percent of white non- Hispanic voters in the state from the 5 th to the 95 th percentile reduces both Democratic and total turnout by about 3%. Conversely, increasing the change in the percent African- American population from the 5 th to the 95 th percentile increases Democratic turnout by a little less than 2.5% and increases total turnout by just over 3%. Changes in the Republican ticket have a substantial impact on overall turnout, but not surprisingly have no impact on Democratic turnout. 5 Auxiliary regressions indicated that multicolinearity is not a problem in our analyses. The auxiliary r- squared for the log of casualties per 100,000 population is only Residuals from both models of vote margin conform closely to a normal distribution, suggesting that our model specification is appropriate.

14 Columns 2 and 4 of Table 2 test our argument that the impact of casualties on turnout will depend on the extent to which the state s population is connected to the military as an institution. Here we find strong support for hypothesis 3. Because the model is interactive, it is important to interpret the coefficients for casualties, military population, and their interaction with some care. For example, in these models the coefficient for the log of casualties per 100k population now represents the impact of that variable when percent military population is equal to zero. However, the percent military population never drops below 5% (or.05) in any state. So the estimated coefficient for casualties does not represent an effect that is ever observed in our dataset. Similarly, the coefficient for the interaction between casualties and military population represents the change in the coefficient for casualties that would result from a one- unit change in military population. But again, military population only varies from.05 to.15 in our data. So the coefficient for casualties can only change by one tenth of the coefficient for the interaction term. Finally, the coefficient for military population represents the impact of a one- unit change in military population when casualties are zero. No state actually suffered zero casualties, although North Dakota and Delaware came close with one casualty each. Thus the best way to evaluate the contingent impact of casualties on turnout is to estimate the coefficient and confidence interval at different levels of military population that we observe in our dataset. Figure 3 displays the estimated coefficient and 95% confidence intervals for the impact of casualties on Democratic turnout and total turnout when the percent military population is set at the 5 th and 95 th percentiles of its variation. Here we can see that when the military does not have a strong presence within a particular state such as New York or New Jersey casualties do not have a significant impact either on Democratic turnout or on total turnout. Specifically the coefficient for Democratic turnout is near zero and does not approach statistical significance, and the coefficient for total turnout is slightly positive (i.e. casualties increase turnout) but does not achieve statistical significance. In states with a higher percentage of the population associated with the military, however, the impact of casualties on turnout becomes sharply negative. Thus for states like Hawaii, Virginia, and Alaska casualties have a significant depressive effect on voter turnout rates. Figure 3 About Here How large are these contingent effects overall and how much did they matter in the 2012 election? Giving a precise answer is somewhat complicated, because casualties have a slightly different impact in every state, depending on the percent military population. However, we provide one estimate of the overall impact of casualties on voter turnout by taking the fourth model in Table 2 and calculating the predicted change in voter turnout between 2008 and 2012 if every state had suffered zero casualties in Afghanistan and Iraq. As noted above, this prediction represents something of an extrapolation outside our dataset. But since some states like DE and ND came close to suffering zero casualties, setting casualties to zero

15 is not so far outside our observed range of data. Moreover, setting casualties to zero for all states provides a hypothetical example of what turnout might have been like in 2012 in the absence of the Afghan War. Figure 4 About Here Predicted changes in turnout rate in our zero casualties scenario are displayed in Figure 4. The left- hand side of Figure 4 displays the impact for states that have a percent military population that is below the median. The vertical axis represents the predicted change in total turnout if the US had suffered no military deaths in Iraq or Afganiststan during Obama s first term. The horizontal axis represents the log of casualties per 100k population. The figure clearly illustrates that casualties had at best a modest impact on turnout in states with low military populations. First, it is worth noting that the casualty burdens for most of the low military states were already relatively low. The predictions for changes in turnout in the zero casualty scenario are clustered fairly closely around zero with only a very modest upward slope to the expected change in turnout if casualties had been zero. The right- hand side of Figure 4, on the other hand, displays the predicted changes in turnout among high military population states if casualties from Iraq and Afghanistan had been zero. It is worth noting that casualty burdens tended to be higher in these states, so making projections with regard to zero casualties is a bit farther outside our observed data here. Nonetheless, there are heavily military states such as Florida and North Carolina that have fairly light casualty burdens relative to their population size that are comparable to the casualty burdens in low military states such as New York and New Jersey. In this half of the figure we can see that the projected changes in turnout are much more substantial and tend to be positive. The predicted effects have a much stronger upward slope as the observed casualty burden increases. Thus the model suggests that highly military state with high casualty burdens such as Maine and Hawaii would have had about an 8% higher turnout rate had they not suffered any casualties in Afghanistan and Iraq. The predictions in Figure 4 demonstrate that the total impact of casualties on turnout was negligible in some corners of America, but quite substantial in others. But as noted above, the zero casualty scenario involves some extrapolation outside our observed data, and so these illustrative projections should be viewed with some caution. As a result, we also made some more modest comparisons of casualty rates between states within our observed data range in order to get a sense of the likely effects that smaller shifts in the geographic distribution of casualties might have had on turnout. For example, instead of imagining a 2012 campaign in which there were no casualties in Iraq and Afghanistan, we could imagine a world in which the casualty burden was distributed evenly on a per capita basis across all 50 states and DC. How would turnout have changed under this scenario in some of the higher military states that shouldered a disproportionate casualty burden? A relatively high 10.8%

16 of Maine residents are active duty military, reservists, or veterans, and the Pine Tree state suffered one soldier killed in action per 100,000 residents. This placed Maine s casualty burden in the 95 th percentile, well above the.62 KIA per 100k population that was the median across the country. Had main suffered only.62 KIA per 100,000 residents, our model suggests turnout in the Pine Tree State would have increased by 3.2% or about 34,000 voters. Similarly, Hawaii is a state with a very high military presence. 13.5% of Hawaiians are active duty military, reservists, or veterans, and the Pineapple State suffered a disproportionate 0.92 soldiers killed in action per 100,000 residents. This number of casualties placed Hawaii in the 90 th percentile in terms of casualty burden. Had Hawaiians suffered only 0.62 KIA per 100,000 residents, our model suggests that turnout in the Pineapple State would have been 4.8% higher. This change would have sent over 47,000 more Hawaiians to the polls. Conclusions The ongoing war in Afghanistan is one of only four conflicts since World War II to result in more than 1,000 American military personnel killed in action. Whether by choice or not, previous incumbent wartime presidents have made the conduct of war a central issue in their reelection campaigns. In stark contrast to Korea in 1952, Vietnam in 1968 and 1972, and Iraq in 2004, Afghanistan received scant attention from both candidates and from the news media in In this paper we have investigated whether and how American military casualties in Afghanistan and Iraq may have shaped voting behavior in 2012 despite the lack of elite attention to the war. Our analysis of state- level voting returns suggests that military casualties from Iraq and Afghanistan did have a significant impact on voting behavior even in the absence of elite cues on the issue. Unlike Bush s reelection bid in 2004, we find that casualties did not influence vote choice as measured in the incumbent s portion of the two- party vote. This result is not surprising, since Obama and Romney took essentially identical stands on the Afghan War. However, we also find that casualties had a significant impact on voter turnout rates both among Democrats and the electorate at large. States with higher casualty burden s as a proportion of their population experienced significantly lower rates of voter turnout. Moreover, we find that the negative impact of casualties on voter turnout was strongly contingent on the size of the military presence within the state. Casualties had a strong negative impact in states where active duty personnel, reserves, and veterans comprised a relatively high proportion of the population, but they had no significant impact on turnout when the military presence in the state was relatively low. These results yield several important implications for our understanding of the impact of war on electoral behavior. First, we find that the public is able to formulate attitudes toward war and express those attitudes in voting behavior even when elites choose not to engage the public on the issue. This result is supportive of

17 rational expectations cost/benefit models of public attitudes toward war (Gartner 2008; Gelpi 2010). Second, consistent with the literature on alienation and voter turnout (Zipp 1985; Plane and Gershtenson 2004), we find evidence that choosing not to vote can be a deliberate expression of policy preferences rather than an expression of indifference towards policy outcomes or the lack of ability or resources to express a preference between candidates. And finally, our results highlight another way in which the unequal sharing of the burdens of war drives a growing wedge into the civil- military divide in America (Kriner and Shen 2010). The constantly shrinking share of the population that remains connected to the US military continues to carry a disproportionate share of the human costs of war. Most Americans, on the other hand, are disconnected from these costs because they are not connected to the military community. In this context, the lack of elite attention to the costs of war seems to have driven many Americans in military communities out of the voting booth and out of the political process.

18 Table 1: Casualties and Changes in Obama s Margin of Victory Constant Effect Contingent Effects Log KIA per 100k (1.00) (0.79) Log Cas X Mil Pop (0.59) % Military Population (0.50) (0.10) Obama Margin (2.49)* (2.50)* GDP Growth (2.70)** (2.63)* Chg in Growth (0.32) (0.33) Change % White (2.28)* (2.21)* Change % Black (2.46)* (2.31)* Change Rep. Ticket (3.69)** (3.74)** Constant (1.40) (0.52) R N Huber-White Robust Standard Errors; * p<0.05; ** p<0.01; Numbers in parentheses are t-ratios

19 Table 2: Casualties and Changes in Democratic and Total Turnout Dem Turnout Constant Effect Dem Turnout Contingent Effect Total Turnout Constant Effect Total Turnout Contingent Effect Log KIA per 100k (2.71)** (1.46) (2.62)* (2.22)* Log Cas X MilPop (2.27)* (2.64)* % Military Pop (1.46) (2.16)* (1.42) (3.11)** Dem Turnout (1.29) (1.37) Turnout (0.32) (0.45) GDP Growth (0.52) (0.44) (2.97)** (2.89)** Chg GDP Growth (0.95) (0.93) (1.68) (1.60) Chg % White (3.11)** (3.17)** (1.95) (2.16)* Chg % Black (2.30)* (2.44)* (1.83) (2.14)* Chg Rep Ticket (1.40) (0.96) (2.73)** (3.00)** Constant (1.54) (2.20)* (1.46) (3.06)** R N Huber-White Robust Standard Errors; * p<0.05; ** p<0.01; Numbers in parentheses are t-ratios

20 Figure 1: Log of Casualties and the Percent Military Population

21 Figure 2: Change in the Margin of Victory and Change in Voter Turnout 08-12

22 Figure 3: Impact of Casualties on Democratic Turnout and Total Turnout by Military Population

23 Figure 4: Change in Total Voter Turnout Predicted if Casualties Were Zero