News from the Frontlines: An Experimental Study of Foreign Policy Issues and Presidential Vote Choice

News from the Frontlines: An Experimental Study of Foreign Policy Issues and Presidential Vote Choice Christopher Gelpi Duke University Draft please do not cite without permission Comments are very welcome and can be sent to gelpi@duke.edu January 18, 2011 Word Count: 11,354

Abstract (148 Words) Recent literature on vote choice suggests that foreign policy issues can have an important impact on voting behavior. But much of the evidence for this claim is based on crosssectional survey data that can severely over-estimate the independent impact of attitudes on vote choice because of endogeneity bias. Data from a survey experiment on voting intentions in September and October 2008 indicate that exposure to news about the Iraq War had a significant exogenous impact on vote choice in 2008. Exposure to news about Afghanistan, however, had no effect. At the same time, these experimental data also suggest that cross-sectional models may significantly over-estimate the impact of foreign policy attitudes. I compare three mechanisms by which news about Iraq could shape voting preferences: candidate competence, issue ownership, and attitude activation. The data provide support for both the issue ownership and attitude activation models, but not for candidate competence.

For decades the consensus view in the vote choice literature held that foreign policy attitudes were of little relevance for voting behavior (Almond, 1950; Campbell et. al. 1960; Converse 1964; Stokes, 1966; Kagay and Caldiera, 1975; Light and Lake, 1985). Much of the early literature on public opinion found little structure in the public s attitudes regarding foreign policy and the Cold War consensus across the American foreign policy establishment meant that the partisan differences over foreign policy were modest at best. 1 But the traumatic experiences of 1968 both in Vietnam and here at home - brought an end to the Cold War foreign policy consensus, and in 1972 we began to see general election contests in which Democrats and Republicans took substantially different positions on foreign policy issues (Aldrich, 1977). At the same time, a growing literature suggested that public attitudes toward foreign policy were coherent and only appeared unstructured because they did not always align with the cleavages that structured domestic issues. Gradually a scholarly consensus developed that the public was generally rational in the sense that its attitudes were relatively stable over time and seemed to respond in reasonable ways to events that occurred in the real world (Caspary, 1970; Page and Shapiro, 1982; Shapiro and Page, 1988; Nincic, 1992, Adrich et. al. 2006). Aldrich, Sullivan, and Borgida (1989) demonstrated that foreign policy views had a large impact on most presidential elections since 1952 and had at least some impact on every presidential general election except for 1976. Subsequent research suggests that 1 Even when foreign policy disputes among political elites were intense, they tended not to align with party cleavages. With regard to the Vietnam War, for example, Democratic nominee Hubert Humphrey held views much closer to those of his general election opponent, Richard Nixon, than he did with Democratic challengers such as Robert Kennedy and Eugene McCarthy (Page and Brody, 1972).

foreign policy issues were also important in shaping electoral choice in the 2000 and 2004 elections (Anand and Krosnick, 2003; Norpoth and Sidman, 2007; Gelpi, Feaver and Reifler, 2007, Karol and Miguel, 2007). But important questions remain regarding the causal impact of foreign policy attitudes on voting. Most importantly, much of the evidence regarding the impact of foreign policy issues on vote choice is based on analyses of cross-sectional data. Recent work on issue framing and vote choice, however, indicates that voters may shape their attitudes to match the policy positions of candidates for whom they already intend to vote (Lenz 2009). This potential endogeneity problem suggests that cross-sectional models of vote choice may substantially over-estimate the impact of foreign policy attitudes. This problem is likely to be especially severe for issues such as the Iraq War that are very closely linked to partisanship (Jacobson 2007). I investigate the exogenous impact of foreign policy attitudes on vote choice through a pair of survey experiments that circumvent the problem of endogeneity bias. The results support the claim that foreign policy attitudes that are cognitively available and accessible can influence vote choice when candidates take distinct positions on the issue. At the same time, the results also suggest that cross-sectional models can substantially over-estimate the impact of foreign policy attitudes on voting behavior. I evaluate three mechanisms by which available and accessible attitudes may influence vote choice: candidate competence, partisan issue ownership, and attitude activation. The data provide support for both the issue ownership and attitude activation models, but not for candidate competence. Consistent with the attitude activation model, I find that priming subjects with news about the Iraq War undermined support for Obama

among those who disapproved of his plans to announce a timetable for withdrawal from Iraq. At the same time, I find that priming subjects with news about Iraq did not increase support for Obama among those who agreed with his policy position. This result is consistent with arguments about Republican ownership of foreign policy issues especially those related to military security. Candidate Competence, Attitude Activation, and Issue Ownership Foreign policy issues may influence voting behavior in a variety of ways. First, foreign policy issues may influence voting preferences by causing voters to gravitate toward the candidate who appears to be more competent or effective in handling this issue. The candidate competence model seems especially plausible with regard to Iraq in 2008, because the core of the debate in this election centered on ending the conflict successfully. The wisdom of launching the Iraq War seems most easily framed as a positional issue. That is, candidates and voters may plausibly support or oppose the launching of the war. The resolution of the Iraq War, on the other hand, is a valence issue. That is, all candidates and voters would prefer the same result: a swift conclusion to the war with as few American casualties as possible while leaving a stable and friendly government in power in Iraq (see Stokes 1963 on positional and valence issues). Valence issues tend to highlight candidate competence in achieving on an outcome that everyone recognizes as good (Enelow and Hinich 1982; Miller and Wattenberg, 1985; Green 2008). Thus the foreign policy debate in 2008 centered not on whether we should try to succeed in Iraq and Afghanistan, but on which candidate was more likely to achieve these goals.

Numerous observers suggested that McCain s candidacy depended critically on the success of the surge because he was so closely associated with that policy (Santora 2007; Balz 2008; Sanger 2008; Shear and Murray 2008). McCain himself seemed to endorse this view in an interview with Chris Matthews in which he repeatedly rejected debates over the launching of the Iraq War in favor of a focus on the successful conclusion of the war. Even Obama, who tended to focus more frequently on the decision to launch the war, framed the question as a matter of judgment. This phrasing also frames Iraq as a valence issue who has better judgment (i.e. competence) in foreign policy rather than different preferences over the use of military force. Afghanistan was also frequently discussed as a valence issue. In this case, Obama and McCain agreed on the decision to attack Afghanistan in the wake of 9/11. Thus when Afghanistan arose as an issue the candidates focused on differences in strategy and tactics over how to win the war. Given the centrality of the debate over the surge as a strategy for winning the Iraq War, the candidate competence model predicts that the impact of foreign policy issues will depend on the content of the information that voters obtain about the candidate s competence on the issue. 2 Exposure to information suggesting that the situation in Iraq is improving, for example, should highlight McCain s leadership (i.e. competence) in choosing effective policies in Iraq. Conversely, exposure to information 2 Much of the literature on pubic support for war suggests that individuals vary their level of support for a conflict as they obtain new information about the costs and the likely success of the mission (Gartner and Segura 1998, Gartner 2008, Gelpi, Feaver and Reifler 2009, Gelpi 2010). This logic is roughly consistent with the first priming mechanism discussed here. It is worth noting, however, that this same literature finds that vote choice operates through somewhat different cognitive mechanisms than war support (Gelpi, Reifler, and Feaver 2007; Gelpi, Feaver and Reifler 2009)

suggesting continued failure in Iraq should highlight Obama s good judgment (i.e. competence) in critiquing ineffective policies. The attitude activation model, on the other hand, treats foreign policy issues as positional rather than valences issue and expects that they will influence voting preferences by drawing attention to issues where voters agree or disagree with the positions taken by candidates. The central mechanism behind the attitude activation model is based on an extensive literature on spatial voting that traces its intellectual roots back to Downs (1957). According to this view, attention paid to foreign policy issues whether through news reporting or campaign messages will cause voters to emphasize those issues in evaluating the distance between their own policy position and those taken by the candidates. The attitude activation model suggests that foreign policy issues will influence vote choice by changing voters estimations of the relative policy distance between themselves and the candidates. 3 Increased attention to a particular foreign policy issue will cause voters to weight that issue more heavily in their spatial calculations, leading them to alter their candidate preferences. The most relevant positional issue with regard to Iraq in 2008 was setting a timetable for withdrawal. McCain and Obama differed substantially and vocally on this issue. Obama was a clear and consistent advocate for a timetable for withdrawal, and setting a firm date for withdrawal from Iraq was one of the central planks of his platform. McCain, on the other hand, was adamantly opposed to any 3 Previous research on the impact of Iraq in the 2004 election suggests that it operated as a positional rather than a valence issue (Gelpi, Reifer, and Feaver 2007; Norpoth and Sidman 2007). Specifically, these studies found that individuals tended to vote based on their evaluation of the wisdom Bush s decision to launch the Iraq War rather than vote prospectively on their perceptions that Bush would deliver success in Iraq.

explicit timetable and he held tightly to this position even as the Bush Administration quietly eased its way toward an implicit timeline for ending America s combat mission in Iraq. Positional differences between McCain and Obama were more muted with regard to Afghanistan. However, the clearest positional stance on this issue was Obama s commitment to increase troop levels in Afghanistan. Unlike the candidate competence model, the issue activation model does not expect that the content of the information that individuals receive about an issue (e.g. success or failure in Iraq) will be important in shaping the impact of the issue on their voting intentions. Instead, the issue activation model suggests that increased attention to a foreign policy issue will increase support for a candidate among voters who agree with his or her position on the issue, and will decrease support for a candidate among voters who disagree with his or her position. This pattern should hold true regardless of the positive or negative nature of the information about the issue. 4 Finally, the issue ownership model suggests that attention to foreign policy issues will shift voter support toward the political party that is perceived as having ownership over that issue. Like the candidate competence model, the issue ownership approach tends to view issues from a valence rather than a positional perspective. That is, the ownership model presumes that voters view issues primarily as problems that require solutions. The central difference between the candidate competence and issue ownership models is that the latter approach presumes that voters attach levels of competence in fixing problems to political parties rather than to candidates (Budge and Farlie 1983; 4 This model is also consistent with arguments about the use of priming to persuade cross-pressured issue voters to support candidates from the opposing party (Hillygus and Shields 2008). Thus we would expect attention to Iraq will push Democrats who are opposed to a timetable for withdrawal most powerfully toward McCain,

Petrocik 1996; Petrocik, et al. 2003). While individual candidates may vary to some extent in terms of their reputations for competence in dealing with particular issues, the issue ownership model expects that parties will develop and maintain fairly stable reputations for handling particular issues that allow them to recruit, maintain, and motivate their core constituents. These reputations may have a powerful effect on voters candidate evaluations, depending on the issues that become salient in a particular campaign. Candidates from a political party that owns a salient issue will attract greater support because the public presumes that the candidate will be competent in addressing that issue. Because of the relative inflexibility of these reputations, candidates will have difficulty in persuading voters of their competence in handling an issue if the opposing party owns the issue. Instead of fighting to be proven right on an opponent s home turf, this argument suggests that candidates are better off changing the focus of the campaign to an issue that their party owns. While not every issue in American politics is clearly owned by the Democratic or Republican Party, foreign policy issues and especially military policy issues have clearly been owned by the Republican Party since at least the end of the Vietnam War (Petrocik 1996; Petrocik, Benoit and Hansen 2003). Polling during the fall of 2008 also suggested that this issue ownership extended to the Republican Party and John McCain (Sussman 2008). Thus according to the issue ownership model, increased attention to foreign policy issues such as Iraq and Afghanistan should decrease support for the Democratic candidate. This pattern should hold regardless of the extent of policy agreement between the candidate and the voter (as emphasized by the attitude activation model), and regardless of foreign policy events that may vindicate or impugn the

competence of the particular candidate on the issue (as emphasized by the candidate competence model). An Experimental Approach to the Study of Foreign Policy Voting In order to assess the exogenous impact of foreign policy issues on vote choice in 2008, I construct a pair of experiments that randomly assign subjects to exposure to news stories about the wars in Iraq or Afghanistan. I created separate experiments to study the impact of each war. In each experiment subjects are randomly assigned to one of three treatment groups: 1) positive news events from the war, 2) negative news events from the war, and 3) no news events. The positive and negative news events stories were similar in both format and length. Each story focused on a raid by US military forces on an Iraqi or Afghan target and each story identified the raid as part of a broader crackdown on insurgents. The positive story then mentions a reduction in the overall level of violence against US forces. The negative story, on the other hand, mentions continued bombings and violence. All events in both the positive and negative news stories were reported in mainstream US newspapers between September 2007 and September 2008. Since news reporting on wartime events is often accompanied by partisan political commentary on the war, I also randomly assigned subjects to be exposed to partisan commentary on the war. The comments in each treatment were made by a fictional US Senator. As with the news events, I randomly varied the positive or negative tone of the commentary. In addition, I randomly varied the party identification of the fictional senator.

Respondent s substantive positions on the wars in Iraq and Afghanistan may be endogenous to their preference for Obama or McCain. But the salience of these issue attitudes will be exogenously manipulated by the exposure to news stories. Changes in voting intention that are associated with an exogenous change in the salience of these issues must result from the increased emphasis that the issue receives in the voter s calculations of candidate preference. Consequently, I attribute changes in voting intentions that result from the experimental treatments to the causal effect of Iraq or Afghanistan on vote choice. This method of estimating the impact of Iraq and Afghanistan is conservative in the sense that it presumes that the substance of issue attitudes are endogenous to vote choice, but it does ensure that the remaining estimated impact is an exogenous influence of a foreign policy issue. Data on Foreign Policy Attitudes and Voting Intentions Data on voting intentions and attitudes toward the wars in Iraq and Afghanistan were gathered from by Polimetrix/YouGov in two surveys fielded from September 27 through October 11, 2008. One survey included the experimental treatments regarding Iraq and the other included the experimental treatments on Afghanistan. Each survey collected 2,000 responses. Polimetrix gathers data from a volunteer opt-in panel of respondents via the internet and produces samples that are representative of the national public through a statistical procedure that selects opt-in respondents that are most similar to a random draw from the 2004 American Communities Study (ACS) conducted by the Census Bureau (Rivers 2007). Malhotra and Krosnick (2007) argue that levels of partisanship, political interest, and even the structure of relationships within the data

differ between a matched opt-in panel and the random sample generated for the American National Election Study (ANES). Hill et. al. (2007), on the other hand, conclude that the sample matching techniques produces modest biases which are analogous to the kinds of biases involved in random digit dialing (RDD) telephone surveys. Since the current study uses an experimental design, random sampling is not central to the estimation of causal effects. Nonetheless, it is useful to evaluate the representativeness of the Polimetrix data for purposes of external validity. Thus as a check on the representativeness of the Polimetrix data, Figure 1 displays the expressed voting intentions of the matched panel respondents from both surveys with an average of all of the probability based surveys recorded by Pollster.com that had field dates from September 27 through October 11, 2008. In addition, the figure compares both of these expressions of voting intentions to the actual electoral result on November 4, 2008. Figure 1 about here The voting intentions of the matched sample are virtually identical to those collected by probability-based sampling methods. Specifically, the matched opt-in panel data from Polimetrix is within 0.3% of the average of the probability sampling polls in their estimate of support for Obama, McCain, and those who are undecided or support other candidates. Moreover, both the matched panel data and the probability samples are quite close to the actual voting result. Of course, the undecided/other category shrank in the final result because those who were undecided either selected a candidate or did not vote. But this process had little impact on the net result, since Obama and McCain each seemed to gain about equally in their share of support. Specifically, the final result shows each candidate gaining about 3% of the 7% undecided/other category from the

September / October polls, while about 1% continued to support other candidates. Thus while representativeness is not necessary for causal inference in an experimental study, the Polimetrix matched opt-in panel data appear to provide a solid foundation for generalizing about American voting behavior in 2008. The Coherence of Public Attitudes Toward Iraq and Afghanistan According to Aldrich et. al. (1989) three conditions must hold in order for a policy attitude to influence vote choice. First, the attitude must be cognitively available. Availability refers to the existence of an attitude as an interconnected set of ideas that can be recalled from memory. An attitude is available for recall if it is a set of ideas that an individual can express with coherence and structure. Second, in order to be influential an attitude must be cognitively accessible. Accessibility refers to the ease with which an attitude can be recalled from memory. It is, of course, possible to make attitudes more accessible through processes such as priming (Iyengar and Kinder, 1987; Krosnick and Kinder, 1990; Mendelberg 2001). But attitudes also have differing levels of chronic accessibility. The greater the level of chronic accessibility, the easier it will be for other factors such as experimental news treatments - to prime that attitude and create additional accessibility. Third, in order for a policy attitude to influence vote choice the candidates in the campaign must take different positions on the issue. A number of studies of public opinion regarding the Iraq War have found that Americans attitudes toward the war are highly coherent and well structured (Jacobson 2007, Berinsky 2007, Berinsky and Druckman 2006; Gelpi, Feaver and Reifler 2009; Gelpi 2010). Much less is known, however, about the structure of attitudes toward

Afghanistan. Given the prominence of Iraq in media coverage and public debate, one might expect that attitudes toward Afghanistan will be less coherent and well structured, since the public has likely spent less time thinking about the issue. Moreover, since the Democratic and Republican parties did not take sharply different positions on the Afghan war, individuals could not easily rely on partisan cues to provide them with opinions on an issue about which they know relatively little. In order to measure the availability of attitudes toward Iraq and Afghanistan, I constructed an index of support for each of these wars. The indices do not seek to estimate or describe the relationships among the various facets of attitudes toward these wars, nor do they seek to determine whether this structure is caused by partisan identification or some other factor. Instead, the indices simply seek to place individuals along a single dimension of support for each war by estimating correlations among their responses to different questions about each war. These linear indices represent generalized support for each war, and I estimate alpha coefficients to describe the reliability of each index. Alpha coefficients range from 0 to 1 with higher values indicating that responses scale together more closely and reliably. While there is no absolute cutoff determining what alpha coefficient reflects a useful index, alpha scores of greater than 0.7 are generally taken as an indication that responses cohere with one another and constitute a reliable dimension (Cronbach, 1951). Moreover, we can estimate 95% confidence intervals around each alpha score to determine whether the attitudes toward one war cohere significantly more closely than another. Drawing on a variety of surveys regarding Iraq and Afghanistan, I asked each respondent five questions about each war to gauge the coherence of their responses.

Specifically, with regard to both wars respondents were asked whether they believed that President Bush had done the right thing by initiating those conflicts, and whether they believe that each mission will succeed. With regard to Iraq, respondents were also asked whether they felt that President Bush s deployment of additional troops to Iraq had improved the situation there, whether they supported a policy of withdrawing US troops only when conditions in Iraq improved, and whether they supported a timetable for withdrawing US troops. With regard to Afghanistan, respondents were also asked whether they believed the situation there was getting better or getting worse, whether they supported keeping US forces in Afghanistan until the situation there is stable, and whether they support sending additional US troops to Afghanistan. The estimation of these Alpha coefficients is described in Table 2. Table 2 About Here These results clearly indicate that the American public had coherent attitudes toward both Iraq and Afghanistan. Specifically, the alpha scores for the Iraq and Afghanistan batteries in this study were 0.87 and 0.78 respectively, indicating that attitudes toward both wars were well structured. At the same time, lower bound of the one-sided 95% confidence interval for the Iraq scale was 0.86, while the upper bound of the one-sided 95% confidence interval for that Afghanistan scale was 0.79. Thus as expected, attitudes toward Iraq were more tightly structured than those toward Afghanistan, but attitudes toward both of these wars clearly meet a standard of availability that should be necessary for them to influence vote choice. The Salience of Iraq and Afghanistan in October 2008

Political psychology models of voting emphasize issue salience because it is necessary for individuals to call up information about an issue into active memory if they are to use that information to make judgments or decisions. One important measure of issue salience is the extent to which individuals retain information about that issue and are able to recall it when asked. Thus as my first measure of the salience of the wars in Iraq and Afghanistan, I used an open-ended question to ask respondents about the number of American soldiers that they thought had been killed in each of these wars. Asking respondents to estimate KIA should measure attitude salience separate from substance because it asks respondents for factual knowledge as opposed to opinions. The ability to estimate US KIA accurately demonstrates that the respondent keeps information about that war easily accessible without conflating that accessibility with any substantive judgments about the issue. Respondents estimates of the number of American soldiers killed in Iraq and Afghanistan are displayed in Figure 2. The actual number of US soldiers killed in Iraq by the end of September 2008 was 4,175, while the actual number of US soldiers killed in Afghanistan by that date was 610. Figure 2 creates bands around these numbers and codes any response within 10% of the correct number as correct. The figure then creates bands that reflect respondents who were 10-20% too high or too low in their estimate, and those who were more than 20% too high or low. Figure 2 About Here Consistent with previous studies, the results in Figure 2 suggest that - in the aggregate - the American public has a pretty good idea of how many soldiers have died in Iraq (Bennett and Flickinger, 2008). Both the median and the modal response to this

question were 4,000, which is a very good estimate especially in light of the reduced media attention paid to the American body count during 2008. According to the Tyndall Report, for example, Iraq was the most reported story on the three major TV network evening news broadcasts in 2007. 5 A total of 1,888 minutes of airtime were devoted to Iraq on the big three evening news programs, nearly double the next most reported story the primary elections at 1,072 minutes. In 2008, on the other hand, election coverage filled 3,677 minutes of airtime and the economic downturn absorbed another 2,767 minutes. Iraq ranked a distant third in coverage at 434 minutes on the major network evening news shows. Nonetheless, approximately 40% of the public could still estimate the number of US battle deaths within 10% of the correct number in the fall of 2008. To be sure, a significant minority did appear to underestimate the human costs of Iraq. About 25% of the public underestimated US KIA by more than 20%. Within this group, the median estimate was 2,000 but the modal estimate (32%) was a more respectable 3,000. This group was balanced by a slightly smaller segment of the public who overestimated the human toll in Iraq. About 10% of the respondents over-estimated US battle deaths by 10-20% and another 15% overestimated the number by more than 20%. The modal estimate in this last group was 10,000. Any assessment of the public s factual knowledge about international issues is likely to of the half-full vs. half-empty variety, but despite the mistakes by some the responses in Figure 2 suggest that in the aggregate the public is quite well-informed about the human costs of the Iraq War. Certainly to the extent that the median voter matters 5 The Tyndall Report is available at http://www.tyndallreport.com.

politically we can say that he or she was very well informed about the human toll in Iraq in the fall of 2008. Afghanistan, on the other hand, does not seem to have attained nearly the same kind of salience in the public mind. This lack of accessibility is not very surprising given the relative paucity of media coverage. In 2007, for example, the Tyndall Report finds that Afghanistan was not among the top 20 most reported stories of the year on the major TV network evening news broadcasts. Afghanistan actually became more salient in 2008 with 126 minutes of network coverage, but even this level of attention left Afghanistan with less than 30% of the airtime of Iraq, even with the sharp decline in Iraq coverage that year. Thus in contrast to the patterns we saw regarding Iraq, the public seems to misunderstand the costs of Afghanistan in a significant way. More specifically, almost 70% of the respondents over-estimated battle deaths in Afghanistan by more than 20%. Both the median and the modal responses to this question were 1,000 KIA an overestimate of almost 40%. Less than 5% of the respondents gave estimates of US KIA in Afghanistan that were within 10% of the correct number. This inability to recall accurately the number of soldiers killed in Afghanistan is not a function of the relatively small number of soldiers killed. For example Steven Kull conducted a similar study of public knowledge of casualties in Iraq in 2004. At the time of his study approximately 740 American soldiers had been killed in Iraq. And yet, Kull (2004) found that approximately 58% of respondents placed the toll at between 500 and 1,000 (i.e. within about 30% of the correct number). By this same standard approximately two-thirds of the public could give a correct estimate of the death toll in

Iraq in 2004, but only about 18% of the public gave a correct estimate in Afghanistan in 2008. As another measure of the salience of Iraq and Afghanistan to voters in 2008, I asked respondents to rank the most important issues facing the federal government after the election. Respondents were provided with a list of seven possible issues and were asked to place them in rank order. The seven issues that I asked respondents to rank were: the war in Iraq, the war in Afghanistan, the economy, health care, immigration, the environment, and the proliferation of weapons of mass destruction. These issues appeared on a web page in a randomized order and then respondents were asked to drag and drop each issue so as to rank them in their preferred order. Figure 3 displays the distribution of rankings given to the wars in Iraq and Afghanistan as well as to the economy. Figure 3 About Here Perhaps the most prominent result that emerges from Figure 3 is the overwhelming priority that the public placed on the economy as the most important issue facing the country. Nearly 50% of respondents ranked the economy as the most important issue, and another 20% placed it second. This result is hardly surprising in light of the stock market collapse and credit market freeze of September 2008. In fact, the war in Iraq did not even receive the second highest number of placements as the most important issue. That distinction went to health care, with 15% of respondents placing that issue first, as compared to the 10% who placed the Iraq War at the top of the list. Nonetheless, Figure 3 does demonstrate that the public did continue to rate Iraq as an important issue. Nearly 50% of respondents placed the Iraq War among

the top three issues facing the country and nearly 70% placed it among the top four issues. Thus while Iraq was clearly eclipsed as the central issue of the campaign, it seems possible that a large enough segment of the electorate gave the issue sufficient weight in their calculations to influence voting decisions. Finally, Figure 3 indicates that Afghanistan continued to lag significantly behind Iraq in terms of the importance the public assigns to it. Less than a quarter of the public placed Afghanistan among the top three issues facing the country, and a majority of the respondents ranked it as fifth or lower on the list of seven issues. This would seem to be consistent with the results concerning the public s low level of information about casualties in Afghanistan. Whether it was due to a lack of media coverage or the pressing nature of other issues, Afghanistan was not chronically accessible to voters in 2008. Candidate Positions on Iraq and Afghanistan in 2008 Finally, if individuals are to rely on attitudes toward Iraq and Afghanistan to make voting choices in 2008, the candidates in the general election must stake out different policy positions on these issues. With regard to Iraq, the candidates appeared to take sharply different positions on a number of aspects of the conflict. For his part, Barack Obama sought to draw a distinction between himself and McCain with regard to their support for the initial decision to invade Iraq. Obama used this policy difference to emphasize his good foreign policy judgment as a counter to McCain s long record of national security experience. In essence, this stance by Obama sought to focus voters on the question of whether attacking Iraq was the right thing to do. McCain, on the other hand, sought to distinguish his own foreign policy judgment as superior to Obama s by

emphasizing his stance in support of the surge in early 2007, a policy that Obama prominently opposed. McCain s position was widely unpopular in the spring and summer of 2007, and some observers suggested that he had hurt his presidential bid by standing with the President on this issue (Ignatius, 2006). By the spring of 2008, however, mounting evidence suggested that the level of violence in Iraq was moving dramatically downward from the height of the civil war in late 2006 and early 2007. Thus when challenged on his initial decision to support the war, McCain would often seek to shift the focus to the surge and his own good judgment and integrity in supporting that strategy even when it was not popular (Santora, 2007). Finally, and perhaps most importantly, the two candidates disagreed on the question of whether the United States should set an explicit timetable for withdrawing its forces from Iraq or whether withdrawals should only occur in accordance with conditions on the ground with the recommendations of local commanders. As noted above, this issue timetable or no timetable became the central organizing issue in the debate over Iraq in 2008 (Balz, 2008). The candidates stances on this question received extensive media coverage during the summer of 2008 prior to the economic collapse and even as President Bush moved toward a de facto acceptance of some kind of timetable, McCain remained steadfastly opposed. With regard to Afghanistan, on the other hand, the candidates policy differences were more modest and muted. Both Obama and McCain, for example agreed on the most critical point that the mission in Afghanistan was the right thing to do. Obama took a strong stand that the US needed to escalate the war in Afghanistan by shifting troops out of Iraq. As noted above, McCain opposed Obama regarding the speed of withdrawal

from Iraq, but he did not take a clear stance on escalation in Afghanistan. McCain did not speak out against deploying additional forces to the Afghan War and seemed open to this idea, but at the same time he did not press the issue because it was not clear where these troops would come from unless the US withdrew from Iraq. Instead, with regard to Afghanistan, McCain focused his criticism of Obama on narrow tactical issues of whether it was wise to say publicly that the US would be willing to bomb remote areas of Pakistani territory in order to hit Al-Qaeda or Taliban forces if the Pakistani government was unwilling or unable to do so (Zeleny, 2008b). Debates such as this seem likely to give voters little traction in choosing candidates since the disagreement is not really over policy both men acknowledge that the US might need to strike in Pakistani territory but rather how the US should speak about its policy. Linking Foreign Policy Attitudes and Vote Choice in 2008 Given the availability of public attitudes toward Iraq and Afghanistan, the relative salience of these wars in the public mind, and the positions staked out by the candidates, it would appear that Iraq clearly meets the criteria outlined by Aldrich et. al. (1989) for a foreign policy issue that can influence vote choice. The public had coherent and wellstructured attitudes toward the Iraq War in the fall of 2008. The Iraq war remained chronically cognitively accessible to many members of the public despite the importance of the economy. And Obama and McCain took distinct stances on the war both retrospectively in terms of the decision to invade and to launch the surge and prospectively in terms of the need for a timetable for withdrawal. Thus we should expect

that exposure to news stories that prime subjects on the Iraq War will influence vote choice. According to these same criteria, however, attitudes toward Afghanistan should not influence vote choice. While voters do have coherent attitudes toward this conflict, they are not as tightly structured as attitudes toward Iraq. More importantly, Afghanistan does not appear to be cognitively accessible to voters, since very few respondents could recall the number of soldiers killed in that conflict, and few respondents placed Afghanistan among the top 3 issues facing the country. Finally, the candidates did not clearly distinguish their stances on Afghanistan from one another. Obama gave Afghanistan greater emphasis in his campaign and took a clear stance in favor of escalating this war. McCain, however, sought to blur these distinctions instead of staking out opposing views. Thus exposure to news stories that prime respondents on Afghanistan should not influence vote choice. A Cross-Sectional Analysis of Iraq, Afghanistan and Vote Choice in 2008 Table 2 displays the cross-sectional relationship between support for the wars in Iraq and Afghanistan on the propensity to express an intention to vote for Barack Obama. In addition to the attitudes toward these two wars, the analysis also controls for age, gender, race, ideology, and partisan identification. Party ID is obviously the most important determinant of vote choice, and as noted above attitudes toward Iraq are strongly polarized along party lines, making this an essential control variable. 6 6 Support for the Iraq War and partisanship are strongly correlated (r=0.68, p<.01). Support for the war in Afghanistan is also strongly correlated with support for the Iraq War (r=0.66, p<.01) and somewhat more weakly with party identification (r=0.42,

The relationship between support for Iraq and support for Obama is strongly negative and statistically significant. And as expected, the impact of attitudes toward Afghanistan is much weaker and does not quite reach statistical significance. The control variables in the model perform largely as one would expect. Younger voters and African American voters were more likely to express and intention to vote for Obama, while conservatives and Republicans were less likely to express such an intention. Interestingly, women were less likely to express an intention to vote for Obama. It is important to remember, however, that this effect arises once one controls for party identification and ideology. As a bivariate relationship, women were more likely to support Obama, but this relationship becomes insignificant once one accounts for the fact that women are more likely to be Democrats. Women become less likely to support Obama once we also account for the fact that women are more likely to be liberal. The apparent impact of attitudes toward Iraq on vote choice in this cross-sectional model is impressive indeed. Specifically, shifting a respondent s support for the Iraq War from the 10 th percentile to the 90 th percentile reduces the probability of voting for Obama by approximately 75%! This effect would appear to be even stronger than the independent impact of party identification, since the shift from strong Democrat to strong Republican only reduces the probability of voting for Obama by 67%. An Experimental Analysis of Iraq, Afghanistan and Vote Choice in 2008 Before concluding that attitudes toward Iraq were more important in shaping vote choice than partisanship, however, we should recall the possibility of endogeneity bias in p<.01).

such cross-sectional analyses. This threat seems especially severe in an election where candidates have longstanding and well-publicized policy positions on issues like Iraq. Obama supporters may have learned from the campaign that they should oppose the Iraq War rather than the other way around (Lenz 2009). As noted above, I addressed this problem by constructing a set of experimental treatments that raised the salience of the wars in Iraq and Afghanistan. By analyzing the effect of exposure to these treatments we can examine the exogenous impact of attitudes towards these wars as the treatments prompt subjects to call them to mind. Table 3 displays impact of these experimental treatments on voting intentions. Exposure to news events about the Iraq War continues to have a significant impact on vote choice. As expected, raising the salience of the Iraq War in voters minds influences their voting preferences because their attitudes toward the war are cognitively available and accessible, and the two candidates take different positions on this issue. Also as expected, exposure to news events about the war in Afghanistan did not influence vote choice. In this instance, while voters have coherent attitudes toward Afghanistan, the issue is not sufficiently cognitively accessible and the differences between the candidates on this issue were complex and somewhat muted. Table 3 About Here These results are broadly consistent with the cross-sectional results displayed in Table 2 in the sense that Iraq influences vote choice but Afghanistan does not. But these results demonstrate that attention to foreign policy issues can influence vote choice even after accounting for the potential endogeneity of policy attitudes. Moreover, the use of experimental treatments allows us to explore the mechanisms by which attention to

foreign policy issues can influence voting behavior: candidate competence, attitude activation, and issue ownership. Table 3 indicates that exposure to news events about Iraq reduced support for Obama regardless of the positive or negative content of the news stories. The negative impact of exposure to bad news from Iraq is statistically significant at the.05 level, while the negative impact of exposure to good news from Iraq is statistically significant at the.07 level. A t-test also revealed that the effects of the positive and negative news treatments did not differ significantly from one another. These results are consistent with the issue ownership model of vote choice and contradict the expectations of the candidate competence model. The candidate competence model suggested that good news about Iraq would be good news for McCain because of his long record of strong support for the war and most especially because of his support for the surge. According to this view, good news about the success of the surge should have persuaded voters that McCain was right on the Iraq issue in the sense that he had a successful plan to ameliorate the problem. Obama s opposition to the surge, on the other hand, should have led voters to conclude that he was wrong on Iraq. But this model also suggested that bad news about Iraq would undermine support for McCain because it would demonstrate that his policies for addressing this problem were not successful and that Obama had been correct in his opposition to the surge. Instead, we find that negative news events - which vindicate Obama s competence on Iraq - reduce support for Obama slightly more than positive news events that vindicate McCain s competence.

The issue ownership model, on the other hand, explains the effect of both Iraq treatments. Exposure to news stories about events in Iraq led subjects to think about the Iraq War, which is a strongly Republican issue given the American public s longstanding preference for Republican handling of foreign policy and security issues. Thus when subjects were asked to think about foreign policy issues they shifted their support away from the Democrat and toward the Republican candidate. The issue ownership mechanism is also bolstered by the fact that numerous polls during the summer and fall of 2008 indicated that foreign policy and security issues were perhaps the only area in which the public generally preferred McCain s ability to handle these problems (Susskind 2008). The control variables in these models continued to operate much as they did in the cross-sectional analysis. Younger voters and African Americans tended to support Obama, while Republicans and conservatives tended to support McCain. Women continued to be less likely to support Obama after taking these other factors into account, but this effect was no longer statistically significant. Interestingly, while both positive and negative news stories about Iraq reduced support for Obama, none of the partisan commentaries that were added to the news stories had any impact on voting intentions. This result also seems consistent with the issue ownership model. The exposure of subjects to news about Iraq raised the salience of a Republican issue regardless of the substance of the story. Exposure to partisan commentary on Iraq does not substantially change the salience of this issue above and beyond the exposure to the news events. The results in Table 3 clearly support the issue ownership model at the expense of the candidate competence model. But in order to examine the impact of voter issue

positions we must interact the experimental treatments with subjects policy positions on Iraq and Afghanistan. As noted above, these issue positions may be endogenous to candidate preference, but we can still examine the impact of issue positions by exogenously raising their accessibility. The issue activation model expects that raising the salience of the Iraq War will reduce support for Obama among voters who oppose a timetable for withdrawal from Iraq, but will increase his support among those who approve of a timetable. The impact of Afghanistan is somewhat more difficult to test for the issue activation model, since the candidates took less clearly distinct positions on this issue. Nonetheless, Obama took a clearer and more salient position in support of increasing the number of US troops deployed in Afghanistan than did McCain. Thus if issue activation is to influence vote choice with regard to Afghanistan, we would expect increasing the salience of this issue to increase support for Obama among those who approve of increasing the number of US soldiers deployed in that war. Support for Obama should decrease in response to the treatments, on the other hand, among those who oppose sending more troops. Figure 4 displays the estimated coefficients for the Iraq and Afghanistan treatments on the propensity to vote for Obama among those who support and oppose his policy position on each war. The vertical bars represent the 95% confidence intervals around each estimated coefficient. Figure 4 About Here Consistent with the results in Tables 2 and 3, we continue to find that Afghanistan has no significant effect on vote choice. This is not surprising given that the candidates positions on this issue were not very distinct. Turning to the Iraq treatments, however,

we can see that the results are at least partly consistent with the attitude activation model. Exposure to either positive or negative news stories about Iraq reduces support for Obama among those who oppose a timetable for withdrawal. This results suggests that raising the salience of Iraq regardless of the positive or negative tone causes voters to match up their own position on Iraq with those of the candidates, and causes voters to reduce support for a candidate with whom they disagree. The 95% confidence interval surrounding the estimated effect of positive news events does not include zero, indicating that this result is statistically significant at the.05 level. The confidence interval surrounding the impact of negative events, on the other hand, just barely crosses zero. This effect is significant at the.09 level. Contrary to the attitude activation model, however, exposure to news events about Iraq does not increase support for Obama among those who support a timetable for withdrawal. Instead, Iraq treatments that draw attention to this issue have no effect on voters who agree with Obama. This combination of results suggests that the issue ownership and attitude activation models have a joint and interactive effect on voting intentions. Specifically, the results suggest that raising the salience of a Republican issue only hurts Democratic candidates among voters who disagree with the Democratic position on that issue. At the same time, the results indicate that raising the salience of a Republican issue cannot increase support for a Democratic candidate even among voters who agree with the Democratic position on the issue. It is worth noting that the impact of issue attitudes in Figure 4 is not simply a proxy for party identification. It is certainly true that most Democrats approved of a timetable for withdrawal from Iraq in 2008, while most Republicans opposed this

position. But the same pattern of treatment effects continues to hold for subjects from each political party. Figure 5 displays the estimated coefficients and 95% confidence intervals for the estimated impact of the Iraq news stories among those who approved and disapproved of a timetable for withdrawal within each party. Interestingly, the impact of exposure to Iraq news is most negative among Democrats who oppose a timetable for withdrawal. This result is strongly consistent with the issue activation model since partisanship would push these voters to ignore their disagreement with Obama on this issue and continue to support their party candidate regardless of the salience of Iraq. This result is also supportive of those who argue that cross-pressured partisans are critical targets for persuasion in presidential campaigns (Hillygus and Shields 2008). Democrats who opposed Obama on Iraq appeared to be very persuadable targets for a McCain campaign that focused on foreign policy issues. Figure 5 About Here But the impact of issue activation is not limited to cross-pressured Democrats. Republicans who opposed a timetable for withdrawal also reduced their propensity to support Obama after exposure to news stories about Iraq. The impact of negative news stories was significant at the.05 level among Republicans, while the impact of good news was only significant at the.11 level. Nonetheless, the pattern of results is consistent across partisanship. This result is important because exposure to the Iraq treatments reduced support for Obama among Republicans who disapproved of a timetable despite the fact that their support for Obama was already low. Thus attitude activation can operate both by persuading cross-pressured voters and by rallying those who already agree with the candidate.

Contrary to the attitude activation model, however, raising the salience of Iraq did not increase support for Obama among voters who supported his Iraq policy regardless of whether the voters in question were Democrats or Republicans. One might expect, for example, that Republican supporters of a timetable should be persuadable crosspressured partisans. But exposure to the Iraq treatments had no effect on these subjects. Nor did exposure to the Iraq story rally greater support from Democrats who agreed with Obama on this issue. These results seem to reflect the continued importance of issue ownership. By raising the salience of the Iraq War, the McCain campaign had the opportunity both to persuade cross-pressured Democrats and to rally committed Republicans. But for Obama, focusing on Iraq could only cause him to lose ground by fighting on Republican home turf. The coefficients displayed in Figures 4 and 5 help us to test the mechanisms by which foreign policy issues can influence voting intentions, but they cannot describe the substantive size of these effects. As noted above, cross-sectional studies of vote choice may substantially overestimate the impact of issue attitudes due to problems of endogeneity. So how much did exogenously raising or lowering the salience of the Iraq War actually influence voting intentions? Figure 6 About Here Figure 6 displays the impact of the Iraq news stories on vote choice among subjects who disapproved of a timetable for withdrawal from Iraq. The predicted probabilities in the figure were generated for a hypothetical respondent 45 year-old, ideologically moderate, and politically independent white male respondent. The grey columns represent the predicted probability of each vote choice for a subject who did not

receive any news events about the Iraq War. The white columns represent the predictions for subjects who received the positive treatment, and the black columns represent the predicted probabilities for those who were exposed to negative news events. The set of columns on the far left represent the predicted probabilities that a respondent will express the unqualified intention to vote for McCain, while the set of columns on the far right represent the predicted probability of expressing an unqualified intention to vote for Obama. As the grey column on the left-hand side of the figure indicates, our hypothetical middle aged white male who disapproves of a timetable for withdrawal from Iraq has about a 35% chance of stating that he is certain that he will vote for McCain in the absence of any exposure to news events. But if this same individual receives either the positive or the negative news story about Iraq, the probability that he will express unqualified support for McCain increases to nearly 50%. Combined with those who lean toward McCain, the overall probability of voting for McCain increases from 54% to about 67%. Conversely, exposure to Iraq news reduces the probability that such an individual will lean toward Obama or will express unqualified support for Obama from 30% to about 20%. Figure 7, on the other hand, displays the impact of exposure to news about Iraq on the same hypothetical respondent who approves of a timetable for withdrawal. Here we can see that Obama fails to gain traction among those who agree with him on this Republican issue. Specifically, the probability of leaning toward or expressing unqualified support for Obama is 77% in the absence of exposure to news about Iraq.

But this probability only increases to 81% after exposure to positive news stories about Iraq and actually decreases to 74% after exposure to negative news events. Figure 7 About Here These results lead to three important implications. First, raising (and lowering) the salience of the Iraq War had a substantial impact on vote choice in 2008. A change in the probability of voting for McCain among timetable disapprovers from 54% to 67% represents a substantial effect. Extrapolating from such an individual level result to making predictions about electoral results is highly complex and beyond the scope of this study. For example, the change in the aggregate electoral result from this individual level effect would depend upon the distribution of timetable opponents across the 50 states, the closeness of those state races, the propensity of these individuals to turn out to vote, and so on. Nonetheless, we can say that the change in individual voting intention is substantial and makes it plausible that the changes in the salience of foreign policy issues like Iraq might well tip the balance of an election. At a minimum, we can say that the declining salience of the Iraq War in the final six weeks of the 2008 presidential campaign almost certainly cost John McCain a large number of votes. Second, estimates of the impact of issue attitudes on vote choice may be significantly biased by problems of endogeneity. While the impact of attitudes toward Iraq remained substantial in the experimental study, they were about one-fifth as large as the estimated impact in the cross-sectional model. Given all that we know about the importance of party identification and the relatively small fraction of the population that will ever express a willingness to vote for a candidate from another party, it seems implausible to suggest as the cross-sectional model did that support for the Iraq War

had a larger independent impact on vote choice than partisanship. In the experimental analysis, however, the exogenous impact of attention to Iraq had a more modest 15% impact on voting intentions, while the shift from being a strong Democrat to a strong Republican continued to reduce the propensity to vote for (or lean toward) Obama by 70%. Finally, these results suggest that a combination of the issue ownership and attitude activation models help to explain the mechanism by which foreign policy issues such as the Iraq War can influence voting behavior. When an issue like Iraq becomes salient in the context of a presidential campaign, voters will shift their support toward a candidate if they agree with the candidate s substantive position on the issue, and the candidate s party has ownership over that issue. It is this convergence of perceived partisan competence and shared policy preference that shapes voting intentions. Conclusions The stock market collapse and credit crisis that began in September of 2008 shifted popular attention decisively toward the economy as the most important issue in the public mind as we approached the general election. Yet the analyses presented here demonstrates that even in an environment dominated by concern over the economy, foreign policy issues can have a substantial and even potentially decisive impact on voting behavior. Drawing on previous studies of the political psychology of voting behavior, I have argued that foreign policy issues should influence voting preferences when three conditions are met: 1) voters attitudes about a foreign policy issue are

available, 2) those attitudes are cognitively accessible, and 3) the candidates policy stances differ on the issue. Applying this framework to public attitudes toward Iraq and Afghanistan I found that while the public has cognitively available views about both of these wars, only Iraq was chronically accessible in the fall of 2008. Barack Obama and John McCain also differed sharply over their policy position on Iraq, but their policy differences on Afghanistan were more muted. Thus I expected that attitudes toward Iraq would influence vote choice in the fall of 2008 but attitudes toward Afghanistan would not. Data on voting intentions from September and October 2008 strongly support these expectations. Moreover, the results support three important conclusions about mechanisms by which Iraq influenced vote choice. First, contrary to the expectations of many commentators on the 2008 election, the public did not appear to be judging Obama and McCain on their competence in handling the Iraq War. Exposure to news events about the Iraq War had the same effect on vote choice regardless of the content of those events. Second, exposure to news about Iraq did cause subjects to match their own policy preferences regarding Iraq against those of the candidates. Attitude activation provides a good explanation for the fact that both Republican and Democratic opponents of a timetable for withdrawal reduced their support for Obama when exposed to news about Iraq. This result also indicates that Democratic opponents of a timetable for withdrawal were persuadable cross-pressured partisans (Hillygus and Shields 2008). Finally, Republican ownership of foreign policy issues played an important role in shaping the impact of Iraq on the election. While Obama was punished among those who

disapproved of his policy stance on a timetable, raising the salience of Iraq did nothing to gain Obama support among cross-pressured Republican partisans who supported a timetable. Nor was Obama able to rally more enthusiasm from Democratic supporters of a timetable by raising the salience of Iraq. It is always difficult to say how artificially created treatments like the ones in this study connect to the real world of political campaigns. But some anecdotal evidence from the summer and fall of 2008 suggests that these patterns may generalize outside of the experimental context. Immediately after securing the Democratic nomination in June 2008, Obama declared his intention to travel to Iraq and Afghanistan prior to the election. Obama s trip in late July and the extensive preparation for it received extensive media coverage, and Iraqi politicians even appeared to endorse Obama s support of a timetable for withdrawal. Yet according to Gallup s data on trial heat questions, Obama s six point lead in the polls over McCain entirely evaporated between mid-june and the end of July. On the day before the stock market crash, September 15, 2009, Gallup showed Obama trailing McCain by one point. The crash immediately shifted the attention of the public and the media almost entirely toward the economy and Iraq rarely became an issue during the rest of the election. The day after the crash Gallup showed Obama with a 4-point lead in the polls that he never relinquished on his way to a landslide victory in November. While one can never say for sure what would have happened in he absence of the stock market crash, it appears that attracting voter attention to the Iraq and Afghanistan substantially harmed Obama s candidacy, while shifting attention to the economy revived it.

Bibliography Aldrich JH. 1977. Electoral Choice in 1972: A Test of Some Theorems of the Spatial Model of Electoral Competition. Journal of Mathematical Sociology. 5: 215-237 Aldrich JH, Sullivan JL, Bordiga E. 1989. Foreign Affairs and Issue Voting: Do Presidential Candidates Waltz Before A Blind Audience? American Political Science Review 81, 1:123-141. Aldrich, JH, Gelpi C, Feaver P, Reifler J, Sharp KT. 2006. Foreign Policy and the Electoral Connection. Annual Review of Political Science 9: 477-502 Almond G. 1950. The American People and Foreign Policy. New York: Harcourt Brace Anand and Krosnick, 2003. Balz D. 2008. Obama Makes War Gains; Maliki s Embrace of Withdrawal Timeline Confounds McCain. The Washington Post. July 22, 2008, p.a6. Bennett SE and Flickinger RS. 2009. Americans' Knowledge of U.S. Military Deaths in Iraq, April 2004 to April 2008. Armed Forces & Society 35, 3:587-604. Berinsky, Adam. 2007. Assuming the Costs of War: Events, Elites, and American Public Support for Military Conflict. The Journal of Politics 69, 4: 975-997. Berinsky, Adam, and James Druckman. 2007. The Polls Review: Public Opinion Research and Support for the Iraq War. Public Opinion Quarterly 71, 1: 126-141. Campbell A, Converse PE, Miller WE, Stokes DE. 1960. The American Voter. New York: John Wiley. Caspary, W. 1970. The Mood Theory': A Study of Public Opinion and Foreign Policy. American Political Science Review 64:536-547. Converse PE. 1964. The Nature and Origin of Belief Systems in Mass Publics. in David Apterl ed. Pp. 206-261 in Ideology and Discontent. New York: The Free Press. Downs, A. (1957). An Economic Theory of Democracy New York: Harper Collins. Enelow, J and MJ Hinich. (1984) The Spatial Theory of Voting, Cambridge: Cambridge University Press. Gelpi C, Feaver PD and Reifler J. 2005/2006. Success Matters: Casualty Sensitivity and the War in Iraq. International Security 30, 3: 7-46

Gelpi C, Reifler J, Feaver P. 2007. Iraq the Vote: Retrospective and Prospective Foreign Policy Judgments on Candidate Choice and Casualty Tolerance." Political Behavior 29:151-174. Gelpi C, Feaver PD and Reifler J. 2009. Paying the Human Costs of War: American Public Opinion and Casualties in Military Conflicts. Princeton: Princeton University Press. Hill S, Lo J, Vavreck L, and Zaller J. 2008. The Opt-in Internet Panel: Survey Mode, Sampling Methodology and the Implications for Political Research. Unpublished Manuscript. Hillygus, DS and T Shields. 2008. The Persuadable Voter. Princeton: Princeton University Press. Ignatius D. 2006. Hagel s Moment? The Washington Post. November 29, 2006, p. A23. Jacobson, G. 2007. A Divider, Not A Uniter: George W. Bush and the American People. New York: Longman Publishers. Kagay, M and Caldeira G. 1975. "I Like the Looks of His Face." Presented at the annual meeting of the American Political Science Association, San Francisco. Malhotra N, and Krosnick JA. 2007. The Effect of Survey Mode and Sampling on Inferences about Political Attitudes and Behavior: Comparing the 2000 and 2004 ANES to Internet Surveys with Nonprobability Samples. Political Analysis, February 2007. Karol D, and Miguel E. 2008. "The Electoral Cost of War: Iraq Casualties and the 2004 U.S. Presidential Election." The Journal of Politics 69:633-648 Kornblut A. 2008. Measured Response to Financial Crisis Sealed the Election. The Washington Post. November 5, 2008, p.a1. Kull S, et al. 2002. Americans on the Conflict with Iraq. The PIPA/Knowledge Networks Poll, Oct. 2. Light, PC, and Lake C. 1985. 'The Election: Candidates, Strategies, and Decision." In The Elections of 1984, ed. Michael Nelson. Washington: Congressional Quarterly. Miller, AH and MP Wattenberg. 1985. Throwing the Rascals Out: Policy and Performance Evaluations of Presidential Candidates, 1952-1980. The American Political Science Review 79(2):359-372 Nincic, M. 1992. A Sensible Public: New Perspectives on Popular Opinion and Foreign Policy. The Journal of Conflict Resolution, vol. 36, no. 4: 772-789

Norpoth, H and AH Sidman. 2007. Mission Accomplished: The Wartime Election of 2004. Political Behavior 29(2):175-195. Page, BI, and Brody RA. 1972. "Policy Voting and the Electoral Process: The Vietnam War Issue." American Political Science Review 66:979-95. Page, BI, and Shapiro RY. 1982. Changes in Americans' Policy Preferences, 1935-1979. Public Opinion Quarterly, vol. 46, no. 1: 24-42. Page, BI, and Shapiro RY. 1988. Foreign Policy and the Rational Public. Journal of Conflict Resolution, vol. 32, no. 2: 211-247. Petrocik, JR. 1996. Issue ownership in presidential elections..american Journal of Science Review. 58 (4): 825-850. Petrocik, JR.,WL Benoit, and GJ Hansen. 2003. Issue Ownership and Presidential Campaigning, 1952 2000. Political Science Quarterly 118:599 626. Rosenau, JN, ed. 1961. Public Opinion and Foreign Policy. New York: Random House. Sanger D. 2008. Events in Iraq Likely to be Key Theme for McCain. The New York Times. September 2, 2008, p.1. Santora M. 2007. Seeing Progress in Iraq, McCain Hopes for Credit. The New York Times. November 10, 2007. Shear MD, and Murray S. 2008. Clash on Iraq Could Be McCain-Obama Preview. The Washington Post. February 28, 2008, p.a1. Sussman D. 2008. Poll Finds McCain Edge on Security. The New York Times. September 26, 2008. Stokes, DE. 1963. Spatial Models of Party Competition, American Political Science Review 57: 368-377. Stokes, DE. 1966. "Some Dynamic Elements of Contests for the Presidency." American Political Science Review 60:19-28. Zeleny J. 2007. As Candidate, Obama Carves Anti-War Stance. The New York Times. February 26, 2007. Zeleny J. 2008. Obama, Drawing Criticism on Two Fronts, Fires Back. The New York Times. February 21, 2008.

Table 1: The Coherence of Attitudes Toward Iraq and Afghanistan Iraq Afghanistan Variable Avg. Interitem Alpha Variable Avg. Interitem Alpha Correlation Correlation Right 0.55 0.83 Right 0.37 0.70 Thing Thing Will 0.54 0.82 Will 0.39 0.72 Succeed Succeed Surge 0.61 0.86 Situtation 0.63 0.87 Better Better Withdraw by 0.58 0.84 More Troops 0.36 0.70 Conditions Withdraw by 0.55 0.83 Maintain 0.37 0.70 Timetable Troops Test Scale 0.57 0.87 Test Scale 0.42 0.79 Lower Bound of One-sided 95% Confidence interval = 0.86 Upper Bound of One-Sided 95% Confidence Interval = 0.79

Table 2: Support for Wars in Iraq and Afghanistan and Vote Choice, 2008 Iraq Support -2.029 (20.63) Afghanistan Support 0.171 (1.93) African American 1.666 (9.02) Hispanic -0.039 (0.29) Age -0.007 (2.37) Gender -0.332 (3.58) Ideology -0.469 (8.40) Point Party Identification -0.554 (19.77) Observations 3291 Absolute value of t statistics in parentheses

Table 3: Impact of Iraq and Afghanistan News Exposure on Voting Intentions 2008 Iraq Afghanistan Positive Events -0.268-0.090 (1.82) (0.61) Negative Events -0.387-0.134 (2.59) (0.91) Negative Senator (D) -0.068-0.416 (0.35) (2.22) Positive Senator (D) -0.099-0.217 (0.52) (1.11) Negative Senator (R) -0.024 0.038 (0.13) (0.20) Positive Senator (R) 0.041-0.049 (0.22) (0.25) African American 1.881 1.387 (7.07) (5.65) Hispanic 0.085-0.135 (0.48) (0.70) Age -0.009-0.009 (2.18) (2.08) Gender -0.142-0.140 (1.16) (1.14) Ideology -0.732-0.903 (9.99) (12.22) Party ID -0.779-0.741 (20.44) (20.20) Observations 1639 1652 Absolute value of t statistics in parentheses

Figure 1 Obama vs. McCain: Probability Sample Polling, Polimetrix Matched Opt-in Sample, and Actual Voting Returns

Figure 2 The Cognitive Salience of Iraq & Afghanistan: Public Estimates of US KIA

Figure 3 The Salience of Iraq, Afghanistan and the Economy: A Ranking of Important Issues