Both scholars and pundits alike have attributed the Democrats' success in gaining control of the House and Senate in 2006 to their “nationalization” of the election and their success in connecting Republican candidates to both the war in Iraq and the highly unpopular George W. Bush presidency. The “election has been nationalized around two standards that could not be more unfavorable to the GOP: an unpopular war and an unpopular president” (Novak 2006). Many have gone so far as to say that the election results provide evidence that former House Speaker Tip O'Neil's famous aphorism, “all politics are local,” has been “swept aside” (Lambro 2006) and does not apply during wartime. “If one trend emerged this year when 435 House seats and 33 Senate seats were at stake, it was this: All politics are not local” (Israel 2006).

We fundamentally disagree. There is no question that national issues were especially important in the 2006 election. We suggest, however, that national issues are frequently experienced and understood through their local contexts. That is, the local effects of national and even international issues shape what individuals know about—and more importantly, frame how people think about—national concerns. Critically, these geographic effects vary considerably, creating a local-political lens for viewing national political phenomena.

In the 2006 elections, Senate Democrats retained all their seats and gained six new, previously Republican, seats. The key question then is why were the Democrats successful in these six states? Our view is that the Iraq War's local human costs made a, if not the, critical difference in deciding these key races, suggesting that the politics of an international war may indeed have an important, and heretofore overlooked, effect on state-level races.

In this war, like previous wars, military casualties—the deaths of fellow citizens—drive public attitudes on the conflict and extant leadership. Casualties are information.¹

They provide both clear knowledge about current and cumulative losses to date, as well as signals about future costs. Widely reported nationally and locally, the death of service men and women are salient, visible, and accountable measures of cost. Losses provide clarity and personality to a distant, anonymous Middle Eastern war. Casualties represent the most common and most important measure of Iraq War costs—especially given the lack of other indicators of success and societal mobilization for war.

This is not to say, however, that other factors do not play a role. For example, the departure of Colin Powell as secretary of state after the 2004 election, and his widely reported disaffection with the conduct of U.S. policy, surely helped influence opinion. Similarly, ongoing revelations about pre-war intelligence failures and manipulations, the continuing violence, the failure to find the alleged weapons of mass destruction (WMDs), and other setbacks certainly served to undermine public confidence in the struggle. While these help explain national levels of support, they fail to capture variation across electorates. Moreover, none of those would be definitive if prevailing in the conflict were costless. It is the costliness of the conflict, particularly with respect to human life, that makes these additional factors corrosive to overall support.

We consider three types of casualty effects. First, as cumulative national casualties go up, support for leaders goes down (Mueller 1973; Gartner and Segura 1998). Individuals see casualties as evidence that a policy is costly and failing—views that lead to disapproval of extant leaders and policy (Gelpi, Feaver, and Reifler 2005). Studies have tied increasing national casualties in Iraq to President Bush's decreasing support and the low levels of support for the war (Karol and Miguel 2007; Voeten and Brewer 2006). While this clearly represents an important part of the Democrats' success in 2006, like the failure to find WMDs or the criticisms of former Secretary of State Powell, cumulative national casualties cannot help to explain the Senate puzzle, because they do not help us understand why the Democrats were successful in exploiting people's negative view of the war in the six states that turned over and not in others. Exit polls on Election Day found on average that, “57 percent of all voters disapprove of the war in Iraq and 58 percent disapprove of Bush's job performance” (CNN 2006a). Yet, in the Senate races, the Republican incumbent median vote share was 53% and their average vote share was 57% (Congressional Quarterly 2006). Republican incumbents clearly did very well in some states and poorly in others—a phenomenon not explained by the nationalization of the election. More generally, we agree with Boettcher and Cobb that this type of research needs to move away from “aggregate-level, cumulative and monotonic explanations” (2006, 850).

National casualties alone would suggest a nationwide, consistent, negative effect when, in fact, elections are a spatially disaggregated process. Recognizing the disaggregated nature of elections, we focus instead on state casualties—and here there is great leverage in explaining Senate, state-wide contests. Geographically distributed casualties, specifically state-level and local casualties, are more identifiable to individuals (Gartner and Segura 2000). While national casualties are reported in summary statistics, reports of local casualties usually include hometowns, community and personal information (school attended, place of employment, family members), pictures, and other detailed information that transform an abstract cost into a loss that is more identifiable and meaningful. There is significant variation in state-level casualties, even when controlling for population. Total per-capita state casualties in the Iraq War up to the time of the 2006 election vary by over 500% (Vermont to Utah) and recent casualties vary by over 1,800% (Montana to New Jersey, excluding states with no recent casualties). Looking at smaller units of analysis leads to even greater variations, for example San Francisco (population 740,000) suffered a total of three Iraq War KIA, while just over 60 miles away the California town of Tracy (population 57,000) experienced five.

Small numbers of war dead can have tremendous influence as their story is repeated again and again. Smaller states that likely experience fewer total casualties can suffer higher disproportionate costs, which also can be a story that generates resentment against extant leadership. For example, a 2005 headline read, “Vermont leads in per-capita war deaths” (Timiraos 2005). Two years later, when that distinction had not changed, we do not think it coincidence that the “state that has lost the most troops per capita became the first to pass a resolution calling on Congress and the president to immediately withdraw U.S. forces from Iraq” (Allen 2007).

Local casualties not only generate more salient information, they also lead to more local news coverage, as local media cover funerals, interview grieving family members and friends, provide details of the life and death of the casualty (such as pictures of their children), and give greater attention to the incident that led to his or her death (Althaus and Coe n.d.; Gartner 2004). By generating both more information and information of greater salience, local casualties play a major role in influencing individuals' assessments of the conflict and its attendant costs (Gartner and Segura 2000).

These negative assessments have significant electoral consequences. Research shows that local casualties affect congressional races (Carson et al. 2001) and state-wide votes on presidential contests (Karol and Miguel 2007; Voeten and Brewer 2006). More pointedly, studies have previously demonstrated the effects of state casualties on Senate elections. As state-wide total per-capita casualties increase, incumbent senators receive a smaller share of the vote—especially if the challenger opposes the conflict and the incumbent supports it (Gartner, Segura, and Barratt 2004).

We also examine a third, and we think relevant, perspective on how human costs influence domestic politics by looking at recent casualties. Casualties that occurred recently—those that are temporally proximate—provide highly salient signals about current and future costs. Recent casualties carry more cognitive “punch,” that is, they have a greater affect on people's perceptions than those deaths that occurred further in the past. “The use of unlogged marginal casualties is both intuitively attractive and supported by framing research in psychology and political science …” (Boettcher and Cobb 2006, 836). These casualties also send signals about the direction that future casualty trends are likely to go. As a result, recent casualties play a major role in influencing opinion on a conflict and its leaders (Gartner and Segura 1998; Voeten and Brewer 2006). Again, these attitudinal effects have real electoral consequences. Studies of Senate elections show that as recent, state-wide casualties increase, incumbent senatorial candidates receive a smaller share of the vote (Gartner, Segura, and Barratt 2004).

Casualties also influence elites, whose views may in turn shape public perceptions (Zaller 1992). Here again, both small numbers and aggregate measures can have an effect. When Pennsylvania Representative John Murtha, a formerly hawkish Vietnam War veteran, turned against the Iraq War, he “cited the shattered life of Mr. Ross, one of his first constituents to be seriously wounded, as a pivotal influence” (Sontag 2007). More generally, as recent state casualties increased, senators increasingly opposed the Vietnam War (Gartner, Segura, and Barratt 2004). By dampening elite support, casualties also have an indirect effect on constituents.

Obviously, many factors, both national and local, affect Senate elections. Some of these factors are idiosyncratic (Virginia Senator George Allen's videotaped “macaca” reference), while some are systematic (e.g., the economy). We want to take a first cut at the possible local influence of the war in Iraq on the six key Senate races and see if there is reason to think that the war had a disproportionate effect in those states. Since national casualties only tell us about the national mood, we need to look at state-level costs. Additionally, recent casualties carry additional salience beyond that captured by cumulative losses. As a result, we examine Iraq War state-level casualties, both recent and cumulative to the election, comparing the six key races where Republican incumbents lost to the other states holding elections. We want to see if the Iraq War exerted an above average effect on those states that experienced seat changes.

The six states in question are Virginia, Missouri, Montana, Rhode Island, Pennsylvania, and Ohio. We look at Recent Casualties (October 2006) and Cumulative Casualties (start of the war in 2003 through October 31, 2006). In each case, we count state-wide military killed in action in the Iraq War (not in Afghanistan) and divide by state population (in millions) to determine the proportionate cost paid by each state.²

Table 1 shows Recent Casualties and Incumbent Vote Share for states holding Senate elections in 2006.³

Did state-level effects of the conflict in Iraq have disproportionate impact on these six states and provide a more negative lens through which constituents viewed the war and their leaders? Yes, the states in which Republican incumbents lost experienced higher recent Iraq War losses. In terms of Recent Casualties, Montana followed by Rhode Island paid the highest recent proportional casualty costs, while Pennsylvania was fourth. Not only are three of the seats that turned over in the top four most recently deadly states, but Virginia and Missouri rank 9^th and 12^th respectively in terms of recent per-capita costs. The sixth state, Ohio, ranks extremely close to the median.

Fifteen of the 33 states that held elections experienced less deadly Octobers than Ohio, the least lethal state that had a Senate seat change hands. Moving beyond October, if we examine a period of 120 days, Ohio jumps above the median and is ranked 13^th in terms of casualties per capita (not shown). In other words: All of the seats that the Democrats captured came from states that paid disproportionate recent Iraq War costs.

These results are supported further by a simple cross-tabulation of recent per-capita state-wide Iraq War costs and victory or defeat of the incumbent party in the 33 states holding Senate elections, reported in Table 2. Despite the obvious limitation of only 33 observations, this relationship approaches significance (p ≈ .08). Additionally, a logistic analysis of recent costs and seat change is positive and significant (p = .029). The results are consistent; as October 2006 per-capita casualties increased, so did the likelihood of a Senate seat change.

Recent casualties pushed all seat turnover states, especially Rhode Island, up the cumulative casualty rankings. Table 3 reports Cumulative Casualties and Incumbent Vote Share. Looking at Cumulative Casualties, we see that five of the six states (Virginia, Montana, Rhode Island, Pennsylvania, and Ohio) experienced total per-capita costs greater than the median state (Hawaii). Only Missouri, which ranked 12^th in terms of Recent Casualties, is below the median for Cumulative Casualties.

Studies of war and elections argue that incumbents of the party in power are more likely to pay the war's electoral price (Gartner, Segura, and Barratt 2004). Looking only at the 16 Republican incumbents, Democrats took three-fourths of the seats in states in the top quartile of Recent Casualties, and all six Republicans losses came from the 11 states that paid the highest Recent Casualties (the same is true when Recent Casualties are measured using 120 days instead of 30). Four of the seven states that paid the highest Cumulative Casualties supported challengers. Even among states that previously voted Republican, senators from states that experienced higher costs had a decreased chance of retaining their seat.

Each of the six seats that changed hands came from states that paid disproportionate recent casualties and five of the six paid higher than most cumulative costs as well. The citizens in these states received salient, visible signals of wartime costs and progress that provided evidence for a conflict that was more costly than those signals received by the citizens of some of the other states. Note that we are not assuming that individuals calculate state KIA per-capita rates, nor do we think they compare their wartime costs with the median or even with other states. We are fully aware of the well-documented information deficits with which many, even most, Americans engage the political system (see Converse 1964 and Delli Carpini and Keeter 1996, among many others). However, casualties represent the most important information from a war about its costs and are readily experienced in that they are widely and repeatedly reported in the communities from which they come. Previous studies show substantial public and media sensitivity to casualty variation at the local level (Althaus and Coe n.d.; Gartner and Segura 2000; Gartner 2004). Those who experience greater casualties, we suggest, are more likely to oppose a conflict and the incumbent leadership to whom they can attribute blame.

Exit poll data from the 2006 elections would appear to support such an explanation. Table 4 presents the CNN Exit Poll marginals on the question of support or opposition to the Iraq War for all but two states where Republicans were defending seats (CNN 2006b).⁴

States are sorted (descending) based on share of respondents opposing the war. Of the nine states in which a majority of respondents disapproved of the war on Election Day, six GOP candidates lost. In the four states where a slim majority still supported the war, all four GOP incumbents won. Despite only 30 observations, an OLS regression finds Senate seat changes are significantly and positively correlated with state levels of disapproval of the war (controlling for Republican incumbents, results not shown). That is, those states that experienced the defeat of the incumbent party had higher levels of disapproval of the Iraq War, just like they had higher levels of casualties. We can also see these patterns by examining some of the races.

Consider the races in Rhode Island and Maine. Both had moderate Republican incumbents who spoke out strongly against the war and the president—views consistent with the attitudes of their constituents, who were more critical of the Iraq War and President Bush than the national average. Both senators had strong state-wide pre-election evaluations. Their state populations are almost identical in size, and both states receive substantial regional news. In addition, Table 4 shows opposition to the war was high in both. Yet Maine's Olympia Snowe was returned to the Senate with over 74% of the vote while Rhode Island's Lincoln Chafee (who had a 63% pre-election approval rating) was defeated. Why? The standard answer is, “Rhode Islanders like Chafee personally, but they voted nationally, not locally” (CNN 2006a). How might an approach that includes the local effects of the war in Iraq explain this differently?

Rhode Island was second in terms of the most costly October experienced by states holding elections and third out of all 50 states. Maine suffered no citizen losses, not just in October but also for more than six months prior to the election, and thus was tied for last in terms of Recent Casualties (however measured). The disproportionately high recent state casualties in Rhode Island brought home the costs of the war to Chafee's constituents, affecting their support of him and the conflict. People from Rhode Island had reason to perceive the war as more costly than did those from Maine. As an incumbent of the party directing this costly war, Chafee was clearly punished at the polls. While a comparison of these two cases is not, alone, sufficient to make our claim, it is clearly consistent with our view that local and recent casualties are part of the story that explains why voters' negative feelings on the war stuck in one race but had no effect in another.

Moreover, there is reason to believe that we have underestimated the impact of local costs, as the effects of casualties are certainly present even in cases where the incumbent did not lose. A simple bivariate regression of Recent Casualties and Incumbent Vote Share is negative, and again despite only 33 observations, statistically significant (p = .01). That is, the effect of recent state casualties on vote share was systematic and not confined to the few instances where it was sufficient to tip the outcome. For example, in Tennessee, the Republican Party narrowly managed to defend one of its seats despite the absence of an incumbent candidate. But the margin of victory, 51-49%, was certainly close, especially when we consider the recent electoral history of the state where Republicans have won every senatorial election since Al Gore became vice president, where the GOP has carried the state in the last two national elections, and where the Democratic nominee was African American, historically a major electoral disadvantage in the South and elsewhere. It is worth noting that Tennessee ranked 6^th in recent casualties.

Casualties “hurt” the Republicans more than they “helped” Democrats. Democrats were not considered “owners” of the Iraq War in the minds of the public. Consistent with previous wars like Vietnam (Gartner, Segura, and Barratt 2004), the effect of casualties and their political ramifications are visited only on the incumbent party—here the GOP. These effects, while logically everywhere, will be most telling in states with genuinely contested elections. For example, Richard Lugar was reelected in Indiana facing only token opposition (CNN did not bother to conduct an exit poll here or in Mississippi). Similarly, Kay Bailey Hutchison ran for reelection in Texas, a state where Democrats have not won a major statewide election since 1990. Thus, some high-casualty states saw their senators safely reelected. Referring to Table 1 (recent casualties), in each instance, the winner was either a Democrat (MI, MD, MA, CA, FL, MN, and WI), or never seriously faced an electoral threat (TX and IN). Among states in the upper half of the casualty distribution, only three (Tennessee, Nevada, and Arizona) saw the GOP hold the seat against meaningful opposition, and with vote shares at 51, 55, and 53, respectively. Looking at Table 3 (cumulative casualties), of the 17 states with casualty rates above the median, Democrats (and Independents caucusing with the Democrats) successfully defended in seven states (ND, NE, VT, DE, WI, NM, and HI) and took five seats from the GOP (MT, PA, OH, VA, and RI). Again, some were effectively uncontested (TX and MS), and both TX and WY are among the very few states where the war still enjoys majority support. Arizona and Nevada had relatively narrower margins.

We are sensitive to the claim that other factors might have played a significant role in determining which of these GOP seats switched to the Democrats. Our effort is intended as exploratory and does not claim to be a general model of Senate election outcomes, which are well-researched and established in the literature. Our goal is to demonstrate that post-election claims about nationalization of the war and the rejection of the local effects of politics likely overreach. Certainly other factors, some substantial, contributed to the outcomes observed. For these factors to supplant casualties entirely, however, they would have to explain away all the variation those measures currently capture. We think this is unlikely, in part because there is no evidence that any standard political factor is correlated with casualties and in part because any evidence of such a correlation—for example, the degree to which the challenger was well-funded—could, itself, be endogenous to the effects of war casualties. That is, if other political factors were uniquely hurting this set of incumbents more than others, there is a strong possibility that the war and its effects are at least partially responsible. In addition, the possibility of an intervening cause would be greater if the lost seats and defeated candidates were all of a type. For example, it would be problematic for our case if all defeated candidates were extremely conservative, or extremely moderate, or closely aligned with religious conservatives, or any other characteristic that would suggest itself as an alternative explanatory variable. In point of fact, these candidates are wildly different from one another. Ideologically, the list of losers includes the GOP's most liberal senator, Chafee, and one of its most conservative, Rick Santorum. Seats were lost by both incumbents as well as a new candidate trying to hold a seat after a retirement (Ohio). Margins of victory range from 18 points to two; the states included range from among the most populous (PA) to among the least (MT), and are dispersed broadly, in New England, the Middle Atlantic, the South, the Midwest, and the Mountain West. Challenger characteristics also varied tremendously.

Wars are international, but they have both national and local effects. We have not attempted to present a model of electoral outcomes here (for example, see Gartner, Segura, and Barratt 2004). Rather, we have tried to make the case that the early dismissal of the local effects of war in the U.S. elections of 2006 is premature. The death of a citizen from one's state is likely to generate both greater and more salient information about the cost of a conflict. In this way, international wars produce widely varying local experiences. It is this local costliness, we suggest, that make other factors—such as the failure to find WMDs—erode the view of the conflict as “necessary,” a process that is particularly damaging to support for the war and its political leaders. Thus, one of the reasons that the Democratic strategy of linking the unpopularity of the Iraq War and President Bush to incumbent Republican senators worked in Virginia, Missouri, Montana, Rhode Island, Pennsylvania, and Ohio is that the citizens of these states suffered disproportionate casualties in Iraq, which generated a more intense and widespread concern about the costs of the war than that experienced by citizens in many other states. The U.S. Senate elections of 2006 contribute further evidence that, even when national issues dominate headlines, advertisements, and campaigning, all politics remain local—especially wartime politics.