Iraq Casualties and GOP Vote Share: A Review of Some Literature

For reasons involving real research, I need to see whether and how much the Iraq war affected the Bush administration’s electoral record. I’m reviewing some of the literature here, partly as a public accountability mechanism, partly as a personal note, and partly to see if anyone else has anything to add.

The theoretical stakes for me here are not, quite, in the realm of voter behavior. Rather, I’m interested in adjudicating whether claims that voters punish incumbents for mishandling foreign policy are well-founded. In particular, what are we to make of the fact that the internationally popular, swift, and decisive 1991 Iraq War was followed by George H.W. Bush’s defeat, while the internationally unpopular, grinding, and essentially doomed 2003 Iraq War was followed by George W. Bush’s victory? Does this mean that Iraq “didn’t matter” for 2004/2006/2008? Such a finding would contrast with the claim that the Iraq war was at the core of Republicans’ electoral reversals in 2006 and 2008.

Although this lit review meandered a bit from a tight focus on the elections, the general findings seem defensible:

  • it’s really hard to establish direct causality between the war and election outcomes–if only we could run experiments!
  • there seem to be clear evidence that war casualties affected evaluations of the president and legislators
  • these effects were mostly driven by local news coverage and local elections (whether that ‘local’ is ‘state’ or ‘county’ remains to be seen)
  • the absolute biggest magnitudes of these effects are distressingly small–enough to shift a presidential election but not to wildly reject a challenger or incumbent on the basis of competence
  • approval for the war and vote returns for presidents seem to track perceptions of success and support for the decision to begin the war as well as costs
  • inference in the first term is compounded by rally effects from 9/11 and the start of the Iraq war
  • to the extent that Republicans suffered because of the war (a finding that seems reasonable), such electoral retribution was largely a result of local casualties  and relatively modest in scope.

Kriner and Shen (2007)

Douglas Kriner and Francis X. Shen, “Iraq Casualties and the 2006 Senate Elections,” Legislative Studies Quarterly

In examining Senate races in 2006, Kriner and Shen find that local casualties in the Iraq war mattered, but not national ones: “These results provide compelling evidence for the existence of a democratic brake on military adventurism, even in small-scale wars, but one that is strongest in communities that have disproportionately shouldered a war’s costs” (Abstract, 507). They present this finding in contrast with the commonplace idea that the 2006 midterms were “a resounding referendum on the Bush administration’s conduct of military operations in Iraq” (507). They look at Senate races, in particular those of Republican incumbents, to see how the “Iraq war’s influence on voting returns was critically mediated by local casualty rates at both the state and county levels” (508).

They highlight that the Iraq war involved very few combat fatalities (510), nearly twenty-fold less than that of Vietnam (as of 2006). They model change in Republican vote share between 2000 to 2006 in senate races at the state and county level. They test two contending explanations:

Iraq casualties might have affected the calculus of American voters [in] at least two ways. First, the total number of combat fatalities suffered in Iraq might have encouraged voters to abandon the Republicans, who … remained the most steadfast supporters of the president’s course in the Middle East. If this were the only mechanism by which the war affected the election outcomes, then Iraq’s adverse efect on Republican vote shares should have been felt nationwide, with little or no geographic variance. In such a world, we would find no evidence that Republican candidates did any better or worse on average in high-casualty states/counties than in low-casualty states/counties.

Alternatively, … the public’s perspectives on the war might also have been moderated by the experience of their local communities. If so, then residents of states and counties that suffered disproportionately high casualty totals or rates might also have felt the war’s costs more acutely and punished the ruling Republicans disproportionately.

They caution that their method would produce a conservative estimate for the war’s effect on voting behavior; finding local variation correlated with casualties may understate the “additional, uniform effect on voters as a whole.”

They analyze all Senate elections at the state level; casualties at the county level; and county-level casualties’ effect on Republican incumbents (14 races), excluding weirdos CT and VT. “Casualties” is KIA (515). Their DV is delta GOP vote share measured as function of casualties, opponent quality, campaign expenditures, 2004 Bush vote share, unemployment rate, percentage of population in the armed forces, and percentage of veterans.

They find a substantial effect of casualty rate at the state and county level on GOP vote share, one even stronger for incumbents. Kriner and Shen conclude (518) that this is a reason to reject the casualty threshold argument: “even a war with comparatively modest levels of casualties can have a substantial effect on congressional elections”. However, this is complicated at the county-level; notably, “the majority of counties as of November 2006 had not suffered a single battle death in Iraq.” (520) The  strongest evidence for blowback comes from county-level tests of GOP incumbents: “A two-standard deviation increase in a county’s casualty count cost the Republican incumbent more than two percentage points at the polls. Similarly, a two standard deviation increase in the county’s casualty rate decreased the Republican incumbent’s expected vote share by almost one percentage point from his or her 2000 performance” (522).

Karol and Miguel (2007)

David Karol and Edward Miguel, “The Electoral Cost of War: Iraq Casualties and the 2004 U.S. Presidential Election,” The Journal of Politics

Karol and Miguel proceed similarly to Kriner and Shen, with much the same model (delta Bush vote share 2000 to 2004), similar geographical precincts (state and county, with the addition of media-market), similar measures of casualties (KIAs), and the same caution: “Although our methodology allows us to gauge the impact of additional Iraq War deaths and wounded on voting outcomes across states, it cannot capture nation-wide trends in support for Bush related to the Iraq war” (635), although they puckishly note that “we cannot categorically exclude the possibility that the cumulative effect of the War was beneficial to President Bush’s campaign”: “Yet to the extent that we find a negative localized effect of casualties, those holding that the war aided Bush on balance would have to posit an even larger positive nationwide effect canceling out the negative local one we estimate.” Karol and Miguel note that their aggregation of casualty rates into media markets should allow them to better explore the impacts of mechanisms working through those other than personal contact (e.g., seeing reports on the news).

Karol and Miguel find a statistically and substantially negative correlation between state-level casualty (deaths and wounded) rate per 100,000 and Bush vote share change (639, 643) and county-level Bush vote share (644). They estimate that, in the absence of the Iraq war, “President Bush’s overall national vote share would have been 52.3% and that he would have won seven more states for a total of 39…[and] 366 Electoral College votes” (643). They note that “Seven scholars’ forecasting models–which largely focused on economic variables and ignored the war–predicted a larger victory for the Preisden than he actually achieved, with a median difference of 2.5 percentage points between their prediction and his actual national vote share”. As they argue, “our findings support the view that substance matters. Real-world events like casualties help account for the year-to-year shifts that relatively stable factors like party identification do not explain” (646).

Grose and Oppenheimer (2007)

Christian R. Grose and Bruce I. Oppenheimer, “The Iraq War, Partisanship, and Candidate Attributes: Variation in Partisan Swing in the 2006 U.S. House Elections”, Legislative Studies Quarterly

This is not quite the House version of Kriner and Shen (2007) but it is close. Grose and Oppenheimer claim that “Voters held individual Republican members of Congress accountable for the local impact of Iraq war deaths in their congressional districts and also held GOP members accountable for their roll-call votes on the Iraq war” (532). Notably, “Voters did not reward or punish Democrats for their votes on the Iraq war or the number of war deaths in their districts (532).

Grose and Oppenheimer seek to explain why different House races saw markedly different swings to the Democrats. They pay attention to the strategic factors that affect House races, including the way that favorable prospects for challengers may influence candidate recruitment (537).  They go a little beyond Karol & Miguel and Kriner & Shen, however, by taking seriously the difference between incumbency, national party brands, and individual legislators. They identify different potential mechanisms and think a little more on the theory side here (I think) than those (otherwise very good!) papers.  Part of this (539) is the fact that House districts are more homogenous in population but more heterogeneous in other ways, like media districts, than Senate districts.

They estimate delta Democratic vote share (2004 to 2006), excluding districts with idiosyncratic reasons not to be in (e.g. redistricted at midterm). I wish they had looked at 2002 to 2006, since midterm-midterm is a better comparison group. They control for Bush vote share percentage, incumebncy, quality challengers, scandal (oh, right, that was a big thing in 2006!), veteran and armed forces in district, and key treatment variables:

  • Iraq War Deaths in District (January 2006-December 2006)
  • Iraq War Vote (if the member voted for the Iraq war, HJR114)

They run separate models for each party’s candidates, implicitly interacting party with each term.

They find that Iraq war deaths in a district are associated with increasing Dem vote share–but only in Republican seats(545). In a rather stunning chart (547), they also found that there were huge effects for Democratic swing for fatalities in Republican districts: at the maximum (9 deaths), there was a predicted eight-point swing to Democrats (when there was an incumbent), which is the same as the open-seat swing (547). They estimate “changes in the predicted vote swing for Republican incumbents…[of] an approximately 1 percentage [point] increase in the Democratic swing for every two deaths per district.” There is no corresponding effect for Democratic-held seats.

Similarly, Republicans who voted for the Iraq war sufferend a 1.6 point Democratic swing; there was no corresponding swing for Democrats (549): “Fifty-four voters running in 2006 voted for the war in Iraq, yet the Democratic vote-share swing for these Democrats was no different than the swing for those Democrats who voted against the war or those Democrats who were not present for the war vote … voters punish the in-party for the Iraq war.” That is, “there was no local effect of the war in Democratic districts–results suggesting that the importance of anticipatory representation and retrospective voting is greater for the majority party” (552).

Grose and Oppenheimer conclude that (551) “as the number of district war deaths increased, the Democratic candidate had an easier time making the war issue resonate with the electorate. Compared wto previous U.S. conflicts, the number of war deaths per district is actually quite low. Perhaps surprisingly, these results suggest that American voters are extremely sensitive to a relatively small number of soldier deaths per district.”

Gelpi, Feaver, and Reifler (2005)

Christopher Gelpi, Peter D. Feaver, and Jason Reifler, “Success Matters: Casualty Sensitivity and the War in Iraq”, International Security 2005/2006

Gelpi et al stake a claim that the public will tolerate costs if the war is worth it:

In this article, we argue that the public will tolerate significant numbers of U.S. combat casualties under certain circumstances. To be sure, the public is not indifferent to the human costs of American foreign policy, but casualties have not by themselves driven public attitudes toward the Iraq war, and mounting casualties have not always produced a reduction in public support. The Iraq case suggests that under the right conditions, the public will continue to support military operations even when they come with a relatively high human cost. Our core argument is that the U.S. public’s tolerance for the human costs of war is primarily shaped by the intersection of two crucial attitudes: beliefs about the rightness or wrongness of the war, and beliefs about a war’s likely success. … Ultimately, we find that beliefs about the likelihood of success matter most in determining the public’s willingness to tolerate U.S. military deaths in combat.”

This paper employs “a close examination of polling data from the beginning of the Iraq war through the 2004 U.S. election—that is, for the first twenty months of the war.” This is, then, explicitly a national level test and one that unfolds over time, rather than an Election Day snapshot.

Gelpi et al find that the war’s public approval passed through three distinct phases: major combat operations, insurgency, and “post-sovereignty” (when the CPA ended). Accounting for this delivers the result that different epochs of the war had different relationships between presidential approval and casualties: total combat casualties are positively and significantly related to presidential approval (!) but those during the occupation had a much larger, negative association with presidential approval: “after the United States transferred sovereignty to an Iraqi authority, the impact of casualties shifts again…the model indicates that, between June and November 2004, casualties had no impact on presidential approval.”

They also employ their own surveys to see if belief in future success affects casualty tolerance. They find it does (34-36). They also find (at 37) that party ID was at the same scale of effect on approving of high military deaths as whether a voter perceived an elite consensus on the war or whether voters perceived Iraq to be a ‘distraction’ or the ‘central front’ of the war on terror.

By the way, I wonder why this is judged to be a measure of prospective, rather than retrospective, support. My favorite findings in political science relate to the strong retrospective bias in current evaluations.

Berinsky and Druckman (2007)
Adam J. Berinsky and James N. Druckman, “Public Opinion Research and Support for the Iraq War,” Public Opinion Quarterly

Berinsky and Druckman respond directly to Feaver, Gelpi, and Reifler, noting that their study formed the basis for a shift in the Bush administration’s rhetoric (2). They launch methodological criticisms of the work focused on the idea that “methodological concerns … [make] it difficult to distinguish between their hypothesized causal hypothesis and that of other leading theories of the determinants of war support” (3). They modestly describe their concerns as relating to “three integrally related dimensions of the FGR analysis: the dependent variable, the independent variables, and, ultimately, their causal claims” (4), which, I suppose, leaves the punctuation.

Dependent variable. Berinsky and Druckman argue that FGR  use too few measures of war support, given that they only use  one explicitly measured dependent variable: “individual casualty tolerance.” As they note, “FGR’s dependent variable inherently assumes that support for war should be measured by the number of American war deaths a respondent is willing to bear … ironic since FGR in fact argue that casualties are not the primary determinant of support for war” (6).

Independent variable.  FGR focus on “perceptions of success” as key explanatory variables. As Berinsky and Druckman argue, though, “it is not clear how best to give meaning to the cross-sectional variation in individual perceptions of success” (8). One problem is that partisanship matters (8) by influencing people’s predictions of success: “In sum, we suspect that FGR’s ‘success’ variable is an indicator of support for war, not a cause of support for war. Like other measures of war support, it is influenced in part by partisan predispositions” (10). They separately dismiss ‘initial support’, FGR’s other key independent variable, by pointing out that “respondents’ evaluation of the initial decision [to go to war]” cannot easily be differentiated from “their contemporaneous evaluation of the war” (10). They suggest instead that, as with the dependent variable, a better measurement would have used (or been derived from) a multiple-measures approach to yield a latent-variable indicator (11-12). And, cuttingly, they argue that (12) “we suspect that the two independent variables of interest may actually be better indicators of latent support for the war in Iraq than their dependent variable.”

Causality. Berinsky and Druckman write that “it is very difficult to establish causal relationships in cross-sectional nonexperimental data with such highly related variables” (13). They further write, “We therefore find ourselves in agreement with former Gallup Vice President David Moore who concludes, ‘The causal model cannot be proved, at least by the data obtained by the three authors; in this case, causality is more an act of faith than a provable dynamic” (13).

Gelpi and Reifler (2008)

Christopher Gelpi and Jason Reifler, “Reply to Berinsky and Druckman: Success Still Matters”, Public Opinion Quarterly

Gelpi and Reifler respond to Berinsky and Druckman. They argue that the dependent variable needs to measure costs somehow; how else to do this? They argue — I think a trifle speciously — that Berinsky and Druckman advocate throwing up our hands and not measuring this. They similarly mostly dodge on questions related to the measurement of the independent variable, including party ID (although I think their response that success is different than casualties is pretty on point). They also have a pretty pointed response to the question of the dimensionality of Iraq attitudes, including the fact that there’s only enough indicators to find one dimension and consequently it’s unsurprising that they find only one. Mostly, the article seems like a defensive attempt — holding onto their claims while giving ground and agreeing that “more work is necessary.”

Fox (2009)

Gerald T. Fox, “Partisan Divide on War and the Economy,” Journal of Conflict Resolution

After seven pages of lit review, this paper gets to the point: “The present analysis tests three partisan influences on Bush approval, consisting of the partisan macroeconomic cleavage, an in-party/out-party rally effect, and a war-economy partisan effect” (910) by testing “the in-party/out-party hypothesis for the 9/11 and the Iraq War rallies on Bush approval” (911). Fox notes (912) that “Out-party identifiers have stronger rallies [in presidential approval] but also faster and deeper defections than in-party identifiers”. Using monthly time-series data (March 2001 to December 2008; 916), Fox estimates approval using a three-equation seemingly unrelated regressions. Notably, he find that log(war fatalities) does not effect Republican support but strongly affects Democratic and Independent support.

Fox finds that “Democrats give fleeting credit to a Republican president for rally events, while Republicans give more sustained credit” (919). He also finds that Democrats and Republicans difered in interpreting economic events: “Gasoline prices and unemployment are significant influences on presidential approval across all groups except Democrats, whereas the war fatalities effect is significant across all groups except Republicans” (927).

Voeten and Brewer 2006)

Erik Voeten and Paul R. Brewer, “Public Opinion, the War in Iraq, and Presidential Accountability,” Journal of Conflict Resolution

Voeten and Brewer take on the Gelpi, Feaver, and Reifler article from a different perspective. They note that voters in a democracy (should) want to reward competence and punish incompetence, but because competence is not directly observable citizens have to figure out whether a president is competent through observing outcomes. Yet “the implications of individual [short-run] outcomes for a leader’s competence may be ambiguous and open to competing interpretations” (811). They contrast a managerial model (“how well is this war going”) of inferring accountability (“if the war is going well, the president must be a good manager”) with a decision-making model (“how good was the decisions to go to war”), which implies a different evaluative process (“if the decision to go to war was good, then the president makes good decisions”). This implies voters use retrospective evaluations to shape their current beliefs about a president.

From this, they argue that “events and casualty reports affect perceptions of the war’s success but have relatively little direct impact on support for the war.” On the other hand, “shifts in aggregate war support have stronger effects on overall presidential approval ratings than do shifts in beliefs about the war’s success or even beliefs about whether President George W. Bush is ‘doing a good job’ in handling Iraq.” Therefore, “citizens primarily held President Bush accountable for his perceived decision-making qualities rather than his managerial qualities” (810-811). They note that “aggregate opinion about the war’s success is likely to be more volatile than opinion about its merits” (813). Therefore, “shifts in war support predict shifts in overall presidential job approval more strongly than do shifts in a more direct measure of perceived managerial abilities: perceptions of Bush’s handling of Iraq” (814).

Brewer and Voeten use a multiple-indicators approach (war support, war success, Bush approval on Iraq, overall Bush approval) to measure (weekly!) changes in aggregate beliefs while controlling for perceptions of the economy. The trick is that “aggregate war support and perceptions of war success followed roughly the same pattern” (819). They measure casualty sensitivities by using the log of cumulative casualties and also the marginal change in the number of cumulative casualties from the preceeding week. They also (and it’s weird that they are among the first to do this!) look to measurements of large-scale Iraqi deaths. They include FGR’s three periods (major combat, occupation, and sovereignty) as well.

They find (822) that lagged casualties affected change in war success but not in war support, Bush approval, or overall Bush approval. Political events had a bigger effect on war support and Bush approval, such as Abu Ghraib (negative), Saddam’s capture (positive), and the stampede of August 2005. (negative). They also find that “weekly shifts in war support had a strong immediate effect on changes in presidential approval” (826): “A 1-percentage point increase in support for the war led to a .71-percentage point immediate increase in overall presidential approval…the corresponding effects for perceptions of war success and Bush approval on Iraq were .28 and .47, respectively.”

They explain these shifts with respect to elite signaling–“the nature of elite discourse about a war is a key determinant of changing patterns in both war support itself and presidential approval during wartime” (828). Therefore, ” a plausible interpretation of the trend in support for the war in Iraq is that increasing elite polarization regarding the war produced not only polarization among the public (with Democratic and Republican partisans diverging on its wisdom…) but also declining support for the war in the aggregate” (828).

Hill, Herron, and Lewis (2010)
Seth J. Hill, Michael C. Herron, and Jeffrey B. Lewis, “Economic Crisis, Iraq, and Race: A Study of the 2008 Presidential Election,” Election Law Journal

This is a fine empirical paper that argues that economic conditions and the Iraq war tended to help Barack Obama in his candidacy for the White House in 2008. The authors employ a county-level study of determinants of vote share; they find significant but rather small effects of unemployment and mortgage delinquency on Obama’s share but huge contextual effects of the Iraq and Afghanistan vote share. From their hierarchical model, they estimate that “a hypothetical 2008 presidential election held in a nation with zero Afghanistan and Iraq war fatalities would have decreased Obama’s national share by 5.2 points” (42).

I am relatively unpersuaded by this paper, however, because its evidence seems so far beyond that of the others surveyed. Unlike Shen and Kriner or Karol and Miguel, for instance, Hill et al do not find any county-level effect of casualties on presidential vote share. Their argument is unpersuasive:

We speculate that this contrast [with earlier findings] is driven by the relative obscurity of county-level war casualties compared to, say, the obviousness of local unemployment, of shuttered storefronts, of a plethora of house for-sale signs, and so forth. Perhaps presidential candidates, with the Electoral College in mind, use state-level Iraq and Afghanistan figures in their advertisements that, due to media market boundaries, cross county lines. (54)

First, I don’t believe that anyone ever says that “the Iraq war has killed N Californians”; I’d be glad to be proven wrong. Second, this is precisely the opposite reasoning from the other, better-specified papers. Third, I can’t believe the counterfactual because neither can the authors. The authors explicitly rule out modeling a counterfactual election with a 90% drop in the financial markets because “a drop of this magnitude [would be] so large and so outof sample, broadly speaking, that suggesting our model can internalize it would strain credibility” (58). However, they then proceed to model a zero-fatality Iraq and Afghanistan election, even though “our models do not have any leverage on what a 2008 presidential race would have looked like if the contesting candidates were, say, Democrat Hillary Clinton and Republican Mitt Romney” (59).