British Public Reactions to Military and Civilian Casualties: Evidence from a Dynamic Panel Experiment.

Transcription

1 British Public Reactions to Military and Civilian Casualties: Evidence from a Dynamic Panel Experiment. Graeme A.M. Davies University of Leeds and Robert Johns 1 University of Essex Paper to be presented at the workshop on Public opinion, foreign policy and the use of force in turbulent times at the ECPR Joint Sessions, Nottingham, April Contact author, rajohn@essex.ac.uk. 1

2 This has been a busy couple of decades for those studying the relationship between public opinion and Western foreign policy. Not only has there been a series of military interventions Iraq in 1998, Kosovo in 1999, Sierra Leone in 2000, Afghanistan in 2001, Iraq again in 2003, Libya in 2011 and Syria in 2014, to name some but by no means all but domestic public opinion has become central to discussions over the feasibility and legitimacy of such interventions. If politicians ever were waltzing before a blind audience when it came to foreign policy, they are certainly not doing so now. When the UK s House of Commons voted against its government s bid to join military action in Syria in 2013, the restraining hand of public opinion on the shoulder of some parliamentarians was obvious, and the return of coalition troops from Afghanistan and Iraq had more to do with the unpopularity than with the accomplishment of their missions. In the near future, Western governments wishing to instigate and maintain military operations will need at the very least to heed the views of their electorates. One reason that those electorates are currently on the sceptical side is that the costs and risks of force have been brought to the fore in recent engagements. Prominent among these are military casualties. In the UK, much political and media attention was given to the country s troop losses in its recent major military operations in Iraq and Afghanistan even if these death tolls, of 179 and 412 respectively, 2 were dwarfed by the corresponding figures for the US military. In that American context, it has long been hypothesised that public support for war falls as the casualty toll mounts, and the form and contingency of the relationship between those two variables has been subject to thorough investigation. Yet, despite the UK s active engagement in many conflicts around the world, we know rather little about British citizens attitudes towards military action and almost nothing about their reactions to troop losses. One major contribution of this study is therefore to test US theories of casualty aversion in a different but strategically relevant context. We track how the British public reacts to military casualty rates and how this factors into a wider calculus of support for war. 2 (accessed 8 th May 2012); (accessed 29 July 2011); (accessed 29 July 2011) 2

3 However, we are not simply replicating US studies of military casualty sensitivity on a different public. This study also extends that research in three important directions. First, while previous work has focused more or less exclusively on military casualties, we take account of the fact that there are other lives lost in war. While British troops have been the major focus of the country s politicians and media, most citizens will nonetheless be at least dimly aware that recent military operations have also resulted in the deaths of foreign civilians. While estimates of civilian deaths often vary wildly (e.g. Burnham et al. 2006; Iraq Family Health Survey Study Group 2008), they are generally agreed to outnumber military casualties by an order of magnitude. This makes the lack of research concerning the effect of civilian casualties on public support for military action particularly striking (a notable exception being Walsh, 2014). Perhaps the neglect of civilian casualties reflected an assumption that the public would be concerned only or mainly with its own military casualties. We put that assumption to the test. Meanwhile, especially in recent years, Western states have deployed private military security company (PMSC) contractors (Krahmann 2010) in place of their own troops. These contractors were especially heavily involved in Afghan operations, with between 38,000 and 107,000 committed to theatre and an estimated 763 losing their lives by mid By comparing public reactions to military and PMSC casualties, we assess whether outsourcing operations can reduce public sensitivity to battlefield casualties and, in turn, allow democratic states to wage longer, costlier operations. Second, we adopt and build on the panel experiment technique method pioneered by Gartner (2008), His work marked a departure from the two principal research designs in work on casualty aversion in the US: observational time series data tracking the correlation of casualty numbers with public opinion polls (e.g. Mueller, 1973) and one-shot experiments manipulating casualty information and observing the effect on public preferences (e.g. Boettcher and Cobb, 2006). With a dynamic panel experiment, we retain the internal validity advantages of experimental designs but can also introduce a longitudinal component, investigating whether and 3 July 31, 2011(accessed December 30, 2015). 3

4 how citizens respond not only to casualty numbers but also trends in those numbers. For example, we can assess whether citizens tolerate more civilian deaths than military deaths but nonetheless eventually turn against military action once the civilian death toll accelerates. Third, building on work demonstrating the importance of the context in which casualty information is provided, we manipulate three different aspects of that context in order to provide a more contingent understanding of public reactions. First, we test whether prior casualty numbers only really provide clear contextual information when they are presented graphically (as by Gartner, 2008) in a way that makes the trend unmistakable. Second, we replicate Boettcher and Cobb s (2006) test of whether reactions to military casualties are moderated by information on the enemy s losses. Third, we test whether the reason for military action and, in particular, whether it is a humanitarian engagement intended to minimise civilian suffering makes the public more averse to civilian casualties than is otherwise be the case. In sum, then, here we present the first systematic analysis of British public responses to trends in wartime casualties, both military and civilian, taking into account how those reactions might be moderated by various contextual factors. Our core conclusions tend to conform to a rationalist model of casualty reasoning, with information on numbers and trends factored into a calculation about the costs, benefits and progress of military action. This leaves governments with very little mileage to maintain support for war by outsourcing losses to PMSCs, some but limited scope to inflict foreign civilian casualties, and not as much scope as previous research might suggest to frame casualties in a way that mollifies the public. Casualty type and casualty aversion A wealth of literature has been devoted to understanding the US public s sensitivity to military casualties. It can be traced back to Mueller s (1973) seminal study of public approval for military action in Vietnam. He found that logged cumulative casualties reduced support for war over 4

5 time. Further research went on to assess the political consequences of military casualties, estimating their effect on the popularity of the leader who instigated that action (Gartner and Segura 2000) and on US Congressional (Carson et. al. 2001; Gartner et. al. 2004) and Presidential elections (Voeten and Brewer 2006; Karol and Miguel 2007). With events in Iraq and Afghanistan a renewed interest in military casualties led again to a huge swathe of research into public responses into casualties (Eichenberg 2005; Gelpi, Feaver, and Reifler 2005/06; Boettcher and Cobb 2006; Eichenberg, Stoll and Lebo 2006; Voeten and Brewer 2006; Althaus and Coe 2007; Berinsky 2007; Gartner 2008a, 2008b; Boettcher and Cobb 2009; Baum and Groeling 2010). Two different perspectives emerged: the rationalist view that members of the public appear to be engaging in simple calculations about the expected value of continuing to engage in armed conflict (Gelpi, Feaver and Reifler 2005); and the elite-driven explanations in which what matters not the unmediated information about casualty rates but the way in which this is framed by elites (Berinsky 2005; Boettcher and Cobb 2006). The persistence of both perspectives strongly suggests that the public is responsive to both numbers and frames, as for example when Boettcher and Cobb (2006) found that also reporting numbers of opposition deaths that is, placing US military casualties in a broader context of battlefield success served to soften but not to eliminate the effect of those US casualties on public approval of war. The focus of all this research has been restricted in two respects. The first is the dominance of the US context. There has been virtually no research examining how the public in other countries respond to casualties. Second, the focus has been largely on military casualties. We know far less about the potential effects of two other types of wartime casualties: PMSC contractors and civilians. Do these also have some influence over public approval of military action, and is that influence dwarfed by the effect of military deaths? The overwhelming focus on military casualties in the literature tends to suggest the latter, but it may be that the public is 5

6 responsive to a broader range of casualty information. Next, we assess reasons why citizens might and might not factor PMSC and civilian casualty information into their support-forwar calculus. PMSC casualties The relationship between citizenship and military service is thought to significantly affect the effectiveness of armed forces. Mercenaries are thought not to be particularly effective soldiers as they are primarily interested in the acquisition of wealth rather than the protection of the state, whereas soldiers selected from the citizenry will be more effective because they are fighting for the survival of their country (Avant and Sigelman 2010). Kant (1795) preferred citizen armies because they would restrain the warlike tendencies of leaders as the population would be cautious about entering into a war which they would directly pay the costs. We are particularly interested in the effect this caution has on the level of public consent for continuing with a war where many citizen soldiers are losing their lives. Would concern about military casualties be lessened if those casualties were modern mercenaries, otherwise known as PMSC contractors? If casualty information is relevant primarily as a signal for the success or otherwise of operations (see Johnson and Tierney, 2006), then the answer will be no. Regardless of the origins and motivations of military personnel, escalating death tolls are a worrying sign. But this sounds too strictly rationalist an expectation. Even if casualty information is ultimately factored into a cost-benefit calculus of support for war, there are reasons to expect PMSC deaths to be costed less expensively. For one thing, their involvement in military operations may be tainted by the profit motive. Even though most Western military action is conducted by volunteer armies rather than conscripts, the dominant discourse about those recruits is one of service rather than financial gain. And, since the service in question is seen as being to their country, we have a related reason to suppose lesser sensitivity to PMSC deaths, namely that the latter do not represent (and typically do not belong to) the national in-group. A third reason is rather 6

7 different, and relates to public ignorance of PMSCs who they are, and what they do. Governments are seldom keen to publicise their outsourcing of military engagement and media coverage of PMSC activities and fatalities is scant and patchy (Avant and Sigelman, 2010; Kruck and Spencer, 2013). Contrasted with the detailed and emotive set of associations triggered by reports of British military deaths, then, reports of PMSC casualties seem unlikely to have the same cognitive or affective impact. To our knowledge, there is only one study comparing public reactions to military and to PMSC casualties, and its null findings are perhaps surprising in the light of the arguments above. Avant and Sigelman s (2010) survey experiment found that the US public showed the same emotional response to soldier and contractor fatalities; moreover, varying casualty type had no impact on public assessments of the perceived justification or effectiveness of military action. We aim to see whether this result replicates to a different time and context. If this result holds, then there is less to gain from outsourcing to PMSCs than political leaders might think. If, however, the public does price contractor deaths less highly, then outsourcing reduces the domestic political costs of casualties and, in turn, eases the constraints on the British and other governments from getting involved in potentially risky and costly conflicts. Civilian casualties As with PMSCs, and via similar ingroup versus outgroup reasoning, we would also expect greater public aversion to domestic military than to foreign civilian casualties. The ingroup/outgroup hypothesis (Simmel 1955; Coser 1956) has been extensively used in the international relations literature to explain why states initiate international conflicts in response to domestic dissatisfaction (Levy 1989; Gelpi 1997; Davies 2002; Foster 2006). Outgroups suffers from negative and stereotypical perceptions that make their deaths less important (Park and Judd, 1990; Read and Urada, 2003; Friedrich and Dood 2009). The relevance of this here is obvious. First, the foreign-ness of civilian casualties already gives them outgroup status (Rothbart and 7

8 Korostelina 2006). And some may reject a distinction between combatants and non-combatants and see civilian deaths as just another example of attacking the out-group. There may also be a reciprocity argument, the public reasoning that foreign civilians would do the same to us or, for example via terrorist attacks, regarding them as having already done so. Then there is the fact that, even if the public wishes to avoid the deaths of foreign civilians, they may see this as conflicting with the even higher priority of keeping (ingroup) troops safe as in the classic strategic choice between air strikes and ground invasion. A final reason to expect lesser public responsiveness to civilian casualties is what is known as proportional numbing, whereby people become less sensitive to losses of life when those statistical losses are construed as small fractions of larger reference groups of at-risk individuals (Friedrich and Dood 2009, 2542). Since the population of foreign civilians dwarfs that of military personnel, we can again expect the public to react more sensitively to the latter type of casualties. Even if these factors lead us to expect a relative insensitivity to civilian casualties, they do not necessarily support a stronger hypothesis that the public will simply ignore mounting civilian death tolls. There are reasons to suppose at least some civilian casualty aversion. Some of these are straightforwardly normative: non-combatants may be regarded as innocent bystanders, there may be a reciprocity argument here, too how would we like it if they attacked our civilians rather than our armed forces? and there. There are also more instrumental reasons for the public to react against civilian casualties. Disregard for civilian lives erodes international support for military action and hardens grievances in the adversary state (Condra and Shapiro 2012). Finally, and most importantly for present purposes, citizens may read use civilian casualties as a cue for judging the success of combat operations (Baum and Groeling 2010). If fewer civilians are dying at time t than at t-1 then it will appear to the public that the intensity of operations is easing and the situation is returning to relative normality. Conversely, if there is an upward trend in civilian casualties then the public will perceive that the situation is deteriorating. It would seem logical that the cueing effect of civilian casualties should be strengthened where the central 8

9 motivation for action in the first place was to alleviate civilian suffering. Boettcher (2004) shows that the number of civilians under threat in humanitarian crises did have a weak positive impact on American (students ) willingness to intervene militarily on their behalf, from which we can infer at least some concern for civilians in support-for-war judgments. As with PMSC casualties, data on public reactions to civilian casualties are limited but provide at least some evidence of casualty aversion. Clarke et al. (2009) used logged civilian casualties during the Iraq war as an indicator of progress during the conflict and included it in a model estimating aggregate approval of Tony Blair. They found that worsening conditions on the ground significantly reduced Blair s approval ratings, arguing that the British public use civilian casualties to evaluate the progress of the war. Eichenberg s (2005, 172) collation of polling data indicates that mentioning civilian casualties in a question stem reduces support for war by about ten percentage points. Attacking the same research question experimentally, Gelpi et al. (2009, 256-7) recorded a 17-point difference between the treatment question asking about approval of a hypothetical military action if it resulted in 2,000 civilian deaths and a control question with no such clause. The size of these effects might be inflated by the blatancy of the primes but, at the very least, they provide evidence of potential aversion to civilian casualties. This is then bolstered by the more traditional survey experimental findings reported by Walsh (2015) and Johns and Davies (2016). Based on population samples in the US and UK, these studies show support for force to be responsive both to the mention of such casualties and to an estimate of numbers (via a manipulation of the likely proportion of civilians among those killed). Indeed, the prospect of civilian casualties in Walsh s first experiment seems to trump that of military casualties, although the comparison can only be rough because only in the military case was an estimate of numbers provided. A similar result is obtained in Friedrich and Dood s (2009) experiments gauging casualty tolerance among small samples of US college students. In each case manipulating whether the question concerned US troops or foreign civilians, they first ran a study asking each participant how many more casualties in Iraq he or she would be willing 9

10 to accept before wanting the US to withdraw. Tolerated death tolls were not significantly higher among participants considering civilian casualties. Overall, then, while we have reason to suppose that the public will react most strongly to their own military casualties, it is at least worth examining whether and how they react to PMSC and civilian deaths. For now, we specify two simple hypotheses: H1(a): The public will be more sensitive to military casualties than to PMSC contractor casualties. H1(b): The public will be more sensitive to military casualties than to civilian casualties. Casualty context and casualty aversion We have already reviewed extensive research suggesting that individuals apply cost-benefit calculations when deciding whether to maintain support for military action, and that casualty numbers feed into these calculations. The key insight from Gartner s (2008) study, the point of departure for our own work, is summed up by its final sentences about the meaning of those casualty numbers: Meaning requires context and casualty patterns provide the lens through which individuals interpret losses. Without this context, casualty figures are just numbers (p. 105). The loss of twenty British soldiers in a given month will take on a very different meaning depending on whether the previous month saw as many as fifty military casualties or as few as five. In particular, an upward casualty trajectory over the months of the war will tend to suggest that there is little chance of military victory, at least in the short term. This is likely to convince waverers among the public that their government should cut its losses and withdraw troops. Gartner (2008) himself finds clear evidence that support for continued military action is influenced not only by recent casualty numbers but also by the trends within which they are interpreted. His work provides a clear basis for H2(a): H2(a): Increasing casualty rates will reduce the duration of military interventions. 10

11 By contrast, Gartner finds no support for a subsidiary hypothesis. If citizens read information about the progress of the war from casualty rates, then we might expect them to be particularly alarmed if those rates begin to accelerate. Yet Gartner finds no significant difference between reactions to linear and to curvilinear increases in casualty rates. We retest that hypothesis here: H2(b): This is especially the case if those increases are curvilinear upwards. Whereas Gartner s work involved only military casualties, we are also interested in reactions to trends in PMSC and civilian casualties. That raises the question of whether we would expect those reactions to differ depending on the casualty type. It may be that there is no interaction here: that the patterns hypothesised in H2 are the same regardless of who is lost, and it is just that as per H1 those reactions remain stronger in the case of military deaths. Alternatively, it may be that the public is not only more responsive to the prospect of military casualties but is also particularly responsive to trends in the military death toll either because they care more about these ingroup deaths or because they see military losses as the clearer signal of success or failure. Since both of those suggestions are well backed up in the literature, we hypothesise thus: H3: The tendency for increasing casualty rates to reduce the duration of military interventions is stronger when these are military than when they are PMSC or civilian casualties. One notable feature of Gartner s study is that he used graphs to report casualty information to respondents. Much research demonstrates that images in general are able to stimulate stronger emotional responses than textual information alone (Joffe 2008; Perlmutter 1998; Iyer and Oldmeadow 2006) and, even if line graphs are themselves are hardly the most poignant of such images, they demonstrate trends more quickly and clearly than text alone (Smith and Joffe 2009). There is reason at least to suggest that this immediate access to contextual information will have amplified its effects on support for military action. We test this conjecture, and the public s capacity to process information about casualties without graphical cues, via: 11

12 H4: The public will be more sensitive to casualty trends if those trends are presented graphically. Another aspect of context for military casualty information is highlighted by Boettcher and Cobb (2006). They find that opposition casualty frames, sometimes vividly described as kill ratios, moderate the effects of military losses. Those reading in a newspaper that five British soldiers were killed in combat today might well react more unfavourably than those reading that five British soldiers and thirty Iraqi insurgents were killed in combat today. The extra information not only helps to place the military casualty estimate in a kind of numerical context, but also provides another cue about the success of the campaign (Boettcher and Cobb 2006; Baum and Groeling 2010). During traditional military operations, the public can judge success by the capturing or recapturing of territory (such as Berlin during WWII, or Port Stanley during the Falklands war), but in many more recent engagements the objectives have been less clear-cut. This makes other cues to success more useful, and a favourable kill ratio may therefore offset the negative effect of troop casualties on support for action. Rightly or wrongly, it may persuade the public that the soldiers did not die in vain. H5: More favourable kill ratios will reduce sensitivity to British military casualties. Our final extension to Gartner s work is to introduce dispute type as another aspect of casualty context. We suggest that the public will be more tolerant of casualties if they perceive that force is being used to defend vital national interests. Effectively, this adds to the benefits side of any cost-benefit analysis, and so makes the human price of war more worth paying. In the face of the detailed academic debate about what constitutes the national interest (Rice 2000; Williams 2005), our empirical approach here is unavoidably rather reductive. We contrast two scenarios: one involves protecting the rights and economic interests of British citizens, and we designate this as a national interests campaign; the other involves the protection of foreign nationals, designated as a humanitarian intervention. Our first hypothesis was already stated above: 12

13 H6(a): The public will be less sensitive to casualties during national interest campaigns than during foreign national protection campaigns. There is an exception to H6(a), however. Since the motivation for a humanitarian intervention is to protect the lives of foreign civilians, it might seem perverse to maintain operations even in the face of rising civilian death tolls. This is not necessarily to suggest that H1(b) will not apply, but we would at least expect any tendency for civilian casualties to be more tolerated than military casualties to be less pronounced in a humanitarian crisis. This is true on strategic as well as emotive grounds: in a humanitarian scenario, civilian deaths are probably a better indicator of failure than they are during a campaign to protect Britain s economic interests. We hypothesise: H6(b): The public will be more sensitive to civilian casualties during a humanitarian than during a national interests intervention. Data and experimental design We are following the approach pioneered by Gartner (2008) in building a dynamic panel experiment that simulates changing casualty rates during an international conflict. This experiment allows us to separate the effects of the most recent report of casualties from those of the longer-term pattern into which it fits. Data The dynamic panel experiment was included in a major three-wave panel study of foreign policy attitudes among the British public. The surveys were administered over the internet by YouGov, the leading on-line opinion research company in Britain. This is the same mode of data collection an experiment embedded in a YouGov survey of a representative sample employed by, for example, Tomz and Weeks (2013) in their study of public attitudes and the democratic peace. YouGov s approximately 300,000 panel members formed the sampling 13

14 frame. 4 The company has an impressive track record of sampling and weighting to achieve representative samples of the British electorate. The survey as a whole is not a traditional panel. The aim was rather to divide a long instrument into three manageable chunks, and to field these at very brief intervals so that the entire process of data collection took less than a month. Since the panel experiment here was taken from a single wave (Wave 3) of the survey, the staggered approach to data collection has minimal implications for this paper. The fieldwork was conducted between 15 and 22 February 2010 (around four weeks after the first wave), with 86% of responses received in the first two days of that period. The sample size is 2,066. This represents a retention rate of 79% from Wave 1 (N=2,615), which itself had a response rate of 61%. In all, then, the relevant wave of data collection was completed by 48% of those approached prior to Wave 1. 5 Experimental design Our experimental design, and specifically the manipulation of casualty trends, is based closely on that of Gartner (2008). The differences stem from our additional interests in whether casualty type, graphical presentation, opposition casualties and crisis type also shape casualty sensitivity. Respondents were presented with a vignette describing an overseas crisis in response to which the UK has already begun a military intervention. All respondents saw the following opening statement: Finally, we would like to ask about another type of situation that might face Britain in the future. Here are some details about the scenario. Rebel military insurgents have taken over Guyana, a former British colony in South America. All respondents were then told that The UK has begun a military intervention, deploying the 16th Air Assault Brigade and 4 Most of these are actively recruited (by targeted campaigns via non-political websites) rather than volunteering for the panel. Similarly, respondents are not able to choose in which surveys to take part: they are either sampled for a given data collection or not. Still, it should be acknowledged that volunteer panels like this not only offer nonprobability samples but also disproportionately politically involved respondents. Since our primary aim here is to compare across experimental groups, though, this unrepresentativeness should have minimal effects. 5 Since members of the opt-in YouGov panel do not have a known probability of selection, it is not feasible to calculate a response rate taking into account all sources of non-response, including panel recruitment and retention. Here, then, response rates are in effect completion rates, representing the proportion of those asked to take part in that survey that agreed to do so. 14

15 3 Commando Brigade around 12,000 troops in total with the objective of defeating the insurgency and restoring the previous government. Some of the manipulated information on casualty type and conflict type was also included in this initial vignette. Once the scenario had been presented, respondents read In the first of its monthly updates on the progress of the military action, the Ministry of Defence reported and were then given information on military, PMSC or civilian casualties in some cases along with reports of opposition insurgent deaths. After this first report, respondents were asked the question that constitutes our dependent variable: Which of these do you think that the British government should do: continue the military intervention in a bid to restore the previous government, or abandon military action and withdraw troops? Choosing the latter option, termination of the action, took respondents to the end of the survey. Those that chose to continue military intervention then received a second casualty report (and, now that at least an initial trend had begun, possibly a graph of the casualties so far), after which they were asked the dependent variable question again. This was repeated up to a maximum of ten casualty reports. The vignette and manipulations are provided in Table I, while Figure I sets out the panel structure of the experiment. 15

16 Table I: Wordings of experimental vignettes and manipulations Manipulation (conditions) Crisis type (2) Casualty type (3) Casualty report (10) Opposition casualty frame (4) Conditions National interests Humanitarian British soldiers Private military security company (PMSC) Civilian casualties See figure 2 Wording Rebel military insurgents have taken over Guyana, a former British colony in South America, and have seized property and assets from the thousand British people living and working there. Many foreign nationals, including some British citizens, have also had their passports confiscated. Rebel military insurgents have taken over Guyana, a former British colony in South America, and have executed dozens of their political opponents, imprisoning and torturing many more without trial. The UK has begun a military intervention, deploying the 16 th Air Assault Brigade and 3 Commando Brigade around 12,000 troops in total with the objective of defeating the insurgency and restoring the previous government. The UK government has hired Ajax Defence Services, a private military contractor that is, a commercial company that sells the services of armed and trained soldiers to intervene. No mention of civilians except in casualty report In the first of its monthly updates on the progress of the military action, the Ministry of Defence reported that (ASSIGNED NUMBER) British troops/ajax personnel/civilians lost their lives during operations in Guyana this month. -20% As above but with and (Assigned number -20% of British Casualties) rebels lost their lives during operations in Guyana this month. +20% As above but with and (Assigned number +20% of British Casualties) rebels lost their lives during operations in Guyana this month +200% As above but with and (Assigned number +200% of British Casualties) rebels lost their lives during operations in Guyana this month No mention No mention Casualty trend (3) Constant Linear increasing See Figure II See Figure II Curvilinear See Figure II increasing Graph Present Casualty trend is represented in graphical form Not present No graph accompanies monthly report 16

17 Figure I: Panel experiment design Scenario Casualties(1) Casualties(3.10) Casualties(2) End Continue (1) Continue (2) Continue (3 10) Stop Fighting (1) Stop Fighting(2) Stop Fighting (3..10) The manipulation of casualty trends across monthly reports is at the heart of our experiment and warrants specific attention. Figure II shows the three casualty patterns that we included in the experiment: linear increasing, curvilinear increasing and constant casualty trends. 6 As the graph shows, the term constant is a slight misnomer: in order to make the experiment feel more realistic and less contrived, we varied casualty numbers by a maximum of ±2 around a mean of 30 monthly casualties. Another modification of Gartner s design is that we reduce the scale of casualty numbers overall, reflecting the smaller size of the British army and population. The cumulative casualty count comes to 277 for all of the different trends. 6 Unlike Gartner (2008), we do not have a curvilinear decreasing trend. This reflects a quest for parsimony given that we have several additional manipulations and need to maintain a sizeable N in the sub-groups. Moreover, since Gartner did not include that curvilinear decreasing trend in all of his models, we can exclude it while still being able to compare our results directly with his. 17

18 Figure II: Casualty trends used in the experiment All constant All linear increasing All curvilinear increasing Some of those in the military casualty report conditions were also told in each monthly update about the number of opposition fighters ( rebel forces ) that had lost their lives. We included three conditions of this opposition casualty manipulation: 20% fewer casualties, 20% more casualties, or 200% more casualties (with the latter serving as a kind of acid test of the capacity of opposition deaths to reduce public sensitiveness to military casualties). These patterns are shown in Figure III. 18

19 Figure III: Example military casualty pattern (curvilinear increasing) with alternative opposition casualty frames Soldier Casualties (Curvilinear) Enemy Casualties 20% More Enemy Casualties 20% Less Enemy Casualties 200% More Results Once again following Gartner (2008), we use Cox proportional hazards models to estimate the impact of casualty variables on support for war. The unit of analysis is a respondent s decision about this military action, and we are modelling the probability that he or she will opt to withdraw troops. We therefore have a minimum of one and maximum of ten observations from each respondent, depending on the point if any at which they advocated withdrawal from the military action. Before modelling those respondent-decision data, it is useful to give a flavour of individual-level responses by graphing the number of months over which respondents continued to support intervention or, in the terms of survival analysis, over which they survived. This is shown in Figure IV. 19

20 Figure IV: Number of months in which respondents advocated continued military action N of months supported intervention The mean duration of support for war is 5.36 of the 10 months but this central tendency conceals the predictable polarisation revealed in the graph: around three quarters of respondents either advocated withdrawal immediately (35%) or were still advocating continued action in the tenth and final month of the experiment (43%). This slightly constrains our scope for assessing the impact of conflict dynamics, given that only around one quarter of respondents actually changed their mind during the course of the campaign (although of course this also reflects the real world of public opinion on military action, in which many minds will be made up from the outset anyway). In any event, our overall sample size is sufficient to give us statistical power even dealing with a smallish proportion of the respondents. The Cox hazard models are reported in Table II. 20

21 Table II: Results of Cox hazard models estimating impact of manipulations on probability of opting to withdraw troops Variable Model 1 Model 2 Model 3 Model 4 National interests conflict.067 (.070).045 (.070).064 (.059).091 (.064) Civilian casualties (.167)** (.322) PMSC casualties (.149) (.149) Linear increase.277 (.087)** (.230)** (.186)** (.186)** Curvilinear increase.361 (.083)** (.211)** (.173)** (.173)** Opposition casualties -20%.053 (.106).049 (.107).048 (.107).050 (.107) Opposition casualties +20%.037 (.106).001 (.106).011 (.106).013 (.106) Opposition casualties +200% (.113) (.113) (.113) (.113) Graphs (.107) (.107) (.107) (.107) Monthly casualties (.002)*.077 (.003)**.071 (.004)*.071 (.003)** Linear increase * monthly casualties (.008)** (.006)** (.006)** Curvilinear increase * monthly casualties (.008)** (.006)** (.006)** Civilian * monthly casualties (.004)**.012 (.004)** PMSC * monthly casualties (.004).002 (.004) National interest * civilian (.180) Log Likelihood Likelihood Ratio Chi 2 N ** p<0.01, * p<0.05-7,245 30*** 8,457-6,727 1,065 8,457-9,705 1,376** 11,970-9,705 1,377*** 11,970 Our initial analysis approximately replicates Gartner s (2008) study by examining only military casualties. In subsequent models we include dummy variables for civilian and PMSC contractor deaths and use British military casualties as a baseline for comparison, so we postpone discussion of H1 for the moment. In Model 1 we include only the main effects of the casualty patterns manipulation and of the monthly casualties variable (that is, the number of military casualties that the respondent read about prior to that particular decision). The casualty patterns effects 21

22 are both significant and positive: those in the casualties increasing conditions were more likely to opt for withdrawal at any given point, giving support to H2(a). There is not much backing for H2(b), however, since the coefficient for curvilinear increasing is not as the standard errors confirm significantly larger than that for linear increasing casualty rates. But there is clear evidence that trends as well as numbers matter. Speaking of numbers, the negative and significant coefficient for monthly casualties i.e. that those who read about more casualties are less likely to favour withdrawal is highly counter-intuitive. It might, however, stem from specification error. In Model 2 we add interactions between the casualty patterns and the monthly casualties variable, reflecting Gartner s (2008) argument that absolute numbers will mean much less when there is a clear trend in casualty rates, and the results are now exactly as we would expect. The monthly casualties effect becomes strongly positive, the interactions are negative (since, in the context of an upward trend in casualties, there is a dampening of the impact of monthly numbers on support), and the main effects of those upward trends not only remain positive and significant but become far stronger bolstering support for H2(a) (but not H2(b), since now the linear increasing coefficient is even stronger than that for curvilinear increasing casualty rates). The other two results worth noting from Model 2 are null findings. First, contrary to H4, the graphical presentation of casualties has no effect on decision-making. This is despite the fact that, with only constant and increasing casualty rates, the graphs will never have been painting a positive picture and so would be expected overall to have boosted the probability of opting to withdraw troops. It appears that, at least under experimental conditions and (for most respondents) over just a few time-points, people were able to keep track of casualty trends in their minds in a way that rendered graphs unnecessary. However, it may be that the key test here is an interaction not yet specified between graphical presentation and the upward casualty trends. Second, perhaps even more surprisingly, enemy losses do not appear to weigh much if at all in the public s decision calculus. None of the opposition casualty manipulations had a 22

23 significant effect and, even if the consistently negative sign for the +200% condition does suggest that those large enemy losses persuaded a few respondents to stay the course, the effect is too small to adduce as support for H5. This is at odds with earlier results reported by Boettcher and Cobb (2006, 2009), a point we return to in concluding the paper. But it does maintain what is becoming a common theme of our results, namely support for a rationalist perspective when it comes to public judgements about military action. Presentational strategies and opposition casualty frames held little sway over those judgements; rather, respondents focused on a combination of hard numbers and the direction in which those numbers are going. Models 3 and 4 extend the previous analyses by examining the effect of casualty type on casualty aversion enabling us to test the remaining hypotheses. Civilian and PMSC casualty types are dummy variables that are compared against the reference category of military casualties. The first thing to note from the main effects is that the coefficient for PMSC casualties is nonsignificant. While surprising in some ways, and contrary to H1(a), these results are not just consistent with those reported by Avant and Sigelman (2010) but maintain that rationalist theme. If casualty rates are assessed with relative dispassion for what they mean about the progress of military action, it makes sense that PMSC death tolls would carry the same implications motivations and nationalities notwithstanding. Meanwhile, there is support for H1(b). The negative and significant coefficient for civilian casualties implies that, when it was foreign civilians whose lives were at stake, there was a lower probability of respondents opting to withdraw troops. What is not clear from this effect is whether it reflects the kicking in of the ingroup/outgroup reasoning that we also expected with PMSC casualties the idea that foreign civilians are associated with the enemy and thus regarded as less worth saving than those, whatever their origin, fighting for the UK or just a continuation of the rationalist calculus, and it is simply that civilian casualties are regarded as a less clear signal of success or failure. Some support for the first contention might be found in 23

24 the positive interaction between the civilian casualties dummy and the monthly casualty variable. The stronger relationship between numbers and the probability of withdrawal in the case of civilian casualties implies that, while smaller civilian losses are easily tolerated, support for war is eroded as numbers escalate. This echoes Friedrich and Dood s (2009) suggestion that civilian casualty insensitivity results from proportional numbing: only when the numbers increase to noticeable levels does the public begin to feel more uneasy about their deaths. Since the numbers were never very large in this experiment, however, they never become a large enough proportion of the civilian population for us to test that conjecture more directly. Finally, there is no clear support for H6(b) 7 We anticipated a significant negative interaction between the national interests conflict type and the civilian casualties dummy, reflecting the notion that civilian casualties would matter particularly little in the case of a national interests conflict (or, as H6(b) has it, matter particularly in the case of humanitarian intervention). The effect is at least in the expected direction and not minuscule in size, but the fit statistics in Table 2 show that including it does nothing to help model fit. Conclusion Casualty dynamics matter. Fluctuating conditions on the battlefield provide important information for the public to make rational assessments about the progress of a conflict. This paper sheds quite a lot of light on how, when and why these dynamics matter. Using a panel experimental design drawn from Gartner (2008), we first replicated in a British context (and with a more representative sample) his key finding that casualty numbers but also casualty context and the interaction between the two are crucial influences on individuals support for military action. Moreover, casualty figures on their own tend to have rather little effect it is the context provided by casualty trends that has the greater influence. 7 To test H6(a) would require an interaction between the conflict type and either or both the monthly casualty and the casualty pattern variables. This will follow in a later draft. 24

25 Strikingly, the effect of casualty numbers and trends was not dampened in any of the following circumstances: if there was accompanying information about greater enemy losses; if trends were concealed by the absence of graphical presentation; and if the casualties were among PMSC personnel rather than the British armed forces. All three of these null results is somewhat unexpected. Boettcher and Cobb (2006) provide evidence from a different experiment in which the effect of casualties on US public support for war could be offset by the use of kill ratios. It is not clear why the British public would appear to have a steely-eyed determination to focus only on their own casualties and not to set them against opposition losses. One possibility is that these enemy deaths have different and cancelling effects among different respondents: some regard them as essentially good news about the progress of the war; others, valuing lives more equally across the two sides, regard them as further human costs of war. We had suspicions that Gartner s (2008) results were more spectacular than they would have been had he not cued casualty trends so strongly by graphing them for respondents. Those suspicions are largely allayed by these results in a way that speaks well of the cognitive capacity and/or commitment of our respondents: they were able to perceive and act on trends without illustrative assistance. It may be that, over a more protracted conflict (of the kind familiar following Iraq and Afghanistan), graphs become more useful in highlighting casualty trends. There remains an external validity question about whether such graphs reflect the way in which casualty information is conveyed to the public in the real media world. The equivalent reactions to military and to PMSC casualties make for perhaps the most striking finding. A key reason why democratic leaders outsource military operations is that, because the public is assumed to be more tolerant of PMSC casualties, this gives them greater leeway in how they conduct war. Our results do not support that underlying assumption about differentials in casualty aversion. Further research could usefully investigate whether this null finding, now obtained in the UK and the US (Avant and Sigelman, 2010), replicates elsewhere. 25

26 Meanwhile, there is again an external validity question. Our experiment brought PMSC casualties to respondents attention just as clearly as it did military casualties. If, in reality, PMSC deaths can be concealed, then the potential negative impact of those casualties on public opinion will not be realised and need not constrain governments. These findings are uniformly consistent with a rationalist approach to casualties, in which the public uses numbers and trends to estimate not only the current and future costs of military action but also the prospects of success. Future data collections along these lines should include post-experimental checks, asking specifically about costs and perceived success, in a bid to separate these two mechanisms whereby casualty rates will affect support for war. For the moment, it is fair to say that our results augur well for the public s ability to hold leaders to account for military casualties and the progress of international conflicts. Our final contribution was to provide the first systematic representative-sample comparison of military and civilian casualties in terms of their impact on public judgements. What is clear is that, as expected, military casualties had a stronger effect on mass opinion. What is less clear is whether this reflected a strategic discrimination, with military casualties providing a stronger signal about the failure of operations, the kind of proportional numbing suggested in one previous analysis of civilian casualty aversion (Friedrich and Dood, 2009), or a more straightforward national chauvinism whereby foreign civilian lives were simply valued less highly than British military lives. The kind of mechanism checks mentioned above would offer one way of disentangling those possibilities. Another is to examine the individual-level moderators of casualty discrimination. If military casualties weigh disproportionately heavily in the calculations of those scoring higher on knowledge of international affairs, then that would support a more strategic interpretation. If, as seems more likely, military casualties weigh disproportionately heavily in the calculations of those scoring higher on nationalism or ethnocentrism, then this would look much less strategic. 26