It’s only money

Abstract

The sunk costs fallacy is an important concept in the academic and policy worlds. It has helped explain consequential national security decisions such as the escalation in Vietnam and the surges in Iraq and Afghanistan. While previous analysis of sunk costs in international relations has made no distinction between financial and human sunk costs, there is evidence in psychology that people treat human lives and financial costs differently. The consensus in the casualty sensitivity literature is that human sunk costs should lower support for a conflict, but there is as yet no evidence on whether financial costs operate in the same way. Using the Environmental Protection Agency’s value of a statistical life to equalize human and financial costs, we create survey experiments through Mechanical Turk and GfK/Knowledge Networks about a hypothetical US military intervention to test if financial and human costs have the same effects on public opinion. We find that public reaction to sunk costs is contingent on the type of costs incurred. Consistent with the growing ‘sunk costs skeptic’ literature we find no evidence that any sunk costs induce greater commitment to a mission. Where the US contribution to the conflict is purely financial, sunk costs induce a desire to cut losses. When intervention involves US lives, sunk costs make no difference to the level of support. Finally, contrary to the implicit assumptions of past policymakers, ex ante levels of public support for sending troops and sending money are indistinguishable. These findings hold true both in situations involving high Iraq War level sunk costs and low Somalia-style costs.

Keywords

behavioral economics foreign policy public opinion

Introduction

The sunk costs fallacy is one of the most influential postulates of behavioral economics (Kahneman & Tversky, 1979). When individuals persist with a project to ‘recoup’ costs which they have already sunk into it even when superior alternatives exist, this is termed the ‘sunk costs fallacy’ (Kahneman, 2011). The concept has been adapted to account for many strategic decisions (Taliaferro, 2004; Jervis, 1976). For instance, Stephen Walt blamed the sunk costs fallacy in part for the Obama administration’s decision to order the ‘surge’ in Afghanistan. ‘If the situation in Afghanistan was exactly the same, but the United States was not already present,’ Walt demanded, ‘would President Obama be getting ready to send 100,000+ troops there?’ (Walt, 2010).

But can a theory designed to explain financial decisions also explain foreign policy choices? Intuitively, it is plausible that individuals treat ‘human sunk costs’ (i.e. soldiers killed in action) in a different manner to financial sunk costs. This intuition is supported by the ‘sacred values protection model’ in political psychology (Tetlock et al., 2000).

The sunk costs fallacy and the sacred values protection model together suggest human and financial sunk costs may be treated differently. Policy practice also appears to support this notion since policymakers often regard sending troops to war as a qualitatively different undertaking to sending materiel. If so, then analysts must take care in applying sunk costs concepts from economics to life and death foreign policy decisions.

This is especially relevant as the Obama administration has shifted to a strategy of combating security threats with US money rather than US lives. The Administration may be influenced in this by the belief that the US public are more likely to support spending treasure rather than blood ex ante, and by the belief that it will be easier for the United States to extricate itself ex post from conflicts over which no blood has been spilt. Our article will aim to shed light on US public support for different policy tools.

If people find it hard to trade off lives and money, then sunk human and sunk financial costs may have opposite effects on public support for US involvement. Most of the casualty sensitivity literature has suggested that human sunk costs should ceteris paribus reduce public support for a conflict (Mueller, 1973; Jentleson, 1992; Gelpi, Feaver & Reifler, 2009; Gartner, 2008), although this literature makes no argument as yet on the effects of financial costs. However, if human lives possess a ‘sacred’ significance, then blood may generate an even larger sunk costs effect than treasure.

This matters for two reasons. First, suppose that policymakers pay a lower penalty for treating financial costs as sunk than they do for human costs. If so, then policymakers should use financial means to intervene in a wider range of conflicts. If the national interest is less clear, then policymakers will choose to send money first rather than personnel precisely because they can more easily withdraw if things go wrong. Second, however, foreign adversaries will place less credibility on the United States’ commitment to these conflicts if this commitment is only financial, precisely because it is easier to disengage from ex post.

By contrast, the deployment of ground troops to a region would be a more credible signal of US commitment, as US policymakers would suffer heavier audience costs if US soldiers were seen to have died in vain. Thomas Schelling made this point about the US garrison in West Berlin, arguing that if US lives were lost there it would solidify US commitment to the defense of Western Europe. As Schelling (2008: 47) said: ‘What can 7,000 American troops do, or 12,000 Allied troops? Bluntly, they can die. They can die heroically, dramatically and in a manner which guarantees that the action cannot stop there.’ Presumably Schelling would not have believed that money would have had the same effect. As many states in the Asia-Pacific region, restive about China’s rise, request a greater demonstration of US commitment to their security, the tripwire strategy may return to fashion. Therefore, it is important to investigate its underlying assumptions about public opinion.

This article disaggregates human and financial sunk costs to determine whether they have a different impact on public opinion. Following a growing body of research, we conduct three online survey experiments, two through Mechanical Turk (MTurk) and one through GfK/Knowledge Networks’ probability-based online panel KNOmni. Respondents were asked to express their opinion on a hypothetical US intervention overseas involving one of the four randomly assigned treatments – human prospective costs, human sunk costs, financial prospective costs, and financial sunk costs. We use the Environmental Protection Agency’s value of a statistical life calculation to equalize the costs across treatments and replicated our experiment with both lower and higher levels of sunk costs. Our key finding is that financial and human costs have different effects on public opinion, but neither in the direction the sunk costs fallacy would expect. Financial sunk costs lower support for conflict, while human sunk costs make no difference.

This article proceeds as follows. First we recap the literature on sunk costs, public opinion and foreign policy, and the sacred values protection model. Then we outline our hypotheses and research design before proceeding to our results. We conclude with our interpretation of the results and their policy implications.

Sunk costs and public opinion in international conflict

Public opinion and foreign policy

The earliest studies on US public opinion and foreign policy argued US voters rarely made rational decisions about national security. The ‘Almond-Lippmann consensus’ posited that US voters were too pacific in peacetime, and too belligerent during wartime (Holsti, 1992). While the Almond-Lippmann consensus did not explicitly address sunk costs, it stressed the public’s suboptimal rationality.

Much recent work on the issue emerged as a reaction to that consensus. John Mueller’s work, for instance, suggested that US voters’ attitudes were predictable and consistent. As casualties mounted, Mueller claimed, support for a given mission would fall logarithmically – the first casualties triggering big falls, later casualties triggering smaller falls (Mueller, 1973). Gartner & Segura (1998) agree that casualties are key, but argue that voters respond to marginal casualties rather than cumulative casualties.

Others argue the impact of casualties is contingent. According to Bruce Jentleson (1992), voters are more tolerant of casualties in some missions (e.g. restraining aggression) than in others (e.g. regime change). Similarly, Gelpi, Feaver, and Reifler claim that voters have a rational attitude towards human costs, viewing them as a price to be paid in order to secure a foreign policy ‘benefit’. The mission’s prospects for success are therefore the key conditioning variable (Gelpi, Feaver & Reifler, 2009). Gartner’s (2008) survey experiments on public support for conflict likewise find support for a ‘rational public’ view, though his conclusions differ from those of Gelpi, Feaver, and Reifler in that Gartner believes casualties to be important because they provide voters with information on the likely outcome and costs of the war. Kull & Ramsay (2001) posit that public support for a conflict will persist in the face of casualties if the mission enjoys widespread elite support. Grieco et al. (2011) extend this notion of elite consensus to the international realm, noting that public support for the war may hold up in spite of casualties if the United States retains the support of key foreign allies.

The majority view in the casualty sensitivity literature is that voters are forward-looking, as expected utility theory would suggest. Human sunk costs are expected to lower the public’s tolerance for US intervention overseas, even if this relationship can be mediated by other factors. This contention is supported by the trajectory of US public opinion on wars from Korea to Afghanistan, where increased casualties correlated with lower support.

There are, however, exceptions to this consensus. Berinsky (2007) believes that partisan cues, not casualties, are the key driver of public opinion towards a conflict. For Berinsky, casualties only appear to lower public support for the war because they move elites to turn against it, and the elites’ cues then shift their supporters’ opinions. Berinsky (2007) carried out a survey experiment showing that correcting individuals’ pre-existing view of the level of US casualties in Iraq had no discernible influence on their support for the war.

Boettcher & Cobb’s (2009) findings suggest that framing the decision to continue a conflict in terms of previous casualties increases support for the mission among some sections of the population. Individuals who believe the Iraq War was the right decision were more likely to support continued US involvement there when primed with information on the number of Americans who had died there. However, this same information had the opposite effect on those who believed the war had been the wrong decision.

Patricia Sullivan (2008) also dissents from the casualty sensitivity consensus by suggesting that democratic publics may be more likely to support a military operation as costs increase. She finds public support is more likely to endure when the intervention employs ground troops in addition to other assets (Sullivan, 2008). There are good reasons why this is plausible. Historical policy practice and theoretical work in cognitive psychology – the sacred values protection model and prospect theory – suggest that human and financial costs will be treated differently by voters, as will prospective and sunk costs.

Lives versus money: The sacred values protection model

Historically, a distinction has existed between sending money and sending troops to a conflict. Financial aid to allies has been seen as less controversial than boots on the ground. When possible, leaders often prefer to pay foreigners to fight their enemies rather than risk their own soldiers’ lives. To send ‘our boys’ is seen as a step change in commitment to the conflict, justified only if other options have failed.

In the Cold War, the United States hoped financial support would allow European nations to defend themselves against the Soviets without a large US manpower commitment (Ferguson, 2004). The historical literature cites different public attitudes towards the commitment of US lives and money as the key driver of the strategy. For instance, in the early years of World War II, opinion polls showed that while 80% of Americans supported sending financial aid to the Allies, 82% were opposed to actually fighting the war. Consequently, Roosevelt’s rhetoric stressed the distinction between blood and treasure. In the presidential campaign of 1940, for instance, he stated that ‘your boys are not going to be sent into any foreign wars’, and later claimed that ‘we must get these weapons to them in sufficient volume and quickly enough so that we and our children will be spared the agony and suffering of war’ (Divine, 1976: 105).

Once blood has been split, however, policymakers’ rhetoric has often claimed that the war must be won in order to redeem the sacrifice of the fallen. For instance, in justifying his decision to ‘stay the course’ in Iraq, President Bush stated that: ‘I’m going to make you this promise. I’m not going to allow the sacrifice of 2,527 troops who have died in Iraq to be in vain because we pulled out before the job was done’ (Schwartz, 2005). Bush’s viewpoint was reinforced by President Clinton, who claimed that ‘we’ve all got a stake in it (Iraq) succeeding’ because ‘We’ve got over 600 dead Americans since the conflict was ended.’ By contrast, consider how much the Gettysburg address would resonate if Abraham Lincoln had proclaimed: ‘we here highly resolve that this money shall not have been spent in vain’ (Schwartz, 2006).

The ‘sacred values protection model’ in psychology suggests why. According to this, lives and money are not viewed as fungible. Instead most people find it transgressive to even think of human lives in dollar terms. Tetlock et al. (2000) produced a number of experiments where individuals were faced with a trade-off between human life and money. In one experiment subjects were presented with the hypothetical case of a hospital administrator who had to weigh a child’s life against $1 million which could be spent elsewhere in the hospital. Not only did they express moral outrage if the administrator chose to save the $1 million rather than the child, they also did so if the administrator chose to save the child but had to even spend time reflecting on it first. Tetlock and his research team have replicated and extended their findings in a number of articles (Tetlock, 2003; McGraw & Tetlock, 2005; Tetlock, McGraw & Kristel, 2004), while the model has attracted attention from evolutionary psychologists (Pinker, 2007).

The extension of the sacred values protection model most closely related to the current survey is Fagley & Miller’s (1997) experiment, where respondents react differently to the framing of situations depending on the ‘outcome arena’. Individuals in their experiment were more risk acceptant in problems where the result of a decision was to save human lives as opposed to money. This experiment suggests caution is required in assuming individuals will be able to trade off human and financial costs, although Fagley and Miller made no attempt to equalize the value of the outcome across the different arenas.

The sacred values protection model suggests human and financial costs will be treated differently. Meanwhile, prospect theory suggests those same costs will have different effects depending on whether they are sunk or future costs. We argue that the intersection of these two theories can provide fruitful insights into public support for different foreign policy tools.

Prospect theory and sunk costs fallacy

The sunk costs fallacy is best known to political scientists and economists through its connection to prospect theory (Kahneman & Tversky, 1979), which suggested that individuals base their decisions on losses and gains relative to some reference point. If the reference point for individuals is the beginning of a course of action, prospect theory suggests that individuals try to maximize the improvement in their welfare relative to that time, rather than their welfare from the present. The use of sunk costs in international relations dates back to Robert Jervis (1976). Taliaferro (2004), for instance, applied the sunk costs fallacy to great power intervention in the periphery. Prospect theory has been suggested by a number of analysts as a potential micro-foundation for theories of international relations (Levy, 1997; McDermott, 2004; Stein, 1992; Berejikian, 2002). It has been used to build theories of compellence and deterrence (Schaub, 2004), international regime design and compliance (Stein 1992), and crisis bargaining more generally (Berejikian, 2002). Prospect theory has also been used to help build statistical tests of realism and deterrence theory (Huth, Gelpi & Bennett, 1992) and to explain state behavior from Soviet relations with Syria in the Six Day War (McInerney, 1992) to British intervention in Egypt during the Suez crisis in 1956 (Richardson, 1991). Furthermore, there is research suggesting that sunk costs can be applied to the public. Berejikian (2002: 782) wrote that ‘there is increasing evidence that prospect theory contours the topography of public opinion concerning national policy decisions […] Support for a government policy depends crucially on the framing adopted in the play of competing public discourse […] (T)his extends to international policy choices.’

Prospect theory holds that individuals become more risk acceptant when they perceive themselves to be in the domain of ‘loss’ rather than gain (Kahneman & Tversky, 1979). This is said to explain why individuals will escalate their commitment to an apparently failing project if they have already devoted resources to it – a phenomenon termed the ‘sunk costs fallacy’. Kahnemann illustrates the sunk costs fallacy with the following scenario. Suppose a company has already spent $10 million on Project A when it discovers that it will need to spend an additional $6 million in order to complete it. At the same time, however, there is a new project, Project B, which would also cost $6 million but which would offer a higher return. The CEO chooses to spend the money on Project A in order to ‘recoup’ the $10 million already spent, mistakenly passing up the higher return from Project B even though successful completion of Project A will not bring the $10 million back (Kahneman, 2011).

Support for the sunk costs fallacy comes from a number of additional experimental and observational studies, although the latter are open to problems of endogeneity and selection bias (Friedman et al., 2007; Arkes & Blumer, 1985). However, recent years have seen a backlash against the sunk costs fallacy in economics. McAfee, Mialon & Mialon (2010) note that there are many reasons why rational, forward-looking decisionmakers would choose not to ignore sunk costs. For instance, if success in an endeavor is the product of a ‘cumulative process’, then the prospects of success will be a positive function of sunk costs. Taking this theoretical work into account, (Friedman et al., 2007) conducted a number of laboratory experiments into the sunk costs fallacy, revealing, in their words, ‘a surprisingly small sunk costs effect, one which is surprisingly insensitive to the proposed psychological drivers’. Finally, using a regression discontinuity design to identify the causal effect of sunk costs in playing-time decisions in basketball, Leeds, Leeds & Motomura’s (2015) work ‘strongly support(s) the neo-classical (i.e. rational, forward-looking) outlook’.

The ‘sunk costs skeptic’ literature in economics has still not influenced international relations, however, where the sunk costs fallacy remains a popular explanation for many decisions. Moreover, until now the international relations literature on sunk costs has treated human and financial sunk costs almost interchangeably. Both Jervis and Taliaffero made no attempt to distinguish human and financial sunk costs (Jervis, 1976; Taliaferro, 2004). Sullivan’s work is the closest the international relations literature has come to disaggregating human and financial sunk costs; however Sullivan does not compare military interventions in which blood is spilt to those where it is not (Sullivan, 2008). Similarly, Berinsky’s (2007) survey experiment primed respondents with separate information about the human and the financial costs of the war in Iraq, finding there to be no significant difference between the different treatment groups. The Iraq War, however, involved both human and financial costs and so is not a clean comparison of the separate effects of the two values on support for the war. Boettcher & Cobb’s (2009) survey experiment similarly did not compare Iraq with another conflict which involved the loss of money but no US lives.

In short, there is evidence that individuals treat human life and money in different ways, because the former is viewed as having ‘infinite and transcendental significance’ (Tetlock et al., 2000) whereas the latter is not. Consequently they have difficulty trading them off against one another (Tetlock et al., 2000). From this, it follows that voters should be more likely to support spending treasure rather than blood on foreign conflicts ex ante.

Ex post, however, a different dynamic could be at play. Richard Ned Lebow (1981) claims that in crises with high stakes, the emotional stress experienced by policymakers may produce a ‘cognitive overload’ which diminishes their ability to make decisions in accordance with procedural rationality. Similarly, the sacred values protection model suggests that ‘sacred’ considerations such as human lives impede voters’ ability to apply rational, forward-looking, cost–benefit analysis to political decisionmaking. Therefore, it could follow that if a political decision is framed in terms of human lives, voters will be more likely to resort to cognitive shortcuts such as the sunk costs fallacy than when the same decision is framed in terms of money. They will therefore be more risk acceptant in the face of human rather than financial losses, so that lives lost will solidify support for a conflict while money spent should not.

We term this combination of the sacred values protection model with the sunk costs fallacy the ‘don’t let them die in vain’ hypothesis. This hypothesis states that the loss of soldiers’ lives will generate a greater desire on the part of the public to persist with a conflict than the loss of money. Note that this hypothesis stands in contrast to the majority view of the casualty sensitivity literature which would predict that, ceteris paribus, loss of lives should reduce support for the war.

Theory and hypotheses

Combining the sacred values protection model with prospect theory, we derive hypotheses about the public’s support for military involvement abroad. First, following the sunk cost literature, we expect previous costs to increase public support for military intervention. If people are more willing to engage in actions in an attempt to recoup losses from previous decisions, we expect increased support in places where the USA has already expended resources:

Sunk cost hypothesis (H1): Sunk costs (both human and financial) will increase support for intervention.

However, not all costs are created equal. Following the ‘sacred values protection model’ we argue voters will treat sunk costs of the same magnitude¹ differently depending on whether those costs involve the loss of human life or the loss of money. Therefore, we expect voters to treat the decision to send money and troops differently:

Sacred values protection hypothesis (H2): Respondents will treat human and financial sunk costs differently.

Combining prospect theory and the sacred values protection model, we expect respondents to react differently to prospective costs and sunk costs. Prospectively, since the sacred values protection model suggests that people are more averse to losing human lives, we expect that citizens will be less likely to support sending troops than sending money ex ante.

Send money first (H3): Respondents will be more willing to send money than troops to help an ally, ex ante.

Retrospectively, however, we argue that the sacred values protection model exacerbates the sunk cost fallacy, which will stiffen voters’ resolve to stick with an intervention.

‘Don’t let them die in vain’ ² (H4): Human sunk costs will have a larger positive influence on support for intervention than financial sunk costs.

Research design

In order to assess the public’s attitudes towards human and financial sunk costs, we use a between-group experimental design where respondents are randomly assigned into one of four groups: two prospective cost groups (human and financial) and two sunk cost groups (human and financial). We ran two survey experiments using MTurk – one with low sunk costs, and one with high sunk costs. However, since subjects select into MTurk’s subject pool, this carries potential drawbacks. Berinsky, Huber & Lenz (2012) show that MTurk’s population is systematically unrepresentative of the US population. We found this to be true in our sample: over 75% of respondents were under the age of 35 and over 50% identified as liberal.³ This raises the additional problem of selection bias – MTurk workers may be systematically different from the general population in important and unmeasured ways (Kahan, 2013). To address these sampling concerns, we reran the low sunk cost experiment using Knowledge Networks’ probability-based sample of the US adult population. KNOmni’s panel is far more representative of the US population, but all three surveys produced similar results.

We compared support for increased military involvement after the USA has already spent money/lost troops to support military involvement where the USA has not been involved. If the sunk cost hypothesis is true, we expect increased public support for military involvement from subjects that are told the USA has already been involved and lost lives/money.

Each participant received a question about the United States’ hypothetical involvement in an unnamed foreign country, and was asked to report their level of support for US military involvement. In all the treatments, the decks were stacked in favor of the respondent supporting the intervention – along with the use of the emotive Al-Qaeda descriptive, the allied president was described as trustworthy, democratically elected, and a friend of the United States. The ‘decisionmaker’ was described as ‘the United States’ not the ‘President’ in the hope of isolating the effects of the treatment from those of partisan cues. Moreover, respondents were told that the USA was responding to the foreign leader’s request for help in all cases and were asked if they approved – also increasing the chances of support.

The aim of loading the decks was simple. After eight years of war, US public support for further overseas engagements is low. A potential danger facing the survey would consequently be zero variance – that is, respondents opposing any US intervention at all. This would make answering the central question impossible. At the same time, however, biasing the responses in favor of intervention in all groups would not bias our causal inferences, which are based on differences between samples.

Subjects were assigned into one of four treatment groups (T1, T2, T3, T4); see Tables I and II. Participants who received the ‘prospective only cost’ treatment (T1 and T2) saw the question:

A country faces an insurgency from al-Qaeda backed Islamic militants. The leader of this country requests US (troops/weapons) to fight alongside his forces against the militants. The leader is democratically elected, trustworthy and a genuine friend of the United States. Independent military experts suggest that there is an approximately fifty-fifty chance that the insurgency will be defeated if the United States sends troops. Fearing the country may become a base for terrorists if the insurgency succeeds, the United States agrees to the request to send (troops/weapons) to the country. On a scale of 1–10, with 1 meaning that you strongly disapprove, and 10 meaning that you strongly approve, what is your opinion of the US decision?

Table I.

Low cost experiments (MTurk & Knowledge Networks)

	Costs
	Prospective only	Low sunk
Human Financial	T1 T4	T3 T2

Table II.

High cost experiment (MTurk)

	Costs
	Prospective only	High sunk
Human Financial	T1 T3	T2 T4

Respondents who received the ‘sunk cost’ treatment saw the same question, except their prompt states: ‘the United States sent [X weapons/troops] to the country’ before the leader requests additional troops or money. In the human sunk costs treatment, the help came in the form of US ground troops, whereas in the financial treatment, the help ‘only’ cost money.

Using the Environmental Protection Agency’s (EPA) formula for calculating the value of a human life, we equalize the monetary value of the sunk costs. According to the EPA’s standard calculation, a statistical life is currently valued at $7.4 million (US Environmental Protection Agency, 2010). For the low sunk cost experiment, the number of troops lost in the conflict was said to be 14, and for the high sunk cost experiment the number of troops lost was said to be 4,500. In the low sunk cost experiment the 14 lives lost equaled $103.6 million for weapons. This is a realistic figure for US expenditure on foreign military aid. For instance, US military aid to the Philippines in 2012 increased to $30 million from the previous year’s total of $15 million.

As an alternative to using the EPA’s value of a statistical life calculation, we considered asking respondents what dollar value they placed on a human life and then using this figure as the basis for our subsequent comparison. However, there were two problems with this approach. First, Tetlock’s work suggests most people not only view human lives and money as being incommensurable, but even the suggestion that they could be traded off against one another to be morally transgressive. Therefore, asking respondents how much money a human life is worth would yield a low response rate. This in turn would create response bias. Individuals who do view human lives and money as incommensurable would not give a dollar value in the first place and so would be excluded from the study. Given the aim of the study this would be problematic. The EPA’s value of a statistical life calculation has the benefit of being based on how people actually choose to value their own life in the real world when they have to trade off, for instance, a safer job with lower compensation against a riskier but better paid one (Sunstein, 2005).

In all treatments the prospective benefit of military action was to prevent an Al-Qaeda terrorist base. The prospects for success of the mission had to be treated carefully. From Gelpi, Feaver & Reifler (2009), we know that the prospects for success are key in determining public support for a war. If the prospects for success were set too low, then we would incur the risk that no respondents would support any intervention, so there would be no variance to examine. If the prospects for success were not mentioned at all, then given the experiences of the United States in Iraq and Afghanistan, respondents would most likely assume the prospects for success to be low. In that case, we could have ended up with the same no variance problem as if we had explicitly stated the probability of victory to be low. If, by contrast, the prospects for success were set too high, then the situation would no longer clearly be one in which the sunk costs fallacy is applicable. The sunk costs fallacy is most powerful in explaining commitment to an apparently failing project, not an apparently successful one. The latter phenomenon is easily explicable in expected utility terms. Consequently, we stated that ‘independent military experts’ estimated the chances of success to be fifty-fifty – high enough to generate at least some support for intervention across all conditions, but not so high as to raise the question of whether increased support for an intervention with sunk costs would constitute genuinely ‘fallacious’ behavior.

Hypothesis testing

We do find support for H2: citizens do treat the loss of human lives differently than the loss of money, but it is not in the way the sunk costs fallacy would lead us to expect. We can see this from the group means presented in Table III. We argued that sunk costs would have a

Table III.

Unconditional means for experiments

		Type of cost treatments
		Human		Financial
Survey sample	Cost treatment	Prospective	Sunk	Prospective	Sunk
Knowledge Networks	Low sunk cost	6.14	6.16	6.12	5.53
MTurk	Low sunk cost	6.23	5.93	6.59	6.09
MTurk	High sunk cost	6.26	5.94	6.73	5.76

Table IV.

Human and financial sunk cost v. prospective costs – OLS

	Knowledge Networks		MTurk
	Knowledge Networks		Low cost experiment		High cost experiment
	(1)	(2)	(3)	(4)	(5)	(6)
Human sunk costs	0.016		–0.316		–0.308
	(0.069)		(–1.474)		(–1.435)
Financial sunk costs		–0.591*		–0.569**		–0.905***
		(–2.421)		(–2.702)		(–4.208)
(Intercept)	6.141***	6.122***	6.248***	6.660***	6.248***	6.660***
	(38.415)	(35.490)	(50.171)	(54.440)	(50.865)	(54.462)
No. of subjects	510	508	566	555	557	544
R²	0.000	0.011	0.004	0.013	0.004	0.032
F	0.005	5.860	2.174	7.299	2.060	17.703
AIC	2,407.266	2,473.137	2,606.712	2,525.973	2,550.029	2,475.556
BIC	2,419.969	2,485.829	2,619.728	2,538.930	2,562.996	2,488.453

Z values in parentheses; ^† p < 0.10, *p < 0.05, **p < 0.01, ***p < 0.001.

positive effect on support for US intervention, but we find no evidence of this hypothesis from the experiment. Table III also shows that H1 – sunk costs will engender increased military support – is not supported by the data. In no experiment are sunk costs, either human or financial, higher than their prospective group counterpart. This also provides contrary evidence for H4, since human sunk costs are associated with no change in support for the mission. If the ‘don’t let them die in vain’ hypothesis were true, then we would find an increase in support when subjects were informed of troop losses.

While we find no evidence for H1 and H4, there is some supporting evidence for H2: subjects treat financial and human sunk costs differently, but not in the way we predicted. Human sunk costs show no change in support while financial sunk costs decrease support, suggesting subjects treat the two differently. We can see this in Tables IV and V, where Financial sunk costs are negative and significant, compared to Financial prospective costs, across all three experiments.⁴ In order to test our hypotheses, we used OLS and logit models. In the case of the logit regressions, we assign responses 5 to 10 from the military support category as a 1, and 1 to 4 responses as 0, allowing us to see if the groups are more likely to approve or disapprove of the war effort. While the OLS estimates are difficult to substantively interpret, the logit analysis provides a more intuitive interpretation: subjects are more likely to disapprove of intervention when money has been lost.

In order to visualize the effect of Financial sunk costs, we graph post-estimated simulations of simple binary regressions with just a constant and a dummy variable for the Financial sunk cost treatment group. Figures 1 and 2

Table V.

Human and financial sunk cost v. prospective costs – logit

	Knowledge Networks		MTurk
	Knowledge Networks		Low cost experiment		High cost experiment
	(1)	(2)	(3)	(4)	(5)	(6)
Human sunk costs	–0.072		–0.258		–0.301
	(–0.329)		(–1.295)		(–1.496)
Financial sunk costs		–0.509*		–0.503*		–0.696***
		(–2.500)		(–2.373)		(–3.299)
(Intercept)	1.386***	1.286***	1.138***	1.484***	1.138***	1.484***
	(8.855)	(8.442)	(9.448)	(11.051)	(9.448)	(11.051)
No. of subjects	510	508	566	555	557	544
Log-likelihood	–259.255	–290.957	–323.353	–285.686	–319.314	–285.424
AIC	522.511	585.914	650.706	575.373	642.629	574.849
BIC	530.980	594.375	659.383	584.011	651.274	583.447

Z values in parentheses; ^† p < 0.10, *p < 0.05, **p < 0.01, ***p < 0.001.

show the result of these simulations for each of the experiments. As we can see, the effect of the Financial sunk costs is non-trivial. Across the three experiments the Financial sunk costs group is associated with a 0.6 decrease in mean support, and a roughly 10% decrease in ‘high’ support for military involvement.

Furthermore Table VI shows evidence against H3. The financial prospective cost group (T1) is not statistically different from the human prospective cost group (T2). The lone exception is the logit analysis of the high sunk costs group from the MTurk survey. Therefore we conclude that subjects are insensitive to the type of support being given to the country.

Our evidence reveals a surprising degree of insensitivity to the level of human costs, given that some have occurred. We see this in two ways. First we see a nearly identical level of support between subjects sending in troops and having lost troops. Second, in the MTurk survey we find respondents are equally likely to support continuing the fight whether after losses from minor skirmishes (14) or Iraq War sized losses (4,500).

Interpretation

What can we learn from these results? First, there is no evidence the sunk costs fallacy influences US public opinion on overseas intervention. No sunk costs treatments increased support for intervention. Second, we find evidence against the consensus view in the casualty sensitivity literature. Casualties had no effect on support in our experiments. Third, these results are consistent across the levels of sunk costs, even though there are over 300 times as many troops lost and money spent in the high sunk costs experiment than the low sunk costs experiment. This may suggest voters lack a clear anchor in terms of the expected costs in military operations (Berinsky, 2007). Finally, spending money on military aid to allies is not ex ante more popular than sending ground troops.

Our results are consistent with a number of schools in the literature on public opinion and foreign policy and on the sunk costs fallacy. First, null effect of casualties is consistent with Berinsky’s work. Information on casualties was presented to our subjects without any partisan cues, and since casualties did not have any effect on support under these conditions, this provides partial confirmation for Berinsky’s theories. Similarly, the insignificance of casualties when holding prospects for success constant suggests that, as Gartner suggested, casualties may be important because they provide voters with information about the likely outcome of the conflict. However, our results are not fully consistent with Gartner’s theory. Unlike us, Gartner found casualties to have an effect net of the information they provide on the prospects for success. Second, he suggested that casualties may provide information on the likely success and future costs of the war. As we found casualties to have no effect even though we did not hold future costs constant, Gartner’s theory cannot fully account for our findings.

Second, our findings provide support for the ‘sunk-costs skeptic’ literature in economics. Sunk costs, when isolated from other factors such as the prospects for success, do not produce a desire to double down. Human sunk costs make no difference while financial sunk costs generate demands for withdrawal. It is possible that sunk costs exert opposite effects upon individuals with

Figure 1.

Estimated effect of financial sunk costs on military support – OLS

different prior views about the wisdom of the war itself, so that the net effect is zero. This would be consistent with Boettcher and Cobb’s findings. However, we found that self-identified conservatives, whom one might imagine to be more likely to favor the conflict in the first place, displayed the same null effect of sunk costs as the other respondents. Perhaps the sunk costs fallacy is a better description of elites’ behavior. Political leaders might find an appeal to sunk costs compelling (especially if they are the ones who sank the costs to begin with). However, it does suggest that invoking sunk costs to rally support for a conflict is not a winning rhetorical strategy, at least in the aggregate. Moreover, our findings do lend some support to a ‘roughly rational’ view of public opinion.

Third, we find some support for the sacred values protection model. Lives and money are treated differently by voters. Voters want to walk away from money which has been spent. The same is not true of lost lives.

Figure 2.

Estimated effect of financial sunk costs on military support – logit

However, subjects were not more willing to use money rather than troops. This is a surprising conclusion given the USA’s historical record of deploying troops to conflicts only after monetary assistance to allies has failed and its preference for capital intensive tactics which minimize the risk to US personnel. There are two potential explanations – information and agency concerns.

First, it could be that the US decision to send troops reveals additional information to subjects about the importance of the conflict to US national interests. This is unlikely, however. The stakes of the conflict in our scenario – preventing the emergence of an Al-Qaeda safe haven – are clearly stated. Little more could be communicated by the US decision to send troops. Second, it could be that subjects are not ‘buying’ the stated premise that the foreign ally can be trusted with the aid. Dangers of blowback might exist even if the ally is trustworthy – the allies may be defeated and the weaponry seized from

Table VI.

Prospective costs – human vs. financial treatments

	Knowledge Networks		MTurk
	Knowledge Networks		Low cost experiment		High cost experiment
	(1) – OLS	(2) – Logit	(3) – OLS	(4) – Logit	(5) – OLS	(6) – Logit
Human prospective costs	0.019	0.100	–0.360	–0.168	–0.472	–0.567*
	(0.081)	(0.459)	(–1.501)	(–0.691)	(–1.929)	(–2.094)
(Intercept)	6.122***	1.286***	6.595***	1.323***	6.734***	1.688***
	(36.615)	(8.442)	(38.975)	(7.530)	(38.402)	(8.057)
No. of subjects	509	509	387	387	356	356
R2	0.000		0.006		0.010
F	0.007		2.252		3.721
Likelihood-ratio		0.211		0.478		4.493
AIC	2,446.225	524.629	1,767.786	416.006	1,609.304	358.001
BIC	2,458.922	533.094	1,779.661	423.922	1,620.929	365.751

Z values in parentheses; ^† p < 0.10, *p < 0.05, **p < 0.01, ***p < 0.001.

them and used against Americans. This possibility constitutes an interesting avenue for further research.

Conclusion

Foreign policy has been a key issue in US politics since 9/11. Iraq arguably won George W Bush re-election in 2004 (Gelpi, Reifler & Feaver, 2007) but cost him Congress in 2006 (Eichenberg, 2009). As Richard Eichenberg (2009) argues, foreign policy similarly allowed Barack Obama to win the Democratic nomination in 2008 and remains a crucial part of his electoral appeal. Public opinion will therefore continue to play a vital role in shaping the decisions which the United States makes in dealing with challenges such as ISIS or the rise of China. In this article we examined the US public’s attitudes towards financial and human tools of statecraft and to the influence of prospective versus sunk costs on support for US intervention overseas. We have unearthed a number of interesting findings. First, the sunk costs fallacy has little purchase over US public opinion on foreign policy. Second, US citizens treat human and financial sunk costs differently. Monetary sunk costs generate a desire to leave – human sunk costs do not. Third, the US respondents do not appear to have an intrinsic preference for countering challenges with money rather than lives. We suggest a number of avenues for future research and policy recommendations.

First, policymakers should not imagine that abandoning missions in which casualties have been incurred will be perceived as a ‘betrayal of the fallen’. Our study suggests that overall, Americans judge interventions overseas in forward-looking terms. If the expected benefits of the intervention are no longer perceived as being worth the costs, voters want their elected leaders to call time, as Ronald Reagan did in Lebanon. Of course, politicians might still have electoral incentives to prolong unpopular wars. As the ‘gambling for resurrection’ literature shows, politicians may prefer a small chance of victory and electoral recovery rather than the certainty of electoral defeat if they admit their original decision for war was wrong (Downs & Rocke, 1994). By doing so, however, policymakers are unquestionably shirking their responsibility as the voters’ agents. Moreover, there is no such incentive for policymakers who inherited conflicts from their predecessors.

Second, relatedly, using sunk costs to rally support for war will not work. Politicians need to justify their policies to voters in terms of the benefits to the nation from now on.

Third, policymakers should concentrate on the best tools for the mission at hand. If bankrolling allies is the best way to get the job done because those allies have superior local knowledge or support, then this strategy is appropriate. However, our research suggests US citizens have no intrinsic preference for fighting battles with cash rather than flesh and blood. If boots on the ground ‘work’ in a way that machines do not, policymakers have leeway from the public to deploy them.

Fourth, our research casts doubt on Schelling’s ‘tripwire’ theory. There is no evidence that casualties would serve to ‘tie the hands’ of US policymakers. Placing US service people in harm’s way in, for instance, flashpoints in East Asia may simply risk lives without having the desired pacifying effects. Of course, the scenario outlined above is rather different from an interstate conflagration in Asia. US casualties incurred in a surprise attack initiated by an opponent of the United States in which US forces are perceived to be the victims, or which are inflicted by another state as opposed to insurgents, may produce quite different results. We view this as the most interesting avenue for future research.

Footnotes

Replication data

The dataset, codebook, and do-files for the empirical analysis in this article, along with the online appendix, can be found at . All analyses were conducted using R or STATA.

Acknowledgements

We thank Peter Feaver, Christopher Johnston, Scott Clifford, Paul Kenny, Kyle Dropp, Iain Henry, Amy Catalinac, and Jan Pierskalla for their input and comments. We would also like to thank the anonymous reviewers for their thoughtful suggestions.

Funding

We would like to thank the Bradley Foundation and the Australian National University’s Strategic and Defence Studies Centre for their generous provision of funding.

Notes

References

Arkes

Hal R

Blumer

Catherine

(1985) The psychology of sunk cost. Organizational Behavior and Human Decision Processes 35(1): 124–140.

Berejikian

Jeffery

(2002) Model building with prospect theory: A cognitive approach to international relations. Political Psychology 23(4): 758–786.

Berinsky

Adam J

(2007) Assuming the costs of war: Events, elites, and American public support for military conflict. Journal of Politics 69(4): 975–997.

Berinsky

Adam

Huber

Gregory

Lenz

Gabriel

(2012) Evaluating online labor markets for experimental research: Amazon.com’s Mechanical Turk. Political Analysis 20(3): 351–368.

Boettcher

William A

III Cobb

Michael D

(2009) Don’t let them die in vain: Casualty frames and public tolerance for escalating commitment in Iraq. Journal of Conflict Resolution 53(5): 677–697.

Divine

Robert

(1976) The Reluctant Belligerent: American Entry into World War II. New York: Wiley.

Downs

George W

Rocke

David M

(1994) Conflict, agency, and gambling for resurrection: The principal–agent problem goes to war. American Journal of Political Science 38(2): 362–380.

Eichenberg

Richard C

(2009) Public opinion and foreign policy in the Obama era. Politique Américaine 14(2): 11–24.

Fagley

Nancy S

Miller

Paul M

(1997) Framing effects and arenas of choice: Your money or your life? Organizational Behavior and Human Decision Processes 71(3): 355–370.

10.

Ferguson

Niall

(2004) Colossus: The Price of America’s Empire. New York: Penguin.

11.

Friedman

Daniel

Pommerenke

Kai

Lukose

Rajan

Milan

Garrett

Huberman

Bernard A

(2007) Searching for the sunk costs fallacy. Experimental Economics 10(1): 79–104.

12.

Gartner

Scott S

(2008) The multiple effects of casualties on public support for war: An experimental approach. American Political Science Review 102(1): 95–107.

13.

Gartner

Scott S

Segura

Gary M

(1998) War, casualties, and public opinion. Journal of Conflict Resolution 42(3): 278–300.

14.

Gelpi

Christopher

Feaver

Peter D

Reifler

Jason

(2009) Paying the Human Costs of War: American Public Opinion and Casualties in Military Conflicts. Princeton, NJ: Princeton University Press.

15.

Gelpi

Christopher

Reifler

Jason

D Feaver

Peter

(2007) Iraq the vote: Retrospective and prospective foreign policy judgments on candidate choice and casualty tolerance. Political Behavior 29(2): 151–174.

16.

Grieco

Joseph M

Gelpi

Christopher

Reifler

Jason

Feaver

Peter D

(2011) Let’s get a second opinion: International institutions and American public support for war. International Studies Quarterly 55(2): 563–583.

17.

Holsti

Ole R

(1992) Public opinion and foreign policy: Challenges to the Almond-Lippman consensus. International Studies Quarterly 36(4): 439–466.

18.

Huth

Paul

Gelpi

Christopher

Scott Bennett

(1992) System uncertainty, risk propensity and international conflict among the great powers. Journal of Conflict Resolution 36(3): 478–517.

19.

Jentleson

Bruce

(1992) The pretty prudent public: Post post-Vietnam American opinion on the use of military force. International Studies Quarterly 36(1): 49–73.

20.

Jervis

Robert

(1976) Perception and Misperception in International Politics. Princeton, NJ: Princeton University Press.

21.

Kahan

Daniel

(2013) Fooled twice, shame on who? Problems with Mechanical Turk study samples, part 2 (http://www.culturalcognition.net/blog/2013/7/10/fooled-twice-shame-on-who-problems-with-mechanical-turk-stud.html).

22.

Kahneman

Daniel

(2011) Thinking Fast and Slow. New York: Farrar, Straus & Giroux.

23.

Kahneman

Daniel

Tversky

Amos

(1979) Prospect theory: An analysis of decision making under risk. Econometrica 47(2): 263–291.

24.

Kull

Steven

Ramsay

Clay

(2001) American public attitudes on military fatalities in the post-Cold War period. In: Everts

Philip P

Isernia

Pierangelo

(eds) Public Opinion and the International Use of Force. London: Routledge, 199–221.

25.

Lebow

Richard N

(1981) Between Peace and War: The Nature of International Crisis. Baltimore, MD: Johns Hopkins University Press.

26.

Leeds

Daniel M

Leeds

Michael A

Motomura

Akira

(2015) Are sunk costs irrelevant? Evidence from playing time in the National Basketball Association. Economic Inquiry 53(2): 1305–1316.

27.

Levy

Jack S

(1997) Prospect theory, rational choice and international relations. International Studies Quarterly 41(1): 87–112.

28.

McAfee, R Preston; Hugo Mialon & Sue H Mialon (2010) Do sunk costs matter? Economic Inquiry 48(2): 323–336.

29.

McDermott

Rose

(2004) Political Psychology in International Relations. Ann Arbor, MI: University of Michigan Press.

30.

McGraw

A Peter

Tetlock

Philip E

(2005) Taboo trade-offs, relational framing, and the acceptability of exchanges. Journal of Consumer Psychology 15(1): 2–15.

31.

McInerney

Audrey

(1992) Prospect theory and Soviet policy towards Syria 1966–1967. Political Psychology 13(2): 265–282.

32.

Mueller

John E

(1973) War, Presidents and Public Opinion. New York: Wiley.

33.

Pinker

Steven

(2007) The Stuff of Thought: Language as a Window into Human Nature. New York: Viking.

34.

Richardson

Louise

(1991) Avoiding and incurring losses: Decision-making in the Suez crisis. International Journal 47(2): 370–401.

35.

Schaub

Gary J

(2004) Deterrence, compellence and prospect theory. Political Psychology 25(3): 389–411.

36.

Schelling

Thomas

(2008) Arms and Influence. New Haven, CT: Yale University Press.

37.

Schwartz

Barry

(2005) The sunk-cost fallacy. Slate 9 September (http://www.slate.com/articles/news_and_politics/hey_wait_a_minute/2005/09/the_sunkcost_fallacy.html).

38.

Schwartz

Barry

(2006) The ‘sunk-cost fallacy’. Los Angeles Times 17 September (http://articles.latimes.com/2006/sep/17/opinion/oe-schwartz17).

39.

Stein

Janice

(1992) International cooperation and loss avoidance: Framing the problem. International Journal 47(2): 202–234.

40.

Sullivan

Patricia L

(2008) Sustaining the fight: A cross-sectional time-series analysis of public support for ongoing military intervention. Conflict Management and Peace Science 25(2): 112–135.

41.

Sunstein

Cass R

(2005) Laws of Fear: Beyond the Precautionary Principle. Cambridge: Cambridge University Press.

42.

Taliaferro

Jeffery W

(2004) Balancing Risks: Great Power Intervention in the Periphery. Ithaca, NY: Cornell University Press.

43.

Tetlock

Philip E

(2003) Thinking the unthinkable: Sacred values and taboo cognitions. Trends in Cognitive Sciences 7(7): 320–324.

44.

Tetlock

Philip E

Kristel

Orie V

Beth Elson

Green

Melanie C

(2000) The psychology of the unthinkable: Taboo trade-offs, forbidden base rates and heretical counterfactuals. Journal of Personality and Social Psychology 78(5): 853–870.

45.

Tetlock

Philip E

Peter McGraw

Kristel

Orie V

(2004) Proscribed forms of social cognition: Taboo trade-offs, blocked exchanges, forbidden base rates, and heretical counterfactuals. In: Haslam

Nick

(ed.) Relational Models Theory: A Contemporary Overview. Mahwah, NJ: Lawrence Erlbaum, 247–262.

46.

US Enviromental Protection Agency, NCEE (2010) Guidelines for preparing economic analysis (http://yosemite.epa.gov/ee/epa/eerm.nsf/vwAN/EE-0568-50.pdf/$file/EE-0568-50.pdf).

47.

Walt

Stephen

(2010) Foreign policy: Bait and switch in Afghanistan. NPR 11 November (http://www.npr.org/templates/story/story.php?storyId=131241642).

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB