The British Accounting Review 44 (2012) 144–156

Contents lists available at SciVerse ScienceDirect

The British Accounting Review

journal homepage: www.elsevier .com/locate/bar

Sunk costs and sunk benefits: A re-examination of re-investmentdecisions

Darren Duxbury*

Leeds University Business School, The University of Leeds, Leeds LS2 9JT, UK

a r t i c l e i n f o

Article history:Received 20 October 2010Received in revised form 2 November 2011Accepted 16 November 2011

JEL classification:C91

Keywords:Escalation of commitmentProspect theorySunk benefitSunk cost

* Tel.: þ44 113 343 4508; fax: þ44 113 343 4459E-mail address: dd@lubs.leeds.ac.uk.

1 It is necessary at this stage make clear the distin(2008), escalation commonly refers to the persistedeterminants of this persistence, including self-justimpact of sunk costs on decision making, and necesdecision making. Thus escalation is seen as a broadecamp, though the two terms are used interchangea

0890-8389/$ – see front matter � 2012 Elsevier Ltdhttp://dx.doi.org/10.1016/j.bar.2012.07.004

a b s t r a c t

Prior experimental studies supporting the prospect theory explanation of the sunk-costeffect manipulate the framing of the initial investment, describing it either in neutral termsor as a prior loss. This paper subjects the prospect theory explanation to further exami-nation, but takes an alternative experimental approach based on the differential risk takingbehaviour predicted by prospect theory’s S-shaped value function. The experimentsmanipulate whether an initial investment produces a sunk cost (prior loss) or a sunkbenefit (prior gain) and investigate the impact of this on the likelihood of authorising anincremental investment held constant across treatment conditions. To ensure the resultsare robust to the type of incremental investment, two experiments are conducted acrosswhich the outcomes of the incremental investment are manipulated to produce poor orgood investment opportunities. In all cases the results fail to support a higher likelihood ofauthorising the incremental investment following a sunk cost than a sunk benefit. Inisolation, therefore, prospect theory is unable to explain fully the sunk-cost effect.

� 2012 Elsevier Ltd. All rights reserved.

1. Introduction

The conventional wisdom in management accounting follows the principles of economic rationality, which dictate thatindividuals should evaluate financial decisions based solely on an incremental analysis and so consider only future costs andbenefits. Behavioural research in accounting, however, provides convincing evidence that individuals’ decisions are influ-enced by prior outcomes. Numerous studies, for example, document escalation of commitment; the tendency to commitadditional funds to a failing project in an attempt to recoup prior sunk costs (e.g. Arkes & Blumer,1985; Cheng, Schulz, Luckett,& Booth, 2003; Chow, Harrison, Lindquist, & Wu, 1997; Denison, 2009; Garland, 1990; Ghosh, 1997; Kanodia, Bushman, &Dickhaut, 1989; Staw, 1976; Thaler, 1980).1 While various explanations of the sunk-cost effect have been proposed (seeWilson & Zhang, 1997, for a review), many authors (e.g. Arkes & Blumer, 1985; Garland, 1990; Whyte, 1993) draw onKahneman and Tversky’s (1979) prospect theory in which prior outcomes impact on subsequent decisions, with increasedrisk seeking behaviour in the presence of prior losses (i.e. sunk costs) and risk aversion in the presence of prior gains (whatmight be referred to as sunk benefits).

.

ction between “escalation of commitment” and the “sunk-cost effect”. Following Navarro and Fantinonce in a course of action, usually in the face of failure. Studies of escalation have identified severalification, but have not manipulated sunk costs. In contrast the sunk-cost effect relates to the specificsitates the manipulation of sunk cost as an independent variable to determine its role in investmentr phenomenon than the sunk-cost effect. This paper manipulates sunk cost and so belongs to the latterbly where no confusion will ensue.

. All rights reserved.

D. Duxbury / The British Accounting Review 44 (2012) 144–156 145

Given that escalation of commitment is routinely related to a failing course of action, the escalation and sunk-cost effectliteratures have tended to focus on the prospect theory explanation of behaviour in the presence of prior losses, ignoring itsprediction relating to behaviour in the presence of prior gains. Whyte (1986, p. 319), however, an early proponent of theprospect theory explanation of the escalation of commitment or sunk-cost effect, recognises the testable implication stem-ming from prospect theory’s prediction in the presence of prior gains; notably that “escalating commitment in the context ofsuccess will occur, but the level of commitment evidenced in the future will tend to be less than is warranted by an objectiveanalysis of the situation.”More specifically, if prospect theory is at the root of the sunk-cost effect, then changes in risk takingbehaviour following prior outcomes shouldmake re-investment more likely following sunk costs (prior losses) thanwould beobserved following sunk benefits (prior gains).

This paper examines this testable implication, thus subjecting the prospect theory explanation of the sunk-cost effect tofurther experimental examination. In the experiments reported here participants are presented with either a sunk cost ora sunk benefit investment scenario and then asked to make an identical, incremental investment decision, the outcome ofwhich feeds directly into the monetary incentive mechanism used to determine payment for participation in the experiment.Holding the incremental investment constant, economic theory would predict no difference in the tendency to authorise theincremental investment across the sunk cost/benefit treatment, while if prospect theory is at the root of the sunk-cost effectthere should be an increased tendency for those participants experiencing the sunk-cost condition to accept the incrementalinvestment than those experiencing the sunk benefit condition.

To ensure the results are robust to the nature of the incremental investment two versions of the experiment are run. InExperiment One the incremental investment represents a poor investment (i.e. it has a negative expected value), while inExperiment Two it represents a good investment (i.e. positive expected value). Contrary to the predictions of both economictheory and prospect theory, the results from Experiment One indicate that individuals are less likely to authorise incrementalinvestments in the presence of a sunk cost than a sunk benefit when the incremental investment represents a poorinvestment opportunity. In Experiment Two, when the incremental investment represents a good investment, contrary toprospect theory there is no effect of the sunk cost/benefit treatment on the tendency to authorise the incremental investment,which is in line with economic theory. In isolation, therefore, prospect theory is unable to explain fully the sunk-cost effect,thus supporting Wilson and Zhang (1997) who conclude that, while many of the explanations of escalation have explanatorypower, none alone can explain fully the phenomenon.

The structure of the paper is as follows. The next section provides a brief review of related literature, explains in moredetail the prospect theory explanation of the sunk-cost effect and states in general terms the hypotheses to be examined.Section 3 discusses the experimental method employed, along with the background of the participants and a manipulationcheck. Section 4 highlights unique features of the design and presents the results from Experiment One, while Section 5 doesthe same for Experiment Two. Section 6 presents additional analyses intended to provide further insight concerning thefactors that influence participants’ decisions to authorise the incremental investment. The final section draws conclusions.

2. Related literature and the prospect theory explanation

The principles of economic rationality dictate that individuals should evaluate decisions based solely on an incrementalanalysis and so consider only future costs and benefits. Thus prior outcomes should not impact upon their decisions. However,prior studies document that individuals’ decisions, whether consumption or investment related, are influenced by prioroutcomes (e.g. Thaler & Johnson, 1990). One area that has received widespread investigation is the sunk-cost effect; thetendency to commit additional funds to a project in an attempt to recoup prior sunk costs. Numerous academic studies, withtheir roots in the early work by Staw (1976) on the escalation of commitment, have documented empirically the effect of sunkcosts on individuals’ decisions to commit incremental investments (e.g. Garland, 1990; Thaler, 1980). Routinely citedexamples of the sunk-cost effect at large in practice range from the prolonging of the Vietnam War, the collapse of Baringsbank due to the investment behaviour of Nick Leeson, continued financial commitment to the supersonic airliner Concordeand NASA’s space shuttle program. The sunk-cost effect is believed to have played an integral part in the irrational decision tocontinue investing (whether in terms of time, money or life) in projects that should have been abandoned on a rationaleconomic basis (i.e. projects that represented poor investment opportunities).

Various explanations of the sunk-cost effect have been proposed in the literature, including the desire to avoid waste (e.g.Arkes & Blumer, 1985), personal responsibility and the need for self-justification (e.g. Schulz & Cheng, 2002; Staw, 1976),reputation and information asymmetries (e.g. Kanodia et al., 1989) and mental accounting effects (e.g. Soman & Cheema,2001; Thaler, 1980). Another popular approach in the literature is to draw on the implications of Kahneman and Tversky’s(1979) prospect theory to explain the sunk-cost effect (e.g. Arkes & Blumer, 1985; Garland, 1990; Northcraft & Neale, 1986;Whyte, 1993). In recognition of the high cost implications of falling foul of the sunk-cost effect, recent studies have begunto investigate ways inwhich the effect can be mitigated (e.g. Cheng et al., 2003; Ghosh,1997; McCain,1986; Tan & Yates, 1995,2002). Such endeavours would be more productive, however, if a clear understanding of why individuals succumb to thesunk-cost effect was available in the first place. The wide range of possible explanations cited above, indicates that a generalconsensus does not exist. This paper contributes to our understanding of the sunk-cost effect by taking one explanation,prospect theory, and subjecting it to further experimental scrutiny.

Whyte (1986) provides a comprehensive discussion of the prospect theory explanation of the sunk-cost effect, thus thefollowing discussion is purposefully brief. The S-shaped value function of prospect theory is concave in the domain of gains

D. Duxbury / The British Accounting Review 44 (2012) 144–156146

and convex in the domain of losses, and steeper in the loss domain than the gain domain, which Kahneman and Tversky(1979) term the reflection effect. Thus, the absolute value of a marginal gain is less than the absolute value of an equiva-lent marginal loss. The implication of the reflection effect is that individuals will display more risk seeking behaviour in thedomain of losses than they would in the domain of gains. The concave/convex shape of the value function implies that thevalue placed on a prospect or gamble is not constant, but is dependent upon the reference point from which it is evaluated.Thus, the marginal value placed on a given gain/loss declines the further an individual’s reference point is from the origin. Forexample, a marginal loss of £X is felt more keenly if the reference point is zero than if the same marginal loss of £X wasevaluated from the position of a prior loss. Kahneman and Tversky (1979) call this the reference effect. Jointly the reflectionand reference effects imply that if a sunk cost is coded as a prior loss, an individual may be predisposed to accept riskygambles that they would not have accepted in the absence of the sunk cost.

In an early examination of prospect theory in a management accounting setting Harwood, Pate, and Schneider (1991)employ a modified version of the classic framing manipulation used by Kahneman and Tversky (1979) to investigate theimpact of framing of decision alternatives on budgeting decisions. While they report evidence of a framing effect, they do notinvestigate escalation or sunk-cost effects specifically. Following the framing manipulation approach of Harwood et al. (1991),Whyte (1993) and Sharp and Salter (1997) subject the prospect theory explanation of escalation to experimental examination.Both studies manipulate the description of the initial investment, framing it either neutrally or as a prior loss. While the initialinvestment is constant across the framing manipulation, when framed as a loss there is an increased tendency to observeescalation or sunk-cost effects in the experiments ofWhyte (1993) and Sharp and Salter (1997). Given the focus in these studiesis on the notion of escalation, which is commonly studied in the context of an initially failing project, both studies consider onlyone dimension of prior outcomes, namely losses. Prospect theory, however, describes decisions under risk more generally,predicting that prior gains, in addition to losses, will influence subsequent behaviour. It is this broader view of decision makingunder risk that this paper exploits in a re-examination of the prospect theory explanation of the sunk-cost effect.

This paper adopts an alternative experimental approach to the framingmanipulations used inprior studies. Prospect theory’sS-shapedvalue functiondepicts risk seekingbehaviour for losses and riskaversion for gains.Whileprospect theorycanbeused toexplainwhy individuals arewilling to invest additional funds in a failing project, evenwhen the incremental investment is on lessthan favourable terms (i.e. poor investment opportunity), it would also predict that in the presence of prior gains an individualmaybepredisposed tomore readily dismiss riskygambles theywould accept otherwise, thus resulting inunderinvestment in thepresence of prior success (a proposition acknowledged byWhyte, 1986, p. 319). Thus a prospect theory explanation of the sunk-costeffect leads to the testable implication, ceterisparibus, that an increased tendency for re-investment shouldbeobservedaftera sunk cost than a sunk benefit. This paper exploits prospect theory’s disparate predictions about individuals’ risk preferences inthe loss domain versus gain domain to investigate experimentally its ability to explain the sunk-cost effect in isolation. The paperexamines the following null and alternate hypotheses expressed in general terms.

Ho: The presence of a sunk cost versus a sunk benefit does not significantly impact upon the decision to authorise additionalfunds to an incremental investment.Ha: The presence of a sunk cost versus a sunk benefit significantly increases the decision to authorise additional funds to anincremental investment.

Premised on the principle of economic rationale, in which prior outcomes are irrelevant to future decisions, the nullhypothesis (Ho) states that the decision to authorise an incremental investment is expected to be unaffected by the presenceof a prior sunk cost (prior loss) or sunk benefit (prior gain). In contrast, the alternate hypothesis (Ha) draws on prospect theoryto predict, ceteris paribus, that authorisation of an incremental investment is more likely following a sunk cost than followinga sunk benefit. The following section discusses the experimental method employed.

3. Experimental method

3.1. General experimental design

Participants are presented with a decision scenario in which they adopt the role of a project manager in a medium sizedfirm (see the Appendix for an example of the experimental instrument). A 2 � 2 between-subjects design is employed assummarised in Table 1, which also displays the experimental parameters employed, and explained below.

In the first part of the decision scenario participants are informed about the cash flows associated with an existing projectthey had previously authorised.2 The two-level SUNK treatment manipulates whether the participants were presented withan existing investment that produced a sunk cost or sunk benefit. In the second part of the decision scenario, the re-investment decision, the expected cash flows associated with an incremental investment to prolong the existing projectare displayed and participants are asked to indicate on a scale of 1–7 how likely they would be to authorise the incrementalinvestment (where 1 ¼ ‘definitely authorise’ and 7 ¼ ‘definitely not authorise’). In this incremental investment stage

2 Prior studies (e.g. Arkes & Blumer, 1985; Staw, 1976) document the importance of responsibility for the sunk cost, thus the decision scenario informsparticipants they are responsible for investment decisions in the organisation, including the existing project.

Table 1Design and parameters: Experiment one and two.

Experiment one (poor project) Experiment two (good project)

Sunk cost Sunk benefit Sunk cost Sunk benefit

Good outcome low probabilityExisting project £m’s £m’s £m’s £m’sInitial investment 3 3 3 3Cash inflow 2 4 2 4Net cash flow �1 1 �1 1

Incremental InvestmentImmediate cash outflow 2 2 2 2Good outcome inflow (prob. ¼ 0.2) 5 5 15 15Bad outcome inflow (prob. ¼ 0.8) 0 0 0 0Net cash flow (expected value) �1 �1 1 1

Good outcome high probabilityExisting project £m’s £m’s £m’s £m’sInitial investment 3 3 3 3Cash inflow 2 4 2 4Net cash flow �1 1 �1 1

Incremental investmentImmediate cash outflow 2 2 2 2Good outcome inflow (prob. ¼ 0.8) 5 5 7.5 7.5Bad outcome inflow (prob. ¼ 0.2) �15 �15 �15 �15Net cash flow (expected value) �1 �1 1 1

The table reports the experimental designs and parameters employed in Experiment one and two. Each experiment employs a 2 (sunk cost/benefit)� 2 (low/high probability of good outcome from incremental investment) design. In Experiment One the incremental investment has a negative expected value and sorepresents a poor opportunity, while in Experiment Two it has a positive expected value and so represents a good opportunity.

D. Duxbury / The British Accounting Review 44 (2012) 144–156 147

a monetary incentive mechanism is employed with one in every 25 participants chosen randomly to play out their decisionfor cash at a rate of £10 for every £1m earned by the incremental investment. The decision facing participants can essentiallybe viewed as a choice between a certain payout and a lottery. Those individuals choosing not to authorise the incrementalinvestment (i.e. they responded with a score of 5–7) effectively choose the certain payout of £20 (i.e. the payout equivalent ofthe £2m cash outflow not invested). Those individuals choosing to authorise the incremental investment (i.e. they respondedwith a score of 1–3) effectively choose the lottery, with the outcome (good or bad) from the incremental investmentdetermined at random and the individual receives the net cash flow from the incremental investment. For example, considerthe versions of the experimental instrument faced by an individual where the good outcome was £15m (probability 0.2) andthe bad outcome was £0m (probability 0.8). If the good outcome of £15m was chosen at random based on the associatedprobability of 0.2, then the individual would receive payment of £130 (¼£15m–£2m, at a rate of £10 per £1m) and zero if thebad outcome was randomly chosen. For ethical reasons if the calculated net cash payment for an individual turned out to benegative themoney owed could not be taken from the participant. Instead, if the cash payment was negative (as would be thecase if the bad outcomewas selected at random in the example above), the individual would undertake departmental clericalwork until the negative cash payment was paid off at a rate of 1 h for every £5. While participants could not lose their ownmoney, therefore, the forgone wage rate of £5 per hour represents a real opportunity cost.3,4

An important feature of the experimental design is that the net cash flows from the existing project (i.e. sunk cost or sunkbenefit) do not form part of the monetary incentive mechanism and thus should be irrelevant to the choice between thegamble and the certain payout. This was an intentional feature of the experimental design to ensure full comparability ofmonetary incentives across the SUNK treatment conditions and to remove the potential for house money effects associatedwith windfall gains (Thaler & Johnson,1990) to distort the results. Soman and Cheema (2001) demonstrate that windfall gainsweaken the sunk-cost effect, thus if real monetary incentives based on the existing project cash flows are employed, such thatparticipants receive windfall gains (sunk benefits) or losses (sunk costs), there is the potential for the house money effect tooverwhelm any sunk-cost effect.5

3 Participants knew and agreed to this in advance. The situation occurred once across the two experiments.4 If selected, an individual with a score of 4 would have been required to make a definitive decision, but this situation never occurred.5 An alternative approach to the one adopted here would be to initially endow participants with a real cash lump-sum. In the sunk benefit treatment an

additional amount would be added to represent the gain on the existing project, while in the sunk cost treatment some of the cash lump-sum would beconfiscated to represent a loss on the existing project. On first consideration such an approach would appear to build in real, prior gains and losses, ratherthan using hypothetical ones. However, unless the amount confiscated exceeded the initial cash endowment (which would be problematic from an ethicalperspective), even those participants in the sunk cost condition would effectively find themselves in a real, net windfall gain situation and so couldpotentially promote house money type behaviour.

D. Duxbury / The British Accounting Review 44 (2012) 144–156148

The two-level PROB treatment is a secondary factor to manipulate the probability of the incremental investmentproducing a good outcome (i.e. the higher of the two possible cash flows) such that the probability is either 0.8 (high) or 0.2(low). The expected value of the incremental investment is held constant across the PROB treatment conditions. The PROBtreatment is included to ensure that the results with respect to the main SUNK treatment factor are robust towhether there isa high or low probability of the incremental investment being a success.6 The experimental parameters employed are dis-played in Table 1.

3.2. Participants

In order to provide as clean a test as possible of the prospect theory explanation of the sunk-cost effect, it was deemeddesirable to draw participants from as homogeneous a participant-pool as possible to ensure that differences in behaviouracross treatment conditions were due solely to the experimental manipulations under investigation and not due to differ-ences between participants. For this reason it was thought that students would represent an ideal participant-pool, henceparticipants in both experiments were undergraduate students at a UK Business School. All participants had undertakenmodules in management accounting and as such had been exposed to investment appraisal techniques such NPV, IRR, ARRand Payback. In addition they had undertaken modules in statistics and so were familiar with probabilities and features ofdistributions includingmeasures of central tendency (mean), dispersion (standard deviation/variance) and symmetry (skew).Given this common background, all participants were well placed to engage with the decision and to evaluate the risksassociated with the investment scenario they faced, thus ensuring the internal validity of the experiments. The participant-pools were approximately evenly split by gender and ages ranged from 19 to 21 years.

It is perhaps prudent at this stage to revisit the long standing debate on the use of students as surrogates for practitionersin accounting experiments. Proponents of the view that students serve as poor surrogates for practitioners point to evidencefrom a variety of studies (e.g. Bouwman, 1984; Frederick, 1991; McAulay, King, & Carr, 1998) demonstrating differencesbetween the decision-making processes of ‘experts’ (practitioners) and ‘novices’ (students). A common feature of such studiesis the elicitation of attitudes or judgements concerning complex, ill-defined scenarios (e.g. case studies with no definitivesolution), where prior experience and expertise are fundamental to the attitude forming process. In contrast, there is evidenceto suggest that students canmeaningfully serve as surrogates for practitioners in scenarios concerning decision-making taskswhere there exists a transparent connection between the decision (choice) and the outcome, whether it be certain orprobabilistic in nature (e.g. Ashton & Kramer,1980; Remus,1986). In summary, while students may not be good surrogates forpractitioners in complex, ill-defined, attitude based tasks, it seems in simple, choice-outcome decisions with well-definedincentive mechanisms, of the kind studied in this paper, that students can be used as surrogates for professionals withoutany obvious loss of external validity. More recently, in a task close in nature to the focus of the current study, Liyanarachchiand Milne (2005) demonstrate that students can be viewed as acceptable surrogates for accounting practitioners in aninvestment decision task, for both short-term and long-term decisions. Their findings provide further evidence to support theview that the results reported in this paper are robust to the use of students as participants.

It is important also to note the purpose of this paper, which is to test theoretical predictions via the manipulation ofexperimental treatment conditions. In this context interpretation of the results reported below is not hindered by thedecision to use student participants. Indeed, an argument could be made that the use of accounting practitioners, who mayhave varying degrees of prior exposure to sunk-cost situations and may bring with them pre-conceived ideas from outside ofthe laboratory, wouldweaken the level of control achievable in the experiments.7 Furthermore, Libby, Bloomfield, and Nelson(2002) argue that accounting practitioners willing to participate in experimental research represent a valuable and scarceresource, only to be used when the research question under consideration necessitates it. That is not the case here and so thedecision was made to use students in the current study, so as to conserve the practitioner participant-pool for other studieswith research agendas for which practitioners are deemed essential.

3.3. Manipulation check

While the hypothetical nature of the sunk costs and sunk benefits in the first part of the decision scenario is a necessarycomponent of the experimental design, as discussed above, the success of the treatment manipulation remains to bedemonstrated. By way of manipulation check, 40 individuals different to those participating in the main experiments re-ported belowwere shown information relating to the existing investment employed in themain experiments. In a task askingparticipants to spontaneously report words to describe the existing investment, 17/20 individuals in the sunk-cost conditionreported words relating to “loss” and 16/20 individuals in the sunk benefit condition reported words relating to “gain”, thusestablishing the effectiveness of the SUNK treatment manipulation.

6 Prior studies (e.g. Arkes & Hutzel, 2000; Harbaugh, Krause, & Vesterlund, 2002) suggest that this may be important in the context of the currentexperiments.

7 It is conceivable that random assignment to treatment conditions could not be relied upon to ensure experimental control had accounting practitionersbeen used as participants in the experiments.

D. Duxbury / The British Accounting Review 44 (2012) 144–156 149

4. Experiment one

4.1. Experimental design

The convention in prior studies of escalation of commitment and sunk-cost effects is for the incremental investment torepresent a poor and risky investment opportunity that is unlikely to lead to a good outcome (e.g. Garland, 1990; Staw, 1981),hence the notion that individuals are “throwing good money after bad” (Garland, 1990). In Experiment One this conventionalfeature of prior studies is retained and the incremental investment represents a poor investment opportunity in the sensethat the expected value is negative. The experimental design and parameters are displayed in Table 1. Note that in all four cellsof the design the expected value of the incremental investment is held constant.

4.2. Hypotheses

The following formal hypotheses are tested in Experiment One and also in Experiment Two. The null hypothesis reflectsthe rational economic perspective, while the alternative hypothesis states the predicted behaviour based on prospect theory.

Ho: There is no difference in mean authorisation score across the sunk cost and sunk benefit treatment conditions.Ha: The mean authorisation score in the sunk-cost condition is lower than that in the sunk benefit condition.

4.3. Analysis and results

Fig. 1 summarises the percentage of responses across the authorisation score categories and Table 2 reports the author-isation score frequencies and descriptive statistics for the decision to authorise the incremental investment in ExperimentOne and Experiment Two. While participants in Experiment One are closely split between willing/unwilling to authoriseprojects with negative incremental expected value (willing ¼ 44.5% scores � 3, unwilling ¼ 52.5% scores � 5), the majoritydisplay a clear preference one way or the other with only 4 participants unable to decide (score ¼ 4). The data provides clearevidence, therefore, that the participants engaged with the decision scenarios and felt able to make a decision aboutauthorising the incremental project or not.

Table 3 reports themean responses by treatment condition for the 137 participants in Experiment One, alongwith the resultsfrom a between-subjects ANOVA. The SUNK and PROB treatments produce significant main effects, but the interaction effect isinsignificant. Contrary to both economic theory and prospect theory, the results indicate a higher likelihood of authorising (i.e.lower score) the incremental investment following a sunkbenefit (3.89) thana sunk cost (4.46). ThusHo andHa are both rejectedin favour of lowermean authorisation scores in the presence of a sunk benefit, supporting a higher likelihood to authorise in thistreatment condition. The results also indicate that the higher the probability of the ‘good’ outcome, the higher the likelihood ofauthorising the incremental investment (3.59 and 4.78, respectively), despite the expected value being held constant.8

In all four cells of the experiment the expected value of the incremental investment is negative, thus authorising theincremental investment clearly constitutes risk seeking behaviour. Contrary to the predictions of prospect theory the resultssupport the view that individuals exhibited, relatively speaking, increased risk seeking behaviour in the presence of a sunkbenefit rather than in the presence of a sunk cost. Thus, when the incremental investment represents a poor opportunity (i.e.negative expected value), as is the norm in the escalation of commitment literature, the predictions of prospect theory are notupheld.9 This result is robust to changes in the probability that the incremental investment delivers a ‘good’ outcome. Thisprobability, however, impacts on individuals’ relative tendency to risk seeking behaviour in its own right. When the prob-ability of a ‘good’ outcome is high (i.e. 0.8) there is an increased tendency to risk seeking behaviour in the absence of changesin expected value. The behaviour across the PROB treatment conditions is at odds with standard assumptions in financetheory where expected return (expected value of outcomes) is traded-off against risk (variance of outcomes). The variance ofthe outcomes in the high (0.8) treatment condition is higher than in the low (0.2) treatment condition, without any increasein expected value to offset this.

8 The data do not meet some of the distributional assumptions for ANOVA. The Shapiro–Wilk test indicates non-normality and Levene’s test suggests thathomogeneity of variance may not hold (p ¼ 0.078). While it is commonly accepted that ANOVA is robust to violations of these assumptions when cellnumbers are similar (see Lindman, 1974), as is the case here, additional analyses are conducted to confirm this. First, given the insignificant SUNK � PROBinteraction in Table 3, one-way ANOVAs are run separately for the SUNK and PROB treatment factors with the Brown–Forsythe andWelch corrections to theF test requested to check for robustness to violations of homogeneity of variance. Results from these analyses are consistent with those reported in Table 3.Second, the above one-way analyses are replicated using the non-parametric medians test and one-way Kruskal–Wallis test, with both tests confirming theoriginal results are robust to departures from normality. Third, the dependent variable is converted to a binary, yes/no authorisation variable (scores of 1–3coded “yes” and scores 5–7 coded “no”, with scores of 4 omitted). The coefficients and significance levels from a binary logistic regression again confirm therobustness of the original ANOVA results, with both the SUNK and PROB treatment factors being significant. It is concluded, therefore, that the results inTable 3 are robust to departures from the distributional assumptions for ANOVA.

9 The finding of behaviour contrary to prospect theory predictions echoes the results in Moreno, Kida, and Smith (2002) who report increased riskseeking for gains and risk aversion for losses in the presence of affect and conclude that “we should not automatically assume that accounting decisionmakers will conform to the prospect theory predictions.” (p. 1346).

Experiment One

0%

10%

20%

30%

40%

50%

60%

70%

1 2 3 4 5 6 7

Authorisation score

Percen

tag

e r

esp

on

se

PROBLOW:SUNK COST PROBLOW:SUNK BENEFITPROBHIGH:SUNK COST PROBHIGH:SUNK BENEFIT

Experiment Two

0%

10%

20%

30%

40%

50%

60%

70%

1 2 3 4 5 6 7

Authorisation score

Percen

tag

e r

esp

on

se

PROBLOW:SUNK COST PROBLOW:SUNK BENEFITPROBHIGH:SUNK COST PROBHIGH:SUNK BENEFIT

Fig. 1. Percentage of responses across authorisation score – Experiment one and Experiment two. The figure reports the percentage of responses across theauthorisation score categories for each treatment condition.

D. Duxbury / The British Accounting Review 44 (2012) 144–156150

5. Experiment two

5.1. Experimental design

While the results from Experiment One fail to support the prospect theory explanation of the sunk-cost effect, the theorymakes predictions under more general conditions than those in Experiment One. Specifically, the prospect theory prediction

Table 2Frequency of authorisation score and descriptive statistics – Experiment one and Experiment two.

Experiment one (poor project) Experiment two (good project)

PROB LOW with SUNK COST SUNK BENEFIT SUNK COST SUNK BENEFITScore Frequency Frequency Frequency Frequency1 3 1 1 22 1 3 3 53 5 6 19 134 – 1 5 45 9 11 15 166 13 9 10 97 5 2 2 2n 36 33 55 51Mean 4.94 4.61 4.24 4.22Median 5.5 5.0 4.0 5.0Mode 6 5 3 5Variance 3.14 2.50 2.04 2.37

PROB HIGH with SUNK COST SUNK BENEFIT SUNK COST SUNK BENEFITScore Frequency Frequency Frequency Frequency1 1 3 1 22 9 8 16 93 8 13 24 364 1 2 4 35 8 3 7 86 4 4 1 47 4 – – –

n 35 33 53 62Mean 3.97 3.18 3.06 3.29Median 3.0 3.0 3.0 3.0Mode 2 3 3 3Variance 3.32 2.15 1.17 1.36

The table reports the frequency of the authorisation scores and descriptive statistics for the likelihood of authorisation. The scores are from a 7-point Likertscale anchored on 1 ¼ ‘definitely authorise’ and 7 ¼ ‘definitely not authorise’.

D. Duxbury / The British Accounting Review 44 (2012) 144–156 151

of risk seeking following losses and risk aversion following gains holds irrespective of the nature of the incrementalinvestment, be it a good or bad opportunity. Experiment Two was intended, therefore, to give prospect theory a secondchance at predicting incremental investment behaviour and to determine whether the results reported for Experiment Oneare robust to the type of incremental investment (i.e. whether it represented a ‘poor’ or ‘good’ project). The design replicatesthat of Experiment One with only one exception; the expected value of the incremental project is positive, thus classing it asa good investment opportunity. Where possible the experimental parameters replicate those in Experiment One, however,some changes are necessary to produce the positive expected value (see Table 1). The monetary incentive mechanism isidentical across the two experiments.

5.2. Analysis and results

Experiment Two investigates the same hypotheses as those examined in Experiment One (see Section 4.2 above). Table 4reports the mean responses by treatment condition for the 221 participants in this experiment, along with the results fromthe between-subjects ANOVA. A significant main effect is reported for the PROB treatment, but insignificant SUNK treatmentand interaction effects. As with Experiment One, the higher the probability of the ‘good’ outcome, the higher the likelihood ofauthorising the incremental investment (3.18 and 4.23, respectively). There is no statistical difference, however, in thelikelihood of authorising the incremental investment across the SUNK cost and benefit treatment conditions (3.71 and 3.66,

Table 3Mean likelihood of authorisation and between-subjects ANOVA results: Experiment one – poor incremental investment.

SUNK COST SUNK BENEFIT Overall mean F test Sig. level

PROB LOW 4.94 4.61 4.78 (1, 133) ¼ 17.584 p ¼ 0.000PROB HIGH 3.97 3.18 3.59Overall mean 4.46 3.89

SUNK*PROB interactionF test (1, 133) ¼ 3.893 (1, 133) ¼ 0.6223 p > 0.4Sig. level p ¼ 0.051

The table reports the results from a 2� 2 between-subjects ANOVA of themean likelihood of authorisation of the incremental investment when it representsa poor opportunity. The scores are from a 7-point Likert scale anchored on 1 ¼ ‘definitely authorise’ and 7 ¼ ‘definitely not authorise’. Participant numbersacross the four cells are approximately equal, ranging from 33 to 36, with a total of 137 participants. All significance levels are two-tailed.

Table 4Mean likelihood of authorisation and between-subjects ANOVA results: Experiment two – good incremental investment.

SUNK COST SUNK BENEFIT Overall mean F test Sig. level

PROB LOW 4.24 4.22 4.23 (1, 217) ¼ 35.501 p ¼ 0.000PROB HIGH 3.06 3.29 3.18Overall mean 3.71 3.66

SUNK*PROB interactionF test (1, 217) ¼ 0.364 (1, 217) ¼ 0.518 p > 0.4Sig. level p > 0.5

The table reports the results from a 2� 2 between-subjects ANOVA of themean likelihood of authorisation of the incremental investment when it representsa good opportunity. The scores are from a 7-point Likert scale anchored on 1 ¼ ‘definitely authorise’ and 7 ¼ ‘definitely not authorise’. Participant numbersacross the four cells are approximately equal, ranging from 51 to 62, with a total of 221 participants. All significance levels are two-tailed.

D. Duxbury / The British Accounting Review 44 (2012) 144–156152

respectively). The rejection of Ha once more challenges the prospect theory explanation of the sunk-cost effect, while thefailure to reject Ho provides support for economic theory.10

In all four cells of the experiment the expected value of the incremental investment is positive, but the probabilistic natureof the outcome still means that authorising the incremental investment constitutes a risky proposal. Consequently, prospecttheory still predicts greater risk seeking behaviour in the presence of a sunk cost than a sunk benefit. Thus, while the evidencefrom Experiment Two is not as damning as that from Experiment One, it still fails to support the prospect theory explanationof the sunk-cost effect. When the incremental investment represents a good opportunity (i.e. positive expected value), as isthe case in Experiment Two, decisions are in line with economic theory in the sense that participants are not influenced bythe sunk outcome, thus appearing more rational. As with Experiment One, the results support the conclusion that when theprobability of a ‘good’ outcome is high (i.e. 0.8) there is an increased tendency to risk seeking behaviour (i.e. authorising theincremental investment) in the absence of changes in expected value.

6. Further insight

This section reports the results of additional analyses that combine to provide further insight into how participantsevaluate the risk inherent in the decision to authorise the incremental project.11 Across the two experiments, the incrementalinvestment differs on a number of dimensions including expected value, maximum loss and probability of the good/badoutcomes (i.e. probability of gain/loss, respectively). The academic backgrounds of the participants ensure that they are wellplaced to evaluate the risks associated with acceptance of the incremental project.

6.1. The potential to misperceive zero outcome cash flows

How participants react when the bad outcome inflow is zero, as is the case in the PROB LOW condition across bothexperiments, can provide further insight into their decision-making processes. If participants discount the initial cash outflowof the incremental project (�2 in all cases) theymaymisperceive12 the bad outcome inflow of zero as a “no loss” position andso view the incremental project as riskless, in which case one would expect a higher likelihood of authorising when the badoutcome is zero (0) than when the bad outcome is negative (�15). However, across both experiments, the number ofparticipants authorising the incremental investment when the bad outcome is 0 and �15 are 65/175 ¼ 35.4% and 130/183 ¼ 71.0%, respectively. Furthermore, the mean authorisation score of 4.45 when the bad outcome is 0 (in the PROB LOWcondition) is significantly (p ¼ 0.000) higher than the mean score of 3.33 when the bad outcome is �15 (in the PROB HIGHcondition). Participants are less likely to authorise the incremental investment when the bad outcome is 0 than when it is�15, indicating that they perceive the risk associated with the incremental investment when the bad outcome is 0 and do notmisconstrue the investment to be riskless.

6.2. The role of expected value and probability of loss

It is possible that participants focus exclusively on expected value when making their decisions, thus ignoring otherindicators of risk. If this is the case then participants on the whole would be unwilling to authorise the incremental project inExperiment One due to its negative expected value. As can be seen from the distribution of authorisation scores reported inTable 2, however, 44.5% of participants are willing to authorise the incremental project in Experiment One, hence expectedvalue alone is not driving their decisions. Staying with Experiment One, the tendency to authorise the incremental

10 The results of identical robustness checks to those reported for Experiment One, confirm that the results in Table 4 are robust to departures from thedistributional assumptions for ANOVA.11 The referee raised many interesting observations that motivated the inclusion of this section in the paper.12 Misperceive because the worst situation is not a “no loss” position, but a negative net outcome of 0–2 ¼ �2 (the 0 cash inflow from the incrementalproject minus the 2 initial cash outflow if a decision to authorise is taken).

D. Duxbury / The British Accounting Review 44 (2012) 144–156 153

investment is influenced greatly by the probability of the outcomes, with the likelihood of authorising significantly higherunder the PROB HIGH condition (42/68 ¼ 61.7%) than the PROB LOW condition (19/69 ¼ 27.5%), indicating that participantsreact to the risk associated with the probability of good/bad outcomes, when the expected value of the incremental project isnegative. The same is true also of the incremental project with a positive expected value in Experiment Two. The strong andsignificant main effects for probability of outcome (PROB treatment factor) reported in both Tables 3 and 4 support stronglythe view that participants evaluate risk in ways other than just looking at expected value, in particular they look to theprobability of good/bad outcomes (i.e. probability of gain/loss) with respect to the incremental project.

The finding that 71.0% of participants are willing to authorise the incremental investment when the bad outcomecondition is �15 (see Section 6.1 above), despite the potential for a much higher loss, provides clear insight into the cognitiveprocess adopted. In this condition the probability of obtaining the good outcome is 0.8 (PROB HIGH), hence the probability ofloss is low at 0.2. This is further evidence, therefore, that participants, at least in part, perceive probability of loss as anindication of the risk associated with authorising the incremental project. This finding is in line with a long establishedliterature in psychology relating risk perception to the probability of loss (e.g. Payne, 1975; Shapira, 1995; Slovic &Lichtenstein, 1968; Yates & Stone, 1992).

6.3. The ability to distinguish good from bad projects

Turning to a comparison across Experiment One and Experiment Two, reveals similar mean authorisation scores acrossthe SUNK BENEFIT conditions of Experiment One (bad incremental project) and Experiment Two (good incremental project).While themean scores for the bad (3.89) and good (3.71) incremental projects are in the expected direction (higher likelihoodto authorise the good project with a positive expected value), there is no significant difference between these mean scores. Apossible interpretation of this result is that participants do not recognize that the good incremental project in ExperimentTwo has a higher expected value than the bad incremental project in Experiment One. However, adding in the data from theSUNK COST conditions of Experiment One and Experiment Two, to allow a full analysis of the effect of good versus badincremental project, produces mean scores of 4.19 and 3.68, respectively. This result is statistically significant (p ¼ 0.005) andindicates that participants recognize that the good incremental project represents a lower risk than the bad incrementalproject.

Further comparing across Experiment One and Experiment Two, the mean authorisation scores are very similar across theSUNK BENEFIT condition in Experiment One and the SUNK COST condition in Experiment Two, suggesting that the twoconditions are perceived to be similar in risk. While both conditions produce a net expected value of 0 (þ1 � 1 and �1 þ 1,respectively) when taking into account the sunk outcome, suggesting that sunk outcomes are incorporated in the evaluationof perceived risk, the incremental project in Experiment One (negative expected value) is inferior to that in Experiment Two(positive expected value), which suggests that the likelihood of authorising should be higher in Experiment Two thanExperiment One. Indeed the number of participants authorising the incremental investment when it represents a badprospect with negative expected value (Experiment One) and when it represents a good prospect with positive expectedvalue (Experiment Two) are 34/64 ¼ 53.1% and 75/110 ¼ 68.2%, respectively. These frequencies are statistically different(p ¼ 0.048) indicating that, even when the net expected value is 0 taking into account the sunk outcome, participants weremore likely to authorise a good incremental project (Experiment Two) than a bad one (Experiment One).

In summary, the additional analyses reported above provide strong evidence to support the view that participants are ableto evaluate the risks present in the decision to authorise the incremental investment, provide insight into the cognitiveprocesses participants adopt when making their decisions, indicating that probability of loss and expected value are stronglylinked to perceived risk.13

7. Conclusions

This paper provides an alternative evaluation of the prospect theory explanation of the sunk-cost effect. Two experimentsare conducted in which participants make incremental investment decisions in the presence of either a sunk cost or a sunkbenefit. The two experiments differ with respect to the type of incremental investment available to participants. In commonwith the approach in prior studies, the incremental investment in Experiment One represents a ‘poor’ proposal witha negative expected value, thus authorising the investment constitutes risk seeking behaviour. However, contrary to thepredictions of both prospect theory and economic theory, the results indicate that individuals exhibit, relatively speaking, anincrease in risk seeking behaviour after a sunk benefit rather than after a sunk cost. This result is robust to changes in theprobability that the incremental investment delivers a positive net cash flow. The incremental investment in Experiment Tworepresents a ‘good’ proposal as it has a positive expected value, but the probabilistic nature of the outcome means thatauthorising the incremental investment is still risky. No effect of the sunk cost/benefit treatment is found, thus observed

13 Further evidence in this regard is provided by the results of a binary logistic regression (not reported here due to space considerations) of the yes/nodecision to authorise against project type (negative versus positive expected value), sunk outcome (cost versus benefit) and probability of good outcome(high versus low). The results confirm that projects with negative expected value and low probability of a good outcome (i.e. high probability of loss) wereless likely to be authorised.

D. Duxbury / The British Accounting Review 44 (2012) 144–156154

behaviour in the presence of a good investment opportunity appears to be in line with economic theory and conventionalaccounting wisdom. However, while the evidence is not as damning as that from Experiment One, it still fails to support theprospect theory explanation of the sunk-cost effect.

In addition to the main result discussed above, there are a number of other interesting features of the results across thetwo experiments that warrant comment. First, while individuals appear to be affected by the sunk cost/benefit manipulationwhen evaluating a poor investment, they appear not to be affectedwhen evaluating a good investment and so appear tomakemore rational choices. Thus in times of adversity, when investment opportunities are poor and good investment decisions areparticularly important, individuals appear to make less rational decisions than in times of prosperity, when investmentopportunities are good and poor investment decisions are not so costly.

Second, Sitkin and Weingart (1995) argue that perception of decision risk is negatively related to risk taking behaviour,thus the lower the perceived risk the higher the risk taken. Following this reasoning, it is possible to gain insight about whatindividuals perceived as risk in the investment scenarios they faced. For both experiments there was a significantly higherlikelihood of authorising the incremental investment in the high probability of a good outcome condition than the lowprobability of a good outcome condition. In both experiments, the high probability condition generated a potential loss thatwas larger than that in the low probability condition (�£17m including initial outlay, compared to �£2m) and also produceda higher variance of outcomes than the low probability condition, but in contrast was associated with a lower probability ofloss (0.2 compared to 0.8 in the low probability condition). The results suggest, therefore, that individuals perceived risk to beassociated with probability of loss, not size of loss or variance of outcomes.

Third, Zeelenberg and van Dijk (1997) investigate the sunk-cost effect and demonstrate that feedback on forgoneoutcomes, with the associated potential to induce anticipated regret, canmodify behaviour. By focussing on probability of lossit is conceivable that individuals in the experiments reported herewere concernedwith the probability of ending upworse offthan theywould have been for certain had they chosen not to authorise (a cash prize value of £20); a situation that would haveclearly prompted regret. Individuals were four times more likely to end up worse off if they authorised than if they did notauthorise in the low probability of a good outcome condition than the high probability condition (probability of loss of 0.8 and0.2, respectively).

A potential limitation of the experimental method employedmay be that participants did not actually experience the prioroutcome, sunk cost or sunk benefit, for real and so it could be questioned whether participants’ reference points wereinfluenced by the treatment manipulation. The results of a manipulation check reported earlier, however, suggest that thetreatment manipulation would have the desired effect on participants’ reference points. In addition, Experiment One reportsa significant effect of the sunk cost/benefit treatment, thus providing clear evidence that the hypothetical nature of themanipulation did not prevent it from impacting upon participants’ reference points, though the change in behaviour was notin the direction predicted by prospect theory. Finally, it is worth noting that many studies, including of course Kahneman andTversky (1979), report behaviour consistent with prospect theory based upon hypothetical decision scenarios, thus there is noa priori reason to question whether participants’ frames of reference, and hence behaviour, could be manipulated bya hypothetical sunk cost (loss) or sunk benefit (gain).

To conclude, if a decisionmaker falls prey to the sunk-cost effect when decidingwhether to commit additional resources toa failing project, they are making decisions on less than rational economic grounds and scarce resources can potentially beallocated inefficiently. Understanding why individuals fall prey to the sunk-cost effect is, therefore, important and will help inthe development of robust techniques aimed at mitigating the effect. Contrary to the predictions of prospect theory, thispaper provides evidence that individuals are not more likely to authorise incremental investments in the presence of a sunkcost than a sunk benefit. In isolation, therefore, prospect theory is unable to explain fully the sunk-cost effect. This view iscorroborated by Friedman, Pommerenke, Lukose, Milam, and Huberman (2007) for whom the sunk-cost effect proves elusiveand whose data provide little evidence of loss aversion. While the findings reported here fail to support the prospect theoryexplanation, it is not to say that it does not have a role to play. Predictions of changing risky behaviour across gains and lossesbased on the S-shape value function are not supported, however, the sunk-cost effect could be seen as a special case relatingto losses only. It may be that prospect theory behaviour under losses coupled with the desire to avoid waste, which cannotmeaningfully be extended to decision scenarios involving sunk benefits, are joint drivers of the sunk-cost effect, though thisremains to be investigated.14

Future research seeking tomitigate the sunk-cost effect should look to other possible explanations of the effect, such as theway inwhich individuals form and usemental accounts. The results in Soman (2001) indicate that when the prior investmentis in the form of time individuals do not fall prey to the sunk-cost effect in the sameway as when the investment is monetary.Soman (2001) provides evidence to suggest this result is due to theway inwhich individuals mentally account for time, whichDuxbury, Keasey, Zhang, and Chow (2005) show differs from how they mentally account for money. Thus, mental accountingeffects, perhaps coupled with prospect theory, may play an important role in the tendency for individuals to succumb to thesunk-cost effect and represent an important avenue for future research, both in terms of causes of the sunk-cost effect andways in which to mitigate its impact on investment and consumption behaviour.

14 It is not immediately apparent how the influence of the prospect theory value function on behaviour in the loss domain can be separated from theinfluence of a desire to avoid waste. 29.

D. Duxbury / The British Accounting Review 44 (2012) 144–156 155

Acknowledgements

I am grateful to Kevin Keasey, an anonymous referee and the joint editors for detailed comments and advice on earlierdrafts of this paper.

Appendix. Example of the decision scenarioYou are the projectmanager of a medium sized engineering firm and are responsible for making investment decisions. You

are considering whether to authorise the investment of additional company funds into an existing project that you previouslyauthorised. The additional investment would prolong the project, which would end otherwise.

The expected incremental cash flows associated with the additional investment are shown below. Note that there are twopossible outcomes if you decide to undertake the additional investment.

Further information

The original project required an initial investment of £3m (cash outflow) and generated a cash inflow of £2m.

Cash flow information for additional investment

£ Million Probability

Immediate cash outflow 2Estimated future cash inflows 5 20%

0 80%

On a rating of 1–7 how likely would you be to authorise investment in this project (where 1 ¼ definitely authorise and7 ¼ definitely not authorise)? Please indicate on the scale below by circling the number corresponding to your decision.

1 2 3 4 5 6 7

Definitelyauthorise

Highly likelyto authorise

Likely toauthorise

Undecided whetherto authorise or not

Unlikely toauthorise

Highly unlikelyto authorise

Definitely notauthorise

References

Arkes, H. R., & Blumer, C. (1985). The psychology of sunk cost. Organizational Behavior and Human Decision Processes, 35, 124–140.Arkes, H. R., & Hutzel, L. (2000). The role of probability of success estimates in the sunk cost effect. Journal of Behavioral Decision Making, 13, 295–306.Ashton, R. H., & Kramer, S. S. (1980). Students as surrogates in behavioral accounting research: some evidence. Journal of Accounting Research, 18, 1–15.Bouwman, M. J. (1984). Expert vs novice decision making in accounting: a summary. Accounting, Organizations and Society, 9, 325–327.Cheng, M. M., Schulz, A. K. D., Luckett, P. F., & Booth, P. (2003). The effects of hurdle rates on the level of commitment in capital budgeting. Behavioral

Research in Accounting, 15, 63–85.Chow, C. W., Harrison, P., Lindquist, T., & Wu, A. (1997). Escalating commitment to unprofitable projects: replication and cross-cultural extension.

Management Accounting Research, 8, 347–361.Denison, C. A. (2009). Real options and escalation of commitment: a behavioral analysis of capital investment decisions. The Accounting Review, 84, 133–155.Duxbury, D., Keasey, K., Zhang, H., & Chow, S. (2005). Mental accounting and decision making: evidence under reverse conditions where money is spent for

time saved. Journal of Economic Psychology, 26, 567–580.Frederick, D. M. (1991). Auditors’ representation and retrieval of internal control knowledge. The Accounting Review, 66, 240–258.Friedman, D., Pommerenke, K., Lukose, R., Milam, G., & Huberman, B. A. (2007). Searching for the sunk cost fallacy. Experimental Economics, 10, 79–104.Garland, H. (1990). Throwing good money after bad: the effect of sunk costs on the decision to escalate commitment to an ongoing project. Journal of

Applied Psychology, 75, 728–731.Ghosh, D. (1997). De-escalation strategies: some experimental evidence. Behavioral Research in Accounting, 9, 81–112.Harbaugh, W. T., Krause, K., & Vesterlund, L. (2002). Risk attitudes of children and adults: choices over small and large probability gains and losses.

Experimental Economics, 5, 53–84.Harwood, G. B., Pate, J. L., & Schneider, A. (1991). Budgeting decisions as a function of framing: an application of prospect theory’s reflection effect.

Management Accounting Research, 2, 161–170.Kahneman, D., & Tversky, A. (1979). Prospect theory: an analysis of decision under risk. Econometrica, 47, 263–291.Kanodia, C., Bushman, R., & Dickhaut, J. (1989). Escalation errors and the sunk cost effect: an explanation based on reputation and information asymmetries.

Journal of Accounting Research, 27, 59–77.Libby, R., Bloomfield, R., & Nelson, M. W. (2002). Experimental research in financial accounting. Accounting, Organizations and Society, 27, 775–810.Lindman, H. R. (1974). Analysis of variance in complex experimental designs. San Francisco: W. H. Freeman & Co.Liyanarachchi, G. A., & Milne, M. J. (2005). Comparing the investment decisions of accounting practitioners and students: an empirical study of the

adequacy of student surrogates. Accounting Forum, 29, 121–135.McAulay, L., King, M., & Carr, C. (1998). Expertise and learning: an alternative perspective on the theory–practice debate. The British Accounting Review, 30,

173–200.

D. Duxbury / The British Accounting Review 44 (2012) 144–156156

McCain, B. E. (1986). Continuing investment under conditions of failure: a laboratory study of the limits to escalation. Journal of Applied Psychology, 71,280–284.

Moreno, K., Kida, T., & Smith, J. F. (2002). The impact of affective reactions on risky decision making in accounting contexts. Journal of Accounting Research,40, 1331–1349.

Navarro, A. D., & Fantino, E. (2008). The sunk-time effect: an exploration. Journal of Behavioral Decision Making, 22, 252–270.Northcraft, G. B., & Neale, M. A. (1986). Opportunity costs and the framing of resource allocation decisions. Organizational Behavior and Human Decision

Processes, 37, 348–356.Payne, J. W. (1975). Relation of perceived risk to preferences among gambles. Journal of Experimental Psychology: Human Perception and Performance, 1, 86–94.Remus, W. (1986). An empirical test of the use of graduate students as surrogates for managers in experiments in business decision making. Journal of

Business Research, 14, 20–30.Schulz, A. K. D., & Cheng, M. M. (2002). Persistence in capital budgeting reinvestment decisions – personal responsibility antecedent and information

asymmetry moderator: a note. Accounting and Finance, 42, 73–86.Shapira, Z. (1995). Risk taking: A managerial perspective. New York: Russell Sage Foundation.Sharp, D. J., & Salter, S. B. (1997). Project escalation and sunk costs: a test of the international generalizability of agency and prospect theories. Journal of

International Business Studies, 28, 101–121.Sitkin, S. B., & Weingart, L. R. (1995). Determinants of risky decision-making behavior: a test of the mediating role or risk perceptions and propensity.

Academy of Management Journal, 38, 1573–1592.Slovic, P., & Lichtenstein, S. (1968). Relative importance of probabilities and payoffs in risk taking. Journal of Experimental Psychology, 78, 1–18.Soman, D. (2001). The mental accounting of sunk time costs: why time is not like money. Journal of Behavioral Decision Making, 14, 169–185.Soman, D., & Cheema, A. (2001). The effect of windfall gains on the sunk-cost effect. Marketing Letters, 12, 51–62.Staw, B. M. (1976). Knee-deep in the big muddy: a study of escalating commitment to a chosen course of action. Organizational Behavior and Human

Performance, 16, 27–44.Staw, B. M. (1981). The escalation of commitment to a course of action. Academy of Management Review, 6, 577–587.Tan, H., & Yates, J. F. (1995). Sunk cost effects: the influence of instruction and future return estimates. Organizational Behavior and Human Decision Processes,

63, 311–319.Tan, H., & Yates, J. F. (2002). Financial budgets and escalation effects. Organizational Behavior and Human Decision Processes, 87, 300–322.Thaler, R. (1980). Towards a positive theory of consumer choice. Journal of Economic Behavior and Organization, 1, 39–60.Thaler, R., & Johnson, E. J. (1990). Gambling with the house money and trying to break even: the effects of prior outcomes on risky choice. Management

Science, 36, 643–660.Whyte, G. (1986). Escalating commitment to a course of action: a reinterpretation. Academy of Management Review, 11, 311–321.Whyte, G. (1993). Escalating commitment in individual and group decision making: a prospect theory approach. Organizational Behavior and Human

Decision Processes, 54, 430–455.Wilson, R. M. S., & Zhang, Q. (1997). Entrapment and escalating commitment in investment decision making: a review. The British Accounting Review, 29,

277–305.Yates, J. F., & Stone, E. R. (1992). The risk construct. In J. F. Yates (Ed.), Risk-taking behavior (pp. 1–25). Chichester: John Wiley & Sons.Zeelenberg, M., & van Dijk, E. (1997). A reverse sunk cost effect in risky decision making: sometimes we have too much invested to gamble. Journal of

Economic Psychology, 18, 677–691.