The Sunk Cost Effect Examined
-
Allard ten Hoopen – 0296996
Datum: 22/11/2007 FEB
UvA
Algemene Economie
Introduction
Sunk costs are expenses, either in money effort or time that have been made and cannot be recovered. According to traditional economic theory such cost should be ignored in a further decision- making process. However, often in the presence of sunk costs decision makers seem to make sub-optimal decisions. This influence of sunk cost on the decision making process is called the sunk cost effect.
Imagine a poker player who knows he is unlikely to win the game but still raises his bet because he has already put lots of money in the pot. Or think of a political leader who refuses to stop fighting a loosing war because so many already gave their lives for the cause. Both examples seem to indicate the presence of a sunk cost effect, as in neither case the decision to commit more resources is likely to change the outcome in a positive way, but still this decision is made. It seems then that the sunk cost effect can sometimes cause a decision maker to make the irrational decision to pursue a course of action that has negative expected utility, as a result of the resources already committed.
Why would a decision-maker make such an irrational decision? This is a question that is often somewhat overlooked in economic literature, but is central in psychological studies. Economists on the other hand focus more on showing whether or not the sunk cost effect actually exists. This paper will therefore focus on the following questions: Firstly, how can sunk cost motivate people to undertake an irrational course of action? And secondly, can the sunk cost effect be shown to exist empirically?
The rest of this paper will organized as follows. The first paragraph will focus on explaining the processes at causes the sunk cost effect. The second paragraph will review the empirical evidence concerning the sunk cost. The third paragraph will interpret the results found in the second paragraph. The fourth paragraph will hold the conclusions.
The sunk cost effect: how does it work?
This section will investigate the underlying reason behind the sunk cost effect. We know that people might take sub optimal decisions in the presence of sunk cost, but why? Staw (1981) gives several explanations for the sunk cost effect, or as he calls it ‘escalation of
The most straightforward explanation is that people are simply unable to determine what the optimal decision is due to limitations in individual information processing (Staw, 1981, 578). Put simply, a person cannot oversee all factors relevant to the decision taking process which may in turn lead to sub optimal outcomes. Of course this would be a proble m in any decision taking process, not just one that involves sunk cost. As such this solution does explain decision-taking mistakes in general, but not how these might be made worse (or better) by sunk cost and for this reason it does not really explain the sunk cost effect.
Self- justification seems to be a more suitable explanation for the sunk cost effect. Staw (1981, 580) distinguishes between two kinds of self- justification: internal and external self-justification. Internal self- justification can be translated as the protection of self-esteem. A person likes to perceive himself as competent and rational. To maintain this image of competence the person might engage in justification his actions (Whyte, 1986, 313). In the face of a failing course of action this could mean that this person further increases his
commitment in order to salvage the situation and prove that the original course of action was in fact rational. Feelings responsibility for a negative outcome might also lead to internal self-justification. A person feeling responsible for the (negative) outcome of his decisions might try to correct this, again by further escalating commitment.
External self-justification explains the breakdown of rationality as a result of group dynamics (Staw, 1981, 578). In this theory, the need to justify previous decisions to other members of a group is sometimes stronger than rationality. A decision taker will want to appear as a good, rational leader. Apparently one characteristic associated with rational leadership is consistency, so a decision may not only be judged on the decision at hand but also on how this decision relates to those that were made earlier. This results in some courses of action being less attractive to the decision taker because they would constitute a deviation from previous decisions and thus would have consequences for his reputation or image (Staw, 1981, 584). This could lead to the rather ironic circumstances where a decision taker, in his desire to appear rational, acts in irrational ways. Perhaps an example can clarify this further:
A manager in a firm has heavily promoted a particular project and considerable amounts of money have already been spent on it. Still it has never made any profit and doesn’t seem likely to do so in the futur e. The manager knows cancellation of the project would be the best option but instead he keeps allocating additional funds. This course of action is explained by the fact that the decision to cancel the project would cast doubts on the decision to set up the project in the first place,
damaging the manager’s reputation, in particular with his superiors. The manager therefore sees no other option than trying to salvage the project at even greater cost.
According to the above interpretation, the sunk cost effect may seem to be a part of the principal-agent problem where the agent can use the existence of asymmetric information to pursue his own interests at the cost of the principal. In the case of the sunk cost effect the information asymmetry would be in the knowledge about what course of action is rational and what is not. However, while the sunk cost effect can occur in the setting of the principal agent problem the two phenomena are quite different because the sunk cost effect is not restricted to the domain of principal-agent relationships. A decision taker who has no responsibility
towards anybody but still feels the desire to be considered rational by others would not be part of the principal agent problem, but still could be under the influence of a sunk cost effect.
How people behave when facing sunk cost is, explained by Whyte (1986) with the use of prospect theory. Prospect theory describes how people evaluate risky choices. Key
assumption under this theory is that people evaluate their decisions relative to some reference point and not on the total change of wealth (Whyte, 1986, 315). Outcomes below the
reference point would be considered losses, while outcomes above the reference point are considered as gains. The theory does not determine what this reference point is. It is therefore theoretically possible that an outcome that for one person is considered a loss is considered a gain for another (and vice versa). What is important is that the reference point does not have to be the wealth at the moment the current decision is taken, it can also be the wealth before earlier decisions were made.
A second property of prospect theory is that the certainty effect is assumed to affect the valuation of gains and losses (Whyte, 1986, 315). This psychological effect holds that a certain gain is valued higher than a possible gain, while a certain loss is valued lower than a possible loss. The result of this effect is that people would be risk averse with respect to gains and risk seeking with respect to losses. In other words, when given the choice between to options with equal expected value a person will prefer the option that has the more certain payout when this outcome is perceived as a gain. When on the other hand the outcome is perceived to be a loss the, person will prefer the option that has lower probability of loss, but of course this option will also have a higher potential total loss.
Now how can prospect theory explain the sun cost effect? The key is that there must be a choice between a certain lo ss and a loss that is only possible. Suppose that the starting
capital at the beginning of a project serves as the reference point. The project does not go well and losses are incurred. A decision to stop the project would mean a certain loss relative to the reference point. Another option is to continue the project, which requires additional
investment but also gives a chance to minimize losses relative to the reference point. Given that people are risk seeking with respect to losses, the expectation would be that the more risky option is preferred even if the certain option would a yield considerably higher expected payoff. Under these conditions prospect theory explains how more resources are committed than would be expected under traditional assumptions of rationality. Apparently the sunk cost effect is generated by a desire to avoid losses that is stronger than rationality.
This overview wouldn’t be complete without the theory that appears in Arkes and Blumer (1985). They subscribe prospect theory but add a psychological explanation to it based on ‘the desire not to appear wasteful.’ According to this theory, people are more likely to undertake a sub optimal action provided that action could lead to avoiding ‘wasting’ the resources already spent. Interestingly a connection can also be made to self- justification. Staw suggests that the demand for consistency drives external self-justification, however aversion of wastefulness might well be another motive as it seems likely that a leader wouldn’t want to appear wasteful any more than to appear inconsistent.
The theories on the sunk cost effect that were discussed above don’t have be to mutually exclusive. They may in fact be closely connected. In this interpretation prospect theory describes how people act when faced with sunk cost, but by itself doesn’t explain why they do so. People can be loss averse, the can dislike being wasteful or they might want to appear rational. On a theoretical basis it is difficult to say what the true motivations are, but most likely they will vary from case to case.
Such an integrated theory doesn’t seem to have been formulated yet, but it seems present implicitly in studies like Garland and Newport (1991). Whyte (1986, 316) suggests that self- justification might influence adaptation to losses and can have an influence on the position of the reference point as described by prospect theory. However, Whyte does not elaborate any further on this line of thought to create a formal theory.
The sunk cost effect: is it real?
After having examined the theories about what might cause the sunk cost effect it is time to see if the effect actually can be shown to exist empirically. The five studies discussed below
have attempted to do just that, but the results are not without contradic tion. Some claim to have found a strong sunk cost effect, others can only find a relatively weak effect. And then there are some studies that hardly find any effect at all. This section will compare these studies by means of a short description of the experimental setup and giving the most important results.
First up is Effects of Absolute and Relative Sunk Costs on the Decision
to Persist with a Course of Action, a study by psychologists Garland and Newport (1991).
This study distinguishes the relative and the absolute sunk cost effect. The underlying idea is that the sunk cost effect depends not on the absolute size of the sunk cost but on the size of the sunk cost relative to the original budget of the project. Assuming that people use some reference point to evaluate decisions resembles prospect theory and indeed the authors claim that their theory is consistent with prospect theory (Garland and Newport, 1991, 59). The theory of Garland and Newport therefore could be regarded as a more specific form of prospect theory. Garland and Newport state that their most important goal is to investigate how the ‘decision to continue an investment in a questionable course of action’ is influenced by both absolute expenditures and the proportion of the budget expended (Garland and Newport, 1991, 59). An interesting aspect of this study is the attention given to framing, or the way the decisions are presented to the subjects. Based on Staw’s theory of external justification, some cases are presented as personal decisions, while others are framed as business decisions.
Garland and Newport (1991, 60) ran their experiment in two sessions, once with 88 students and once with 30 MBA. The subjects were required to consider four cases, two framed as business decisions and two as personal decisions. In total there were four treatments that used the same four scenarios except that each treatment varied in the sizes of the absolute and relative sunk costs. For each scenario the subjects had to indicate their likelihood of continuing on the described course of action on a scale of 0 to 100.
Garland and Newport (1991, 65) find only one significant effect on the probability of continuing with a course of action and that is the relative size of the sunk cost, so absolute size had no significant effect and neither had the scenario. As they got almost identical results in both sessions Garland and Newport feel very confident about the result. The experiment indeed seems to provide some powerful evidence for the relation between the sunk cost effect and the relative size of the sunk cost, however interpretation is tricky. Subjects are only asked about their probability of continuing, actual behavior is not observed. It is therefore possible
that when asked to actually take the decision whether to continue or not, the subjects would behave differently than implied in the results found by Garland and Newport. However, if true, these findings have important implications. It would mean that sunk costs themselves don’t influence decision taking unless they are sufficiently large compared to some reference point. It would also mean that it could be expected that no sunk cost effect will be found when the reference point is regularly reset in the decision takers mind.
That no effect of framing is found is not too surprising given the experimental design. The scenarios are supposed to present cases as business or personal decisions. But it just seems unlikely that the subjects sitting in a classroom will really identify with the story they are told. It is implied that the expenditures already made are the subjects’ responsibility, but in reality the subject never made such a decision (Garland and Newport, 1991, 67). So why would they feel any responsibility? It might have been better to let the subjects actually make the initial decision, but without the subject knowing this, setting it up in such a way that the outcome always will be negative. This would ensure that subjects see the initial decision as there own which might lead to different results. Also this study lacks any financial incentives to make to right decisions.
The Psychology of Sunk Cost by Arkes and Blumer (1985) describes ten experiments on the
sunk cost effect, the most interesting of these being experiment 2 because it is a field study instead of a laboratory setting (Arkes and Blumer, 1985, 127). The experimenters sold (unknowing) subjects season tickets to the university theater at various prices. They then measured how often the subjects would go to the theater over the season. If the subjects would adhere to economic theory they should on average all go to the theater the same number of times as preferences on visiting the theater should be randomly distributed. However, Arkes and Blumer find that the subjects with more expensive tickets visit significantly more often, indicating a sunk cost effect. What is particularly striking about the result is that the effect was observed to work over several months and that the size of the sunk cost was quite small, only $15 for the most expensive ticket.
The other experiments are mainly interesting for the effect of framing. For example, in experiment 3 subjects are asked to decide whether or not continue investment in an aircraft in which large investments have already been made but that has a superior competitor (Arkes and Blumer, 1985, 129). A strong sunk cost effect is found here. In experiment 8 subjects basically have the same decision to make, but the question is subtly changed to place less
responsibility with the subject (Arkes and Blumer, 1985, 135). The sunk cost effect is much weaker in this experiment.
In Sunk and Opportunity Costs and Valuation of Bidding Phillips et al. (1991) investigate the sunk cost effect in valuing lottery tickets and at auctions. In their first set of experiments the authors ask the subjects to value lottery tickets from a set of tickets that when selected will cost different amounts but have identical payouts. One ticket than is randomly selected. It should be noted that paying for the selected ticket can not be avoided so that at the moment the valuation is made the costs of the ticket price have already been sunk and thus should not influence the value. This means that all tickets should be valued the same. Any deviation from this would indicate a sunk cost effect. The above treatment is a one-shot game, but to allow for learning effects a second experiment is run where a very similar game is played for several rounds.
The authors believe that the sunk cost effect might disappear in a market environment with learning opportunities (Phillips et al., 1991, 114). This is based on the conjecture that in competitive market subjects who fall for the sunk cost fallacy will be forced out of the market or learn to avoid it in later rounds. Therefore another experiment is performed in an auction setting. They investigate whether or not subjects’ bids are influenced by having to pay a fee before entering a market as compared to a baseline setup where no entry fee is required. This game was repeated over multiple periods.
As the authors expect, the sunk cost effect becomes less pronounced as they go from the one-shot game to the repeated game and then to the market setting. They believe this might be explained by the increasing transparency of the later games, he lping subjects to ignore sunk cost (Phillips et al., 1991, 117). Still even in the market game a (small) sunk cost effect is observed. This study provides evidence that a sunk cost effect does indeed exist, particularly in an out of market setting and in one-shot games.
Next is Searching for the Sunk cost Fallacy by Friedman et al. (2004). In this study subjects play in a treasure hunt scenario. Each subject gets a budget of 200 points that can be used to dig for treasure. The treasure hunting takes place on islands with 20 dig sites. On every island between 2 and 18 treasures can be found (Friedman et al., 2004, 7). When subjects decide they are done on one island they can move to the next. Traveling to and from islands is costly and these costs are randomly assigned to be either high or low. Upon reaching a new island these travel costs are sunk and should not influence decisions on how many sites are going to
be explored. However, if there is a sunk cost effect, the authors expect to find that subjects will continue digging longer high cost islands than on low cost ones, despite this being sub-optimal (Friedman et al., 2004, 2). If this turns out to be the case that would be evidence for the existence of the sunk cost effect.
The experiment includes nine different treatments to test for different hypotheses (Friedman et al., 2004, 8). In one treatment information is given on how many treasures are on the island. In another treatment subjects can’t decide at which sites to dig but only whether or not they dig up all 20 sites on the island (costing 20 points) or move on to the next island. Some variations are designed with the theoretical explanations for the sunk cost effect in mind. The treatment where subjects can chose to travel north or south is designed to capture the sunk cost effect if caused by self-justification. To find the effect if caused by risk
aversion, a treatment exists were a thermometer indicates the net cumulative points earned on the island. This indicator than might stimulate a sunk cost effect by influencing the subjects’ reference point. There is also a treatment that has a smaller budget than the baseline treatment. If the sunk cost effect is correlated with the relative size of the sunk cost to the original
budget, an effect would be expected here.
To their amazement the authors find only weak indication of a sunk cost effect for any of their hypotheses (Friedman et al., 2004, 20). A statistically significant effect is found with respect to sunk cost size in some cases. Subjects do display a higher probability to stay on an island longer is the cost of getting there was high, but the effect seems to be rather small (Friedman et al., 2004, 19). It is however an indication that relative size of the sunk costs is important, in line with the findings by Garland and Newport (1991).
Of the psychological explanations, self-justification produces only inconsistent and mostly insignificant effects (Friedman et al., 2004, 19). This is remarkable as the experiment seems to be setup specifically with these explanatory theories in mind, maximizing chance of finding an effect. Note that in any event, external self- justification effects cannot be found using this experimental setup, as there is no group interaction, but that internal
self-justification yields so little effect is important. The subjects do make the initial decision (with likely negative outcomes) themselves, which increases personal involvement in the decision and makes it more likely for self- justification to come into play. Nevertheless hardly any effect is found, yielding some strong evidence that at least internal self-justification is no important factor in the sunk cost effect.
Variables designed to capture risk aversion do not do much better at explaining the sunk cost effect. Again only very weak or insignificant results are found. What is interesting
though is that the relative size of the sunk cost does seem to matter. Garland and Newport interpret the same finding, using prospect theory, as an indication that people are in fact risk averse. This makes the result very puzzling. It is therefore disappointing that Friedman et al. don’t devote more time to analyze the treatment that varies budget size.
The last study discussed here is Licences and Collusion (Offerman and Potters, 2006). This paper investigates whether or not higher market prices will be charged when the license to enter that market have to be bought, as compared to when the license are given away for free. Economic theory of course suggests that there shouldn’ t be any difference because the cost of acquiring the license is sunk when the market price is being decided. If there is one profit-maximizing price, that’s the price that should be chosen regardless of whether or not an entry fee was paid (Offerman and Potters, 2006, 770).
To test if economic theory is right in this case an experiment was setup with different three treatments. In the baseline treatment two subjects out of a group of four are randomly selected. These two subjects then play a duopoly price setting game, where players’ profits depend both on the own price chosen and on the price of the other player. This price setting game is repeated with the same players for five rounds. The other two treatments are the same except for the way the players enter the market. In the auction treatment all four subjects bid for the right to enter the market. In the fixed cost treatment two subjects were randomly selected to pay an exogenous entry fee. To make the data more comparable the fees in the fixed cost treatment were made to match the outcomes of the auction treatment (Offerman and Potters, 2006, 775).
Two mechanisms are suggested how sunk cost could influence prices. Firstly it is argued that firms use markup pricing. They simply calculate total costs and then ad a certain percentage of that price as profit margin. As an entry fee increases total cost the market price would be higher as well. Alternatively auctioning licenses might select those players that expect to be most able to engage in collusion and thus be able to make greater profits than could be expected in either the competitive or the Nash-equilibrium (Offerman and Potters, 2006, 775). Trying to collude does make the game more risky as the other player can defect by setting a lower price to take over the market.
The results confirm that prices are indeed higher when entry fees must be paid (Offerman and Potters, 2006, 776). Average prices per round are almost always significantly higher for the fixed cost and auction treatment compared to the baseline treatment. This is a clear indication there is a sunk cost effect in this setting.
The prices of the fixed cost and auction treatment move very closely together. This suggests that there is no selection of more collusive players in the auction treatment (Offerman and Potters, 2006, 776).
In addition to the above experiment a monopoly price setting game was conducted. It had both a baseline and auction treatment similar to the previous experiment, but only one player was selected to enter the market. This experiment indicated there was no difference in average prices between the baseline and auction treatments (Offerman and Potters, 2006, 789), refuting the conjecture of markup pricing. In fact prices were fairly close to the theoretical optimum. Apparently subjects were able to ignore sunk costs in this experiment, but perhaps only for the lack of any risky alternative with possible higher payoffs.
The above results indicate that people do tend towards risky strategies in the face of sunk costs if such a strategy is available. According to Offerman (personal interview), loss aversion is best explanation for the observed behavior.
The sunk cost effect: explained?
What can be learned from the empirical studies discussed above? If the question simply is: ‘does the sunk cost effect exist?’ then the answer must be yes. All of the empirical studies did at least find some sunk cost effect in some way or another albeit sometimes very weak. It is clear then that behavior can be influenced by sunk cost. Much less clear however, is how the sunk cost effect actually works. This is mainly due to the fact that most studies do not include tests to differentiate between the theoretical explanations. The only one that does (Friedman et al., 2004) can at best only find small effects for the variables associated with the various psychological drivers. Therefore, based on these studies, it is not possible to determine what psychological mechanism is really causing the sunk cost effect beyond what intuition might tell us. At best one explanation can appear more appropriate for the particular situation at hand, but there is no way that this can be proven given the available information.
Still some general observations can be made based on the data the literature does provide. As a rule these findings are not reproduced in other studies but for lack of any counterevidence they might be tentatively accepted. Arkes and Blumer (1985) show the importance of framing. By subtle changes in the setting they created significant differences in outcome. Both Garland and Newport (1991) and Friedman et al. (2004) show that the relative size of the sunk cost relative to the budget does matter. This finding is consistent with what
would be expected under prospect theory. Phillips et al. (1991) indicates that people can learn from bad outcomes and that observation of the sunk cost effect is therefore less likely under market conditions. From Offerman and Potters (2006) can be learned that subjects don’t choose the risky policy by accident. They were in fact able to determine the optimal policy quite well, but when faced with sunk costs preferred a riskier policy.
Now some observations on methodology. As already noted most studies lack a way to identify exactly what mechanism is causing the sunk cost effect. Because of this these studies only succeed in showing that a sunk cost effect does exist. The explanation that the authors may give for such findings can really only be considered conjectural as it cannot be proven that the explanation given works better than the other theories. The only study that does it do right in this regard is the one by Friedman et al. (2004) which does have separate treatments for the different theoretic explanations. What is also quite striking is that no study makes any attempt to include any group effects. This is surprising, as external self- justification
intuitively seems to be a powerful motivation for engaging is risky strategies making it an interesting topic for further research.
Conclusion
The sunk cost effect has been shown to exist in every study discussed. There can therefore be little doubt that the effect is real. This is a significant result for economists who generally assume that the sunk effect does not occur. Fortunately for the economists it seems that the effect is not very strong under market conditions.
The exact working of the sunk cost effect remains uncertain. Three theories explaining the sunk cost effect exist. These are self- justification, prospect theory and aversion of
wastefulness. However these cannot be considered totally separate (in fact aversion of wastefulness is defined as an extension of prospect theory by the authors themselves), but an overall theory combining all three theories is lacking. In addition the empirical evidence is disappointingly weak on this subject. Further research into the exact workings of the sunk cost effect is therefore necessary.
References
Arkes, H.R. and Blumer, C (1985). The Psychology of Sunk Cost, Organizational Behavior
and Human Decision Processes, 35, 124-140.
Friedman, D., Pommerenke K., Lukose, R., Milam, G., Huberman, B., (2004). Searching for
the Sunk Cost Fallacy, 1-33.
Garland, H. and Newport, S. (1991). Effects of Absolute and Relative Sunk Costs on the Decision to Persist with a Course of Action, Organizational Behavior and Human Decision
Processes, 48, 55-69.
Offerman, T. and Potters, J. (2006). Does Auctioning of Entry Licences Induce Collusion? An Experimental Study, Review of Economic Studies, 73, 769-791.
Phillips, O.R., Battalio, R.C. and Kogut, C.A. (2001). Sunk and Opportunity Cost in Valuation and Bidding, Southern Economic Journal, 58(1), 112-128.
Staw, B.M., (1981). The Escalation of Commitment to a Course of Action, The Academy of
Management Review, Vol. 6, No. 4., 577-587.
Whyte, G., (1986). Escalating Commitment to a Course of Action: A Reinterpretation, The
