University of Amsterdam, Faculty of Economics and
Business
Behavioral Economics and Game Theory Master Program
The behavioral sunk cost
effect revisited
Master Thesis
Eleni Papadopoulou (Student number: 10038450)
Supervisor:
Gönül Doğan
2 CONTENTS
1. Introduction……….. 3
2. Literature Review………. 5
2.1Related Work on the Sunk Cost Effect……….. 5
2.2 Behavioral Sunk Cost Literature………. 6
2.3 Experimental methods in previous research………. 8
3. Experimental Design……… 9
4. Hypotheses……… 12
5. Results………... 13
6. General Discussion and Conclusions………. 16
6.1. Summary of research and interpretation of results……… 16
6.2. Contributions………. 18
6.3. Limitations………. 18
6.4. Discussion and directions for future research……….. 19
7. References……….. 20
3 1. Introduction
The sunk cost effect is a classic topic that continues to be the reason why individuals keep making bad resource management decisions and why they find it very hard to ignore the resources they spent in the past. It is manifested in a greater tendency to continue an endeavor once an investment in money, effort, or time has been made (Arkes& Blumer, 1985). There are various decisions where the sunk cost effect can take place, from managerial and economic contexts to all sorts of everyday-life situations; all decisions that involve prior investments in money, time or effort which are irretrievably spent are exposed to this fallacy. Especially the behavioral sunk costs (i.e. time or effort) are often underestimated because they are harder to accurately measure and usually perceived as less important than money investments, but they can also lead to irrational decisions like the monetary sunk costs do. The strength of this effect has not been estimated in such situations and relating the amount of effort sunk in a certain course of action to the decision to keep investing or not will deepen our understanding of the behavioral aspects of the sunk cost effect. Therefore, this research focuses primarily on the investigation of the following questions: How much does the sunk cost effect affect a decision to continue with the current course of action when the decision-maker deals with accumulated behavioral sunk costs? And is the effect stronger or weaker in comparison to behavioral sunk costs that are unrelated to any monetary value, the so-called purely behavioral sunk costs?
The sunk-cost effect has been observed in both business and personal decisions and it can involve both investments in money as well as investments in time or effort. It is a robust judgment error that occurs in all kinds of decisions, from issues related to interpersonal relations or change of career orientation to simpler everyday matters as, for example, whether or not to keep watching a terrible movie just because you have already spent an hour watching it. The matter of time and effort as considerable costs should not be under-estimated or ignored as it is equally important to the monetary investments ; Research into behavioral-resources allocation documents that people recognize the costs invested in cognitive tasks.
However, previous research has failed to successfully replicate the sunk cost effect when investigating behavioral investments, except when such investments relate to monetary values (Devoe & Pfeffer, 2007; Heath, 1995; Soman, 2001; Zeelenberg & van Dijk, 1997). My motivation is to provide a suitable experimental framework that can replicate the sunk cost effect in terms of effort and with real, and not induced, accumulated costs that is missing from the existing literature. Furthermore, this work’s contribution to the literature is mainly methodological. I am interested in examining the sunk cost effect in a direct experimental environment with an appropriately designed game without sticking to the usual hypothetical-scenario questionnaires that are primarily used in previous relevant experiments.
In order to investigate the sunk cost effect in time and effort let’s consider the hypothetical situation where an individual has the ambitious goal to climb Mount Everest. Apart from the monetary costs that such a goal implies which we will overlook for the purposes of this paper, we will focus on the time and effort that should be invested given that they
4
constitute a considerable cost. After having invested a lot of effort and time climbing, she reaches the first base-camp and prepares herself for the next one. Deciding whether or not she should continue, she keeps investing her time and effort and finally makes it to the second base-camp but, she soon realizes that the weather is getting really unfavorable for the rest of the way. Therefore, on the one hand, she knows that she might have to quit pretty soon but, on the other hand she doesn’t want to go home without having climbed Mount Everest especially after having invested so much time and effort so far. Will she decide to quit or will she take the risk of trying to climb until the end? What does economic theory predict about this kind of decisions and do empirical data confirm this prediction? This situation involves decision making in every stage of the process and constitutes one of many examples where the sunk cost effect might take place. Each decision to be made in each base-camp should be regarded as a new, forward-looking choice, independent of the past, already sunk costs that are related to the situation. Even though there is no correct and wrong choice from an objective point of view in this type of decisions, the conditions for the sunk cost effect to take place appear to be present. This study aims at investigating how this situation could be translated into a direct experimental environment and how the magnitude of effort invested in an ongoing course of action can influence future investment decisions. Moreover, a second objective is to examine if the proposed experimental framework can locate any differences between the presence and the strength of the behavioral versus the purely behavioral sunk costs.
For the investigation of the above objectives I carried out an online experiment in which participants exerted real effort in order to complete all stages of an appropriately designed game. I analyzed the difference among situations where the participants had a chance to win a monetary prize and situations where there was no link to money whatsoever (purely behavioral costs). The first finding of my study was that the behavioral sunk cost effect was successfully replicated proving that it can be demonstrated for both behavioral and purely behavioral investments in a direct experimental setting. As far as the magnitude of effort is concerned, I found that the more the effort people invest in an ongoing course of action, the more risk averse they are, thus the more likely it is to switch their course of action. In addition, the presence of the sunk cost effect in behavioral investments was found to be as strong as in the purely behavioral investments. Last but not least, the most quitters were found to quit in the intermediate stages of the game.
The following part of this paper provides more details on prior research and a literature review in order to have a clearer and global view of the behavioral sunk cost effect and the way mental accounting drives the decision makers to this fallacy. Thereafter, I continue with an extensive description of my experimental design and the hypotheses and, finally, there is a thorough presentation of the results and the conclusions, a discussion on their interpretation, its limitations and also, directions for future research.
5 2. Literature Review
2.1 Related work on the Sunk Cost Effect
A basic principle in economics, that past costs and benefits should be irrelevant to current decisions and that only the incremental costs and benefits should be taken into account by the decision maker, is very often violated. In other words, economic theory fails to predict economic behavior when sunk costs are involved; it dictates that decisions should be made “on the margin”, examining if the marginal benefits which refer to the probability of success times the payoff can compensate for the marginal costs, namely the investment fee. Empirical findings show that there is an irrational tendency, in the psychology literature referred to as escalation of commitment, that people tend to let their decisions be influenced by costs made at an earlier time in such a way that they are more risk seeking than they would be had they not made these costs and thus, their commitment to a certain ongoing project is escalating. The theory of escalating commitment also predicts that subjects should never quit investing once they start investing, leading them to be increasingly willing to invest additional amounts of resources even when it seems as it was a bad investment in the first place.
But how can we explain the sunk cost effect and which are the basic underlying mechanisms of this irrational tendency? One explanation attempted by Thaler (1980) is based on prospect theory (Kahneman & Tversky, 1979) and uses the following two basic features: the loss frame of the value function of prospect theory which refers to the diminishing sensitivity individuals have to losses, and the certainty effect. What the first feature explains is the tendency of a person who has already sunk a cost to be more risk seeking. Compared to the gains frame, a person who is in the loss frame (the prior investment made is seen as a loss), is more likely to make a risky investment by continuing adding funds to the sunk cost. The second feature refers to the fact that absolutely certain gains are greatly overvalued and certain losses are greatly undervalued. That means that if there is a sunk cost dilemma between the choice of a certain loss (stop investing in a project you are close to considering as a waste) and a long shot (throwing good money after bad for the exact same project), the certainty effect favors the latter option because of the undervaluation of the certain losses. According to Zeelenberg (1997), sunk costs may also result in risk aversion and this hypothesis is explained on the basis of the aspiration level of the decision-maker. As Simon (1955) suggested, people usually simplify choices by coding the outcomes as satisfactory if they are above the aspiration level and unsatisfactory if they are below it. Therefore, an aspiration level may also be satisfied not only by risk seeking choices but also by risk averse choices and the famous phrase “Too much invested to quit” (Teger,1980) in certain cases can be rephrased as “Too much invested to gamble” (Zeelenberg,1997).
Cognitive dissonance theory also appears to be related to the sunk cost effect (Festinger, 1957). The definition of cognitive dissonance in the literature has been established as the mental discomfort provoked by trying to believe two mutually contradictory propositions (Aronson & Mills, 1959). More strictly, it refers to the emotional investment people make in a particular belief, their tendency to try to line up belief and behavior when the dissonance between them becomes salient. Somewhat counter-intuitively, beliefs will often adapt to
6
justify behavior, though interventions focused on the disjunction between behavior and belief can “force” the process in the opposite direction. So, as to establish the link between cognitive dissonance and the behavioral sunk cost effect let’s think of the fact that, as numerous studies have shown, once a subject is induced to expend effort on an onerous task, the task is revalued upward (e.g., Aronson & Mills,1959). Such revaluation would probably result in increased willingness to expend further resources on the task compared to the resources which would be willingly allocated by a subject not having made a prior expenditure (Park & Jang, 2014). In other words, because cognitively demanding tasks are undesirable, dissonance arises whenever people engage in effortful tasks. Another account of the sunk cost effect is self-justification (Brockner, 1992; Staw, 1976): if one abandons continuing the investment, then it seems that she admits the previous investment in her responsibility is wrong; to avoid this bad appearance, she has to continue the investment. Conclusively, the sunk cost effect is explained by the self- justification mechanism that is based on the cognitive dissonance theory.
Arkes and Blumer (1985) on the other hand, believe that cognitive dissonance theory adds little to our understanding of the sunk cost effect as it does not give answer to the question of why sunk costs are so difficult to ignore. Alternatively, they propose that the later suffering is provoked by the fear of admitting to one-self that the prior investment was a waste. This admission can be avoided by continuing to act as if the prior spending was sensible, and a good way to foster that belief would be to invest more money into that option than if their prior decision was successful. Aligned to that theory, Arkes and Ayton (1999) proposed that the sunk cost effect is fundamentally grounded in the use of a relatively simple ‘don’t waste’ rule, which can usefully be seen as a kind of abstract goal or heuristic.
Moreover, another possible underlying mechanism behind the sunk-cost effect is maintaining a reputation for finishing what you started. In some cases this mechanism might lead people to be more risk seeking in order to live up to other people’s expectations trying to maintain a reputation that they are reliable. There are certain situations a manager chooses to continue an unprofitable project than to cancel it and be disapproved by its supporters (Milgrom and Roberts, 1992.)
A recent neuroeconomics research sheds light to the neural basis of the phenomenon of the sunk-cost effect (Zeng et al., 2013). They found that increasing sunk cost is associated with certain brain regions that are typically involved in risk taking, suggesting that the sunk cost effect as a matter of fact is closely related to risk taking. On the other hand, smaller incremental cost induced stronger activity in the parts of the brain that are sensitive to rewards. Also, no overlapping areas were found to respond to both incremental and sunk costs and together with the afore-mentioned results lead us to point out the certainty effect over self-justification or the diminishing sensitivity to losses as the primary explanation of the sunk-cost effect.
2.2 Behavioral sunk cost literature
As far as the behavioral sunk costs are concerned, prior research has mainly focused on the difference in mental accounting of the behavioral versus the monetary resources. Thaler in
7
his work on mental accounting (1980, 1985, and 1999) provides a framework that is useful for the demonstration of sunk-cost effect for behavioral resources. He proposes that individuals open a transaction-specific mental accounting on incurring a payment and close that account on the accrual of benefits. Sunk investments set a mental account “in the red”, a fact that escalates commitment to the current course of action in an attempt to close the account “in the black”. In the example of the climbing of mountain Everest, assuming that you regard your time and your effort as considerable costs, the more you proceed the more you invest in the current course of action so your mental account is set “in the red”. The most probable outcome is that you cannot close the account “in the black” if you quit and go home, so you keep going until the end. In the case of the monetary sunk costs the mental accounting works in the way explained above, but for the behavioral sunk costs, booking and tracking process for mental accounting become impaired, so there is not only one possible direction that the mental accounting might lead you to. Behavioral investments are harder to book and track so they might not lead to escalation of commitment for as long the mental account is not set “in the red” and consequently the sunk-cost effect is less likely to be observed (Soman, 2001; Heath, 1995).
Individuals tend to choose the option that presents a higher likelihood of changing the balance of their account from “in the red” to “in the black”. That involves cases where higher behavioral sunk costs lead to more risk seeking behavior but also cases where individuals who invest greater effort being more prone to switch to the safer option when uncertainty is imperceptible. On the latter case, Cunha and Caldieraro (2009) propose that according to a mental accounting perspective, individuals who invest greater effort should be more likely to switch their course of action than those who invest smaller effort.
Soman (2001), on a more time-specialized research, examines the effect of past time investments on current decisions. He demonstrates with a series of experiments that the sunk cost effect is not observed for past investments of time, except for the cases where the investments are expressed as monetary quantities, so he concludes that when the accounting of time investments is facilitated by some monetary links the sunk cost effect reappears. Nonetheless, there is also what is called “duration neglect” in the evaluation of experiences referring to the fact that in retrospective evaluations of an experience the actual duration does not matter (Fredrickson and Kahneman (1993)). Consequently, since duration does not seem to be an important predictor of the evaluation of sunk time costs, the sunk-cost effect is weaker in the domain of temporal costs than in the domain of monetary costs. The sunk cost effects are more likely to occur when resources are framed in terms of money rather than time or effort, perhaps because people are somewhat less capable of mentally keeping track of “expenditures” of the latter compared to the former. As Cuhna and Caldieraro (2009) also observe, in the absence of any contextual cues that facilitate the tracking and accounting of investments, for example when there is no link of time to a monetary equivalent, then in such cases previous research has failed to provide evidence for purely behavioral sunk-cost effects. In their paper they also conclude that an effort-justification mechanism, which is basically a compensation process that can reduce dissonance, can predict the sunk cost effect in purely behavioral investments even though no contextual cues were needed.
8
Last but not least, Heath (1995), in his work about when escalation and de-escalation occurs in response to sunk costs argues that, as the cumulative expenditure on a project approaches the budget limit, individuals reduce their commitment to the project and show a reverse sunk-cost effect. Moreover, the basic rule that he proposes that can help us explain better the appearance or the lack of the sunk cost effect due to behavioral investments is that individuals are only likely to fall prey to the sunk-cost effect when they fail to set a budget or when expenses are difficult to track.
2.3 Experimental methods in previous research
Almost all previous research on the sunk cost effect and the escalation of commitment is based on responses to hypothetical survey questions and scenarios. The only exception is the work of Friedman et al. (2007) that used a game as their experimental tool; however their study reported a rather small sunk cost effect. In the majority of the papers using hypothetical scenarios, serious doubts are raised about the interpretations of the results while skepticism could offer opposite and equally valid interpretations and the issue of a better and more direct experimental tool emerges. For example, in numerous studies subjects are asked to imagine that they spent $50 on a ticket for event A and $100 on a ticket for event B. Their preferences are given, so subjects are asked to imagine that they really prefer A to B, and they are also informed that the events are mutually exclusive and that the tickets have no redeem value. Half of the subjects when asked to choose between A and B chose the more expensive but less preferred option B and psychologist authors interpret this result as a proof that the sunk cost effect can be stronger than “true” preferences. A skeptic could explain the results from a different point of view since the choices are not salient, for instance that subjects attend more to their actual homegrown preferences between A and B, or to the impression they make on the person asking the question (D. Friedman et al., 2007). This example consists one of many that express reservations about the experimental tool used so far when examining the sunk cost effect. It is widely reported in psychological research papers (e.g. , Garland and Newport, 1991) that the sunk cost effect increases in the size of the hypothetical sunk cost, especially in proportional terms .The effect is very sensitive to framing, and is reduced by emphasizing the salience of the incremental costs ( Northcraft and Neale, 1986; Tan and Yates, 1995; D. Friedman et al., 2007). Indeed, Heath (1995) finds a strong reverse sunk cost effect when the sum of incremental and sunk costs would exceed the total benefit. Consequently, it is important to pay specific attention to framing and an efficient way towards that direction could be to avoid the hypothetical scenario method.
In the part of the experimental design that follows I defend in more details my intention to make an experiment that translates the Mount Everest hypothetical scenario into a direct experimental framework.
9 3. Experimental Design
In order to answer my research questions I designed an experiment that deviates from all previous research methods. Instead of asking hypothetical questions I made an experiment that consists of stages where participants are asked to complete a task where they actually invest their time and effort and they are not just asked to imagine that they had. The experimental design consists of three treatments. In the Baseline treatment I look into the behavioral sunk costs and it is used as a baseline for comparing the other two treatments. Participants have to exert effort in each stage to complete a task and there is a monetary prize for those who continue until the last stage of the experiment, so it is a Low Effort- Payment treatment. In the second treatment, which we will refer to as High Effort- Payment treatment, higher behavioral sunk costs are induced and participants have to exert bigger effort in each stage. In this treatment I chose a slightly more difficult task but there is still a monetary link as in the baseline treatment. Finally, in the third treatment, which is the Low Effort - No Payment treatment, the purely behavioral sunk costs are examined. The task is identical to the Low Effort - Payment treatment but it is unrelated to monetary values so there is no monetary prize at the end.
For High Effort-Payment treatment and Low Effort-No Payment treatment the design should be created in such a way that they are both comparable to the Baseline ( Low Effort-Payment), so there should not be more than one changes between each treatment pair. Therefore, the task to be used had to be a task that would appeal to the intrinsic motivation of the participants because there would be no extrinsic incentives in this particular treatment. However, at the same time, it should also be a task that makes participants exert effort and invest their time in it so it should call for their intellectual engagement and not be one hundred percent enjoyable.
Consequently, I designed a game to be played individually by the participants and that embodies the above features. It’s a game that consists of 5 equally difficult stages and in each stage participants are asked to guess a code. Each stage in every treatment is made equally difficult so that the behavior of the participants will not vary depending on the difficulty of each stage and thus the sunk cost effect is clearer. In order to win the game one should combine all the codes from all stages to guess the password, which is a popular song title.
More specifically, the task was the following and I named it “Can you crack the code?” : For the Low Effort- Payment and the Low Effort-No Payment treatment, in each stage, participants had to find a 4-digit code. In the 1st Stage they get a hint about one of the five digits and in the next Stages the hint is the last digit of the previous Stage’s code. When a Stage starts they have to guess the code by typing one to the computer, the computer then will give them feedback about how far their answer is from the correct one and then they can try again by reformulating. The computer will inform them about the correct digits and the correct position of the digits by responding to each code with: a “#” if there is some right digit in the exact right place, a “*” if there is some right digit in the wrong place and a “-” if they are all wrong. Also, a number cannot appear twice in the same code (for example 1770 is not possible). The participants will have to play the game (keep typing codes) until
10
they have the code correctly when they get the message “#### Congratulations”. Note that the order of the feedback characters do not give any information about which digit refers to a “#” and which one refers to a “*”. The “#” always appear before the “*”.
To demonstrate how it works I cite part of the instructions:
For example, let’s say that the code we are looking for in the second Stage is 6458. The hint is that the code contains the digit 5. Every first line is a guess you might make and every second is the computer’s response.
6458
5087
** (which means that two digits are correct but need to change order) 2703
- (which means that none of these digits are contained in the code) 9856
#** (three digits are correct; one of them is in the right place) 9816
**
And you keep going until you get as a response ‘ ####’
The participant has the option to quit before the start of each Stage getting a relevant message as soon as she clears one Stage.
After the end of the 5th Stage, the participant has to combine all the codes to find the winning password which is a popular song title. This part is made in order to add to the intrinsic motivation of the participants with a more holistic regard of the game and not just involve the completion of tasks that have no relation to each other. In this last part of the experiment, the participant has to write all the codes in a row without spaces between them and try to translate the numbers she sees into letters from the alphabet. For example, a part of the number a participant has in front of her might be 20154125 which can be translated into “today”. (20=t, 15=o, 4=d, 1=a, 25=y). The participant receives more specific instructions and a hint when she reaches this point in the game.
For Low Effort-Payment treatment: After a participant has successfully completed the 5th
and final Stage of the game, then she has a 20% chance to be picked to receive the winning prize of 15 euros and this information is given to the participant at the start of the game, in the instructions, where she also has the choice not to participate at all. Therefore, the risky choice in this situation is to keep investing her time because she knows that the odds are against her to win the monetary prize. So, the sunk cost effect should be the force influencing her on deciding to invest more of her time on the basis of the already spent time to complete the first Stages.
For High Effort- Payment treatment: I used a slightly more difficult game:”The higher
effort game”. The same rules apply in this game with the game of High Effort-Payment treatment, but now the code is a 5-digit code. This will imply a bigger effort from the participants’ part. The hint will continue to be one digit and there is still a link to a monetary
11
equivalent. There is a clear effort manipulation between treatments; a 5-digit code requires more thinking and intellectual engagement than a 4-digit code.
After the end of the experiment the participants were asked to complete a small questionnaire for additional information but also with questions that aimed to examine the underlying mechanism driving their choices and also their response to a situation similar to the one they experienced during the experiment but expressed with a hypothetical scenario. The full questionnaire is cited in the Appendix.
The risky choice is clear through the experimental design; the decision maker in every stage deals with the dilemma of quitting or keep going and investing more time and effort. For the treatments that have a link to a monetary equivalent (Low Effort-Payment and High Effort-Payment treatments) there is only a 20% chance to receive payment at the end, therefore the participant is aware that the risky choice is to keep investing more resources in a choice that is highly unlikely to pay off, and the safe choice is to quit now. As the participant completes more and more Stages, then the sunken time/effort costs start to play a role in their decision-making process and the sunk cost effect is present. For the Low Effort-No Payment treatment, that represents the purely behavioral sunk cost treatment, the sunk cost effect is also present in the choice of the decision maker to continue until the end, letting the past investments (in time and effort spent in previous stages) influence the choice to continue or not.
It was important to do the experiment individually for each participant because the presence of more people in the room who might be quitting and leaving early or staying until the end could influence the decisions and the behavior of an individual. I started the experimental sessions at the Amsterdam Science Park library which belongs to the University of Amsterdam, having reserved a private room where each recruited participant would be doing the experiment individually and on his own time. Soon I realized that this is not the best way to proceed as the participants kept asking exactly how much time the entire game takes, and therefore they decided to participate only if they had available to spend exactly the amount of time the experiment should take up, so since the beginning they aimed at completing the whole game, which altered the concept and role of having Stages. I regarded this path as biased as it ruled out the possibility of starting the task for a certain amount of time like a try-out and then decide if they will proceed or not in the next stages, after having sunk a time/effort cost. Consequently, I believe that there weren’t the right conditions for the sunk cost effect to appear, so I decided to change my method. Based on the above reasons, I made the experiment available online in July 2013, and I sent it to students currently studying at different faculties of the University of Amsterdam and also graduates educated in the Netherlands. This way, with an online experiment, I had the opportunity to recruit more participants and more importantly, the afore-mentioned bias was no longer present. The fact that the experiment was available online gave the opportunity to the participants to quit whenever they wanted without feeling guilty that they did not finish the whole experiment, as it was the case in the direct experiment at the Science Park library. The experiment was available online for a certain period of time, one week, and also the participants had a maximum of 2 hours during which they could complete the whole experiment, so this way they had in mind that they shouldn’t interrupt
12
the experimental procedure to do other activities. Moreover, I controlled for cheating as I made 4 different versions of the game for each treatment, namely with different codes to be cracked and different song titles to be found at the end.
4. Hypotheses
The hypotheses to which I based the experimental design are the following:
1. The sunk cost effect is present in all treatments. That means that a large number of participants will choose to continue until the end even though they should not let past behavioral investments influence their current decision to continue or not. The participant comes face to face with this decision after the completion of every Stage, therefore the effort /time spent on the previous Stages is regarded as a sunk time/effort cost. This hypothesis is in line with a large number of studies about monetary sunk costs but it is rarely confirmed in behavioral sunk-cost studies. 2. There are going to be more quitters in the intermediate stage, stage 3. According to
He and MIttal (2007), motivation to commit resources reaches its lowest point during the intermediate stage of a project as the escalation of commitment is no longer influenced by the need for project information or the need for project completion. The former is observed to be present at the beginning of a project, fact that strengthens the escalation of commitment, and the latter is present mostly at the final stages of a project. So, during the intermediate stage, a participant is more likely to quit the game. I am expecting that most of the participants will quit before or after the intermediate Stage 3. The Stages are 5 in total but each participant has 4 chances to quit throughout the entire game, meaning that after Stage 5 it automatically moves to the guessing of the song and does not have an option to quit, as all stages are completed.
3. When comparing the presence of the sunk cost effect with regards to the magnitude of the effort invested, there are no standard expectations. The following pair of competing hypotheses captures this reasoning:
A)The more effort you invest the more risk seeking you are, so I am expecting that participants in the High Effort-Payment treatment will exhibit more the sunk cost effect, so they will continue until the end of the game, staying loyal to the rule “Too much invested to quit”.
B) The reverse hypotheses can also be present: Too much invested to gamble, meaning that the more effort you invest the more risk averse you are. Also, as mentioned in the literature review, individuals who invest greater effort should be more likely to switch their course of action than those who invest smaller effort (Cuhna and Caldieraro, 2009)
13
4. The outcome of behavioral sunk costs versus purely behavioral costs is also controversial in the literature. Therefore we have the following competing hypotheses:
A) Relation to monetary values facilitates the accounting of time and effort so the Low Effort-Payment treatment in comparison to the Low Effort-No Payment treatment will exhibit a stronger sunk cost effect.
B) According to Heath (1995) people escalate commitment when they do not set a budget in certain cases like when information about benefits is not available. My experimental design for all treatments makes information about costs difficult to calculate precisely and plan beforehand so I hypothesize that both the Low Effort-Payment and Low Effort-No Effort-Payment treatments will equally exhibit the sunk cost effect.
5. Results
One hundred and six (106) participants, all of which were between 20 and 33 years-old and were students or graduates educated in the Netherlands, participated in the online experiment. The average time needed to complete all stages from the Baseline (Low Effort-Payment) treatment was 25 minutes, and from the Low Effort-No Payment treatment that was equally difficult was 23 minutes. There was a clear distinction as far as time is concerned from the High Effort-Payment treatment, where the average time was 43 minutes and it shows that the effort manipulation in this treatment was mostly successful. The higher standard deviation for the High Effort-Payment treatment suggests that some participants did not find the game as difficult as others, but clearly more effort was needed in comparison to the other two treatments. The Table below presents the number of participants, the time needed to complete the game, standard deviation of time spent and the percentage of quitters across the three treatments.
14 Low Effort-Payment (Baseline) High Effort -Payment Low Effort-No Payment Number of participants 26 26 49
Average Time needed (in
min) 25 43 23
Std Deviation of time spent 13.9 38.44 22.78 Percentage of quitters 3.8% 23% 6.1% Table 1
The research question seeks answer primarily to whether the effort manipulation in High Effort-Payment treatment had an effect on the risky decision of the participants in comparison to the baseline (Low Effort-Payment). Also, a second basic question refers to whether the decision makers that dealt with purely behavioral sunk costs exhibited more the sunk cost effect in comparison to the baseline.
Graph 1
Presence of the sunk cost effect: We can conclude that Hypothesis 1 is confirmed: the sunk cost effect is present in all treatments, and this result is expressed by the fact that the big majority of the participants chose to continue until the end; 90% of all subjects from all three treatments decided to complete all five stages of the game even though they were aware of the fact that the chance to get paid was very low (for the Baseline treatment and High Effort-Payment treatment) and after realizing that it is a time-consuming activity. Results concerning the magnitude of the effort invested: The difference in proportions of subjects quitting among the second and the baseline treatment was statistically significant (one-sided Fisher’s exact test, p-value =0.049). From the 26 subjects in the Baseline
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Subjects that completed all 5 Stages in %
Low Effort- Payment High Effort-Payment Low Effort-No Payment
15
treatment only 1 chose to quit, whereas in the High Effort-Payment treatment that had the same number of subjects, 6 of them quit. Therefore, Hypothesis 3b is confirmed demonstrating that individuals who invest greater effort are more likely to switch their course of action than those who invest smaller effort.
Behavioral sunk costs versus purely behavioral costs: The Low Effort- No Payment treatment, had 49 subjects and only 3 of them quit. Performing the same test again between the Low Effort-No Payment treatment and the Baseline (Fisher’s exact test) we establish that the difference in proportions of subjects quitting was not statistically significant (Fisher’s exact test, p-value =0.61), consequently these two treatments exhibited equally the sunk cost effect, as hypothesized in Hypothesis 4b.
As far as the stage with the most quitters is concerned, the results indicated that it was the intermediate stage, thus Hypothesis 2 is confirmed as well. From the 10 in total quitters of all treatments combined, 8 of them quit after the completion of Stage 3 (because participants were given the option to quit only after the completion of a Stage, not during), and the other two after Stage 4 and after Stage 2 respectively. Graph 2 below presents the ratio of quitters across the Stages.
Graph 2
Questionnaires:
With a Mann Whitney non parametric test I checked if the amount of fun the participants experienced during the experiment influenced their likelihood to go through all the stages, and the result was not significant (p=0.832). This is considered to be a positive result as, the factor of the sunk cost effect and not the factor of fun should be driving the participants to continue until the end. I also tried to control for the fact that some subjects might have chosen to go through until the end because they felt obliged to in order to help the experimenter with her research. Considering that the participants that answered this question with a 5 (in a 1-5 scale, 5 being the highest) did not regard their time as an investment and their decision to continue did not rely on their own personal criteria, I decided to exclude all such participants from my sample (5 in total). In addition, it is
16
important to see if some of the participants did not regard time as being a rather considerable type of cost in general. A Spearman's Rank Order correlation was run to determine the relationship between the value that participants gave to time as a cost in general and the reported sunk cost effect. There was a moderate, negative correlation between the value of time and the sunk cost effect, which was statistically significant (rs = -0.312, p = .003). The more they estimated their time as an inconsiderable cost, the less they reported to exhibit the sunk cost effect, confirming our hypothesis about the conditions in which the sunk cost effect appears, which is the fact that they regard their time as an investment .
The questionnaire also asked the participants that finished the whole game to state with a number from 1 to 5, the degree to which the time already spent on the previous stages influenced their decision to go until the end. The mean was 3.7 so the average subject reported to have encountered a sunk cost effect of time. A Mann Whitney non parametric test was performed and it can be concluded that there was a statistically significant difference between the High Effort-Payment and the Baseline treatment group's rating of exhibiting the sunk cost effect ( p = 0.025). It can be further concluded that the second treatment with the effort manipulation elicited statistically significant lower self-valuation score on the presence of the sunk cost effect, a fact that is in accordance with the results from the experiment performance. The same test was performed for the Low Effort-No Payment treatment and the Low Effort-Payment but no statistically significant relation was observed. Furthermore, for the participants that quitted, the questionnaire included a question of how much the factor of difficulty of the game influenced their decision to quit, and all of the quitters gave a very low rating as far as difficulty was concerned, indicating that the reason they quit was not related to the difficulty. Conversely, they gave a very high rating on the fact that they didn’t want to make an even bigger investment in time.
Moreover, the vast majority of the participants chose not to quit also in the simulation of the experiment scenario question. Examined with a Fischer’s exact test, there was no statistically significant correlation between the likelihood of quitting in the Baseline Treatment of the experiment and the likelihood of quitting reported by the scenario questions on the same treatment.
6. GENERAL DISCUSSION AND CONCLUSIONS 6.1 Summary of research and interpretation of the results
My objective in this paper was to investigate the sunk-cost effect in situations where the past investment was in time or effort rather than monetary in nature and with an effort manipulation to observe the strength of the sunk cost effect. A second objective was to compare the presence of the sunk cost effect between behavioral and purely behavioral sunk costs. Using an online experiment in which participants exerted real effort in order to complete all stages of an appropriately designed game, I provided empirical evidence for my hypotheses.
17
Overall, my first and more general hypothesis that the sunk cost effect was present in all treatments was confirmed, proving that the sunk cost effect can be demonstrated for behavioral and purely behavioral investments in a direct experimental setting. In addition, the second finding of my study was that the most quitters were found to quit in the intermediate stages of the game, confirming my hypothesis and joining the results of other related studies. As far as the magnitude of effort is concerned, I found support for the hypothesis that the more effort people invest in an ongoing course of action, the more risk averse they are. Last but not least, the presence of the sunk cost effect in behavioral investments was as strong as in the purely behavioral investments. The rest of this section investigates the possible interpretations of the results and their affiliation with relevant results of the literature.
According to the results, it seems that the more effort people invest in an ongoing course of action, the more risk averse they are, most probably because they dislike feeling that they have worked for nothing. This result can be explained by Soman(2001) who argues that the absence of the sunk cost effect occurs due to the inability or unwillingness of individuals to account for time or effort using the same principles as those used with money. However, he found that changing the size of the sunk-time investment did not impact current decision making, which is opposed to my results. His research focuses on mental accounting of time costs exclusively, leading us to think that maybe there is a difference between sunk time and sunk effort investments. On the other hand, Cuhna and Caldieraro (2009) proposed that individuals who invest greater effort should be more likely to switch to the better alternative option than those who invest smaller effort, a prediction that was confirmed in the current study. It seems that as investment in an option increases, willingness to abandon this option decreases. However, if the cost to be sunk is relatively big, then the accumulation of these costs leads to increased willingness to abandon the ongoing course of action.
No significant difference was observed in the presence of the sunk cost effect between the behavioral versus the purely behavioral sunk cost investments leading us to conclude that purely behavioral sunk cost effect can be confirmed in the lab and the effect can have the same magnitude as with behavioral sunk costs. Even without the help of contextual cues, the sunk cost effect was also present in the No Payment Treatment. Cuhna and Caldieraro (2009) argue that because they used an actual effort manipulation rather than hypothetical scenarios, the activation of the effort justification mechanism was possible. This could well be the case in my study where participants exerted real effort and the results confirmed the appearance of the sunk cost effect in purely behavioral investments.
The difference in my results from other previous research can also be explained by whether and which experimental set-up choices satisfy the participant’s aspiration level. The aspiration level of an individual typically increases with additional sunk costs, which can be a compelling explanation for the risk aversive behavior of the bigger-effort Treatment. As Zeelenberg and Van Dijk (1997) state in their paper, “opting for the risky option does not only involve the risk of losing the gamble, but also the risk of experiencing the feeling of having worked for nothing”, so it seems that we have too much invested to gamble, making the safer choice of quitting more attractive as it satisfies the individual’s aspiration level.
18
One possible explanation about my results regarding hypothesis 1 (the presence of the sunk cost effect in all treatments) and why they differ from the results of other studies that could not replicate the sunk cost effect in behavioral investments in the lab could be that, as Heath (1995) stated in his work about mental budgeting, people escalate commitment when they do not set a budget. When information about benefits is not available, people may not immediately set a budget. It could as well be the case that this rule also applies when information about costs is not available. In the case of my experiment, the subjects had all the information needed for the marginal benefits but they could not fully calculate their budget as the marginal cost, namely the investment fee which in this case was the effort or time needed to complete a stage, was not given precisely because it was a factor that can differ from individual to individual. These kinds of situations may be the reason why people invest longer and it results difficult for individuals to use their normal budget setting procedure.
6.2 Contributions
The present research extends the literature in two important ways. First of all, my contribution is mainly methodological, as it is one of the few studies to avoid using hypothetical survey questions and instead, urge the participants to make real investments. Moreover, it is the first study among the relevant literature to use an online experiment in order to examine the sunk cost effect, and for the reasons listed in the experimental design section, I believe this is a very useful and appropriate tool for answering a research question about behavioral investments. The second contribution is providing a deeper insight into the ambiguous issue of the cases in which the sunk cost effect of behavioral investments is observed. My research suggests that the bigger the effort we invest in an ongoing, risky course of action, the bigger the probability to switch to a safer alternative. In addition, there is not always need for a link to a monetary equivalent to value effort and time as a considerable cost.
6.3 Limitations
The present research was not without limitations. First of all, my study revolved around behavioral investments mostly expressed in effort from the part of the participants. This raises two basic issues; Firstly, effort is not easily measured and is a cost that can vary among the participants which incommodes the estimation of it as well as the efficiency of the effort manipulation in High Effort-Payment treatment. Secondly, the effort manipulation in the experiment was also confounded in many instances with the total time spent on the task. In general, effort is deeply interwoven with the sense of time making these two factors almost inseparable when it comes to isolating them in a laboratory setting. A more complete treatment of the sunk-cost effect should involve an orthogonal
19
manipulation of time, effort and monetary investments in order to isolate the effects of each other on decision making.
Secondly, there is a possibility that the task was more enjoyable than expected and this might have influenced the propensity of participants to keep going until the last stage of the game. The questionnaires did not report such an incident but a skeptic could argue that, when choosing to play a game, individuals usually invest a lot of time without measuring parameters other than fun because it is mostly perceived as an enjoyable activity no matter how much cognitive effort it requires. This angle should probably be further investigated as in most situations in real life where the sunk cost effect occurs the context is slightly more serious than the context of a game.
Last, my investigations did not allow me to determine the single underlying mechanism for the purely behavioral sunk cost effect. I provided some possible explanations to interpret my results but a deeper view into the psychological rules that govern our behavior when we cling to a bad time or effort investment need a broader testing.
6.4 Discussion and directions for future research
In this experiment I tried to control for the quality of the experience and I only changed the quantity of effort, and subsequently, of time. However, research on time perception suggests that it is possible that the same quantity of time investment may have a different sunk-cost impact in different settings, for example when varying the quality of the experience. That remark leads us to consider the possibility of the task of the experiment being something rather enjoyable than tedious, a factor that did not vary among the treatments but it might have influenced the overall willingness of participants to finish all the stages of the game. It would be a promising path for future research to further investigate the effect of the quality of time and effort regarding a task on its sunk-cost impact by varying the quantity and controlling for the quality of the experience.
Moreover, further research would be useful in order to acquire a deeper understanding of the mechanisms than can explain the sunk cost effect as a result of purely behavioral investments. Among my findings was that in direct experimental settings the sunk cost effect of purely behavioral investments reappears, most probably due to the effort-justification mechanism that could take place in the process. A more extensive and targeted to the psychological side research could broaden our understanding on the issue.
Last but not least, I would encourage future research to look into the importance of how to better exploit our knowledge on the sunk cost effect towards our own benefit and how this exploitation could perhaps better our personal management of time and cognitive resources allocation.
20 7. References:
Arkes, H. R., & Ayton, P. (1999). The sunk cost and Concorde effects: are humans less rational than lower animals?.Psychological Bulletin,125(5), 591.
Arkes, H. R., & Blumer, C. (1985). The psychology of sunk cost. Organizational behavior and
human decision processes, 35(1), 124-140.
Aronson, E. & Mills, J. (1959). The effect of severity of initiation on liking for a group. Journal of Abnormal and Social Psychology, 59(2), 177-181.
Brockner, J. (1992). The escalation of commitment to a failing course of action: Toward theoretical progress.Academy of Management Review,17(1), 39-61.
Cunha Jr, M., & Caldieraro, F. (2009). Sunk‐Cost Effects on Purely Behavioral Investments.Cognitive Science,33(1), 105-113.
DeVoe, S. E., & Pfeffer, J. (2007). When time is money: The effect of hourly payment on the evaluation of time. Organizational Behavior and Human Decision Processes, 104(1), 1-13. Festinger, L., & Carlsmith, J. M. (1959). Cognitive consequences of forced compliance. The Journal of Abnormal and Social Psychology, 58(2), 203.
Fredrickson, B. L., & Kahneman, D. (1993). Duration neglect in retrospective evaluations of affective episodes.Journal of personality and social psychology,65(1), 45.
Friedman, D., Pommerenke, K., Lukose, R., Milam, G., & Huberman, B. A. (2007). Searching for the sunk cost fallacy.Experimental Economics,10(1), 79-104.
Garland, H., & 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(1), 55-69.
He, X., & Mittal, V. (2007). The effect of decision risk and project stage on escalation of commitment.Organizational Behavior and Human Decision Processes,103(2), 225-237. Heath, C. (1995). Escalation and de-escalation of commitment in response to sunk costs: The role of budgeting in mental accounting. Organizational Behavior and Human Decision
Processes, 62(1), 38-54.
Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econometrica: Journal of the Econometric Society, 263-291.
Milgrom, P. R., & Roberts, J. (1992).Economics, organization and management(Vol. 7). Englewood Cliffs, NJ: Prentice-Hall.
Northcraft, G. B., & Neale, M. A. (1986). Opportunity costs and the framing of resource allocation decisions.Organizational Behavior and Human Decision Processes,37(3), 348-356.
Park, J. Y., & Jang, S. S. (2014). Sunk costs and travel cancellation: Focusing on temporal cost.Tourism Management,40, 425-435.
Simon, H. A. (1955). A behavioral model of rational choice. The quarterly journal of
21
Soman, D. (2001). The mental accounting of sunk time costs: Why time is not like money. Journal of Behavioral Decision Making, 14(3), 169-185.
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(1), 27-44. Tan, H. T., & Yates, J. F. (1995). Sunk cost effects: The influences of instruction and future return estimates.Organizational Behavior and Human Decision Processes,63(3), 311-319. Teger, A. I., & Cary, M. (1980).Too much invested to quit. New York: Pergamon Press. Thaler, R. H. (1980). Toward a positive theory of consumer choice. Journal of Economic Behavior and Organization,1, 39–60.
Thaler, R. H. (1985). Mental accounting and consumer choice. Marketing Science, 4, 199– 214.
Thaler, R. H. (1999). Mental accounting matters. Journal of Behavioral Decision Making, 12, 183–206.
Thaler, R. H., & 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. 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(6), 677-691.
Zeng, J., Zhang, Q., Gong, Q., Yu, R., & Chen, C. (2013). An fMRI study on sunk cost effect.Brain research.
22
APPENDIX
INSTRUCTIONS FOR BASELINE TREATMENT: Instructions Welcome to this experiment!
The experiment consists of 5 Stages in which you will be asked to complete a task. If you successfully pass all stages you have a chance to receive the winner’s prize which is 15 euros. The chance to win the prize is 20% .
Before the start of each Stage you always have the option to quit and the option to move on to the next Stage. If you decide to take part to the next Stage you will have to do the following task-game:
Can you crack the code?
In each Stage you have to correctly guess a 4-digit code. The goal of the game is to combine all the codes you crack to find the password that wins the game. In order to find the 4-digit codes you start by making a guess and after receiving feedback from the computer about how far your answer is from the correct one you can try again and again until you find the correct code. The code cannot contain the same digit more than once, for example the code 1223 is not possible. In every Stage you will be given a hint about a number that is in the code to help you.
The feedback that you will get from the computer will be as follows:
The computer will inform you about the correct digits and the correct position of the digits by responding to each code with: a “#” if there is some right digit in the exact right place, a “*” if there is some right digit in the wrong place and a “-“ if they are all wrong. Note that the order of the feedback characters do not give any information about which digit refers to a “#” and which one refers to a “*”. The “#” always appear before the “*” .
For example, let’s say that the code we are looking for in the second Stage is 6458. The hint is that the code contains the digit 5. Every first line is a guess you might make and every second is the computer’s response.
6458 5087
** (which means that two digits are correct but need to change order) 2703
- (which means that none of these digits are contained in the code) 9856
23 9816
**
And you keep going until you get as a response ‘ ####’
After the end of the last Stage, you combine all the codes to find the winning password which will be a popular song title. You will receive more specific instructions and a hint when you reach this point in the game.
At the end of the experiment all participants will have to complete a very small questionnaire.
Payment will be determined on the 20th of June where 20% of the participants who completed the game will be randomly picked to receive the payment.
Please complete the experiment in one go without pauses, because after 2 hours you will automatically be disconnected.
24 SAMPLE OF THE ONLINE EXPERIMENT DISPLAY
25
