Content uploaded by Corina Haita-Falah
Author content
All content in this area was uploaded by Corina Haita-Falah on Apr 06, 2021
Content may be subject to copyright.
Sunk-cost fallacy and cognitive ability in individual
decision-making
Corina Haita-Falah
University of Kassel, Nora-Platiel-Strasse 4, 34109 Kassel, Germany
article info
Article history:
Received 9 September 2015
Received in revised form 14 June 2016
Accepted 12 December 2016
Available online 23 December 2016
Keywords:
Cognitive ability
Cognitive dissonance
Sunk-cost fallacy
Loss aversion
abstract
This paper reports on a laboratory experiment aiming at documenting the sunk-cost
fallacy in individual decision-making and at identifying the role of the cognitive abil-
ity in its manifestation. For this purpose, the design rules out loss aversion and cog-
nitive dissonance, identified by the literature as being the main psychological drivers
of the bias. The sunk-cost fallacy is identified by comparing a low and a high sunk-
cost treatment, respectively, against a control group that does not incur a sunk cost.
There is evidence of a weak manifestation of the sunk-cost fallacy, which is statisti-
cally significant only for the high sunk-cost treatment. However, strong evidence of
the fallacy was found among the high-cognitive-ability subjects. Finally, although cog-
nitive ability is predictive of status-quo bias, it was not found to reduce the sunk-
cost bias.
Ó2016 Elsevier B.V. All rights reserved.
1. Introduction
Normative economic theory indicates that costs incurred in the past are irrelevant for future marginal payoffs, i.e. sunk
costs must be ignored. Nevertheless, there is evidence that the actual human behavior violates this normative prescription
and people tend to account for historical costs. In common language, the sunk-cost fallacy (bias) is the irrational behavior of
”throwing good money after bad,” i.e. once found on a course of action to which they committed an investment (e.g. time,
money, effort), people continue to stay on that course of action and invest even more resources despite it being unprofitable.
While less discussed in the literature, the bias can also manifest into a premature abandoning of an otherwise profitable
enterprise. For example, concerning firm’s short versus long-run production decisions, a (competitive) firm should only exit
the industry when the price is below the long-run average total cost. However, shutting down before the price falls below the
minimum average variable cost in the short-run (premature exit) is a manifestation of the bias since fixed costs are sunk in
this case.
1
As Thaler (1980) points out, efforts of identifying the sunk-cost fallacy from field data are often hindered by a selection
bias. Therefore, evidence of the sunk-cost fallacy has, thus far, been limited to hypothetical scenarios and field experiments,
while efforts for documenting it in laboratory are still surprisingly scarce and provide mixed evidence (Ashraf, Berry, &
Shapiro, 2010). On the one hand, hypothetical questions lack salience since subjects are asked to imagine decision scenarios
http://dx.doi.org/10.1016/j.joep.2016.12.001
0167-4870/Ó2016 Elsevier B.V. All rights reserved.
E-mail address: haita-falah@uni-kassel.de
1
This is, probably, the most common example that microeconomics textbooks use when teaching firm’s production decisions and the distinction between
sunk fixed costs and non-sunk fixed costs. I thank to an anonymous referee for pointing out this instance of the sunk-cost bias.
Journal of Economic Psychology 58 (2017) 44–59
Contents lists available at ScienceDirect
Journal of Economic Psychology
journal homepage: www.elsevier.com/locate/joep
and state their decisions. On the other hand, field experiments are most of the time contextual and use real commodities
(Harrison & List, 2004), which limits the validity of the findings to the particular context. Moreover, decisions in the field
interfere with subjects’ unobserved prior beliefs and experience in relation to the particular experimental context. At the
same time, it is conceivable that (consumption) decisions in the field do not elicit individual, but rather group decisions,
being, thus contaminated by the relative bargaining power within the group.
2
The latter, however, remains unobserved to
the experimenter. Therefore, more controlled laboratory experiments can provide cleaner evidence for the manifestation of
the fallacy in individual decision making and help identifying the roots of the bias.
In the sunk-cost fallacy literature, two main psychological mechanisms have been made responsible for the
manifestation of the bias. First, Staw (1976) argues that the state of cognitive dissonance between one’s actions
and the cognition of rational behavior creates a state of mental discomfort. One common mechanism that reduces
this discomfort is a post hoc rationalization of past decisions, i.e. self-justification of past decisions. In the context
of the sunk costs, the best way one can justify past decisions is by continuing to pour resources into a failing course
of action. Supporting this argument, the author finds that people are more committed to a previously chosen alter-
native if made responsible for that decision at an earlier point in time. Similarly, Bazerman, Giuliano, and Appelman
(1984) find that being made responsible for the existence of a sunk cost increases the amount of resources allocated
for the continuation of a project. Arkes and Blumer (1985) also discuss the cognitive dissonance theory as being
related to the manifestation of the sunk-cost bias, but conclude that this explanation is insufficient for understanding
the bias. Instead, the authors advocate loss aversion (Kahneman & Tversky, 1979) as a suitable explanation for the
sunk-cost fallacy. In fact, the connection between loss aversion and the sunk-cost fallacy was firstly noted by
Thaler (1980) starting right from the experiments conducted by Kahneman and Tversky (1979). The author explains
that the convexity of the utility function in the domain of losses, i.e. risk-seeking behavior, is responsible for the
escalation on an initial investment.
In this paper I use a laboratory experiment in which the above-mentioned psychological drivers of the sunk-cost fallacy
are made impotent. Hence, the first endeavor of this study is to show that the two psychological channels of the sunk-cost
fallacy are not necessary for the bias to manifest itself. Second, I investigate the potential of the cognitive ability to alleviate
the bias. Cognitive ability was found to reduce several biases such as conjunction fallacy, base rate fallacy, conservatism bias
and overconfidence (Hoppe & Kusterer, 2011; Oechssler, Roider, & Schmitz, 2009). However, virtually all the evidence relat-
ing the sunk-cost fallacy to the cognitive ability is supplied by the psychology literature (see Section 2). Although subjects in
psychological experiments are paid for their participation, they are not paid in accordance to their decisions. In fact, these
experiments use hypothetical-scenario tasks without economic consequences for the participants and, therefore, they do not
guarantee actual behavior. Economic experiments, on the other hand, are likely to provide a more accurate measure of peo-
ple’s actual behavior in an economic environment. Nevertheless, I am not aware of any economic experiment in a controlled
laboratory setting which investigates the relationship between cognitive ability and the decision to ignore sunk costs. I take
up this endeavor in the experiment reported here. For this, I use the Cognitive Reflection Test (CRT) developed by Frederick
(2005), together with three mathematical questions from Benjamin, Brown, and Shapiro (2006), as a measure of cognitive
ability.
The experimental manipulation consists of one control and two treatment groups. The participants in the control
group are endowed with a number of units of an asset A and an amount of cash, whereas the participants in the
two treatment groups are endowed with cash and offered the possibility to purchase the same number of units of
asset A as the control group. The two treatment conditions differ with respect to the price of asset A, a low
sunk-cost and a high sunk-cost treatment, respectively, in order to test whether the sunk-cost fallacy is related to
the size of the investment. In a subsequent stage, all participants have the possibility to trade their holdings of asset
A and buy an alternative asset B that has the same redemption value but a lower ask price than asset A. For this rea-
son, selling all the endowment of asset A is the profit-maximizing decision, though the selling price of asset A is lower
than the initial purchase price, i.e. part of the initial investment remains sunk. Comparing trades in this second stage
allows to identify a sunk-cost bias if participants in the treatment groups sell fewer units of asset A than those in the
control group.
The experimental data indicates behavior consistent with the manifestation of the sunk-cost fallacy. The non-parametric
analysis confirms the existence of a statistically significant trend across the three experimental groups, though two-by-two
comparisons show a significant difference only between the control group and the high sunk-cost treatment. Similarly,
regression analysis shows a significant treatment effect only for the high sunk-cost treatment. While the treatment effect
in the case of the high sunk cost survives controlling for cognitive ability, the interaction between the treatment dummy
and the measure of cognitive ability does not confirm an effect of the latter on the sunk-cost bias. However, cognitive ability
appears to be responsible for status-quo bias.
The paper proceeds as follows. In the next section I review the existing evidence of the sunk-cost fallacy. In Section 3I
present the experimental design and I discuss how it relates to the psychological causes of the fallacy. Section 4presents
the experimental results, while Section 5includes a discussion of the results and the limitations of the study. Section 6sum-
marizes the findings and concludes.
2
See, for example, Ashraf et al. (2010) which studies consumption decisions at household level.
C. Haita-Falah / Journal of Economic Psychology 58 (2017) 44–59 45
2. Existing literature
Most of the experimental literature investigating the sunk-cost fallacy makes use of contexts and situations, particularly
in field studies where real goods are used. For this reason, the results are rather confined to the context, the particular com-
modity used or the population treated. Along these lines, Tan and Yates (1995) shows that the decision to escalate on an
initial course of action is sensitive to the context in which the problem is formulated. Using hypothetical scenario questions,
the authors show that students who had prior instructions in sunk-cost principles did ignore it when the context of the prob-
lem was similar to the textbook examples. However, they failed to do so when the decision reflected a real-life situation.
Probably the most prominent study documenting the sunk-cost fallacy is the field experiment conducted by Arkes and
Blumer (1985). The authors are able to capture differences in behavior among three groups of theater season tickets buyers,
who were randomly chosen to pay different prices: full price and two levels of discounted prices. The experiment shows that
those who paid the full price of the ticket visited the theater more often during the season than those who paid a discounted
price.
Considered to be the second field experiment investigating the sunk-cost fallacy, Ashraf et al. (2010) employ a random-
ized control trial in Zambia to test whether higher prices induce more product use. Their experimental design is able to iso-
late the sunk-cost effect from the self-selection effect, but they find no evidence of the sunk-cost effect, at least in the domain
of health products used in their study. Their experimental manipulation is inspired by the unexpected random discount in
the offer price manipulated by Arkes and Blumer (1985). However, unlike Arkes and Blumer (1985) and similar to my design,
they include a treatment with zero transaction price. Using this treatment the authors test the hypothesis of paying a pos-
itive price versus paying zero price and find a sunk-cost effect, although not statistically significant. Interestingly, Ashraf
et al. (2010) find evidence of the sunk-cost effect in households’ answers to hypothetical questions, which is, however, incon-
sistent with households’ actual behavior. This result seems to undermine the reliability of the findings from previous studies
based on hypothetical questions, and reinforces the need for laboratory experiments in order to clarify the mixed evidence.
I am aware of only two economic experiments that explicitly investigate the sunk-cost fallacy in the laboratory. First,
using lottery valuations as a measure of escalation of commitment, Phillips, Battalio, and Kogut (1991) show that when
the sunk costs are made more transparent, they are more likely to be ignored. Nearly half of their subjects failed to ignore
the sunk cost when this was not explicitly paid, but was only a verbal commitment. However, only 19% of their subjects
exhibited the bias when the sunk cost was made more salient through the physical act of paying the lottery ticket (the sunk
cost in their experiment). Second, Friedman, Pommerenke, Lukose, Milam, and Huberman (2007) devised a computer game
to isolate factors which determine the sunk-cost fallacy. Precisely, their design eliminates rational motives for the manifes-
tation of the fallacy, but can identify the effects of the cognitive dissonance and loss aversion on the bias. The authors report
surprisingly small and inconsistent evidence of the sunk-cost fallacy, while the psychological drivers manipulated by their
study also have a small and inconsistent impact on the manifestation of the fallacy.
The experimental psychology literature points to cognitive ability as a candidate to explaining the sunk-cost fallacy,
though the evidence from these experiments is rather mixed. For example, Larrick, Nisbett, and Morgan (1993) find some
correlation between subjects’ recognition of economists’ position with respect to sunk costs and the Scholastic Assessment
Test (SAT) verbal score, which is their measure of cognitive ability. However, the SAT score did not correlate with subjects’
own reported behavior. Similarly, Strough, Mehta, McFall, and Schuller (2008) find small or insignificant correlations
between the sunk-cost fallacy and the scores of various cognitive tests. On the other hand, using the self-reported SAT scores
as a measure of cognitive ability, Stanovich and West (2008) find a significant effect of the cognitive ability on the manifes-
tation of the sunk-cost fallacy across the cognitive groups (low and high). However, the interaction of the cognitive ability
with their measure of the sunk-cost fallacy was not found significant, suggesting that sunk-cost fallacy is not attenuated by
cognitive ability. Parker and Fischhoff (2005) measure the correlation between knowledge and reasoning, as proxies for cog-
nitive ability, and the resistance to sunk costs. The authors find a weak and overall statistically insignificant correlation
between the sunk-cost fallacy index and their measure of cognitive ability. Extending the scale of the sunk-cost questions
of Parker and Fischhoff (2005) and using a much larger sample of adults, Bruine de Bruin, Parker, and Fischhoff (2007) find
similar small correlations between the resistance to sunk cost and the two measures of cognitive ability, i.e. knowledge and
reasoning. As in my study, Toplak, West, and Stanovich (2011) use the CRT to measure cognitive ability and find that the test
results correlate significantly with their measure of rational-thinking which contains a sunk-cost task. However, the authors
do not report a separate correlation between the CRT and the performance on the sunk cost task itself, which is one of the
goals of the present paper.
3. Experimental design
The first part of this section describes and discusses the design of the main sunk-cost experiment, together with the tes-
table hypothesis. In the second part I detail the procedure applied in each experimental session.
3.1. Treatments and hypothesis
The experimental task reported in this study consists of a one-shot individual decision, with no interaction among sub-
jects. The decision environment can be summarized as follows. Consider a situation in which, at the time of receiving new
46 C. Haita-Falah / Journal of Economic Psychology 58 (2017) 44–59
information, a decision maker has already invested into a course of action towards achieving a certain goal. In the experi-
ment, the initial course of action is represented by an asset labeled ‘‘A.” At this point she learns that, for achieving the goal,
a cheaper alternative course of action becomes available, represented in the experiment by an asset labeled ‘‘B.” Moreover,
the initial investment in A is partially reversible, which should provide a nudge towards abandoning the initial course of
action when this is no longer profitable. Failure to abandon the initial course of action A or, conversely, to adopt the alter-
native course of action B, is interpreted as sunk-cost fallacy. In this experiment the fallacy can manifest in various degrees,
depending on how much the decision maker adopts the alternative course of action.
The experimental manipulation consists of one control group (CT) and two sunk-cost treatments: a low sunk cost (LSC)
and a high sunk cost (HSC) treatment, respectively. Considering two sunk-cost conditions, the aim is to investigate whether
the size of the sunk cost has any effect on the manifestation of the fallacy. The CT group receives a free endowment of 20
units of an asset A and 200 Experimental Euro (EE), while the treatment groups are endowed only with cash: 1000 EE in
the LSC condition and 1400 EE in the HSC condition. The sunk-cost groups have the opportunity to either invest in 20 units
of asset A or keep their cash endowment and make no further decisions (see Section 3.2). However, the unit price of the 20
units of asset A depends on the treatment condition, i.e. 40 EE in the LSC and 60 EE in the HSC, making the total investment
equal to 800 EE and 1200 EE, respectively. Hence, subjects in the CT group do not incur a sunk cost, whereas the investments
of the LSC and the HSC subjects constitute sizable amounts of the initial cash endowments. This makes the sunk cost salient.
With this choice of parameters, subjects in both treatments have the same financial positions after the investment: 20
units of asset A and 200 EE, which is identical to the initial endowment in the CT group. This eliminates wealth effects across
the experimental groups with respect to further decisions. When offered to invest, subjects are informed that if invested, the
20 units of asset A will be redeemed at the end of the experiment for a unit price of 70 EE. Therefore, investment is always
profitable, fact that was also emphasized in the experimental instructions which informed subjects truthfully that, regard-
less of further decisions, investing cannot lead to lower earnings than the initial cash endowment.
After the completion of the investment stage (or the presentation of the initial position in the case of the CT group), sub-
jects learn simultaneously three pieces of new information. First, they are informed that their goal is to accumulate exactly
30 units of assets A and B, in any combination of the two. The general instructions read aloud before the experiment began
informed subjects about the existence of asset B and the fact that they had to collect a number of units of Asset A and/or
Asset B, but not about the actual number of units to collect.
3
Second, they learn that they can accomplish the goal by the means
of a market in which they can trade (buy or sell) asset A with the experimenter and buy asset B from the experimenter. The
trading price of asset A is above the price of B but below that at which A was bought initially, such that part of the initial invest-
ment in A remains sunk and asset B is comparably more attractive.
4
Third, the unit price of asset B is 10 EE and its redemption
value is the same as that of asset A, i.e. 70 EE. All prices are known to the subjects before they make their trading decisions. The
reader is referred to Appendix A for the theoretical basis of the parameters’ choice, while a summary of the experimental treat-
ments and parameters can be found in Table B.1 of Appendix B.
Thus, at the time of receiving new information, the investment in 20 units of asset A is sunk and, given the relationship
between the prices of asset A and B, it is rational to sell all units of asset A and buy 30 units of asset B to complete the goal.
Therefore, the number of units of asset B bought measures the degree to which the decision maker reverts from or escalates
on the initial course of action. In order to leave open the possibility of fully escalating on the initial course of action (i.e. asset
A), the remaining cash endowment of 200 EE was chosen to allow for the purchase of the extra 10 units of asset A to reach
the goal of 30 units.
Given the experimental design and parameters described above, I am now ready to formulate the sunk-cost fallacy
hypothesis of this paper:
Hypothesis 1. Subjects in the control group buy more units of asset B than subjects in the LSC group, who, in turn, buy more
units of asset B than subjects in the HSC group.
It may be argued that holding on asset A, which is interpreted as manifestation of the sunk-cost fallacy in this design, can
also be explained by endowment effect with respect to the 20 units of asset A. Note, however, that the endowment effect
predicts that the value associated to an item increases after the property right has been assigned, regardless of whether
the assignment was free or at a cost. Hence, any endowment effect with respect to the 20 units of asset A should be the same
across the three experimental conditions. By contrast, under the sunk-cost fallacy the initial price paid plays a disproportion-
ate role in the valuation of the item. Therefore, it is exactly the difference in how subjects were endowed with these units
that identifies the sunk-cost fallacy in my experiment.
As discussed in Section 1, the previous literature argued that cognitive dissonance and loss aversion are the leading expla-
nations for the manifestation of the sunk-cost fallacy. In order to examine the role of the cognitive ability in the manifesta-
tion of the bias, the experimental design rules out these psychological explanations.
To see this, first note that when faced with the decision to invest in asset A, subjects of my experiment knew the exact
consequences of this investment, i.e. the sure profit resulted from the difference between the redemption value and the pur-
chase price. Since there is no deception in the experiment, the decision to invest is both ex-ante and ex-post optimal. Hence,
3
Details about the general instructions are presented in Section 3.2 and the instructions are reproduced in Appendix C.
4
This price was randomly drawn with equal probability from the integers in ½16;20, with the purpose of creating variation in the re-sale price, which,
nevertheless, did not have a significant effect on trade.
C. Haita-Falah / Journal of Economic Psychology 58 (2017) 44–59 47
in this experiment investing does not have negative consequences and, therefore, cannot create a need for self-justification.
This rules out cognitive dissonance as an explanation for the sunk-cost bias in this experiment.
Next, loss aversion cannot explain why subjects in this experiment escalate on the initial commitment, i.e. hold on asset A
or even buy more units of asset A when a cheaper alternative is available. Note that the key condition for loss aversion to
explain this escalation is that the decision-maker is in the domain of losses.
5
Quite the opposite, the investment decision
in my experiment is always profitable since the final redemption value is above the initial price of asset A, for all treatment
conditions. Therefore, the decision to invest increases subjects’ financial positions relative to their initial cash endowments
(the reference point), such that they are in the gains domain.
6
After ruling out the psychological factors explaining the sunk-cost bias, cognitive ability remains one factor that could
play a role in its manifestation. Hence, in line with the evidence from the psychology literature, the second hypothesis of
this study can be formulated:
Hypothesis 2. High cognitive-ability subjects are less prone to the sunk-cost fallacy than the low cognitive-ability subjects.
To measure cognitive ability I use the CRT developed by Frederick (2005) and three mathematics questions selected from
Benjamin et al. (2006). The CRT is a three-item test that requires reflection that leads to the correct answer, but at the same
time calls for the temptation to give an intuitive, but wrong answer. The mathematics questions had the purpose of collect-
ing information about subjects’ numeracy skills, particularly the ability to perform arithmetic operations and compare
numbers.
3.2. Procedure
Six experimental sessions (two per treatment) were run in the experimental laboratory at the University of Hamburg, in
July 2014. Each session lasted approximately 45 minutes. A total of 138 subjects, recruited online through the hroot system
(Bock, Nicklisch, & Baetge, 2012), participated in the experiment. The participants were mostly undergraduate students from
both social and natural sciences. No subject participated in more than one session. The interface of the experiment was pro-
grammed in z-Tree (Fischbacher, 2007).
Each experimental session consisted of two parts: the main sunk-cost experiment described above, and a cognitive ability
test. Both parts of the experiment were monetary-incentivized. The earnings from each part were added up, converted into
Euros at the exchange rate of 200 EE for 1 Euro and paid out in private at the end of the session. Per subject earnings ranged
between 5 and 13.90 Euro, with an average payoff of 12.35 Euro.
Before the experiment started, the instructions reproduced in Appendix C were read aloud. The general part of the
instructions informed the subjects about the two parts of the experiment and the calculation of the final payoff. Subse-
quently, the instructions described the main sunk-cost experiment, providing an overview of the experimental task with
the purpose of familiarizing the subjects with the elements and the sequence of decisions, without introducing the actual
parameters. Subjects learned the specific parameters only on the experimental screens as proceeding through the sequence
of decisions. The experimental screens are presented in Appendix D. Before the experiment started, all subjects answered a
set of control questions. In case of incorrect answers, the next screen provided the correct answer along with an
explanation.
7
The stages of the main sunk-cost experiment are as follows. After the initial endowment stage, the LSC and HSC groups
entered the investment stage in which they were offered the possibility to invest in 20 units of asset A. Those subjects who
chose not to invest could keep their cash endowment and were asked to wait idle in their seats for the second part of the
experiment – the cognitive quiz. Subjects who chose to invest proceeded to the trade stage, which for the CT group followed
immediately after the initial endowment. At the trade stage, subjects in all groups could actively buy or sell units of asset A
and buy units of asset B, under the constraint of holding exactly 30 units of asset A and/or B. Since there was no room for
speculation in this experiment and in order to eliminate confusion, the experimental interface did not allow subjects to buy
and sell units of asset A at the same time (see Fig. D.4). Moreover, as there was only one round of trade, subjects were asked
to confirm their trading choice before this became effective (see Fig. D.5). At every stage, subjects could see their financial
positions on the experimental screens.
After the completion of the main sunk-cost experiment, all subjects (including those who did not invest) answered the
CRT and the mathematics questions. The two types of questions were shuffled and presented to the subjects in the order
shown in Appendix E, where questions 1, 4 and 6 constitute the CRT. Every correct answer in the cognitive ability test
was worth 50 EE. The experiment ended with a final questionnaire in which all subjects answered a set of demographic
questions.
5
Loss aversion predicts risk seeking in the domain of losses, due to the convexity of the psychological value associated with losses. This means that after an
unsuccessful investment the decision-maker is willing to take further risks in the hope of an eventual gain.
6
Indeed, right after the investment the financial position of a subject in any of the experimental conditions is (20 units of asset A) 70 EE + 200 EE = 1600
EE, which is larger than any of the initial endowments. Moreover, subjects remained in the gain domain regardless of the rationality of their further decisions,
with minimum possible earnings of 2100 EE which is well above either of the initial endowments. The Supplementary Appendix contains the results from an
additional experiment aiming at understanding the sunk-cost fallacy in the loss domain, using the strategy method.
7
The control questions verified subjects’ understanding of the exchange mechanism between the two assets, the rules of the trade and the fact that the two
assets had the same redemption value. The questions are available upon request from the author.
48 C. Haita-Falah / Journal of Economic Psychology 58 (2017) 44–59
4. Results
The following analysis is based on the sample composed of all subjects in the control group and those in the treatment
groups who invested in asset A and, thus, continued the experiment with further decisions.
8
The summary statistics of the
treatment groups are presented in Table B.2. The last column in the table shows the p-values of the Kruskal-Wallis equality-
of-populations rank test, which tests whether at least two of the three treatment groups differ significantly from each other.
Apart from slightly more males in the LSC group, the three treatment conditions do not exhibit statistically significant differ-
ences. Most importantly, they are the same with respect to the score of the cognitive ability quiz, both overall and on its
two components, the CRT and the mathematics questions.
The variable of interest for testing Hypothesis 1 is the number of units of asset B bought by the subjects. This measures
the degree to which the subjects recognized the optimality of adopting the alternative course of action and acted accord-
ingly. The possible values of this variable range from 0 units, indicating full escalation of commitment, to 30 units, meaning
full abandonment of the initial course of action. Any value below 30 is interpreted as evidence of the sunk-cost fallacy and
the lower this value, the higher the manifestation of the fallacy. Hence, in this experiment subjects can manifest the sunk-
cost bias in a continuous manner.
4.1. Treatment effects
Fig. B.1 presents the kernel density estimates of the distribution of the number of units of asset B bought by the exper-
imental subjects in each of the treatment conditions. The upper tails of the distributions (11–30 units of asset B) correspond
to the region in which subjects sold at least half of their holdings of asset A - the region of highest rational behavior. Con-
sistent with the manifestation of the sunk-cost fallacy, this region features the lowest frequency among the HSC group (the
dotted line) followed by the LSC (the dashed line), with the CT group (the continuous line) exhibiting the highest frequency.
The lower tails of the distributions (0–10 units of asset B) correspond to the region in which the experimental subjects
retained the entire 20 initial units of asset A to which they added more units of asset A towards the completion of the
required 30 units of assets. This is the region of lowest rational behavior. Again, consistent with committing the sunk-
cost fallacy, the lowest frequency of this type of behavior is manifested by the CT group, followed by the LSC and HSC groups,
respectively.
I further analyze the treatment differences using non-parametric analysis. Table B.3 shows the treatment averages and
the standard errors for the units purchased from asset B, along with the number of units of asset A bought or sold.
9
In both
sunk-cost treatments the subjects recognized less the optimality of switching to the cheaper asset B than in the CT group. How-
ever, the difference in behavior is statistically significant only between the CT and the HSC group (p¼:01).
10
Nevertheless, a test
of the joint hypothesis that qB
CT >qB
LSC >qB
HSC , which is the hypothesis of this study, confirms the existence of a trend across the
treatments with respect to the number of units of asset B bought (Jonckheere-Terpstra trend test, 1-sided p¼:01). Treatment
differences between the CT group and the HSC group are also found in the loss domanin (see the Supplementary Appendix for
details on extra sessions involving the loss domain).
The above results are also confirmed by the regression analysis presented in Table B.4. Column (1) shows the treatment
effects in which the control group is the baseline category. Thus, the constant shows the average number of units of asset B
bought by the control group: 22 out of the maximum possible of 30 units. Indeed, the coefficients on LSC and HSC (the treat-
ment effects) have signs consistent with the manifestation of the sunk-cost bias, but they are statistically significant only for
the HSC group. A subject in the HSC treatment bought, on average, 5–6 units of asset B less than a subjects in the CT group.
Hence, the sunk-cost hypothesis of this study is only partially confirmed:
Result 1. Subjects’ behavior in the experiment is consistent with the manifestation of the sunk-cost fallacy, but this behavior is
statistically significant only for the HSC group.
4.2. Cognitive ability
In order to investigate the causal relationship between cognitive ability and the sunk-cost bias, I add the two measures of
the cognitive ability (CRT and Math) as covariates to the regression from column (1) of Table B.4. Both measures are stan-
dardized. The results are presented in column (2). The treatment effects remain unchanged, both in magnitude and statistical
significance. The coefficient on Math is both economically small and statistically insignificant, but the CRT test score is
strongly predictive of behavior consistent with rationality. Specifically, a subject in the CT group with a CRT score of one
standard deviation above the mean bought 4 more units of asset B as compared to a subject who had an average CRT score.
The reason for the lack of significance of the mathematics test is the very small variation with respect to this dimension of
8
11 out of the 96 subjects who were offered to invest chose not to do so. A discussion of the potential selection effect due to these subjects is deferred to
Section 5.
9
Note that while all these variables are measures of the sunk-cost fallacy, the latter two are mutually exclusive by design.
10
Unless otherwise specified, all the reported p-values are 1-sided, Mann-Whitney-Wilcoxon test.
C. Haita-Falah / Journal of Economic Psychology 58 (2017) 44–59 49
the cognitive ability (the standard deviation is 0.4 with mean 2.84 correct answers). In fact, 99% of the subjects answered
correctly at least two out of the three mathematics questions and no subject gave zero correct answers. By contrast, the pro-
portion of correct answers for the CRT was only 45%, 60% and 68%, respectively, for each of the three questions that com-
posed the test. These proportions are in line with those obtained by Oechssler et al. (2009) and Hoppe and Kusterer (2011).
Column (3) includes the variable Stock market which indicates familiarity with the stock-market trading and stock-
market news, as reported by the subjects in the final questionnaire.
11
Holding the treatment and the CRT test score constant,
a subject who reported reading financial newspapers and following the stock market bought on average 5 more units of asset B
than a subject who reported not having this habit. However, the inclusion of this variable did not affect the results on the treat-
ment effects and the CRT score established in the regression from column (2).
Next, since the mathematics score was not found significant, I further investigate the effect of the cognitive ability on the
sunk-cost fallacy using only the CRT score. To this end I interact the treatment dummies with the standardized CRT score.
The results are presented in column (4). Compared to the previous specifications, the treatment effects remain unchanged,
both in magnitude and significance level. However, the CRT score has no effect on the manifestation of the sunk-cost fallacy,
as the interaction coefficients are statistically insignificant. This is summarized in the following result:
Result 2. Cognitive ability has no effect on the manifestation of the sunk-cost fallacy.
Hence, Hypothesis 2 of this study is not confirmed. This result adds to the findings of Hoppe and Kusterer (2011) who
could not confirm an effect of the CRT test on the endowment effect, and concluded that the CRT has predictive power only
for biases that arise due to errors in reasoning and for which analytical skills are helpful in deriving the correct solution.
Finally, I split the experimental sample into a high and a low-cognitive ability group, according to the number of correct
answers in the CRT. Precisely, a score of 2 or 3 correct answers belongs to the high-cognitive-ability group (76 subjects) and a
score of 0 or 1 correct answers belongs to the low-cognitive-ability group (51 subjects). Fig. B.2 presents the treatment aver-
ages of the number of units of asset B by cognitive ability group. For each of the treatment conditions, the high-cognitive-
ability subjects bought significantly more units of asset B than the low-ability subjects (LSC: p¼:06, HSC: p¼:012, CT:
p¼:006). In fact, the average of the low-cognitive-ability CT group is below that of the high-cognitive-ability HSC group,
though not statistically significant (p¼:22). This indicates that the low-cognitive-ability subjects are affected by the
status-quo bias, though statistically significant treatment differences are confirmed for the HSC group (p¼:036). Further,
I conduct a non-parametric analysis of the treatment differences within the group of high-cognitive-ability subjects only.
Statistically significant differences in the number of units of asset B are, indeed, confirmed between the CT group and the
LSC group (p¼:075) and between the CT group and the HSC group (p¼:039), but not between the two sunk-cost groups
(p¼:306). Hence, the following result can be established:
Result 3. There is evidence of the sunk-cost fallacy within the group of high-cognitive-ability subjects. However, this is
independent of the size of the sunk cost.
5. Discussion and caveats
Although statistically significant evidence of the sunk-cost fallacy was found only for the HSC treatment, the sign of the
treatment effect for the LSC group is also consistent with the fallacy. These treatment differences survived in a regression
controlling for cognitive ability and socio-demographic variables. Moreover, the statistically significant test of a trend across
the three experimental groups supports the sunk-cost fallacy hypothesis. These results are surprising in light of the obvious
character and the simplicity of the experimental design, but suggest that sunk-cost fallacy could manifest itself even in the
absence of the psychological roots usually discussed in the literature, i.e. loss aversion and cognitive dissonance.
In this experiment, the sunk-cost fallacy was identified by the reluctance of giving up on the initial holdings of asset A for
a lower price than the purchase price. An explanation for this reluctance can be adapted from the realization utility theory
developed by Barberis and Xiong (2012). According to the authors, people feel a burst of pleasure when a gain is realized
and a burst of pain when a loss is realized, right in the moment of its realization. In other words, people derive utility
not only from consumption of goods and services, as economic models often assume, but also from the mere act of selling
an asset at a gain, right in the moment of executing the sale.
12
This line of argument seems to also explain the reluctance of the subjects to part with asset A when offered a price
below the purchase price.
13
There were significantly fewer subjects in the HSC treatment as compared to both the CT group
and the LSC treatment who sold at least 1 unit of asset A (37% compared to 86% and 70% in the CT and LSC groups, respec-
tively). Hence, despite recognizing the optimality of selling the inventories of asset A, the subjects in the sunk-cost treatments
11
While gender, field of study and years of education where also reported in the final questionnaire, they were not found significant in the regressions.
12
This theory was confirmed by Frydman, Barberis, Camerer, Bossaerts, and Rangel (2012) using an experimental stock market in which they scanned
subjects’ brain activity at the moment of submitting their trading decisions.
13
Realization utility has also been proven suitable for explaining the closely related, but different, disposition effect (Barberis & Xiong, 2012). While the
disposition effect explains the greater propensity to selling ‘‘winning” stocks rather than ‘‘losing” stocks, the sunk-cost fallacy is a phenomenon that accounts
for the size of the loss.
50 C. Haita-Falah / Journal of Economic Psychology 58 (2017) 44–59
failed to do so completely and only chose to sell intermediate amounts of their inventories. This is consistent with subjects
being willing to experience the pain from the realization of a loss only up to a point, i.e. the existence of a threshold for the
realization utility.
The surprising evidence of the sunk-cost fallacy is, however, weakened by the presence of a high degree of status-quo bias
among the experimental subjects. Indeed, a post-estimation coefficients test shows that the constant in the regressions from
Table B.4 is significantly different from 30, the number of units corresponding to rational behavior. The regressions from
Table B.4 suggest that the status-quo bias is attributed to cognitive ability. The strong status-quo bias exhibited by the group
of low-cognitive-ability subjects (see Fig. B.2), which represents about 40% of the experimental sample, partially explains the
low treatment differences between the CT group and the sunk-cost groups. Therefore, I conjecture that had the status-quo
bias been less severe, the sunk-cost bias would have been more pronounced.
Status-quo bias was also observed at the Investment stage of the experiment. From the total of 96 subjects who
were offered to invest (the sunk-cost treatments), 3 subjects in the LSC group and 8 subjects in the HSC group chose
to keep the initial cash endowment. This choice occurred despite the fact that the experimental parameters guaranteed
that investing was always profitable and despite the experimental instructions emphasizing the benefit of investing.
This situation raises worries of self-selection bias.
14
However, the decision to invest in asset A is, in this experiment,
solely motivated by the sure profit resulting from the difference between the purchase price and the redemption value.
Hence, investing as opposed to not investing is the rational payoff-maximizing choice in both sunk-cost treatments and it
is therefore, not driven by self-selection. Moreover, those subjects who chose not to invest were asked to remain in the
laboratory for the second part of the experiment, which also excludes the opportunity cost of time as an explanation for
their decision.
It is, nevertheless important to asses the extent to which the self-selection could affect the results. For this, note that cog-
nitive ability is the main difference (p¼:04) between the subjects who invested and the 11 attriters. This difference is driven
by the CRT (p¼:02) and it is also confirmed by the Probit regression presented in Table B.5 (column (1)). Apart from the CRT
score, the decision to invest is not explained by any of the co-variates collected via the final questionnaire (column (2)).
Therefore, the refusal to engage in the cognitive effort entailed by the continuation of the experiment appears to be the most
pertinent explanation for these subjects’ attrition. This is, in turn, predictive of status-quo bias. Moreover, the irrational
behavior they exhibited at the Investment stage is an indicator of potential further irrational behavior with respect to sunk
costs. Therefore, had they been forced to invest the 11 subjects would have, in fact, increased the gap between the CT group
and the two treatment groups, thus strengthening the evidence for sunk-cost fallacy.
6. Conclusions
This paper presents a laboratory experiment to test for the manifestation of the sunk-cost fallacy. The experimental
manipulation consists of two groups which differ with respect to the size of the sunk cost, low and high, and an additional
control group which incurs no sunk cost. Despite the obviousness of the optimal course of action and the absence of the psy-
chological drivers of the bias, the data shows behavior consistent with the manifestation of the sunk-cost fallacy. However,
this behavior is statistically significant only for the high-sunk cost treatment. This result survives controlling for other
covariates in a regression analysis. Therefore, the first result of this study is that sunk-cost fallacy may manifest itself even
in the absence of the psychological mechanisms that typically explain it.
The second goal of this experiment was to understand the relationship between cognitive ability and sunk-cost fallacy.
First, the CRT score was found to account for the status-quo bias, but had no effect on the sunk-cost bias. Second, non-
parametric analysis of treatment differences confirms the sunk-cost fallacy hypothesis for the group of high-cognitive-
ability subject, who were also less status-quo biased.
Acknowledgments
This research was funded by the University of Hamburg, Cluster of Excellence ‘‘Integrated Climate System Analysis and
Prediction” (CliSAP). The sessions reported in the Supplementary Appendix were financially supported by the University of
Kassel. The funding sources have no involvement in the collection, analysis and interpretation of the data or in the writing of
the paper. I thank Jana Freundt for help in the lab and for proofreading the manuscript. I am indebted to Botond K}
oszegi and
Anke Gerber for comments and support.
Appendix A. Theoretical framework
Let us assume that there are two types of assets in the economy: asset A and asset B. Next, let us suppose that the decision
maker has initially invested in q
A
0
units of asset A at the price p
A
0
per unit. As explained in the text, once the investment is
completed she learns the unit price of asset B, p
B
, the trade price of asset A, p
A
1
per unit and the number of units she must
14
Self-selection is an issue in identifying the sunk-cost fallacy because those subjects who choose to buy an asset do so because they are also more likelyto
use it.
C. Haita-Falah / Journal of Economic Psychology 58 (2017) 44–59 51
accumulate, Q>q
A
0
, in any combination of asset A and B. Finally, the redemption value of each the Qunits is p, regardless of
being of type A or B. Let q
A
1
be the number of units of asset A she decides to sell (q
A
1
60) or buy (q
A
1
P0) and q
B
P0 the
number of units of asset B she decides to buy. The final payoff of the decision maker is given by the revenue from holding
the Qunits of assets minus the cost of buying asset B, minus the cost (plus the revenue) from trading asset A and minus the
sunk cost. A rational decision maker chooses q
A
1
and q
B
to maximize this payoff. Formally, this writes:
max
q
A
1
;q
B
P
¼pQ p
B
q
B
p
A
1
q
A
1
p
A
0
q
A
0
such that
Q¼q
A
0
þq
A
1
þq
B
q
A
1
Pq
A
0
and q
B
P0
ðA:1Þ
Two cases must be analyzed. First, if p
B
Pp
A
1
, then q
A
1
¼Qq
A
0
and q
B
¼0. Second, if p
B
<p
A
1
, then q
A
1
¼q
A
0
and q
B
¼Q. The
former case implies that if asset A is cheaper than asset B, then it is optimal for the decision maker to keep the initial units of
asset A and buy more units of the same asset to complete the Qunits. However, only the latter case allows for the identi-
fication of the sunk-cost fallacy because it predicts the total abandonment of the initial investment. Therefore, the experi-
mental parameters are chosen to generate this situation. In addition, the re-sale price of asset A is chosen to be below
the initial purchase price, p
A
1
<p
A
0
, such that part of the initial investment remains sunk.
Appendix B. Tables and figures
See Tables B.1–B.5 and Figs. B.1 and B.2.
Table B.1
Experimental parameters.
Parameters Experimental condition
CT LSC HSC
Initial cash endowment (EE) 200 1000 1400
Initial price of A per unit (EE) 0 40 60
Initial endowment of A (units) 20 0 0
Assets A and/or B to hold at the end (units) 30 30 30
Trading price of A (EE) Random in f16;17;18;19;20g
Price of B (EE) 10 10 10
Redemption value of A and B (EE) 70 70 70
Table B.2
Descriptive statistics: means by treatment.
Treatment Control LSC HSC p-value
N 424342
Males .45 (.077) .58 (.076) .33 (.073) .07
Cognitive score 4.62 (.187) 4.65 (.182) 4.45 (.174) .64
CRT score 1.76 (.163) 1.79 (.154) 1.64 (.155) .75
Math score 2.85 (.055) 2.86 (.063) 2.81 (.061) .66
Behav. Econ. .17 (.058) .16 (.057) .12 (0.50) .79
Stock market .29 (.070) .19 (.060) .17 (.058) .36
Note: Standard errors in parentheses. ‘‘Behav. Econ.” and ‘‘Stock market” are self-reported variables indicating whether the subject attended a Behavioral
Economics course and reads financial newspapers or follows the stock market, respectively. p-value is reported from the Kruksal-Wallis equality-of-
populations rank test.
Table B.3
Decision variables (averages).
N Purchases of Asset B Sales of Asset A Purchases of Asset A
(1) (2) (3)
Control 42 21.83 (1.49) 12.43 (1.31) .59 (.35)
LSC 43 19.18 (1.59) 10.47 (1.29) 1.28 (.50)
HSC 42 16.26 (1.67) 8.09 (1.31) 1.83 (.57)
Note: Standard errors in parentheses.
52 C. Haita-Falah / Journal of Economic Psychology 58 (2017) 44–59
Appendix C. Experimental instructions (translation from German)
Welcome to the Experiment!
This is an experiment in the economics of decision making. It is very important that you read these instructions carefully.
If you follow the instructions and make good decisions, you will earn a considerable amount of money. All earnings on your
computer screen are in Experimental Euro (EE) and they will be converted in real euro at the exchange rate:
200 EE ¼1 euro:
The experimental session consists of two independent parts. Your decisions and earnings from one part do not affect your
decisions and earnings of the other part. Everyone in this room will participate in both parts of the experiment. You will first
receive instructions for Part 1 and then make your decision at the computer terminal. After this, Part 1 is done. Next, you will
receive instructions for Part 2 and again make your decision for Part 2. However, each part will start only after everyone has
Table B.4
GLS regression results.
Dependent variable: Units of asset B
(1) (2) (3) (4)
Constant 21.956 21.494 20.063 20.186
(1.479)
⁄⁄⁄
(1.408)
⁄⁄⁄
(1.480)
⁄⁄⁄
(1.488)
⁄⁄⁄
LSC 2.075 2.254 1.905 2.052
(2.154) (1.985) (1.98) (2.008)
HSC 5.635 5.058 4.432 4.509
(2.229)
⁄⁄
(2.060)
⁄⁄
(2.009)
⁄⁄
(2.016)
⁄⁄
CRT 4.158 4.336 3.331
(0.872)
⁄⁄⁄
(0.860)
⁄⁄⁄
(1.377)
⁄⁄
Math 0.745 0.654 0.703
(0.844) (0.838) (0.844)
Stock market 4.815 4.772
(2.013)
⁄⁄
(2.047)
⁄⁄
LSC X CRT 1.199
(2.038)
HSC X CRT 1.997
(2.03)
N 127 127 127 127
Note: Heteroskedasticity robust standard errors in parenthesis. CT is the baseline category.
p<:10.
⁄⁄
p<:05.
⁄⁄⁄
p<.01.
Table B.5
The investment decision.
Dependent variable Invest (1 if invested, 0 if not invested)
(1) (2)
CRT score 0.404 0.334
(0.162)
⁄⁄
(0.179)
⁄
Math score 0.239 0.251
(0.408) (0.434)
Male (1 if male, 0 if female) 0.264
(0.391)
Read financial newspapers 0.354
(0.483)
Trimester of study 0.117
(0.147)
Difficulty of the experimental tasks 0.199
(0.243)
Constant 1.314 1.405
(1.163) (1.415)
N9696
Note: Heteroskedasticity robust standard errors in parentheses.
p<0:01.
⁄
p<0:1.
⁄⁄
p<0:05.
C. Haita-Falah / Journal of Economic Psychology 58 (2017) 44–59 53
made their decisions for the current part. Your final earnings will be the sum of your earnings from both parts and they will
be privately paid to you in cash at the end of the session. After both parts of the experiment are completed, you will be asked
to answer some general questions.
Important rules:
1. From my side: NO DECEPTION. I promise that this experiment will be conducted exactly as described in these instruc-
tions. This is the rule in economics experiments. Without this rule the results of the research cannot be published.
2. From your side: NO COMMUNICATION. This is an experiment on individual decision making. Your earnings in this exper-
iment are NOT affected by the decisions of any other participant and your decisions do NOT affect the earnings of any
other participant in this experiment. Therefore, please do not communicate with other participants during this experi-
ment and take your decisions individually. If you have any question during the experiment, please raise your hand
and you will receive assistance.
0.01 .02 .03 .04
Density
010 20 30
Units of Asset B
CT LSC HSC
Fig. B.1. Distribution of units bought from Asset B.
18.35
15.33
12.16
24.2
21.25
19.65
0 5 10 15 20 25
CT LSC HSC CT LSC HSC
Low−cognitive ability group High−cognitive ability group
Units of asset B (means)
Fig. B.2. Average units of Asset B by cognitive ability groups.
54 C. Haita-Falah / Journal of Economic Psychology 58 (2017) 44–59
Part 1: General Instructions
In this part of the experimental session there are two types of assets, A and B. Your task will be to collect a required num-
ber of units of assets, in any combination of A and B that you wish. This means that you can have only asset A or only asset B,
or any other combination of the two assets which gives you exactly the required total number of units of assets. You will
learn the required number of units you must collect during the experiment. The stages of the experiment proceed as follows.
In the first stage of the experiment you will be endowed with an amount of cash in EE [Control condition: and a number of
units of asset A. This number will be lower than the required number of units you must collect.] [Sunk-cost conditions:
You will be offered the possibility to invest part of your cash endowment in order to purchase a certain number of units of
asset A. This number will be lower than the required number of units you must collect. If you decide to invest, you will
continue the experiment with further decisions. If you decide not to invest, you will keep your cash endowment, but you
will be asked to wait quietly in your seat for the next part of the experiment.]
[Sunk-cost conditions: If you have purchased asset A in the first stage,] In the next stage, a market will open in which you
will have the possibility to trade, i.e. to buy or sell asset A and to buy asset B. You will learn the prices of asset A and asset
B once the market opens. At this stage you will see a table like the one below:
In column ‘Asset A’ you can decide how many units of asset A to buy or how many units of asset A to sell by entering a
number in the corresponding box. Note that you cannot buy and sell asset A at the same time. Please leave empty or enter
0 in the box you do not want to use. If you want to neither buy nor sell asset A (that is, keep all units you currently hold),
please leave empty or enter 0 in both boxes. If you choose to sell, you cannot sell more units of asset A than you hold in
your account. If you choose to buy, you cannot buy more units of asset A than the number of units you need in order to
have the required total number of units of assets. In column ‘Asset B’ you only have the option to buy. Note that you can-
not buy more units of asset B than the number of units you need in order to have the required total number of units of
assets. If you do not want to buy any unit of asset B, please leave the box empty or enter 0.
Regardless how you decide to trade, please make sure that the total number of units of assets you hold at the end
equals the required total number of units of assets. Note that regardless of the type of asset you want to buy, you will
always have enough cash in your account to buy as many units of assets as you want and are allowed to.
As a result of your trading decision, your account will contain a final amount of cash endowment and the required total
number of units of assets. At the end of the session, the experimenter will buy all your collected units at the same
redemption value of 70 EE per unit, regardless of the type of assets you hold. Your payoff for this part of the experiment
will be calculated as:
Payoff ¼½final cash endowmentþ½redemption value½required total number of units:
Fig. D.1. Initial position screen.
C. Haita-Falah / Journal of Economic Psychology 58 (2017) 44–59 55
Before the experiment begins, please answer the following questions on your screen to make sure you have a good
understanding of the process and the decisions in the experiment.
Appendix D. Experimental screens (translation from German)
See Figs. D.1–D.5.
Appendix E. The cognitive quiz
Question 1: A bat and a ball cost 1.10 euro in total. The bat costs 1.00 euro more than the ball. How much euro does the
ball cost?
Question 2: Which number is larger?
(A) 250
(B) ð800 1=2Þþð01=2Þ
Fig. D.2. Investment screen.
Fig. D.3. Investment confirmation screen.
56 C. Haita-Falah / Journal of Economic Psychology 58 (2017) 44–59
Question 3: y and z are two numbers with the following properties: If we subtract two from y, z is obtained and by mul-
tiplying y and z, we obtain 48. Which of the following CANNOT be NEITHER y NOR z?
(A) 6
(B) 8
(C) 12
(D) 6
(E) 8
Question 4: If it takes 5 machines 5 minutes to make 5 widgets, how long would it take 100 machines to make 100 wid-
gets (in minutes)?
Fig. D.5. Trade screen – confirmation.
Fig. D.4. Trade screen – decision.
C. Haita-Falah / Journal of Economic Psychology 58 (2017) 44–59 57
Question 5: Which number is larger?
(A) 250
(B) ð200 1=2Þþð01=2Þ
Question 6: On a lake there is a patch of lily pads. Every day, the patch doubles in size. If it takes 48 days for the patch to
cover the entire lake, how long would it take for the patch to cover half of the lake (in days)?
Appendix F. Supplementary Appendix
Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.joep.
2016.12.001.
References
Arkes, H. R., & Blumer, C. (1985). The psychology of sunk cost. Organizational Behavior and Human Decision Processes, 35(1), 124–140.
Ashraf, N., Berry, J., & Shapiro, J. M. (2010). Can higher prices stimulate product use? Evidence from a field experiment in Zambia. American Economic Review,
100, 2383–2413.
Barberis, N., & Xiong, W. (2012). Realization utility. Journal of Financial Economics, 104, 251–271.
Bazerman, M. H., Giuliano, T., & Appelman, A. (1984). Escalation of commitment in individual and group decision making. Organizational Behavior and
Human Performance, 33(2), 141–152.
Benjamin, D. J., Brown, S. A., & Shapiro, J. M. (2006). Who is ‘Behavioral’? Cognitive ability and anomalous preferences. Harvard University.
Bock, O., Nicklisch, A., & Baetge, I. (2012). ‘hroot: Hamburg registration and organization online tool.’ WiSo-HH working paper series no. 1.
Bruine de Bruin, W., Parker, A. M., & Fischhoff, B. (2007). Individual differences in adult decision-making competence. Journal of Personality and Social
Psychology, 92(5), 938–956.
Fischbacher, U. (2007). z-Tree: Zurich toolbox for ready-made economic experiments. Experimental Economics, 10(2), 171–178.
Frederick, S. (2005). Cognitive reflection and decision making. Journal of Economic Perspectives, 19(4), 25–42.
Friedman, D., Pommerenke, K., Lukose, R., Milam, G., & Huberman, B. A. (2007). Searching for the sunk cost fallacy. Experimental Economics, 10(1), 79–104.
Frydman, C., Barberis, N., Camerer, C., Bossaerts, P., & Rangel, A. (2012). ‘Using neural data to test a theory of investor behavior: An application to realization
utility’. Yale ICF working paper no. 12-30.
Harrison, G. W., & List, J. A. (2004). Field experiments. Journal of Economic Literature, XLII, 1009–1055.
Hoppe, E. I., & Kusterer, D. J. (2011). Behavioral biases and cognitive reflection. Economic Letters, 110(2), 97–100.
Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decisions under risk. Econometrica, 47(2), 263–291.
Larrick, R. P., Nisbett, R. E., & Morgan, J. N. (1993). Who uses the cost-benefit rules of choice? Implication for the normative statues of microeconomics
theory. Organizational Behavior and Human Decision Processes, 56, 331–347.
Oechssler, J., Roider, A., & Schmitz, P. W. (2009). Cognitive abilities and behavioral biases. Journal of Economic Behavior and Organization, 72, 147–152.
Parker, A. M., & Fischhoff, B. (2005). Decision-making competence: External validation through an individual-differences approach. Journal of Behavioral
Decision Making, 18, 1–27.
Phillips, Q. R., Battalio, R. C., & Kogut, C. A. (1991). Sunk and opportunity costs in valuation and bidding. Southern Economic Journal, 58(1), 112–128.
Stanovich, K. E., & West, R. F. (2008). On the relative independence of thinking biases and cognitive ability. Journal of Personality and Social Psychology, 94(4),
672–695.
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.
Fig. D.6. Payoff screen.
58 C. Haita-Falah / Journal of Economic Psychology 58 (2017) 44–59
Strough, J., Mehta, C. M., McFall, J. P., & Schuller, K. L. (2008). Are older adults less subject to the sunk-cost fallacy than younger adults? Association for
Psychological Science, 19(7), 650–652.
Tan, H.-T., & Yates, J. F. (1995). Sunk cost effects: The influences of instructions and future return estimates. Organizational Behavior and Human Decision
Processes, 63(3), 311–319.
Thaler, R. (1980). Toward a positive theory of consumer choice. Journal of Economic Behavior and Organization, 1, 39–60.
Toplak, M. E., West, R. F., & Stanovich, K. E. (2011). The cognitive reflection test as a predictor of performance on heuristics-and-biases tasks. Memory and
Cognition, 39, 1275–1289.
C. Haita-Falah / Journal of Economic Psychology 58 (2017) 44–59 59
