MANAGEMENT SCIENCE
Vol. 50, No. 5, May 2004, pp. 575–586
issn 0025-1909 eissn 1526-5501 04 5005 0575
informs®
doi 10.1287/mnsc.1030.0148
© 2004 INFORMS
Keeping Doors Open: The Effect of Unavailability on
Incentives to Keep Options Viable
Jiwoong Shin, Dan Ariely
Massachusetts Institute of Technology, 38 Memorial Drive, Cambridge, Massachusetts 02142
{jishin@mit.edu, ariely@mit.edu}
Many of the options available to decision makers, such as college majors and romantic partners, can become
unavailable if sufficient effort is not invested in them (taking classes, sending flowers). The question
asked in this work is whether a threat of disappearance changes the way people value such options. In four
experiments using “door games,” we demonstrate that options that threaten to disappear cause decision makers
to invest more effort and money in keeping these options open, even when the options themselves seem to be
of little interest. This general tendency is shown to be resilient to information about the outcomes, to increased
experience, and to the saliency of the cost. The last experiment provides initial evidence that the mechanism
underlying the tendency to keep doors open is a type of aversion to loss rather than a desire for flexibility.
Key words: options; aversion to loss; search
History: Accepted by Martin Weber, decision analysis; received March 12, 2002. This paper was with the
authors 8 months for 3 revisions.
Introduction
Imagine a student who is uncertain about whether
he wants to become a computer programmer or a
poet. If he wants to keep both options available, he
has to keep taking classes in both majors. On the
other hand, keeping both options open has its own
cost. Double majoring implies that the student has to
divide his time and effort and take classes in both
fields—leading him to become proficient in both, but
an expert in neither. Along similar lines, consider a
person pursuing two potential relationships. As long
as this romantic decision maker spends sufficient time
with each of her potential romantic partners, she can
keep them both as viable future relationships. However,
once she starts spending more time with one and
neglecting the other, the neglected party is likely to
move on and become unavailable. Given the possible
loss of the second romantic option, our enthusiastic
dater might try to spend at least some of her time
with her less-preferred partner, largely to maintain
the viability of the relationship. However, much like
the student with the double major, “keeping doors
open” has its costs, drawing valuable time and energy
away from the more promising relationship.
Double majoring and dating are just two examples
of cases where one must invest extra time and effort
to keep options available. The main questions asked
here are whether the threat of future unavailability
makes less-desirable options seem more appealing
and whether this causes individuals to overinvest in
these options. In other words, do doors that threaten
to close appear more attractive than doors that remain
open? And if so, will individuals overinvest just to
keep them open?
From a naive, rational perspective, one could expect
that the value of an option (having the ability to make
a choice) would be based solely on the expected utility
of the outcomes it represents. From a psychological
perspective, however, there are two primary reasons
why the subjective value of an option can exceed its
expected value: a desire for flexibility and aversion
to loss.
Initial evidence for the value of flexibility was proposed
by Brehm (1956), who showed that people are
willing to sacrifice consumption pleasure to increase
freedom of choice (see also Simonson 1990, Gilbert
and Ebert 2002). The desire for flexibility is not limited
to humans; even pigeons exhibit it (Catania 1975).
Such preference for flexibility implies that individuals
can get utility (pleasure) from simply “having the
right to choose” (keeping options open) prior to making
a final choice.
Evidence for aversion to loss dates back to
Kahneman and Tversky (1979).1
The most relevant
application of this aversion to loss is the case
1
The general reluctance to give up (aversion to loss) is related to
loss aversion (Tversky and Kahneman 1991). However, this general
reluctance to give up does not require any comparison between the
gains and losses. General reluctance to give up can, therefore, be
regarded as a related phenomenon capturing the general human
tendency to try to avoid losses. For instance, the endowment effect
can be seen as related to both loss aversion and aversion to loss.
575
Shin and Ariely: Effect of Unavailability on Incentives to Keep Options Viable
576 Management Science 50(5), pp. 575–586, © 2004 INFORMS
of endowment effect (Kahneman et al. 1990, 1991;
Bar-Hillel and Neter 1996; Carmon and Ariely
2000), showing that ownership, or even deliberation
(Carmon et al. 2003), can increase attachment
and hence valuations. Support for aversion
to loss was also provided in the context of risky
choice, in particular the rejection of a pair of mixed
gambles (Markowitz 1952, Williams 1966). Although
options for items are very different from the items
themselves—for example, the possibility of dating a
person is a very different experience from actually
dating that person—and although it is not possible to
own an option in the same way it is to own an item,
losing an option (opportunity loss) is closely related
to the loss of an item. Namely, the loss of an option
also implies the loss of the item. Based on this similarity
in terms of loss and the large influence of loss
on decision making (Tversky and Kahneman 1991),
it can be argued that individuals will also experience
the general aversion to loss and a pseudo-endowment
effect for options. The general aversion to loss implies
that the utility that individuals get from simply having
the “right to choose” (keeping options open) is not
a utility, but rather disutility or pain that can accompany
the loss of options.
In summary, the current work asks two questions:
First, whether the threat of unavailability increases
the perceived value of an option; and second, if so,
whether the higher valuation comes from a desire for
flexibility or from aversion to loss. Four experiments
were designed to provide initial answers to these
questions.
The Experiments: General
Because all four experiments employ the same basic
design, it is simpler to first describe the overall
paradigm (the “game”) and provide more details
about specific differences as they pertain to the individual
experiments.
The general structure of the game involved a
sequential search task (Camerer 1995, Ratchford
and Srinivasan 1993, Zwick et al. 2000), in which
respondents were faced with multiple alternatives,
each associated with a different payoff distribution.
Respondents playing the game faced a dilemma
similar to many real-life search tasks: They wanted to
maximize their earnings by finding the best alternative
(payoff in this game is based solely on performance),
yet search is costly. Thus, respondents had
to trade off the possible value of additional searching
against its cost to determine their stopping rule (Saad
and Russo 1996).
As a metaphor for “keeping options open,” we
created a computer game with three doors to three
rooms (for a schematic illustration of the game, see
Figure 1 Schematic Illustration of the “Door Game”
Main Screen: The Three Doors
Door
1
Door
2
Door
3
Door
1
Door
2
Door
3
ClickClickClick
Note. Respondents first encountered three doors to three rooms. Clicking
on any door opened that door, allowing the respondents to either click within
that room or move to another room. Clicking in a room resulted in a payoff
randomly sampled from the distribution of that room. Moving to a different
room cost the respondents a click. Respondents were given a total click budget,
and the experiment was completed when the click budget was depleted.
Figure 1). One door was red, another blue, and the
third green. By clicking with the mouse on one of
the doors (door-click), respondents opened that door
and entered the room. Once in the room, respondents
could either click in that room (room-click) or click
on a door to a different room (door-click). Each roomclick
resulted in a payoff gain sampled randomly
from that room’s distribution, and each door-click
transferred the respondent to another room (without
a payoff). Respondents were given a click budget to
use on door- and room-clicks as they wished. Once
respondents used all their clicks, the game was over
and they were paid the sum of their door-click payoffs.
Note that charging the respondent a click to
switch rooms created a switching cost. The total number
of clicks was indicated clearly on the screen, in
terms of both how many clicks the respondent had
used and how many clicks they had left until the end
of the experiment.
The main manipulation of interest was the relationship
between the actions of the player and door availability
(option availability), which was varied on two
levels: constant availability and decreased availability. In
the constant-availability conditions, all three rooms
remained as viable options throughout the experiment,
irrespective of the action of the respondent.
In the decreased-availability conditions, availability
depended on the action of the respondent. Every time
a respondent clicked either on a door or within a
room, the doors to the other two rooms were reduced
in size by 1/15 of their original width. A single
Shin and Ariely: Effect of Unavailability on Incentives to Keep Options Viable
Management Science 50(5), pp. 575–586, © 2004 INFORMS 577
door-click on a shrinking door revitalized it to its
original size and the process continued. Once the size
of a door reached zero, it was eliminated for the rest
of the game. With this shrinking factor, an option
(room) that was not clicked on within 15 clicks was
eliminated and was no longer visible or available.2
In sum, at each point, respondents had to decide
whether to remain with their current choice or to
continue searching while incurring switching costs.
In addition, respondents in the decreased-availability
condition also had to decide whether to invest in
options that threaten to disappear to maintain their
viability.
The analogy between the experimental game and
the examples presented earlier should be clear. The
three doors represent different academic or romantic
options. In the decreased-availability conditions, the
viability of an option is threatened when there is no
investment in, or attention to, that option. Moreover,
after a certain amount of neglect, options become
unavailable, a state that is irreversible.
Experiment 1: Effect of
Decreased Availability
Experiment 1 was designed to determine whether
the mere fact that options could become unavailable
would influence decision makers’ behavior. Our
hypothesis was that the decreasing-availability condition
would cause respondents to invest in keeping
options viable. By providing an initial answer to the
question of whether people switch rooms more often
when there is a threat of disappearance, Experiment 1
served as a starting point for examining the possible
motivation to invest in keeping options open.
Method
Respondents. Advertisements were placed around
campus to recruit 157 respondents, including some
from within the computer lab where the experiment
took place. The experiment lasted about 15 minutes.
Respondents were randomly assigned to one of
the two option-availability conditions (constant and
decreased availability).
Design. The overall structure of the game was as
described in the general description of the game. For
this experiment, the expected value of each room-click
was 3¢, but the three rooms were associated with
three different distributions (Table 1). Door 1 was
highly concentrated around mean 3 (normal with
2
For a robustness test, we manipulated this visual saliency of
the disappearance of the doors in a separate experiment. The
results showed that there was no observable impact on the player’s
actions, suggesting that the effect of availability was not due to the
visual saliency that was used in this game.
Table 1 Distributions of Payment in the Three Doors Across the Four
Experiments
Experiment #
(clicks) Manipulation Door 1 Door 2 Door 3
Experiment 1 Option availability
(100) Distribution Normal Normal Chi-square
Average ¢/variance 3/2 25 3/0 64 3/10
Min ¢/Max ¢ 0/7 1/5 −2/10
Experiment 2 Information level
(50) Distribution Normal Normal Chi-square
Average ¢/variance 6/9 6/2 25 6/16
Min ¢/Max ¢ 0/14 2/9 −4/19
Experiment 3 Saliency of the cost
(100) Distribution Normal Normal Chi-square
Average ¢/variance 10/9 10/2 25 10/20
Min ¢/Max ¢ 4/18 6/13 0/20
Experiment 4 Reactivation
(100) Distribution Normal Normal Normal
Average ¢/variance 2 5/1 25 3/1 25 3 5/1 25
Min ¢/Max ¢ −0 6/5 9 0 1/6 9 1 2/8 1
Note. Door 3 was a chi-square distribution with a degree of freedom, which is
larger than the expected mean by 2¢. We subtracted 2¢ from the distribution
to keep the same average, but encounter a few negative outcomes. For example,
in Experiment 1, door 3 was a chi-square distribution with 5 degrees of
freedom, where we subtracted 2¢.
variance 0.64); door 2 was symmetric around the same
mean, but much more diffused (normal with variance
2.25); and door 3 was highly skewed toward
high numbers (chi square with 3 degrees of freedom).
The payoff distributions across these three rooms
ranged from −2¢ to 14¢, with the lower numbers
being more frequent than the higher numbers (so that
the mean value was 3¢). Respondents were given
a total of 100 clicks in the experiment, which they
could allocate as they saw fit between switching
rooms (door-clicks) and getting payoffs within a room
(room-clicks).
Procedure. Upon arrival at the lab, respondents
were seated individually and given instructions for
the game. All respondents received instructions that
emphasized that their goal in the experiment was
to make as much money as possible and that the
amount they made would be paid to them at the
end of the experiment. In the decreased-availability
condition, respondents were also given the description
of the rules governing the shrinking, revitalizing,
and disappearance of the doors. The instructions
did not include any information about the different
payoff distributions of the three doors; respondents
had to learn about the distributions while playing
the game.
Results and Discussion
First, we compared how door-switching behavior
varied across the two conditions. A comparison of
the average number of room switches (door-clicks)
Shin and Ariely: Effect of Unavailability on Incentives to Keep Options Viable
578 Management Science 50(5), pp. 575–586, © 2004 INFORMS
revealed that switching was more likely to occur
in the decreased-availability condition M = 16 70
than in the constant-availability condition (M = 7 47;
t 156 = 7 82, p < 0 001).
Next, we examined how the tendency to switch
rooms in the two option-availability conditions
changed as a function of the total number of clicks
used (click number). Note that the click number is a
measure of both the learning and the expected value
of keeping options open, both reducing the motivation
for switching. First, as the click numbers increase,
respondents have more experience, better estimation
of the distributions, and thus a reduced need to
explore the different options. Second, the expected
benefit of exploring different options is reduced with
the click number because the time horizon during
which this information can be used is reduced.
To analyze the effect of the click number, clicks
were divided into 10 blocks of 10 clicks each. An
overall 2 (option-availability) by 10 (block) ANOVA
revealed a significant main effect for option availability
(F 1 1550 = 306 27, p < 0 0001), a significant main
effect for block (F 9 1550 = 5 61, p < 0 0001), and a
significant interaction effect between option availability
and block (F 9 1550 = 3 82, p = 0 0001). As can
be seen in Figure 2, there was a decreased tendency
to switch rooms later in the game. However, even in
the last block of 10 clicks, more switching occurred
in the decreased-availability condition (M = 1 27)
than in the constant availability condition (M = 0 75;
F 1 155 = 8 23, p = 0 0047). More important, there
were interesting differences in how the tendency to
open other doors changed as a function of block in the
two conditions, as indicated by the interaction. In particular,
while respondents in the constant-availability
condition switched the most during the first block,
respondents in the decreased-availability condition
switched the most during the second block—which
Figure 2 Average Number of Door Switches for Decreased- and
Constant-Availability Conditions Within Each Block of
10 Clicks in Experiment 1
Click Number (blocks)
BL-1 BL-2 BL-3 BL-4 BL-5 BL-6 BL-7 BL-8 BL-9 BL-10
#ofswitches
0.0
0.5
1.0
1.5
2.0
2.5
3.0 Decreased Availability
Constant Availability
D
C
Note. Error bars are based on standard errors.
was the first time they encountered a threat of option
elimination.
It is worth contrasting the behavior of the respondents
to an optimal strategy benchmark, which in
this experiment was to select a single room and
remain there during the entire game, which would
have earned the highest possible payoff due to the
implicit opportunity cost of 3¢ for each room switch
(door-click). Relative to this standard, the respondents
in Experiment 1 gave up 11% of their profits (8%
in the constant-availability condition and 14% in the
decreased-availability condition) as a consequence of
switching rooms, which occurred on the average of
12 times per respondent. Note that in this experiment,
respondents had to discover the underlying
payment distribution based on experience, and therefore
had to switch to learn about the doors—that is,
payoffs. Accordingly, the reduction in payment cannot
be taken as evidence of any irrational behavior.
Experiment 2 more carefully examined normative
expected behavior in such cases.
In summary, Experiment 1 showed a main effect for
option availability. Decision makers’ interests in alternative
options seemed to increase when they were
threatened by their unavailability.
Experiment 2: Effects of Knowledge
on the Desire to Keep Doors Open
Although the results of Experiment 1 suggest that
the respondents were willing to invest to keep their
options open, it remains unclear as to whether this
investment can be classified as an overinvestment. It
is possible, for example, that in the face of uncertainty,
the optimal strategy is to keep options open until
sufficient information about distribution accumulates.
Experiment 2 manipulated the level of knowledge
respondents had about the distributions, the logic
being that if the reason for keeping options available
is lack of knowledge, providing respondents with
more information about payoff distributions should
eliminate, or at least substantially decrease, the difference
in switching between the decreased- and
constant-availability conditions. On the other hand, if
the tendency to keep options open is caused by mechanisms
such as preference for flexibility or aversion
to loss, providing additional information should not
influence the effects of option availability on room
switching.
Method
Respondents. Advertisements were placed around
campus to recruit 105 respondents, including some
from within the computer lab where the experiment
took place. Respondents were randomly assigned to
one of six conditions.
Shin and Ariely: Effect of Unavailability on Incentives to Keep Options Viable
Management Science 50(5), pp. 575–586, © 2004 INFORMS 579
Design and Procedure. The main manipulation in
Experiment 2 was a manipulation of information,
which was varied on three levels: no prior information,
practice information, and descriptive information, which
was crossed with the manipulation of option availability.
The distributions of the three rooms had the
same mean value of 6¢ (Table 1), and respondents
were allocated 50 clicks rather than 100 clicks as their
clicking budget. The no-prior-information conditions
were a basic replication of Experiment 1. In these
two conditions (constant and decreased availability),
respondents did not get any prior information about
the distributions. They were simply given the opportunity
to play the game. In the practice-information
condition, respondents played the same game twice,
first for 50 practice trials without getting paid, and
then for 50 real trials. Respondents were clearly
informed that the distributions associated with each
room were the same in the practice and real parts,
thus increasing their knowledge about these distributions
for the real part of the experiment (the part for
which they got paid). Respondents in the descriptiveinformation
condition were told that the averages
of the distributions of all three rooms were identical.
They were also shown a graph in which the
means, skewnesses, and variance of each distribution
were depicted. Although the respondents in the
descriptive-information condition knew the three distributions,
they did not know which room corresponded
to which distribution. Thus, if they were
not satisfied with the equal expected value across
the three rooms, they could have searched the three
rooms for their preferred distribution.
Results and Discussion
As in Experiment 1, the main dependent measure was
the frequency of room switches across the different
conditions, analyzed in a 2 (option-availability) by
3 (information) between-subjects ANOVA. The overall
ANOVA (Figure 3a) revealed a main effect for option
availability (F 1 99 = 56 66, p < 0 001), replicating
the main results of Experiment 1. The overall ANOVA
also revealed an effect for information (F 2 99 = 6 99,
p < 0 001), showing that the no-prior-information conditions
induced more switching than did the other
two conditions (F 1 101 = 12 78, p < 0 001), which
were not different from each other (F 1 61 = 1 85,
p = 0 18). Finally, the analysis showed a nonsignificant
interaction between option availability and information
(F 2 99 = 1 32, p = 0 27), demonstrating that
the addition of information did not change the effect
of option availability on switching behavior; that is,
respondents with no prior information about the distributions
exhibited the same reaction to the threat of
disappearance as respondents who had more information
(either descriptive or practice) about these
distributions. There were a few respondents who
wanted to end the experiment as fast as possible, not
switching rooms at all. These respondents increased
the standard errors in general but most profoundly
when the mean switching was higher, which is the
decreased-availability condition.
While these results demonstrate that additional
information does not reduce the effect of option
availability, they do not rule out rational explanations
for the observed effect. For example, had respondents
needed 15 clicks per room to learn its payoff distribution,
respondents in the decreased-availability conditions
would have had to switch rooms at least six
times, while respondents in the constant-availability
conditions would have had to switch only twice. To
examine more carefully such possible explanations,
we constructed three other measures: pecking, elimination
point, and click investment.
First, we examined pecking, the number of times
that respondents switched to another room, clicked
in that room once, and switched back (the result
remains the same if we define pecking as switching
to another room and switching back without clicking
inside the room, or as a combined measure).
From the perspective of gaining information about
the payoffs, we could consider such pecking behavior
as an irrational overinvestment in keeping options
open because it provides little information (one
more sample) at a high cost (three clicks—one
for switching away, one for sampling the payoffs,
and one for switching back). ANOVA analysis
revealed that pecking behavior was more frequent
in the decreased-availability condition (M = 0 36)
than in the constant-availability condition (M = 0 07;
F 1 99 = 5 97, p = 0 016), suggesting that in the face
of a threat that options could become unavailable,
respondents showed “irrational” behavior more often.
More important, the effect of information on pecking
was not significant (F 2 99 = 0 682, p = 0 508),
nor was the interaction between option availability
and information (F 2 99 = 0 435, p = 0 649), suggesting
that the different amounts of information had
no effect on respondents’ overinvestment in keeping
options open.
In a second attempt to examine the irrational aspect
of keeping doors open, the number of clicks from
the start of the experiment in which each respondent
stopped visiting each of the three rooms was
computed and compared across the different conditions.
For each respondent, the smallest number of
the three was the first time he or she eliminated a
door from his or her consideration—which we termed
the elimination point. We reasoned that the comparison
of this elimination point could demonstrate the
amount of investment in learning across different
conditions. If respondents overinvested in options to
Shin and Ariely: Effect of Unavailability on Incentives to Keep Options Viable
580 Management Science 50(5), pp. 575–586, © 2004 INFORMS
keep them, then their elimination point would be later
(larger). An overall ANOVA (Figure 3b) revealed a
main effect for option availability (F 1 99 = 44 67,
p < 0 001), a nonsignificant effect for information
(F 2 99 = 0 322, p = 0 725), and a significant interaction
effect between option availability and information
(F 2 99 = 4 76, p = 0 011). These results indicate
that although respondents felt that they did not need
Figure 3 (a) Average Number of Door Switches Across the Two OptionAvailability
and Three Information Conditions in Experiment
2
Decreased
Availability
Constant
Availability
#ofswitches
0
1
2
3
4
5
6
7
8
9
10
(b) Average Elimination Point in the Two Option-Availability
and Three Information Conditions in Experiment 2
Decreased
Availability
Constant
Availability
Eliminationpoint
0
5
10
15
20
25
30
35
40
No prior info
Practice info
Descriptive info
Note. Error bars are based on standard errors.
to revisit their least preferred room relatively early
in the process (as indicated by the elimination point
in the constant-availability condition: M = 9 8), they
kept the least preferred option viable for longer in the
decreased-availability condition (M = 27 14). Moreover,
the practice-information condition showed that
the addition of practice information actually increased
the difference between the constant-availability and
decreased-availability conditions, as the interaction
suggested (Figure 3b).
The third attempt to examine the irrational aspect
of keeping doors open used the behavior of respondents
in the constant-availability condition to create
a normative standard from which to evaluate the
behavior of the decreased-availability condition. This
analysis assumed that clicks that took place early
in the process are best viewed as an investment of
search costs to accumulate enough information3
to
determine which room to stay in. Based on this idea,
we computed click investment, which is the number
of clicks participants invested before they settled
down in one of three doors. This measure captures
the amount of information that respondents felt they
need to determine which option to pursue. This
analysis is particularly useful as a test of whether
the increased number of switching in the decreasedavailability
condition was due to rational information
search, as illustrated in the example with 6 and
2 switches above. The overall ANOVA revealed a
main effect for option availability (F 1 99 = 64 99,
p < 0 001), showing that the decreasing availability
leads the higher click investment in options (M =
10 07), compared with the constant-availability condition
(M = 4 49). Moreover, the results also showed
a nonsignificant effect of information (F 2 99 = 0 33,
p = 0 72), suggesting that respondents overinvest in
information search in the face of the possibility that
the option would become unavailable, irrespective of
their informational state. These results are also in
accord with the results of the later trials in Experiment
1 (Figure 2), showing that even when participants
had more information (in the last block
of 10 clicks), the effect of option availability was
pronounced.
In summary, the results of Experiment 2 replicated
Experiment 1 by showing that decreased availability
increases the tendency to invest in keeping options
open. More important, Experiment 2 demonstrates
that this effect could not simply be attributed to information.
Providing respondents with more experience
(in the practice-information condition) or telling them
3
What kind of information people need depends on individual
preference. For some, the mean might be sufficient, while others
might need more information about the distribution.
Shin and Ariely: Effect of Unavailability on Incentives to Keep Options Viable
Management Science 50(5), pp. 575–586, © 2004 INFORMS 581
explicitly about the distributions (in the descriptiveinformation
condition) decreased overall switching
behavior a bit, but it did not change the effect of
decreased availability on switching (the difference
between the two option-availability conditions). Combined
with the results of Experiment 1, these findings
suggest that there is an inherent tendency to keep
options open, even when doing so is costly. Experiment
2 also provided initial evidence that people are
overzealous in their preference for keeping options
open beyond the level that could be attributed to
investment in learning (based on the analyses of pecking,
elimination point, and click investment).
Experiment 3: Effects of Cost Saliency
on the Desire to Keep Options Open
Experiments 1 and 2 both demonstrated that the
threat of option disappearance causes decision
makers to sacrifice payoffs to keep options viable.
Moreover, Experiment 2 showed that this tendency
remained even when it became more apparent that
keeping these options open had no expected value
(such as in the descriptive-information condition). It
is possible, however, that while respondents understood
that keeping options available had little value,
they nonetheless wanted them open because they
did not understand the costs of keeping them available.
Specifically, the cost in Experiments 1 and 2
was implemented as an opportunity cost; respondents
lost a click every time they switched to a different
room. While we carefully explained to the
respondents that they lost a click for every door-click
and presented them with an updated click-counter
after every click, opportunity costs might have been
less heavily weighted compared with out-of-pocket
explicit costs (Thaler 1980). Respondents may have
simply failed to carefully consider the value of opportunity
cost, leading them to frequent switching.
Experiment 3 examined this issue by including a
condition in which respondents paid explicitly for
switching rooms. We reasoned that if the high level
of switching in Experiments 1 and 2 were due to the
low saliency of the cost, then making the cost explicit
(and higher) would decrease switching and eliminate
the effect of option availability. On the other hand,
to the extent that room switching is not influenced
by the cost, we would increase our confidence that
individuals have the desire to keep options open.
Method
The basic design of Experiment 3 differed from Experiment
1 in two ways. First, all respondents engaged in
100 practice clicks before beginning with the 100 real
clicks (as in the practice-information condition in
Experiment 2, but with 100 clicks). Second, there were
explicit penalties for door-clicks (room switching).
Respondents. Advertisements were placed around
campus to recruit 86 respondents, including some
from within the computer lab where the experiment
took place. Respondents were randomly assigned to
one of four conditions.
Design and Procedure. Experiment 3 included the
same option-availability manipulation as in Experiment
1, with an additional manipulation of cost,
which was varied on two levels: implicit cost and
explicit cost, crossed with option availability. In the
implicit-cost conditions, the cost of switching rooms
was the loss of a click (as in Experiments 1 and 2).
In the explicit-cost conditions, the cost of switching
rooms was loss of a click (implicit cost) and a loss
of 3¢. The loss of 3¢ per switch was noted on the
screen for every door-click in the same way that payoffs
for each room-click were posted. We selected 3¢
as the explicit cost because it was the expected value
of a room-click (Table 1), making the total cost of
switching in the explicit conditions twice as much as
in the implicit conditions.
Results and Discussion
An overall ANOVA of door-clicks indicated a significant
main effect for option availability (F 1 82 =
13 41, p < 0 001), a marginal effect for cost (F 1 82 =
3 48, p = 0 066), and a nonsignificant interaction
between option availability and cost (F 1 82 = 0 38,
p = 0 539). As can be seen in Figure 4, the effect
of option availability replicated the previous experiments,
showing that decreased availability caused
more switching behavior (M = 13 26) than constant
availability (M = 5 36). The effect of cost revealed that
switching was more frequent, but only marginally so,
in the implicit-cost condition (M = 10 8), compared
with the explicit-cost condition (M = 6 65). Although
the cost manipulation was marginally significant, the
important aspect is that the magnitude of the cost
Figure 4 Average Number of Door Switches Across the Two OptionAvailability
and Two Cost Conditions in Experiment 3
Decreased Availability Constant Availability
#ofswitches
0
3
6
9
12
15
Implicit
Explicit
Note. Error bars are based on standard errors.
Shin and Ariely: Effect of Unavailability on Incentives to Keep Options Viable
582 Management Science 50(5), pp. 575–586, © 2004 INFORMS
effect ( = 3 48) was much lower than that of the
option-availability effect ( = 13 41). Most important,
the nonsignificant interaction between option availability
and cost illustrates that the desire to keep
options open persisted even when the cost was more
explicit, and even when its magnitude was twice as
large. Finally, the amount of experience in this experiment
was higher (100 clicks instead of 50), which
allowed us to look at trials in which respondents had
more experience—the effects of availability and cost
persisted throughout the 100 clicks.
In summary, Experiment 3 suggested that the tendency
to keep options open persists when the cost is
explicit and doubled. While making the cost explicit
and twice as large increased the amount of attention
respondents paid to switching, and thus slightly
reduced it, this cost did not prevent them from having
increased interests in alternatives when there was
a threat that these alternatives will be eliminated.
Experiment 4: Aversion to
Loss vs. Flexibility
The previous three experiments demonstrated that a
threat to availability has a strong influence on the
desire to keep options open. Both Experiments 2 and 3
demonstrated that neither information nor saliency
of cost can account for the effect of option availability.
Experiment 4 examined two possible psychological
mechanisms that could provide an explanation for
respondents’ tendency to keep doors open: the desire
to keep or increase flexibility in future choices, and
the desire to protect against possible losses. To test
these two hypotheses, we added a new manipulation
of reactivation to the decreased-availability condition,
allowing respondents to reactivate a door that had
previously disappeared. To do this, respondents simply
pressed a button, paid a known payment, which
was varied (0¢, 6¢, or 30¢), and then the door would
reappear and the respondent would be in that room
(as if the door had been clicked). By using this reactivation
manipulation, options could disappear without
changing future flexibility—disassociating desire
for future flexibility from disappearance (loss) of an
option. If the increased switching in the decreasedavailability
conditions is caused by the desire for
future flexibility, adding the possibility of reactivation
should decrease or eliminate the effect of option availability.
On the other hand, if the increased switching
in the decreased-availability conditions is caused
by aversion to loss, reactivation should not influence
the effect of option availability. The argument here
is that reactivation following the disappearance of a
door can revitalize it, but it does not eliminate its
disappearance (loss). Returning to the initial dating
example, the reactivation is analogous to a case where
our romantic decision maker knows that even if a
potential romantic partner becomes unavailable, this
unavailability could always be reversed at a known
cost, such as a gift, flowers, or jewelry.
Method
Respondents. Advertisements were placed around
campus to recruit 91 respondents, including some
from within the computer lab where the experiment
took place. Respondents were randomly assigned to
one of five experimental conditions.
Design and Procedure. Experiment 4 had five conditions,
all of which offered respondents 100 clicks.
The first two conditions were constant availability
and decreased availability (as in Experiment 1). The
novel conditions in Experiment 4 introduced the reactivation
mechanism, which guaranteed the future
flexibility of the doors in the decreased-availability
settings. In the reactivation conditions, once a door
disappeared, a small box appeared above the location
of the door. By clicking on this box, the respondents
reactivated the door and entered the room (as if the
respondent had clicked on the door), and the cost of
reactivation was deducted from the payoff. The cost
of reactivation varied: 0¢, 6¢, or 30¢.
The expected pattern of results depended on
whether switching behavior was motivated by the
desire to keep, or increase, flexibility in future choices,
or by protection against possible losses. If the flexibility
account is correct, it would be expected that
respondents would expend a similar level of effort in
the three reactivation conditions as in the constantavailability
condition, while a higher level of effort
would be expected to keep options open in the basic
decreased-availability condition. Furthermore, within
the reactivation conditions, it would be expected that
the level of effort would depend on the cost of exercising
the reactivation. On the other hand, if the
tendency to keep options open largely relates to a
general aversion to losses, the reactivation manipulation
should have no influence on the effect to keep
options open, because reactivation does not prevent
the disappearance (loss) of the doors. In terms of the
cost of reactivation, the aversion to loss account suggests
that because the main motivation is to eliminate
loss, there will be a low sensitivity to the magnitude
of the cost (because the cost of reactivation matters
only after losing some options). Thus, to the extent
that the aversion to loss account is correct, room
switching should be the same in all four decreasedavailability
conditions, irrespective of reactivation.
There is another way to look at these conditions.
Based on the future flexibility account, there is only
one condition where the future flexibility is not guaranteed
(decreased-availability condition), while in
Shin and Ariely: Effect of Unavailability on Incentives to Keep Options Viable
Management Science 50(5), pp. 575–586, © 2004 INFORMS 583
four other conditions (constant-availability condition
and the three reactivation conditions), the future flexibility
remains the same. In contrast, from the perspective
of the aversion to loss account, there is only one
condition without a threat of availability (constantavailability
condition), while in the other four conditions,
availability is threatened (decreased-availability
condition and the three reactivation conditions).
A final prediction based on aversion to loss relates
to the difference in switching between the 0¢ reactivation
and the constant-availability condition. Note that
these two conditions are identical from the perspective
of a rational agent, but that they differ in terms
of framing, such that the 0¢ reactivation condition can
involve the aversion to loss mechanism.
One additional feature of Experiment 4 was that it
introduced a case in which the expected payoffs of
the different doors were not the same (Table 1). This
was done to test whether respondents in the previous
three experiments might have expended efforts
to keep all options open simply because they had
no clear reason to keep one and discard the others
(Kahneman et al. 1991, Shafir et al. 1993, Inman and
Zeelenberg 2002). In Experiment 4, where all options
were not equal, respondents could more easily find a
reason to make decisions and thus could justify less
switching than when all options were created equally
and, therefore, it was difficult to find a reason to make
decisions. The distributions were normal, with variance
1.25, and means of 2.5, 3, and 3.5.
Results and Discussion
There were five conditions in Experiment 4, two
of which were a replication of the main optionavailability
manipulation (constant availability and
decreased availability) and three of which were
reactivation-related decreased-availability conditions
with reappearance cost of 0¢, 6¢, and 30¢. An overall
ANOVA of the switching behavior revealed a
main effect for option availability (F 4 90 = 2 73,
p = 0 034).
First, we examined whether these results replicated
the previous experiments. As can be seen in the left
side of Figure 5, the main result was replicated—
switching in the constant-availability condition
(M = 6 06) was lower than switching in the decreasedavailability
condition (M = 12 76; t 31 = 2 83,
p < 0 01). Next, we investigate the effect of using distributions
of different means for the different rooms,
comparing the constant and decreased-availability
conditions in Experiment 4 with Experiment 1, where
all options had the same expected value. Using
a 2 (experiment: equal/unequal distributions) by 2
(option availability) between-subjects ANOVA, the
results revealed an effect of availability (F 1 186 =
32 52, p < 0 0001), confirming our previous finding
Figure 5 Average Number of Door Switches in the Two Replication
Conditions (Left) and the Three Reactivation Conditions in
Experiment 4.
Constant Decreased 0¢ 6¢ 30¢
#ofswitchs
0
2
4
6
8
10
12
14
16
18
Note. Error bars are based on standard errors.
of the effect of option availability. The results also
showed a marginally significant effect of the experiment,
where respondents switched more in Experiment
1 (M = 12 11) than in Experiment 4 (M = 9 52;
F 1 186 = 3 66, p = 0 057). Although marginally significant,
this result is consistent with the idea that
the different means provided the respondents with
reasons to switch less. Furthermore, the interaction
between experiment and option availability was not
significant, demonstrating that unequal distributions
did not change the effect of option availability on
the desire to keep options open. To further support
this idea, we replicated Experiment 1 (N = 35), with
distributions averaging 2.5, 3, and 3.5. These results
showed that respondents in the decreased-availability
condition switched significantly more (M = 10 13)
than respondents in the constant-availability condition
(M = 4 26; t 33 = 3 17, p < 0 001).
With the knowledge that Experiment 4 replicated
the previous experiments, we next examined which of
the two theories (future flexibility of choices and aversion
to loss) was better supported. Recall that we were
interested in the relationship between the reactivation
conditions to the constant- and decreased-availability
conditions, and particularly in the comparison
between these conditions and the 0¢ reactivation
condition.
First, in comparing the reactivation conditions with
the constant- and decreased-availability conditions,
we asked whether the three reactivation conditions
would be similar to the constant-availability condition,
thus supporting the future flexibility explanation,
or whether they would be similar to the
decreased-availability condition, thereby supporting
the aversion to loss explanation. As can be seen in
Figure 5, the switching behaviors in the three reactivation
conditions were not different from each other
(M = 11 58; F 2 55 = 0 74, p = 0 484), and they were
also not statistically different from the decreasedavailability
condition (M = 12 76; F 1 73 = 0 32,
Shin and Ariely: Effect of Unavailability on Incentives to Keep Options Viable
584 Management Science 50(5), pp. 575–586, © 2004 INFORMS
p = 0 5735). The three reactivation conditions, however,
were significantly different than the constantavailability
condition (M = 6 06; F 1 72 = 9 34,
p < 0 001). These results provide support for aversion
to loss over future flexibility as the driving force
underlying the desire to keep doors open.
Next, in comparing the 0¢ reactivation condition
with the constant- and decreased-availability conditions,
we asked whether the 0¢ reactivation condition
would exhibit similar switching to the constant availability
(to which it was logically equivalent) or to
the decreased availability, which could be the case
if aversion to loss is the force that causes individuals
to switch more in the face of the threat of
options unavailability. This result (Figure 5) indicates
that the switching behavior in the 0¢ reactivation
condition (M = 10 38) is more similar to that
of the decreased-availability condition (t 40 = 0 72,
p = 0 475) than that of the constant-availability condition
(t 40 = 2 50, p = 0 016), suggesting that in
our set-up, aversion to loss plays a larger role than
flexibility.
These results can also provide a hint as to whether
the effort to keeping doors open is driven by utility
(or pleasure from having more options) or disutility
(or pain from having options disappear). The higher
switching in the reactivation conditions (in particular,
the 0¢ reactivation condition) compared with the
constant-availability condition suggests that it is the
disutility of having options disappear that is the driving
force.
It is also interesting to examine the impact of the
magnitude of a reactivation fee on switching behavior.
The lowest amount of switching occurred in
the 0¢ reactivation condition (M = 10 38), followed
by the 30¢ reactivation condition (M = 12 12) and
the 6¢ reactivation condition (M = 13). But there
was no statistical difference between these conditions
(F 2 55 = 0 74, p = 0 484). This lack of sensitivity to
the magnitude of the cost can be taken as another
indication that the tendency to keep doors open is not
due to a rational cost-benefit analysis.
In sum, the different ways of looking at the results
of Experiment 4 all point to the same conclusion—that
the threat of availability of options is aversive, and
hence, respondents are willing to invest to reduce the
possible experience of loss. This effect can be termed
aversion to loss, or disappearance aversion, similar in
some ways to the general principle of loss aversion.
General Discussion
The current work examines a basic aspect of human
behavior that extends from interpersonal relationships
to abstract monetary options—valuations of
options. The experiments attempted to shed some
light on how individual decision makers evaluate
options by examining how the threat of option
unavailability influences the value of the options.
Experiment 1 demonstrated that the possibility that
the options will become unavailable in the future
increases investments in them to keep them from disappearing.
Experiment 2 tested whether this effect can
be due to information, and, in addition, added three
more fine-grained measures (pecking, click investment,
and elimination point) to test whether the effort
respondents expanded to maintain options open can
be rationally explained; it cannot. Experiment 3
tested whether the distinction between implicit and
explicit cost is the reason that our respondents overinvested
in keeping doors open; it was not. Finally,
Experiment 4 contrasted two psychological theories—
flexibility and aversion to loss—as possible mechanisms
for the overinvestment in keeping options
open. The results from this experiment point to aversion
to loss as being the more powerful of the two (at
least in our set-up).
In a further test of aversion to loss, we created
a new measure aiming at examining whether the
room that respondents “gave up on” first (elimination
point) was one for which they had more or less
information about compared with the one they “gave
up on” second (second elimination point). We argue
that from an informational point of view, subjects
should abandon a room they have more information
about, because the amount of information indicates
their certainty in the quality of the room. On the
other hand, from an aversion to loss perspective, a
room that had attracted more clicks might also have
a higher attachment associated with it, thus leading
to a lower tendency to abandon such a room. Analyzing
this measure in Experiment 2 revealed that
the respondents were four times more likely to first
abandon rooms they have less information about,
thus supporting the attachment and aversion to loss
ideas. Moreover, the increased impact of availability
on the practice-information condition in Experiment 2
strongly supported the aversion to loss explanation
(Figure 3). The experience of actual feeling of the
losses of the options during the practice trials seemed
to cause respondents to be even more resistant to
experiencing more losses during the actual trial.
In summary, the experimental evidence presented
suggests that individuals value options in a way that
is different from the expected value of these options,
and, in particular, that decision makers overvalue
their options and are willing to overinvest to keep
these options from disappearing. Based on the results
of Experiment 4, we believe that the desirability of
keeping options open is a kind of disutility from
loss rather than utility from “having more options to
choose from.”
Shin and Ariely: Effect of Unavailability on Incentives to Keep Options Viable
Management Science 50(5), pp. 575–586, © 2004 INFORMS 585
In a world where maintaining options has no cost,
such a tendency would have been nonconsequential.
However, we believe that in most day-to-day cases,
there is substantial cost to keeping options open,
which would lead to erroneous behavior. There are
many situations in which decision makers encounter
trade-offs between the future availability of options
and their maintenance costs. We have already mentioned
dating and choosing a major in college. Other
examples include trade-offs between focusing on
one’s current work and looking for new employment
elsewhere; whether to specialize in a way that suits
one’s current employer or instead to invest in skills
that are valued by other potential employers. These
results might also shed light on one of life’s greater
mysteries: Why do some people channel surf rather
than, for example, enjoy a single movie? The answer
might be the fear of losing other options.
These results might also be generalized to one-shot
cases. For example, when buying a new computer,
consumers face the dilemma of deciding whether to
buy a system that suits their current needs or purchase
an expandable system (e.g., more slots for cards,
and more memory) that is more expensive but could
better fit their uncertain future needs. In this case, the
main source of the dilemma is the uncertainty as to
whether future expansion will be needed, compared
with the current additional cost. Our computer buyer
is faced with a situation that is analogous to the door
game one click before a door disappears. She can take
a costly action at purchasing time to ensure that the
expansion option remains available to her whether
she subsequently decides to expand or not.
Other examples in which consumers face “disappearing”
options are deciding whether to purchase
an extended warranty when buying a new electronic
product and deciding whether to buy pictures of oneself
on whitewater rafting trips. In such cases, consumers
are given the opportunity to act on the options
(the warranty or the pictures), while realizing this is
their only opportunity to take this action, and that not
acting on the options is irreversible and may cause
the “pain” of losing these options. We suspect that the
effectiveness of such tactics is based on the option’s
nonavailability in the future, which would cause these
options to be perceived more favorably and to be
acted on more frequently.
There remain numerous unanswered questions. For
example, what are the mechanisms that underlie
the fear of losing options? What is the relationship
between keeping options open and indecision, particularly
when deciding means committing to one out
of a multitude of other possibilities (see also Amir
2004)? What is the impact of options’ prospective
lifetime and unavailability on their subjective value?
Faced with a large number of options, would decision
makers still value options (Iyengar and Lepper 2000)?
What is the number of options people would like to
keep? Finally, under what conditions will individuals
want to actively eliminate options? We keep these
research opportunities open for the future.
Acknowledgments
The authors thank Jim Bettman, Shane Frederick, and
Duncan Simester for their insightful comments; Leonard
Lee for his help with data collection; Dongwook and
Amit for expanding our options in life; and MasterCard’s
Future Transactions Lab for continued support. Comments
from the area editor and three anonymous referees were
extremely helpful.
References
Amir, O. 2004. The pain of deciding: Indecision, flexibility and consumer
choice online. Working paper, Yale School of Management,
Yale University, New Haven, CT.
Bar-Hillel, M., E. Neter. 1996. Why are people reluctant to exchange
lottery tickets? J. Personality Soc. Psych. 70 17–27.
Brehm, J. W. 1956. Post-decision changes in desirability of alternatives.
J. Abnormal Soc. Psych. 52 384–389.
Camerer, C. 1995. Individual decision making. J. H. Kagel,
A. E. Roth, eds. Handbook of Experimental Economics. Princeton
University Press, Princeton, NJ, 587–703.
Carmon, Z., D. Ariely. 2000. Focusing on the forgone: How value
can appear so different to buyers and sellers. J. Consumer Res.
27(3) 360–370.
Carmon, Z., K. Wertenbroch, M. Zeelenberg. 2003. Option attachment:
When deliberating makes choosing feel like losing.
J. Consumer Res. 30(1) 15–29.
Catania, A. C. 1975. Freedom and knowledge: An experimental
analysis of preference in pigeons. J. Experiment. Anal. Behavior
24 89–106.
Gilbert, D. T., J. E. J. Ebert. 2002. Decisions and revisions: The
affective forecasting of changeable outcomes. J. Personality Soc.
Psych. 82(4) 503–514.
Inman, J. J., M. Zeelenberg. 2002. Regret in repeat purchase versus
switching decisions: The attenuating role of decision justifiability.
J. Consumer Res. 29(June) 116–128.
Iyengar, S. S., M. Lepper. 2000. When choice is demotivating: Can
one desire too much of a good thing? J. Personality Soc. Psych.
76 995–1006.
Kahneman, D., A. Tversky. 1979. Prospect theory: An analysis of
decision under risk. Econometrica 47(2) 263–292.
Kahneman, D., J. L. Knetsch, R. H. Thaler. 1990. Experimental tests
of the endowment effect and the Coase theorem. J. Political
Econom. 98 1325–1348.
Kahneman, D., J. L. Knetsch, R. H. Thaler. 1991. The endowment
effect, loss aversion, and status quo effect: Anomalies.
J. Econom. Perspectives 5(Winter) 193–206.
Markowitz, H. 1952. The utility of wealth. J. Political Econom. 60
151–158.
Ratchford, B. T., N. Srinivasan. 1993. An empirical investigation of
return to search. Marketing Sci. 12 73–87.
Saad, G., J. E. Russo. 1996. Stopping criteria in sequential choice.
Organ. Behavior Human Decision Processes 67 258–270.
Shafir, E., I. Simonson, A. Tversky. 1993. Reason-based choice.
Cognition 49(2) 11–36.
Simonson, I. 1990. The effect of purchase quantity and timing on
variety-seeking behavior. J. Marketing Res. 27 150–162.
Shin and Ariely: Effect of Unavailability on Incentives to Keep Options Viable
586 Management Science 50(5), pp. 575–586, © 2004 INFORMS
Thaler, R. 1980. Toward a positive theory of consumer choice.
J. Econom. Behavior Organ. 1 39–60.
Tversky, A., D. Kahneman. 1991. Loss aversion in riskless
choice: A reference-dependent model. Quart. J. Econom. 106
1039–1061.
Williams, C. Arthur, Jr. 1966. Attitudes toward speculative risks as
an indicator of attitudes toward pure risks J. Risk Insurance
33(4) 577–586.
Zeelenberg, M., W. W. van Dijk, J. van der Pligt, A. S. R. Manstead,
P. van Empelen, D. Reinderman. 1998. Emotional reactions to
the outcomes of decisions: The role of counterfactual thought
in the experience of regret and disappointment. Organ. Behavior
Human Decision Processes 75(August) 117–141.
Zwick, R., E. Weg, A. Rapoport. 2000. Invariance failure under subgame
perfectness in sequential bargaining. J. Econom. Psych. 21
517–544.