Ritov, I., & Baron, J. (1995). Outcome knowledge, regret,
and omission bias. Organizational Behavior and Human
Decision Processes, 64, 119-127.
Outcome knowledge, regret, and omission bias
Ilana Ritov1
Ben Gurion University
Jonathan Baron
University of Pennsylvania
Abstract
Subjects made or evaluated decisions in hypothetical scenarios.
We manipulated knowledge about the outcome and act vs. omission
in four cases. In case 1 (production processes), acts (changing
the process) were considered better than omissions when the
decision maker did not know the outcome or knew that it was
better than the status-quo. Acts were considered worse than
omissions once the decision maker learned that the foregone
option would have led to an even better outcome. In case 2
(medical treatment) act vs. omission again interacted with gain
vs. loss (relative to the status-quo) unless the outcome of the
foregone option was known, in which case act vs. omission
interacted with better vs. worse (of the two options). In case
3 (fetal testing), subjects tolerated less risk of miscarriage
when a potential for regret was present (because the test with
the risk of miscarriage, although better, might miss disorders
that another test would detect). This effect was greater for
actions than omissions. In case 4 (vaccination) subjects showed
less tolerance of vaccine risk when the decision maker would know
about the outcome of vaccination or nonvaccination. Thus, the
bias toward omission (not vaccinating) is greater when potential
regret is present, and potential regret is greater when knowledge
of outcomes is expected.
Introduction
When a decision leads to a bad outcome, relative to what might
have been, people think that the decision was worse if the
outcome resulted from action than if it resulted from inaction
(Baron & Ritov, 1994; Kahneman & Tversky, 1982; Miller &
Taylor, in press; Ritov & Baron, 1990; Spranca, Minsk, & Baron,
1991). This result has been called omission bias. One way in
which omission bias might occur is that people might expect
greater regret for bad effects of actions than for bad effects of
omissions. This is because acts tend to be seen as more causal
than omissions (Spranca et al., 1991), and blame, including
self-blame and regret, depends on perception of causality (e.g.,
Fincham & Jaspers, 1980). Regret can affect decisions if
decision makers avoid options that could produce it. Such
anticipated regret might be especially conspicuous when the
decision maker expects to know the outcome of the choice not
taken, because learning that the outcome was bad and could have
been avoided will evoke regretful feelings. This paper provides
evidence for the link between omission bias, outcome knowledge,
and regret. It also shows an effect of knowing only the outcome
of the chosen option, as opposed to the forgone option.
Regret, as Sugden (1985) points out, has two components. One
component has to do with feelings of responsibility, blame, and
subjective evaluation of the quality of the decision. The other
component pertains to evaluation of a chosen option by comparison
of the outcome that occurred to the one that might have occurred.
The first component of regret may occur even without knowing what
would have happened had one chosen differently. The second
component, however, must depend to a large extent on the
availability of outcome information. Hence, the extent of
uncertainty resolution is expected to affect evaluation of
outcomes, and consequently choice as well, through regret.
(However, it is possible to think about how we would feel if we
knew the outcome, even if we never will.)
Comparison of the obtained outcome to alternative outcomes has
been shown to affect omission bias. Baron and Ritov (1994) found
that acts leading to a worse outcome than the alternative outcome
were judged as worse than omissions that lead to the worse of the
two outcomes, regardless of whether these outcomes were gains or
losses relative to the status-quo. Miller and Taylor (in press)
obtained the same result, although they did not manipulate the
status-quo. This bias is at least partially explained by norm
theory. What actually happened is compared to what might have
been, with the outcome of omission usually considered a neutral
reference point. People think that a bad outcome of an act
is particularly bad if it could have been avoided by doing
nothing, but they are less inclined to think that a bad outcome
of an omission is particularly bad because it could have been
avoided through some action.
The bias toward omission has been found also in cases in which
the outcome of the option not chosen will never be known. People
are reluctant to vaccinate their child, for example, even if the
risk of death from the vaccine is lower than the risk of death
from the disease (Ritov & Baron, 1990). Hence, the role of
comparison to alternative outcomes in generating omission bias in
evaluation and choice needs to be further examined.
Decisions under uncertainty vary in the extent to which
uncertainty is resolved after the decision is made. For example,
in choosing to invest in stock A rather than stock B, one
eventually comes to know the value of both stocks. We shall call
this type of situation, in which uncertainty is resolved for all
options, complete knowledge. In other situations, uncertainty
is resolved only for the chosen alternative. For instance, in
deciding whether to vaccinate one's child against a disease, only
one outcome obtains: that of the chosen option. If one has
chosen to vaccinate, the outcome of vaccination will obtain, but
one will never know for certain what would have happened had one
chosen not to vaccinate. We call this partial knowledge.
Finally, there are situations in which the decision maker knows
no outcome, neither the outcome of the chosen option nor that of
the foregone option. We call this no knowledge.
In the present studies, subjects evaluated hypothetical
decisions, both before and after the outcome is known. We ask
whether the evaluation of the decision depends on whether the
outcome resulted from an act or an omission, and on whether the
outcome will be known or not, that is, on knowledge. We
contrast all three knowledge conditions - full, partial, and
none - whenever possible. We ask about the extent of omission
bias in each of these three knowledge conditions. If omission
bias is caused by anticipated regret over actions more than
omissions, then we would expect more bias with more knowledge.
In particular, with full knowledge, regret will be greatest when
the outcome of action leads to a worse outcome than the outcome
of inaction (which will be eventually known). If such an event
could happen with an action, the subject will consider the action
less good as a choice, because of the anticipation of this
regret.
In partial knowledge we hypothesize that the outcome that
occurred is compared to the most salient or most easily
imaginable alternative state. This could be the status-quo, or
some other state. It may also be the predicted outcome of the
foregone option, but this is not necessarily the case. Whatever
the comparison state, outcomes are likely to be regarded as
better or worse than this state. Consequently, omission will be
preferred if the outcome is better than the comparison state,
whereas action is likely to be preferred if the outcome is worse.
In the case of no knowledge, people should be less inclined to
think at all about how they will feel about the comparison of
outcomes to what might have occurred. Instead, they will make
decisions more on the basis of expected consequences.
Alternatively, subjects might think about various possible ways
of filling in what might happen for both chosen and unchosen
options, as assumed implicitly by regret theory in its original
form (Bell, 1982; Loomes & Sugden, 1982). Our main hypothesis,
however, is that this does not happen all the time, so that the
effects of anticipated regret on decisions are at least greater
when some knowledge is expected than when it is not.
The four experiments vary in the situations used and in whether
the subjects are asked to predict the feelings of someone else
(Experiment 1), put themselves in someone else's position
(Experiments 2 and 4), or imagine themselves in a situation
(Experiment 3). We know of no previous studies of regret or
omission bias that find effects of such differences, and in some
of our own studies we have looked explicitly for them (Ritov &
Baron, 1990; Baron & Ritov, 1994).
A few previous studies have examined the effect of anticipated
knowledge on possible regret effects, although none has examined
the effect of knowledge on omission bias as such. Ritov (1994)
found that anticipation of complete knowledge led to some
preference reversals in choice between gambles. When complete
knowledge was associated with high probability of regret,
expectation of complete knowledge increased preference for
regret-minimizing choices. Kelsey and Schepanski (1991) found no
effect of anticipated knowledge in hypothetical taxpayer
decisions, but they used a between-subject design with a small
number of cases and subjects. Simonson (1992) found that
subjects who were led to anticipate their feelings made choices
that were more easily justified, such as purchasing a familiar
brand or buying an item on sale now rather than waiting for a
better sale. Subjects may have seen these options as defaults,
in which case Simonson's results would agree with those we find
here. Josephs, Larrick, Steele, & Nisbett (1992) reported
evidence that expected outcome knowledge affected choices of
subjects low in self-esteem but not of subjects high in
self-esteem. They suggest that regret is a threat to self-esteem
so it is avoided when self-esteem is fragile. (We do not measure
self-esteem in the present studies.)
The present study also provides additional evidence for
anticipated-regret effects of any sort. Early demonstrations of
anticipated-regret effects (e.g., Loomes, 1987) showed that
decisions were affected by the pairing of outcomes with uncertain
states. For example, lottery A (with 100 tickets) yields $24 if
any of tickets 1-9 are drawn; B yields $16 for 10-21. Most
people choose A because more regret will occur if 1-9 are drawn
than if 10-21 are drawn. But most people prefer lottery C,
yielding $16 for 1-12, over A because the regret with 1-9 is
much reduced. Note that the probability and prize in B and C are
the same. Starmer and Sugden (1993), however, found that these
results were artifactual; they resulted from presenting the
outcomes from C in two columns (1-9 and 10-12). Thus, the
evidence for anticipated-regret effects comes largely from the
studies just cited, plus one other study, that of Beattie, Baron,
Hershey, and Spranca (1994), which did find an effect of the
pairing of outcomes within a state: subjects were less willing to
make decisions at all when their choice led to different outcomes
in different states.
Experiment 1
In this experiment subjects were presented with a hypothetical
managerial decision, expected to increase efficiency. The
manager described had decided either to implement a proposed new
procedure or not, in a between-subject design. Subjects were
asked to judge the manager's satisfaction with his decision under
three knowledge conditions, one at a time: the outcome of his
decision is not known yet; the outcome is known but the outcome
of the alternative option (adopted by another factory) is not;
and, finally, both outcomes are known. The outcome of the
decision in all cases is positive: the efficiency has increased.
However, the alternative outcome, when it is known, turns out to
be better still.
We hypothesize that, in the complete knowledge condition, as the
manager learns that his decision led to the worse of the two
possible outcomes, satisfaction will be lower in case of action
than in case of omission. In the partial knowledge condition,
on the other hand, the manager who learns that his decision led
to an efficiency increase is likely to be more satisfied if this
resulted from an act rather than an omission. As suggested
earlier, assumptions concerning the outcome of one's choice,
relative to the outcome of the foregone option are less likely to
be made if the outcome of neither alternative is known yet.
Hence, under no knowledge, the distinction between omission and
commission might be reduced.
Method
Ninety-one industrial engineering students at Ben Gurion
University completed the questionnaire during regular class
sessions. Each subject was randomly assigned to one version,
act or omission.
In both versions subjects were presented with a situation in which
a company, owning two "twin" production plants, was considering
the possibility of implementing a certain change in the
production system, designed to increase efficiency. The
implementation of the change would not incur any extra costs.
The proposed change was brought before the board of directors,
and pros and cons were voiced. As the board could not reach a
unanimous decision, each of the two managers of the production
plants was given the choice whether to implement the proposed
change in his plant.
Next, subjects in the act version were told that the manager of
plant A decided to implement the change (in the omission version
they were told that the manager decided not to implement
the change). Following the description of the problem and the
decision of the manager, subjects were told that it is getting
close to a year's time since the decision, but the calculation of
the year's production costs has not yet been completed. Subjects
were asked to imagine the manager's satisfaction with his
decision, at this point in time, when the outcome information is
not yet available to him. A 9 points scale was used, ranging from
"little satisfaction" to "very high satisfaction." After
making the rating, half the subjects in each version were told,,
"The manger of plant A knows that the manager of plant B has
made the opposite decision. The announcement of production costs
in plant B is expected shortly."
In both versions, subjects completed the first part without
knowing that a second part of the questionnaire was to follow.
After subjects completed the first part, the second part was
handed out. In this part subjects were told, "At the end of
this year it was found that the production costs of the plant
were 7% lower than the production costs the previous year."
Subjects were then asked to re-judge the manager's satisfaction
with his decision concerning the change in the production system,
now that the outcome of the decision has become known.
Finally, after the second part has been completed, a third part
of the questionnaire was handed out. In this part the following
information was added: "The data for both plants were now made
public. The manager of plant A has just learned that, in plant
B, where the opposite decision was made, the production costs
were 9% lower than the production costs the previous year."
Subjects were asked, once again, to judge the satisfaction of
plant A manager with his decision, now that he knows the outcome
of both options. The same scale was use as before.
Results
The between-subject manipulation (within each version) of
knowledge that the other manager made the opposite decision did
not significantly affect the responses, so the results were
combined across this factor. The mean responses to the three
parts of each version are presented in Table 1. Clearly
satisfaction was dramatically affected by the availability of
information concerning outcomes (F2,88=27.60, p < .001 for
the main effect of outcome information). In all versions
satisfaction increased upon learning the positive outcome of the
manager's decision, and decreased, later, when the subjects
learned that the alternative choice had produced even better
results than the choice made by the manager of plant A.
Table 1
Mean responses, Experiment 1.
| Knowledge | |
| None | Partial | Full | Mean |
Omission | 5.43 | 6.80 | 4.61 | 5.61 |
Action | 6.07 | 8.26 | 4.00 | 6.11 |
Mean | 5.72 | 7.47 | 4.33 | |
The overall omission effect is almost significant
(F1,89=3.31, p=.07 for the between-subject omission
factor). However, the omission factor interacted with outcome
information (F2,88=9.75, p < .001). Action was rated higher
than omission in the first part of the questionnaire, in which no
outcome information was available (F1,89=4.10, p < .05), as
well as in the second part, once the positive outcome of the
manager's choice was made known (F1,89=18.64), p < .001).
Action ratings became lower than omission ratings in the final
part, when the better outcome of the foregone option was made
known, although this effect by itself was not significant
(F1,89=1.76, p=.19). (The failure to get a significant
difference did not result from a floor effect for satisfaction
ratings: only six subjects used the "little satisfaction"
response, five in the omission condition.)
Experiment 2
In this experiment, subjects rated happiness with outcomes for
partial and complete outcome knowledge for medical treatments, in
a within-subject design. Both gains and losses from the
status-quo were examined. In the complete-knowledge condition,
gain vs. loss was crossed orthogonally with better vs. worse.
The terms better and worse refer to comparison of the outcome
that occurred to the outcome that would have occurred if the
alternative option had been chosen. Thus, even a loss could be
the better outcome if the alternative option would have led to an
even worse outcome. Baron and Ritov (1994), using completely
described outcomes, found an interaction between act vs.\
omission and better vs. worse: when the act led to the worse
outcome, ratings were particularly low. We found no interaction
between act vs. omission and gain vs. loss, although gain
ratings were higher than loss ratings.
Baron and Ritov did not examine partially described outcomes. In
this condition, subjects might compare the outcome achieved to
the status-quo rather than to the outcome of the alternative
option. Act vs. omission might then interact with gain vs. loss:
subjects might regard an act leading to a loss as particularly
bad. They might use the status-quo as a reference point, or they
might even assume that the outcome of the foregone option would
be the status-quo.
Method
Subjects were 64 students solicited by a sign on a walkway at the
University of Pennsylvania. They came from both that university
and the Philadelphia College of Pharmacy Science. They were paid
$6/hour for completing this questionnaire and others.
We asked subjects to take the role of a physician making
decisions about treatment. The scenarios involved, respectively,
a patient with high blood pressure (50% above normal) and a
patient with high cholesterol (also 50% above normal). The
patient had just moved from another city, and the previous
physician had prescribed a treatment, which was the default
(omission) option. The physician in question could either change
the treatment or leave it. Outcomes were described in terms of
percentages above normal after treatment. For example, in the
gain-worse condition, the treatment led to 30% above normal but
the foregone option would have led to 10% above normal. The
treatment always led to a 20% increase (loss) or decrease (gain),
and, when complete information was provided, the foregone option
was always either 20% better or worse than the actual outcome.
One of the first two scenarios was presented with only partial
information. The other was presented with partial information
and then complete information; subjects were told that the
outcome of the foregone option would be known after results of a
research study were reported, and they were asked how they would
feel about the outcome that occurred both before and after they
were told the results of the research. The partial-information
condition was presented first (with blood pressure) for about
half of the subjects and second (with cholesterol) for the other
half. Subjects rated their feelings about each outcome "on a
scale from 100 to -100, where zero represents neither happiness
nor unhappiness."
Results
Table 2 shows the mean ratings for each condition. (The columns
labeled "means of better & worse" are based on the complete
information condition.) The overall omission bias is not
significant in any condition. When better vs. worse was not
specified, acts were rated higher than omissions for gains
(p=.000 for partial knowledge; p=.001 for complete; Wilcoxon
tests were used because many differences were zero) but lower
than omissions for losses (p=.000 for partial; p=.005 for
complete).
Table 2
Mean ratings for each condition of Experiment 2.
| Partial | Complete | Mean of |
| information | information | better & worse |
| omit | act | omit | act | omit | act |
gain | 42 | 57 | 45 | 56 | 23 | 23 |
loss | -51 | -65 | -50 | -61 | -14 | -16 |
gain better | | | 59 | 64 | | |
loss better | | | 11 | 17 | | |
gain worse | | | -12 | -18 | | |
loss worse | | | -39 | -50 | | |
When better vs. worse was specified, however, this interaction
between act vs. omission and gain vs. loss was no longer
significant: the interaction term for gain vs. loss and act vs.\
omission was higher before the alternative outcome was specified
than after it is specified (p=.014 by a Wilcoxon test comparing
the two interaction terms). Instead, acts were rated higher than
omissions for the better outcome but lower than omissions for the
worse outcome. Although neither of these effects was significant
by itself, the difference between them was significant (p=.026,
Wilcoxon test). (Order of conditions did not affect any results
reported here.) This interaction was found by Baron and Ritov
(1994) in several experiments. In sum, act vs. omission
interacts with better vs. worse when complete outcomes are
specified but with gain vs. loss when partial outcomes are
specified (so that better vs. worse is undefined). Subjects seem
to compare the outcomes of action (especially) to the alternative
outcome when it is known and to the status-quo otherwise.
The first two experiments examined subjects' evaluation of
outcomes, and their satisfaction with the decisions which led to
resolved outcomes. We found that evaluation of both decisions
and outcomes was affected by knowledge of alternative outcomes.
However, the regret associated with knowing the alternative
outcome interacted with the omission factor in evaluation of acts
and omissions. To the extent that people consider potential
regret when making decisions, regret effects in evaluation of
outcomes should reflect on actual choice in decision making. The
next two experiments examine the way anticipation of possible
regret, and expectation of outcome knowledge (possibly leading to
regret) affect decisions in two hypothetical situations.
Experiment 3
In this experiment, we presented subjects with hypothetical
situations in which they had to choose between two fetal testing
procedures, one completely safe, and the other involving some
risk of a miscarriage. However, the "risky" test has the
advantage of having a higher detection rate of serious disorders.
The potential for regret was manipulated within subjects by
varying the degree of overlap between the particular disorders
detected by the "safe" test and those detected by the risky
test. Thus, for example, if every disorder detected by the safe
test is also detected by the risky test then taking the risky
test eliminates the possibility of regret concerning the choice
of test, associated with giving birth to a disordered child. If,
on the other hand, there is no substantial overlap in detection
potential, one could experience regret following the birth of a
disordered child, even though one had chosen (and taken) the
risky test. (A within-subject comparison of these conditions is
helpful because subjects vary considerably in their tolerance for
risk.)
These two regret conditions were crossed, between subjects, with
action vs. omission. Thus, the default (omission) could be
either the safe or the risky test, or there could be no default.
If subjects are more sensitive to the possibility of regret when
it is associated with the outcome of action, then they might be
more affected by the regret manipulation when the safe test
serves as the default than in other cases.
Method
Subjects were 115 industrial engineering students at Ben Gurion
University, tested during regular class sessions, and 109
students solicited as in Experiment 2. Subjects were randomly
assigned to the different versions of the questionnaire.
Nationality of the subjects - or sex, or age, or marital status,
or number of children - did not affect the results, so these
variables are ignored henceforth, although they were asked about
at the end of the questionnaire.
The questionnaire was described as dealing "with decisions
concerning fetal testing during pregnancy. These tests are
designed to determine whether the fetus suffers from any severe
disorder (such as Down's syndrome). If the tests uncover such a
severe disorder, it is possible to terminate the pregnancy."
Subjects were instructed to imagine that they are expecting
a child. They were informed of two different possible tests
(here, the no-regret condition):
Test A: the disorders detected by this test exist in 3 out of
1000 embryos. This test does not involve any risk.
Test B: The disorders detected by this test exist in 6 out of
every 1000 embryos. Every disorder detected by Test A will also
be detected by Test B. In addition, other disorders, which can
not be detected by Test A, will be detected as well. Test B
involves some risk of miscarriage.
The questionnaire went on to explain that for various
(unspecified) reasons it is not possible to take both tests.
Subjects are asked to decide which one of the two tests they
would prefer to take. The response was made by indicating the
highest level of risk (of miscarriage) at which Test B will still
be preferred to Test A. The specific wording of the response
was: "I will prefer Test B to Test A, as long as the risk of
miscarriage, following administration of Test B, will not be over
percent." The blank was filled by circling a
number on a scale. The scale range from 0% to 2.6%, with an
additional point described as "more than 2.6%."
In the second part (regret condition) subjects were asked to
compare test A and test C, assuming that Test B does not exist.
The disorders detected by Test C exist in 6 out of 1000 embryos.
However, the disorders detected by test C are different from the
ones detected by Test A. That is, some disorders detected by
Test A will not be detected by Test C. Test C also involves some
risk of miscarriage.
As in the first part, here too, subjects were asked to indicate
the highest level of risk (associated with Test C) at which they
would still choose Test C over Test A.
The order of the two parts was reversed, for approximately half
of the subjects. Order did not affect the results.
Three versions of the questionnaire differed in terms of a
specified default option. In the No-default version, no default
was specified. In the Risky-default version, the default was the
riskier test. Thus, subjects were told: "By the routine
procedure you were assigned to take Test B (or Test C, at the
second stage). However, if you prefer to take Test A you can be
re-assigned with no difficulty." The Safe-default version had
the safe test (Test A) as the default. The description of the
default was parallel to the one given in the risky-default
version. The three versions of the questionnaire were otherwise
identical.
Results
The mean response to the two parts of each version are shown in
Table 3. Order of presentation of Test B and Test C did not
significantly affect the response, hence the results for the two
orders were combined. Across the three versions, subjects were
willing to take higher risk in the no-regret condition than in
the regret condition (87 vs. 24 subjects, p=.000, Wilcoxon test).
However, the three versions differed in the magnitude of this
effects (p=.052 by Kruskal-Wallace "nonparametric analysis of
variance" on the difference between regret and no-regret; p=.020
for Risky-default vs. Safe-default). The Safe-default condition
also differed from the No-default condition (p=.050, one tailed),
but the Risky-default did not differ from the No-default
condition. The basic regret vs. no-regret effect was not quite
significant in the Risky-default condition (21 vs. 9 subjects,
p=.055 one-tailed Wilcoxon test), but it was clearly significant
in each of the other conditions (29 vs. 9 for No-default, 37 vs.\
6 for Safe-default, p=.000 for each). (The three conditions did
not differ significantly in mean risk tolerance combining the
regret and no-regret conditions.)
Assuming that the birth of a child with severe disorders is much
worse than a miscarriage, the potential for regret in taking Test
C is much higher than the potential for regret in taking Test B.
The present results indicate that people are more sensitive to
regret when it is associated with the outcome of action rather
than omission. When the riskier option is the default, keeping
it requires no action. Indeed, in that case, subjects did not
show notable differentiation between Test B (no regret) and Test
C (regret). However, when taking the riskier test required
action, subjects' responses reflect greater willingness to take
the risk when the potential for regret is lower (Test B).
Table 3
Mean percent of risk tolerance for Test B and Test C, in each
version, Experiment 3.
Version | Test B (no-regret) | Test C (regret) | Mean |
No default | .685 | .510 | .598 |
Risky default | .759 | .668 | .713 |
Safe default | .776 | .497 | .637 |
Mean | .738 | .554 | |
Experiment 4
To examine the relation between act vs. omission and knowledge,
we used a vaccination case similar to that used by Ritov and
Baron (1990). The decision was whether or not to vaccinate
children against a potentially fatal flu, given that the vaccine
itself can also be fatal. The case was presented with full,
partial, and no knowledge, in a within-subject design. In the
full-knowledge case, the decision maker would know both the
outcome that occurred and the outcome that would have occurred if
the other option (vaccinate vs. not vaccinate) had been chosen.
Omission bias is defined as unwillingness to vaccinate even
though vaccination reduces overall loss of life.
Method
The questionnaire was given to 137 subjects solicited as in
Experiment 2. It began, "A type of flu can be fatal to young
children. A vaccine can prevent the flu. The vaccine is
extremely inexpensive and is given along with other vaccines, so
no extra shots are required. The vaccine is also well
researched: nothing can be learned from new data. However, the
vaccine can also be fatal. Children who die from the vaccine
would not necessarily have died anyway from the flu. Ten out of
every 10,000 children will die from the flu if they are not
vaccinated. You are a public health official in a foreign
country, and you must decide whether to vaccinate the children in
your district. You plan to leave soon, so you will not see the
individual children after you make your decision."
The no-knowledge condition began, "You will never know what
happens to any of the children, and you will never know what
would have happened if you had decided differently." The
partial-knowledge condition began, "After you leave, you will
see reports about each child in your district. You will know
which children died from the flu, if any, and which died from the
vaccine, if any. But you will not know what would have happened
if you had decided differently." The full-knowledge condition
began, "After you leave, you will see reports about each child in
your district. You will know which children died from the flu,
if any, and which died from the vaccine, if any. Also, you will
see, for each child, the results of a test to determine which
vaccinated children would have died from the flu and which
unvaccinated children would have died from the vaccine. (These
tests take a long time, and the decision whether to vaccinate has
to be made now.) So you will know, for each child, both what
happened and what would have happened had you decided
differently."
To make sure that subjects understood the conditions, we asked
four questions after each of these introductions: "Will you know
which children died? If you vaccinated all the children, will
you know what would have happened to each child if you had not
vaccinated? If you do not vaccinated any children, will you know
what would have happened to each child if you had vaccinated?
How many children out of 10,000 will die from the flu if you do
not vaccinate them?" Half of the subjects received the
conditions in this order, half in the reverse order. (Order did
not affect any results.)
Subjects were then asked, "Would you want to vaccinate your
patients if the death rate from the vaccine were each of the
following?" They were given a list of rates: 0, 1, 3, 5, 7, 9,
and 11 out of 10,000. They were asked to indicate whether they
would vaccinate all children in each case. At the end of the
questionnaire, subjects were asked why the difference between
full and partial knowledge was relevant or not, and likewise for
partial vs. no knowledge.
Results
Omission bias was highest in the full-knowledge condition and
lowest in the no-knowledge condition. Subjects' justifications
supported the role of anticipated regret in this effect. Few
subjects contributed to this result, however, because many
subjects showed no difference between the conditions or
misunderstood the task. Specifically, 40 subjects failed one or
more of the test questions, leaving 97. Of these 97, 55 were
willing to vaccinate at rates of 9 or 11 in all three cases - a
result consistent with Ritov and Baron (1990) - and 4 were
unwilling to accept any risk. Of the 38 remaining subjects, 19
showed no effect of knowledge, giving the same answer in all
three cases. Of the remaining 19, 16 showed the hypothesized
difference between full and no knowledge (the extreme cases) and
3 showed the reverse difference. A sign test is significant
(p=.002, one tailed), indicating that our hypothesis accounts
for most of the unequal responses. In sum, of the 38 subjects
who willing to accept some risk but not always the optimal
amount, almost half were influenced by knowledge.
For the 19 subjects who were affected in either direction by
knowledge, the means for the three conditions (of the maximum
acceptable vaccine risk) were 6.5, 5.1, and 4.1 for no, partial
and full knowledge, respectively. The difference between no and
partial was significant (11 as predicted, 1 reversed, p=.003) and
the difference between partial and full was almost significant (8
vs. 2, p=.055).
Subjects' explanations of their answers were consistent with our
hypothesis, in that all mention anticipated feelings about
outcomes as reasons for not vaccinating. Examples of
justifications for no vs. partial knowledge were:
This difference would affect my decision because I would know the
individual children and families. It is easier on the conscience
to make a decision not knowing who it affects. I wouldn't want
to personally know who was affected because I would feel guilty.
The difference of knowing who dies or not comes into play because
I needed to make a moral decision. Knowing what happened to the
children made me realize how deadly the vaccine can really be.
With this knowledge at hand, it would be hard for me to NOT
change my position.
It does affect my decision because of the guilt involved. ...
No guilt, no paranoia.
Examples for full vs. partial were:
This would affect my decision because it would be torture
knowing that you are responsible for a child's death. At
least with questions 1 and 2 [no and partial] you can't be sure.
I probably would be very upset and feel very guilty if I knew
what kids died from the vaccine and not the flu. I would feel
very responsible for taking a life that wasn't meant to be taken
yet. Very upset!!
Knowing if the vaccine would or would not help is a very
important factor. This makes the decision more of a reality. I
could feel good about helping others or I could feel guilty and
wonder what could have been. ...
I suppose it all comes down to a question of conscience. After
all, who wants that on your mind?
If children died because I vaccinated them, I would have to live
knowing I killed them. If they died because I didn't vaccinate
them, I would still feel that I killed them.
It is noteworthy that we found no cases in which subjects
referred to outcomes as positive, e.g., as saving a life that
might have otherwise been lost. This kind of framing of the
situation is consistent with the existence of omission bias in
this case, for it is found when the outcome is considered to be
worse than the point of comparison.
Discussion
We examined the effects of knowledge - full, partial, and none -
and of action vs. omission on several judgments and hypothetical
decisions. Anticipated-regret effects were hypothesized to be
greatest under full knowledge, when it is possible to know what
would have happened if the foregone option had been chosen. If
regret is greater for actions than omissions, then differences
between action and omission should be greatest under full
knowledge. Under partial knowledge, people might make reasonable
assumptions about the foregone option, or they might compare the
outcome to the status-quo. Here, regret effects might still be
obtained. Without any knowledge, when people think about
decisions, they may pay less attention to the feelings they would
have about the outcomes associated with different options.
Experiment 1 found that a manager's satisfaction with his
decision was higher for acts than omissions before any outcome
was known. This result conflicts somewhat with previous findings
(Spranca et al., 1991), but it is possible that subjects expected
a good outcome, in which case the result is consistent with other
findings of a preference for acts when the outcome is good (Baron
& Ritov, 1994). Satisfaction was also higher for acts when the
outcome was revealed as better than the status-quo, but it was
worse for acts with full knowledge, when it was revealed that
the outcome of the foregone option was even better.
Experiment 2 supported our interpretation of Experiment 1 by
showing that action vs. omission interacts with gain vs. loss
under partial knowledge but interacts with better vs. worse (and
not with gain vs. loss) under full knowledge. Recall that
better vs. worse was defined with respect to the foregone option,
while gain vs. loss was defined relative to the status-quo.
Thus, in thinking about acts and omissions under partial
knowledge, people compare outcomes to some expected outcome
(perhaps a good outcome in Experiment 1) or to the status-quo.
Once information about the outcome of foregone options is
revealed, however, people compare the chosen option to the
foregone option.
Experiment 3 compared a decision in which regret was possible for
the more effective fetal-testing option to a decision in which
regret was not possible. The potential for regret arose because
the more effective test might miss some disorders detected by the
other test. Subjects would tolerate less risk of a miscarriage
in the regret condition, unless the risky test was the default,
in which case no action was required to stick with it. This
result amounts to an interaction between omission vs. commission
and potential regret. Of course, this experiment involved only
partial knowledge.
Experiment 4 found an interaction between act vs. omission and
knowledge. Subjects were less willing to tolerate less risk
from a vaccine when they would know the outcome of vaccination
than when they would not, and even less risk when they would know
both the outcome of vaccination and nonvaccination (the foregone
option). The case of full knowledge, of course, is the one in
which potential regret is most salient. Subjects' explanations
were consistent with the hypothesis that potential regret is
greater for action than omission.
In sum, then, we have found evidence that people anticipate
regret when they expect to be able to compare a bad outcome to a
better outcome that would have resulted from a foregone option.
They evaluate decisions as worse when such a situation exists,
and they are reluctant to choose options that might lead to such
a situation, especially when these options involve action rather
than inaction. The effect of potential regret is reduced when
people do not expect to know the outcome of the option they will
choose or the option that they did not choose. In these cases,
people may think more in terms of expected utility.
We should note that many of our results may lack generality in
one way or another. Gilovich and Medvec (1994), for example,
found that regret for decisions made recently was - in line with
our findings - greater for acts than omission, but over the
course of people's lives their greatest regrets were for
omissions. Advice from the old to the young is thus often, "Do
it. You'll regret it if you don't." In addition, specific role
relationships, like that between a doctor and a patient, might
make omissions more regrettable (Haidt & Baron, in press) even
to the point of being more regrettable than acts leading to the
same outcome. Finally, regret may be greater when decisions
affect others - as they do in our examples - than when they
affect the self only (as suggested by Beattie et al., 1994).
Still, despite the lack of generality, we have shown that, in
some situations, anticipated regret is greater for acts than
omissions and that the anticipation of regret is greater when
outcomes will be known.
What is the normative status of these effects? Regret itself can
be considered a rational emotion insofar as it helps us make
better decisions in the future. Likewise, the anticipation of
regret is in general a good reason (although not a decisive
reason) against a choice. It is more difficult to see
normatively why regret should depend as it does on the
distinction between action and inaction, or why the effect of
regret on decision making should depend on knowledge of outcomes.
People might try to avoid this effect by forcing themselves to
work through the possible outcomes of all options, thinking about
how they would feel if they knew the outcome, even if they know
that they will never know it.
One implication of our results is that, if people know about the
effects of outcome knowledge, they may try to avoid such
knowledge, especially if the effects of regret over a bad outcome
are greater than the effect of rejoicing over a good one. Thus,
investors who make more risky investments may try to make it more
difficult for themselves to follow the ups and downs of their
investments day by day (e.g., by making them through pension
funds). More poignantly, some women who decided against having
children before the genetic test for Huntington's chorea was
discovered are reluctant to have the test after their
childbearing years are over: the happiness that they are not
subject to the disease would be mitigated by their regret over
not having children because of their fear of passing it on.
References
Baron, J. & Ritov, I. (1994). Reference points and omission
bias. Organizational Behavior and Human Decision Processes,
59, 475-498.
Beattie, J., Baron, J., Hershey, J. C., & Spranca, M. (1994).
Determinants of decision seeking and decision aversion. Journal
of Behavioral Decision Making, 7, 129-144.
Bell, D. E. (1982). Regret in decision making under uncertainty.
Operations Research, 30, 961-981.
Fincham, F. D., & Jaspers, J. M. (1980). Attribution of
responsibility: from man the scientist to man as lawyer. In L.\
Berkowitz (Ed.), Advances in Experimental Social Psychology,
13, 81-138.
Gilovich, T., & Medvec, V. H. (1994). The temporal pattern to
the experience of regret. Journal of Personality and Social
Psychology, 67, 357-365.
Haidt, J. & Baron, J. (in press). Social roles and the moral
judgment of acts and omissions. European Journal of Social
Psychology.
Josephs, R. A., Larrick, R. P., Steele, C. M., & Nisbett, R. E.\
(1992). Protecting the self from the negative consequences of
risky decisions. Journal of Personality & Social Psychology.
62, 26-37.
Kahneman, D., & Tversky, A. (1982). The psychology of
preferences. Scientific American, 246, 160-173.
Kelsey, D., & Schepanski, A. (1991). Regret and disappointment in
taxpayer reporting decisions: An experimental study. Journal of
Behavioral Decision Making, 4, 33-53.
Lomes, G. (1987). Testing for regret and disappointment in
choice under uncertainty. Economic Journal, 97, 119-129.
Loomes, G., & Sugden, R. (1982). Regret theory: An alternative
theory of rational choice under uncertainty. Economic Journal,
92, 805-824.
Miller, D. T., & Taylor, B. R. (in press). Counterfactual
thought, regret, and superstition: How to avoid kicking yourself.
In N. J. Roese and J. M. Olson (Eds.), What might have been: The
social psychology of counterfactual thinking. Hillsdale, NJ:
Erlbaum.
Ritov, I. (1994). Probability of regret: anticipation of
uncertainty resolution in choice. Manuscript, Ben Gurion
University.
Ritov, I., & Baron, J. (1990). Reluctance to vaccinate: omission
bias and ambiguity. Journal of Behavioral Decision Making, 3,
263-277.
Simonson, I. (1992). The influence of anticipating regret and
responsibility on purchase decisions. Journal of Consumer
Research, 19, 105-118
Spranca, M., Minsk, E., & Baron, J. (1991). Omission and
commission in judgment and choice. Journal of Experimental
Social Psychology, 27, 76-105.
Starmer, C., & Sugden, R. (1993). Testing for juxtaposition and
event-splitting effects. Journal of Risk and Uncertainty, 6,
235-254.
Sugden, R. (1985). Regret, recrimination and rationality.
Theory and Decision, 19, 77-99.
Footnotes:
1This research was partly supported by
N.S.F. grants SES91-09763 and SBR92-23015 to J. Baron. We
thank Jane Beattie and an anoymous reviewer for comments. Send
correspondence to Ilana Ritov, Department of Industrial
Engineering and Management, POB 653, Ben Gurion University, Beer
Sheeva 84105, Israel, or (e-mail) ritov@bgumail.bgu.ac.il.
File translated from
TEX
by
TTH,
version 3.40.
On 9 Dec 2003, 20:22.