Baron, J. & Ritov, I. (1994). Reference points and omission
bias. Organizational Behavior and Human Decision Processes,
59, 475-498.
Reference points and omission bias
Jonathan Baron
Department of Psychology
University of Pennsylvania
Ilana Ritov
Department of Industrial Engineering and Management
Ben-Gurion University
Acknowledgments
This work was supported by grants from the National Science Foundation (SES-88-09299 and SES-91-09763). Address
correspondence to J. Baron, Department of Psychology, University
of Pennsylvania, 3815 Walnut St., Philadelphia, PA 19104-6196;
e-mail baron@psych.upenn.edu.
Abstract
Subjects were asked to evaluate the choice of options leading to
known outcomes, or to say how they would feel about a chance
outcome, in hypothetical decisions. We independently manipulated
the value of the status-quo and the assignment of the better or
worse outcome to an act or an omission. Acts leading to the
worse outcome were always considered worse than omissions leading
to the worse outcome. This act-omission difference was reduced
or reversed for the better outcome. Most experiments showed an
overall bias toward omissions (combining better and worse).
Little evidence was found for greater omission bias for losses
relative to the status-quo than for gains. A bias toward
maintaining the status-quo itself was found, however,
independently of omission bias. Most of the results can be
explained by norm theory and by loss aversion, but other possible
accounts are inconsistent with the results.
Introduction
People often judge acts to be worse than omissions with the same
consequences. Many people will not terminate life-sustaining
medical treatment, although they would not initiate such
treatment for an otherwise identical case. Likewise, it was
argued that hurricanes should not be seeded, even if the total
harm would be reduced, because different people would be hurt by
the seeded hurricane, and the decision makers would be
responsible for this harm (Howard, Matheson, & North, 1972). Our
legal system honors the distinction even when it seems
irrelevant: pharmaceutical companies are held liable for the harm
caused even by well-produced, compulsory vaccines, but not for
the harm caused by failing to produce vaccines (Inglehart, 1987).
Many countries have no laws against failure to rescue
(Feldbrugge, 1966). More generally, we tend not to hold
ourselves responsible for harm that we could easily prevent
(Singer, 1979).
These examples suggest the operation of a general bias in
decision making, a systematic failure to make decisions in ways
that best achieve people's goals concerning outcomes. This bias
occurs even when people are aware of it. Arguments that it is a
bias can therefore not depend on demonstrations that people
violate decision rules that they themselves endorse. Rather, it
is a bias because it prevents us from achieving our goals (Baron,
in press a,b).
Empirical research supports the existence of this bias. Sugarman
(1986) found that subjects judged active euthanasia as worse than
passive euthanasia. Sugarman asked about `responsibility,'
however, which could be interpreted as a legal question rather
than one about morality or advisability. Spranca, Minsk, & Baron
(1991) avoided this problem and still found a bias toward harmful
omission. In one case, for example, subjects were told about
John, a tennis player who thought he could beat Ivan Lendl only
if Lendl were ill. John knew that Ivan was allergic to cayenne
pepper, so, when John and Ivan went out to the customary dinner
before their match, John planned to recommend to Ivan the house
salad dressing, which contained cayenne pepper. Subject were
asked to compare John's morality in different endings to the
story. In one ending, John recommended the dressing. In another
ending, John was about to recommend the dressing when Ivan chose
it for himself, and John said nothing. Ten out of 33 subjects
thought that John's behavior was worse in the commission ending,
and no subject thought that the omission was worse.
Ritov and Baron (1990) examined a set of hypothetical vaccination
decisions. We compared omission and commission as options within
the same choice. In one experiment, subjects were told to
imagine that their child had a 10 out of 10,000 chance of death
from a flu epidemic, a vaccine could prevent the flu, but the
vaccine itself could kill some number of children. Subjects were
asked to indicate the maximum overall death rate for vaccinated
children for which they would be willing to vaccinate their
child. Most subjects answered well below 9 per 10,000. Of the
subjects who showed this kind of reluctance, the mean tolerable
risk was about 5 out of 10,000, that is, half the risk of the
illness itself. Some subjects would accept no risk at all from
the vaccine. Asch et al. (1993) found that the existence of this
bias correlated with mothers' resistance toward DPT vaccination
(which may produce death or permanent damage in a few children).
In these cases, the action and omission options being compared
always lead to bad outcomes compared to the status quo. (In the
status quo, Lendl is not sick and all children are alive.)
Another kind of decision has been studied in which two attributes
of an outcome trade off so that an improvement in one could be
compensated by a decrement in the other. For example, in most
monetary transactions, money is given up in return for some good,
or vice versa.
A common finding in these situations is that people tend toward
inaction or toward the status-quo. They require more money to
give up a good than they are willing to pay for the same good
(Knetsch & Sinden, 1984; Samuelson & Zeckhauser, 1988; Viscusi,
Magat, & Huber, 1987). In studies of valuation of public goods,
willingness-to-accept (the amount of money required for giving up
a good) exceeds willingness-to-pay (for the good) (Mitchell &
Carson, 1989). Kahneman, Knetsch, and Thaler (1991) showed that
these effects were not the result of wealth effects or other
artifacts. They are, at least in part, true biases.
In all of these studies, keeping the status quo requires no
action, and changing the status quo requires action. Ritov and
Baron (1992) found that the status-quo bias - the attachment to
the status-quo that is common to all of these findings - was
largely a consequence of omission bias. Specifically, when
subjects were told that keeping the status-quo required action
and that giving it up required inaction, subjects then favored
giving up the status-quo. When both options required action, no
preference for the status-quo was found, but when both options
yielded new outcomes (neither one matching the status-quo), a
preference for omissions was still found. Schweitzer (in press)
found both omission bias without a status-quo option, thereby
supporting our conclusion that this bias is not dependent on the
status-quo bias, and status-quo bias without an omission option,
a result that we did not find. His results suggest that the
status-quo bias as usually measured consists of two different
effects, a bias toward omissions and (in some cases) a bias
toward the status-quo.
In sum, omission bias has been found in three different ways:
First, subjects judge omissions leading to the worse outcome as
less bad than acts leading to the same outcome (as in the Lendl
case). This result requires comparisons of options in different
choice sets; only in this way is it possible for the worse of the
same two outcomes to be caused by an act in one case an an
omission in the other. Second, subjects sometimes judge harmful
acts to be worse than more harmful omissions within the
same choice set (as in the vaccine cases). We shall call this
result a `reversal,' since the act-outcome distinction reverses
the ordering of options from what it would otherwise be. Third,
subjects judge omission to be better than acting to bring about
an equally desirable outcome that is better in one way but worse
in another way. We shall call this situation a `tradeoff.' All
the results described so far involve outcomes that can be seen as
bad in some way. Later we shall review other results concerning
outcomes that are good.
The aim of the present paper is to explore the possible
explanations of these findings in terms of comparisons of
outcomes to reference points. We do not assume that this is the
only way to explain these results. Other explanations could
involve the use of heuristics such as `let nature take its
course.' Even such heuristics, however, may operate through
comparison of outcomes to reference points, such as the effect of
nature.
Explanations of omission and status-quo biases
Mechanisms that could explain these effects differ in their
assumptions about the reference point that is used and about
whether some outcomes or options are weighed more heavily than
others. Some possible reference points are the status-quo and
the outcome of the omission option. Both options could be
compared to the same reference point, but it is also possible
that acts are compared to the status-quo while omissions are
compared to themselves or not evaluated. Losses could be weighed
more heavily than gains, or the act (or its outcomes) could be
weighed more heavily than the omission (or its outcomes). All of
the explanations considered can explain the basic finding of the
Lendl story, but we shall review other findings that are
consistent with some mechanisms but not others.
Norm-theory
Kahneman and Miller (1986, p. 145) suggested that `the affective
response to an event is enhanced if its causes are abnormal' and
that acts are considered abnormal because `it is usually easier
to imagine abstaining from actions that one has carried out than
carrying out actions that were not in fact performed.' This
account was suggested by the finding of Kahneman and Tversky
(1982) that subjects anticipated more regret when bad outcomes
resulted from action (e.g., buying the less profitable of two
stocks) than when they resulted from omission (failing to buy the
more profitable stock). We interpret norm theory as asserting
that the omission is the reference point: u(O,X)=0 for all X,
where O represents an omission and X represents the outcome.
Acts leading to the worse outcome are seen as worse than any
omission (whether from the same choice set or a different choice
set), and acts leading to the better outcome are seen as better
than any omission. More generally - if the use of omission as
the reference point were true only some of the time, or if it
were a kind of argument that pulls subjects' evaluations in a
certain direction - omission bias for the worse outcome should be
greater than that for the better outcome. That is, an
interaction is predicted between better vs. worse and act vs.
omission. This is often called an amplification effect: acts are
amplified.
Landman (1987) and Gleicher et al. (1990) found that anticipated
joy in response to positive outcomes was stronger when the
outcomes were the result of action rather than inaction. In
these studies, as in Kahneman and Tversky (1982), the outcome was
uncertain, and the better vs. worse distinction resulted from
chance. Spranca et al. (1991, Experiment 3) found evidence for
better vs. worse difference in omission bias in the evaluation of
decisions with known outcomes: A train was rolling down a track
toward a switch. On one branch of the track after the switch,
two men were working; on the other branch, three men. Subjects
evaluated four decisions: the train was rolling toward three men
and the agent switched it to two (A,-2); the agent did not switch
the train (O,-3); the train was rolling toward two and the agent
switched it to three (A,-3); the agent did not switch to three
(O,-2). Most subjects ranked A,-3 worse than O,-3, but subjects
were about equally divided on whether O,-2 was better or worse
than A,-2, and many subjects rated them equal. Better and worse
outcomes were clearly treated differently. This result, however,
could also be explained if A,-3 was perceived as intentional
endangering of an extra person while O,-3 was not.
Loss-aversion
The simple assumption that omission is the reference point cannot
explain omission bias in tradeoffs (e.g., Ritov & Baron, 1992).
We can explain this result, however, if we add the loss-aversion
assumption that losses are weighed more heavily than equivalent
gains (Kahneman & Tversky, 1979; Tversky & Kahneman, 1991). The
commission option is represented as a gain in one dimension and a
loss in the other, relative to the omission (default), which is
taken as the reference point. Because the negative utility of
losses is greater than the utility of equivalent gains, people
will prefer the default (Thaler, 1980). For example, money paid
out is valued more highly than money received, and a good given
up is valued more than a good received, so people who are
unwilling to spend $X for a good will not accept $2X for the same
good.
This same account can explain reversals, if outcomes are
perceived in terms of different dimensions. For example, in the
vaccination case of Ritov & Baron (1990), deaths from the vaccine
and deaths from the disease may be considered as different
dimensions. Compared to the omission, the act increases vaccine
deaths but decreases disease deaths. Consistent with this
account is the finding that omission bias was sharply reduced
when subjects were told that the children who would die from the
vaccine were the same ones who would have died anyway from the
flu; this instruction may have encouraged subjects to think of
the outcomes as one dimensional. Loss aversion, however, cannot
explain extreme reversals in which subjects are willing to accept
no risk at all from the vaccine, unless gains are given no weight
relative to losses.
Loss aversion can also explain the results of Experiment 4 of
Spranca et al. (1991), where a medical treatment removed the risk
of brain damage from disease but incurred its own risk. The risk
was 15% for the treatment and 20% for the disease, or the
reverse. Subjects rated each option in each choice set on a
scale of -100 (bad) to 100 (good). Omissions leading to the
worse outcome (20%) were rated higher than acts leading to that
outcome, and omissions leading to the better outcome (15%)
were also rated higher than acts leading to that outcome. An act
leading to the better outcome would be seen as trading a 20% gain
in disease risk for a 15% loss of treatment risk, so loss
aversion would still lead to a lower rating than for the
omission. Consistent with this account, more zero ratings were
assigned to omissions than to acts, as would be predicted if the
omission were often taken as the reference point, and mean
ratings were higher overall when the worse option was an omission
than when the worse option was an act. (A somewhat different
account was provided by Spranca et al.)
The same experiment also found the interaction predicted by norm
theory. Spranca et al. (1991) reported no such interaction, but
a reanalysis of their data finds a modest but significant effect
when it is tested directly by comparing the omission bias for
better vs. worse outcomes (t(47)=2.03, p=.048; it was originally
tested as an interaction an analysis of variance in which order
was a factor, but order did not affect the size of the
interaction significantly).
Unequal weighting
Another account holds that outcomes are compared to the perceived
status-quo (or some other salient reference point) rather than to
the omission. Acts are simply weighed more heavily, perhaps
because more attention is paid to them. In the extreme,
omissions are not evaluated at all. (Omissions are also not
evaluated according to norm theory, but that is because they are
the reference point.1) This account
predicts a greater omission bias for losses than for gains, where
gain and loss are defined relative to the status-quo. If no
other mechanism were operating, acts would be considered better
than omissions for gains over the status-quo, as well as being
considered worse than omissions for losses.
Whether reversals were found for losses would depend on the
weights and the outcomes. For example, if acts were weighed more
than twice as much as omissions, and if utility were linear, then
u(A,-5) < u(O,-10), but, if acts were weighed less than twice as
much as omissions, then u(A,-5) > u(O,-10). (The outcomes are
defined relative to the status-quo.) If omissions were given no
weight, then u(O,-10)=0, and u(A,-X) would always be lower for
any X > 0. In Ritov & Baron's (1990) study, many subjects were
willing to tolerate some risk from the vaccination, but others
were not willing to tolerate any risk. The former may have been
giving some weight to the outcome of the omission (compared to
the status-quo), while the latter were giving no weight at all to
it, simply evaluating the effect of the act relative to the
status-quo.
Unequal weighting can account for Landman's (1987) results, if
subjects perceived the outcomes as losses compared to the
status-quo or some other reference point. Unequal weighting by
itself cannot account for omission bias in tradeoffs unless both
options are worse than the status-quo.
The main purpose of the present studies is to manipulate
systematically the relation of outcomes to each other and to the
status-quo. We manipulate gains vs. losses (relative to the
status-quo) and better vs. worse independently by manipulating
the status-quo independently of the outcomes. In this way, we
can distinguish unequal weighting, in which the status-quo plays
a special role, from other accounts, in which it does not. We
also examined a variety of different scenarios. Finally, some
previous studies have examined emotional reactions, and others
have examined judgments of decision quality or advisability. We
examined both, sometimes in the same experiment.
Experiment 1
The first experiment was a replication of Gleicher et al. (1990),
Kahneman & Tversky (1982), and Landman (1987), except that our
cases had a clear status-quo, different from both the act and the
omission. We could therefore examine omission bias as a function
of gain vs. loss (relative to the status quo) as well as a
function of better vs. worse.
Subjects were told about a decision involving either switching or
not switching an investment. They were told the outcomes of both
options, regardless of which one the agent chose. Both outcomes
were either gains or losses relative to the status-quo, but one
outcome was better than the other. The subject was asked to
compare the feelings of two agents who reached the same outcome
out of the same pair of possible outcomes, one agent reaching the
outcome through an act and the other through an omission.
Method
Sixty-two subjects in a class at Ben-Gurion University were given
a questionnaire describing four pairs of descriptions of
situations. `In each pair of descriptions the final situation of
both people is identical, but the way in which they reached this
situation is different.' Subjects were asked to `read the
descriptions carefully and try to imagine how each of them feels.
For each pair of descriptions ... indicate which one of these two
people feels better (or less bad) at the end.' Subjects were
discouraged from indicating that the two people feel the same.
The first pair read (in translation from Hebrew):
X owns shares of company A. During the past year he considered
switching to shares of company B, but decided not to do so. At
the end of the year he finds out that his shares in company A
have earned a net profit of 6000 shekels. Had he switched to
shares of company B his net profit would have been 4000 shekels
higher (that is, his net profit would have been 10,000).
Y owned shares of company A. During the past year he switched
to shares of company B. At the end of the year he finds out that
his shares in company B have earned a net profit of 6000 shekels.
Had he kept his shares in company A his net profit would have
been 4000 shekels higher (that is, his net profit would have been
10,000).
Note that the two cases were the same except for whether the
outcome resulted from an act or an omission. In this pair, the
outcome was the worse of the two possible outcomes, but each
possible outcome was a gain. In another pair, the outcome
was the better of the two, and both were gains. The two
other pairs were Loss-worse and Loss-better. The four pairs were
presented in four different orders for counterbalancing, and the
order of act vs. omission within each pair was likewise
counterbalanced. Roughly equal numbers of subjects were randomly
assigned to each possible order.
Results
Order did not affect any of the measures. Table 1 shows the
number (and percent) of subjects in each condition who thought
that the omission or act would lead to feeling better (or less
bad).
Table 1
Number (percent) of subjects in each condition of Experiment
1 who thought that the omission or act would lead to better (or
less bad) feeling.
Condition
Gain-worse Loss-worse Gain-better Loss-better
Option
Omission 42 (67.7%) 42 (67.7%) 24 (38.7%) 24 (38.7%)
Act 19 (30.6%) 18 (29.0%) 36 (58.1%) 34 (54.8%)
Equal 1 (1.6%) 2 (3.2%) 2 (3.2%) 4 (6.5%)}
When the outcome was the worse of the two possible outcomes, the
omission led to better (less bad) feeling (p=.005 for Gain-worse,
p=.003 for Loss-worse, by two-tailed sign tests). When the
outcome was the better one, the act tended to lead to better
feeling, but not significantly so. Overall, combining both
Better and Worse conditions, the tendency of omissions to lead to
better feeling is not quite significant (p=.061, two-tailed sign
test across subjects on whether omission or act is favored more
often). Differences among the four conditions were significant
(p=.004, Friedman test), each Better condition differed from each
Worse condition in the relative ranking of act and omission
(p=.005 or less for the four comparisons, Wilcoxon test), and the
two remaining Gain vs. Loss comparisons did not differ
significantly.
In sum, the results did not support the prediction of unequal
weighting that omission bias would be greater for losses than for
gains, but they did support the prediction of norm theory that
the bias would be greater for worse outcomes than for better
outcomes. For the worse of the two possible outcomes, the act
led to more negative feeling than does the omission.
Experiment 2
Experiment 2 examined two different kinds of decisions, one
concerning pension funds (which differed in interest rates) and
another concerning trade policies (which affected unemployment
rates in a particular industry). Outcomes varied in their
relation to the status-quo and their relation to each other, as
in Experiment 1. Here, however, subjects made two different
kinds of judgment. In one judgment the expected effects of each
option were known, and the subject rated the advisability of
each option as a choice. This judgment is intended to be similar
to judgments made in other studies in which subjects were asked
about the goodness of decisions (Spranca et al., 1991; Ritov &
Baron, 1990). In the other judgment, the subject rated emotion,
as in Experiment 1. In the emotion judgment, the two options
were expected to be equally good, and one turned out to be better
than the other by chance. (The emotion judgments were always
made after the advisability judgments, lest the subject not
believe that the outcomes were predictable for the latter.) This
experiment therefore allowed us to ask whether emotion ratings
are affected by different factors than those that affect
advisability ratings.
Method
Sixty-five university-student subjects, solicited by a sign on a
walkway at the University of Pennsylvania, were paid $6 per hour
to fill out a questionnaire individually or in small groups. Two
were eliminated because of apparent serious misunderstanding
attributable to poor command of English, leaving 63. A few other
subjects inadvertently skipped a few questions each, but their
remaining data were used. The questionnaire had four forms to
counterbalance for order.
Advisability ratings. In the first half of the
questionnaire, subjects were presented with two-option choices
and were asked to `rate the advisability of each option, on
a scale from -100 (bad decision) to 100 (good decision).
Decisions with higher numbers should always be preferable to
decisions with lower numbers. Feel free to go beyond -100 or 100
if you need to.'
For the advisability ratings about pension funds, subjects were
told, `In these cases, imagine yourself working for an employer
in a job that you think you will be in for some time. You have a
pension fund, which pays a certain rate of interest this year.
For next year, your employer gives you the option of switching to
a new fund by checking a box on a form. The only information you
have about the two funds is the expected rate of interest for the
next year.' The cases were of the form,
1. Your pension fund now earns 7% interest. Your options are:
A. Switch to the new fund, expected to earn 9%.
B. Stay with the old fund, expected to earn 8%.
In the unemployment cases, subjects were told, `In these cases,
imagine yourself as a U.S. government official charged with
deciding whether to change a trade policy. If you recommend a
change, your recommendation will almost certainly be accepted,
because it will be part of a much larger bill and nobody will
study it that carefully. The possible change affects mainly the
projected unemployment rate of the workers in a particular
industry for the next year.' Cases were of the form:
1. The unemployment rate is now 7%. Your options are:
A. Change the policy: 9% unemployment expected.
B. Do not change the policy: 8% unemployment expected.
Table 2 shows the cases used for the four forms of the
questionnaire. The percentages for Pension cases and
Unemployment cases were identical; note, however, that a high
percentage was negative for unemployment but positive for
pensions. Forms C and D of the questionnaire were identical to
forms A and B, respectively, except that the order of the Pension
and Unemployment cases were switched. In Forms A and C, the act
always came first (as in the example just given), and in forms B
and D, the omission came first. Note that, for Forms A and D,
cases 1, 2, 5, and 6 involved gains and cases 3, 4, 7 and 8
involved losses. However, in cases 5-8, one outcome was always
identical to the status-quo, so that we can look for a bias
toward this outcome. In cases 9 and 10, the outcomes were 11%
and 12%, as a check for sensitivity to outcome level.
Table 2.
Items used in Experiment 2.
Forms A and C Forms B and D
Case Status-quo Omission Act Status-quo Omission Act
1 7% 8% 9% 10% 9% 8%
2 7% 9% 8% 10% 8% 9%
3 10% 8% 9% 7% 9% 8%
4 10% 9% 8% 7% 8% 9%
5 8% 8% 9% 9% 9% 8%
6 8% 9% 8% 9% 8% 9%
7 9% 8% 9% 8% 9% 8%
8 9% 9% 8% 8% 8% 9%
9 9% 10% 11% 9% 11% 12%
10 9% 11% 10% 9% 12% 11%
Note: Act came before omission in Forms A and C, omission before
act in B and D. Pension came first in A and B, Unemployment in C
and D.
Emotion ratings. In the second half of the questionnaire,
subjects were presented with two outcomes and were asked to `rate
how you think you would feel, on a scale from -100 (bad) to 100
(good) in each case.' Subjects were again told that they could
go beyond -100 or 100, and they were reminded to rate both
outcomes.
The cases were in the same order as those in the first half. In
the pension cases, subjects were told, `Imagine yourself working
for an employer in a job that you think you will be in for some
time. You have a pension fund, which pays a certain rate of
interest this year. Last year, your employer gave you the option
of switching to a new fund by checking a box on a form. Your
employer thought that the two funds were equally good, and your
co-workers had different opinions about whether to change or
not.' The first case of one form (corresponding to the case
given above) was:
1. Last year, your pension fund earned 7% interest.
It earned 8% this year, and the new fund earned 9%.
A. You switched to the new fund, and you earned 9%.
B. You stayed with the old fund, and you earned 8%.
In the unemployment cases, subjects were told, `Imagine yourself
as a U.S. government official. Last year you were charged with
deciding whether to change a trade policy. If you recommended a
change, your recommendation was accepted. The effect of the
change was mainly on the unemployment rate of the workers in a
particular industry in the current year. At the time you made
the decision, you were very uncertain about which policy was
better.' The first case of this form was:
1. Last year's unemployment rate was 7%.
A. You changed the policy. This year's unemployment rate
was 9%. If you had not changed, it would have been 8%.
B. You did not change the policy. This year's unemployment
rate was 8%. If you had changed, it would have been 9%.
Results
Presentation order had no effect on any measures. The results
are summarized in Table 3, which shows the mean ratings in each
condition. Note that the alternative options in each choice are
diagonally opposite in each group of four conditions. For
example, in the Pension Advisability ratings when the status-quo
was 7%, one choice was between the act leading to the better
outcome of 9% (GB, rated 69 on the average) and the omission
leading to the worse outcome of 8% (GW, rated 7); the letter G
indicates that both outcomes were gains compared to the
status-quo. The rightmost two columns represent 11% and 12%
outcomes. These were included to check the response to large
changes. It is apparent that subjects gave higher ratings to
better outcomes in these cases. These data are ignored in all
subsequent analyses because they are not included in the
counterbalancing scheme.
Table 3.
Means of ratings for all conditions in Experiment 2. Main
conditions are in the same order for each scenario. (Standard
deviations ranged from 37 to 77.)
Advisability
Pension
Status-quo: 7% 10% 8% 9% 9%
Outcome: 9% 8% 9% 8% 9% 8% 9% 8% 12% 11%
Condition: GB GW LB LW gb gw lb lw
------------------------------------------------------------
act 69 -7 54 -52 54 -17 48 -38 69 33
omission 79 7 59 -36 66 4 61 -20 76 43
Effect 10 14 5 16 12 21 13 18 7 10
Unemployment
Status-quo: 10% 7% 9% 8% 9%
Outcome: 8% 9% 8% 9% 8% 9% 8% 9% 11% 12%
Condition: GB GW LB LW gb gw lb lw
------------------------------------------------------------
act 70 3 36 -46 61 -11 51 -40 3 -58
omission 64 7 35 -48 56 -11 42 -33 0 -59
Effect -6 4 -1 2 -5 0 -9 7 -3 -1
Emotion
Pension
Status-quo: 7% 10% 8% 9% 9%
Outcome: 9% 8% 9% 8% 9% 8% 9% 8% 12% 11%
Condition: GB GW LB LW gb gw lb lw
------------------------------------------------------------
act 70 15 34 -34 58 -13 48 -28 81 51
omission 69 21 39 -23 53 5 48 -11 81 50
Effect -1 6 5 11 -5 18 0 17 0 -1
Unemployment
Status-quo: 10% 7% 9% 8% 9%
Outcome: 8% 9% 8% 9% 8% 9% 8% 9% 11% 12%
Condition: GB GW LB LW gb gw lb lw
------------------------------------------------------------
act 65 -12 39 -38 53 -20 43 -30 20 -37
omission 61 -7 41 -38 44 -22 42 -29 21 -39
Effect -4 5 2 0 -9 -2 -1 1 1 -2
Note: In labels of conditions, G=gain, L=loss, B=better outcome,
W=worse outcome. Upper case letters indicate choices in which
both options are gains or both are losses; lower case letters
indicate choices in which one option is a gain or loss and the
other is equivalent to the status-quo. Conditions not analyzed
(the two rightmost columns) are not labeled.
All hypotheses concerning omission effects were tested by
comparing the ratings assigned to acts and omissions that led to
the same outcome with the same status-quo. If the omission
received a higher rating, the comparison was considered positive;
if the act received a higher rating, the comparison was negative;
and if the ratings were the same, the comparison was neutral.
When conditions were combined, the number of positive comparisons
was compared to the number of negative comparisons. The result
was considered positive if there were more positive than negative
comparisons. Thus, each subject was counted as positive,
negative, or neutral for each question asked, and two-tailed sign
tests were used throughout. (Wilcoxon tests yielded
substantively identical results.) These conservative procedures
were adopted because about half of the subjects gave the same
rating to the the act and the omission in each comparison (range,
36% to 64% across comparisons), contrary to the error assumptions
required for parametric statistics.
One question was whether omissions were rated higher than acts
across all conditions (Better vs. Worse, Gain vs. Loss). This
general bias was found for Pension (49% positive vs. 13%
negative; p=.000) but not for Unemployment (45% vs. 37%). Within
Pension, it was found for both Advisability (43% vs. 18%, p=.015)
and Emotion (56% vs. 13%, p=.000).
A second question was whether omission bias was greater for
losses than for gains, as predicted by unequal weighting. To
measure this effect, we counted a subject as positive if there
were more positive Loss and negative Gain comparisons than
negative Loss and positive Gain comparisons, with everything
changed accordingly for counting a subject as negative. This
effect was not found anywhere. Combining all conditions and
scenarios, 47% of the subjects were positive, 42% negative.
Recall that Experiment 1 also found no such effect.
A third question was whether the omission bias was greater for
the worse of the two outcomes than for the better of the two, as
predicted by norm theory and as found in Experiment 1. Subjects
were classified according to whether positive Worse and negative
Better comparisons outnumbered negative Worse and positive
Better. This effect was found for Pension (61% vs. 25%, p=.004)
and Unemployment (60% vs. 23%, p=.003). Within Pension, it was
found for Emotion (56% vs. 21%, p=.004) but not for Advisability
(29% vs. 24%). Within Unemployment, it was found for both
Advisability (51% vs. 22%, p=.012) and Emotion (47% vs. 23%,
p=.045). It was significant for Advisability with Pension and
Unemployment combined (56% vs. 22%, p=.004).
Recall that omission bias was found across all conditions for
Pension but not for Unemployment. That result combines with the
result just found to produce the following effects: for
Unemployment, omission bias was found for the worse outcome (53%
vs. 25%, p=.020), and a reverse bias was found for the better
outcome (30% vs. 55%, p=.050); for Pension, a bias was found for
the worse outcome (67% vs. 13%, p=.000) but not for the better
outcome (43% vs. 34%).
In sum, the interaction predicted by norm theory and found in
Experiment 1 was replicated for both emotion and advisability
judgments (except for Pension Advisability). Omission bias was
stronger when it led to the worse outcome, regardless of the
relation of the possible outcomes to the status-quo.
A fourth question concerns the bias toward the status-quo itself.
In a sense, the omission in this study is associated with the
status-quo, since it is the current pension fund or government
policy that stays in effect if nothing is done. But another
feature of the status-quo is the current interest rate or
unemployment rate, and we can ask whether subjects tend to favor
the option that maintains this rate the same, other things being
equal. To assess this status-quo effect, we subtracted the raw
ratings of cases with 8% status-quo and 9% outcome or with 9%
status-quo and 8% outcome from the ratings of cases with 8%
status-quo and 8% outcome or with 9% status-quo and 9% outcome.
This effect was significant for Advisability ratings (Pension
Advisability effect=8.4, t(62)=2.69, p=.009; Unemployment
Advisability effect=6.7, t(62)=2.35, p=.022; Advisability ratings
combined, t(62)=3.45, p=.001) but not for Emotion ratings
(Pension Emotion effect=4.4, t(60)=1.97, p=.054; Unemployment
Emotion effect=1.0, t(59)=0.55, p=.586, Emotion ratings combined,
t(59)=1.60, p=.115). The difference between Advisability and
Emotion in the size of the effect was almost significant
(t(59)=1.93, p=.058). The effect was significant over
Advisability and Emotion ratings combined (t(59)=3.09, p=.003).
In sum, maintaining the most important feature of the outcome,
the percent interest or the unemployment rate, seems to be
preferred, especially for Advisability ratings.
A fifth question is whether omission was used as the reference
point, as predicted by loss aversion and norm theory. If so,
then the mean rating for both options in a pair would be higher
when the omission is associated with the worse option than when
it is associated with the better option. Also, we would expect
`zero' ratings to be more frequently given to omissions than to
acts. Both of these results were found by Spranca et al. (1991,
Experiment 4). The first result was found here too: for
Advisability (t(62)=2.58, p=.012); for Emotion (t(59)=2.81,
p=.007; for Pension (t(59)=2.82, p=.007); and for Unemployment
(t(59)=2.64, p=.011). However, zero ratings did not differ for
acts (5.9%) and omissions (5.2%, Wilcoxon test z=1.45). Zero
ratings were rarely used except for options that maintained the
exact percentage of the status-quo, regardless of whether this
was associated with an act or omission. (12.9% of these options
received zero ratings, as opposed to 2.2% of the other options in
the same decisions.) In sum, evidence for the use of the
omission as the reference point is mixed.
In general, Experiment 2 found evidence for an overall omission
bias (better and worse outcomes combined) in one scenario. In
all scenarios, omission bias was stronger for the worse outcome
than for the better outcome. We also obtained further evidence
for a status-quo effect that is independent of omission bias, and
we found some evidence for the use of omission as a reference
point.
Experiment 3
Experiment 2 asked subjects to give ratings, but other studies
have asked for direct comparisons. Experiment 3 was a
replication of Experiment 2 using direct comparisons instead of
ratings.
Method
89 subjects, solicited as in Experiment 2, were given a
questionnaire introduced as follows, for pension funds:
`In each of these cases, imagine two people, P and Q, working for
different employers in jobs that they will be in for some time.
They each have different pension funds. Each fund pays a certain
rate of interest this year. For next year, each employer gives
each person a chance to switch to a new fund by checking a box on
a form. The only information they have about their options is
the expected interest rate for the next year.
In each case, indicate which person, P or Q, made the better
decision (or the less bad decision). If you think that their
decisions were exactly equally good, write `equal,' or an equal
sign. The two decisions to compare always yield the same
expected interest in the next year.'
The first item for one version then read:
`P's current pension fund earns 7% interest. Next year, this
fund is expected to earn 8%, and the new fund is expected to earn
9%. P switched to the new fund.
Q's current pension fund earns 7% interest. Next year, this
fund is expected to earn 9%, and the new fund is expected to earn
8%. Q did not switch to the new fund.'
Parallel changes were made for the unemployment cases. For the
emotion ratings, the subjects were asked to indicate which of two
people felt better (or less bad) or whether the two felt the
same. The outcomes were unanticipated, as in Experiment 2. The
items used were based on numbers 1-8 in Table 2. Only Forms A
and B were used. 17 subjects were omitted from analysis because
they gave `equal' in every case, and one subject was omitted for
giving extremely idiosyncratic responses, leaving 71.
Results
Data were analyzed as in Experiment 2. Of course, the comparison
of act and omission was done directly by the subject rather than
through comparison of numerical ratings.
The overall bias toward omissions (better and worse combined) was
found for the decision comparisons but not for the emotion
comparisons. The effect was significant for Pension decisions
(49% vs. 20%, p=.006), for Unemployment decisions (43% vs. 20%,
p=.024), but not for Pension emotions (30% vs. 35%) or
Unemployment emotions (34% vs. 31%). These results contrast
inexplicably with the results of Experiment 2, where an overall
bias toward omissions was found for Pension items but not for
Unemployment items.
The act-omission effect was greater for worse outcomes than for
better outcomes in Pension decisions (56% vs. 13%, p=.000),
Pension emotions (69% vs. 11%, p=.000), and Unemployment emotions
(47% vs. 16%, p=.002), but not in Unemployment decisions (37% vs.
26%). Here, it is of interest that the overall bias toward
omission was greater for decisions than for emotions, but the
difference between better and worse outcomes in this effect was
greater for emotions than for decisions: for 20 subjects this
pattern was true, but for only 6 the reverse was true (p=.011).
It would be difficult, then, to explain the overall bias toward
omissions in terms of just the fact that acts leading to the
worse outcome are judged to be particularly bad. (For the worse
outcome, omissions were not as bad as acts: p=.000 for Pension
decisions; .059 for Unemployment decisions; .000 for Pension
emotions; .017 for Unemployment emotions. For the better
outcome, acts were better than omissions for Pension emotions
[p=.000] and Unemployment emotions [p=.018], but the difference
for the decision items was not significant.)
The act-omission effect was no greater for losses than for gains
in emotions, in decisions, for Pension, or for Unemployment.
Although the results from Experiment 3 differ from those from
Experiment 2 in the conditions for overall omission bias, both
experiment show an omission bias for the worse of two outcomes,
this replicating the major finding of previous studies. The
results of both experiments also agree strongly in finding an
effect of better vs. worse on omission bias, as predicted by norm
theory, and in finding no effect of gain vs. loss, thus failing
to support unequal weighting.
Experiment 4
The unequal weighting hypothesis has received no support from the
experiments reported so far. Changing the status-quo did not
affect omission bias. The evidence that led to this hypothesis,
however, came from the Ritov and Baron (1990) study of
hypothetical vaccination decisions. Experiment 4 returns to the
scenario used in that study to test again the prediction that
omission bias for losses will increase as the status-quo becomes
better.
Method
41 subjects, solicited as in Experiments 2-3, were given a
questionnaire in which they were told, `Imagine that you are
married and have one child, who just became one year old. You
live in a country in which one-year-olds sometimes die from a
kind of flu. A vaccine for this kind of flu has just become
available in your country. It has been tested extensively
elsewhere. The vaccine completely prevents the flu, but it
sometimes causes side effects that can be fatal. Aside from
death, neither the flu nor the vaccine causes any long-lasting
ill effects. The vaccine is free and is given along with other
vaccines that you must have anyway, so no extra effort or expense
is required.
In each of the following situations, you are to indicate the
probability that you would vaccinate your child, on the
basis of three facts:
1. The overall current death rate from the flu in the last
year. This number has been about the same for several years.
2. The predicted overall death rate for unvaccinated
children in the coming year. You can assume that this
prediction is accurate. It can differ from the current death
rate because of a sudden and unexplained increase or decrease in
the flu worldwide.
3. The predicted overall death rate for vaccinated
children in the coming year.
Noone can predict which children will die from the flu and
which children will die from the vaccine. A child who dies from
the vaccine would not necessarily have died anyway from the flu,
and vice versa.'
In case 1, subjects were told: `10 out of 10,000 children
currently die from the flu. 10 out of 10,000 unvaccinated
children will die from the flu in the coming year.' They were
then asked to indicate the probability that they would vaccinate
for four different death rates for vaccination, 1, 5, 9, and 13
out of 10,000. The status-quo was changed from 10 out of 10,000,
for about half of the subjects, to 5 for case 2 and 15 for case
3. For the other half it was 15 for case 2 and 5 for case 3.
All other information, including the expected death rate without
vaccination (the omission), was left unchanged. Case 4 was
identical to case 1 except that subjects were told, `In the
following case, imagine that the vaccine causes no deaths, but it
is not fully effective. Vaccinated children could die from the
flu, because the vaccine could fail.' Ritov and Baron (1990)
found increased willingness to vaccinate in this kind of case
even in subjects who would accept little risk from the vaccine in
a case like case 1.
Four additional cases were presented, in which subjects were
asked to `rate the quality of the decisions in each case on
a scale in which -100 indicates a very bad decision and 100
indicates a very good decision.' In case 1 of this group,
subjects were told, `7 out of 10,000 children currently die from
the flu. 7 out of 10,000 unvaccinated children will die from the
flu in the coming year.' They then rated the quality of
vaccinating and not vaccinating for vaccination death rates of 1,
5, 9, and 13 out of 10,000. The next two cases had a status-quo
of 13 or 1 death, in balanced order. The last case again had a
status-quo of 7, and subjects were asked to assume that all
deaths of vaccinated children were from vaccine failure.
Results
The basic omission bias was demonstrated in some of the measures.
In the first case, 19.5% of the subjects gave less than 50% as
their chance of vaccinating even when vaccination led to a risk
of only 1 out of 10,000. When the risk was 5 (half the risk of
the flu), 34.1% indicated that they probably would not vaccinate;
and, when the risk was 9, 51.2% so indicated. (75.6% indicated
they would probably not vaccinate when the risk was 13, and 12.2%
indicated that they probably would vaccinate, despite the
increased risk of doing so.) The probability ratings for the
first case were significantly lower than those in the last case,
vaccine failure (t(40)=4.37, p=.000). The same results were
found for the quality ratings: ratings of vaccination were higher
for the last case (t(40)=4.24, p=000), and ratings of
nonvaccination were lower (t(39)=3.67, p=.001). One measure that
did not show an omission bias was the overall difference between
the quality ratings of vaccination and nonvaccination
(t(39)=1.04). The finding of omission bias even when
nonvaccination is equal to the status-quo (as it was in case 1)
is new; in Ritov & Baron (1990), the status-quo was always
assumed to be zero risk.
As hypothesized, omission bias was greater (probability ratings
lower) when the status-quo was better (lower death rate). The
ratings of probability of vaccination were higher when the
status-quo death rate was higher (mean ratings of 52 vs. 47,
t(40)=3.30, p=.002). Likewise, the quality ratings of
vaccinating were higher (27 vs. 14, t(40)=2.89, p=.006) and the
ratings of not vaccinating were lower (-6 vs. 8, t(40)=3.40,
p=.002).
This result appears to support unequal weighting. An additional
experiment, however, suggests that the result was an artifact.
After an essentially identical introduction to that described
above, 29 subjects were asked about the probability of their
vaccinating in each of 9 cases. The status-quo, nonvaccination
outcome, and vaccination outcome were described for each case.
(The status-quo and nonvaccination outcomes were therefore
repeated every time, even when they were held constant across
several values of the vaccination outcome, unlike the version
described above.) The status-quo was 10 out of 10,000 for the
first three cases and 5 and 15 for the next two groups of 3, in
balanced order. The nonvaccination outcome was always 10. The
vaccination outcome was 0, 5, and 10 for the members of each
group of three cases, respectively. At the end of the
questionnaire, subjects were asked if they gave different answers
to any of the cases that differed only in status-quo (cases 4 and
7, 5 and 8, or 6 and 9), and, if so, they were asked to explain
their ratings.
Out of the 29 subjects, 16 showed the effect of status-quo, with
higher probability ratings when the status-quo was higher
(worse); 2 showed the reverse effect; 8 gave identical ratings;
and 1 gave mixed ratings (some higher, some lower). Out of the
16 subjects who showed the effect, six subjects gave
justifications indicating that the status-quo should be taken
into account in predicting the results of nonvaccination, for
example: `In [the high death-rate status-quo case] it seemed as
the flu struck more people. Therefore it would be more likely to
take more precious children the next year. It seemed more
serious.' `In [low status-quo], past numbers show that in the
past, the flu hasn't been fatal. In [high status-quo] the fact
that 20/10,000 died of flu the year before makes me think the
vaccine is still safer than risking the flu despite predictions.'
An additional 6 subjects referred to the status-quo as
justification but without spelling out their reasoning (as these
other subjects did), e.g., `because 0 people died from the flu,
so it probably wasn't that big.' The remaining 4 subjects gave
incomprehensible or irrelevant justifications, e.g., `The amount
of children that die from the vaccination increases, the more
probably my child will die.'
In sum, the status-quo seems to be relevant because subjects
think it should be taken into account in predicting the results
of nonvaccination. No subject said anything that could be
interpreted as supporting unequal weighting, e.g., that the
result of vaccination was worse than the status-quo or that it
was more important to consider the effect of acting. The results
leave us skeptical about the existence of any true effect of the
status-quo on the magnitude or direction of omission bias, as
predicted by unequal weighting.
Experiment 5
In Experiments 1-3, the omission could be seen as maintaining the
status-quo (e.g., the stock or fund owned). It is not clear,
then, whether the results we have found concerning the
interaction between act vs. omission and better vs. worse are
specific to omission bias. They could also pertain to the
status-quo bias. In Experiment 4, the omission did not maintain
the status quo, but we could not carry out a clear test of the
interaction. (The omission outcome varied across cases, and the
act outcome varied within cases.) In Experiment 5, we compared
the status-quo and omission biases in different items. We
examined the effect of gain vs. loss and of better vs. worse on
each bias. The cases were new, so they also bear on whether our
results are robust across different kinds of items.
Method
89 subjects, solicited as in Experiments 2-4, completed a 16-item
questionnaire (42 in the order to be described, 47 with the items
in reverse order).
Subjects were asked to `rate the advisability of each
option, on a scale from -100 (bad decision) to 100 (good
decision). Decisions with higher numbers should always be
preferable to decisions with lower numbers. Feel free to go
beyond -100 or 100 if you need to.
In the following cases, imagine that you have high blood
cholesterol and that you are taking medication to reduce it.
Because of other changes in your medical condition, your
cholesterol level will not remain the same from year to year,
even if you keep taking the same medication. All possible
medications are equivalent in side effects and cost, but they are
chemically different.'
The status-quo condition was presented first. The subject had to
make a decision; no default was offered. The first item read
`With your current medication, your level is 50% above normal.
You must choose whether to:
A. stay with that medication, in which case your level will be
40% above normal, or
B. switch to a new medication, in which case your level will be
30% above normal.'
In the next case, the 30% and 40% were switched, so as to
unconfound better vs. worse from stay (i.e., the status-quo) vs.
change. In the next pair of cases, the current level was 20%
instead of 50%, so that these cases were losses rather than
gains.
The next four cases repeated the first four except that they
concerned omission bias rather than the status-quo bias. For
example, item 5 read, `With your current medication, your level
is 50% above normal. You will continue to receive this
medication automatically from the pharmacy unless you ask the
doctor to change the prescription. Your options are:
A. do nothing, in which case your level will be 40% above
normal, or
B. switch to a new medication, in which case your level will be
30% above normal.'
Finally, in the last 8 cases were a repeat of the first 8 except
that the subject was told, `In the following cases, you are a
physician. Your patient has high blood cholesterol and is taking
medication to reduce it. ...' (As noted, the 16 cases were
reversed for 47 of the subjects.)
Results
Table 4 shows the mean ratings, collapsed over the two
perspectives (patient vs. physician) and over the two orders.
Perspective and order did not affect any of the comparisons
reported. Results were analyzed as in Experiment 2.
Table 4.
Means of advisability ratings for all conditions in
Experiment 5. (Standard deviations ranged from 42 to 68.)
Status-quo or omission:
50% 20%
Outcome:
30% 40% 30% 40%
Status-quo condition
stay 70 0 29 -51
change 72 -7 25 -56
Omission condition
omit 68 1 27 -54
act 72 -14 27 -61}
The omission bias was present overall, that is, omissions were
rated higher than acts (44% vs. 22%, p=.019). The status-quo
bias was not present (30% vs. 25%).
As found in Experiments 1-3, omission bias was greater for the
worse option than for the better one (33% vs 13%, p=.012), but
the bias was no different for losses than for gains (25% vs.
25%). The status-quo bias did not differ as a function of worse
vs. better (17% vs. 21%) or losses vs. gains (19% vs. 20%). In
sum, the main result that omission bias is greater for losses
does not seem to depend on the confounding of omission and
status-quo.
As in Experiment 2, to test whether the omission was the
reference point, we compared the mean ratings for both options in
cases in which the omission was the worse outcome with the
ratings in cases in which it was the better outcome. Again,
ratings were higher in the former cases (t(88)=3.22, p=.002). We
also did the same test for the status-quo, and this effect was
not significant (t(88)=1.30); the effect for omissions was larger
than that for the status-quo (t(88)=2.11, p=.038). We have
further evidence, then, that the omission tends to be considered
as a reference point, or as closer to the reference point, but
the status-quo does not.
On the other hand, the reference point would also be expected to
be given a rating of zero, insofar as zero ratings were given at
all. An analysis of zero ratings found (as in Experiment 2)
little tendency to assign them to the omission (3.4% for
omissions, 2.7% for acts, Wilcoxon z=1.18). Instead, zero
ratings tended to be assigned to the option closer to the
status-quo, in both status-quo and omission conditions (4.4% for
that option, 1.5% for the option farther from the status-quo).
In sum, the results of Experiment 5 agreed with previous results
indicating that acts that lead to the worse outcome are
particularly bad. This effect is peculiar to acting and not to
simply changing the status-quo. Conflicting evidence was found
for the use of the omission as the reference point, as predicted
by norm theory and loss aversion.
General discussion
Our results consistently support the hypothesis that acts leading
to the worse of two possible outcomes are considered particularly
bad, whether these outcomes are predictable or not. Subjects do
compare outcomes to other outcomes in the choice set, not just to
the status-quo. The comparison between these outcomes is more
relevant to the evaluation of acts than to the evaluation of
omissions. This result is consistent with norm theory as we
interpret it.
Although subjects may also compare outcomes to the status-quo,
they do not do so differentially for acts or omissions. The
devaluation of acts that lead to the worse outcome is just as
great when both outcomes are gains than when both are losses
relative to the status-quo. The last experiment indicates that
acts that lead to the worse outcome are considered bad because
they are acts, not because they change the status-quo.
Most experiments contained at least one scenario or condition in
which acts were considered worse than omissions, averaging over
better vs. worse and gain vs. loss. Experiment 3 found that this
general devaluation of acts was distinct from the particular
devaluation of acts leading to the worse outcome (leading to an
interaction between act vs. omission and better vs. worse): the
general act-devaluation effect was found only for decisions, but
the interaction effect was found only for emotions. The general
effect is not predicted by norm theory, but it is explained by
loss aversion. Alternatively, subjects might simply favor
omissions, giving them `extra points.' The conditions for this
general bias toward omissions are as yet unknown. Although this
effect was not found consistently, the general bias toward
omissions for the worse outcome was consistent across the present
experiments as well as previous ones.
Both norm theory and loss aversion assume that the omission is
taken as the reference point. Our results support this
prediction inconsistently. On the one hand, overall ratings of
both options in a choice are higher when the omission was
assigned to the worse option than when it is assigned to the
better option. (This does not hold for the assignment of the
status-quo to the better or worse option.) On the other hand,
subjects are no more likely to assign zero ratings to omissions
than to acts. (Spranca et al., 1991, Experiment 4, did, however,
find such an effect for zero ratings.) Both of these tests
assume that subjects will tend to assign zero, or ratings closer
to zero, to whatever they take to be the reference point. This
assumption itself could often be incorrect, so the weak results
of these tests does not refute the theories in question.
The unequal-weighting hypotheses did not fare well. Changing the
status-quo had essentially no effect. Moreover, in Experiment 4,
reversals were found even when the status-quo was in between the
outcomes of acts and omissions. Reversals, then, are more easily
explained by loss aversion applied to separate dimensions (e.g.,
losses from disease vs. losses from vaccination), as argued
earlier. Consistent with this account is Ritov and Baron's
(1990) finding, replicated in Experiment 4, that the bias toward
nonvaccination was reduced when the children who would die from
the vaccine were the same ones who would have died anyway from
the flu. In this case, it might be more difficult for subjects
to segregate the dimensions.
Most of our results can be accounted for by a heuristic rule:
avoid acts that lead to harm (compared to the outcome of
omissions) even when they are compensated by benefit (again,
compared to the outcome of omissions). Such a heuristic also
accounts for reversals, as well as for demonstrations of the
status-quo bias in which the status-quo is the default. This
heuristic also accounts for previous results in which harmful
acts were compared to harmful omissions across cases. Such a `do
no harm' heuristic does not account for all the results, however.
In some situations, we find an overall bias toward omissions,
regardless of whether they lead to the better or worse outcome.
This effect is not fully accounted for by the devaluation of acts
that lead to the worse outcome. We do not know why this effect
occurs.
References
Asch, D., Baron, J., Hershey, J. C., Kunreuther, H., Meszaros,
J., Ritov, I., & Spranca, M. (1993). Determinants of
resistance to pertussis vaccination. Manuscript, Department of
Psychology, University of Pennsylvania.
Baron, J. (in press a). Morality and rational choice.
Dordrecht: Kluwer.
Baron, J. (in press b). Nonconsequentialist decisions.
Behavioral and Brain Sciences.
Becker, S. W., Ronen, J., & Sorter, G. H. (1974). Opportunity
costs - an experimental approach. Journal of Accounting
Research, 12, 317-329.
Feldbrugge, F. J. M. (1966). Good and bad Samaritans: A
comparative survey of criminal law provisions concerning failure
to rescue. American Journal of Comparative Law, 14,
630-657.
Gleicher, F., Kost, K. A., Baker, S. M., Strathman, A. J.,
Richman, S. A., & Sherman, S. J. (1990). The role of
counterfactual thinking in judgments of affect. Personality
and Social Psychology Bulletin, 16, 284-295.
Howard, R. A., Matheson, J. E., & North, D. W. (1972). The
decision to seed hurricanes. Science, 176, 1191-1202.
Inglehart, J. K. (1987). Compensating children with
vaccine-related injuries. New England Journal of Medicine,
316, 1283-1288.
Kahneman, D., Knetsch, J. L., & Thaler, R. H. (1990).
Experimental tests of the endowment effect and the Coase theorem.
Journal of Political Economy, 98, 1325-1348.
Kahneman, D. & Miller, D. T. (1986). Norm theory: Comparing
reality to its alternatives. Psychological Review, 93,
136-153.
Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis
of decisions under risk. Econometrica, 47, 263-291.
Kahneman, D., & Tversky, A. (1982). The psychology of
preferences. Scientific American, 246, 160-173.
Knetsch, J. L., & Sinden, J. A. (1984). Willingness to pay and
compensation: Experimental evidence of an unexpected disparity in
measures of value. Quarterly Journal of Economics, 99,
508-522.
Landman, J. (1987). Regret and elation following action and
inaction: Affective responses to positive versus negative
outcomes. Personality and Social Psychology Bulletin, 13,
524-536.
Mitchell, R. C., & Carson, R. T. (1989). Using surveys to
value public goods: The contingent valuation method.
Washington: Resources for the Future.
Ritov, I., & Baron, J. (1990). Reluctance to vaccinate: omission
bias and ambiguity. Journal of Behavioral Decision Making,
3, 263-277.
Ritov, I., & Baron, J. (1992). Status-quo and omission bias.
Journal of Risk and Uncertainty.
Samuelson, W., & Zeckhauser, R. (1988). Status quo bias in
decision making. Journal of Risk and Uncertainty, 1, 7-59.
Schweitzer, M. E. (in press). Disentangling status-quo and
omission effects: An experimental analysis. Organizational
Behavior and Human Decision Processes.
Singer, P. (1979). Practical ethics. Cambridge University
Press.
Sugarman, D. B. (1986). Active versus passive euthanasia: An
attributional analysis. Journal of Applied Social
Psychology, 16, 60-76.
Spranca, M., Minsk, E., & Baron, J. (1991). Omission and
commission in judgment and choice. Journal of Experimental
Social Psychology, 27, 76-105.
Thaler, R. (1980). Toward a positive theory of consumer choice.
Journal of Economic Behavior and Organization, 1, 39-60.
Tversky, A., & Kahneman, D. (1991). Loss aversion in risky and
riskless choice: A reference dependent model. Quarterly
Journal of Economics, 107, 1039-1061.
Viscusi, W. K., Magat, W. A., & Huber, J. (1987). An
investigation of the rationality of consumer valuation of
multiple health risks. Rand Journal of Economics, 18,
465-479.
Footnotes:
1Another case of failure to evaluate
outcomes may occur in the neglect of opportunity costs (e.g.,
Becker et al., 1974). Opportunity costs are described as
foregone gains relative to the status-quo.
File translated from
TEX
by
TTH,
version 3.59.
On 21 May 2004, 18:42.