Content uploaded by Hal Richard Arkes
Author content
All content in this area was uploaded by Hal Richard Arkes
Content may be subject to copyright.
Current Directions in Psychological
Science
22(5) 356 –360
© The Author(s) 2013
Reprints and permissions:
sagepub.com/journalsPermissions.nav
DOI: 10.1177/0963721413489988
cdps.sagepub.com
The hindsight bias manifests in the tendency to exagger-
ate the extent to which a past event could have been
predicted beforehand. First systematically investigated by
Fischhoff (1975), the bias is sometimes called “Monday
morning quarterbacking” or the “I knew-it-all-along
effect” (Wood, 1978). The hindsight bias has particularly
detrimental effects in the domain of medical decision
making. I begin with the classic study demonstrating how
the bias diminishes the salutary impact of a medical edu-
cation exercise.
The Hindsight Bias as an Impediment
to Learning
A clinicopathologic conference (CPC) is a dramatic event
at a hospital. A young physician, such as a resident, is
given all of the documentation except the autopsy report
that pertains to a deceased patient. After studying the
material for a week or so, the physician presents the case
to the assembled medical staff, going over the case and
listing the differential diagnosis, which consists of the
several possible diagnoses for this patient. Finally, the
presenting physician announces the diagnosis that he or
she thinks is the correct one. The presenter then sits
down, sweating profusely, as the pathologist who did the
autopsy takes the podium and announces the correct
diagnosis. The cases are chosen because they are diffi-
cult, so the presenting physician’s hypothesis often is
incorrect.
The CPC is supposed to be an educational experience.
Its goal is to enlighten the audience members about diag-
nosing a particularly challenging case. However, after
hearing the pathologist’s report, which contradicts the
diagnosis made by the presenting physician, many audi-
ence members think, “Why aren’t we hiring residents as
astute as my cohort of residents? This diagnosis was
easy.” The audience members do not learn from the
instructive case presented at the CPC. Instead, they criti-
cize the presenter, because in hindsight, they think the
case was relatively obvious.
A CPC is fertile ground for the manifestation of the
hindsight bias; after the correct diagnosis is disclosed, it
seems as if it could easily have been discerned before-
hand. Dawson et al. (1988) interrupted eight CPCs at two
critical junctures. At each CPC, after the presenter listed
the five possible diagnoses, the researchers asked half of
the audience to assign a probability to each diagnosis
(i.e., the likelihood that it was correct). These participants
were in the foresight group, because they were asked to
provide data before the correct diagnosis was revealed.
These data were collected, the pathologist then revealed
the true diagnosis, and the CPC was paused a second
time for the other half of the audience to provide data.
489988CDPXXX10.1177/0963721413489988ArkesHindsight Bias in Medical Decision Making
research-article2013
Corresponding Author:
Hal R. Arkes, Department of Psychology, Ohio State University, 1827
Neil Ave., 240N Lazenby Hall, Columbus, OH 43210
E-mail: arkes.1@osu.edu
The Consequences of the Hindsight Bias
in Medical Decision Making
Hal R. Arkes
Ohio State University
Abstract
The hindsight bias manifests in the tendency to exaggerate the extent to which a past event could have been predicted
beforehand. This bias has particularly detrimental effects in the domain of medical decision making. I present a
demonstration of the bias, its contribution to overconfidence, and its involvement in judgments of medical malpractice.
Finally, I point out that physicians and psychologists can collaborate to the mutual benefit of both professions.
Keywords
hindsight bias, overconfidence, malpractice, decision support systems
Hindsight Bias in Medical Decision Making 357
The researchers asked them to assign a probability to
each of the five possibilities as if they had not just been
informed of the right answer. These participants were in
the hindsight group, because they were asked to provide
data after the correct diagnosis was revealed. For each
case, Dawson et al. asked two experts who had attended
their domain-appropriate CPC to indicate, on a 10-cm
line, what proportion of good clinicians would choose
the correct diagnosis. The average of the two experts’
ratings was used to divide the eight CPCs into two
quartets—one being the four more difficult cases and the
other being the four less difficult ones. Dawson et al.
divided the attendees into groups of less experienced
and more experienced physicians on the basis of their
training and seniority. As Figure 1 shows, in three of the
four experience–diagnostic-difficulty groups, the hind-
sight participants estimated the correct answer to be
more likely than the foresight group did. The difference
averaged approximately 12 percentage points. Hindsight
physicians mistakenly think that the case was easier than
it really was (as evidenced by the predictions of the fore-
sight subjects). The educational benefit of the CPC is con-
sequently diminished, because the audience members
think there is little or nothing to be learned, given that
they retrospectively judge their diagnoses to have been
relatively accurate.
Note that the most senior physicians diagnosing the
most difficult cases escaped the hindsight bias. They real-
ized that they were not likely to have made the correct
diagnosis because a particular disease was so very rare or
the presentation of the disease was so very abnormal.
Given that the hindsight bias compromises the educa-
tional value of a CPC, it would be helpful if there were
some way to diminish its negative impact. Following the
guidance of Slovic and Fischhoff (1977), Lord, Lepper,
and Preston (1984), and Koriat, Lichtenstein, and Fischhoff
(1980), Arkes, Faust, Guilmette, and Hart (1988) modified
the typical hindsight design. Before some neuropsychol-
ogists stated in hindsight the probability that they would
have assigned to the correct diagnosis, they first had to
explain how each alternative diagnosis might have been
correct. What symptoms might have been consistent with
the diagnoses that were not the correct one? This simple
exercise reduced the magnitude of the hindsight bias
by 71%.
Overconfidence
The hindsight bias also plays an insidious role in generat-
ing unwarranted overconfidence. Consider the case of
right-sided heart catheterization, a procedure in which a
long slender tube is inserted into an artery, usually in the
groin area, and then is very carefully snaked through the
circulatory system, ultimately reaching the heart. There
the catheter can be used to monitor various aspects of
blood flow (“hemodynamic functioning”). Unfortunately,
the procedure of catheterization poses some risk to the
patient; adverse events do infrequently occur during this
procedure. Some physicians assert that various indices of
hemodynamic functioning can be accurately and confi-
dently estimated without the use of the catheter; that is,
some physicians think that by using only such noninva-
sive procedures as assessing blood pressure, they can
gather enough information about hemodynamic func-
tioning that a catheterization is unnecessary. Thus,
adverse events due to catheterization would be avoided.
However, when only noninvasive procedures are used,
are confident physicians more likely to be accurate in
their estimates than physicians who are not confident in
their estimates based solely on noninvasive procedures?
Are confident physicians justified in eschewing catheter-
ization because it is not necessary given the presumed
high accuracy of their estimates based on noninvasive
measures? Dawson et al. (1993) checked whether physi-
cians’ confidence in three indices of hemodynamic func-
tioning was appropriate. Before the catheter was inserted
into the patient, each physician estimated these three
indices and stated his or her confidence in each estimate.
Then the catheter was inserted, and the three levels were
directly measured. The results were startling: There was
no relation whatsoever between the accuracy of an esti-
mate and the confidence a physician assigned to that
estimate. One relation involving confidence was statisti-
cally significant, however: the relation between years of
experience and expressed confidence. Veteran physi-
cians were more confident in their estimates than were
junior physicians, even though confidence and accuracy
were unrelated. Yang and Thompson (2010) reported a
similar finding among nurses.
Foresight Hindsight
Probability Est. of Correct Diagnosis
24
26
28
30
32
34
36
38
40
42 41.1 More Experienced
40.2 Less Experienced
38.5 Less Experienced
24.5 More Experienced
Fig. 1. Mean estimated probabilities of the correct diagnosis as a func-
tion of the timing of the estimates (foresight vs. hindsight), experience
of the estimators (less = thin lines vs. more = bold lines), and case
difficulty (less difficult = solid lines vs. more difficult = dashed lines).
358 Arkes
This research provided important guidance: Physicians’
confidence in their estimate of hemodynamic functioning
should not be used as a basis for deciding whether a
catheterization is needed.
Why was physician confidence unrelated to accuracy?
This experiment constituted the first time these physi-
cians ever had to provide a confidence rating for their
estimates of hemodynamic status. In their prior experi-
ence, they had inserted the catheter, noted the levels of
each index, and concluded that each was “pretty much
what I thought it would be.” This is a manifestation of
the hindsight bias, the belief that they “knew it all along.”
In fact, they did not know it all along. The hindsight
bias merely provides unwarranted post hoc confirmation
of their ghost estimate of hemodynamic functioning.
Physicians that are more senior were more confident
because they had experienced more of these bogus “con-
firmations.” By being forced to give an a priori estimate in
our experiment, the physicians had experienced their first
learning trial. In order to improve one’s estimates, one has
to make an actual estimate and then receive feedback. No
prior overt estimates had apparently occurred in the
experience of these physicians.
Note that most of the studies cited in the hindsight and
overconfidence sections of the article by Dawson et al.
(1993) are 20 years old or more. Because it is difficult to
do naturalistic “on-the-job” psychological research with a
statistically sufficient number of physicians working in
high-stakes situations, research that is more recent is
scarce but greatly needed.
Malpractice
Malpractice verdicts are always rendered from the per-
spective of hindsight. An adverse event has already taken
place, and jurors are asked to consider whether a physi-
cian has met the standard of care in his or her treatment of
the patient. I once asked a physician if he practiced “defen-
sive medicine,” which is generally defined as treatment not
designed to promote the health of the patient but instead
to reduce the possibility of successful malpractice claims
against the practitioner. The physician, who was well
versed in the psychology of medical decision making,
promptly answered that he did practice defensive medi-
cine. He thought that jurors would succumb to the hind-
sight bias. They would think that he should have easily
been able to make the correct diagnosis. Due to the hind-
sight bias the physician would test for every possible diag-
nosis no matter how unlikely. Defensive medicine was his
way of counteracting the hindsight bias.
This physician made a reasonable argument that was
a testament to the power he attributed to the hindsight
bias. One possible way to lessen one’s vulnerability to
the hindsight bias is to use a computer-based decision
support system (DSS). Such systems are designed to
assist physicians in diagnosis and treatment. They pro-
vide advice to practitioners, and they have been shown
to be superior to physicians in a wide variety of diagnos-
tic contexts (Dawes, Faust, & Meehl, 1989). Because the
use of a DSS might represent modern medicine in the
eyes of a juror, perhaps any physician who used such an
advanced tool might be insulated from jurors’ ire com-
pared with a physician who did not use computer assis-
tance. On the other hand, if the use of a DSS were to be
perceived as an abrogation of the physician’s responsibil-
ity, then the use of a DSS might foster greater probability
of being found liable for malpractice.
Arkes, Shaffer, and Medow (2008) tested these two
opposing hypotheses using a realistic video recording of
the key portions of a staged malpractice trial. The same
adverse medical outcome occurred regardless of whether
the physician used a DSS. The eight scenarios also varied
the severity of the symptoms and whether the physician
heeded the advice of the DSS (or in the case of the physi-
cians who used no DSS, whether the physician happened
to choose the course that would have been recom-
mended by a DSS). The good news is that the use of the
aid did not increase mock jurors’ willingness to find the
physician liable for malpractice. However, if a physician
used the aid but defied its recommendation and was
found liable, the jurors were more punitive toward the
physician than if the aid was not used or was used and
heeded. This has led some physicians I have met to
decide that they would not want to use a DSS, because
our results suggest that if they were to be found liable for
malpractice, they would have had to heed the DSS, even
if they disagreed with its advice, in order to escape the
wrath of punitive jurors. Many times, physicians want to
defy the aid even if it would have been better for them to
heed it (Dawes et al., 1989).
Why Should a Psychologist Be
Interested in Medical Decision
Making?
There are at least two reasons why psychologists should
become involved in the domain of medical decision mak-
ing. First, as illustrated in many of these examples, psy-
chologists can contribute to the medical education
of physicians, both in their formal training and in their
on-the-job experience. Some of the studies reviewed sug-
gest how physicians can reduce their susceptibility to
the hindsight bias and improve the calibration of their
confidence estimates. Gigerenzer (2002) has demon-
strated ways to substantially improve physicians’ consid-
eration of risk. Given physicians’ surprising difficulty in
Hindsight Bias in Medical Decision Making 359
comprehending health statistics (Wegwarth & Gigerenzer,
2011), techniques that improve physicians’ performance
in this domain would be extremely valuable. Second, to
improve health care and health care decisions, patients
and journalists must also improve their understanding of
health statistics and medical data. Psychologists have
already contributed very substantially in this endeavor
(e.g., Garcia-Retamero & Galesic, 2010; Tait, Voepel-
Lewis, Zikmund-Fisher, & Fagerlin, 2010), but much more
needs to be done.
Not only can psychologists contribute to medicine, but
also research in the domain of medical decision making
can contribute to psychological theory in at least two
ways. First, medical decisions generally involve impor-
tant, highly consequential situations. However, most psy-
chological theories are tested in much less significant
contexts. For example, the earliest demonstration of the
hindsight bias (Fischhoff, 1975) used laypersons consid-
ering the potential outcomes of obscure historical events.
It is important to ascertain whether psychological theo-
ries and findings also apply when lives and malpractice
verdicts are at stake. Second, most psychology experi-
ments in the domain of decision making are done with
descriptions of lotteries, gambles, and risky situations.
However, Hertwig, Barron, Weber, and Erev (2004) have
shown that decisions based on experience can differ sub-
stantially from those based on mere descriptions. Note
that physicians use their experience in rendering their
decisions, but patients—even the best informed ones—
generally have only a description of the probabilities and
possible outcomes. Thus, the medical situation is a good
venue in which to study the difference between decision
making based on personal experience and decision
making based solely on descriptions. I am suggesting
that medicine can provide an important realm for the
testing and furtherance of psychological theory, and psy-
chologists can contribute to the performance and training
of physicians. Members of both professions can benefit
from mutual collaboration.
Recommended Reading
Arkes, H. R., & Gaissmaier, W. (2012). Psychological research
and the PSA test controversy. Psychological Science, 23,
547–553. Illustrates how a psychology-based explanation
can cast light upon a current medical controversy.
Chapman, G. B., & Sonnenberg, F. A. (2000). Decision mak-
ing in health care: Theory, psychology, and applications.
Cambridge, England: Cambridge University Press. Contains
a number of excellent chapters on the psychology of medi-
cal decision making.
Gigerenzer, G., & Gray, J. A. M. (2011). Better doctors, better
patients, better decisions. Cambridge, MA: MIT Press. Contains
important information on why both physicians and patients
are not making good decisions and how their decision
making could be improved.
Marks, M. A. Z., & Arkes, H. R. (2008). Patient and surrogate dis-
agreement in end-of-life decisions: Can surrogates accurately
predict patients’ preferences? Medical Decision Making, 28,
524–531. Another example of the analysis of an important
decision—one made at the very end of life.
Declaration of Conflicting Interests
The author declared no conflicts of interest with respect to the
authorship or the publication of this article.
Funding
Some of the research upon which this article is based
was funded by the Program in Decision, Risk, and Manage-
ment Science at the National Science Foundation (Grant SES
0326468).
References
Arkes, H. R., Faust, D., Guilmette, T. J., & Hart, K. (1988).
Eliminating the hindsight bias. Journal of Applied Psychol-
ogy, 73, 305–307. doi:10.1037/0021-9010.73.2.305
Arkes, H. R., Shaffer, V. A., & Medow, M. A. (2008). The influ-
ence of a physician’s use of a diagnostic decision aid on
the malpractice verdicts of mock jurors. Medical Decision
Making, 28, 201–208. doi:10.1177/0272989X07313280
Dawes, R. M., Faust, D., & Meehl, P. E. (1989). Clinical versus
actuarial judgment. Science, 243, 1668–1674. doi:10.1126/
science.2648573
Dawson, N. V., Arkes, H. R., Siciliano, C., Blinkhorn, R.,
Lakshmanan, M., & Petrelli, M. (1988). Hindsight bias:
An impediment to accurate probability estimation in clini-
copathologic conferences. Medical Decision Making, 8,
259–264. doi:10.1177/0272989X8800800406
Dawson, N. V., Connors, A. F., Jr., Speroff, T., Kemka, A.,
Shaw, P., & Arkes, H. R. (1993). Hemodynamic assess-
ment in the critically ill: Is physician confidence warranted?
Medical Decision Making, 13, 258–266. doi:10.1177/02729
89X9301300314
Fischhoff, B. (1975). Hindsight is not equal to foresight:
The effect of outcome knowledge on judgment under
uncertainty. Journal of Experimental Psychology: Human
Perception and Performance, 1, 288–299. doi:10.1037/0096-
1523.1.3.288
Garcia-Retamero, R., & Galesic, M. (2010). How to reduce
the effect of framing on messages about health. Journal
of General Internal Medicine, 25, 1323–1329. doi:10.1007/
s11606-010-1484-9
Gigerenzer, G. (2002). Insight. In Calculated risks (pp. 39–54)
New York, NY: Simon & Schuster.
Hertwig, R., Barron, G., Weber, E. U., & Erev, I. (2004).
Decisions from experience and the effect of rare events in
risky choice. Psychological Science, 15, 534–539.
Koriat, A., Lichtenstein, S., & Fischhoff, B. (1980). Reasons for
confidence. Journal of Experimental Psychology: Human
Learning and Memory, 6, 107–118. doi:10.1037/0278-
7393.6.2.107
Lord, C. G., Lepper, M. R., & Preston, E. (1984). Considering the
opposite: A corrective strategy for social judgment. Journal
360 Arkes
of Personality and Social Psychology, 47, 1231–1243.
doi:10.1037/0022-3514.47.6.1231
Slovic, P., & Fischhoff, B. (1977). On the psychology of experi-
mental surprises. Journal of Experimental Psychology:
Human Perception and Performance, 3, 544–551.
Tait, A. R., Voepel-Lewis, T., Zikmund-Fisher, B. J., & Fagerlin,
A. (2010). The effect of format on parental understanding
of the risks and benefits of clinical research: A compari-
son between text, tables, and graphs. Journal of Health
Communication, 15, 487–501. doi:10.1080/10810730.2010
.492560
Wegwarth, O., & Gigerenzer, G. (2011). Statistical illiteracy in
doctors. In G. Gigerenzer & J. A. M. Gray (Eds.), Better
doctors, better patients, better decisions: Envisioning health
care 2020 (pp. 137–151). Cambridge, MA: MIT Press.
Wood, G. (1978). The knew-it-all-along effect. Journal
of Experimental Psychology: Human Perception and
Performance, 4, 345–353. doi:10.1037/0096-1523.4.2.345
Yang, H., & Thompson, C. (2010). Nurses’ risk assessment
judgements: A confidence calibration study. Journal of
Advanced Nursing, 66, 2751–2760. doi:10.1111/j.1365-
2648.2010.05437.x
