Organizational Behavior and Human Decision Processes 103 (2007) 197–213

www.elsevier.com/locate/obhdp

Overconfidence and underconfidence: When and why people

underestimate (and overestimate) the competition q
a,* b,1
Don A. Moore , Daylian M. Cain
a
Carnegie Mellon University, 5000 Forbes Avenue, Pittsburgh, PA 15213, USA
b
Harvard University, 1805 Cambridge Street, Cambridge, MA 02138, USA

Received 27 September 2005

Available online 3 November 2006

Abstract

It is commonly held that people believe themselves to be better than others, especially for outcomes under their control. However,
such overconfidence is not universal. This paper presents evidence showing that people believe that they are below average on skill-
based tasks that are difficult. A simple Bayesian explanation can account for these effects and for their robustness: On skill-based
tasks, people generally have better information about themselves than about others, so their beliefs about others’ performances tend
to be more regressive (thus less extreme) than their beliefs about their own performances. This explanation is tested in two exper-
iments that examine these effects’ robustness to experience, feedback, and market forces. The discussion explores the implications for
strategic planning in general and entrepreneurial entry in particular.
 2006 Elsevier Inc. All rights reserved.

Keywords: Entrepreneurial entry; Overconfidence; Controllability; Skill; Competence; Entrepreneurship; Better-than-average; Reference group
neglect; Egocentrism; Differential regression; Comparative judgment

One of the most popular social psychology textbooks their ability to get along with others, and their chances
states, ‘‘For nearly any subjective and socially desirable of obtaining jobs that they like (College Board, 1976–
dimension. . .most people see themselves as better than 1977; Svenson, 1981; Weinstein, 1980). Some have
average’’ (Myers, 1998, p. 440). For example, people argued that the most important business decisions,
report themselves to be above average in driving ability, including the decision to found a new firm, enter an
existing market, or introduce a new product are routine-
ly biased by such overconfidence (Cooper, Woo, & Dun-
q
The authors appreciate the insightful comments, on earlier versions kelberg, 1988; Dunning, Heath, & Suls, 2004; Hayward
on this manuscript, by Linda Babcock, J. Nicolas Barbic, Max
& Hambrick, 1997; Malmendier & Tate, 2005; Odean,
Bazerman, Jason Dana, Paul Geroski, P.J. Healy, Chip Heath, Erik
Hoelzl, George Loewenstein, Daniel Lovallo, Rob Lowe, John Oesch, 1998; Zajac & Bazerman, 1991).
John Patty, Vahe Poladian, Jesper Sorensen, Lise Vesterlund, and Recent evidence, however, has cast doubt on the gen-
Roberto Weber. Thanks to Sapna Shah and Sam Swift for help with erality of overconfidence. There are a number of differ-
data collection. The authors also appreciate the support of National ent domains in which people are systematically
Science Foundation Grant SES-0451736, a Berkman Faculty Devel-
underconfident. For example, people believe that they
opment Grant at Carnegie Mellon, and the assistance of John Duffy in
the use of the Pittsburgh Experimental Economics Laboratory at the are below average in unicycle riding, computer program-
University of Pittsburgh for collecting the experimental data. ming, and their chances of living past 100 (Kruger, 1999;
*
Corresponding author. Fax: +1 412 268 7345. Kruger & Burrus, 2004). It turns out that people tend to
E-mail addresses: [email protected], [email protected]. predict that they will be better than others on easy tasks
edu (D.A. Moore), [email protected] (D.M. Cain).
1 where absolute performance is high, but worse than
Fax: +1 617 495 7730.

0749-5978/$ - see front matter  2006 Elsevier Inc. All rights reserved.
doi:10.1016/j.obhdp.2006.09.002
198 D.A. Moore, D.M. Cain / Organizational Behavior and Human Decision Processes 103 (2007) 197–213

others on difficult tasks where absolute performance is It is simple to extend this logic to a multi-item test: If
low (Hoelzl & Rustichini, 2005; Moore & Kim, 2003; one begins with the assumption that one is just as likely
Windschitl, Kruger, & Simms, 2003). A number of as others to get any given item correct, after having tak-
researchers have explained this effect as egocentrism: en the test, one should estimate that others tend to score
People focus on their own performances and neglect somewhere between one’s own score and one’s prior
consideration of others’ (Camerer & Lovallo, 1999; expectation. For example, suppose you initially expected
Kruger, 1999). In this paper, we present a new explana- everyone to score about 70%, but you think you scored
tion for these better-than-average (BTA) and worse- about 90%. Depending on how indicative you feel your
than-average (WTA) effects.2 Our explanation holds score is of others’ scores, you might predict others to
that BTA and WTA effects are a natural consequence score, say, 80%. If you scored 50%, you might predict
of regressive estimates of others, which result from the others to score, say, 60%. Notice that this perspective
fact that people have better information about them- does not imply a belief in differences of overall ability
selves than they do about others. We test this explana- between you and others—across both tests you would
tion using two experiments that examine the predict the same average score for everyone, namely
robustness of BTA and WTA effects to experience, feed- 70%. But, on each test, you would be right to expect dif-
back, and market forces. The results are consistent with ferences between you and others, given better informa-
our hypotheses, and have some provocative tion about your own score on that test. For a more
implications. formal development of this differential regression theo-
For the sake of exposition, let us introduce our ry, see Appendix A.
theory by considering beliefs about performance on a Naturally, if the task includes no skill component
one-question test where the answer is either right or whatsoever and performance is yet to be determined
wrong. Before having seen the problem, and without entirely by chance factors or factors outside one’s con-
any information regarding its ease or difficulty, how trol, then there would be little reason for people, on
likely are you to solve it? One assumption might be that average, to predict that they would be above or below
performance will be uniformly distributed across possi- average. Consistent with this reasoning, a number of
ble outcomes (Fischhoff & Bruine De Bruin, 1999; Fox researchers studying BTA effects have found that they
& Rottenstreich, 2003), leaving a 50% chance that you tend to be stronger on controllable tasks than on uncon-
will solve the problem. Such an ‘‘ignorance prior’’ might trollable tasks (for a review, see Harris, 1996). For
make sense in the absence of better information. What- instance, Camerer and Lovallo (1999) found that poten-
ever it is, this prior is simply your baseline expectation tial market entrants were excessively confident about
for your performance. winning when competition was based on their skill but
After taking the test, let us say that you know not when winners were selected randomly. The authors
whether you solved the problem. What are you to used this evidence to claim that high rates of entrepre-
believe about others’ performances? If your own perfor- neurial entry might be attributable to entrepreneurial
mance is useless for predicting others’ (e.g., if you think overconfidence. However, because prior studies have
that your good performance was based entirely on luck), employed easy tasks, the conclusion that people believe
your estimation of others’ performances ought to they are better than average on all skill-based tasks
remain unchanged from your prior beliefs. Therefore, is unwarranted. Instead, our theory would predict
doing well should leave you thinking that you did better WTA effects when the task is more difficult than expect-
than others; and doing poorly should leave you thinking ed. We test this prediction in our first experiment. The
that you did worse than others. Even if your beliefs first experiment also tests our theory that BTA and
about your own performance are helpful for predicting WTA effects are attributable to the differential regres-
others’, so long as there remains uncertainty about oth- siveness in estimates of self vs. others. Experiment 2
ers’ performances, your predictions of them should addresses some shortcomings of Experiment 1 and
depend on—and thus regress towards—the ignorance provides further support for our theory that better infor-
prior. The upshot is that, when your absolute perfor- mation about self than others produces differential
mance is better (or worse) than your prior expectations, regressiveness.
sensible Bayesian inference will lead you to make predic-
tions of others’ performances that are between these pri-
ors and your current beliefs about your performance. Experiment 1: The market entry game

2
Our design builds on that of Camerer and Lovallo
We use the terms better- and worse-than-average to be consistent (1999). They devised an N-player coordination game in
with prior work. We acknowledge that with skewed distributions, it is
indeed possible for the majority of people to be above (or below)
which, in each round, N players decide simultaneously
average. This concern, while valid, does not represent a problem for and without communication whether to enter a market
the results of the experiments we present. or not. Each market had a pre-announced capacity, c,
 D.A. Moore, D.M. Cain / Organizational Behavior and Human Decision Processes 103 (2007) 197–213 199

which determined how many entrants earned money: reference group neglect, as well as other plausible alter-
Entrants ranked below c lost money, entrants ranked c native explanations, and show that the differential
or above earned money, while non-entrants neither regression hypothesized by our theory can better
earned nor lost any money. Each entrant’s payoffs account for our results.
depended on his or her rank within the market, such The design of Experiment 1 includes several features
that more money was earned by better performance rel- that should help people avoid the mistake of ignoring or
ative to other entrants. neglecting the competition: First, competitors are phys-
Camerer and Lovallo’s key contribution over prior ically present, salient, and individuated. Second, partic-
market-entry experiments was manipulating whether ipants engage in a series of competitions over several
rankings were determined by either (a) a chance device, rounds with full feedback each round, giving them the
or (b) the entrant’s skill (in answering trivia questions). opportunity to learn.
This manipulation was implemented within-subject, so
the same participants saw several rounds in which Method
entrants were ranked based on a skill-based task and
several rounds in which entrants were ranked randomly. In each round of our experiment, all seven partici-
They found that skill-dependent payoffs encouraged pants in each experimental session were ranked relative
overconfidence and excess entry. Furthermore, excess to each other, according to a pre-announced method.
entry was highest in sessions for which it was common Before the rankings were made public, we asked partici-
knowledge that all participants were trivia enthusiasts, pants whether or not they wanted to enter into a compe-
suggesting that participants were neglecting consider- tition in which only the three top-ranked entrants would
ation of the reference group (similar enthusiasts) with make money. After they decided whether to enter, partic-
which they would be competing. ipants answered a number of questions regarding their
However, our theory predicts underconfidence and own performances and the performances of others.
insufficient entry as well. To test this, the new feature Finally, participants received full feedback regarding
of our design is that skill-dependant payoffs are based absolute performances (of self and others), how many
on either an easy or a difficult trivia game. Contrary to participants chose to enter each round, and the relative
the notion that overconfidence tends to be pervasive rankings of all (anonymously identified) entrants. The
on all skill-based competitions, we predict that partici- entire process was repeated over 12 rounds, with the
pants will only believe they are better than others on three ranking methodologies (scores on a simple trivia
simple tasks, and thus, we expect excess entry only there. quiz, scores on a difficult trivia quiz, or randomly gener-
We also test Camerer and Lovallo’s explanation: that ated scores) manipulated within session between rounds.
people focus on themselves and simply neglect consider- There were 13 experimental sessions, each with 7 peo-
ation of others (rather than miscalculating others’ per- ple for a total of 91 individual participants. Participants
formance) when making comparative judgments. were students at Carnegie Mellon University. Each par-
Camerer and Lovallo called this ‘‘reference group ticipant was endowed with $10. In each of the 12 rounds,
neglect’’ and others have simply called it egocentrism participants decided whether to enter the market or
(Chambers & Windschitl, 2004; Kruger, 1999). For whether to stay out and risk nothing. Entering the mar-
example, as examinations become more difficult, stu- ket meant either a loss or a gain, based on the entrant’s
dents become more pessimistic about their final grades, rankings within that market. These payoffs are shown in
even when it is common knowledge that the test will be Table 1.
graded on a forced curve (Windschitl et al., 2003). While
our differential regression explanation would predict the
Table 1
same effect, the reference group neglect explanation Payoffs as a function of number of entrants and market rank
holds that such false pessimism arises because students
Rank Payoff Cumulative Cumulative Cumulative
neglect to consider the fact that other students are also entrants payoff expected payoff
likely to find the test difficult. In other words, students per entrant
trying to estimate their curved grades put too much (assuming ignorance
weight on their own absolute performances. The differ- about rankings)
ential regression explanation, on the other hand, 1st $14 1 $14 $14
hypothesizes that, regardless of the weighting attached 2nd $10 2 $24 $12
3rd $5 3 $29 $9.67
to it, estimates of others will be more regressive than
4th $10 4 $19 $3.50
estimates of self. In summary, reference group neglect 5th $10 5 $9 $1.80
is about errors in the weight one puts on estimates of 6th $10 6 $1 $0.17
others’ performance, while differential regression is 7th $10 7 $11 $1.57
about errors in the estimate that are weighed. We The table shows how much money was paid out in total (column 4)
will measure the differential weighting hypothesized by and per entrant (column 5).
200 D.A. Moore, D.M. Cain / Organizational Behavior and Human Decision Processes 103 (2007) 197–213

The system by which players were ranked was 5. If you chose to enter the market this round, what
announced publicly at the beginning of each round. In rank do you think you will get?
four of the 12 rounds, rankings were determined ran-
domly. After they decided whether to enter, participants At the end of each round, participants received full
were assigned a randomly generated score from the set feedback regarding each of the seven players’ individual
of real numbers between 0 and 5, inclusive. In the scores, entry decisions, and rankings. In the eight skill-
remaining eight (skill-based) rounds, rankings were rank rounds, these scores were their trivia quiz perfor-
based on trivia quizzes taken at the beginning of the mances; in the four random-rank rounds, these were
round. Quizzes had five questions and a sixth tiebreaker their randomly generated scores. This information was
question. Four of these eight rounds’ quizzes were sim- posted using anonymous participant numbers on a
ple (with a mean score of 4.58 out of 5) and four were blackboard in view of all participants. Each participant
difficult (with a mean score of .41 out of 5). The tie- knew his or her own number, but did not know how the
breaker questions were scored based on the answer’s dis- other numbers corresponded to those individuals pres-
tance from the correct numerical answer. The presence ent. All of the 12 rounds’ results were left up for the
of this tiebreaker question virtually eliminated the entire experimental session. Experiment 1 did not mea-
chance of a tied score (there were none). The four simple sure prior expectations regarding difficulty, but Experi-
and four difficult quizzes appear in Appendix B. ment 2 did.
In order to rule out idiosyncratic effects of order, we At the end of the 12 rounds, three rounds were
varied the sequence in which participants encountered chosen at random to determine payoffs. The payoffs
the three different ranking systems as follows. The differ- for these three rounds were averaged, and this amount
ent ranking systems (R = random, S = simple, and was added to (or subtracted from) participants’ $10
D = difficult) were arranged in three different orders endowment. Thus, the maximum possible payoff was
which varied across experimental sessions: RSD, DRS, $24 for a participant who entered and was ranked first
and SDR. Whatever sequence was arbitrarily chosen in each of the three payoff rounds ($10 endowment plus
for the session was repeated four times, making 12 an average of $14 in total over the three selected payoff
rounds in four three-round blocks. So, for example, if rounds). It was also possible for a participant to leave
the first three rounds used the sequence RSD, all partic- the experiment empty-handed if he entered and was
ipants in that session faced the same quizzes at the same ranked 4th or below on each of the three payoff rounds
time, with ranking systems in the order: RSD–RSD– ($10 endowment minus an average loss of $10 in total).
RSD–RSD. The order in which participants encountered Across all participants, the mean final payoff was $13.01
the four different simple and difficult trivia quizzes was (with a range of $4 to $24).
also counterbalanced between experimental sessions.
In the eight skill-rank rounds, after taking the given Equilibrium predictions
quiz but before seeing their scores, all participants As Table 1 (column 5) shows, entry has a positive
simultaneously made their entry decisions (to enter or expected value so long as five or fewer players enter
stay out). In the four random-rank rounds, there were the market, assuming players have no information
no quizzes, and all participants merely made their entry about their own relative ranks. If players are risk-neu-
decisions prior to learning their ranks. In all 12 rounds, tral, then it is rational (i.e., there is a set of pure-strategy
after participants made their entry decisions they then Nash equilibria) for five players to enter each round.
answered the following questions: Lacking some coordinating device for deciding which
of each session’s seven total players enter and which stay
1. How many people total do you think will enter the out, there is a rational strategy (i.e., a mixed-strategy
market this round? Include yourself in this figure if equilibrium) that is somewhat more complicated to
you chose to enter. compute, but the result is that all players enter with a
2. What percentage of the other six entrepreneurs in this probability of 84%. Naturally, since only the top three
round do you think will score lower than you will ranks actually win money, if all players know what their
(regardless of whether anyone enters)? ranks will be, then only the top three players (3/7 or 43%
3. How many questions (out of 5) do you think you got of the potential entrants) will enter. Therefore, if all
correct this quiz? In random-rank rounds, this ques- players were unbiased and imperfectly informed, we
tion was replaced with the question: What score should expect between 43% and 84% of participants to
(out of 5) do you think you will get this round? enter each round.
4. How many questions (out of 5) do you think the aver- Predicting the equilibrium outcome without the
age participant will get correct this round? In assumption of risk neutrality is more difficult. Even
random-rank rounds, this question was replaced with without information on their rankings, if everyone was
the question: What score (out of 5) do you think the sufficiently risk averse, no one would enter, and if every-
average participant will get this round? one was sufficiently risk seeking, everyone would enter.
 D.A. Moore, D.M. Cain / Organizational Behavior and Human Decision Processes 103 (2007) 197–213 201

So, following Camerer and Lovallo (1999), we use the 7

Difficult
random-rank condition (when players cannot possibly
6 Random
have useful information on their rankings) to provide
Simple
an empirical estimate of behavior given players’ risk 5

Entrants per round

preferences. Deviations from entering 84% of the time
in random-rank conditions suggest particular deviations 4

from risk neutrality. And since all participants see all

3
conditions, their entry decisions in the different condi-
tions serve as within-subject controls for risk preferenc- 2
es. The difference in entry rates between the different
conditions (random, simple, and difficult) is the depen- 1

dent measure of primary interest.

0
1 2 3 4
Hypotheses Block
Consistent with the differential regression explana- Fig. 1. Entry rates in the three different ranking systems across the
tion, we predict that participants will believe themselves four blocks. Error bars show standard errors.
to be above average (and above median) on simple tests
but below average (and below median) on difficult tests.
2.9, 4.2, and 5.7, respectively, in the last block. Fig. 1
As a result, we predict that participants will enter too
shows these means.
frequently on simple-rank rounds and too rarely on
difficult-rank rounds. We will take entry rates into
Explaining differences in rates of entry
random-rank rounds as indicators of participants’
There are four possible explanations for the systemat-
behavior given ignorance about their relative ranks
ic effect of ranking systems on rates of entry: our differ-
and given their risk preferences. We predict that entry
ential regression explanation and three alternatives. The
rates in random-rank rounds will lie between entry rates
first alternative explanation is that participants believed
in simple- and difficult-rank rounds.
that others would stay out of simple-rank rounds and so
entry would have a higher expected value in simple
Results rounds. The data contradict this explanation: Partici-
pants predicted that there would be 5.2 entrants in the
The average random-rank round saw 4.27 entrants average simple-rank round (there were 5), 4.5 entrants
(61% entry rate—suggesting slight risk aversion, on the average random-rank round (there were 4.27), and
average). In contrast to this baseline, the average sim- 3.2 entrants in the average difficult-rank round (there
ple-rank round saw 5.0 entrants and the average diffi- were 2.94). These predictions are consistent over time
cult-rank round saw 2.94 entrants. In order to test for and do not systematically get either better or worse over
the statistical significance of these differences, we con- the 12 rounds of play. So participants expected more
ducted a (4) · (3) within-subjects ANOVA in which each competition in simple rounds, but more entered there
of the 13 experimental sessions served as a single inde- anyway.
pendent case. The four three-round blocks served as The second alternative explanation is that potential
the first within-subjects factor and the three ranking sys- entrants systematically overestimated their own scores
tems served as the second within-subjects factor. The more on simple tasks than on difficult tasks. This expla-
results reveal a significant effect of the experimental con- nation is also contradicted by the data—the opposite is
dition, F (2, 24) = 39.17, p < .001, g2 = .77. Contrast actually true. Participants underestimated their scores
tests confirm the significance of both the difference on the simple quiz, reporting that they had gotten 4.41
between entry rates in difficult-and the random-rank out of 5 correct, when in fact they actually got 4.58. This
rounds (p < .001) and the difference between the simple- difference is revealed to be significantly different by a
and random-rank rounds (p = .018). comparison of actual vs. self-reported scores in a
The main effect of block is not significant, F (3, (2) · (4) within-subjects ANOVA performed at the level
36) = .82, p = .49, g2 = .06. Although the interaction of the individual, where the four blocks served as the
between block and ranking system (as shown in Fig. 1) second within-subjects factor, F (1, 90) = 19.03, p <
is marginally significant in the overall ANOVA, F (6, .001. On the difficult test, by contrast, participants over-
72) = 2.08, p = .07, g2 = .15, this does not appear to estimated their scores, reporting that they had gotten .95
result from a consistent reduction in the effect of ranking correct when in fact they had only gotten an average of
system as participants gained experience: Entry rates in .41 correct, F (1, 90) = 78.20, p < .001.
the difficult, random, and simple markets were 2.9, 4.4, While this pattern at first seems incongruous, it
and 5.1, respectively in the first block and were similarly ought not to be surprising: The tendency for people to
202 D.A. Moore, D.M. Cain / Organizational Behavior and Human Decision Processes 103 (2007) 197–213

overestimate their own performances more on difficult 70

Difficult
than on simple tasks is one of the more robust findings 65 Random
in the literature on overconfidence and calibration (Bur- Simple

Estimated percentile rank

son, Larrick, & Klayman, 2006; Lichtenstein, Fischhoff, 60

& Phillips, 1982). It can be readily explained using the 55

same regressive logic that we used to predict BTA effects
on simple tasks and WTA effects on difficult tasks: 50

Because people have imperfect knowledge of their own 45

scores, their estimates of their own performances are
40
slightly regressive (Erev, Wallsten, & Budescu, 1994;
Juslin, Winman, & Olsson, 2000). If people’s estimates 35
of their own performances are slightly regressive, then
30
their estimates of the performances of others are likely
1 2 3 4
to be even more regressive. This follows from the fact
Block
that people have better information about themselves
than they do about others, and so, people underestimate Fig. 2. Participants’ estimated percentile rankings (percentage of
others more on simple tasks than on difficult tasks. others worse than them) in the three different ranking systems across
the four blocks. Error bars show standard errors.
In simple rounds, our participants underestimated
their scores (which rules out the second alternative
explanation) and they expected more competition—yet information about each quiz that might have under-
more entered there anyway. Before we test the reference mined their willingness to attend to this general histori-
group neglect explanation, let us turn to our explanation cal fact. Even if an individual notices that she has
for the observed entry rates: differential regression. The consistently overestimated her relative performance on
data are consistent with differential regression. People simple quizzes, if she takes a new quiz and scores highly
underestimated others’ scores on simple quizzes more relative to her prior expectations, the inference that she
than their own, reporting that others would score a is likely to be above average may still be a sensible one.
regressive 4.2 out of 5, but that they themselves would So long as there is more uncertainty about others’ scores
score 4.41, F (1, 90) = 13.92, p < .001. On the other than about her own, her predictions of others’ scores
hand, participants overestimated others’ scores on diffi- will remain more regressive.
cult quizzes more than their own, reporting that others
would score a regressive 1.49 out of 5, but that they Fourth explanation: reference group neglect
themselves would score .95, F (1, 90) = 41.63, p < .001. We have presented evidence for the idea that esti-
Because participants’ estimates of others are so mates of others are regressive. The remaining question
regressive, they believe themselves to be above median is whether the differential regression explanation alone
on the simple quiz and below median on the difficult can account for the observed differences in entry rates
quiz. On the simple test, participants reported that they between experimental treatments, or whether there is
expected to outscore 63% of the other participants any evidence of reference group neglect. Reference
taking the same test. On the difficult test, by contrast, group neglect posits that participants chose to enter on
participants only expected to outscore only 46% of the simple rounds and stay out on difficult rounds not
others. On the simple test, participants expected there because they actually believed that they would score
to be 5.2 entrants and expected that their rank among any differently from others—but because they just were
entrants would be 2.6. On the difficult test, participants not paying enough attention to others (Klar & Giladi,
expected only 3.2 entrants yet expected to rank 2.7. 1997; Kruger, 1999; Windschitl et al., 2003). If people
Fig. 2 shows patterns in participants’ beliefs about per- neglect to consider the group and instead focus on
centile rankings across the three treatments and four themselves when estimating their relative standing, we
blocks. Differences between simple and difficult treat- should observe that beliefs about own performance are
ments persist throughout the experiment, despite the weighted more heavily than are beliefs about others.
provision of feedback. There is little evidence for learn- In what follows, we test this prediction in a pair of
ing in this figure. regression analyses (in Table 2) predicting comparative
These results provide a hint as to the reasons for the judgments using performance by self and others. As
durability of differences in entry rates across different the remainder of the results section details, the substan-
ranking systems. Participants got consistent feedback tial majority of the effect of difficulty can be accounted
showing that they tended to underestimate their relative for by greater regressiveness in estimates of others than
performances on the difficult quizzes and that they tend- of self.
ed to overestimate their relative performances on the Model 1 in Table 2 is the optimal model, predicting
simple quizzes; but they nevertheless had specific new participants’ actual percentile ranks within each round
 D.A. Moore, D.M. Cain / Organizational Behavior and Human Decision Processes 103 (2007) 197–213 203

Table 2
Actual and perceived value of participants’ own scores and the scores of others for estimating percentile rank
Model 1 predicting actual percentile rank Model 2 predicting self-reported percentile rank
Independent variable B SE Independent variable B SE
Own actual score .201* .005 Own estimated score .111* .006
Actual average score .201* .006 Estimated average score .077 * .008
R2 .56* R2 .27*
*
p < .001.

using their own scores and actual average scores for the statistical test.3 The Nagelkerke R2 value of this logistic
round as independent variables. The results show, not regression reveals that the experimental treatment
surprisingly, that participants’ own actual scores and accounts for 7.9% of the variation in individual entry
average scores have B coefficients that are of similar size decisions, and is statistically significant, v2 (1) = 65.67,
but opposite signs. We compare this result with partici- p < .001. What this means is that 7.9% is the total size
pants’ self-reported beliefs, using participants’ beliefs of the effect of difficulty on entry, and we must now
about their own scores and beliefs about the average determine how much of it can be accounted for by dif-
score to predict their self-reported percentile rank. The ferential regression and how much of it cannot be.
first apparent difference between these two analyses is In order to assess the effect of differential regressive-
more noise in self-reported beliefs than in actual perfor- ness, we next regressed entry decisions on participants’
mance, as shown by the smaller value of R2. It ought not beliefs about their relative performance, as measured
to be a shock that people’s estimates of their own scores by the difference between their estimated absolute scores
and their percentile rankings are imperfect. for self and for others. Beliefs about relative performance
The second and more important difference is that par- account for 24.4% of the variation in entry rates,
ticipants’ beliefs about their own scores were weighted v2 (1) = 218.96, p < .0001. The mere fact that partici-
more heavily than were their beliefs about others’ scores. pants’ beliefs about relative performance are predictive
To be precise, the weight attached to other (jBj = .077) of their entry decisions is neither impressive nor interest-
is 69% the size of the weight attached to self ing—it would be surprising if they were not. The interest-
(B = .111), and this difference is statistically significant, ing question is whether these beliefs about relative
t (1088) = 3.4, p < .001. This shows evidence of refer- performance can account for the effect of the experimen-
ence group neglect but raises the following question: tal manipulations on entry decisions. In order to test for
What proportion of our primary result (the effect of dif- such a mediation effect, we conducted a third regression
ficulty on entry rates) can be accounted for by differen- that included both experimental treatment and partici-
tial regression and how much can be accounted for by pants’ self-reports of relative performance. The resulting
reference group neglect? In order to answer this ques- Nagelkerke R2 value indicates that these two variables
tion, we first begin by assessing the experimental treat- combined account for 26.0% of the variation in entry
ment’s effect on entry decisions. We did this by decisions. The inclusion of experimental treatment pro-
regressing entry rates on experimental treatment. The vides only a 6.7% increase in variation explained (over
independent variable in this regression was equal to 1 the 24.4% using only the relative performance). However
for simple-rank rounds, 0 for random-rank rounds, the B coefficient associated with experimental condition
and -1 for difficult-rank rounds. When we conduct this remains significant (B = .34, SE = .09, p < .001). The
analysis at the level of the round, the R2 value of this significance of experimental treatment suggests that
regression shows that the experimental treatment there is an effect of difficulty that is distinct from partic-
accounts for 28% of the variation in entry rates across ipants’ beliefs about their relative standing. Of the total
all rounds, F (1, 154) = 59.67, p < .001. However, more 7.9% of variation in entry decisions accounted for by
useful to our purposes is this analysis performed at the our experimental treatment, 1.6% (or 26%  24.4%) can-
level of the individual. There are two major reasons to not be accounted for by participants’ self-reported beliefs
expect R2 to be lower in the regression conducted at
the individual level: First, participants’ entry decisions
are partially driven by their actual relative performance, 3
For the sake of simplicity, we present logistic regression analyses in
which is uncorrelated with the experimental treatment; which each subject in each round serves as the unit of analysis (91
second, idiosyncratic individual-level factors such as risk subjects · 12 rounds = 1092 observations). The results we present are
not appreciably different when the same analyses are conducted using a
preferences affect entry decisions. At the individual level, hierarchical linear model which treats subjects as random effects and
since the dependent variable is dichotomous (entry accounts for the fact that experimental treatments are nested within
or not), a logistic regression is the more appropriate trial blocks which are in turn nested within experimental sessions.
204 D.A. Moore, D.M. Cain / Organizational Behavior and Human Decision Processes 103 (2007) 197–213

about their own performance relative to others. This BTA effects (Alicke, Klotz, Breitenbecher, Yurak, &
1.6% represents 20% of the variation due to the experi- Vredenburg, 1995). Here, we test whether it can also
mental treatment that remains unexplained. Reference reduce WTA effects.
group neglect is the most viable alternative explanation
for this unexplained 20%, but the substantial majority Methods
of the effect of difficulty (80%) can be accounted for by
greater regressiveness in estimates of others than of self. Participants
We recruited 128 undergraduate students at Carnegie
Discussion Mellon University by offering them a base payment of
$2 plus from $0 to $8 on top of that. Experimental
The results of the first experiment show that confi- sessions were advertised under the name ‘‘Games of
dence regarding one’s competitive performance depends skill’’ with the following description: ‘‘Participants will
on the type of competition. Contrary to prior evidence be playing a game in which they can earn money. How
(Camerer & Lovallo, 1999; Harris, 1996; Klein & much you get paid will depend on exactly how things come
Kunda, 1994), we show that controllable tasks do not out.’’
necessarily elicit more overconfidence than chance tasks.
In difficult-rank rounds, people avoided entering. People Design
overestimated others’ performances, leading them to The experiment had a 2 (quiz difficulty: simple vs. dif-
stay out of the competition despite the fact that they ficult) · (3) (time of wager: before quiz vs. after quiz vs.
accurately forecast few other entrants. Thus, skill-based after results) mixed design. Quiz difficulty was manipu-
tasks do not always elicit overconfidence and entry rates lated between subjects and time served as a within-
depend in part on how difficult potential entrants see subjects factor.
the task.
Participants’ prior expectations regarding difficulty Procedure
play an important role in our theory, but the first exper- Participants were each given $4 and invited to bet as
iment did not measure them. Experiment 2 was designed much as they wanted on winning a trivia competition
to address this shortcoming. Furthermore, our theory against a randomly chosen opponent. Participants
posits a fundamental role for information about perfor- were truthfully told that their opponents’ scores would
mance—one’s own and others’. Experiment 2 allows us be selected at random from a group of 144 students
to observe the effect of information on participants’ who had previously taken these same quizzes as a part
beliefs as they learn first about their own performances of a different study (reported in Moore & Kim, 2003,
and then about the performances of others. Experiment 3). None of the 128 participants in the
present study had participated in that prior study.
The test would consist of 10 items plus an 11th tie-
Experiment 2: The effect of information on comparisons breaker question that virtually eliminated the chance
of a tied score. Winning participants would double
Because our differential regression explanation the amount they bet; those who lost would keep
describes the mechanisms by which errors in entry only the un-wagered portion of their $4. Note that
occur, it also offers useful insights into which interven- the second and third time they bet, participants were
tions might be useful for reducing errors and which told that the most recent bet would be the one that
interventions are unlikely to be effective. Experiment 2 counted.
tests these interventions. Participants were first told that Participants in the simple quiz condition were told
they would be taking either a difficult or simple quiz and they would take a simple trivia quiz and be shown the
were then asked to predict the outcome (Time 1). After following example question and answer:
taking the quiz (Time 2), participants were invited to What is the common name for the star inside our own
revise their answers to their prior estimates of absolute solar system?
and relative scores. Our theory would predict that infor- Answer: The Sun
mation about one’s own performance provided at Time Participants in the difficult quiz condition were told
2 would produce BTA on easy tasks and WTA on diffi- they would be taking a difficult trivia quiz and shown
cult tasks. Finally, participants were given full informa- the following example question and answer:
tion about how others scored on the same quiz they What is the name of the closest star outside our solar
took, and they were asked to report the same compara- system?
tive judgments (Time 3). Our theory would not predict Answer: Proxima Centauri
BTA and WTA effects at Time 3, in the presence of Participants were then asked how much they wanted
excellent information about others. Previous research to bet. After they bet, participants were given a ques-
has shown that information about others can reduce tionnaire that asked:
 D.A. Moore, D.M. Cain / Organizational Behavior and Human Decision Processes 103 (2007) 197–213 205

(1)‘‘How many of the 10 questions do you think you Results and discussion
will get right?’’
(2)‘‘How many of the 10 questions do you think your Manipulation check
opponent will get right?’’ As expected, the simple quiz resulted in higher scores
(3)‘‘What percentage of the group will have scores (M = 8.25 out of 10, SD = 2.01) than did the difficult
below yours? (If you expect your score will be the quiz (M = 1.54 out of 10, SD = 1.34), F (1, 126) =
very best, then put 100. If you expect your score will 490.39, p < .001, g2 = .80.
be exactly in the middle, put 50. If you expect your
score will be the lowest, put 0.)’’ Participants’ predictions at Time 1
At Time 1, participants who had seen only an easy
Questions 1 and 2 were objective measures of abso- sample question (and were about to take—but had not
lute evaluation for self and for opponent. Question 3, yet taken—the simple test), predicted that they would
like the bet, was a direct measure of beliefs about rel- score 7.22 (SD = 1.57) and that others would score
ative standing. After participants had answered all 6.41 (SD = 1.79) out of 10. Those who saw only a diffi-
these questions, they were given an actual trivia test. cult sample question predicted that they would score
The questions from the difficult and simple quizzes 5.22 (SD = 1.90) and that others would score 4.92
are listed in Appendix C. Participants were then told, (SD = 1.65). We analyzed these predictions using a 2
‘‘Now that you have taken the quiz, you may choose to (difficulty) · (2) (target: self vs. other) mixed ANOVA.
revise your answers to these questions. Please answer The results reveal a main effect of target, F (1, 124) =
all the questions, whether or not you put the same 11.97, p = .001, g2 = .09, since people predicted that
answers as before.’’ Then participants were asked they would do better than would others. The differential
how much they wanted to bet and were asked regression explanation cannot account for this effect; the
the same list of questions again. These were their results suggest some basic amount of self-enhancement.
Time 2 responses. The main effect of difficulty is, of course, also significant,
After they had answered all the questions at Time 2, F (1, 124) = 46.13, p < .001, g2 = .27. The target · diffi-
participants were then given truthful feedback about culty interaction effect does not attain statistical signifi-
the scores of the previous test-takers from whose ranks cance, F (1, 124) = 2.77, p = .099, g2 = .02.
their randomly selected opponent would be drawn. For
example, those who took the simple quiz were Tests of differential regression at Time 2
informed that: ‘‘The average score is 8.71 out of 10, At Time 2, the differential regression explanation
with a standard deviation of 1.1.’’ Those who took the would hypothesize that people predict better relative
difficult quiz were told: ‘‘The average score is 1.48 out performance (BTA) on simple tasks and (WTA) worse
of 10, with a standard deviation of 1.01.’’ Participants relative performance on difficult tasks. This would man-
were also given a breakdown of the percentage of oth- ifest itself in a significant interaction between difficulty
ers who got each of the 11 possible scores (from 0 to (simple vs. difficult) and target (self vs. other). Indeed,
10) on the quiz. After they had a chance to review this when we subjected estimates of absolute performance
information, participants were told, ‘‘Now that you to this 2 · (2) ANOVA, the difficulty · target interaction
have seen how others did, you may choose to revise your emerges as significant, F (1, 120) = 20.77, p < .001,
answers to these questions. Please answer all the ques- g2 = .15.4 At Time 2, participants reported believing
tions, whether or not you put the same answers as that they scored better (M = 8.30, SD = 1.49) than their
before.’’ Then participants were asked how much they opponents (M = 7.83, SD = 1.28) on the simple quiz,
wanted to bet and were asked the same list of three t (61) = 2.94, p = .005, g2 = .12. But they also reported
questions again. These were their Time 3 responses. believing that they scored worse (M = 2.39, SD = 1.31)
The bet that was counted for computing payoffs was than their opponents (M = 3.30, SD = 1.61) on the dif-
this third and final one. ficult quiz, t (59) = 3.48, p = .001, g2 = .17.
Our differential regression explanation holds that Consistent with our theory, the increase in BTA and
BTA effects and WTA effects result when people have WTA effects from Time 1 to Time 2 is largely attribut-
good information about themselves but lack informa- able to changes in beliefs about one’s own score. On
tion about others, such as at Time 2 after taking the average, participants changed their estimates of their
quiz. At Time 3, after getting good information about own scores by 2.34 points (SD = 1.77). However, they
others, these effects should go away. Time 1 beliefs are
useful for assessing participants’ priors, but are based 4
on so little information that our theory would not make Naturally, the main effect of difficulty is also significant, since
participants predict higher scores on the simple than on the difficult
strong predictions regarding their beliefs. We shall test test, F (1, 120) = 606.16, p < .001, g2 = .84. The main within-subjects
both our differential regression explanation and that of effect of target (self vs. opponent) is not significant, F (1, 120) = 1.40,
reference group neglect. p = .24, g2 = .01.
206 D.A. Moore, D.M. Cain / Organizational Behavior and Human Decision Processes 103 (2007) 197–213

70 different from one another t (123) = 4.57, p < .001.

Difficult And consistent with our theory, the reduction in BTA
65
Simple and WTA effects on bets is mediated by changes in peo-
60 ple’s beliefs about others, b = .31, p = .003, not the
Percentile rank

55
self, b = .16, p = .07.

50 Tests of reference group neglect

45 The reference group neglect hypothesis predicts that
direct comparisons (like estimates of percentile rank)
40 will show stronger BTA and WTA effects than will indi-
35 rect comparisons (computed by subtracting absolute
estimates of others from self) which make others’ perfor-
30
mances salient. The standard test is to regress compara-
Time 1 Time 2 Time 3
tive judgment on absolute evaluations of target and
Fig. 3. Participants’ estimated percentile rankings (percentage of referent. Using this standard test, we replicate the result
others worse than them) in the two difficulty conditions at three that comparative evaluation is strongly associated with
points in time. Error bars show standard errors. self-evaluation but more weakly predicted by absolute
evaluation of others. We regressed percentile estimates
only changed their estimates of others’ scores by 1.92 on predictions of point scores by self and other for
points (SD = 1.76), and this difference is significant by responses at Time 1, before participants had taken the
paired-samples t-test, t (123) = 2.34, p = .02. Further- actual test. As Table 3 shows, the b coefficient for abso-
more, these changes mediate the difference on bets lute self-evaluation is .86, p < .001, indicating that abso-
between difficulty conditions from Time 1 to Time 2. lute and relative self-assessment are strongly correlated.
We included these two change measures in a regression The b coefficient for other-evaluation, however, is .53,
predicting change in participants’ bets from Time 1 to p < .001, is only 62% the magnitude of the coefficient for
Time 2, along with a dummy variable for difficulty. self. This finding is consistent with reference group
Their inclusion renders difficulty non-significant, neglect.
b = .02, p = .87. As our theory would predict, changes Note that this differential weighting changes as people
in self-estimates were a significant predictor of changes gain information. At Time 2, when participants had more
in bets, b = .47, p < .001. However, changes in other-es- information about themselves, the weight put on other-es-
timates were not significant, b = .13, p = .16. timates (.49) is only 53% the size of the weight put on
self-estimates. But at Time 3, when people had better
Tests of differential regression at Time 3 information about others, other-estimates (b = 1.69) car-
The differential regression explanation would not pre- ry 92% the weight placed on self-estimates (b = 1.84). If
dict BTA and WTA effects at Time 3, when participants reference group neglect affects how people bet, then we
have good information not only about themselves but ought to observe some effect of test difficulty on bets, over
about others. Indeed, the same 2 · (2) ANOVA on abso- and above the effect of differential regressiveness on esti-
lute evaluations at Time 3 does not produce a significant mates of self and others’ actual performances. We tested
target · difficulty interaction, F (1, 124) = .02, p = .89.5 this as we did in Experiment 1.6 The result was that 74%
On the simple quiz, participants predicted similar scores of the effect of difficulty on bets at Time 2 could be
for themselves (M = 8.31, SD = 1.62) and for their explained by differential regression. However, this test
opponents (M = 8.14, SD = 1.37). Likewise on the diffi- may claim too much credit for differential regression. As
cult quiz, participants predicted similar scores for them- the results in Table 3 highlight, better information about
selves (M = 2.08, SD = 1.19) and for their opponents self than others appears to produce both differential
(M = 1.90, SD = .71). Fig. 3 shows participants’ self-re- regression and differential weighting. When they are con-
ported percentile ranks. Furthermore, as our theory pre- founded, this test will give all the credit to differential
dicts estimates of others’ scores changed more from regression over differential weighting.
Time 2 to Time 3 than did estimates of self. Estimates At Time 3, the effect of test difficulty on bets shrinks
of others changed by an average of 1 point dramatically: Difficulty accounts for only 3.4% of the
(SD = 1.18), whereas estimates of self only changed by
.4 points (SD = 1.05), and these two are significantly 6
First, we began with the primary effect of test difficulty on bets. At
Time 1, test difficulty only accounted for a statistically insignificant
1.9% of the variance in bets, as shown by the R2 value associated with a
5
Naturally, the main effect of difficulty remains significant, F (1, regression predicting bets using a dummy variable for experimental
124) = 931.98, p < .001, g2 = .88. The main effect of target remains condition. At Time 2, however, difficulty accounts for 15% of the
insignificant, F (1, 124) = 2.42, p = .07, g2 = .03. variance in bets.
 D.A. Moore, D.M. Cain / Organizational Behavior and Human Decision Processes 103 (2007) 197–213 207

Table 3
Experiment 2’s results for the three different measures of comparative judgment at three points in time.
Time Self-reported comparative judgment Simple vs. Difficult Regression results
2
Effect size (g ) b (Self) b (Other)
1 Bet .02 .53*** .27*
1 Percentile rank .04* .86*** .53***
1 Indirect comparison .03 1.07  1.0 
2 Bet .15*** .92*** .48**
2 Percentile rank .16*** .93*** .49***
2 Indirect comparison .15*** 1.86  1.53 
3 Bet .03* 1.67*** 1.35***
3 Percentile rank <.001 1.84*** 1.69***
3 Indirect comparison .00 3.10  3.01 
The third column shows the effect size of the difference between simple and difficult conditions, and asterisks show the significance of the t-test
comparing difficulty conditions. Regression results predicting indirect comparative judgment for the three different measures of comparative
judgment appear in the fourth and fifth columns.
*
p < .05.
**
p < .01.
***
p < .001.
 
Independent variables perfectly account for dependent variable.

variance in bets. Our theory would not predict that test downward and a majority will conclude that they are bet-
difficulty would affect comparative judgments when ter than others. When a task is more difficult than expect-
people possess complete information regarding perfor- ed, estimates of others’ performances will regress upward
mance by self and others. Indeed, statistically speaking, and a majority will conclude that they are worse than oth-
differential regressiveness accounts for only 3% of this ers. Experiment 2 shows that these WTA and BTA effects
small effect. The remaining 97% is most likely attribut- are strongest when people are confident regarding their
able to the egocentric overweighting of self-knowledge own performances but unsure of the performances of oth-
over other-knowledge. ers. This may also explain why BTA and WTA effects are
stronger when people compare themselves to some vague
group than when they compare themselves to a specific,
General discussion known individual (see Hoorens & Buunk, 1993; Klar,
Medding, & Sarel, 1996; Klein & Weinstein, 1997; Perloff
The present results offer two primary findings. First, & Fetzer, 1986; Price, 2001; Windschitl et al., 2003). It
the results of the first experiment show that confidence may also help explain why BTA effects have been shown
regarding one’s competitive performance depends on to be stronger for observable performances than for tasks
the ease of the task. Contrary to prior evidence (Camer- or traits where people only know about themselves and
er & Lovallo, 1999; Harris, 1996; Klein & Kunda, 1994), cannot observe others’ performances directly (Dunning,
we show that controllable tasks do not necessarily elicit Meyerowitz, & Holzberg, 1989; Suls, Lemos, & Stewart,
more overconfidence than chance tasks. People underes- 2002). Given that differential regression explains most
timated others’ performances on simple tasks but over- of the effect of difficulty on entry and on bets in our exper-
estimated them on difficult tasks, leading them to enter iments, providing decision-makers with better informa-
with confidence on simple rounds despite the fact that tion about the performance of others is likely to be the
they accurately forecast numerous other entrants. Yet most effective way to eliminate this cause of myopic
in difficult-rank rounds, people decided not to enter: comparisons.
They overestimated others’ performances, leading them Our studies experimentally manipulated the informa-
to stay out of the competition despite the fact that they tion people had, making it impossible for us to measure
accurately forecast few other entrants. Thus, skill based differences in the degree to which people seek out infor-
tasks do not always elicit overconfidence; instead, confi- mation about themselves and others (for other studies
dence and entry rates depend in part on how difficult that have measured this, see Moore, Oesch, & Zietsma,
potential entrants see the task. in press; Radzevick & Moore, 2006). However, our
The second contribution of this paper is that we results nevertheless show that people do not always use
identify the cause for what appear to be myopic interper- information as they should. When making social compar-
sonal comparisons, namely, better information about isons, participants ought to have given the same weight to
one’s own performance than about the performance of information about themselves as to information about
others. When a task is simpler than people expect it to others. The fact that one’s competitors are doing poorly
be, people’s estimates of others’ performances regress on some task is just as important as the fact that one is
208 D.A. Moore, D.M. Cain / Organizational Behavior and Human Decision Processes 103 (2007) 197–213

doing poorly. Consistent with the reference group neglect low entry rates, at least in the short term. Yet the correla-
explanation, however, our participants’ self-reported tion between entry and exit in any given year is around .7.
beliefs regarding their own performance were weighted The results presented in this paper suggest a possible
more heavily than their beliefs regarding opponents’ per- explanation. Perhaps industries that see persistent high
formance. But this egocentrism effect is not the driving rates of entry are those that potential entrants view as
factor behind WTA and BTA effects and their conse- ‘‘easy’’ or in which they feel capable (Greico & Hogarth,
quences for behavior. Differential weighting (reference 2004). Such industries are then likely to see more intense
group neglect) accounts for a small proportion of the competition, lower profits, and higher rates of failure.
effect of difficulty on entry rates and bets. Experiment 1 shows that this might occur in spite of
We should also note that, besides reference group entrants’ correct prediction that they will have lots of
neglect, there are other viable explanations for why peo- competition in ‘‘simple’’ markets. Indeed, even when they
ple’s judgments would appear to weight self more heavily underestimate their own absolute abilities, entrants will
than others. For example, people easily conflate relative underestimate the abilities of their competition even
and absolute evaluation on vague subjective measures more.
(Baron, 1997; Biernat, Manis, & Kobrynowicz, 1997). However, given that our experimental participants
The conflation error occurs when people answer the were not actual entrepreneurs with substantial quantities
question, ‘‘How good are you relative to others?’’ as if of money at stake, we must be cautious about generaliz-
they were answering the question ‘‘How good are ing our results from errors made in the lab. Might actual
you?’’ (Klar & Giladi, 1999). Unlike differential weight- entrepreneurs learn to avoid the biases in comparative
ing due to reference group neglect, the conflation judgment shown by participants in the present experi-
explanation is exceedingly mundane: Vague measures ments? While it is possible that experience may allow
facilitate confusion between relative and absolute actual entrepreneurs to learn to overcome these errors,
evaluation (Burson & Klayman, 2005; Moore, 2005). it is unclear how much experience is needed for such
Conflation can make comparative evaluations appear learning to take place. Participants in the market entry
to overweight the self because the person does not take game were students at a selective university and also
herself to be making a comparative evaluation. experienced 12 rounds with full feedback. Perhaps this
Although we have attempted to distinguish the differ- experiment included too few trials for them to learn to
ential regression explanation from reference group solve this problem. If this is the case, however, entrepre-
neglect, we should also note their fundamental compati- neurs are likely to make the same mistake. After all,
bility. Both are caused by the greater accessibility and even the most experienced entrepreneurs rarely get the
quality of information about the self than about others. opportunity to start more than a handful of firms.
Less information about others leads people to make Furthermore, our experimental setup was more trans-
more regressive estimates of others but also probably parent, assessment of the competition was clearer, and
leads to further underweight of those regressive estimates the causes of success were more obvious than they are
(Kruger, Windschitl, Burrus, Fessel, & Chambers, 2006). likely to be for most entrepreneurs.
We must note that both our theory and our results are
Managerial and economic implications bounded. We do not claim that the differential regression
explanation (a cognitive explanation) accounts for all
Our results have implications for understanding entre- BTA and WTA effects. Motivation and bias do influence
preneurial entry. While we do not take a stand on the comparative interpersonal judgments in important ways
question of whether overall rates of actual entrepreneurial (Kunda, 1990). It is also clear that other cognitive expla-
entry are excessive, we instead note that rates of entry vary nations such as reference group neglect can account for
considerably between industries. Some industries, such as some biases in comparative judgment and in strategic
retail clothing stores, restaurants, and bars, are marked decision making, as shown in the present findings as well
by persistently high rates of entry and high rates of subse- as in other work (Klar, 2002; Moore, 2004; Rose &
quent exit (US Small Business Administration, 2003). Windschitl, 2006; Windschitl et al., 2003). However, the
Indeed, one of the stylized facts to emerge is that rates errors showcased in this paper seem to be more about
of entry and exit are highly and positively correlated not having good information about one’s competition,
(Dunne, Roberts, & Samuelson, 1988; Geroski, 1996; as opposed to merely ignoring the competition.
Mata & Portugal, 1994). Differences in rates of entry are
not well accounted for by the size of an industry, the prof-
itability of its firms, or barriers to entry (Geroski, 1996). Appendix A. Formalization of the differential regression
So, if new firms enter an industry because of above-nor- explanation
mal profits and exit an industry because of below-normal
profits, then one would instead expect entry and exit to be In this appendix, we attempt to formalize our theory,
negatively correlated such that high failure rates meant first in general terms, then with a specific example. Our
 D.A. Moore, D.M. Cain / Organizational Behavior and Human Decision Processes 103 (2007) 197–213 209

point in this appendix is not to suggest that it is possible useful for estimating performance by self than others,
to predict with certainty what individuals will believe— it will lead to greater updating of S than of O. There-
given a multitude of constraints that may or may not be fore, estimates of others will tend to be closer to the
realistic (we let our data speak to how real people prior than will estimates of self. In other words, O
behave)—nor do we intend to provide a generalized regresses to the prior more than does S. We call this
proof or justification of the decision process we are sug- rule ‘‘Rule O’’:
gesting is at work. Here, we wish only to provide insight Rule O: Since (whenever possible) O is more regres-
on why WTA/BTA effects might occur in groups of rea- sive than S, O must be closer to the prior than is S.
sonable people. (When S is maximally regressive, O is equally
Let us restrict our analysis to some specific group of n regressive.)
individuals who all expect to take a test together. These There are three possible relationships between S and
people, on average, expect that their performances will the prior:
be average, relative to the others in the group. It is not Case 1: S > prior, meaning that, on average, people
necessary that everyone believe that they are exactly think that they did better than they expected to do; thus
average, just that beliefs are balanced within the group. S > O (or else a contradiction follows; if we assume
In other words, for every person (or subset of people) S > O to be false, i.e., if O P S, and (as given) S > prior,
who believes that they are better than others (or better then O P S > prior, and O would be farther from the
than some individual) there is another person (or subset prior than S, violating Rule O), and people will, on aver-
of people) who believes that they are worse than others age, believe they performed better than others.
(or worse than some individual) to the same degree, and Case 2: S < prior, meaning that, on average, people
vice versa. think that they did worse than they expected to do; thus
Within the group, let S be the average of all individ- S < O (ELSE: O 6 S < prior, and O would be farther
uals’ estimates of the absolute performance of self; let from the prior than S, violating Rule O), and people
O = the average of all individuals’ estimates of the will, on average, believe they performed worse than
absolute performance of others. To be precise: others.
Case 3: S = prior, meaning that, on average, people
ðs1 þ s2 þ . . . sn Þ think that they did as they expected to do; thus S = O
S¼
n (ELSE: O > S = prior (or O < S = prior), and O would
be less regressive than S, violating Rule O), and people
where sn is the nth person’s self-estimate.
will, on average, believe they performed the same as
ðo1 þ o2 þ . . . on Þ others.
O¼ Example: Suppose there are three test takers, A, B, C,
n
each completing a test that is scored out of 100. Suppose
where on is the nth person’s estimate of (the average of) that, prior to taking the test, the average expected
others’ scores (not including self). score is 50. Suppose A’s actual score = 90; B’s actual
score = 65; C’s actual score = 40. The average actual
ðs1 þ s2 þ . . . sn Þ  ðo1 þ o2 þ . . . on Þ test-score = 65. On average, the test takers did better
SO¼
n than expected (65 > 50), even though some did worse
and than expected (e.g., C scored 40). Suppose, for the sake
of simplicity, that all three know exactly how well they
ðs1  o1 þ s2  o2 þ . . . sn  on Þ themselves did, but they know their sense of others is
SO¼
n imperfect. Granted, specific numerical examples of esti-
S  O is equal to the average comparative judgment of mates will depend on individual test-takers and specific
self to others. When S  O is positive, it implies a tasks. With imperfect information, however, as in
BTA effect: On average, people believe that they are Bayesian updating, people’s best estimates (in this case
better than others. When S  O is negative, it implies of others) will tend to fall between actual values and
a WTA effect: People believe they are worse than others.
At the outset, assume that S = O (on average, people
expect that their performance will be average, relative
Table 4
to the others).
(Row A, Col B) = B’s prediction of A’s score
Let people acquire additional information about
Predictor
their own (past or anticipated) performance. To the
extent that this information about self is different A B C
than prior information, it will justify updating beliefs Target A 90 70 70
about one’s own performance (away from the prior). B 57.5 65 57.5
C 45 45 40
To the extent that information about the self is more
210 D.A. Moore, D.M. Cain / Organizational Behavior and Human Decision Processes 103 (2007) 197–213

their prior expectations. Table 4 depicts reasonable esti- 3. What does MTV stand for?
mates. In keeping with the idea that estimates of self are 4. On what continent is the country of Egypt?
less regressive than estimates of others, suppose that 5. What is the most widely spoken language in the US,
estimates of others are computed using a somewhat after English?
arbitrary equal weighting of the prior and actual
score: (prior + actual)/2. Since the average expected Tiebreaker: What is the height of the Eiffel Tower
score is 50, assume that priors of all people’s scores by (in feet)?
all others = 50. So, for example, if A scores 90, A will
predict that B scores (90 + 50)/2 = 70. B.2. Simple Test 2
Next we calculate S and O for the group:

S ¼ ðA’s estimate of self þ B’s estimate of self 1. What was the first name of the Carnegie who founded
the Carnegie Institute of Technology?
þ C’s estimate of selfÞ=3 2. How many states are there in the United States?
¼ ð90 þ 65 þ 40Þ=3 3. In which month is Thanksgiving celebrated in the
United States?
¼ 65
4. Harrisburg is the capital of what US state?
O ¼ ½A’sðaverageÞestimate of others 5. On what continent is the country of France
þ B’sðaverageÞestimate of others located?

þ C’sðaverageÞestimate of others=3 Tiebreaker: How many films did Alfred Hitchcock

¼ ½1=2ð57:5 þ 45Þ þ 1=2ð70 þ 45Þ þ 1=2ð70 þ 57:5Þ=3 direct?
¼ ½51:25 þ 57:5 þ 63:75=3 B.3. Simple Test 3
¼ 57:5

Result: S (65) > O (57.5). On average, relying on 1. Which American civil rights leader gave a famous
sensible rules of inference, but using systematically speech in which he repeated the lines, ‘‘I have a
imperfect information (and which is known to be dream. . .’’
imperfect), people believe that they (S) are better than 2. What American director was behind the movies, A.I.,
others (O). E.T., Minority Report, Saving Private Ryan, and
Note that C actually does (40) worse than expected Jurassic Park?
(50) and everyone knows it [but C knows it best; as 3. What is the name of Pittsburgh’s professional hockey
shown in the preceding table, where (C, C) = 40, while team?
(C, A) = 45, and (C, B) = 45]. Nevertheless, on average, 4. What Pennsylvania city is know for being at the
the group thinks it did better than expected (actu- confluence of the Allegheny and Monongahela
al = 65 = estimated > expected = 50). Our theory holds Rivers?
that, when this occurs, people will (on average) think 5. What country lies directly north of the United States?
they did better than average. The S  O calculation
bears this out: S  O = 7.5 > 0. Tiebreaker: How many member states are there in the
The logic outlined above works just the same if United Nations?
each person’s estimate of his or her own score is
also imperfect and known to be imperfect, and B.4. Simple Test 4
therefore it also regresses toward the prior. The only
key requirement is that estimates of others be 1. What American became the first person to ever win
more regressive than estimates of self, and Rule O the Tour de France 6 times?
holds. 2. Paris is the capital of what country?
3. In what large US city is the famous Times Square
located?
Appendix B. Trivia quizzes (four simple, four difficult), 4. Where in the human body is the cerebellum located?
experiment 1 5. What famous act of military aggression by Japan
happened on Dec 7, 1941 that brought the United
B.1. Simple Test 1 States into World War II?

1. Who was the first president of the United States? Tiebreaker: How many men signed the Declaration
2. How many inches are there in a foot? of Independence?
 D.A. Moore, D.M. Cain / Organizational Behavior and Human Decision Processes 103 (2007) 197–213 211

B.5. Difficult Test 1 B.7. Difficult Test 3

1. In what European city would you find the famous 1. Blues musician Huddie Ledbetter is better known by
Tivoli Gardens? what name?
2. Truth or Consequences is a city in what US state? 2. What make and model of car holds the record
3. What company’s research and development lab was for being the most widely produced car in the
once known as the ‘‘House of Magic?’’ world?
4. What is the largest moon of Saturn? 3. Laudanum is a form of what drug?
5. What African country lies directly south of Egypt? 4. Who was the president of Indonesia, as of August
2002?
Tiebreaker: In the 2000 US Census, what was the 5. What is the capital of Nepal?
population of Walla Walla, Washington?
Tiebreaker: Approximately how many pieces of art
B.6. Difficult Test 2 did Pablo Picasso create during his lifetime?

1. Thomas Hooker is associated with the founding of B.8. Difficult Test 4

which of the thirteen American colonies?
2. Who is the only US president to have served two non- 1. Which team won the first NBA Draft Lottery?
consecutive terms in office? 2. The Nobel Prizes are awarded in what two
3. In Quentin Tarrantino’s Reservoir Dogs, what is the cities?
alias of the man who is revealed to be an undercover 3. Dr. Faustus is best known for selling what item?
police officer? 4. What two South American countries are land-
4. In The Odyssey, who was the son of Ulysses locked?
(Odysseus)? 5. Pro football announcer John Madden coached which
5. Who was voted Time magazine’s Man of the Year in team to a Super Bowl victory?
1938?
Tiebreaker: How many consecutive weeks did the
Tiebreaker: What is the land area of Morocco (in Pink Floyd album Dark Side of the Moon spend on
square kilometers)? the billboard music charts?

Appendix C

Trivia questions used in the simple and difficult trivia quizzes (Experiment 2).

Simple Difficult
1. How many inches are there in a foot? Which creature has the largest eyes in the world?
2. What is the name of Pittsburgh’s professional How many verses are there in the Greek national
hockey team? anthem?
3. Which species of whale grows the largest? What company produced the first color television
sold to the public?
4. Who is the President of the United States? How many bathrooms are there in the White
House (the residence of the US President)?
5. Harrisburg is the capital of what US state? Which monarch ruled Great Britain the longest?
6. What was the first name of the Carnegie who founded The word ‘‘planet’’ comes from the Greek word
the Carnegie Institute of Technology? meaning what?
7. How many states are there in the United States? What is the name of the traditional currency of
Italy (before the Euro)?
8. What continent is Afghanistan in? What is Avogadro’s number?
9. What country occupies an entire continent? Who played Dorothy in ‘‘The Wizard of Oz’’?
10. Paris is the capital of what country? Who wrote the musical ‘‘The Yeoman
of the Guard’’?
Tiebreaker question: How many people live in Pennsylvania?

Answers—Simple: (1) 12 (2) Penguins (3) Blue (4) George W. Bush (5) Pennsylvania (6) Andrew (7) 50 (8) Asia (9) Australia (10) France. Difficult:
(1) Giant squid (2) 158 (3) RCA (4) 32 (5) Queen Victoria (6) wanderer (7) Lira (8) 6.02 · 1023(9) Judy Garland (10) Gilbert and Sullivan. Tiebreaker:
12,281,054.
212 D.A. Moore, D.M. Cain / Organizational Behavior and Human Decision Processes 103 (2007) 197–213

Appendix D. Supplementary data Hoelzl, E., & Rustichini, A. (2005). Overconfident: Do you put your
money on it? Economic Journal, 115(503), 305–318.
Hoorens, V., & Buunk, B. P. (1993). Social comparison of health
Supplementary data associated with this article can risks: Locus of control, the person-positivity bias, and unreal-
be found, in the online version, at doi:10.1016/j.obhdp. istic optimism. Journal of Applied Social Psychology, 23(4),
2006.09.002. 291–302.
Juslin, P., Winman, A., & Olsson, H. (2000). Naive empiricism and
dogmatism in confidence research: A critical examination of the
hard-easy effect. Psychological Review, 107(2), 384–396.
References Klar, Y. (2002). Way beyond compare: Nonselective superiority and
inferiority biases in judging randomly assigned group members
Alicke, M. D., Klotz, M. L., Breitenbecher, D. L., Yurak, T. J., & relative to their peers. Journal of Experimental Social Psychology,
Vredenburg, D. S. (1995). Personal contact, individuation, and the 38(4), 331–351.
better-than-average effect. Journal of Personality and Social Psy- Klar, Y., & Giladi, E. E. (1997). No one in my group can be below the
chology, 68(5), 804–825. group’s average: A robust positivity bias in favor of anonymous
Baron, J. (1997). Confusion of relative and absolute risk in valuation. peers. Journal of Personality and Social Psychology, 73(5), 885–901.
Journal of Risk and Uncertainty, 14, 301–309. Klar, Y., & Giladi, E. E. (1999). Are most people happier than their
Biernat, M., Manis, M., & Kobrynowicz, D. (1997). Simultaneous peers, or are they just happy? Personality and Social Psychology
assimilation and contrast effects in judgments of self and others. Bulletin, 25(5), 585–594.
Journal of Personality and Social Psychology, 73(2), 254–269. Klar, Y., Medding, A., & Sarel, D. (1996). Nonunique invulnerability:
Burson, K. A., & Klayman, J. (2005). Judgments of performance: The Singular versus distributional probabilities and unrealistic opti-
relative, the absolute, and the in-between. Ann Arbor, unpublished mism in comparative risk judgments. Organizational Behavior and
manuscript. Human Decision Processes, 67(2), 229–245.
Burson, K. A., Larrick, R. P., & Klayman, J. (2006). Skilled or Klein, W. M. P., & Kunda, Z. (1994). Exaggerated self-assessments
unskilled, but still unaware of it: How perceptions of difficulty and the preference for controllable risks. Organizational Behavior
drive miscalibration in relative comparisons. Journal of Personality and Human Decision Processes, 59(3), 410–427.
and Social Psychology, 90(1), 60–77. Klein, W. M. P., & Weinstein, N. D. (1997). Social comparison and
Camerer, C. F., & Lovallo, D. (1999). Overconfidence and excess unrealistic optimism about personal risk. In B. P. Buunk & F. X.
entry: An experimental approach. American Economic Review, Gibbons (Eds.), Health, coping, and well-being: Perspectives from
89(1), 306–318. social comparison theory (pp. 25–61). Mahwah, NJ: Lawrence
Chambers, J. R., & Windschitl, P. D. (2004). Biases in social Erlbaum Associates.
comparative judgments: The role of nonmotivational factors in Kruger, J. (1999). Lake Wobegon be gone! The ‘‘below-average
above-average and comparative-optimism effects. Psychological effect’’ and the egocentric nature of comparative ability judg-
Bulletin, 130(5). ments. Journal of Personality and Social Psychology, 77(2),
College Board. (1976–1977). Student descriptive questionnaire. 221–232.
Princeton, NJ: Educational Testing Service. Kruger, J., & Burrus, J. (2004). Egocentrism and focalism in unrealistic
Cooper, A. C., Woo, C. Y., & Dunkelberg, W. C. (1988). Entrepre- optimism (and pessimism). Journal of Experimental Social Psy-
neurs’ perceived chances for success. Journal of Business Venturing, chology, 40(3), 332–340.
3(2), 97–109. Kruger, J., Windschitl, P. D., Burrus, J., Fessel, F., & Chambers, J. R.
Dunne, T., Roberts, M. J., & Samuelson, L. (1988). Patterns of firm (2006). The rational side of egocentrism in social comparisons,
entry and exit in US manufacturing industries. Rand Journal of unpublished manuscript.
Economics, 19(4), 495–515. Kunda, Z. (1990). The case for motivated reasoning. Psychological
Dunning, D., Heath, C., & Suls, J. M. (2004). Flawed self-assessment: Bulletin, 108(3), 480–498.
Implications for health, education, and business. Psychological Lichtenstein, S., Fischhoff, B., & Phillips, L. D. (1982). Calibration of
Science in the Public Interest, 5(3), 69–106. probabilities: The state of the art in 1980. In D. Kahneman, P.
Dunning, D., Meyerowitz, J. A., & Holzberg, A. D. (1989). Ambiguity Slovic, & A. Tversky (Eds.), Judgment under uncertainty: Heuristics
and self-evaluation: The role of idiosyncratic trait definitions in and biases (pp. 306–333). Cambridge, England: Cambridge Uni-
self-serving assessments of ability. Journal of Personality and Social versity Press.
Psychology, 57(6), 1082–1090. Malmendier, U., & Tate, G. (2005). CEO overconfidence and
Erev, I., Wallsten, T. S., & Budescu, D. V. (1994). Simultaneous over- corporate investment. Journal of Finance, (60), 6.
and underconfidence: The role of error in judgment processes. Mata, J., & Portugal, P. (1994). Life duration of new firms. Journal of
Psychological Review, 101(3), 519–527. Industrial Economics, 42(3), 227–246.
Fischhoff, B., & De Bruin, Bruine (1999). Fifty-fifty = 50%?. Journal of Moore, D. A. (2004). Myopic prediction, self-destructive secrecy, and
Behavioral Decision Making 12(2), 149–163. the unexpected benefits of revealing final deadlines in negotiation.
Fox, C. R., & Rottenstreich, Y. (2003). Partition priming in judgment Organizational Behavior and Human Decision Processes, 94(2),
under uncertainty. Psychological Science, 14(3), 195–200. 125–139.
Geroski, P. A. (1996). What do we know about entry? International Moore, D. A. (2005). When good = better than average. Pittsburgh:
Journal of Industrial Organization, 13(4), 421–441. Tepper Working Paper 2004-E38.
Greico, D., & Hogarth, R. M. (2004). Excess entry, ambiguity seeking, Moore, D. A., & Kim, T. G. (2003). Myopic social prediction and the
and competence: An experimental investigation, unpublished solo comparison effect. Journal of Personality and Social Psychol-
manuscript. ogy, 85(6), 1121–1135.
Harris, P. (1996). Sufficient grounds for optimism? The relationship Moore, D. A., Oesch, J. M., & Zietsma, C. What competition? Myopic
between perceived controllability and optimistic bias. Journal of self-focus in market entry decisions. Organization Science, in press.
Social and Clinical Psychology, 15(1), 9–52. Myers, D. G. (1998). Social psychology (5th ed.). New York: McGraw-
Hayward, M. L. A., & Hambrick, D. C. (1997). Explaining the Hill.
premiums paid for large acquisitions: Evidence of CEO hubris. Odean, T. (1998). Volume, volatility, price, and profit when all traders
Administrative Science Quarterly, 42, 103–127. are above average. Journal of Finance, 53(6), 1887–1934.
 D.A. Moore, D.M. Cain / Organizational Behavior and Human Decision Processes 103 (2007) 197–213 213

Perloff, L. S., & Fetzer, B. K. (1986). Self-other judgments and Svenson, O. (1981). Are we less risky and more skillful than our fellow
perceived vulnerability to victimization. Journal of Personality and drivers? Acta Psychologica, 47, 143–151.
Social Psychology, 50(3), 502–510. US Small Business Administration. (2003). Longitudinal Establish-
Price, P. C. (2001). A group size effect on personal risk judgments. ment and Enterprise Microdata. Washington, DC: Office of
Memory and Cognition, 29, 578–586. Advocacy (202-205-6530).
Radzevick, J. R., & Moore, D. A. (2006). For the love of the game? Weinstein, N. D. (1980). Unrealistic optimism about future life events.
Betting, prediction, and myopic bias in athletic competition. Journal of Personality and Social Psychology, 39(5), 806–820.
Pittsburgh: Tepper Working Paper 2005-E7. Windschitl, P. D., Kruger, J., & Simms, E. (2003). The influence of
Rose, J. P., & Windschitl, P. D. (2006). How egocentric optimism egocentrism and focalism on people’s optimism in competitions:
change in response to feedback in repeated competitions, unpub- When what affects us equally affects me more. Journal of
lished manuscript. Personality and Social Psychology, 85(3), 389–408.
Suls, J. M., Lemos, K., & Stewart, H. L. (2002). Self-esteem, construal, Zajac, E. J., & Bazerman, M. H. (1991). Blind spots in industry and
and comparisons with the self, friends, and peers. Journal of competitor analysis: Implications of interfirm (mis)perceptions for
Personality and Social Psychology, 82(2), 252–261. strategic decisions. Academy of Management Review, 16(1), 37–56.