How Chronic Self-views Influence (and Potentially Mislead) Estimates of Performance

Abstract

An important source of people's perceptions of their performance, and potential errors in those perceptions, are chronic views people hold regarding their abilities. In support of this observation, manipulating people's general views of their ability, or altering which view seemed most relevant to a task, changed performance estimates independently of any impact on actual performance. A final study extended this analysis to why women disproportionately avoid careers in science. Women performed equally to men on a science quiz, yet underestimated their performance because they thought less of their general scientific reasoning ability than did men. They, consequently, were more likely to refuse to enter a science competition.

Full text

ATTITUDES AND SOCIAL COGNITION
How Chronic Self-Views Influence (and Potentially Mislead)
Estimates of Performance
Joyce Ehrlinger and David Dunning
Cornell University
An important source of people’s perceptions of their performance, and potential errors in those
perceptions, are chronic views people hold regarding their abilities. In support of this observation,
manipulating people’s general views of their ability, or altering which view seemed most relevant to a
task, changed performance estimates independently of any impact on actual performance. A final study
extended this analysis to why women disproportionately avoid careers in science. Women performed
equally to men on a science quiz, yet underestimated their performance because they thought less of their
general scientific reasoning ability than did men. They, consequently, were more likely to refuse to enter
a science competition.
How’m I doin’?
Ed Koch, Mayor of New York City, 1978–1989
The knowledge of whether one is succeeding or failing at a task
can be a precious piece of information. The degree to which
individuals think that they performed well often has substantial
consequences for the actions they take and the outcomes they
encounter. A young person’s impression of how well he or she did
on a first date can determine if he or she pursues the relationship.
A law student’s impression of how well he has answered practice
questions for a bar exam will determine whether he continues to
study. A young actor’s impression of her performance in the latest
community play may determine whether she decides to hop on the
next bus to Hollywood rather than enroll in the local college. Thus,
much like Ed Koch did during his 12 years as mayor of New York
City, people frequently ask and reflect on how they are doing at a
particular task.
However, gaining accurate insight into one’s performance can
be an elusive goal. Abstract beliefs about one’s ability, as well as
beliefs about how well one is performing in particular instances,
often correlate rather imperfectly with the reality of perfor-
mance—and at times fail to correlate at all. For example, the
accuracy with which people convey their feelings to others fre-
quently does not correlate with their perception of how well they
have conveyed them (Riggio, Widaman, & Friedman, 1985). Doc-
tors’ estimates of their knowledge about a variety of disorders fail
to correlate with demonstrated knowledge (Tracey, Arroll, Rich-
mond, & Barham, 1997); nurses’ estimates of their basic life
support skills are not related to their actual level of knowledge
(Marteau, Johnston, Wynne, & Evans, 1989). For adolescent boys
taking a quiz of condom use, their assessment of their knowledge
correlates only slightly with their actual knowledge (Crosby &
Yarber, 2001). Gun owners have little insight into how they have
performed on a test of gun use, safety, and knowledge (Ehrlinger,
Johnson, & Dunning, 2002). When attempting to discover who
might be lying, the confidence people imbue in their judgments
correlates only .04 with their accuracy (DePaulo, Charlton, Coo-
per, Lindsay, & Muhlenbruck, 1997). The confidence of eyewit-
nesses when making identifications correlates very modestly and
unreliably with whether they have accurately identified the culprit
out of the lineup (Sporer, Penrod, Read, & Cutler, 1995).
Although extreme, these examples are not necessarily isolated
cases. In general, the perceptions people hold, of either their
overall ability or specific performance, tend to be correlated only
modestly with their actual performance. In 1982, Mabe and West
surveyed 55 studies that had examined the relationship between
perception and reality of performance across a wide variety of
domains, ranging from managerial duties, clerical duties, scholas-
tic performance to athletic skill, and found the average correlation
to hover around .29. To be sure, such correlations did tend to be
positive and statistically significant, but they also tended to be far
from perfect. Whether people’s perceptions were expressed as
specific task predictions (e.g., what grade will you get in this
Editor’s Note. James Olson served as the guest editor for this arti-
cle.—PD
The research described herein was supported financially by National
Institute of Mental Health Grant RO1-56072 awarded to David Dunning.
We thank Leah Doane, Kevin Van Aelst, and Nathalie Vizueta for
assisting in the collection of data and Dick Darlington, Thomas Gilovich,
and Dennis Regan for their comments on previous versions of this article.
Elaine Wethington’s help with AMOS is appreciated.
Correspondence concerning this article should be addressed to either
Joyce Ehrlinger or David Dunning, Department of Psychology, Uris Hall,
Cornell University, Ithaca, New York 14853-7601. E-mail: jme15@cornell
.edu or dad6@cornell.edu
Journal of Personality and Social Psychology, 2003, Vol. 84, No. 1, 5–17
Copyright 2003 by the American Psychological Association, Inc. 0022-3514/03/$12.00 DOI: 10.1037/0022-3514.84.1.5
5

particular class?) or as more abstract assessments (e.g., their over-
all assessment of their organizational skill) mattered only slightly
in influencing the correlation.
1
Studies in the workplace also tend to find that perceptions of
performance are only modestly related to more objective measures
of performance. Across a number of studies, employees’ percep-
tions of their job performance correlate, on average, only .35 with
their supervisors’ and .36 with their peers’ opinion of them, even
after correcting for unreliable measurement of ability. By contrast,
the perceptions of peers and supervisors correlate roughly .62
(Harris & Schaubroeck, 1988). Peer perceptions of leadership
ability have been shown to better predict such outcomes as rec-
ommendations for promotion among naval officers than do self-
perceptions (Bass & Yammarino, 1991). Peer and supervisor per-
ceptions also predict how well surgical residents will do on an
objective test of surgical skill, whereas self-perceptions do not
(Risucci, Tortolani, & Ward, 1989).
In short, people often possess an imperfect degree of what
educational, cognitive, and social psychologists refer to as meta-
cognitive insight, which, among its many meanings, refers to the
skill of anticipating the likely accuracy and error of one’s re-
sponses (Metcalfe & Shimamura, 1994; Yzerbyt, Lories, & Dar-
denne, 1998)—and this lack of insight often extends to situations
in which people attempt to estimate their performance on a par-
ticular task or test. In recent years, research has increasingly
focused on why metacognitive knowledge can be so imperfect. For
example, Kruger and Dunning (1999) suggested that some indi-
viduals, namely, the incompetent, are just not in a position to judge
their performances accurately. Not only does their lack of skill
prevent the incompetent from forming current responses to situa-
tional demands, but it also prevents them from recognizing when
judgments will be accurate and when they will be erroneous. In a
series of studies, Kruger and Dunning demonstrated that incom-
petent individuals (i.e., those performing poorly relative to their
peers) were the least able to assess the quality of their performance
as well as the performances of others (see also Bem & Lord, 1979;
Fagot & O’Brien, 1994; Kunkel, 1971; Maki, Jonas, & Kallod,
1994; Moreland, Miller, & Laucka, 1981; Shaughnessy, 1979;
Sinkavich, 1995, for similar findings).
Sources of Performance Estimates: The Importance of
Chronic Self-Views
Although demonstrating one reason why people fail to evaluate
their performance accurately, Kruger and Dunning (1999) did not
explore an equally important question: Where do people’s evalu-
ations of their performances come from, particularly instances in
which they assess their performance on a specific test? If not
strongly based on their actual achievement, what are people’s
perceptions of their performance based on?
In this article, we identify a potentially important source of
people’s impressions of their performance on a specific task, one
that can be independent of their actual performance and that can
potentially produce errors in estimation. In four studies, we sug-
gest that people’s evaluation of how well they have done at a
particular task is not exclusively determined by their experience
with the current task or the cues they encounter as they complete
it. Rather, these evaluations stem from general views individuals
chronically hold about their abilities, ones they hold before they
even begin a task. People who think they have terrific skills at
logical reasoning will think they did better on a test of logic than
will those who think they do not have the skill, independent of the
actual performance they post. In a phrase, performance evaluations
on specific tasks may not be so much bottom-up as they are
top-down, formed by referring to a person’s chronic view about his
or her abilities in the specific domain in question.
One could argue that it is proper to base performance estimates
on chronic self-views because they are likely to be based on a
lifetime of performances in the domain in question. Thus, differing
self-views would largely reflect differences in objective compe-
tence, and relying on self-views would be an important strategy
toward accurately assessing one’s current performance at a task. In
part, this is true. As Mabe and West (1982) discovered, chronic
self-views did tend to predict performances, albeit modestly.
That said, there are many reasons to believe that self-views
would not necessarily objectively reflect one’s past performances.
Self-views are likely to be influenced by a host of features that
have little to do with true ability level. For example, assessments
of ability tend to be self-serving (for reviews, see Dunning, 1993;
Kunda, 1990). People do not dispassionately count up their success
and failures to form a self-impression as much as they actively
interpret them to fit chronic views, usually positive ones, of the
self (Dunning, 2001; Kunda, 1990; Miller & Ross, 1985). Positive
feedback is more likely to be accepted unquestioningly; negative
feedback is placed under close scrutiny with an eye toward dis-
counting it (Ditto & Lopez, 1992). Finally, people’s memories of
past events become distorted according to their self-views, and so
the act of accurately recording some feedback provides little
insurance that it will be accurately recalled in the future (Story,
1998). Given that self-views are unlikely to be a completely
objective accounting of how one has performed in the past, they
may not necessarily lead to accuracy over error in performance
estimates.
Some fragments exist in the psychological literature indirectly
suggesting that performance evaluations are top-down in nature.
People rely heavily on self-views for specific domains when
estimating how well they pass other “tests.” Kenny and DePaulo
(1993) discovered that people’s beliefs about how they were
viewed by others were more related to their own beliefs about
1
Of course, one well-developed literature systematically contradicts this
conclusion, finding that self-perceptions are tightly correlated with objec-
tive performance—namely, work on self-efficacy (Bandura, 1977, 1997).
This contradiction may be instructive. In much of the work reviewed in the
text, participants are faced with tasks in which good performance depends
on knowledge, intelligence, wisdom, or common sense. In self-efficacy
work, performance depends on something else—participants’ ability to
control or regulate their actions, such as controlling a phobia (Bandura,
1977, 1997). Self-assessment questions in self-efficacy work are also very
tightly crafted to specifically match the tasks confronting participants. It
might be the case that people can provide much more accurate self-
assessments when the issue is what they can control versus what they
know, and, given the lessons learned from self-efficacy research, when
self-assessments of control are quite specific to the task at hand (Bandura,
1977, 1997). Given that Mabe and West (1982) found a wide range of
self-view/performance correlations across the studies they reviewed, this
speculation might be one well worth pursuing in determining when self-
perceptions are accurate and when they are not.
6
EHRLINGER AND DUNNING

themselves than they were to the actual impressions that others
held. McFarland, Ross, and DeCourville (1989) discovered that
women’s memories of a menstrual cycle were biased toward their
general theories about how they thought they were affected by
menstrual cycles. With respect to performance situations, both
self-esteem and depression have been shown to predict people’s
predictions about how well they will perform in the future (Dun-
ning & Story, 1991; McFarlin & Blascovich, 1981). Self-esteem
also predicts what personality traits were reflected in performances
(Lewinsohn, Mischel, Chaplin, & Barton, 1980), as well as
whether people will evaluate their performances as “good” or
“bad” (Jussim, Coleman, & Nassau, 1987; Lindeman, Sundvik, &
Rouhiainen, 1995).
However, to date, virtually no work has examined whether
self-views influence what people think their performance objec-
tively was rather than their satisfaction or dissatisfaction with a
performance. That is, in a situation in which the quality of a
person’s performance can be objectively determined, do self-views
influence what level of performance people think they have at-
tained? And do self-views produce estimates of performance that
are independent of the reality of the performance? To our knowl-
edge, there is only one extant study showing that self-views
influence estimates of performance independently of an actual
performance that can be verified. In 1976, Shrauger and Terbovic
asked participants who possessed high or low self-esteem to com-
plete a conceptual learning task. Although high- and low-esteem
participants did not differ in the number of items they answered
correctly, high-esteem participants estimated that that they had
gotten a higher number of items right than did their low-esteem
peers. High-esteem participants also believed they outperformed a
higher percentage of their peers than did low-esteem individuals.
In this article, we extend this initial finding by Shrauger and
Terbovic (1976) to provide a more comprehensive examination of
the role played by chronic self-views in estimates of performance.
Like, Shrauger and Terbovic, we examine whether the relationship
between self-views and performance estimates exists independent
of any relationship to actual performance. Unlike Shrauger and
Terbovic, we examine more specific self-views (e.g., perceptions
of scientific talent), ones that might have a better chance of being
related to a person’s actual performance than more general im-
pressions of self-esteem. But more than that, we examine whether
self-views play a causal role in producing performance estimates
that are independent of actual performance, focusing on self-views
that are specific to a domain rather than a general sense of
self-esteem. We do so by manipulating which specific self-view is
seen as relevant to a task, or by raising or lowering a person’s
self-view, to see if we can alter performance estimates even when
no change in actual performance is observed. We place the issue of
accuracy and error under closer scrutiny, to see if self-views are
related to whether people under- or overestimate their perfor-
mances. Finally, we examine whether self-views, independent of
actual performance, produce performance estimates that have an
impact on behavior.
Overview of Present Research
In four studies, we examined the extent to which individuals rely
on self-views of ability in a top-down fashion in order to estimate
performance levels. In Study 1, we explored whether perceptions
of performance were predicted by chronic self-views independent
of how individuals actually performed. In Study 2, we manipulated
which self-view was supposedly relevant to a task to see if we
could alter participants’ performance estimates independently of
any change in performance. In Study 3, we manipulated the
favorability of self-views directly to see if that influenced perfor-
mance estimates independently of any difference in actual perfor-
mance. Finally, in Study 4, we extended our analysis to behavioral
choices, specifically, the opportunity to enter into a science com-
petition. We predicted that perceptions of performance would be
driven significantly by self-views. We expected that perceptions of
performance would influence choices to participate in the contest,
independent of actual performance. Thus, we compared the per-
ceptions, performances, and choices of men and women in this
experiment in an effort to contribute to an understanding of why
women leave scientific careers more frequently than do men.
Study 1: Do Self-Views Matter?
Study 1 was designed to fulfill a two-fold task. The first was to
examine the exact role played by chronic self-views in estimates of
performance. We expected self-views to be related to performance
estimates. However, that relationship could come in two forms.
First, self-views might be related to performance estimates simply
because self-views track differences in actual performance. Once
that actual performance was controlled for, the relationship be-
tween self-views and estimates would disappear. Second, and
actually what we predicted, self-views could be significantly re-
lated to performance estimates, independent of any relation to
actual performance. Even after accounting for actual performance,
more positive self-views could be associated with more optimistic
performance estimates, more negative self-views with more pes-
simistic estimates.
The second task was to gauge how important self-views were in
predicting performance estimates. We did this by comparing how
well self-views predicted performance estimates with what should
be predicting performance estimates, the actual performance par-
ticipants had attained.
In Study 1, we first measured self-views of abstract reasoning
ability. We then gave students a test of that ability and asked them
to estimate how well they had done on the test, in terms of their
raw score as well as their percentile ranking relative to their peers.
Method
Participants. Participants were 59 (44 women, 15 men) Cornell Uni-
versity undergraduate students who were recruited from large lecture
psychology courses to take part in the study in exchange for extra credit
toward their course grades.
Procedure. We described the first study as one about perceptions of
strengths and weaknesses. Participants rated the extent to which they
possessed 14 abilities on a scale ranging from 1 (not at all)to9(to an
extreme degree). One item asked the degree to which they possessed an
“ability to reason abstractly”; the others referred to irrelevant abilities.
After completing the scale, participants completed a supposed second
study, described as focusing in on a single ability. Participants were asked
to complete a 10-item multiple-choice test consisting of items taken from
a Law School Aptitude Test (LSAT) preparation guide (Orton, 1993). We
labeled the test as one measuring “logical reasoning ability” rather than
“abstract reasoning” to reduce a demand for consistency by obscuring the
relationship between the initial ability questionnaire and the test. Finally,
7
SELF-VIEWS AND PERFORMANCE ESTIMATES

participants provided an estimate of how many of the 10 items they had
answered correctly as well as a percentile estimate of their performance
relative to other Cornell University students in the experiment (between 1
and 100).
Results and Discussion
Gender did not interact with any of the measures in this or the
next two studies. Thus, all analyses are reported without mention
of gender until Study 4.
Accuracy of performance estimates. Replicating past research,
participants tended to overestimate how well they had done on the
test relative to their peers. Participants on average believed their
performance fell in the 61st percentile, significantly higher than
the true average (50th percentile), t(58) ⫽ 4.54, p ⬍ .0001.
Participants, however, did not overestimate the raw number of
items they got right (t ⬍ 1). Estimates of performance were
correlated with actual performance, but the strength of that corre-
lation depended on the specific measure examined. Percentile
estimates were only marginally correlated with the actual percen-
tile participants attained, r(58) ⫽ .22, p ⬍ .10, but estimates of raw
score were correlated with the actual raw score that participants
achieved, r(58) ⫽ .42, p ⬍ .001.
The role of self-views. Our key hypothesis was that self-views
would be a significant source of performance estimates, indepen-
dent of any relationship with actual performance. Looking at
zero-order correlations, we found, as predicted, that self-views did
significantly predict performance estimates—r(58) ⫽ .31 and .42,
ps ⬍ .05, for percentile and raw estimates, respectively.
To address whether self-views contribute to performance esti-
mates when actual performance is held constant, we conducted
multiple regression analyses predicting performance estimates
from both self-views and actual performance. For percentile esti-
mates of performance, the overall model was marginally signifi-
cant, F(2, 56) ⫽ 3.53, p ⬍ .10. Of more importance, performance
estimates turned out to be significantly related to self-views (
␤
⫽
.27, p ⬍ .05), but not to participants’ actual percentile rankings
(
␤
⫽ .13, ns). Estimates of raw scores showed a similar pattern.
Self-views of abstract reasoning ability and actual performance
(measured as raw score) predicted raw score estimates, F(2,
56) ⫽ 9.81, p ⬍ .0005. Taken separately, each also significantly
predicted raw score estimates—and did so equally (both
␤
s ⫽ .31,
p ⬍ .05).
Summary. In sum, self-views of logical ability appeared to be
a significant source of participants’ perceptions of how well they
had performed on the test. Indeed, depending on the measure,
self-views were just as, and sometimes more, important a source as
actual performance. Participants’ estimates of their percentile
ranking were significantly correlated with their self-views, but not
with their actual percentile ranking. Participants’ estimates of their
raw score on the test were as related to their preconceived notions
of self as they were with their actual raw scores.
Of key importance, self-views significantly influenced perfor-
mance estimates after controlling for actual performance regard-
less of the measure. This suggests that self-views are partially
responsible for the mistakes people make when they evaluate how
well they have performed on a task. Relative to people with low
opinions of their abilities, people with high opinions will estimate
that they are doing much better, even when both groups are posting
equal performances. Conversely, even when people with little
confidence in their abilities perform just as well as their high
self-view counterparts, they estimate that they are doing worse.
Study 2: Which Self-View to Apply?
Study 1 suggested that self-views are an important source of
performance estimates, but did not do so conclusively. The design
of Study 1 was correlational, and although it showed that self-
views were related to performance estimates at times more
strongly than actual performance was, it could not show that
self-views are a causal influence over those estimates.
Thus, in Studies 2 and 3, we performed experiments to show
more directly that chronic self-views influence performance esti-
mates, and do so independently of actual performance. In Study 2,
we did so by varying which of two possible self-views was
purportedly relevant to a test that we asked participants to com-
plete. We gave participants a test based on the Graduate Record
Examination (GRE) analytical section, and capitalized on the fact
that the test can be described as measuring several different abil-
ities. We told one group of participants that the test measured
abstract reasoning ability, an ability our participant population
tended to think they had in abundance. We told another group that
the test assessed computer programming ability, an ability our
population tended to think they did not have.
We predicted, independent of actual performance, that partici-
pants in the abstract reasoning group, consulting self-views of
abstract reasoning ability, would estimate they did much better
than would participants in the computer programming group, con-
sulting self-views of programming ability, even though each group
confronted exactly the same test. Through mediational analyses,
we also sought to show that these differences in performance
estimates would arise precisely because students hold chronic
notions that they possess more abstract reasoning ability than they
do computer programming skills.
Method
Participants. Participants were 91 (64 women, 27 men) Cornell Uni-
versity students who participated in exchange for extra credit in large
lecture psychology courses.
Procedure. Participants first rated the extent to which they possessed
a number of abilities, including the ability to program a computer and the
ability to think about abstract concepts, using the 9-point scale described
for Study 1. Participants also rated the extent to which they thought each
of these abilities was desirable on a scale ranging from 1 (not at all
desirable)to9(extremely desirable). The order of these two sets of
measures was counterbalanced.
The experimental task was then introduced. Participants were given a
short test of 10 GRE analytical items taken from a GRE test preparation
guide (Educational Testing Service, 1996). Those in the abstract reasoning
group were told in written instructions that “the logic necessary to under-
stand abstract concepts generally makes more sense to those who score
highly on this test.” Those in the computer programming condition were
told “the logic of how to build a computer program generally makes more
sense to those who score highly on the test.” The test was exactly identical
for both conditions except for what it was said to measure and a large label
on the front page reading either “Abstract Reasoning Ability Exam” or
“Computer Programming Ability Exam.”
After completing the exam, participants provided percentile and raw-
score estimates of performance, using measures similar to those in Study 1.
8
EHRLINGER AND DUNNING

They also estimated how many questions they believed Cornell University
students in the experiment correctly answered on average.
Results and Discussion
Self-views. As expected, participants overall rated their ab-
stract reasoning ability more highly (M ⫽ 6.6) than their computer
programming skills (M ⫽ 2.7), t(90) ⫽ 16.38, p ⬍ .0001. Partic-
ipants also rated the ability to think abstractly as more desirable
(M ⫽ 7.2) than computer skills (M ⫽ 5.6), t(90) ⫽ 7.02, p ⬍
.0001.
Performance estimates. These self-views mattered when it
came to participants’ performance estimates. Across participants,
the self-view considered relevant to the task (ratings of abstract
reasoning in the abstract reasoning condition and of computer
ability in the computer programming condition) predicted both raw
(partial r ⫽ .33, p ⬍ .005) and percentile (partial r ⫽ .44, p ⬍
.0001) estimates of performance even after controlling for the
actual performance obtained. Thus, it was not surprising that
estimates of performance differed across the abstract reasoning
and computer programming conditions. When providing percentile
estimates of their performance on the test, abstract reasoning
participants (those who had been told that the test was a measure
of abstract reasoning) rated themselves more highly (M ⫽ 70.8)
than did their computer programming counterparts (M ⫽ 58.4),
t(89) ⫽ 2.80, p ⬍ .01. They also estimated that they had achieved
a higher raw score (M ⫽ 7.7) than their computer programming
peers (M ⫽ 6.7), t(89) ⫽ 2.34, p ⬍ .05. Of importance, these
different perceptions arose in the absence of a difference in actual
performance on the test (Ms ⫽ 8.2 and 7.9 for the abstract
reasoning and computer programming conditions, respectively),
t(89) ⫽ 0.81, ns. In addition, we created a single index of perfor-
mance estimates by standardizing participants’ percentile and raw
score estimates and then averaging them into one overall measure.
A multiple regression predicting this index from condition and
actual performance showed a significant effect of condition inde-
pendent of actual score, F(1, 88) ⫽ 7.42, p ⬍ .001. Participants
across conditions also did not differ in how many items they
estimated the average Cornell University student in the experiment
would get right (Ms in both conditions ⫽ 6.9).
Mediational analyses. In sum, despite equivalent perfor-
mance, participants in the abstract reasoning group thought they
had done much better than their counterparts in the computer
programming condition. We conducted a mediational analysis,
using the series of tests suggested by Kenny, Kashy, and Bolger
(1998), to affirm that these different performance estimates arose
because of a difference in the self-views participants had brought
to bear in their performance estimates. We have already estab-
lished that the condition to which participants were assigned
influenced the proposed mediator, namely the favorability of the
self-views they thought were relevant to the test (
␤
⫽ .67, p ⬍
.0001). We have also shown that the condition was related to the
key dependent measure, our overall index of performance esti-
mates (
␤
⫽ .22, p ⬍ .05). A final test revealed that self-views of
ability continued to predict performance estimates even after con-
trolling for the condition to which participants had been assigned
(
␤
⫽ .39, p ⬍ .005). The impact of condition on performance
estimates was significantly reduced by a Sobel test (z ⫽ 3.04, p ⬍
.005), and was, in fact, nonsignificant (
␤
⫽⫺.04, ns).
Summary. In sum, Study 2 provided more conclusive evidence
that people rely on self-views to determine how well they have
performed on a task. In the study, we manipulated whether a high
or low self-view was relevant on a test, and that manipulation had
a significant impact on how people perceived their performance
without impacting their actual performance. Mediational analyses
provided further support that participants differed in their perfor-
mance estimates as a direct result of the manipulation changing the
self-view considered relevant.
Study 3: Does Altering the Self-View
Alter Performance Estimates?
Study 3 was designed to further demonstrate that people rely on
self-views when estimating their performance, independent of
actual performance. We manipulated whether participants had a
high or low opinion of their proficiency in a single domain,
knowledge of U.S. geography. We then gave them a test of
geographical knowledge and investigated whether our manipula-
tion of self-views altered participants’ estimates of performance.
Our double-barreled manipulation of self-views for Study 3 was
inspired by two manipulations found in previous research. First,
Salancik and Conway (1975; for a similar design, see Chaiken &
Baldwin, 1981) found that participants could be led to believe they
were more or less religious by the way they were asked about their
previous religious behaviors. Participants given items for which it
is easy to give the “religious” response (e.g., “Do you occasionally
go to church?”) afterwards described themselves as more religious
than did participants given items for which it is hard to provide the
religious response (e.g., “Do you go to church every week?”).
Second, Schwarz et al. (1991) found that they could alter partici-
pants’ self-views by varying the amount of behavioral evidence
they asked participants to provide about those views. For example,
after asking participants to provide 12 examples of their assertive
behaviors, participants rated themselves as less assertive than did
those who were asked to list only 4 examples.
In our study, we used both techniques to influence participants’
views of their knowledge of U.S. geography. In one part of the
manipulation, we asked participants whether they had visited a
series of locations that we had determined, through pretesting,
were likely or unlikely locations for our participants to have
visited. We also asked them a series of questions that placed their
knowledge of geography in a favorable light (e.g., could they
name 1, 2–3, or more than 3 Canadian provinces) or an unfavor-
able one (e.g., could they name 1–5, 6–10, or more than 10
Canadian provinces). One group of participants received a ques-
tionnaire that was designed to give them a favorable impression of
their geographical knowledge, in that they were asked whether
they visited locations that in all likelihood they had, and also asked
questions that placed their knowledge in a favorable light. The
other group received a negative questionnaire that asked about
locations they had likely not visited and presented them with
questions that placed their geographical knowledge in some
disrepute.
After this self-view manipulation, participants in both condi-
tions were given the exact same geography quiz. They were given
a blank map of North America and asked to place 15 U.S. cities.
Independent of actual performance on this test, we expected par-
ticipants who had received a positive self-view manipulation
9
SELF-VIEWS AND PERFORMANCE ESTIMATES

would provide more favorable performance estimates than those
who received a negative manipulation. In addition, we predicted
that these differences in performance estimates would be mediated
by the manipulated favorability of self-views.
Method
Participants. Participants were 55 (40 women, 15 men) Cornell Uni-
versity undergraduate students who took part in exchange for extra credit
toward their grades in large lecture psychology courses.
Self-view manipulation. We sought to manipulate self-view of U.S.
geographical knowledge by asking participants a series of questions pur-
portedly designed to see how much they had traveled and how much
geography they knew. The questions were designed to lead participants to
provide favorable responses about themselves in one condition and unfa-
vorable responses in the other. In the first part of the questionnaire,
participants were asked whether they had ever visited six given locations.
In the positive condition, the places were ones that, according to pretesting,
Cornell undergraduates were likely to have visited (i.e., New York City,
Connecticut, California, Pennsylvania, Florida, and Massachusetts). In the
negative condition, the locations were ones likely not to have been visited
(i.e., Mississippi, Missouri, Wyoming, North Dakota, Nebraska, and
Oregon).
Participants were next asked six questions about their knowledge, for
example, of state capitals and Canadian provinces. On each question, they
were asked to rate their knowledge on a 3-point scale—but the responses
indicated by each point in the scale differed across conditions. For exam-
ple, when asked how many Canadian provinces they could name, partici-
pants in the positive condition confronted a 3-point scale that read 1, 2–3,
and 4 or more. In the negative condition, the scale read 1–5, 6–10, and 11
or more.
Geography test. The geography test consisted of a blank map of the
United States that was drawn to scale. The map contained a small amount
of topological information (e.g., rivers, lakes) but contained no information
about state or national borders. Participants were given a list of 15 cities
and asked to indicate where each city was located on the map. The location
of Boston, Massachusetts was marked on the map to serve as an example,
and was surrounded by a circle of 2.1 cm in diameter indicating to
participants the range that would be considered correct for each city.
Participants were told that the circle corresponded to a 150-mile radius
around the true location of each city.
Procedure. Upon arriving at the laboratory, participants first com-
pleted the self-view manipulation questionnaire and then completed a
manipulation check, for which they rated their knowledge of U.S. geog-
raphy on a scale ranging from 1 (very weak)to10(very strong). Partici-
pants then completed the geography test and, finally, estimated how well
they had performed using the percentile and raw-score scales described in
previous studies.
Results and Discussion
Self-views. Manipulation checks showed that the positive and
negative surveys did influence responses. Participants reported
having visited 83% of the locations listed in the positive manipu-
lation survey and only 17% of those in the negative one,
t(52) ⫽ 6.83, p ⬍ .0001. On the six questions testing geographical
knowledge with use of a 3-point scale, the mean response of
participants in the positive condition was at 2.51 scale points,
whereas in the negative condition it was 1.32, t(53) ⫽ 9.68, p ⬍
.0001. These different responses then prompted participants to
report different self-impressions when it came to their knowledge
of U.S. geography (Ms ⫽ 6.0 and 4.4 for the positive and negative
groups, respectively), t(53) ⫽ 2.85, p ⬍ .01.
Performance estimates. Across the sample, self-views pre-
dicted both raw (partial r ⫽ .55, p ⬍ .005) and percentile (partial
r ⫽ .68, p ⬍ .0001) estimates of performance, controlling for
actual performance. Further, as predicted, participants in the pos-
itive condition evaluated their performance on the geography quiz
more favorably than did those in the negative condition. This was
evident in their percentile estimates (Ms ⫽ 62 and 44 for positive
and negative conditions, respectively), t(52) ⫽ 2.37, p ⬍ .05, as
well as in their estimates of raw score (Ms ⫽ 9.4 and 6.7, for
positive and negative conditions, respectively), t(52) ⫽ 3.00, p ⬍
.005.
These significant differences in performance estimates arose in
the absence of a comparable difference in actual performance.
Although we found that positive-manipulation participants an-
swered a greater number of questions correctly than their negative-
manipulation counterparts (Ms ⫽ 8.6 and 7.1, respectively), this
difference achieved only marginal significance, t(52) ⫽ 1.86, p ⬍
.10. Further, performance estimates remained significantly differ-
ent across the two conditions even after controlling for differences
in actual performance. As in Study 2, we collapsed percentile and
raw score estimates into a single performance estimate index. A
multiple regression predicting this index from condition and actual
performance showed a significant effect of condition independent
of actual score, F(1, 52) ⫽ 4.24, p ⬍ .05.
Mediational analysis. We performed a mediational analysis to
assess whether the self-view manipulation influenced performance
estimates by virtue of its impact on self-views of geography
knowledge. The combined performance estimate index, as com-
puted above, was significantly influenced by our manipulation
(
␤
⫽ .23, p ⬍ .05). In addition, the self-view manipulation had a
significant impact on the self-ratings of knowledge of geography
(
␤
⫽ .24, p ⬍ .05). Of key importance, self-rated knowledge of
geography strongly predicted how participants would estimate
their performance on the geography test (
␤
⫽ .70, p ⬍ .0001),
even after controlling for the impact of the self-view manipulation.
After controlling for self-views, the impact of the manipulation fell
significantly (z ⫽ 2.19, p ⬍ .05) and, indeed, was nonsignificant
(
␤
⫽⫺.06).
Summary. The data from Study 3 showed that altering an
individual’s self-view had a measurable impact on his or her
performance estimates that operated independently of how well
that person had actually performed. Participants manipulated to
hold a more positive view of their knowledge of U.S. geography,
relative to those manipulated toward a more negative self-view,
estimated that they had performed better on a geography test. A
mediational analysis showed that the impact of the manipulation
on performance estimates was a result of changed self-views.
Study 4: Do Differing Self-Views Underlie Gender
Differences in the Pursuit of Science?
Study 4 was designed to fulfill a two-fold purpose. The first
purpose dealt with a minor methodological point. In the first three
studies, we measured or altered participants’ self-views immedi-
ately before we asked them to take a test and estimate how well
they had done. Perhaps self-views influenced later performance
estimates only by virtue of having been brought to mind minutes
before performance estimates were elicited. Thus, in Study 4, we
examined whether performance estimates were related to chronic
10
EHRLINGER AND DUNNING

self-views measured weeks before participants faced the relevant
test.
The second purpose was more extensive in scope, to see whether
performance estimates influenced by chronic self-views carry be-
havioral consequences. In particular, we wanted to examine the
relevance of our analysis of performance estimates for a significant
and enduring issue in social life—the persistent gender gap in
those who pursue scientific careers. Despite recent improvements,
there exists a discernible gender difference in participation in
scientific and engineering careers (National Research Council,
1991). From early on, girls and women are less enthusiastic about
taking science courses than are their male counterparts (DeBacker
& Nelson, 2000; Farenga & Joyce, 1999). According to a recent
National Science Foundation (2000) report, women consistently
account for a minority of the bachelor’s degrees awarded in
chemistry (43%), computer science (28%), chemical engineering
(32%), and earth sciences (35%)—accounting for less than 20% of
those awarded in physics as well as electrical, aerospace, and
mechanical engineering. This gender gap increases as people get
older, with a smaller percentage of women earning postgraduate
degrees than those earning their bachelor’s. Women make up only
22% of the science and engineering labor force, despite making up
46% of the labor force overall. These differences arise despite no
apparent difference in true ability (Seymour, 1992a, 1992b).
Researchers from a variety of perspectives have made contribu-
tions toward understanding this gender disparity, focusing, for
example, on the influence of cultural norms and a lack of encour-
agement by teachers and family (Fox, Benbow, & Perkins, 1983),
as well as the dual demands placed on women to start families as
well as engage in careers (Tittle, 1986). It has been demonstrated
that women are less likely to value science than men (Weinburgh,
1995), and that they think less of their science and mathematical
abilities (Eccles, 1987). Eccles (1994) suggests that these factors
lead women to avoid scientific and engineering careers.
We speculated that our analysis of performance estimates might,
in part, explain the relative reluctance of women to pursue scien-
tific activities. Specifically, because women tend to hold more
negative self-views about their scientific ability, they may mises-
timate how well they perform on science-related tasks. As a
consequence, they decide to avoid scientific activities even when
they show just as much proficiency as their male counterparts. We
decided to explore this by, first, surveying male and female college
students on their self-views of scientific ability, and then inviting
them to the laboratory where they take a science quiz. We pre-
dicted that women would hold more negative views of their
scientific ability relative to men, and that this would lead them to
estimate their performance on the quiz more negatively than would
men. The relatively negative impressions of performance on the
part of women would be mediated by the self-views reported on
our preliminary survey.
After taking the quiz, we invited male and female participants to
take part in a science competition taking place sometime in the
future. We expected that women would decline this invitation
more frequently than the men, and this reluctance to take part
would be predicted more strongly by the perception they held of
their performance on the science quiz than by the reality of their
performance. In short, in linking gender to differential interest in
scientific activities, we proposed a cascade. Women, holding more
negative self-views of scientific ability than men, would provide
more negative estimates of their performance on a test, which
would then lead to more disinterest in taking part in a science
competition. This cascade would take place even if male and
female participants did not differ in actual performance.
Method
Participants. Participants were 119 (75 women, 44 men) Cornell Uni-
versity students who took part in exchange for extra credit in their large
lecture psychology courses.
Materials and procedure. We asked several hundred students in large
lecture psychology courses to rate their ability in a number of domains,
including their ability to “reason about science,” on a scale ranging from 1
(I do not possess this ability at all)to9(I possess this ability to an extreme
degree). Several weeks after answering this question, we e-mailed an
invitation to participate in our study, with little description of the study’s
focus. A total of 116 students responded and participated in the study. Data
from an additional 3 participants were excluded because of suspicion.
Upon arriving at the laboratory, participants were given a test consisting
of 10 scientific reasoning items taken from an American College Test
(ACT) preparation guide (American College Test, 1999). The test required
participants to interpret scientific passages, as well as tables and graphs, to
answer multiple-choice questions. As in past studies, after completing the
test, participants estimated their performance by providing percentile ranks
and estimates of their raw score. They also reported the extent to which
they thought scientific reasoning ability was desirable on a scale ranging
from 1 (not at all)to9(extremely desirable).
Participants were next given a flyer advertising a Science Jeopardy
contest being organized by several science departments on campus in
collaboration with the psychology department. Prizes were said to include
$200 in cash and a free dinner for two at a local expensive restaurant.
Participants were asked three questions with respect to this contest. They
were asked whether they would like to receive an e-mail with more
information, like to sign up for the contest, or be added to an e-mail list that
would ask weekly science jeopardy questions for smaller prizes. Partici-
pants’ responses were coded as to whether they expressed interest in the
context on any of these three queries. If a participant responded “yes” to
any of the questions, he or she was categorized as having some interest in
the contest. If a participant answered “no” to all questions, he or she was
coded as uninterested.
Results and Discussion
Gender and self-views. As predicted, women harbored more
negative views of their scientific ability than did men (Ms ⫽ 6.5
and 7.6, respectively), t(114) ⫽⫺3.95, p ⬍ .0001. Women, to a
marginal degree, also considered scientific skills to be less desir-
able than did men (Ms ⫽ 6.5 and 7.2, respectively), t(114) ⫽ 1.69,
p ⬍ .10.
Performance estimates. Also as expected, women evaluated
their performance on the science quiz more negatively than men
did, both on the percentile measure (Ms ⫽ 56.0 and 73.4, respec-
tively), t(114) ⫽⫺3.82, p ⬍ .0005, and on the raw-score estimate
(Ms ⫽ 5.8 and 7.1), t(114) ⫽⫺2.83, p ⬍ .01. These differential
estimates arose even though women scored just as well on the test
as men (Ms ⫽ 7.5 and 7.9, respectively), t(114) ⫽⫺1.03, ns. We
should note that both men and women underestimated their raw
score on the test—but women did so more to a significant degree,
t(114) ⫽ 2.83, p ⬍ .01.
Multiple regressions showed that, for men and women together,
self-views predicted estimates of both raw (partial r ⫽ .51, p ⬍
.0001) and percentile (partial r ⫽ .49, p ⬍ .0001) score, indepen-
11
SELF-VIEWS AND PERFORMANCE ESTIMATES

dently of actual performance. A mediational analysis showed that
male and female participants provided different performance esti-
mates because they walked into the laboratory with different views
of their scientific ability. To conduct the mediational analysis, we
first dummy-coded gender (women ⫽ 1, men ⫽ 0) and constructed
the combined performance estimate index described in previous
studies. We then conducted regression analyses that showed that
(a) gender predicted performance estimates (
␤
⫽⫺.15, p ⬍ .05)
and (b) gender predicted the self-views of scientific ability mea-
sured weeks before participants arrived in the laboratory (
␤
⫽
⫺.27, p ⬍ .001). Finally, even after controlling for the influence of
gender, self-views still predicted performance estimates (
␤
⫽ .22,
p ⬍ .005), suggesting that these views mediated the link between
gender and estimates of performance on the science quiz. Further
affirming that conclusion, after controlling for self-views, the
impact of gender on performance estimates was significantly re-
duced (z ⫽⫺3.34, p ⬍ .001) and, indeed, was no longer signif-
icant (
␤
⫽⫺.09, ns).
Interest in scientific activities. Women also were less enthu-
siastic about signing up for the science jeopardy competitions.
Although 71% of the men showed some interest in signing up, only
49% of the women did (z ⫽ 2.26, p ⬍ .025). This lack of interest
among women was linked more to their perceptions of how well
they had done on the quiz than it was to reality. In a series of
simple logistic regression analyses, we found that signing up for a
science competition was related to perceptions of performance, as
measured by the combined performance estimate index (b ⫽ .67,
p ⬍ .005), but not to actual performance (b ⫽ .07, ns).
Further analyses revealed that perceptions of performance on
the science quiz were the crucial link connecting gender to lack of
interest. In a multiple logistic regression analysis, we examined
how well interest in the science competitions was predicted from
the combined performance estimate index, actual score on the test,
ratings of the desirability of science ability, participant gender, and
prerating of scientific ability. Only perceptions of performance
predicted interest (b ⫽ .99, p ⬍ .05), and controlling for percep-
tions of performance caused the relationship between gender and
interest to become nonsignificant (b ⫽⫺.38, ns). Actual perfor-
mance and perceptions of desirability did not significantly predict
interest in further scientific activities.
Linking gender to interest in scientific activities: A path model.
Our theoretical framework suggests a specific path model linking
gender to interest in the science jeopardy contest. Gender is linked
to chronic self-views of scientific ability, which affects perceptions
of performance on the science quiz, which in turn influences
interest in the science competition. We assessed this account with
the structural equation modeling (SEM) program within the
AMOS procedure (Arbuckle & Wothke, 1999). Five measures
were included. Four were suggested by our theoretical framework:
gender, chronic self-views of scientific ability, perceptions of
performance (indexed by the combined raw score and percentile
measure described above), and interest in the science jeopardy
contest (a binary variable indexed, as above, by whether partici-
pants indicated any interest in participating on any of the three
chances we gave them). We also included actual performance on
the science quiz.
The model is depicted in the left half of Figure 1. In the model,
the cascade from gender to self-views to performance estimates to
interest in the scientific contest is depicted. Actual performance is
also allowed to predict performance estimates. As can be seen in
the figure, all paths were significant (Zs ⬎ 2.4, all ps ⬍ .01). In
fact, the model fit the data rather well (comparative fit index ⫽
.99). Adding a direct link from gender to interest in the science
competition did not significantly increase the fit to the data,
␹
2
(1,
Figure 1. Two path models linking gender to interest in the science jeopardy contest. The model on the
left-hand side is the predicted model in which gender is linked to interest through perceived performance on the
science quiz. The alternative model of the right-hand side connects gender to interest through actual perfor-
mance. All correlations marked with an asterisk are significant at p ⬍ .005.
12
EHRLINGER AND DUNNING

N ⫽ 119) ⫽ 2.09, ns, suggesting that the cascade accounted for the
link between gender and interest.
This model also fit the data better than an alternative model,
depicted in the right half of Figure 1. In this model, the cascade
from gender to interest in the science jeopardy competition runs
through actual performance, not perceived performance. This
model,
␹
2
(6, N ⫽ 119) ⫽ 59.07, did not fit the data as well as the
one above,
␹
2
(6, N ⫽ 119) ⫽ 20.14, which is not surprising given
that actual performance was not significantly related to interest in
the science competition.
2,3
Summary. In sum, for women, lack of interest in pursuing
science-related activities could be traced to different perceptions of
how well they had performed on a science test. Women provided
much more pessimistic self-estimates than men, even though
women performed just as well as men. These less favorable esti-
mates could, in turn, be traced back to the lowered self-views of
scientific ability that women brought to the laboratory. This pattern
of data suggests, by extension, that women might disproportion-
ately avoid scientific pursuits because their self-views lead them to
mischaracterize how well they are objectively doing on any given
scientific task. Because they think they are doing more poorly than
do men, they are more likely than men to avoid science when given
an option.
General Discussion
We began this article by noting how perceptions of performance
are often loosely tethered to actual performance, and sought an
explanation of what leads these perceptions astray from actual
performance. Four studies suggested that a significant source of
performance estimates were participants’ chronic views of their
ability within the relevant domain. Independent of actual perfor-
mance, participants with high opinions of their ability thought they
had done well on any given test; those with unfavorable self-views
thought they had done poorly. Study 1 demonstrated this pattern by
showing that previously held self-views of logical reasoning abil-
ity predicted performance estimates as strongly, and sometimes
more strongly, than did actual performance on a logic test.
Study 2 went further to reveal the causal role played by these
self-views. When participants were told that a test measured ab-
stract reasoning skills—an ability they believed they had in abun-
dance—they thought they had done much better than when told
that the test assessed computer programming skills—an ability
they did not believe they had. These changes in performance
estimates arose even though participants confronted the exact same
test and achieved the same performance regardless of what the test
was said to measure. Study 3 supplied convergent evidence. In-
ducing participants to hold more or less favorable views of their
knowledge about U.S. geography caused them to raise or lower,
respectively, their estimate of how well they had done on a test of
geographic knowledge, independent of any actual change in per-
formance. Finally, Study 4 demonstrated how overreliance on
inaccurate self-views can have behavioral consequences. When
deciding whether to participate in a science competition, partici-
pants’ decisions were based on perceptions of their performance,
which were significantly linked to their chronic self-views, but not
to their actual performance.
Participants across our four experiments displayed a fair degree
of accuracy in predicting how well they have performed, yet the
influence that self-views imposed on performance estimates ex-
isted independently of any relationship to actual performance. In
fact, the relation of self-views to performance estimates, indepen-
dent of actual performance, was quite strong. Across the four
studies, self-views predicted estimates of one’s raw score on a test
as strongly as did actual raw score (mean
␤
s ⫽ .40 and .43 for
self-views and actual raw scores, respectively). Self-views pre-
dicted perceptions of one’s percentile standing more strongly than
actual percentile scores did (mean
␤
s ⫽ .55 and .17 for self-views
and actual percentile scores, respectively).
How Do Self-Views Influence Estimates?
A potentially valuable area for future research would be the
specific pathways by which chronic self-views influence perfor-
mance estimates. There seem to be several possible mechanisms
linking self-views to performance estimates, independent of actual
performance. For example, self-views may potentially serve as an
anchor on which people base their performance estimates, away
from which they adjust (insufficiently) to take into account any
specific experiences they have had with the task.
Or, as people review their experience with the task, their mem-
ories may be biased by their self-views. As research on memory
2
We also ran models comparing how well the cascade from gender to
interest flowed through the participants’ beliefs about the desirability of
having scientific talent rather than through perceptions of performance on
the test. These comparisons revealed that models involving desirability as
part of the cascade were not as good fits to the data compared with those
involving perceptions of performance. Desirability did marginally predict
interest in the science jeopardy contest (
␤
⫽ .19, p ⬍ .06), but did not do
so after performance perceptions were controlled for (
␤
⫽⫺.03, ns).
3
We also examined the potential role played by mood in the link
between gender and interest in scientific activities. Past research has shown
mood congruency in a wide variety of judgments (for reviews, see Clore,
Schwarz, & Conway, 1994; Forgas, 1995). Thus, it might be possible that
the mood of female participants was more negative than that of male
participants, and this difference carried over to performance estimates. We
collected the Positive and Negative Affect Schedule measure of positive
and negative affect (Watson, Clark, & Tellegen, 1988) after completion of
the exam to determine if facing a test in a domain for which one holds, for
example, a negative self-view creates a negative mood, which then leads
one to make negative performance estimates. In fact, there was no gender
difference in either positive or negative mood (ts ⬍ 1). Further, neither
positive nor negative mood correlated with estimates of performance (rs ⬍
.13). Thus, it seems transitory mood did not moderate the proposed path
between gender and performance evaluations in our data. On further
reflection, the lack of a mood effect in this domain is not surprising. Mood
tends to influence judgments of global life satisfaction, but is much less
common for judgments that are familiar and regarding specific domains
like the one we presented to participants (e.g., Levine, Wyer, & Schwarz,
1994; Schwarz et al., 1991, reviewed in Forgas, 1995).
A second alternative explanation worthy of attention concerns the role of
self-esteem, which has been shown to predict self-views (e.g., Dutton &
Brown, 1997), as well as performance evaluations (Shrauger & Terbovic,
1976). As such, it might be the case that a gender difference in self-esteem
led to the reported effects. The Rosenberg Self-Esteem Scale was included
within the larger pretest questionnaire from which we recruited participants
(Rosenberg, 1965). We matched self-esteem scores to 69 participants
(59%) in our sample. The present results occur in the absence of any gender
difference in self-esteem (t ⬍ 1) or correlation between level of self-esteem
and performance evaluations (r ⫽ .01, ns).
13
SELF-VIEWS AND PERFORMANCE ESTIMATES

has shown from the classic work of Carmichael, Hogan, and
Walter (1932) and Bartlett (1932), all the way to more recent work
(Halberstadt & Niedenthal, 2001), memories can be significantly
altered by the conceptual labels people apply to them or the
abstract expectations they have. A third possible mechanism might
be that the self-views provide an initial hypothesis of how one
might perform. Individuals commonly seek out information that
will confirm a favored hypothesis (e.g., Mynatt, Doherty, &
Tweeney, 1977; Wason, 1966, 1968) and, as such, interpret new
information as more supportive of a favored hypothesis than might
objectively be the case (Russo, Medvec, & Meloy, 1996; Russo,
Meloy, & Medvec, 1998). As a result, individuals will seem to
have considerable evidence supporting an estimate of performance
that is consistent with the preexisting self-view, even if one con-
sulting the same evidence in the absence of an initial hypothesis
would estimate performance quite differently. Future research
would look closer at the process through which self-views influ-
ence estimates of performance.
Appropriateness of Using Self-Views in Performance
Estimates
A careful reader at this point may wish to return to the issue of
whether it is normatively correct to rely on self-views when
forming performance estimates. It would seem plausible that using
self-views to estimate performance would be a quite useful strat-
egy, simply because such self-views would have been formed by
noting feedback one has received from past performances in a
domain. If past performances influence self-views, and also predict
current performance, then self-views would be a valuable predictor
of performance.
This analysis is correct, but it rests on two assumptions. The first
is that people receive unambiguous and unbiased feedback about
their performances. This assumption is not necessarily true in
every case. People might not always be in a position to receive
complete and unambiguous feedback about their failures and suc-
cesses (Einhorn, 1982). In addition, the feedback that other people
provide might be distorted by many different influences, such as
the reluctance to transmit bad news (Blumberg, 1972; Tesser &
Rosen, 1975).
Even when objective complete feedback is readily available, a
second assumption must be fulfilled—that people dispassionately
note and incorporate the feedback into their self-views. On this, the
literature contains a few difficulties. Feedback often has no mea-
surable effect on future predictions and later estimates of perfor-
mance (Fischer, 1982; Keren, 1988). Further, those cases in which
feedback has been shown to influence performance estimates are
quite circumscribed, such as for underconfident but not overcon-
fident judgments (Subbotin, 1996), when feedback is delivered by
computer rather than by a person (Zakay, 1992), or in the presence
of an explicit directive to consult feedback when estimating con-
fidence (Winman & Juslin, 1993).
Even with these difficulties, it is still possible that self-views
contain enough valid information about one’s abilities and capac-
ities to aid in providing accurate assessments. Thus, what is there
to say about the value of using self-views to inform performance
estimates? Answering this question is not a simple matter. Self-
views may have some value in informing performance estimates,
but they can also lead people astray.
First, the optimistic side. Consistent with past research, self-
views, on average, did significantly predict actual performance on
the tasks we gave to participants. As the top half of Table 1 shows,
chronic self-views on average, tended to significantly predict the
performances that participants attained (collapsed across condi-
tion, where relevant), measured both in raw score and percentile
terms (mean r ⫽ .27, weighted by the number of participants in
each study, for both performance estimate measures). The rela-
tionship between self-views and performance was modest in mag-
nitude for Studies 1, 2, and 4 (hovering between .00 and .39) and
quite strong for Study 3 (in excess of .60). Thus, overall, there was
some usefulness in considering self-views when predicting perfor-
mance (mean rs ⫽ .29 and .38, zs ⫽ 4.83 and 4.82, for raw-score
and percentile estimates, respectively, ps ⬍ .001).
However, there also is a pessimistic side. Although chronic
self-views accurately anticipated actual performance, such self-
views also led to systematic errors in performance appraisals.
Analyses we have described for each study already suggest this:
Holding actual performance constant, self-views still predicted
performance estimates, indicating that such self-views exerted
“unwarranted” influences that could lead to mistaken performance
estimates. A more conservative test, depicted in the bottom half of
Table 1, also shows that self-views were related to errors in
estimation (collapsed across conditions, where relevant). Table 1
shows how strongly errors in overestimation or underestimation
(estimated performance minus actual performance) were predicted
by chronic self-views. For Studies 2 and 4, self-views significantly
predicted errors in performance estimates (all ps ⬍ .05). The more
positive the self-view, the more participants overestimated their
performance; the more negative their self-views, the more they
underestimated them. Study 3 showed a similar, albeit nonsignif-
icant, trend. Study 1 was mixed. Across the four studies, this
pattern was significant (mean r ⫽ .29, Z ⫽ 5.68, p ⬍ .001 for
raw-score estimates, and mean r ⫽ .18, Z ⫽ 3.41, p ⬍ .001 for
percentile estimates, weighted by number of participants in each
study).
In addition, even when self-views correlate with objective per-
formance, they still might lead more to error than to accuracy.
They will do so when the self-views people hold overall are either
too high or too low relative to objective performance. If people, for
example, believe their skill on average hovers around the 70th
percentile, when the true average is the 50th percentile, then
Table 1
Relationship of Chronic Self-Views to Actual Performance and
Over- or Underestimation of Performance
Correlation between self-views and
Study
1 234
Actual performance
Raw score .34** .00 .66** .27**
Percentile score .31* .39** .64** .29**
Over-/underestimate of performance
Raw score .08 .30* .19 .44**
Percentile score ⫺.10 .31* .13 .23*
Note. For each study, the correlations represent the overall data, collaps-
ing across conditions.
* p ⬍ .05. ** p ⬍ .01.
14
EHRLINGER AND DUNNING

relying on those positive self-views will prompt people in general
to overestimate their performances, even if there is some under-
lying correlation between self-view and objective performance.
Indeed, a wealth of research has shown that people tend to hold
self-views that are too favorable given objective criteria, lending
credence to this concern. More people, for example, claim to have
above-average talents and capacities than is possible given the
logic of descriptive statistics (Alicke, 1985; Dunning, Meyerowitz,
& Holzberg, 1989; Kruger & Dunning, 1999; Weinstein, 1980).
An examination of these issues, as well as the psychological
literature in general, suggests that the value of self-views for
assessing performance will most likely depend on the domain
under consideration. In some domains, the correlation between
self-views and objective performance is more substantial than it is
in others. For athletics, the correlation between self-view and
performance tends to be higher (roughly .48) than it is for domains
involving interpersonal (roughly .17) or managerial skills (roughly
.04) (Mabe & West, 1982). Indeed, in this article, self-views
correlated with both actual performance and error in estimates, but
the strength of those correlations varied across the four domains
represented. In addition, the extent to which self-views overesti-
mate objective performance overall also varies by domain (Alicke,
1985; Dunning et al., 1989; Kruger, 1999; Weinstein, 1980).
Although psychological research has provided some pointers about
which domains generate more accurate self-assessments and which
generate more erroneous ones, the extant work is hardly compre-
hensive. Once a greater understanding of the domains in which
self-views tend to be accurate is achieved, more can be said about
exactly when one is well-advised to base judgments on said
self-views.
Women in Science
Perhaps the lesson of this article is most important when applied
to Study 4, which demonstrated the potential behavioral conse-
quences of the link between self-views and performance estimates.
Women, relative to men, walked into our laboratory with more
unfavorable beliefs about their level of talent at scientific reason-
ing. As a consequence, their performance estimates on a science
quiz were lower then those of men, even though in actuality they
did just as well. Furthermore, these lowered perceptions had con-
sequences. When asked whether they would like to enter into a
science competition, women were more likely than men to decline.
Further, this higher rate of refusal among women was linked to
their perceptions of their performance on the test we gave them,
not to the reality of their performance.
The findings of Study 4 highlight the importance of learning
how clear and unambiguous feedback might be incorporated to
improve the accuracy of self-views. Left to their own devices,
people can be inaccurate about how well they are doing, as were
our female participants with respect to the quiz we gave them.
Feedback from an outside agent might have gone a long way
toward bringing their perceptions into line, which could have
important behavioral consequences, but only if that feedback in-
fluences future perceptions and estimates. One wonders, for ex-
ample, if female participants in Study 4 would have expressed
more interest in the science competition if we had told them,
before they made the choice, how well they had actually done on
the quiz.
To be sure, the issue of why women disproportionately avoid
activities involving science is a complex one. There are undoubt-
edly many forces—psychological, sociological, and structural—
that cause women to avoid the small science competition we
suggested to our participants and to withdraw from careers in
science and engineering. Past research, for example, has high-
lighted the specific hurdles that women face when it comes to
science, such as discrimination as well as the difficulty of starting
a family life at the difficult first stages of one’s career. Research
has also shown how the threat of societal stereotypes leave women,
but not men, with special psychological hurdles that they have to
deal with and that might inhibit their performance (Spencer,
Steele, & Quinn, 1999; Steele, 1997).
That said, what Study 4 suggests is that even when women
perform just as well as men, psychological processes may still
prevent them from recognizing that fact. Perhaps as women tackle
engineering and scientific tasks, they come to evaluate their per-
formances more negatively than men do, even if their perfor-
mances are every bit as good as the men’s. Reviewing what
appears to be a more lackluster series of achievements, women
decide that it is best to pursue some other livelihood. Conversely,
perhaps men, convinced of the sterling quality of their achieve-
ments, decide to pursue a career at which they are not as uniquely
talented as they believe themselves to be.
Concluding Remarks
We are fairly certain that, with a few minutes of reflection, any
reader can generate examples of people whose perceptions of their
performances fails to match a plausible form of reality. Perhaps it
is that stiff and gangly dancer who prides himself on his grace, or
the brilliant student who is convinced that every single comment
she makes is banal. In this research, we sought to provide insight
into how such disparities between perception and reality are gen-
erated, and perhaps more importantly, why they are maintained.
Even when people desire to know themselves accurately, their
performance estimates may vary from the truth because those
estimates are importantly influenced by inaccurate self-views of
ability.
Thus, when gaining an accurate impression of one’s current
performance is important, it may be wise to assume that one does
not already know the answer but to seek out information from the
outside world—and then to pay it heed. In sum, it may pay to be
a little more like Ed Koch, making sure to ask the question
“How’m I doin’?” when accuracy is important, or at least not to
assume that one already knows the answer.
References
American College Test. (1997). Getting into the ACT: Official guide to the
ACT assessment. San Diego, CA: Harcourt Brace.
Alicke, M. D. (1985). Global self-evaluation as determined by the desir-
ability and controllability of trait adjectives. Journal of Personality and
Social Psychology, 49, 1621–1630.
Arbuckle, J. L., & Wothke, W. (1999). Amos 4.0 User’s Guide. Chicago,
IL: SmallWaters Corporation.
Bandura, A. (1977). Self-efficacy: Toward a unifying theory of behavioral
change. Psychological Review, 84, 191–215.
Bandura, A. (1997). Self-efficacy: The exercise of control. New York:
Freeman.
15
SELF-VIEWS AND PERFORMANCE ESTIMATES

Bartlett, F. C. (1932). Remembering: A study in experimental and social
psychology. New York: Macmillan.
Bass, B. M., & Yammarino, F. J. (1991). Congruence of self and others’
leadership ratings of naval officers for understanding successful perfor-
mance. Applied Psychology, 5, 437–454.
Bem, D. J., & Lord, C. G. (1979). Template matching: A proposal for
probing the ecological validity of experimental settings in social psy-
chology. Journal of Personality and Social Psychology, 37, 833–846.
Blumberg, H. H. (1972). Communication of interpersonal evaluations.
Journal of Personality and Social Psychology, 23, 157–162.
Carmichael, L. C., Hogan, H. P., & Walter, A. A. (1932). An experimental
study of the effect of language on the reproduction of visually perceived
form. Journal of Experimental Psychology, 15, 73–86.
Chaiken, S., & Baldwin, M. W. (1981). Affective-cognitive consistency
and the effect of salient behavioral information on the self-perception of
attitudes. Journal of Personality and Social Psychology, 41, 1–12.
Clore, G. L., Schwarz, N., Conway, M. (1994). Affective causes and
consequences of social information processing. In R. S. Wyer & T. K.
Srull (Eds.), Handbook of social cognition (2nd ed., Vol. 1, pp. 323–
419). Hillsdale, NJ: Erlbaum.
Crosby, R. A., & Yarber, W. L. (2001). Perceived versus actual knowledge
about correct condom use among U.S. adolescents: Results from a
national study. Journal of Adolescent Health, 28, 415–420.
DeBacker, T. K., & Nelson, R. M. (2000). Motivation to learn science:
Differences related to gender, class type, and ability. The Journal of
Educational Research, 93, 245–254.
DePaulo, B. M., Charlton, K., Cooper, H., Lindsay, J. J., & Muhlenbruck,
L. (1997). The accuracy-confidence correlation in the detection of de-
ception. Personality and Social Psychology Review, 1, 346–357.
Ditto, P. H., & Lopez, D. F. (1992). Motivated skepticism: Use of differ-
ential decision criteria for preferred and nonpreferred conclusions. Jour-
nal of Personality and Social Psychology, 63, 568–584.
Dunning, D. (1993). Words to live by: The self and definitions of social
concepts and categories. In J. Suls (Ed.), Psychological perspectives on
the self (Vol. 4, pp. 99–126). Hillsdale, NJ: Erlbaum.
Dunning, D. (2001). On the motives underlying social cognition. In N.
Schwarz & A. Tesser (Eds.), Blackwell handbook of social psychology,
Volume 1: Intraindividual processes (pp. 348–374). New York: Black-
well Publishers.
Dunning, D., Meyerowitz, J. A., & Holzberg, A. D. (1989). Ambiguity and
self-evaluation: The role of idiosyncratic trait definitions in self-serving
assessments of ability. Journal of Personality and Social Psychol-
ogy, 57, 1082–1090.
Dunning, D., & Story, A. L. (1991). Depression, realism, and the over-
confidence effect: Are the sadder wiser when predicting future actions
and events? Journal of Personality and Social Psychology, 61, 521–532.
Dutton, K. A., & Brown, J. D. (1997). Global self-esteem and specific
self-views as determinants of people’s reactions to success and failure.
Journal of Personality and Social Psychology, 73, 139–148.
Eccles, J. S. (1987). Gender roles and women’s achievement-related deci-
sions. Psychology of Women Quarterly, 11, 135–172.
Eccles, J. S. (1994). Understanding women’s educational and occupational
choices: Applying the Eccles et al. model of achievement-related
choices. Psychology of Women Quarterly, 18, 585–609.
Educational Testing Service. (1996). GRE: Practicing to take the general
test: Big book. Princeton, NJ: Educational Testing Service.
Ehrlinger, J., Johnson, K. L., & Dunning, D. (2002). Are the unskilled
really unaware? Further explorations of self-assessments among the
incompetent. Manuscript in preparation, Cornell University, Ithaca, New
York.
Einhorn, H. J. (1982). Learning from experience and suboptimal rules in
decision making. In D. Kahneman, P. Slovic, & A. Tversky (Eds.),
Judgment under uncertainty: Heuristics and biases (pp. 268–286). New
York: Cambridge University Press.
Fagot, B. I., & O’Brien, M. (1994). Activity level in young children:
Cross-age stability, situational influences, correlates with temperament,
and the perception of problem behaviors. Merrill Palmer Quarterly, 40,
378–398.
Farenga, S. J., & Joyce, B. A. (1999). Intentions of young students to enroll
in science courses in the future: An examination of gender differences.
Science Education, 83, 55–75.
Fischer, G. W. (1982). Scoring-rule feedback and the overconfidence
syndrome in subjective probability forecasting. Organizational Behavior
and Human Performance, 29, 352–369.
Forgas, J. P. (1995). Mood and judgment: The affect infusion model
(AIM). Psychological Bulletin, 117, 39–66.
Fox, L. H., Benbow, C. P., & Perkins, S. (1983). An accelerated mathe-
matics program for girls: A longitudinal evaluation. In C. P. Benbow &
J. C. Stanley (Eds.), Academic precocity: Aspects of its development (pp.
133–139). Baltimore, MD: Johns Hopkins University Press.
Halberstadt, J. B., & Niedenthal, P. M. (2001). Effects of emotion concepts
on perceptual memory for emotional expressions. Journal of Personality
and Social Psychology, 81, 587–598.
Harris, M. M., & Schaubroeck, J. (1988). A meta-analysis of self-
supervisor, self-peer, and peer-supervisor ratings. Personnel Psychol-
ogy, 41, 43–62.
Jussim, L., Coleman, L., & Nassau, S. (1987). The influence of self-esteem
on perceptions of performance and feedback. Social Psychology Quar-
terly, 50, 95–99.
Kenny, D. A., & DePaulo, B. M. (1993). Do people know how others view
them? An empirical and theoretical account. Psychological Bulletin,
114, 145–161.
Kenny, D. A., Kashy, D. A., & Bolger, N. (1998). Data analysis in social
psychology. In D. T. Gilbert, S. T. Fiske, & G. Lindzey (Eds.), The
handbook of social psychology (4th ed., Vol. 1, pp. 233–265). New
York: McGraw-Hill.
Keren, G. (1988). On the ability of monitoring non-veridical perceptions
and uncertain knowledge: Some calibration studies. Acta Psycho-
logica, 67, 95–119.
Kruger, J. M. (1999). Lake Wobegon be gone! The “below-average effect”
and the egocentric nature of comparative ability judgments. Journal of
Personality and Social Psychology, 77, 221–232.
Kruger, J. M., & Dunning, D. (1999). Unskilled and unaware of it: How
difficulties in recognizing one’s own incompetence lead to inflated
self-assessments. Journal of Personality and Social Psychology, 77,
1121–1134.
Kunda, Z. (1990). The case for motivated reasoning. Psychological Bul-
letin, 108, 480–498.
Kunkel, E. (1971). On the relationship between estimate of ability and
driver qualification. Psychologie und Praxis, 15, 73–80.
Levine, S. R., Wyer, R. S., & Schwarz, N. (1994). Are you what you feel?
The affective and cognitive determinants of self-judgments. European
Journal of Social Psychology, 24, 63–78.
Lewinsohn, P. M., Mischel, W., Chaplin, W., & Barton, R. (1980). Social
competence and depression: The role of illusory self-perceptions. Jour-
nal of Abnormal Psychology, 89, 203–212.
Lindeman, M., Sundvik, L., & Rouhiainen, P. (1995). Under- or overesti-
mation of self? Person variables and self-assessment accuracy in work
settings. Journal of Social Behavior and Personality, 10, 123–134.
Mabe, P. A., III, & West, S. G. (1982). Validity of self-evaluation of
ability: A review and meta-analysis. Journal of Applied Psychology, 67,
280–296.
Maki, R. H., Jonas, D., & Kallod, M. (1994). The relationship between
comprehension and metacomprehension ability. Psychonomic Bulletin
and Review, 1, 126–129.
Marteau, T. M., Johnston, M., Wynne, G., & Evans, T. R. (1989). Cogni-
tive factors in the explanation of the mismatch between confidence and
16
EHRLINGER AND DUNNING

competence in performing basic life support. Psychology and Health, 3,
173–182.
McFarland, C., Ross, M., & DeCourville, N. (1989). Women’s theories of
menstruation and biases in recall of menstrual symptoms. Journal of
Personality and Social Psychology, 57, 522–531.
McFarlin, D. B., & Blascovich, J. (1981). Effects of self-esteem and
performance feedback on future affective preferences and cognitive
expectations. Journal of Personality and Social Psychology, 40, 521–
531.
Metcalfe, J., & Shimamura, A. P. (Eds.). (1994). Metacognition. Cam-
bridge, MA: MIT Press.
Miller, D. T., & Ross, M. (1985). Self-serving biases in the attribution of
causality: Fact or fiction? Psychological Bulletin, 82, 213–225.
Moreland, R., Miller, J., & Laucka, F. (1981). Academic achievement and
self-evaluations of academic performance. Journal of Educational Psy-
chology, 73, 335–344.
Mynatt, C. R., Doherty, M. E., & Tweeney, R. D. (1977). Confirmation
bias in a simulated research environment: An experimental study of
scientific inference. Quarterly Journal of Experimental Psychology, 29,
85–95.
National Research Council: Committee on Women in Science and Engi-
neering. (1991). Women in science and engineering: Increasing their
numbers in the 1990s. Washington, DC: National Academy Press.
National Science Foundation. (2000). Women, minorities, and persons with
disabilities in science and engineering. Arlington, VA: National Science
Foundation.
Orton, P. Z. (1993). Cliffs law school admission test preparation guide.
Lincoln, NE: Cliffs Notes Inc.
Riggio, R. E., Widaman, K. F., & Friedman, H. S. (1985). Actual and
perceived emotional sending and personality correlates. Journal of Non-
verbal Behavior, 9, 69–83.
Risucci, D. A., Tortolani, A. J., & Ward, R. J. (1989). Ratings of surgical
residents by self, supervisors, and peers. Surgical and Gynecological
Obstetrics, 169, 519–526.
Rosenberg, M. (1965). Society and the adolescent self-image. Princeton,
NJ: Princeton University Press.
Russo, J. E., Medvec, V. H., & Meloy, M. G. (1996). The distortion of
information during decisions. Organizational Behavior and Human De-
cision Processes, 66, 102–110.
Russo, J. E., Meloy, M. G., & Medvec, V. H. (1998). Predecisional
distortion of product information. Journal of Marketing Research, 35,
438–452.
Salancik, G., & Conway, M. (1975). Attitude inference from salient and
relevant cognitive content about behavior. Journal of Personality and
Social Psychology, 32, 829–840.
Schwarz, N., Bless, H., Strack, F., Klumpp, G., Rittenauer-Schatka, H., &
Simons, A. (1991). Ease of retrieval as information: Another look at the
availability heuristic. Journal of Personality and Social Psychology, 61,
195–202.
Seymour, E. (1992a). “The problem iceberg” in science, mathematics, and
engineering education: Student explanations for high attrition rates.
Journal of College Science Teaching, 21, 230–232.
Seymour, E. (1992b). Undergraduate problems with teaching and advising
in SME majors: Explaining gender differences in attrition rates. Journal
of College Science Teaching, 21, 284–292.
Shaughnessy, J. J. (1979). Confidence judgment accuracy as a predictor of
test performance. Journal of Research in Personality, 13, 505–514.
Shrauger, J. S., & Terbovic, M. L. (1976). Self-evaluation and assessments
of performance by self and others. Journal of Consulting and Clinical
Psychology, 44, 564–572.
Sinkavich, F. J. (1995). Performance and metamemory: Do students know
what they don’t know? Instructional Psychology, 22, 77–87.
Spencer, S. J., Steele, C. M., & Quinn, D. M. (1999). Stereotype threat and
women’s math performance. Journal of Experimental Social Psychol-
ogy, 35, 4–28.
Sporer, S. L., Penrod, S., Read, D., & Cutler, B. (1995). Choosing,
confidence, and accuracy: A meta-analysis of the confidence–accuracy
relation in eyewitness identification studies. Psychological Bulletin, 118,
315–327.
Steele, C. M. (1997). A threat in the air: How stereotypes shape intellectual
identity and performance. American Psychologist, 52, 613–629.
Story, A. L. (1998). Self-esteem and memory for favorable and unfavor-
able personality feedback. Personality and Social Psychology Bulle-
tin, 24, 51–64.
Subbotin, V. (1996). Outcome feedback effects on under- and overconfi-
dent judgments (general knowledge tasks). Organizational Behavior and
Human Decision Processes, 66, 268–276.
Tesser, A., & Rosen, S. (1975). The reluctance to transmit bad news. In L.
Berkowitz (Ed.), Advances in experimental social psychology (Vol. 8,
pp. 193–232). New York: Academic Press.
Tittle, C. K. (1986). Gender research and education. American Psycholo-
gist, 41, 1161–1168.
Tracey, J. M., Arroll, B., Richmond, D. E., & Barham, P. M. (1997). The
validity of general practitioners’ self assessment of knowledge. British
Medical Journal, 315, 1426–1428.
Wason, P. (1966). Reasoning. In B. Foss (Ed.), New horizons in psychology
(pp. 135–151). London: Penguin.
Wason, P. (1968). Reasoning about a rule. Quarterly Journal of Experi-
mental Psychology, 20, 273–281.
Watson, D., Clark, L. A., & Tellegen, A. (1988). Development and vali-
dation of the brief measures of positive and negative affect: The PANAS
scales. Journal of Personality and Social Psychology, 54, 1063–1070.
Weinburgh, M. (1995). Gender differences in student attitudes toward
science: A meta-analysis of the literature from 1970–1991. Journal of
Research in Science Teaching, 32, 387–398.
Weinstein, N. D. (1980). Unrealistic optimism about future life events.
Journal of Personality and Social Psychology, 58, 806–820.
Winman, A., & Juslin, P. (1993). Calibration of sensory and cognitive
judgments: Two different accounts. Scandinavian Journal of Psychol-
ogy, 34, 135–148.
Yzerbyt, V., Lories, G., & Dardenne, B. (Eds.). (1998). Metacognition:
Cognitive and social dimensions. London: Sage.
Zakay, D. (1992). The influence of computerized feedback on overconfi-
dence in knowledge. Behaviour and Information Technology, 11, 329–
333.
Received January 17, 2002
Revision received July 22, 2002
Accepted July 24, 2002 䡲
17
SELF-VIEWS AND PERFORMANCE ESTIMATES