ArticlePDF Available

How Chronic Self-Views Influence (and Mislead) Self-Assessments of Task Performance: Self-Views Shape Bottom-Up Experiences With the Task


Abstract and Figures

Self-assessments of task performance can draw on both top-down sources of information (preconceived notions about one's ability at the task) and bottom-up cues (one's concrete experience with the task itself). Past research has suggested that top-down self-views can mislead performance evaluations but has yet to specify the exact psychological mechanisms that produce this influence. Across 4 experiments, the authors tested the hypothesis that self-views influence performance evaluations by first shaping perceptions of bottom-up experiences with the task, which in turn inform performance evaluations. Consistent with this hypothesis, a relevant top-down belief influenced performance estimates only when learned of before, but not after, completing a task (Study 1), and measures of bottom-up experience were found to mediate the link between top-down beliefs about one's abilities and performance evaluations (Studies 2-4). Furthermore, perception of an objectively definable bottom-up cue (i.e., time it takes to solve a problem) was better predicted by a relevant self-view than the actual passage of time.
Content may be subject to copyright.
How Chronic Self-Views Influence (and Mislead) Self-Assessments
of Task Performance: Self-Views Shape Bottom-Up Experiences
With the Task
Clayton R. Critcher and David Dunning
Cornell University
Self-assessments of task performance can draw on both top-down sources of information (preconceived
notions about one’s ability at the task) and bottom-up cues (one’s concrete experience with the task
itself). Past research has suggested that top-down self-views can mislead performance evaluations but has
yet to specify the exact psychological mechanisms that produce this influence. Across 4 experiments, the
authors tested the hypothesis that self-views influence performance evaluations by first shaping percep-
tions of bottom-up experiences with the task, which in turn inform performance evaluations. Consistent
with this hypothesis, a relevant top-down belief influenced performance estimates only when learned of
before, but not after, completing a task (Study 1), and measures of bottom-up experience were found to
mediate the link between top-down beliefs about one’s abilities and performance evaluations (Studies
2– 4). Furthermore, perception of an objectively definable bottom-up cue (i.e., time it takes to solve a
problem) was better predicted by a relevant self-view than the actual passage of time.
Keywords: self-views, expectations, bottom-up cues, self-assessment, performance evaluation
Despite spending more time with themselves than with any
other person, people often have surprisingly poor insight into their
skills and abilities (Dunning, 2005; Dunning, Heath, & Suls, 2004;
Harris & Schaubroeck, 1988; Mabe & West, 1982). Nurses’ esti-
mates of their basic life-support skills do not correlate with their
performance on objective skill tests (Marteau, Johnston, Wynne, &
Evans, 1989). Adolescent boys’ sense of their knowledge of con-
dom use correlates very modestly with their actual knowledge
(Crosby & Yarber, 2001). Doctors show similar poor self-insight
into their knowledge of various disorders (Tracey, Arroll, Rich-
mond, & Barham, 1997). Summing up the literature on the rela-
tionship between self-views and actual knowledge, Dunning
(2005) stated that there tends to be “a real relationship between
perception and reality, just not a very strong one” (p. 5; see also
Dunning et al., 2004).
Recently, Ehrlinger and Dunning (2003) offered one reason for
why self-assessments tend to be so modestly correlated with the
reality of performance. Estimates of performance are driven, at
least in part, by something else— by chronic self-views people
have about their abilities, preconceived notions about whether they
are skilled or unskilled at a task. If John, for example, thinks that
he has a good deal of logical reasoning ability, but Marvin does
not, John will assume he did better on the logical reasoning quiz
they both just completed than will Marvin, even if both performed
equally. Ehrlinger and Dunning discovered that these preconceived
notions of self often correlate equally strongly, and sometimes
more strongly, with people’s performance estimates than do their
actual performances.
To be sure, one might think this would be a useful way to
estimate one’s performance. What better way to estimate one’s
performance in today’s task than to refer to one’s beliefs about
one’s skill, based on a lifetime of experience, in the relevant
domain? The fly in the ointment of this analysis, however, is that
these chronic self-views tend to be only modestly correlated with
objective performance (for reviews, see Dunning, 2005; Dunning
et al., 2004; Mabe & West, 1982). Thus, although these views
provide a modicum of validity to performance estimates, relying
on them too much can lead to bias and error (Ehrlinger & Dunning,
Ehrlinger and Dunning (2003) provided several demonstrations
that people’s chronic self-views influence performance assess-
ments, irrespective of actual performance. Performance estimates
Clayton R. Critcher and David Dunning, Department of Psychology,
Cornell University.
This research was supported in part by a National Science Foundation
Graduate Research Fellowship awarded to Clayton R. Critcher, National
Institute of Mental Health Grant RO1 56072, and National Science Foun-
dation Grant 0745806, both awarded to David Dunning. We thank Jane
Risen for her comments. We thank Sally Apuzzo, Claire Chung, Latonia
Coryatt, Adi Kochavi, Sarah Mayefsky, Ryan Middleton, Nurul Mohamed,
Bernadette Park, and Jeanette Zambito for assistance with data collection.
Correspondence concerning this article should be addressed to Clayton
R. Critcher, Department of Psychology, Cornell University, 211 Uris Hall,
Ithaca, NY 14853. E-mail:
Journal of Personality and Social Psychology, 2009, Vol. 97, No. 6, 931–945
© 2009 American Psychological Association 0022-3514/09/$12.00 DOI: 10.1037/a0017452
on a logic quiz were equally or more correlated, depending on the
measure, with self-views than with actual performance. Changing
which self-view was putatively relevant to a task caused people to
provide different performance estimates, although actual perfor-
mance levels were unchanged. Inducing people to have more
positive or negative self-views about their knowledge of North
American geography prompted them to change their performance
estimates on a quiz that asked them to place famous cities (e.g.,
St. Louis, Missouri) on a North American map, although actual
performance was unaffected. Men and women, who walked into an
experiment with divergent ideas about their scientific talent, dif-
fered in how well they thought they did on a pop quiz on science,
although they did not differ in objective performance. Taken
together, these findings are consistent with prior research showing
that more global self-views (i.e., self-esteem) predict how people
evaluate their performance on specific tasks (Jussim, Coleman, &
Nassau, 1987; Lindeman, Sundvik, & Rouhiainen, 1995; Shrauger
& Terbovic, 1976).
But in all these demonstrations, Ehrlinger and Dunning (2003)
left unanswered two important questions. The first is how do these
preconceived self-notions influence performance estimates? What
are the psychological mechanisms that link chronic self-views to
estimates of today’s specific performance? What psychological
processes or mechanisms lie in the path between self-views and
performance estimates?
The second mystery is why the impact of top-down self-views is
so strong, given other strong influences in the environment that
should dampen or eliminate that impact. In particular, it is well
known in the psychological literature that people’s estimates of
performance are importantly driven by the bottom-up experiences
they have as they complete a task. People are more confident in
their performance, for example, if they come to answers quickly,
rather than slowly (Benjamin, & Bjork, 1996; Costermans, Lories,
& Ansay, 1992; Kelley & Lindsay, 1993; Reber & Schwarz, 1999;
Schwarz, 2004; Schwarz, Sanna, Skurnik, & Yoon, 2007). They
are also more confident if the terms appearing in questions and
answers feel familiar to them rather than novel (Arkes, Boehm, &
Xu, 1991; Schwartz & Metcalfe, 1992). Why do these bottom-up
experiences fail to reduce the impact of top-down self-beliefs?
In this article, we address these two mysteries by examining the
relationship between top-down self-beliefs, bottom-up experi-
ences, and performance estimates. We suggest that bottom-up
experiences may actually provide the link between self-views and
performance estimates. Ehrlinger and Dunning (2003) stated, “Per-
formance 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” (p. 6). In essence, Ehrlinger and Dunning speculated that
bottom-up cues and top-down self-views independently contribute
to performance evaluation and that bottom-up evidence is set
aside, in part, in favor of top-down notions. We, instead, assert that
the influence of top-down and bottom-up cues may not be inde-
pendent. Rather, top-down self-views influence performance esti-
mates because they first influence people’s experience and inter-
pretation of the bottom-up cues they encounter as they complete a
task, which then, in turn, informs their estimates about how well
they are doing. Bottom-up cues do not dilute the impact of top-
down self-views. Instead, they are the mediator responsible for the
influence of those top-down views. For example, if people think
they are skilled at logic, they will perceive the amount of time they
took to solve a logical brainteaser to be shorter than those who
think they are unskilled. We assert that these bottom-up experi-
ences are not different in reality, but merely in perception and
interpretation. People who think they are skilled perceive them-
selves as having an easier bottom-up experience with the task,
regardless of what objective measures may indicate.
There are many empirical demonstrations that top-down beliefs
influence the phenomenology of bottom-up experience. Yogurt
labeled as “full-fat” is rated as tastier than identical yogurt labeled
“low-fat” (Wardle & Solomons, 1994; see also Sanford, Fay,
Stewart, & Moxey, 2002). Meat labeled 75% lean tastes better than
meat with the semantically identical label of 25% fat (Levin &
Gaeth, 1988). A bottle of wine seems more pleasant, and activates
the medial orbitofrontal cortex more, when it is priced at $90 than
at $10 (Plassman, O’Doherty, Shiv, & Rangel, 2008). Attitudes
toward samples of cola (McClure et al., 2004), turkey (Makens,
1965), seltzer water (Nevid, 1981), and beer (Allison & Uhl, 1964)
were all assimilated toward the attitudes held toward these prod-
ucts’ brands, but only when the label was showing. Moving
beyond demonstrations of mere liking, Wansink, Park, Sonka, and
Morganosky (2000) found that a protein bar labeled as “soy”
begins to taste more grainy and less flavorful. Furthermore, bitter
coffee tastes milder if tasters are first misinformed that the coffee
is not actually bitter (Olson & Dover, 1978). Such effects arise at
the social level, too. People who dispositionally expect to see
interpersonal hierarchies were more likely to see a hierarchy
difference between two photographed people (Mast, 2005). Label-
ing a face as African American causes people to see its skin as
darker than when the face is labeled as European American (Levin
& Banaji, 2006). A push looks more aggressive if it comes from an
African American rather than a European American protagonist
(Sagar & Schofield, 1980).
Here, we assert that people’s top-down self-beliefs similarly
color the phenomenology and interpretation of bottom-up cues
while completing a task. To our knowledge, there have been no
empirical demonstrations of whether top-down self-beliefs influ-
ence people’s in vivo bottom-up experience and whether that
influence, in turn, shapes estimates of performance. The closest
study to have looked at such phenomena is that of Bunz, Curry,
and Voon (2007), who did not examine experience with an actual
task but did find that “perceived fluency” (i.e., perceived skill) at
computer tasks correlated with a report that one had “computer-
related anxiety.” Furthermore, this anxiety was not related to
actual ability. Bunz et al.’s purpose was not to measure bottom-up
experience, and in fact, the items on their anxiety scale seem to
measure a sense of ability (e.g., “I feel insecure about my ability
to interpret a computer printout”), rather than a bottom-up expe-
rience. Furthermore, this computer-related anxiety reflected a gen-
eral, decontextualized belief, not an assessment of actual experi-
ence with a computer task. Instead, we propose (and test) that
top-down beliefs color in vivo bottom-up experience. Also, Bunz
et al.’s study did not examine the second part of our hypothesis,
that these contaminated bottom-up experiences mediate the link
between self-views and performance estimates on a specific task.
Herein, we expressly focused on whether chronic self-views in-
fluenced one’s subjective experience as one sits down to actually
perform a task—and whether those altered experiences ultimately
led to different notions about how well one had done on a task.
Overview of the Present Studies
In four studies, we tested whether chronic self-views influence
performance evaluation because they influence the perceptions and
interpretations of the bottom-up experience people have with the
task. Study 1 explored this analysis by varying the timing of when
participants learned that a certain skill (e.g., computer program-
ming) was supposedly relevant to the task at hand. We hypothe-
sized that if participants learned about the skill before beginning
the task, then their self-beliefs of skill would color bottom-up
experiences and, thus, influence performance estimates. However,
if they learned about the skill after completing the task but before
providing performance evaluations, there would be no impact of
top-down self-views on performance estimates, given that their
bottom-up experiences had already been set. Study 2 provided a
more direct test by actually testing statistically whether bottom-up
experience mediates the link between self-views and performance
estimation. Studies 3 and 4 did the same but added more direct
tests of mediation by manipulating which self-view was suppos-
edly relevant to the task at hand.
Study 1
The goal of Study 1 was to provide an initial test of the idea that
top-down self-views influence performance evaluations through
their impact on bottom-up experience. The study extended and
replicated Ehrlinger and Dunning (2003, Study 2), in which par-
ticipants were told that a Graduate Record Examination (GRE)-
type test measured either abstract reasoning (a skill participants
tended to believe that they had in abundance) or computer-
programming skill (a skill participants did not think they possessed
to any positive degree). Ehrlinger and Dunning found that partic-
ipants provided more favorable evaluations of their own perfor-
mance (e.g., thought they got more questions right) when they
thought it measured abstract reasoning, rather than computer skill,
although there was no difference in actual performance.
Our extension involved varying exactly when participants were
told that the test focused on abstract reasoning versus computer-
programming skills. Some participants were informed before they
took the test, with the remainder being told after they had com-
pleted the test but before they evaluated their performance. If
top-down views influence performance estimates directly, then the
timing should not matter. However, because we suggest that top-
down views influence performance estimates indirectly through
participants’ bottom-up experiences, we predicted that the timing
would matter. Participants told before the test had a chance to have
their self-views influence their bottom-up experiences with the test
and, thus, their performance estimates. Those informed afterward
would already have concluded their bottom-up experiences with
the test, and thus, their performance evaluations should remain
There is empirical precedent for varying the timing of providing
top-down information to test whether top-down beliefs exert their
influence by acting through bottom-up perceptions. Learning that
balsamic vinegar had been added to a sample of beer (which
apparently sounds much worse than it tastes) contaminates one’s
enjoyment only when learning of this mystery ingredient before
sampling it, not afterward (Lee, Frederick, & Ariely, 2006). More
generally, when top-down views influence perceptual (bottom-up)
processing, the biasing view must be in place at the time of
exposure to the bottom-up, perceptual information (see von Hip-
pel, Sekaquaptewa, & Vargas, 1995).
Participants and Design
Forty-five undergraduates at Cornell University (Ithaca, New
York) participated in exchange for course extra credit. Participants
were randomly assigned to one of the cells in a 2 (Timing: pretest
or posttest) 2 (Skill Related: abstract reasoning vs. computer
programming) full-factorial design.
As participants arrived at the laboratory, we first assessed their
self-views in a number of domains, including abstract reasoning
and computer programming but also filler domains (e.g., emotional
intelligence, geography). Participants indicated their skill level on
percentile scales, ranging from 0 to 100, that asked them to
compare their skill with those of other students at Cornell Univer-
sity by noting the percentage of peers they thought they outper-
formed for each skill. Then, participants indicated for the same
domains “the extent to which you believe you possess the follow-
ing abilities or aptitudes” on 11-point Likert scales, anchored at 1
(not at all) and 11 (completely).
At this point, participants in the pretest timing condition learned
that they would be taking a test of abstract reasoning or computer-
programming ability. Those in the posttest timing condition were
simply told they would be taking an “ability test.” At this point,
participants took the 10-item logic games test used in Ehrlinger
and Dunning (2003, Study 2). At the end of the test, participants in
the posttest condition learned that they had just taken a test of
abstract reasoning or computer-programming ability. Finally, par-
ticipants estimated how many of the 10 items they believed they
answered correctly and at what percentile they believed their
specific performance fell relative to other students taking part in
the experiment.
The zero-order correlations between performance and the mea-
sured variables (for all four studies) are listed in Table 1.
Results and Discussion
As a check that self-views were more positive for abstract
reasoning than for computer-programming ability, we conducted
two 2 (Timing: pretest or posttest) 2 (Relevant skill: abstract
reasoning vs. computer programming) 2 (Skill Rated: abstract
reasoning vs. computer programming) mixed-model analyses of
variance, with the last factor measured within-subjects. For the
percentile scores, participants saw their own abstract reasoning
ability skills (M64.7, SE 2.4) as superior to their computer-
programming skills (M41.2, SE 4.9), F(1, 41) 26.62, p
.001. On Likert-scale responses, abstract reasoning skills (M
7.2, SE 0.29) were also seen to be superior to computer pro-
gramming skills (M4.7, SE 0.46), F(1, 41) 32.01, p
.001. None of the higher order effects involving condition ap-
proached significance (all Fs1), speaking to the success of
random assignment.
For the performance evaluations taken after completion of the
test, both raw score and percentile estimates were highly corre-
lated, r(42) .71, p.001. One participant’s performance
estimates fell more than seven standard deviations from the con-
dition mean; this outlier was excluded from all analyses reported
below. At this point, we standardized the two performance-estimate
items and summed them to create a performance-estimate composite
and submitted this composite to a two-way Timing Relevant Skill
analysis of covariance, with test label and timing as between-
subjects variables and actual test performance as a covariate.
Timing Skill interaction was significant, F(1, 38) 4.41, p
.04. As seen in Figure 1, the significant interaction reflected the
tendency for the relevant skill to affect performance estimates only
when participants learned of it before taking the test, t(38) 3.08,
p.004, but for it to have no effect when learned after the test
We then tested whether individual differences in participants’
self-views could predict performance estimates when participants
learned of the test label before the test but not after the test. Thus,
we regressed the performance composite estimate on the timing
condition, the relevant self-view (as measured by self-rating), the
irrelevant self-view (again, the self-rating), the Relevant Self-
View Timing interaction, the Irrelevant Self-View Timing
interaction, and the actual score. As predicted, the Relevant Self-
view Timing interaction reached significance, ␤⫽⫺.31,
t(36) 2.67, p.01, whereas the Irrelevant Self-View Timing
interaction did not (␤⫽.10, t1). We conducted simple slopes
analyses to examine this significant interaction (Aiken & West,
1991). Consistent with hypotheses, performance estimates were
tightly correlated with self-views in the pretest condition, ␤⫽.73,
t(36) 3.39, p.002, and showed no relationship in the posttest
condition (␤⫽.08, t1).
This shows that the key interaction
emerges not merely using a dichotomous coding for the relevant
self-view but using an idiographically assessed measure as well.
Finally, to confirm that the effect of test label on performance
estimates (in the pretest condition) could be entirely explained by
the idiosyncratic self-views, we regressed performance estimates
on the test-label condition and the relevant self-view, while con-
trolling for actual performance. As expected, the relevant self-view
predicted performance estimates, ␤⫽.61, t(16) 3.46, p.003,
whereas the effect of test label became nonsignificant, ␤⫽.19,
t(16) 1.21, p.24. In other words, differences in perceived skill
at abstract reasoning and computer programming explained the
gap between the two pretest conditions.
In sum, self-views influenced performance estimates, but only
when participants learned what self-view was relevant before
taking the test. These results are consistent with our hypothesis that
top-down views only contaminated performance estimates when
they had the potential to distort one’s bottom-up experience. When
the relevant self-view was learned after the test, it no longer could
contaminate one’s bottom-up experience with the test. As ex-
pected, making different self-views relevant to the test left perfor-
mance estimates unaffected.
These results may seem somewhat surprising in light of research
on the perseverance effect (Ross, Lepper, & Hubbard, 1975), the
tendency for task feedback to continue to influence performance
estimates for the future, even after the feedback has been discred-
ited. Feedback in perseverance studies is given after, not before,
the person completes the task (e.g., Guenther & Alicke, 2008;
McFarland, Cheam, & Buehler, 2007), leading to an apparent
discrepancy with the present study. That is, perseverance research-
ers see an effect for post-task information, whereas we do not.
We would point to a key difference between our work and the
research done on perseverance that may explain the different
findings for post-task feedback. In their feedback, perseverance
researchers give their participants a specific and pointed event to
explain. Participants are told that they either succeeded brilliantly
at a task or that they failed. Perseverance theories like that of Ross
et al. (1975) have argued that this false performance feedback
prompts people to construct causal stories to explain their perfor-
mance—and that these performance-explaining explanations re-
By covarying out actual performance, we could examine the influence
of top-down or bottom-up cues that were not simply sensitive to true
performance. In other words, we were not interested in the pathway by
which accurate self-views influenced performance estimates through
bottom-up experience because one’s bottom-up cues detected one’s actual
The astute reader may have noticed that those who learned of the
relevant self-view after taking the test gave performance estimates similar
to those who learned that the test was one of abstract reasoning before
taking the test. Given that the test actually was one of abstract reasoning
abilities (logical reasoning), this pattern is perhaps unsurprising. But to
further confirm that those in the after-test condition were spontaneously
drawing upon their abstract reasoning self-views in completing the test, we
regressed these after-test participants’ performance estimates on their ab-
stract reasoning and computer-programming self-views, while controlling
for actual performance. Consistent with our assumption that participants in
the posttest condition spontaneously labeled the test as one of abstract
reasoning, there was an almost-significant influence of abstract-reasoning
self-view on performance estimate, ␤⫽.27, t(19) 2.03, p.06,
whereas there was no relationship between computer-programming self-
views and estimated performance (␤⫽⫺.10, t1).
Table 1
Correlations Between Actual Performance, Relevant Self-View,
Bottom-Up Experience, and Estimated Performance, by Study
Study Self-view
Study 1
Actual performance .43
Self-view .37
Study 2
Actual performance .03 .13 .15
Self-view .16
Bottom-up experience .39
Study 3
Actual performance .09 .53
Self-view .26
Bottom-up experience .65
Study 4: GRE test
Actual performance .27
Self-view .32
Bottom-up experience .67
Study 4: High school test
Actual performance .30
Self-view .39
Bottom-up experience .68
main even after the feedback is discredited. In the language of
top-down and bottom-up cues, performance feedback leads people
to call upon other top-down theories that would explain their
success or failure. That is, it does not appear that false feedback
leads people to reconsider what their experience with the task was
like—that is, to review their bottom-up experience— but, rather, to
bring to mind top-down theories that would explain that event.
Note that, for us, the post-task information given in Study 1 was
different. Participants were not given an event to explain but,
rather, just one top-down cue that they could use to guess what that
event (i.e., their performance) was. As an example, participants in
the posttest computer-programming condition may have been sur-
prised that they did not have more difficulty with the test, learning
of the tested domain did not demand that they revise their under-
standing of the test experience. As such, their performance esti-
mates were not influenced.
There does remain one alternate model with which Study 1 is
also compatible. It is possible that knowledge of the relevant
self-view before the test did not distort one’s bottom-up experience
but, instead, discouraged one from carefully attending to
bottom-up information about a task. This perspective is reflected in
the cognitive miser approach (Fiske & Taylor, 1991), which holds
that people rely on prior top-down beliefs so they do not have to
expend the cognitive resources to attend to bottom-up cues (Bel-
more, 1987; Fiske, Neuberg, Beattie, & Milberg, 1987). This
alternative predicts that only those in the posttest condition at-
tended to performance-relevant bottom-up cues and that these
undistorted cues eliminated the influence of the self-view manip-
ulation. To rule out this alternative, in the remaining studies, we
actually measured participants’ bottom-up experience to test
whether self-views did in fact distort it.
Study 2
Study 1 provided indirect evidence that top-down views have an
impact on performance evaluation because they alter the ways in
which bottom-up cues are perceived. However, in Study 1, we did
not directly measure bottom-up cues, and thus, it was possible that
top-down views contaminated performance estimates, not because
they changed bottom-up experience, but perhaps because the in-
formation the self-view putatively conveyed discouraged people
from attending to bottom-up cues. Thus, in Study 2, we decided to
examine directly whether the link between chronic self-views and
performance assessments flowed through perceptions of
bottom-up experiences.
Participants completed a 15-item DVD-based test of interper-
sonal perception abilities. Before the test, we asked participants to
report on their interpersonal perception ability. Then, after answer-
ing each question on the test, participants answered five further
queries tapping into their bottom-up experience. We examined
whether self-views would be correlated with these bottom-up
experiences, controlling for actual performance, and whether these
bottom-up perceptions would explain any relationship between
self-views and summary performance evaluations that participants
provided after taking the test. In other words, we tested whether
the influence of self-views on performance estimates could be
statistically mediated by bottom-up experience.
Two hundred forty-two members of the Cornell University
community participated in exchange for course extra credit. Par-
ticipants completed the study as part of an hour-long session in
which they also completed an unrelated experiment.
First, all participants stated in what percentile among students at
their university they believed they fell (from 0th to 100th) on
several abilities. It was crucial that participants indicated their
Figure 1. Performance estimates by test label and when the test label was revealed (Study 1). Error bars
indicate 1 standard error.
percentile placement for social perception ability, which was de-
fined for participants as “the ability to accurately observe and
interpret the expressive behavior of others, to decode behavioral
and verbal cues from others in order to reach accurate assessments
about them, and to interpret subtle expressive behaviors.” Partic-
ipants wrote their percentile estimate in a blank provided.
Participants then learned that they would be taking the Interper-
sonal Perception Task (IPT), “a measure that indicates accuracy in
social perception.” The IPT–15 (Costanzo & Archer, 1989) pre-
sents 15 audiovisual scenes of interpersonal interaction. The test
taker must use verbal and nonverbal cues to discern the relation-
ships between the characters. For example, in one scene, a man and
a woman have a short discussion about teaching at a university. At
the end of the scene, the test-taker must indicate, on the basis of
what he or she observed, which person is the higher status person.
After participants indicated the answer to each question, we had
them answer five questions designed to tap into their bottom-up
experiences. On 5-point scales, participants indicated how long it
took them to figure out the answer, whether they immediately
knew the answer, to what extent they had to guess, the perceived
difficulty of the item, and the extent to which they went back and
forth between the answer choices. Finally, at the end of the test,
participants estimated how many of the 15 items they believed that
they answered correctly and in what percentile they believed that
their performance placed them.
Results and Discussion
As expected, participants’ estimated test score and estimated
percentile score on the test were correlated, r(240) .45, p
.001. Accordingly, we standardized each and summed these values
to create a performance-estimate composite. We then conducted
four analyses.
First, to test whether we replicated Ehrlinger and Dunning
(2003), we regressed the performance-estimate composite on rat-
ings of social perception ability that participants provided before
taking the test, controlling for participants’ actual test score. Al-
though actual performance predicted performance estimates, ␤⫽
.14, t(239) 2.24, p.03, participants’ self-views were espe-
cially predictive, ␤⫽.27, t(239) 4.38, p.001.
Second, to assess whether participants’ self-views influenced
bottom-up experiences with the test, we first conducted a
principal-components analysis on the five bottom-up queries for
each of the 15 test items separately. In each case, a single-factor
solution emerged. Thus, we took each participant’s factor score for
each of the 15 items and averaged them together to create a bottom-up
composite of bottom-up experience for each participant.
numbers indicated that the item was experienced as easier. We
regressed this bottom-up composite on self-views while control-
ling for actual performance. Participants’ actual performance was
only marginally predictive of their bottom-up experience, ␤⫽.12,
t(237) 1.92, p.06, but their self-views significantly predicted
their bottom-up experience with the items, ␤⫽.15, t(237) 2.43,
Third, we tested whether bottom-up experiences were predictive
of participants’ performance estimates. We regressed performance
estimates on bottom-up experiences while controlling for actual
performance. Again, actual performance was only marginally pre-
dictive of performance estimates, ␤⫽.11, t(237) 1.92, p.06,
but bottom-up experiences closely corresponded with participants’
performance estimates, ␤⫽.39, t(237) 6.51, p.001.
Finally, to examine whether bottom-up experiences mediated
the link between chronic self-views and performance estimates, we
tested whether bottom-up experiences would continue to predict
performance estimates after we controlled for self-views (Baron
and Kenny, 1986). Bottom-up experiences continued to predict
performance estimates, ␤⫽.35, t(237) 6.02, p.001. In
addition, the direct effect of self-views on performance estimates
was significantly reduced, as indicated by a Sobel test (z2.25,
p.02), although the effect of self-views on performance esti-
mates remained significant, ␤⫽.22, t(237) 3.73, p.001.
Taken together, these tests suggested that bottom-up experience
partially mediated the link between self-views and performance
estimates (see Figure 2).
We should note that participants’ self-views did not correlate
with their actual performance, r(240) .03, ns. Thus, to the extent
that self-views contaminated bottom-up cues, the self-views led
performance estimates astray. However, bottom-up experience did
correlate, albeit weakly, with participants’ actual performance,
r(240) .13, p.05.
In sum, even when we controlled for their actual performance,
participants’ performance estimates on a test of social perception
ability was predicted by their self-views of their social perception
ability, with this relationship being partially mediated by their
ratings of bottom-up cues. Although self-views exerted a top-down
influence on self-evaluation, they appear to do so partially by their
impact on bottom-up experience. Of course, this study was corre-
lational, so it would be a stronger test of our hypothesis to show
that manipulating self-views could influence participants’ perfor-
mance estimates by altering the bottom-up experiences they be-
lieve they had while completing a task.
Study 3
Study 3 was designed with two goals in mind. First, we returned
to the methods of Ehrlinger and Dunning (2003), using a test that
we could reasonably claim measured either abstract reasoning or
computer-programming ability, and this time, we asked partici-
pants to report their bottom-up experiences as they completed the
test. We expected that participants’ bottom-up experience would
be driven by the self-view we said was relevant to the test, but not
the self-view that was irrelevant, and that it would be the relevant
self-view that would again predict performance evaluation. This
would demonstrate that the manipulated self-view changed partic-
ipants’ bottom-up experience.
Second, we wanted to test whether the top-down beliefs changed
only the subjective bottom-up experience with the task, rather than
objective bottom-up experience. Although participants were not
aware of this, the computer timed how long it took them to
For all bottom-up items except one (knew immediately) higher numbers
tended to indicate more trouble with the question. Many participants
seemed to miss this difference, giving uniformly high or low numbers to all
five bottom-up items for each test question. By using the factor scores
instead of simply averaging the bottom-up responses together, this “noisy”
item was deweighted, given that it, of course, loaded less highly on the
single factor. Factor scores are used for the same reason in the analyses for
Study 3 as well.
complete each question. In this way, we could see if the expecta-
tion that a task would be easier actually led participants to speed
through it more quickly, or whether self-views only produced the
perception that they were flying through the task, even if that
perception had no relationship to actual experience.
One hundred thirty-three undergraduates at Cornell University
participated in the study in exchange for extra course credit.
Participants first rated four of their abilities on 11-point Likert
scales, which included ratings of their abstract reasoning ability
and computer-programming ability on scales anchored at 1 (not at
all) and 11 (completely). Then, depending on condition, partici-
pants were told they would be taking a test that tapped into abstract
reasoning ability or computer-programming aptitude. Participants
completed the same 10-item test used in Study 1, although items
were presented one at a time on a computer, rather than on paper.
After completing each item, participants answered five ques-
tions designed to assess their bottom-up experience. Because the
format of the test was different from Study 2, some of the
bottom-up cues seemed less relevant (e.g., going back and forth
between answer choices). Instead, participants indicated the extent
to which they questioned the logic they used; how long it took
them to solve a problem; the extent to which it felt like they were
guessing; the extent to which the way to solve the problem came
to them immediately, leading them to answer the question quickly;
and the perceived difficulty of the question. After completing the
test, participants estimated how many of the 10 items they an-
swered correctly and in what percentile they believed their perfor-
mance fell.
Results and Discussion
As expected, participants rated their abstract reasoning ability
more favorably (M6.9) than their computer-programming abil-
ity (M3.3), paired t(132) 16.93, p.001. As in Study 1,
participants’ percentile estimates and score estimates were posi-
tively correlated, r(131) .45, p.001. We once again stan-
dardized and summed these values to create a performance-
estimate composite.
To assess whether the self-view manipulation led to between-
condition differences in performance assessments, we submitted
participants’ performance estimates to a one-way analysis of co-
variance with actual performance as a covariate. Participants who
believed they were taking the abstract reasoning test believed they
had performed better (M0.32, SE 0.18) than those who
completed the computer programming test (M⫽⫺.24, SE
0.16), F(1, 130) 5.38, p.02, even though participants’ actual
performance did not differ between the abstract (M7.6, SD
1.84) and computer programming (M7.7, SD 1.97) condi-
tions (F1). Contrary to our expectations, bottom-up experience
did not differ by condition (Ms⫽⫺.01 and .01), F1. None-
theless, our remaining analyses suggest our manipulation did in-
deed change bottom-up experience, and we return to this oddity in
Study 4 with a methodological modification.
Mediation Analyses
To examine whether the relevant, but not the irrelevant, self-
view predicted performance estimates through an effect on
bottom-up experience (despite the lack of a main effect of condi-
tion on bottom-up experience), we conducted mediational analyses
for each condition. All regression analyses controlled for actual
test performance, so we concentrated on influences on bottom-up
experience and performance estimates that did not stem from
actual performance. As in Study 2, we combined the responses to
the bottom-up questions for each pair of items using principal-
components analysis. For each participant, we averaged across
these five factor scores to create a bottom-up composite for each
participant. Higher scores indicated a more positive bottom-up
experience with the test. A summary of analyses is presented in
Figure 3.
Abstract reasoning test. We regressed participants’ estimated
performance on the abstract reasoning and computer-programming
self-views. As expected, participants’ abstract reasoning self-
views predicted performance estimates, ␤⫽.25, t(54) 2.15, p
.04, whereas their computer-programming self-views did not relate
to performance estimates (t1). We then regressed bottom-up
experience on the two self-views. Again, abstract reasoning self-
Figure 2. The mediation model demonstrating that bottom-up experience partially mediates the relationship
between chronic self-views and performance estimates (Study 2). Note that the standardized betas in parentheses
are those from the regression model in which performance estimate was regressed on self-view and bottom-up
experience simultaneously.
views predicted bottom-up experience with the test, ␤⫽.40,
t(54) 3.48, p.001, whereas computer-programming self-
views’ influence was nonsignificant (t1). Bottom-up experience
also drove participants’ performance estimates, ␤⫽.63, t(55)
6.50, p.001. When regressing performance estimates on the
bottom-up experience and both self-views, the bottom-up experi-
ence remained a strong predictor of performance estimates, ␤⫽
.64, t(53) 5.86, p.001, whereas both self-views no longer
offered any explanatory power in predicting performance esti-
mates (ts1). A significant Sobel test (z2.02, p.04),
suggested that the influence of abstract self-views on test perfor-
mance was fully mediated by the self-views’ influence on
bottom-up experience.
Computer-programming test. For participants told the test
tapped into computer-programming ability, computer-programming
self-views predicted performance estimates, ␤⫽.27, t(71) 3.33,
p.001, although abstract reasoning self-views did as well, ␤⫽
.19, t(71) 2.31, p.02. However, in looking at the influence of
self-views on bottom-up experience, only computer-programming
self-views affected bottom-up experience, ␤⫽.19, t(71) 2.02,
p.05, whereas abstract reasoning self-views did not, ␤⫽.13,
t(71) 1.46, p.14. Bottom-up experience, in turn, was a
predictor of performance estimates, ␤⫽.36, t(71) 3.42, p
.001. When we regressed performance estimates on bottom-up
experience and both self-views, bottom-up experience remained a
significant predictor of performance estimate, ␤⫽.24, t(70)
2.37, p.02, as did the computer programming self-view, ␤⫽
.23, t(70) 2.79, p.01. The abstract reasoning self-view
dropped to marginal significance, ␤⫽.16, t(70) 1.94, p.06.
Because the abstract reasoning self-view did not influence
bottom-up experience, bottom-up experience was only a candidate
mediator for the relationship between computer-programming self-
view and performance estimate. A Sobel test provided evidence for
partial mediation (z1.93, p.05).
Subjective Versus Objective Bottom-Up Experience
We next examined whether self-views influenced only the par-
ticipants’ subjective experiences with the task or whether they also
influenced objective experiences. We did so by examining the
experience of time. To create an index of perceived time, we
averaged the five ratings (one for each pair of items) of the extent
to which participants solved the problem quickly (reverse scored)
with the five items asking how long it took participants to solve the
This composite had good internal reliability (␣⫽.75),
especially given the fact that we aggregated across bottom-up
experience from different items. As a measure of actual time, we
examined the average number of milliseconds that participants
took from the time the multiple-choice question appeared on the
screen to the point at which they indicated a response. Because,
across participants, there was evidence of positive skew, we log
transformed these times. We then regressed the perceived time
index on both self-views and the log-transformed actual times.
For those in the abstract reasoning condition, the abstract rea-
soning self-view predicted a sense that participants were taking
less time to complete the items, ␤⫽⫺.46, t(54) 3.87, p.001,
whereas the computer-programming self-view had no effect (t
1). The actual time participants spent on the question did not relate
to how long they felt they were taking, ␤⫽.12, t(54) 1.01, p
.31. For those in the computer-programming condition, more pos-
itive computer-programming self-views led participants to believe
they were completing the items more quickly, ␤⫽⫺.22, t(71)
1.97, p.05, whereas the abstract reasoning self-view had a
marginal influence, ␤⫽⫺.19, t(71) 1.70, p.09. Again, the
actual time participants spent on the items did not predict their
sense of how long it was taking them, ␤⫽.18, t(71) 1.61, p
.11. Even collapsing across conditions, there was no significant
correlation between the log-transformed actual times and the per-
ceived time, r(131) .13, p.12. Furthermore, there was no
tendency for those with more positive self-views to actually com-
plete the items any more quickly; if anything, they completed the
items more slowly r(131) .10, p.23.
In sum, despite taking the same test, participants’ performance
evaluations corresponded to their self-rated views of skill regarding
abstract reasoning ability or computer-programming ability, depend-
ing on how we labeled the test. The relationships we observed
between self-ratings of skill and performance estimates were medi-
ated by the influence of the appropriate self-view on bottom-up
experience, which then connected to performance estimates. Also,
participants’ perceptions of the time did not correspond to the actual
passage of time. Instead, subjective time perception was predicted by
the self-view purportedly relevant to the test.
To determine the reasonableness of this composite, we presented 60
participants with the five original items and asked them to select the item
or items that essentially asked about “the perceived length of time it took
to solve the problem” and to rate each item on a 1– 4 scale for whether it
should be included in a perceived time composite. A majority selected
these two items. A majority rejected the other three items. On the contin-
uous measure, only these two items had means greater than 3. Regardless
of whether one looks to the dichotomous or continuous measure, the two
items making up our composite were judged superior to the other three
items ( ps.001).
Figure 3. The mediation models demonstrating that bottom-up experi-
ence mediated the relationship between participants’ abstract reasoning and
computer-programming self-views on their performance estimates on what
was labeled an abstract reasoning and computer-programming test, respec-
tively (Study 3). Note that all numbers are standardized betas. The betas in
parentheses are from the regression model in which performance estimate
was regressed on the test-appropriate self-view and bottom-up experience
Given that bottom-up experience was differentially tied to different
self-views in each condition, it is clear that our manipulation of
top-down views changed participants’ bottom-up experience. For
example, the only way that computer-programming self-views could
be unrelated to self-views in the abstract reasoning condition but
related to self-views in the computer programming condition is if the
manipulation shifted participants’ bottom-up experience into align-
ment with this self-view. Furthermore, our condition manipulation
changed participants’ performance estimates. Like the bottom-up
experience, performance estimate was differentially tied to the two
self-views depending on participants’ condition. Finally, the two
significant mediation models reflect that self-views influenced per-
formance estimates by influencing bottom-up experience.
Study 4
Nonetheless, there remained one oddity in the data from Study
3: Making a different self-view relevant to the test changed per-
formance estimates between conditions but did not produce a
similar main effect for bottom-up experiences. That is, participants
in the abstract reasoning condition did not rate their bottom-up
experiences of the test as more benign relative to those in the
computer-programming condition. How can there be clear evi-
dence that self-views changed bottom-up experience at the indi-
vidual level but no overall main effect between conditions?
One explanation for this oddity is that participants in Study 3
may have used the bottom-up scales differently by condition
because they were using different reference points to inform their
reports. People in the computer-programming condition, for ex-
ample, may have actually experienced the test as more difficult,
but then moderated their bottom-up reports to indicate that al-
though the test was tough, it was “not that tough for a computer
science test.” This would explain why the mediation models within
each condition supported our hypothesis and why there was a
between-condition main effect on performance estimates but not
bottom-up experience. Other work has shown that people shift
their usage of rating scales depending on their expectation or
experience in a way similar to what we suspect here. For example,
people shift how they use a response scale about height or aggres-
sion depending on whether they are talking about a man or a
woman. That is, a woman need not be as tall as a man to receive
a rating of tall from perceivers (for a review, see Biernat, 2005), a
response tendency known as the shifting standards phenomenon.
Respondents have also been shown to shift how they use response
scales depending on their cultural context. For Asian respondents
to rate themselves as respectful of elders, they need to show more
of that respect on an objective, behavioral level than would a
respondent from North America (Peng, Nisbett, & Wong, 1997).
To minimize this possible artifact in Study 4, we manipulated
self-views within participants, rather than between them, so that
participants would be more likely to define the bottom-up scales in
a consistent way across both tasks they completed. In other words,
if we asked participants between subjects to rate on a 9-point scale
the height of a 6 ft. 3 in. (1.9 m) man and a 5 ft. 9 in. (1.75 m)
woman, we imagine that both targets would be rated an 8 or a 9.
But if we asked participants to place both targets on the same scale,
it would be more likely that the man would receive a higher
numerical judgment than the woman.
In Study 4, participants took two 15-item tests of U.S. history.
They were told that one of the tests included items from the state
of Maine’s high school exit exam. The other test supposedly
included items from a graduate entrance exam for history PhD
programs. We predicted that people would report a more confident
performance estimate and easier bottom-up experience when the
task was described as a high school history test (for which partic-
ipants should have a favorable self-view) than when it was de-
scribed as a graduate-level history test (for which participants
would have a more negative self-view). Further, we expected that
a within-subjects mediation analysis (Judd, Kenny, & McClelland,
2001) would indicate that differences in bottom-up experience
explained the link between self-view and performance assess-
Finally, we should mention that we altered the focus in our
measures of bottom-up experience. Bottom-up experience ques-
tions related to whether one is confident in one’s logic in working
through a problem are appropriate for tests of logical reasoning but
are less so for tests of history. As we have mentioned above,
research shows that people are more confident in their responses to
the extent that they find test material to feel familiar, rather than
novel (Arkes et al., 1991; Schwartz & Metcalfe, 1992). Therefore,
our bottom-up experience questions dealt with the perceived fa-
miliarity and fluency of the material.
One hundred fifty-nine undergraduates at Cornell University
participated in exchange for extra course credit.
Participants were told they would be completing two tests of U.S.
History that contained past questions from the GRE: History and the
Maine High School Exit Exam. To lend validity to our cover story, we
asked participants to indicate whether they had taken either test
between 1994 and 1998, the years from which the questions were
supposedly taken. In addition, we explained that during that interval,
one of the tests had shifted from four to five answer choices, which
explained why one of the tests had a variable number of answer
choices. We hoped that this small difference would lend credence to
our cover story that the tests indeed came from different sources. Each
test was in a separate test booklet, and the cover pages contained the
official seal of the GRE and the Educational Testing Service or of the
state of Maine. In addition, the text “Must return to experimenter at
end of session. Officially approved for in-lab use only!” appeared in
a box at the bottom of both cover pages.
The order in which participants completed the tests was coun-
terbalanced. We also used two tests. For roughly half of the
participants, Version A was described as the high school-level test
and Version B the graduate school version. For the remainder, the
pairing of version with its label (high school vs. graduate school)
was reversed. As such, participants overall confronted tests of
equal difficulty across high school and graduate school conditions.
Before beginning the test, participants indicated on 1 (not at all)
to 11 (completely) scales the extent to which they knew U.S.
history as would be required on a high school exit exam and on a
graduate school entrance exam. Participants went through each
15-item test, writing their multiple-choice answers on a provided
page. After every three questions, participants were to indicate
their bottom-up experiences from each of the previous three ques-
tions. Those items included: knew the answer immediately upon
reading the answer choices; regardless of how I know, the answer
feels like the right one; and I have a specific memory of having
been exposed to this information before. Participants indicated
their degree of agreement with each item on 4-point scales, with
higher numbers indicating greater agreement.
After completing both tests, the experimenter collected the
answers and bottom-up cues and provided participants with a
follow-up sheet. Participants first indicated how many items out of
15 they believed they answered correctly on both tests. Then,
participants completed four questions that asked them to retrospec-
tively evaluate their bottom-up experience on each test. Partici-
pants indicated their confidence in their answers to the questions,
the extent to which it felt like the questions dealt with content they
had learned before, to what extent they marked items that were
backed up with things they knew or had once learned, and how
often it felt like their responses were simply random guesses.
Results and Discussion
Indicating that the manipulation was successful, participants had
more confidence in their high school U.S. history abilities (M
7.1, SD 2.1) than in their graduate entrance exam U.S. history
abilities (M3.6, SD 1.9), paired t(158) 22.50, p.001.
We begin below by presenting the results of our within-subjects
analytic approach. We then also tested our hypotheses using a
between-subjects approach that is more consistent with our earlier
studies by looking only at the test that participants took first. Even
though the between-subjects approach would not correct for a
shifting-standards problem, and thus would be expected to produce
weaker results, it is possible that we would see less of a shifting-
standards problem than in Study 3, given that all participants knew
while taking the first test that they would be taking both a high
school and a GRE test and that they would be making their
bottom-up ratings on the same scale.
Within-Subjects Analyses
We first tested whether self-views influenced participants’ per-
formance estimates independent of any actual performance differ-
ences. Controlling for which test version was paired with which
test label, participants performed no better on the high school test
(M6.5, SE 0.16) than on the GRE test (M6.7, SE 0.17),
F(1, 157) 1.74, p.18, suggesting that the test labels them-
selves did not systematically influence actual performance. How-
ever, controlling for participants’ actual performance and version
of the test/label pairing, participants estimated that they had an-
swered more items correctly on the high school test (adjusted M
5.8, SE 0.20) than the GRE version (adjusted M4.9, SE
0.20), F(1, 156) 30.53, p.001.
To determine whether the self-view influenced bottom-up ex-
perience with the task, we began by averaging together the
bottom-up ratings for each question.
We then had a composite for
each of the three cues— knew immediately; regardless of how I
know, it just felt right; felt like I had learned this before—for each
of the tests. For each comparison, we included both the test-
version label pairing and participants’ actual performance as co-
variates. While taking the high school test, participants’ bottom-up
experience with the task suggested they experienced the task as
easier, F(1, 153) 18.16, p.001.
We have shown that participants had more positive self-views
regarding the high school test compared with the GRE, reported
having a more positive bottom-up experience with the high school
test, and then incorrectly perceived that they had performed better
on the high school test. To demonstrate within-subject media-
tion—that the influence of self-view on performance estimates can
be explained by different bottom-up experiences—we computed
the correlation between performance estimates and bottom-up ex-
perience, while controlling for actual performance (Judd, Kenny,
& McClelland, 2001). Consistent with our mediation model, online
bottom-up experience strongly correlated with performance esti-
mates for both the high school test, pr(153) .64, p.001, and
the GRE test, pr(153) .64, p.001.
To further confirm the mediation model, we examined how par-
ticipants responded to the high school test relative to the graduate
school test along three difference scores (high school minus graduate
school rating) for actual performance, perceived performance, and
bottom-up cues. If the impact of the self-view manipulation is medi-
ated by bottom up cues, then the difference score of bottom-up cue
ratings should be correlated with the difference score of perceived
performance (Judd et al., 2001). That was, indeed, the case (control-
ling for the actual score difference), pr(153) .67, p.001. In short,
to the extent that participants rated their bottom-up experience with
the high school test as easier than the graduate school test, they also
perceived their performance on the high school test to be higher
relative to the graduate school test.
We used averaged composites, instead of factor scores (as in Studies 2 and
3), for two reasons. First, there was no reverse-scored item, so the motivation
for using the factor scores was reduced (see Footnote 3). Second, because
factor scores by definition have a mean of zero, then by standardizing each of
our two composites (the GRE bottom-up experience and the high school
bottom-up experience), we would eliminate the ability to observe a main effect
on this repeated-measures factor. For our between-subjects analyses, we re-
conducted them using factor scores and found substantively equivalent results.
We thank an anonymous reviewer for pointing out that one of our
bottom-up items, “think I have learned this before” may not reflect a
bottom-up reaction to the task (i.e., “This feels quite familiar to me!”) but
instead an abstract inference drawn from the manipulation (i.e., “I went to a
good high school, so I must have learned this”). Although we intended the item
to measure the former, to the extent that some people answered the item
consistent with the latter interpretation, then this would merely show an
alternate top-down influence on performance estimate (i.e., “Given that my
excellent high school offered a comprehensive American history class, I must
know this stuff.”). But given that participants indicated their feeling that they
learned the item before for each individual item, it seems likely that this
focused participants on their bottom-up experience with each particular item,
instead of on a more abstract inference that would apply to the test as a whole.
Nonetheless, the mediation model was significant using each bottom-up item
individually, so even if one were to conservatively exclude this particular item,
the evidence that bottom-up experience mediates the relationship between
self-views and performance estimates does not change.
Between-Subjects Analyses
We should note that the design of Study 4 also allowed us to
conduct a between-subjects analysis of mediation, if we set aside
the data for the second test that each participant took. Looking at
only the first test that participants took, those who took the high
school test did not in reality score any higher (M6.5, SD 2.2)
than those taking the GRE test (M7.0, SD 2.1), F(1, 156)
1.75, p.18. But once again, controlling for actual performance,
those taking the high school test perceived that they did better
(M5.8, SE 0.29) than those taking the GRE exam (M4.7,
SE 0.30), F(1, 155) 6.55, p.01. Furthermore, the
bottom-up experience of those who took the high school test
suggested that they had less difficulty with the test (M2.2, SE
0.060) than those who completed the GRE test (M2.0, SE
0.060), F(1, 152) 5.60, p.02. It is worth noting that this
between-condition difference is substantially smaller than the one
found using the within-subject analysis. This is, of course, not
surprising given the within-subjects analysis controls for system-
atic participant-level error.
To test whether bottom-up experience mediated the impact of
test label (high school vs. graduate school) on performance esti-
mates, we regressed performance estimates on test label and
bottom-up experience ratings, while controlling for actual perfor-
mance and which specific version of the test the participant took
first. The bottom-up experience strongly predicted performance
estimate, ␤⫽.66, t(151) 10.55, p.001, but there was no
longer a significant effect of test condition, ␤⫽.09, t(151) 1.46,
p.14. A Sobel test confirmed the significance of the full
mediation model (z2.31, p.02).
To provide yet a more stringent test for our hypotheses, we reran
our mediation model not simply coding test condition dichoto-
mously but by using the Likert-scale self-view rating provided by
each participant. The relevant self-view predicted bottom-up ex-
perience, ␤⫽.36, t(152) 5.04, p.001, and performance
estimates, ␤⫽.38, t(152) 5.42, p.001. Furthermore, when
performance estimate was regressed on both the relevant self-view
and the bottom-up experience, bottom-up experience continued to
predict performance estimates, ␤⫽.61, t(151) 9.32, p.001,
whereas the influence of relevant self-view was reduced, ␤⫽.18,
t(151) 2.85, p.005. A much stronger Sobel test confirmed the
significance of this partial mediation model (z4.43, p.001).
It is not surprising that this mediation model is much stronger,
given that the idiographically measured self-views can account for
the influence on the mediator (bottom-up experience) much better
than a condition-specific dichotomously coded variable.
In sum, by manipulating within-subjects the self-view relevant
to performance evaluation, we found that (a) self-views influenced
performance estimates, (b) self-views influenced participants’ on-
line bottom-up experience, and (c) these differences in on-line
bottom-up experiences mediated the link between self-views and
performance estimates.
General Discussion
In this article, we set out to examine the role played by top-down
self-views and bottom-up experiences in influencing estimates of
one’s performance. Although past research repeatedly found that
chronic self-views drive performance estimates (Ehrlinger & Dun-
ning, 2003), the specific psychological mechanisms responsible
for this influence have, to date, remained unclear. In the four
studies reported herein, we consistently found that chronic self-
views shaped bottom-up experiences with the tests we presented to
participants, which then informed their performance estimates.
Those who thought they were highly skilled in the domain in
question, relative to those who believed they were less skilled,
thought they were taking less time to answer questions, were
expending less effort, and were feeling more familiarity with the
possible answers they could choose.
Specifically, Study 2 showed that the naturally occurring link
between chronic self-views and performance estimates were sta-
tistically mediated by perceptions of bottom-up experiences. Stud-
ies 3 and 4 manipulated which self-view was relevant to a task and
showed that bottom-up experiences and performance estimates
were influenced by that relevant self-view, but not by an alterna-
tive, irrelevant self-view. Again, the impact of relevant self-views
on performance estimates importantly involved how participants
perceived their bottom-up experiences with the task. Study 3
demonstrated this at the individual participant level; bottom-up
experiences were correlated with whether participants rated them-
selves high or low on the relevant ability. Study 4 demonstrated
this at the normative level. When participants were asked about
performance on what was supposedly a high school-level history
exam, they thought they were more skilled, rated their bottom-up
struggles with the exam as more benign, and estimated that they
answered more questions correctly.
Finally, Study 1 showed that self-views influence performance
estimates if they were manipulated before participants confronted
the task, when those self-views could then influence perception of
bottom-up experience, but not when they were manipulated after
confronting the task, presumably because bottom-up experiences
with the task had already been fixed. In addition, Study 3 showed
that the influence of these top-down self-views was so strong that,
in at least one case, their power far exceeded the power of reality
to influence perception. Participants’ views of their own ability
strongly affected how much time they thought they were taking to
Given that our between-subjects analytic technique uncovered a
between-condition difference in participants’ bottom-up experience, does
this mean that the mere knowledge that participants would be taking and
rating a second test of (expected) different difficulty eliminated the “shift-
ing scale” problem that may have produced the single oddity in the data of
Study 3? Possibly, but there is an additional concern that leads us to put
more weight in our within-subjects analytic approach. It is possible that
those who first took the high school [GRE] test expected the second test to
be much more difficult [easy]. This may have pushed the high school and
GRE test-takers toward the “I am doing well [poorly]” side of the scales,
respectively, leaving themselves room to rate the second test in opposition
to the first one. This potential artifact only applies to the between-subjects
analyses, which is why we believe the within-subjects analyses are the
stronger test.
We conducted all analyses again using the retrospective reports of
bottom-up experiences participants provided at the end of the experimental
session. All analyses with these measures led to the same statistical
conclusions we reached when focusing on online measures of bottom-up
experience. This is not surprising, given that retrospective measures of
bottom-up experience were highly correlated with online measures (rs
.72 and .70 in the GRE and high school conditions, respectively).
answer questions. Their estimates, however, bore no relationship
with how much time they actually took, objectively measured.
The Nature of the Mediation
Careful readers may speculate about an alternative to our me-
diational model. What if we flipped the roles played by perfor-
mance estimates and bottom-up cues? Could self-views lead di-
rectly to performance estimates, which then inform bottom-up
cues? We think this alternative is less plausible than our account
given two issues: (a) the order in which bottom-up cues and
performance estimates were assessed and (b) the psychology im-
plied by this chronology. First, we asked participants to report their
bottom-up experiences as they confronted the test and asked for
their performance estimates only after the test was completed.
Thus, our measures of bottom-up experience had temporal priority
over our measures of performance estimates. That is, bottom-up
experiences came first, and it is easier to imagine that they influ-
enced subsequent perceptions of the task (i.e., summary perfor-
mance estimates) than for those subsequent perceptions to reach
back in time and influence reports of bottom-up cues. In any event,
work in eyewitness testimony shows that explicitly asking partic-
ipants to consider their bottom-up experiences, as we did here,
inoculates them from top-down influences that may come later in
the experience of giving testimony (Wells & Bradfield, 1999).
Had we asked participants to recall their bottom-up experience
at the end of the task, this alternative model might have been more
psychologically plausible. Given that we measured bottom-up
experience online, one would have to posit a hidden variable of a
tentative performance estimate that influences ratings of
bottom-up experience. That is, for each question, participants first
decided their confidence that they had gotten it right and then
decided to base their reports of bottom-up experience on this
tentative individual-item assessment. Although this process is the-
oretically possible, it is less plausible than the one we favor, in
terms of parsimony.
It is also somewhat hard to imagine that, for
example, people decided how much they were concretely vacillat-
ing between answer choices only after they decided whether they
answered a question correctly, rather than having the experience at
the time they worked toward answering the test question.
Thus, across these studies, we found consistent evidence that
bottom-up experiences mediated the link between chronic self-
views and performance estimates. The individual analyses, how-
ever, were inconsistent about whether this mediation was full or
only partial. For our theoretical purposes, we think it is essential
merely to show partial mediation. Self-views might influence
performance evaluations through multiple routes; we wanted to
examine whether one of those routes led through the experience of
bottom-up cues.
In studies that revealed only partial mediation, we think it is
important to keep in mind that our measures of bottom-up expe-
riences may not have been comprehensive. Bottom-up experience
is a broad, varied construct, and capturing bottom-up experience as
a whole is not a straightforward task. Thus, if we failed to capture
all of the bottom-up experiences that were influenced by top-down
views, our statistical analyses may have missed the full role being
played by bottom-up experience in accounting for the link between
self-views and performance estimates. Thus, some mediation may
have been partial, not because there was a direct effect of chronic
self-views on performance estimates, but rather because the
bottom-up cues mediator was only partially measured. Results
from Study 1 buttress this account. By varying when participants
learned of the relevant self-view, we could test the extent to which
top-down views required subsequent bottom-up experience to con-
taminate performance estimates, without having to directly mea-
sure bottom-up experience. Given that there was not a hint of a
difference between the two self-view conditions in the posttest
condition, we believe this tilts the scale in favor of the full
mediation model.
However, it would be premature to conclude that self-views
always exert their effect on performance estimates by altering
bottom-up experience. For example, we imagine that there are
times in which self-views exert a direct effect on performance
estimates because bottom-up cues to task performance are rela-
tively unavailable. An extremely confident stand-up comic per-
forming in front of a live audience may interpret the laughter of her
audience as more raucous than it actually is, leading her to con-
clude she is doing a great job. An extremely confident stand-up
comic performing in front of a TV audience will not have as many
bottom-up cues to evaluate his performance, and he may decide
that he must have done a great job merely by relying upon his prior
self-views. We leave it to future research to find additional ways
by which self-views may influence performance estimates and in
what contexts and for what tasks self-views influence performance
estimates in these different ways.
Notes on the Use of Bottom-Up Cues in
Performance Estimates
In addition, a critic might also ask whether the apparent heavy
reliance on (contaminated) bottom-up experience accurately re-
flects the way people make judgments in the real world. Unlike in
real-world contexts, we impelled participants to explicitly report
on their bottom-up experience in a rather heavy-handed way.
Could this have led them to artificially attend to and rely on these
cues in their performance estimates? We think the answer to this
question is “no.” Study 1 provided the strongest evidence that the
distorting influence of top-down beliefs on bottom-up cues occurs
even in the absence of an experimenter-instructed focus on
bottom-up cues. In that study, participants did not report their
bottom-up experience, but self-views continued to influence esti-
mates when this information was provided before participants
were exposed to the performance task. In contrast, self-views did
not influence performance estimates when participants learned of
it after the task. This suggests that even in the absence of explicit
If performance estimates were generated online, one could assume that
our measure of performance estimates would be closely tied to online
bottom-up experience, rather than to retrospective bottom-up experience
measures. If performance estimates were not generated until the end of the
test, then performance estimates should be, instead, more closely tied to
retrospective bottom-up experience. In a multiple-regression analysis, the
retrospective bottom-up experience remains very closely tied to perfor-
mance estimates, ␤⫽.71, t(152) 11.51, p.001, whereas the online
bottom-up experience becomes a much weaker predictor, ␤⫽.18,
t(152) 2.90, p.004. This suggests that the performance estimates
really were formulated at the end, which is consistent with our favored
mediation model.
measures of bottom-up experience, self-views continued to influ-
ence performance only when they could influence bottom-up cues
to task performance.
A second question is the extent to which self-views override or
simply compete for influence with objective reality in shaping
bottom-up experience. Consistent with the overriding perspective,
Study 3 found that although self-views influenced perceived time
perception (i.e., how long it seemed to take to answer a question),
actual time passage did not. It seems, however, that some cues may
be more constrained by reality than others. Determining the
bottom-up cues most likely to be used in different types of per-
formance tasks and the degree to which they are constrained by
objective reality will help to determine the circumstances in which
self-views are likely to exert their biggest influence on perfor-
mance evaluation. This issue is particularly important in examin-
ing how best to decontaminate one’s self-assessments.
A third question is whether bottom-up experience is only a
source of error, or whether it is also a source of accuracy. We
returned to Studies 2– 4 and tested whether any of participants’
accuracy (the relationship between actual performance and esti-
mated performance) could be explained through their bottom-up
experience. Although actual performance did not relate to
bottom-up experience in Study 2, it did in Studies 3 and 4. Indeed,
in follow-up analyses, we found that bottom-up experience par-
tially (Study 3) or fully (Study 4) mediated the relationship be-
tween actual and estimated performance in the final two studies.
But when we regressed bottom-up experience on self-views and
actual performance simultaneously, self-views tended to be a
stronger predictor. In other words, although bottom-up experience
may in part be a function of actual performance, it may equally (if
not more so) be a function of misleading self-views. Nonetheless,
this point raises hope that we may be able to find cues to self-
insight in bottom-up experience, if only future research can deter-
mine exactly where we should look.
What conditions must be satisfied for bottom-up experience to
become a source of accuracy instead of error? First, the cue must
be perceived accurately. The measurement of perceived time
(Study 3) that was not related to the actual passage of time shows
that accurate detection can be a challenge. But note that even if the
perception of a particular bottom-up cue is not distorted by top-
down beliefs, two additional conditions must be satisfied for
reliance on this cue to increase self-assessment accuracy. Second,
the cue must actually be diagnostic of performance. Accurate
perception of a nondiagnostic cue is unhelpful. Third, people must
have the appropriate naı¨ve theory linking the particular bottom-up
cue to actual performance. For example, participants in Studies 2
and 3 believed that solving an item quickly indicated that one had
answered it correctly. But when we returned to examine how the
actual time it took to solve a problem correlated with performance
(in Study 3), we actually found that the longer participants took on
a problem, the more likely they were to answer it correctly,
r(131) .19, p.03. Thus, even if participants’ assessments of
the bottom-up cue (i.e., the passage of time) had been accurate,
they would have applied an incorrect theory to understand this
cue’s actual implications for performance. On an optimistic note,
correcting these inaccurate naı¨ve theories may be a relatively
simple way to improve accuracy in self-assessment.
In addition, beyond showing that bottom-up experiences medi-
ate the link between self-views and performance estimates, find-
ings across the four studies serve to rule out other ways in which
bottom-up cues could influence performance estimates. If, for
example, top-down views and bottom-up cues merely exert sepa-
rate and independent influences on performance estimates, then
reliance on top-down cues would not have been sensitive to our
timing manipulation in Study 1, nor would the observed mediation
have emerged. In addition, if the relevant top-down self-view had
its impact through memory, leading people to later recall details of
one’s performance consistent with one’s self-view (cf. Guenther &
Alicke, 2008), then, again, the timing manipulation of Study 1
should not have mattered. Furthermore, bottom-up experience
assessed online (instead of at the time of performance estimation,
when one’s memory could be distorted) would not have been a
successful mediator.
Implications for Stereotyped Groups
Often, members of stereotyped groups confront a performance-
debilitating concern as they perform tasks for which their group is
thought to have low ability. Such stereotype threat has been shown
to impede African Americans’ performance on a test of verbal
ability (Steele & Aronson, 1995), the performance of children of
low socioeconomic status on an intelligence test (Croizet & Claire,
1998), and White men’s performance on a math test that would be
compared with Asian men’s performance (Aronson et al., 1999). In
these performance settings, stereotyped target members are con-
stantly monitoring for cues of how they are performing, assessing
whether they are behaving in a stereotype-consistent way
(Schmader, Johns, & Forbes, 2008). Some research suggests that
this means that they are especially vigilant to actual signs of
failure. Neurophysiological techniques like event-related potential
and functional magnetic resonance imaging have found that under
conditions of stereotype threat, there is evidence of increased
monitoring for and vigilance to performance errors (Amodio et al.,
2004; Forbes, Schmader, & Allen, 2008).
The four studies contained herein suggest another burden for tra-
ditionally stigmatized groups: When a task is ambiguously difficult,
negative expectancies tied to stereotypes may lead one to experience
it as especially difficult, thereby confirming negative stereotypes
about oneself and one’s group, initiating or reinforcing the negative
processes that underlie stereotype threat. Such a subjective sense of
failure may only reinforce the anxiety of stereotype confirmation,
leading to actually worse performance, thereby exacerbating the neg-
ative performance effects of stereotype threat.
In addition, self-views’ coloring of bottom-up experience may
have other negative longer term impacts on stereotype-relevant
tasks. Ehrlinger and Dunning (2003) described how the link be-
tween self-views and performance assessments may underlie dif-
ferences in the enthusiasm with which men and women pursue
scientific careers. Replicating past work, Ehrlinger and Dunning
found that women thought less of their scientific talent than did
men and that this difference led to women thinking they had done
more poorly on a pop quiz on science than did men, even though
there were no differences in objective performance. This percep-
tion of performance, however, mattered in that it led women to
volunteer less often than did men for a science game show taking
place later on. In short, lowered estimates of performance led to
greater lack of interest in science.
Given our data, we can speculate on an additional route by which
preconceived self-views can prompt disinterest in a relevant activity.
Even if people with negative self-views are provided with objective
feedback about their performance, indicating they had done well, their
negative self-views may still leave them with the impression that the
process followed to achieve that performance was aversive and labo-
rious. That is, someone might be told that she did well on a science
quiz, but she may still not want to go on in science because self-views
lead her to believe that she struggled with the quiz. To be sure, we did
not explore this possibility in the research described in this article, but
it would be interesting to pursue whether perceptions of bottom-up
cues shape not only performance evaluations but also subsequent
interest in the activity itself.
The top-down influence of chronic self-views does not vie with
bottom-up experience in influencing self-evaluation. Rather, self-
views define bottom-up experiences. Even though self-estimates of
task performance closely aligned with self-reported bottom-up
experience with the task, this experience was constructed, at least
in part, on the basis of preconceived self-notions coming into the
task. This research may seem to paint a somewhat bleak picture for
the ability of people to evaluate themselves accurately, but we
hope that future research can uncover what objective cues actually
are most diagnostic of task performance. Until then, it remains
important that our work be evaluated by those who do not have
clear top-down beliefs about what to expect from us, even if they
at times don’t realize just how great that work is.
Aiken, L. S., & West, S. G. (1991). Multiple regression: Testing and
interpreting interactions. Newbury Park, CA: Sage.
Allison, R. I., & Uhl, K. P. (1964). Influence of beer brand identification
on taste perception. Journal of Marketing Research, 1, 36 –39.
Amodio, D. M., Harmon-Jones, E., Devine, P. G., Curtin, J. J., Hartely,
S. L., & Covert, A. E. (2004). Neural signals for the detection of
intentional race bias. Psychological Science, 15, 88 –93.
Arkes, H. R., Boehm, L. E., & Xu, G. (1991). Determinants of judged
validity. Journal of Experimental Social Psychology, 27, 576 – 605.
Aronson, J., Lustina, M. J., Good, C., Keough, K., Steele, C. M., & Brown,
J. (1999). When White men can’t do math: Necessary and sufficient
factors in stereotype threat. Journal of Experimental Social Psychology,
35, 29 – 46.
Baron, R. M., & Kenny, D. A. (1986). The moderator–mediator variable
distinction in social psychological research: Conceptual, strategic and
statistical considerations. Journal of Personality and Social Psychology,
51, 1173–1182.
Belmore, S. M. (1987). Determinants of attention during impression-
formation. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 13, 480 – 489.
Benjamin, A. S., & Bjork, R. A. (1996). Retrieval fluency as a metacog-
nitive index. In L. M. Reder (Ed.), Implicit memory and metacognition:
The 27th Carnegie Symposium on Cognition (pp. 309 –338). Hillsdale,
NJ: Erlbaum.
Biernat, M. (2005). Standards and expectancies: Contrast and assimilation
in judgments. New York: Psychology Press/Taylor & Francis.
Bunz, U., Curry, C., & Voon, W. (2007). Perceived versus actual
computer-email-web fluency. Computers in Human Behavior, 23, 2321–
Costanzo, M., & Archer, D. (1989). Interpreting the expressive behavior of
others: The Interpersonal Perception Task. Journal of Nonverbal Behav-
ior, 13, 225–245.
Costermans, J., Lories, G., & Ansay, C. (1992). Confidence level and
feeling of knowing in question answering: The weight of inferential
processes. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 18, 142–150.
Croizet, J. C., & Claire, T. (1998). Extending the concept of stereotype and
threat to social class: The intellectual underperformance of students from
low socioeconomic backgrounds. Personality and Social Psychology
Bulletin, 24, 588 –594.
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, 19, 134 –139.
Dunning, D. (2005). Self-insight: Roadblocks and detours on the path to
knowing thyself. New York: Psychology Press.
Dunning, D., Heath, C., & Suls, J. (2004). Flawed self-assessment: Impli-
cations for health, education, and the workplace. Psychological Science
in the Public Interest, 5, 69 –106.
Ehrlinger, J., & Dunning, D. (2003). How chronic self-views influence
(and potentially mislead) estimates of performance. Journal of Person-
ality and Social Psychology, 84, 5–17.
Fiske, S. T., Neuberg, S. L., Beattie, A. E., & Milberg, S. J. (1987).
Category-based and attribute-based reactions to others: Some informa-
tional conditions of stereotyping and individuating processes. Journal of
Experimental Social Psychology, 23, 399 – 427.
Fiske, S. T., & Taylor, S. E. (1991). Social cognition (2nd ed.). New York:
McGraw Hill.
Forbes, C. E., Schmader, T., & Allen, J. J. B. (2008). The role of devaluing
and discounting in performance monitoring: A neurophysiological study
of minorities under threat. Social Cognitive and Affective Neuroscience,
3, 253–261.
Guenther, C. L., & Alicke, M. D. (2008). Self-enhancement and belief
perseverance. Journal of Experimental Social Psychology, 44, 706 –712.
Harris, M. M., & Schaubroeck, J. (1988). A meta-analysis of self super-
visor, self peer, and peer supervisor ratings. Personnel Psychology, 41,
43– 62.
Judd, C. M., Kenny, D. A., & McClelland, G. H. (2001). Estimating and
testing mediation and moderation in within-subject designs. Psycholog-
ical Methods, 6, 115–134.
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.
Kelley, C. M., & Lindsay, D. S. (1993). Remembering mistaken for
knowing: Ease of retrieval as a basis for confidence in answers to
general knowledge questions. Journal of Memory and Language, 32,
Lee, L., Frederick, S., & Ariely, D. (2006). Try it, you’ll like it: The
influence of expectation, consumption, and revelation on preferences for
beer. Psychological Science, 17, 1054 –1058.
Levin, D. T., & Banaji, M. R. (2006). Distortions in the perceived lightness
of faces: The role of race categories. Journal of Experimental Psychol-
ogy: General, 135, 501–512.
Levin, I. P., & Gaeth, G. J. (1988). How consumers are affected by the
framing of attribute information before and after consuming the product.
Journal of Consumer Research, 15, 374 –378.
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.
Makens, J. C. (1965). Effect of brand preference upon consumers’ per-
ceived taste of turkey meat. Journal of Applied Psychology, 49, 261–
Marteau, T. M., Johnston, M., Wynne, G., & Evans, T. R. (1989). Cogni-
tive factors in the explanation of the mismatch between confidence and
competence in performing basic life support. Psychology and Health, 3,
Mast, M. S. (2005). Interpersonal hierarchy expectation: Introduction of a
new construct. Journal of Personality Assessment, 84, 287–295.
McClure, S. M., Li, J., Tomlin, D., Cypert, K. S., Montague, L. M., &
Montague, P. R. (2004). Neural correlates of behavioral preference for
culturally familiar drinks. Neuron, 44, 379 –387.
McFarland, C., Cheam, A., & Buehler, R. (2007). The perseverance effect
in the debriefing paradigm: Replication and extension. Journal of Ex-
perimental Social Psychology, 43, 233–240.
Nevid, J. S. (1981). Effects of brand labeling on ratings of product quality.
Perceptual and Motor Skills, 53, 407– 410.
Olson, J. C., & Dover, P. A. (1978). Cognitive effects of deceptive
advertising. Journal of Marketing Research, 15, 29 –38.
Peng, K., Nisbett, R., & Wong, N. (1997). Validity problems comparing
value across cultures and possible solutions. Psychological Methods, 2,
329 –344.
Plassman, H., O’Doherty, J., Shiv, B., & Rangel, A. (2008). Marketing
actions can modulate neural representations of experienced pleasantness.
Proceedings of the National Academy of Sciences, USA, 105, 1050 –
Reber, R., & Schwarz, N. (1999). Effects of perceptual fluency on judg-
ments of truth. Consciousness and Cognition, 8, 338 –342.
Ross, L., Lepper, M. R., & Hubbard, M. (1975). Perseverance in self-
perceptions and social perception: Biased attributional processing in the
debriefing paradigm. Journal of Personality and Social Psychology, 32,
880 – 892.
Sagar, H., & Schofield, J. W. (1980). Racial and behavioral cues in Black
and White children’s perceptions of ambiguously aggressive acts. Jour-
nal of Personality and Social Psychology, 39, 590 –598.
Sanford, A. J., Fay, N., Stewart, A., & Moxey, L. (2002). Perspective in
statements of quantity, with implications for consumer psychology.
Psychological Science, 13, 130 –134.
Schmader, T., Johns, M., & Forbes, C. (2008). An integrated process model
of stereotype threat effects on performance. Psychological Review, 115,
336 –356.
Schwartz, B. L., & Metcalfe, J. (1992). Cue familiarity but not target
retrievability enhances feeling-of-knowing judgments. Journal of Exper-
imental Psychology: Learning, Memory, and Cognition, 18, 1074 –1083.
Schwarz, N. (2004). Metacognitive experiences in consumer judgment and
decision making. Journal of Consumer Psychology, 14, 332–348.
Schwarz, N., Sanna, L. J., Skurnik, I., & Yoon, C. (2007). Metacognitive
experiences and the intricacies of setting people straight: Implications
for debiasing and public information campaigns. In M. Zanna (Ed.),
Advances in experimental social psychology (Vol. 39, pp. 127–161). San
Diego, CA: Elsevier.
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.
Steele, C. M., & Aronson, J. (1995). Stereotype threat and the intellectual
test performance of African Americans. Journal of Personality and
Social Psychology, 69, 797– 811.
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.
von Hippel, W., Sekaquaptewa, D., & Vargas, P. (1995). On the role of
encoding processes in stereotype maintenance. Advances in Experimen-
tal Social Psychology, 27, 177–254.
Wansink, B., Park, S. B., Sonka, S., & Morganosky, M. (2000). How soy
labeling influences preference and taste. International Food and Agri-
business Management Review, 3, 85–94.
Wardle, J., & Solomons, W. (1994). Naughty but nice: A laboratory study
of health information and food preferences in a community sample.
Health Psychology, 13, 180 –183.
Wells, G. L., & Bradfield, A. L. (1999). Distortions in eyewitness recol-
lections: Can the postidentification-feedback effect be moderated? Psy-
chological Science, 10, 138 –144.
Received July 30, 2008
Revision received July 3, 2009
Accepted July 7, 2009
E-Mail Notification of Your Latest Issue Online!
Would you like to know when the next issue of your favorite APA journal will be available
online? This service is now available to you. Sign up at and you will be
notified by e-mail when issues of interest to you become available!
... As a result, several studies have focused on finding the causes of this cognitive bias, indicating some of them the origin in the motivation and self-image protection (Blanton et al., 2001), while others are inclined toward a limited information processing (Chambers & Windschitl, 2004), or previously conceived beliefs about their skill and knowledge (Critcher & Dunning, 2009). A recent work suggests that metacognitive judgments may be affected by motivation, for example, when students make predictions about their performance on a forthcoming exam, they may, explicitly or implicitly, take into account the desired grades (Bol et al., 2005). ...
... Students who held illusions of competence on the midterm exam tended to do so throughout the end of semester, failing feedback to obtain a more accurate self-awareness of their low performance. That is, as Critcher and Dunning, (2009) point out, cognitive bias on performance is not affected by students' concrete experiences of similar exams, but by preconceived notions of their ability. There are different forms of intervention that can reduce this effect and that can serve as metacognitive training, such as home activities or group work, with the aim of achieving a better perception of one's own performance (de Bruin et al., 2017). ...
Full-text available
This work analyzes whether there is a cognitive bias between the ideal perception of the skills and the real performance in an introductory physics class, and additionally, whether predictions of students’ performance are related to various motivational variables. We examined through a validated survey and network analysis the relationship between several motivational aspects and volitional variables with the accuracy of their predictions. The results show that the students present a motivational bias when students’ desires were considered, mainly in the students with low academic performance. Finally, it is necessary to explore the development of specific interventions that target the motivations of students, in order to be effective, and to reduce the gap between expected and actual grades, increasing students’ metacognitive skills and thus their academic performance.
... Students in the rllrbl condition seemed to over-rate their capacities and clearly experienced at a later stage the complexity of what they judged easier earlier. Comparable findings have been reported by Critcher and Dunning (2009). As these authors indicate, people often have surprisingly poor insight into their skills and abilities. ...
This study reports on the impact of two alternative interventions to increase undergraduate students’ research competence. In one condition students started early in the research-based learning (rbl) approach and later followed a research-led learning (rll) approach. In the second condition students started early in the rll approach and followed an rbl approach later. Research activities in both conditions were linked to a regular university course in social science. Following a 12-week crossover design, the differential impact was studied by looking at actual changes in students’ (1) research competence, (2) research self-efficacy and (3) motivation to do research – before, during and after the intervention. Focus group discussions (fgd s) helped to collect qualitative data at the end of the intervention. Analysis of the results pointed to a significantly higher impact on students’ research competence of studying first in the rbl context. Students starting in the rll condition and experiencing rbl only after the crossover moment also improved, but did not catch up. The qualitative data further underline the stronger positive impact of rbl on students’ research competence.
... Another potential limitation associate with our panel research is the use of self-assessment data, which poses the risk for overly favorable responses that may not accurately reflect actual performance (Critcher & Dunning, 2009). However, the results in this study do demonstrate that younger, less experienced professionals did respond in ways that showed awareness of their weaknesses. ...
Practices of ethical leadership in public relations can be context-specific and they can influence organizational effectiveness. By conducting a national survey, this study examines female public relations professionals’ perspectives on ethical leadership. The results suggest that the majority of female professionals feel ready and confident in providing ethics counseling as needed. Most importantly, the highest ranked public relations leaders’ ethical conduct help reinforce female professionals’ ethical practice. Female professionals indicate it is necessary to use multiple strategies to build and enact influence as an ethical leader in public relations. Theoretical and practical implications are discussed.
... While overarching evaluation frameworks regarding the assessment of ILOs exist (e.g., DEVISE; Phillips et al. 2014), only a few instruments to assess science inquiry skills are available, and SR skills are mentioned in only one percent of the literature reviewed (Stylinski et al. 2020). Therefore, the evaluation of SR skills in CS projects less often relies on standardized tests than on surveys of self-reported confidence in performing the SR skills (see overview in Stylinski et al. 2020)-despite validity concerns regarding self-reports (Critcher and Dunning 2009). To ensure that conclusions that are drawn from evaluations of ILOs in CS are valid, assessment instruments that do not rely solely on self-reports should be developed (Phillips et al. 2018). ...
Full-text available
Citizen science (CS) projects engage citizens for research purposes and promote individual learning outcomes such as scientific reasoning (SR) skills. SR refers to participants’ skills to solve problems scientifically. However, the evaluation of CS projects’ effects on learning outcomes has suffered from a lack of assessment instruments and resources. Assessments of SR have most often been validated in the context of formal education. They do not contextualize items to be authentic or to represent a wide variety of disciplines and contexts in CS research. Here, we describe the development of an assessment instrument that can be flexibly adapted to different CS research contexts. Furthermore, we show that this assessment instrument, the SR questionnaire, provides valid conclusions about participants’ SR skills. We found that the deep-structure and surface features of the items in the SR questionnaire represent the thinking processes associated with SR to a substantial extent. We suggest that practitioners and researchers consider these item features in future adaptations of the SR questionnaire. This will most likely enable them to draw valid conclusions about participants’ SR skills and to gain a deeper understanding of participants’ SR skills in CS project evaluation.
... People like to think of themselves as attractive and intelligent, and this positive bias is partially maintained through motivated beliefs (Critcher & Dunning, 2009). That is, people typically draw towards information that supports positive perception of the self and reject information that disconfirms positive biases (Vignoles, Regalia, Manzi, Golledge, & Scabini, 2006). ...
... In particular, just as people's memories are sometimes shaped by their self-beliefs, so too can people's experiences in the moment be shaped by their self-beliefs (e.g. Critcher & Dunning, 2009). Indeed, people do not always engage in experiential processes governed by their low-level associative reactions in the moment, as evidenced by various studies that employ experimental manipulations to increase reliance on these processes (e.g. ...
People have a fascinating capacity to picture their actions from an external vantage point. Much of the research on this third-person imagery has focused on the specific effects it has on cognition due to the elements of episodic experience that it lacks relative to first-person imagery. Other research focuses on the information that the third-person provides that first-person imagery lacks. We propose a more systematic approach that conceptualises how third-person imagery’s various effects interrelate due to a common underlying social-cognitive function. Specifically, we outline an integrative model proposing that third-person and first-person imagery cause people to adopt qualitatively distinct processing styles. This model explains many of the diverse effects that have been documented in the literature and helps reconcile seemingly discrepant findings. We conclude with recommendations for strategies to more systematically investigate the functions of visual perspective in mental imagery to build more comprehensive understanding of this phenomenological variable.
... Students responded to nine queries (Table 2) with a numeric response to each question ranging from "strongly disagree" (1) to "strongly agree" (5) (Figure 1). Indeed, self-reported data falls short of the gold standard for measuring impacts (Critcher & Dunning, 2009), and future iterations where schedules are less unexpected should rigorously explore any efficacy of novel teaching techniques. How such outcomes compare to other potential uses of class and laboratory time have not been empirically explored but are important alternative models (e.g., structured online laboratories) that should be used to determine whether engaging in such an activity optimally targets learning outcomes amidst distancing. ...
Full-text available
Inquiry‐based components of ecology curricula can be valuable, exposing students to what it means to do science, from conceiving of a meaningful question to effectively disseminating results to an audience. Here, we describe two approaches for implementing independent, remote research for undergraduates enacted in the spring semester of 2020 at Reed College in Portland, OR, reporting case studies from an intermediate‐level ecology course and an interdisciplinary environmental science course. We report on both the challenges as well as the novel opportunities for independent research projects in such a setting, the details of how projects were implemented, the tools and resources that may help facilitate such endeavors, as well as perceptions on the effectiveness of this endeavor by students. As institutes of higher education continue to operate in an online learning environment, we hope these materials help spark a discussion about how to engage in meaningful research experiences as part of coursework in the COVID‐19 era and beyond. We report on both the challenges as well as the novel opportunities for independent ecology research projects in a remote learning setting, the details of how projects were implemented, the tools and resources that may help facilitate such endeavors, as well as perceptions on the effectiveness of this endeavor by students.
... This self-reported knowledge gain can be due the participants' flawed understanding of their own capability (Chiaburu et al., 2014). This could also be an influence of their personality traits or situational indications from others for acceptance or approval (Berry, Page and Sackett, 2007;Critcher and Dunning, 2009;Steenkamp, Jong and Baumgartner, 2010). ...
Full-text available
Sexual Harassment (SH) in the workplace is evidenced to decrease employees’ productivity as well as negatively impact financial performance of an organisation. Training literature re-establishes the need for SH training and advice on how to implement it. In India, according to the POSH Act 2013, SH training is mandatory for organisations with more than 50 employees. While mandatory, there is a knowledge gap on understanding whether the existing training outlined by POSH Act 2013 is effective in reducing SH in the workplace. This study adds to the body of knowledge and literature on the effectiveness of a SH training in an Indian context. Firstly, there was a review of existing literature on SH training to examine its effectiveness on participants in educational and workplace settings, with more emphasis on the latter. Key findings of the literature established an effect on attitude of participants, knowledge of legal aspects and identification of SH incidents. Furthermore, a survey questionnaire was employed to evaluate these three variables namely, legal knowledge, attitude towards SH and ability to identify SH in an Indian organisation. The results showed that while training did not have any effect on knowledge gain, participants had a positive attitude of reporting SH and there was significant ability to identify SH situations. The findings also highlighted participants’ self-reported knowledge gain on POSH Act and its importance in the workplace. Additionally, the findings indicated a correlation between organisational culture (senior management’s intolerance towards SH) and participants’ attitude towards SH. The results also revealed that gender of the participants had no significant difference in attitude or ability to identify SH.
... Low performers (individuals who do not earn high scores on a test using an objective scale) tend to over-estimate their performance percentile on a task while high performers (individuals who earn high scores measured on an objective scale) tend to under-estimate their performance percentile on the same task, with the direction of this perceptual mismatch extending in both directions (Sieber, 1979). Empirically, this paradigm has been used successfully in many different tasks to elicit the DKE on such tasks as microeconomics college examinations (Ryvkin, Krajč, & Ortmann, 2012), logical reasoning (Schlösser, Dunning, Johnson, & Kruger, 2013), cognitive reflection (Pennycook, Ross, Koehler, & Fugelsang, 2017), size judgements (Sanchez, 2016), finance (Atir, Rosenzweig, & Dunning, 2015) and computer programming (Critcher & Dunning, 2009). More broadly, the effect has been referred to in contexts of driving (Svenson, 1981), aviation (Pavel, Robertson, & Harrison, 2012) and professors rating their own teaching skills (Cross, 1977). ...
The Dunning‐Kruger Effect (DKE) is a metacognitive phenomenon of illusory superiority in which individuals who perform poorly on a task believe they performed better than others, yet individuals who performed very well believe they under‐performed compared to others. This phenomenon has yet to be directly explored in episodic memory, nor explored for physiological correlates or reaction times. We designed a novel method to elicit the DKE via a test of item recognition while electroencephalography (EEG) was recorded. Throughout the task, participants were asked to estimate the percentile in which they performed compared to others. Results revealed participants in the bottom 25th percentile overestimated their percentile, while participants in the top 75th percentile underestimated their percentile, exhibiting the classic DKE. Reaction time measures revealed a condition by group interaction whereby over‐estimators responded faster than under‐estimators when estimating being in the top percentile and responded slower when estimating being in the bottom percentile. Between‐group EEG differences were evident between over‐estimators and under‐estimators during Dunning‐Kruger responses, which revealed FN400‐like effects of familiarity supporting differences for over‐estimators, whereas ‘old‐new’ memory event‐related potential (ERP) effects revealed a late parietal component (LPC) associated with recollection‐based processing for under‐estimators that was not evident for over‐estimators. Findings suggest over‐ and under‐estimators use differing cognitive processes when assessing their performance, such that under‐estimators may rely on recollection during memory while over‐estimators may draw upon excess familiarity when over‐estimating their performance. Episodic memory thus appears to play a contributory role in metacognitive judgments of illusory superiority.
... For example, these construction processes might be shaped or constrained by people's preexisting beliefs about how their minds work or by people's general beliefs regarding their judgmental abilities, confidence, or opinionatedness. Such effects might be analogous to the ways in which general self-beliefs guide people's specific performance estimates (Ehrlinger & Dunning, 2003), including the possibility that general beliefs might bias interpretation of the available lower-order cues (e.g., by biasing interpretation of retrieval fluency; Critcher & Dunning, 2009). Another potential top-down influence is that a person's general sense of certainty or any confidence-laden experiential mindsets (e.g., chronic anger or positive affect vs. chronic fear or negative affect) might be mis-attributed to any salient mental experience Clore & Parrott, 1994), including an attitude one is currently considering. ...
The certainty with which people hold their attitudes is an important consideration because attitudes held with certainty better predict judgment and behavior than attitudes held with doubt. However, little is known about whether people's assessments of their certainty reflect a disposition to hold attitudes with confidence. Adapting methods used to document individual differences in people's attitudes, the present research demonstrates that the certainty with which people hold any given attitude is in part a reflection of a relatively stable disposition. Across 5 studies and 6 samples (total N = 106,050), we demonstrate dispositional variability in attitude certainty and show that it is related to but distinct from confidence in other judgmental domains. We also demonstrate that dispositional attitude certainty may be useful in predicting certainty in newly formed evaluations (Study 3) and an important consequence of certainty-attitude-behavior correspondence (as indicated by reports of behavioral intentions and recent behavior; Study 4 and Student Sample in Study 5). Furthermore, we demonstrate that dispositional attitude certainty is relatively stable over time (Study 5). Results are discussed with respect to potential mechanisms and boundary conditions relating to dispositional attitude certainty, the implications of these individual differences for attitudes and persuasion, as well as the potential origins of dispositional attitude certainty. (PsycInfo Database Record (c) 2020 APA, all rights reserved).
Full-text available
In this article, we attempt to distinguish between the properties of moderator and mediator variables at a number of levels. First, we seek to make theorists and researchers aware of the importance of not using the terms moderator and mediator interchangeably by carefully elaborating, both conceptually and strategically, the many ways in which moderators and mediators differ. We then go beyond this largely pedagogical function and delineate the conceptual and strategic implications of making use of such distinctions with regard to a wide range of phenomena, including control and stress, attitudes, and personality traits. We also provide a specific compendium of analytic procedures appropriate for making the most effective use of the moderator and mediator distinction, both separately and in terms of a broader causal system that includes both moderators and mediators. (46 ref) (PsycINFO Database Record (c) 2012 APA, all rights reserved)
Although much has been written about deception in advertising, no studies have been reported in which a deception and its impact on consumers were demonstrated empirically. The authors present a behavioral definition of deception and illustrate its operationalization in the context of a longitudinal experiment in which the effects of an explicit, deceptive product claim on a variety of cognitive variables were measured both before and after product trial. Issues related to the measurement of deception seriousness are emphasized. The basic approach appears generalizable to nonexperimental studies of real-world deception.
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.
This book examines how standards and expectancies affect judgments of others and the self. Standards are points of comparison, expectancies are beliefs about the future, and both serve as frames of reference against which current events and people (including the self) are experienced. The central theme of the book is that judgments can be characterized as either assimilative or contrastive in nature. Assimilation occurs when the target of evaluation (another person, the self) is pulled toward or judged consistently with the standard or expectation, and contrast occurs when the target is differentiated from (judged in a direction opposite) the comparative frame. The book considers factors that determine whether assimilation versus contrast occurs, and focuses on the roles of contextual cues, the self, and stereotypes as standards for judging others, and the roles of internalized guides, stereotypes, and other people for judging the self.
As a company tries to find the factors accounting for strong and weak markets, typical consumer explanations for both tend to be in terms of the physical attributes of the product. Carling Brewing Company used a relatively inexpensive experiment to help dichotomize contributing influences as being either product or marketing oriented and, also, to indicate the magnitude of the marketing influence for various brands. The experiment involved the use of groups of beer drinkers that tasted (drank) and rated beer from nude bottles and from labeled bottles.
Although much has been written about deception in advertising, no studies have been reported in which a deception and its impact on consumers were demonstrated empirically. The authors present a behavioral definition of deception and illustrate its operationalization in the context of a longitudinal experiment in which the effects of an explicit, deceptive product claim on a variety of cognitive variables were measured both before and after product trial. Issues related to the measurement of deception seriousness are emphasized. The basic approach appears generalizable to nonexperimental studies of real-world deception.
The primary hypotheses of this study were that students high in self-esteem would evaluate their own performance more favorably, and interpret a teacher's evaluative feedback more favorably, than would students low in self-esteem. Students were prescreened using the short form of the Rosenberg Self-Esteem Scale and those who were high and low in self-esteem were selected to participate in this study. In response to a student's performance on an analogies test, the teacher conveyed either positive feedback, negative feedback or no feedback. Questionnaires assessed students' self-evaluations and their perceptions of the teacher's evaluation of their performance. Results demonstrated that students high in self-esteem evaluated their own performance more favorably, and saw the teacher as evaluating their performance more favorably, than students low in self-esteem. Implications for consistency perspectives on the role of the self in social information processing are discussed.
The product quality of two status-oriented brands of carbonated bottled water and of one low-status popularly-priced brand was rated by 24 college students each in brand labeled and unlabeled conditions. The results supported the influence of product image on consumers' judgments of product quality, but it was suggested that the salience of such extrinsic cues might depend on the breadth of the consumers' evaluative frame of reference for judging particular classes of products.