Perspective Taking as Egocentric Anchoring and Adjustment

Abstract

The authors propose that people adopt others' perspectives by serially adjusting from their own. As predicted, estimates of others' perceptions were consistent with one's own but differed in a manner consistent with serial adjustment (Study 1). Participants were slower to indicate that another's perception would be different from--rather than similar to--their own (Study 2). Egocentric biases increased under time pressure (Study 2) and decreased with accuracy incentives (Study 3). Egocentric biases also increased when participants were more inclined to accept plausible values encountered early in the adjustment process than when inclined to reject them (Study 4). Finally, adjustments tend to be insufficient, in part, because people stop adjusting once a plausible estimate is reached (Study 5).

Full text

Perspective Taking as Egocentric Anchoring and Adjustment
Nicholas Epley
Harvard University
Boaz Keysar
University of Chicago
Leaf Van Boven
University of Colorado, Boulder
Thomas Gilovich
Cornell University
The authors propose that people adopt others’ perspectives by serially adjusting from their own. As
predicted, estimates of others’ perceptions were consistent with one’s own but differed in a manner
consistent with serial adjustment (Study 1). Participants were slower to indicate that another’s perception
would be different from—rather than similar to—their own (Study 2). Egocentric biases increased under
time pressure (Study 2) and decreased with accuracy incentives (Study 3). Egocentric biases also
increased when participants were more inclined to accept plausible values encountered early in the
adjustment process than when inclined to reject them (Study 4). Finally, adjustments tend to be
insufficient, in part, because people stop adjusting once a plausible estimate is reached (Study 5).
We have endeavored to show . . . that thought in the child is ego-
centric, i.e., that the child thinks for himself without troubling to make
himself understood nor to place himself at the other person’s point of
view. . . . If this be the case, we must expect childish reasoning to
differ very considerably from ours, to be deductive and above all less
rigorous. (Piaget, 1959, p. 1)
Children view their perceptions of the world as accurate reflec-
tions of its actual properties and only with experience come to
learn otherwise. People’s perceptions of the world are construc-
tions rather than veridical reflections and thus are not only occa-
sionally wrong but also occasionally differ from the perceptions of
others. Childhood thinking is less rigorous than adults’ because it
fails to correct for the subtle but significant fact that our percep-
tions are perspective bound.
Adults certainly know better. Parents, for instance, often excuse
themselves from formally evaluating their own children because
they know their assessments are positively biased (Wegener &
Petty, 1995). People sometimes recognize that their first impres-
sions have been influenced by idiosyncratic associations, expect-
ancies, or stereotypes (Gilbert & Malone, 1995). People also
generally expect those on the opposing side of an argument to
interpret an ambiguous action differently than they have them-
selves (Robinson, Keltner, Ward, & Ross, 1995). People recognize
that their assessments can be influenced by their expectations, that
behavior is not always what it seems, and that motivation can
influence judgment, making one’s own perspective on the world
potentially unique.
As Piaget recognized, adults come to view the world less ego-
centrically than children, although they do not outgrow their child-
hood tendencies altogether. Many social judgments, even among
adults, are still egocentrically biased. People tend to believe, for
example, that their internal states and intentions are more trans-
parent to others than they actually are (Gilovich, Savitsky, &
Medvec, 1998); they overestimate the extent to which others attend
to those states (Gilovich, Medvec, & Savitsky, 2000) and exag-
gerate the extent to which others will share their thoughts and
feelings (Keysar, 1994; Nickerson, 1999; Ross & Ward, 1996; Van
Boven, Dunning, & Loewenstein, 2000). It is clear that even adults
can have a difficult time setting aside their own perspective when
considering the perceptions of others. Our goal is to better under-
stand why these egocentric biases arise by examining the psycho-
logical processes that guide perspective taking.
Like other theorists, we contend that people do not set aside
their own perspective when adopting another’s but instead use it as
a starting point, or judgmental anchor (e.g., Davis, Hoch, &
Ragsdale, 1986; Nickerson, 1999). Because people generally have
eyes, ears, and other sensory organs that operate more or less the
same, it is reasonable for people to assume that others will “see”
or experience the world similarly. However, because adults also
recognize that people have motivations, beliefs, and backgrounds
that can lead to different perceptions or interpretations of the same
stimulus, they understand that this egocentric anchor sometimes
needs adjustment to accommodate differences between themselves
and others.
Nicholas Epley, Department of Psychology, Harvard University; Boaz
Keysar, Department of Psychology, University of Chicago; Leaf Van
Boven, Department of Psychology, University of Colorado, Boulder;
Thomas Gilovich, Department of Psychology, Cornell University.
This research was supported by a visiting fellow position to the Peter
Wall Institute to Nicholas Epley; by Research Grants SES0241544 and
SBR9809262 from the National Science Foundation; by PHS Grants R01
MH49685-06A1 and R29 MH49685 from the National Institutes of Health;
and by a Social Sciences Division Research Grant. We thank Patricia
Alman, Johnny Chueh, Ryan Larrenaga, Gilbert Lo, Erin Rapien, Bree Tse,
and Kevin Van Aelst for conducting the experiments; Linda Ginzel, Rachel
Giora, William Horton, Marcel Just, Howard Nusbaum, and Steven Slo-
man for advice during the course of this research; and Daniel Gilbert for
comments on a draft of this article. We also thank Mark Andrews and
Michael Owren for their assistance creating the audio materials in Study 4.
Correspondence concerning this article should be addressed to Nicholas
Epley, Harvard University, Department of Psychology, William James
Hall, 33 Kirkland Street, Cambridge, MA 02138. E-mail: epley@wjh
.harvard.edu
Journal of Personality and Social Psychology Copyright 2004 by the American Psychological Association
2004, Vol. 87, No. 3, 327–339 0022-3514/04/$12.00 DOI: 10.1037/0022-3514.87.3.327
327

Specifically, we suggest that people adopt others’ perspectives
by initially anchoring on their own perspective and only subse-
quently, serially, and effortfully accounting for differences be-
tween themselves and others until a plausible estimate is reached,
using what Tversky & Kahneman (1974) described as the “anchor-
ing and adjustment heuristic.” We argue that this adjustment is
comprised of a series of discrete miniadjustments coupled with
hypothesis tests: People “jump” some amount from their original
egocentric anchor and evaluate whether this new perspective plau-
sibly captures the other’s perception (Epley & Gilovich, 2001). If
so, adjustment stops. If not, another jump is made to a perspective
more discrepant from their own, and so on, until a plausible
assessment is reached—one that accommodates the difference
between one’s own and another’s perspective.
Like other judgmental heuristics, anchoring and adjustment
simplifies the complicated assessment of another’s perspective by
substituting one’s own perception and adjusting as needed. This
model of perspective taking is therefore most likely to be engaged
when one’s own perspective is readily accessible but another’s
perspective must be inferred. It may apply less frequently to
predictions of family or close friends whose perspectives are
already well known and need not be inferred or for members of
radically different outgroups for whom one’s own perspective may
seem irrelevant. Anchoring and adjustment may also apply less
frequently to predictions of outgroups for which one’s own per-
spective may seem irrelevant or for which one has highly acces-
sible stereotypes (Ames, in press).
Although the anchoring and adjustment heuristic is clearly use-
ful, the adjustments made from one’s own perspective should tend
to be insufficient—and give rise to egocentric biases—because
they terminate once a plausible estimate is reached (Quattrone,
1982). This “satisficing” produces responses biased toward the
egocentric side of the distribution of plausible estimates. This
model differs from a simpler model of egocentric projection
(Krueger & Clement, 1994) because of the effortful, deliberate,
and serial adjustment process that moderates a more extreme
egocentric perception. It also differs from more general correction
models of human judgment that may work through a process of
integration or substitution (e.g., Gilbert, 1989, 2002; Trope &
Gaunt, 2000; Wegener & Petty, 1995) in that the present model
clearly specifies the exact nature of the serial adjustment process
and articulates a stop rule that can explain why adjustments tend to
be insufficient. This model therefore builds on previous research
by treating anchoring and (serial) adjustment as a literal descrip-
tion of perspective taking and not simply as a metaphorical de-
scription of a behavioral outcome (e.g., Krueger & Clement, 1995;
Nickerson, 1999).
Although the anchoring and adjustment heuristic has been in-
voked to explain egocentric biases in perspective taking (Gilovich
et al., 1998, 2000; Keysar & Barr, 2002; Nickerson, 1999), no
direct evidence for this mechanism has been reported. The five
experiments in this article were designed to provide such evidence
by testing a set of predictions derived directly from this model. In
particular, if perspective taking entails effortful serial adjustment
from an egocentric anchor, then factors that inhibit or facilitate this
process should increase or decrease egocentric biases. For exam-
ple, diminishing the time allowed to render a judgment or taxing
one’s cognitive resources should lower the threshold for accepting
an estimate as plausible and may therefore shorten the process of
adjustment. In addition, if adjustment proceeds by testing a series
of hypotheses increasingly discrepant from an egocentric anchor,
then egocentric judgment should be influenced by factors that raise
or lower individuals’ threshold for accepting hypotheses.
This anchoring and adjustment model of adult perspective tak-
ing fits at least three empirical findings. First, manipulations that
hinder people’s ability to expend cognitive resources also diminish
their ability to correct for biases in their immediate sensory expe-
rience. Individuals whose heads are tilted, for example, will judge
straight lines as tilted if they are simultaneously distracted but will
recognize the biasing influence of their tilted head and identify the
line as straight if they are not distracted (Rock & Nijhawan, 1989).
Similarly, people are more likely to use their chronic mood state as
a guide for evaluating ambiguous stimuli when forced to respond
quickly—an extraneous influence that diminishes when people
respond at their leisure and are better able to correct for their
mood’s influence (Gilbert & Gill, 2000). Although this research
was not directly concerned with perspective taking, we suspect that
the psychological process of correcting one’s perception to accom-
modate biasing sensory experiences may involve a process of
adjustment very similar to that in our proposed model of perspec-
tive taking.
Second, measurements of people’s eye gaze during perspective
taking tasks reveal an egocentric anchor: People frequently fixate
on locations that only an egocentric analysis would suggest. In one
experiment, for example, participants followed another’s instruc-
tions to move objects around a rectangular array of boxes (Keysar,
Barr, Balin, & Brauner, 2000). Although most of the objects were
visible to both participants, some were occluded from the “direc-
tor’s” view but easily seen by the “addressee.” Most addressees
initially interpreted the director’s instructions egocentrically, look-
ing at and occasionally reaching for objects that were visible only
from their own perspective and only later looking at the mutually
visible target.
Third, participants’ self-reports of their perspective-taking ef-
forts are consistent with an egocentric anchoring and adjustment
account. In the only study we know of that presents such self-
report data, participants were asked to don an embarrassing T-shirt
(emblazoned with the image of John Tesh or Vanilla Ice) and
asked to estimate the percentage of their peers seated in the room
who would correctly identify the person pictured on their shirt
(Study 3, Gilovich et al., 2000). Participants were obviously well
aware of the person pictured on their shirt and, as a result, tended
to overestimate the number of peers who would notice it as
well—a phenomenon dubbed the Spotlight Effect. More important
for our present purposes, 73% of participants reported considering
another number before making their estimate, and of these, 72%
reported considering a number higher than their final estimate.
Because serial adjustments are deliberate and effortful, they should
frequently be available to conscious awareness and thus appear in
self-report data (Epley & Gilovich, 2001).
Our goals in this research were to go beyond previous work by
directly testing whether people adopt others’ perspectives by se-
rially adjusting from their own and to explain why such adjust-
ments tend to be insufficient. We conducted five experiments, each
with a unique methodology, to provide converging evidence for
our anchoring and serial adjustment account of adult perspective
taking. These experiments go beyond previous work by precisely
articulating the nature of the adjustment involved, by directly
328
EPLEY, KEYSAR, VAN BOVEN, AND GILOVICH

documenting its existence, and by providing a mechanism to
explain residual egocentric biases in adult perspective taking.
Study 1: Sarcastic Messages
Communication is often ambiguous. The same expression may
be interpreted as either an insult or a compliment, as helpful or
hurtful, as serious or sarcastic. Although listeners seem adept at
decoding the underlying meaning of a speaker’s ambiguous mes-
sage (Keysar, 1994; Keysar & Barr, 2002), they may not be as
adept as speakers think. Because speakers are aware of their own
intentions, they may initially assume that their intentions are
obvious and only subsequently adjust in a serial fashion to accom-
modate possible ambiguities.
This anchoring and adjustment account makes two predictions.
First, individuals who have privileged information that clarifies the
meaning of an ambiguous message should be aware of their need
to adjust and therefore expect the communication’s meaning to be
less clear to people who do not have clarifying information.
Second, because such serial adjustments are likely to terminate as
soon as a plausible estimate is reached, egocentric biases should
emerge. As a result, individuals with clarifying information should
still overestimate the clarity of those messages to listeners without
clarifying information.
We tested these predictions by presenting participants with
messages ostensibly left on someone’s answering machine. The
messages were intentionally ambiguous and could be interpreted
as either sincere or sarcastic. The ambiguity was resolved for
participants by giving them clarifying information about the events
leading up to the message, making it either clearly sincere or
clearly sarcastic. Participants in the intention condition then indi-
cated the speaker’s intended meaning; participants in the interpre-
tation condition anticipated how a person who lacked clarifying
information would interpret the messages. We predicted that par-
ticipants would recognize that the message would sound more
ambiguous to listeners without the clarifying information and that
responses in the interpretation condition would therefore be more
moderate than in the intention condition. However, because ad-
justments tend to be insufficient, we expected egocentric biases to
remain and predicted that participants would expect the speaker’s
intention to be clearest to uninformed listeners when it was clearest
to them.
Method
Participants. Seventy-two undergraduates at the University of Chi-
cago participated either for pay or partial fulfillment of a course require-
ment. Two participants were not native English speakers and were there-
fore excluded, leaving 70 in the following analyses.
Procedure. Twelve different scenarios described events in the life of
“Tom Reton.” Each scenario appeared in one of two versions, one describ-
ing a negative event and the other describing a positive event. For instance,
one scenario described a recommendation that Tom received from his
friend Steve about a comedy show:
The other day, Tom was having dinner with two of his friends, Gina
and Steve. Gina urged them to go see a new comedian whose show
just opened in the area. “You’ve got to see him. I heard he’s just
hilarious.”
Each scenario appeared in one of two versions, one describing a negative
event and the other describing a positive event. For example, the positive
event version of the comedian scenario ended with: “Tom thought it was a
great show. He loved the comedian—in fact, he laughed so hard his
stomach hurt.” The negative event version ended with: “Tom followed her
advice but hated the comedian. He thought the guy was arrogant and
tedious.”
After reading each scenario, participants listened to a tape-recorded
message that Tom had ostensibly left on an answering machine. For both
positive and negative event versions of the comedian scenarios, for exam-
ple, participants heard Tom leave the following message:
Steve, this is Tom. How are things going? By the way, remember that
comedian Gina mentioned at dinner? I just saw him yesterday. All I
can say is that you have to see him yourself to believe how hilarious
he really is. Well, call me back when you get a chance and we’ll make
plans for the weekend.
We assumed that participants who read the negative event version (that
Tom hated the comedian) would interpret the message as sarcastic but
participants who read the positive event version (that Tom loved the
comedian) would interpret it as sincere. In all of the scenarios, the negative
event information implied that the proper interpretation of the message was
that it was sarcastic.
The speaker’s tone for all messages was relatively monotonous. The
scenarios were carefully constructed to make it clear that the listener (e.g.,
Steve) would not have access to the clarifying information. The scenarios
were presented in booklets, two per page. Each booklet contained all 12
scenarios, half with the positive event information and half with the
negative event information. To avoid confounding the valence of the event
information with the specific scenarios, we created two versions of the
booklets—one with one random order of positive and negative event
information and a second with its mirror image. These two versions
therefore allowed positive and negative event information to be manipu-
lated between participants for each of the 12 items. The order of the
scenarios was determined by a coin flip, with the provision that no more
than 3 scenarios of the same valence appear consecutively.
After reading each scenario and listening to its corresponding answering
machine message, participants randomly assigned to the intention condi-
tion (n ⫽ 30) judged whether Tom intended the message to be sarcastic.
Participants in the interpretation condition predicted whether the listener
would interpret the message as sarcastic. Specifically, participants indi-
cated whether the message was intended to be, or would be interpreted as,
sarcastic by circling “YES,”“MAYBE,” or “NO.”
Results and Discussion
We predicted that the event information participants read would
influence their interpretation of the message left on the answering
machine, leading participants who read a negative version of the
event to perceive a more sarcastic message than participants who
read a positive version. We also predicted that participants in the
interpretation condition would recognize that the message would
be less clear to uninformed listeners and so would use their own
interpretation as an anchor and adjust away from it to account for
the listeners’ lack of information.
To test this prediction, we first coded YES responses (indicating
the message was sarcastic) as 1, MAYBE as 0, and NO (indicating
the message was not sarcastic) as –1. Then, for each participant,
we averaged these scores across the six negative event scenarios
and the six positive event scenarios, and submitted them to a 2
(condition: intention or interpretation) ⫻ 2 (event information:
positive or negative) mixed model analysis of variance (ANOVA)
with repeated measures on the last factor. This analysis yielded a
significant main effect for event information, F(1, 68) ⫽ 172.65,
329
PERSPECTIVE TAKING AS ADJUSTMENT

p ⬍ .01, qualified by the predicted interaction, F(1, 68) ⫽ 15.86,
p ⬍ .01. Participants in the intention condition who read a negative
scenario rated the message as more sarcastic (M ⫽ .61) than did
those who read a positive scenario (M ⫽ –.52), t(68) ⫽ 11.33, p ⬍
.01. This difference remained, but was significantly smaller,
among participants in the interpretation condition (Ms⫽ .34 and
–.27, respectively, t(68) ⫽ 6.99, p ⬍ .1). The adjustments for both
information conditions were also significant. The mean sarcasm
score was significantly lower in the interpretation condition than in
the intention condition for negative information, t(68) ⫽ 3.19, p ⬍
.01, and it was significantly higher for the positive information
condition, t(68) ⫽⫺2.84, p ⬍ .01. These findings suggest that
participants used their own interpretation of the ambiguous mes-
sage as an anchor from which they adjusted to account for the
uninformed listeners’ lack of information.
We also predicted that participants would perceive the speakers’
intention to be clearer to an uninformed listener (i.e., Steve) to the
extent that the intention was clear to them. We tested this predic-
tion by first subtracting, separately for each condition, the mean
sarcasm score of the negative event version of each scenario from
the mean sarcasm score for the positive event version. We then
correlated, across scenarios, the magnitude of these difference
scores in the intention and interpretation conditions. As expected,
this analysis yielded a strong positive correlation between the
difference scores in the intention and interpretation conditions,
r(10) ⫽ .70, p ⬍ .05, indicating that scenarios that created a large
difference between the perceived intention of the speaker in the
positive and negative event versions also created large differences
in the interpretation condition. In other words, participants thought
the speaker’s intention would be clear to uninformed listeners to
the extent that the scenario made the speaker’s intention clear to
participants themselves.
These results are consistent with our claim that people adopt
others’ perspectives by adjusting from their own. Participants’
interpretations of ambiguous messages, as measured by their judg-
ments of the speaker’s intention, were influenced by the back-
ground information they received. Participants realized that the
speakers’ intention would be less clear to listeners who did not
have access to the same background information and therefore
adjusted from their interpretational anchor to accommodate this
informational difference.
Note that these results also indicate that participants’ adjust-
ments were insufficient: Participants in the interpretation condition
who read negative scenarios thought that uninformed listeners
would interpret the messages as somewhat more sarcastic than
participants who read positive scenarios. Such insufficiency is a
common, albeit not inevitable, result of serial adjustment (Epley &
Gilovich, 2004a, 2004b; Tversky & Kahneman, 1974). As we
directly address in Study 5, such insufficient adjustment can ex-
plain why perspective taking tends to be egocentrically biased.
Study 2: Under Pressure
We sought more direct evidence for adjustment in Study 2 by
both measuring and manipulating the amount of time involved in
perspective taking. Adjustment takes time because it involves
evaluating hypotheses ever further from an original anchor or
starting point (Epley & Gilovich, 2001). Responses made under
time pressure should therefore be closer to the egocentric anchor
than leisurely responses. In addition, participants should respond
more quickly when rendering a judgment that is consistent with
their own perspective (because it requires little or no adjustment)
than when rendering one that is inconsistent (because it requires
more adjustment). We tested these hypotheses in Study 2.
The procedure in Study 2 was similar to that in the interpretation
condition in Study 1. Participants read 12 scenarios—half describ-
ing a negative event and half describing a positive event—and
estimated whether an uninformed listener would interpret the
message as sincere or sarcastic. Some participants made these
judgments at their leisure, as in Study 1; others made them under
time pressure. If perspective taking involves serial adjustment
from the anchor of one’s own perspective, then limiting the time
allowed for adjustment should result in judgments closer to one’s
own perspective. We therefore expected hurried participants to be
more egocentric than unhurried participants.
We also measured the speed with which participants rendered
their judgments. If people engage in an effortful, time-consuming
process of serial adjustment, then they should take longer to
conclude that others will see the world differently than to indicate
that others will see it similarly. We therefore predicted that hurried
participants would take less time to indicate that uninformed
listeners would share their interpretation than to indicate that they
would have a different interpretation.
1
Method
Participants. Ninety-six undergraduates at the University of Chicago
participated either for pay or for partial fulfillment of course credit. All
were native English speakers.
Procedure. Study 2 used the same 12 scenarios and messages as Study
1, each accompanied by either negative or positive event information. All
participants evaluated whether an uninformed listener would perceive a
message as sarcastic (yes or no).
The procedure was similar to Study 1 except that the scenarios were
presented on a computer screen rather than in booklets. After reading each
scenario, participants pressed a key that removed the scenario from the
screen and played the answering machine message over headphones.
Because response latency was an important variable in this experiment,
each message was divided into two parts to more carefully isolate the key
components. The first part of the message included the potentially sarcastic
phrase (e.g., “you have to see him yourself to believe how hilarious he
really is”) and was immediately followed by a beep intended to prompt the
participant’s judgment. Participants indicated whether the message would
be interpreted as sarcastic by pressing keys labeled either “yes” or “no.”
2
After responding, participants heard the second part of the message that
included the concluding remarks (e.g., “well, call me back when you get a
chance, and we’ll make plans for the weekend”).
Before starting the computer program, the experimenter encouraged all
participants to take their time reading each scenario. Participants randomly
assigned to the leisurely condition (n ⫽ 48) were then instructed—in
1
Response times in the leisurely condition are not analyzed because they
are less diagnostic of the actual judgment process than response times in
the hurried condition. Long response times in the leisurely condition
include not only the time spent to arrive at a judgment but also subsequent
deliberations and other extraneous noise that is absent in the hurried
condition.
2
Because we were interested in the speed with which participants
provided expectancy-consistent versus expectancy-inconsistent responses,
Study 2 did not include the “maybe” response option from Study 1.
330
EPLEY, KEYSAR, VAN BOVEN, AND GILOVICH

standardized fashion on the computer—to consider their judgments care-
fully and respond at their leisure. Participants in the hurried condition were
instructed to make their judgments as soon as they could following the
beep and were told that they had 3 seconds in which to answer. If they
failed to respond within the time limit, a second beep sounded and the next
scenario appeared.
3
Results and Discussion
Fewer than 1% of the data points were missing because of
keyboard errors, and none of the participants under time pressure
failed to respond within the 3-second deadline.
Our key prediction was that participants under time pressure
would be more egocentric than those who responded at their
leisure. To examine this prediction, we created two sarcasm scores
for each participant by averaging their predictions (“yes” coded as
1, “no” coded as 0) for the six positive scenarios and the six
negative scenarios. These scores indicated the percentage of com-
munications in each condition that participants expected would be
perceived as sarcastic. We then submitted these scores to a 2 (time
pressure: hurried or leisurely) ⫻ 2 (scenario: positive or negative)
mixed model ANOVA. This analysis yielded a significant main
effect for scenario, F(1, 94) ⫽ 39.6, p ⬍ .01, indicating that those
who received negative event information believed that others were
more likely to detect sarcasm than those who received positive
event information (see Figure 1).
More important, there was also a significant interaction, F(1,
94) ⫽ 23.75, p ⬍ .01, indicating that hurried participants were
more egocentric than unhurried participants. Participants who read
about negative events thought listeners were more likely to inter-
pret the messages as sarcastic when they were under time pressure
(M ⫽ 66%) than when they were not (M ⫽ 50%), t(94) ⫽ 2.97,
p ⬍ .01. In contrast, participants who read about positive events
thought listeners were less likely to interpret the messages as
sarcastic when they were under time pressure (M ⫽ 28%) than
when they were not (M ⫽ 45%), t(94) ⫽ 3.80, p ⬍ .01.
Additional evidence of adjustment comes from hurried partici-
pants’ response times. Recall that we predicted longer response
latencies for perspective-inconsistent responses than for
perspective-consistent responses because an inconsistent response
requires more adjustment (and hence more time) than a consistent
response. As expected, perspective-inconsistent responses (re-
sponding “no” after receiving negative event information or “yes”
after receiving positive event information) were made more slowly
(M ⫽ 1,013 ms) than perspective-consistent responses (M ⫽ 820
ms), t(11) ⫽ 3.61, p ⬍ .01.
4
One additional finding—or, rather, lack of a finding—warrants
discussion. In contrast to Study 1 and past research using similar
stimulus materials (Baldwin & Keysar, 1998; Keysar, 1994, 1998;
Weingartner & Klin, 2001), participants in the leisurely condition
(the white bars in Figure 1) did not exhibit a significant egocentric
bias, t(47) ⫽ 1.03, p ⫽ .31. Although the findings of this exper-
iment are consistent with serial adjustment from an egocentric
anchor, this particular finding might seem inconsistent with a
general tendency for insufficient adjustment. We suspect this oc-
curred because participants in the leisurely condition were explic-
itly instructed to take their time rather than having received no
instructions regarding pace, as in previous experiments. Because
adjustment takes time, instructions to respond leisurely should
increase adjustment because it diminishes the need to accept
quickly a response as a plausible estimate for another’s perception.
Indeed, the amount of time involved in perspective taking is one of
several moderators of egocentric biases that we address in the
General Discussion.
The results of this study further support our contention that adult
perspective taking follows a process of anchoring and serial ad-
justment. Because adjustment from one’s own perspective takes
time, hurried participants adjusted less and were consequently
more egocentric than those who responded at their leisure. In
addition, hurried participants responded more quickly when indi-
cating that others would share their interpretation than when indi-
3
Participants received several memory quizzes during the experiment to
ensure that they attended carefully to the scenarios. Participants knew that
they would occasionally be tested about the details of the scenarios but did
not know when. All participants performed well on the quizzes (each above
80% accuracy).
4
Not all participants provided a response in each of these four cells. As
a consequence, this analysis is done at the level of the item across the 12
scenarios.
Figure 1. Percentage of hurried and leisurely participants who expect a sarcastic interpretation after receiving
positive or negative event information (Study 2).
331
PERSPECTIVE TAKING AS ADJUSTMENT

cating that others would have a different interpretation, presum-
ably because the former involves less adjustment than the latter.
If, as these results indicate, hindering people’s ability to expend
attentional resources systematically shortens adjustment, then mo-
tivating people to expend additional resources should increase
adjustment and produce less egocentric judgments. We examined
this possibility in Study 3 by offering people a monetary incentive
for accurate assessments. Although it is well known that incentives
for accuracy do not influence many of the most well-known
anchoring effects (e.g., Tversky & Kahneman, 1974; Wilson,
Houston, Etling, & Brekke, 1996; Wright & Anderson, 1989), this
is because most anchoring effects involving anchors provided by
an experimenter or some other external source do not bias judg-
ment through a process of insufficient adjustment but rather
through a process of biased hypothesis testing and the selective
recruitment of anchor-consistent information (for reviews see
Chapman & Johnson, 2002; Epley, 2004). We have shown else-
where that true adjustment from “self-generated” anchors known
to be close to the right answer but wrong is, in fact, increased when
participants are provided incentives for accuracy (Epley & Gilov-
ich, 2004b). We thus expected participants in Study 3 to be less
egocentric, and therefore more accurate, in adopting another’s
perspective when they were given financial incentives to do so.
Study 3: Cola Wars
We conducted a taste test to examine our hypotheses. Partici-
pants tasted two unmarked drinks, one containing Coca-Cola
(Coke; Coca-Cola Company, Atlanta, GA) and the other Pepsi-
Cola (Pepsi; PepsiCo, Purchase, NY). Before tasting the drinks,
participants were told the identity of the beverages. As with the
clarifying information in Study 2, we assumed that informing
participants of the drinks’ identities ahead of time would lead them
to taste a difference between them—a difference that would not be
apparent to those uninformed of the identities. Indeed, in a pilot
test of this assumption, 85% of students in the condition informed
of the drinks’ identities (N ⫽ 27) believed they could have cor-
rectly distinguished between the drinks even if they had not been
informed—a figure that differed substantially from the 50% accu-
racy rate actually obtained from a sample of uninformed partici-
pants (N ⫽ 32),
␹
2
(1, N ⫽ 59) ⫽ 8.09, p ⫽ .004.
Participants in this experiment were asked to estimate the per-
centage of their peers who, uninformed of the drinks’ identities,
would be able to distinguish between Coke and Pepsi. On the basis
of the results of Study 1, we assumed that participants would
recognize that uninformed peers would have a harder time identi-
fying the difference than they would themselves and that they
would thus need to adjust from an initially high percentage (“it’s
obvious—everyone will taste the difference”) to a more moderate
estimate.
We manipulated participants’ motivation to engage in adjust-
ment by offering half of them a financial incentive for accuracy.
Because serial adjustment requires effortful thought, we predicted
that those offered a financial incentive would exert more effort,
and therefore be less egocentric, than those who were not offered
an incentive. In particular, we predicted that participants offered an
incentive would expect a lower percentage of uninformed peers to
identify the drinks correctly.
Method
Sixty-one University of British Columbia (Vancouver, British Colum-
bia, Canada) students were approached as they exited a university dining
hall and asked if they would be willing to take part in a brief taste test in
exchange for a chocolate bar. Participants were told that they would taste
two colas—one Coke and one Pepsi— from two different cups labeled “A”
and “B.” After learning the identities of the two colas and tasting each of
them, participants were told that some participants in this experiment
would remain uninformed about the identities of the drinks before tasting
them.
5
Participants then estimated the percentage of these uninformed
participants who would correctly identify the two drinks. Before making
these estimates, participants randomly assigned to the incentive condition
(n ⫽ 31) were told that the participant who made the most accurate
estimate would receive a $50 gift certificate to the University of British
Columbia bookstore—with a tie being resolved by a random drawing.
Participants in the no incentive condition were not offered this incentive.
Finally, participants indicated whether they themselves could taste the
difference between the two colas.
Results and Discussion
Most participants in the control condition (79%, 95% confi-
dence interval ⫽ 61%–92%) and in the incentive condition (84%,
95% confidence interval ⫽ 66%–95%) believed they could taste
the difference between the two drinks—substantially more than the
50% of uninformed participants in the pretest who could distin-
guish the two drinks.
Our key prediction concerned participants’ beliefs about the
ability of uninformed participants to correctly identify the two
drinks. As predicted, participants offered an incentive for accuracy
estimated that uninformed tasters would be less likely to identify
the two drinks correctly (M ⫽ 54.4%) than did participants not
offered an incentive (M ⫽ 64.8%), t(59) ⫽ 2.60, p ⬍ .01. The
estimates made by those offered a financial incentive were not
only more conservative, but their proximity to the 50% accuracy
rate obtained in the pretest indicates they were more accurate as
well. Participants not offered an incentive overestimated the like-
lihood that uninformed tasters—represented by the 50% accuracy
in the pretest—would be able to taste the difference, t(29) ⫽ 4.55,
p ⬍ .01, whereas participants who were offered an incentive did
not, t ⬍ 1. These results offer further evidence that people adopt
others’ perspectives by initially anchoring on their own perception
and then effortfully adjusting for differences between themselves
and others.
Study 4: Backmasked Messages
Our account specifies that people adjust from their own per-
spective to adopt another’s by evaluating a series of hypotheses
ever further from their own perspective until they reach a value
that plausibly captures the other person’s point of view. If so, then
influencing individuals’ willingness to accept these tentative hy-
potheses ought to influence the magnitude of egocentric biases that
emerge when people adopt another person’s perspective.
5
The order in which participants were informed about the identities of
the colas as well as the order in which they drank them were counterbal-
anced. Neither order manipulation influenced any of the results in this
experiment, however, and so we collapsed across these conditions.
332
EPLEY, KEYSAR, VAN BOVEN, AND GILOVICH

To influence this willingness, we asked participants in Study 4
to nod or shake their heads while trying to adopt another person’s
perspective. Nodding one’s head while evaluating a proposition
makes one more likely to accept the proposition, whereas shaking
one’s head makes one less likely to accept it (Forster & Strack,
1996, 1997; Wells & Petty, 1980). These head movements have
this effect because they serve to validate one’s thoughts in a
manner consistent with the everyday implications of these head
movements (Brinol & Petty, 2003). Individuals assessing whether
a persuasive message is true, for instance, are more likely to
confirm this assessment when nodding their heads up and down
(subtly suggesting that, yes, it is true) than when shaking their
heads from side to side. When adopting another’s perspective, we
predicted that participants’ head movements would serve to vali-
date plausible estimates that come to mind early in the process of
adjustment.
Indeed, head nodding and shaking was used in previous research
to confirm the operation of serial adjustment in numerical anchor-
ing tasks (Epley & Gilovich, 2001). For example, most people do
not know when George Washington was first elected President of
the United States, but they readily estimate the year by anchoring
on 1776 and adjusting to a later date. In one experiment, partici-
pants thought Washington was elected President earlier when they
made their estimate while nodding their heads up and down than
while shaking them from side to side. Similarly, we predicted in
this experiment that participants engaged in head nodding would
accept hypotheses earlier in the adjustment process, and thus be
more egocentric, than participants engaged in head shaking.
Participants in this experiment listened to a short segment, in
reverse, from the Queen (1980/1991) song, “Another One Bites the
Dust” (interested readers are encouraged—before reading fur-
ther—to listen to this clip at http://gsbwww.uchicago.edu/fac/
nicholas.epley/research/clip.html). The chorus of this song has
long been used to support the claim that rock and roll bands
“backmask” illicit or immoral statements into their music in an
attempt to influence listeners (Vokey & Read, 1985). Proponents
of this view report being able to hear the phrase “It’s fun to smoke
marijuana” when the chorus from “Another One Bites the Dust” is
played backward. This prodrug message, however, is not produced
by the subversive actions of the artists but by the creative inter-
pretations of listeners, specifically by the human tendency to create
order out of randomness. Knowing what to listen for further
facilitates this tendency.
Accordingly, before listening to the critical selection, approxi-
mately half of our participants were told to listen for the phrase
“It’s fun to smoke marijuana.” The other half were told nothing.
The selection played for participants is sufficiently unclear to
ensure that very few (if any) of those who were not told about the
critical phrase would hear it on their own, but virtually everyone
told about the phrase in advance would have no difficulty detecting
it. After listening to the selection, participants estimated the per-
centage of their peers who, told nothing about the selection, would
be able to hear the critical phrase. Approximately half estimated
while nodding their heads and half while shaking their heads.
We made three predictions. First, participants informed about
the critical phrase would hear the backmasked message more
clearly than those who were uninformed. Second, participants
would be egocentric: Informed participants would expect a higher
percentage of their (uninformed) peers to hear the phrase than
would uninformed participants. Third, and most important, partic-
ipants’ head movements would moderate their egocentrism: Par-
ticipants nodding their heads would be more egocentric than
participants shaking their heads. We expected informed partici-
pants to start with a high egocentric anchor (e.g., “everyone will
hear it”) and adjust downward to account for their private infor-
mation (e.g., “but they weren’t told about the hidden message
beforehand”). Because head nodding leads people to accept values
earlier in the process of adjustment than head shaking (Epley &
Gilovich, 2001), we expected informed participants who were
nodding their heads to provide higher estimates than those shaking
their heads. In contrast, we expected uninformed participants to
start with a low anchor (“nobody will hear it”) and adjust upwards
because the experimenter explicitly told them that at least “some”
people report hearing the phrase. Stopping the adjustment process
early in this context would lead to lower estimates, and thus we
expected uninformed participants who were nodding their heads to
provide lower estimates than those shaking their heads.
Method
Fifty-three Cornell University undergraduates participated in exchange
for extra credit in their psychology or human development courses. On
arrival at the laboratory, the experimenter informed all participants that this
was a study of product evaluations being conducted in conjunction with the
Marketing Department at Cornell University (Ithaca, NY). The experi-
menter explained that participants would evaluate a pair of headphones and
would be asked, at some point in the experiment, to simulate common head
movements one might make while wearing these headphones—either
nodding one’s head up and down or shaking it from side to side.
Participants listened to the Queen song as well as the critical dependent
measures over the headphones. To justify this procedure and eliminate any
apparent connections between the head movements and critical questions,
the experimenter explained that he wished to examine “implicit evaluations
[that people] form without conscious intention or effort.” He thus needed
to busy participants with another task while they were evaluating the
headphones, in this case listening to the music and questions on the tape.
Once the procedure was clear, participants were first played the critical
segment from Queen’s “Another One Bites the Dust” as it was originally
recorded (i.e., forward). Participants randomly assigned to the informed
condition were then told that they would be asked to listen to this same
selection backward and that some people report being able to hear the
words “It’s fun to smoke marijuana.” These informed participants then
listened to the selection three times, each time being reminded to listen
carefully for the critical phrase. Participants assigned to the uninformed
condition, in contrast, received no information about the reported content
of the backward selection and listened to it only once.
After listening to the backward selection, all participants were asked to
move their heads either up and down or from side to side (following the
experimenter’s demonstration) while they listened to a series of questions
over their headphones. The first question informed all participants (some
for the first time) that some people report hearing the words “It’s fun to
smoke marijuana” when the chorus from “Another One Bites the Dust” is
played backward. Participants then indicated (verbally) whether they heard
those words (yes or no) and how clearly they heard them on a scale ranging
from 0 (completely unclear) to 100 (completely clear). All participants
were then asked to imagine that “we played the backwards selection to 100
Cornell University psychology students without giving them any informa-
tion about its source or what might be contained within the selection” and
to estimate the percentage who would indicate being able to hear the
critical phrase.
333
PERSPECTIVE TAKING AS ADJUSTMENT

Results and Discussion
We expected the backward selection to sound very different to
informed and uninformed participants. Indeed, nearly everyone in
the informed condition (88%—all but 3) reported being able to
hear “It’s fun to smoke marijuana,” whereas none (0%) in the
uninformed condition did so,
␹
2
(1, N ⫽ 53) ⫽ 42.20, p ⬍ .01. Not
surprisingly, participants in the informed condition also rated the
phrase as more clear (M ⫽ 58.46) than did those in the uninformed
condition (M ⫽ 9.26), t(51) ⫽ 11.25, p ⬍ .01. Participants’ head
movements did not influence responses to these questions either
across or within the informed and uninformed conditions (all
ts ⬍ 1).
We also expected informed participants to estimate that a higher
percentage of their peers would hear the phrase than participants
who were uninformed. However, we expected this effect to be
qualified by an interaction with participants’ head movements.
Participants who were nodding their heads should be more ego-
centric and give more extreme responses than participants who
were shaking their heads. As predicted, a 2 (knowledge: informed
versus uninformed) ⫻ 2 (head movement: nodding versus shaking)
ANOVA yielded a significant main effect for knowledge, F(1,
49) ⫽ 29.34, p ⬍ .01, qualified by the predicted significant
interaction, F(1, 49) ⫽ 4.10, p ⬍ .05. As can be seen in Figure 2,
participants in the informed condition expected a higher percent-
age of peers to hear the phrase when they were nodding their heads
(M ⫽ 55%) than when they were shaking their heads (M ⫽ 44%).
As predicted, this difference was reversed among participants in
the uninformed condition who expected a smaller percentage of
peers to hear the phrase when they were nodding their heads (M ⫽
16%) than when they were shaking their heads (M ⫽ 28%).
Neither of these simple effects, however, were statistically signif-
icant, ts(49) ⫽ 1.38 and 1.62, ps ⫽ .18 and .12, respectively.
These results provide further evidence that people adopted oth-
ers’ perspectives by serially adjusting from their own. Participants
informed about the critical phrase expected a higher percentage of
their peers to hear the phrase than participants who were told
nothing about the phrase. The effect of the head movement ma-
nipulation suggests that those in the informed condition were
adjusting downward from a relatively high anchor, whereas those
in the uninformed condition were adjusting upward from a low
anchor. Head nodding produced higher estimates in the informed
condition but lower estimates in the uninformed condition com-
pared with head shaking, consistent with a shortened process of
serial adjustment.
Study 5: Satisficing
The preceding four experiments support our claim that people
often adopt others’ perspectives by serially adjusting from their
own. What they do not establish is why such adjustments are often
insufficient. One interesting possibility was suggested by Quat-
trone (1982), who argued that serial adjustment is typically char-
acterized by some degree of “satisficing.” That is, given the
uncertainty surrounding the true value being estimated, people are
likely to have a range of values they would consider to be plausible
estimates. In the absence of sufficient motivation for accuracy,
people are likely to terminate adjustment once a plausible estimate
is reached—arriving at a satisfactory estimate rather than the most
accurate estimate. Because adjustments terminate at the boundary
of the range of plausible values closest to the original (egocentric)
anchor, they tend to be insufficient (see also, Epley & Gilovich,
2004a; Mussweiler & Strack, 2001). This satisficing account of
serial adjustment is consistent with the effects of time pressure and
incentives observed in Studies 2 and 3, but these studies do not test
this account directly. We sought to do so in Study 5.
If this satisficing account provides an accurate description of the
stop rule for serial adjustment when people try to adopt another’s
perspective, then participants’ estimates of others’ perceptions
should be skewed toward the egocentric side of their range of
plausible estimates. To test this possibility, we asked participants
in Study 5 to estimate whether others would interpret an ambigu-
ous message as sincere or sarcastic, as in Studies 1 and 2. As
before, all participants received privileged positive or negative
background information that clarified the intended meaning of the
message. This time, however, only half of the participants (those in
the answers condition) estimated the number of uninformed peers
who would interpret the message as sincere versus sarcastic. The
other half of the participants (those in the ranges condition) stated
the range of values they found to be subjectively plausible by
Figure 2. Estimated percentage of peers who would hear the “backmasked” phrase among informed and
uninformed participants who were either nodding or shaking their heads (Study 4).
334
EPLEY, KEYSAR, VAN BOVEN, AND GILOVICH

estimating the upper and lower bounds on the number of unin-
formed peers who would interpret the message as sincere versus
sarcastic.
6
We predicted that the point estimates provided by
participants in the answers condition would be skewed toward an
egocentric response within the range of plausible values provided
by participants in the range condition. Not only would such a result
illuminate the stop-rule for serial adjustment, but it would also
provide additional evidence in support of our serial adjustment
account of adult perspective taking.
We also sought to investigate our satisficing account of adjust-
ment more fully by including a time pressure manipulation. Study
2 demonstrated that time pressure shortened the process of adjust-
ment, and we believe it did so (like incentives in Study 3 and head
movements in Study 4) by influencing participants’ threshold for
accepting a value as a plausible estimate during the course of serial
adjustment. We tested this possibility in Study 5 by placing half of
the participants under time pressure to provide their responses. As
in Study 2, we expected this manipulation to produce more ex-
treme responses among those in the answers condition. We made
no predictions about the influence of time pressure on participants’
ability to generate plausible estimates, as any result would be
informative. On the one hand, if time pressure influences point
estimates but not range estimates, this would suggest that range
estimates are made by some process other than serial adjustment.
This pattern would suggest that time pressure influences partici-
pants’ threshold for accepting any particular value as a plausible
estimate only while engaging in adjustment. On the other hand, if
time pressure influences both point estimates and range estimates,
then this suggests that it influences what participants find to be
plausible estimates rather than simply the threshold for accepting
a value as plausible during the course of adjustment.
Method
Seventy-eight Harvard University (Cambridge, MA) undergraduates
participated in exchange for $6 or research credit in one of their psychol-
ogy courses. The procedure was the same as that in Study 1, with the
following exceptions. First, all materials were presented on a computer
rather than in booklets. An audio file of the experimenter reading the
materials was presented along with the text on the computer screen to
ensure that all participants attended to the instructions and scenarios.
Second, the ambiguous messages were presented as e-mail messages rather
than telephone messages, primarily to simplify the experimental procedure
but also to increase the variety of experimental manipulations. Third, 10
scenarios involving ambiguous messages were used instead of 12 to allow
2 of the earlier scenarios from Study 2 to be used as practice items. This
was deemed necessary in order to familiarize participants in the ranges
condition with their task. Of these 10 items, half included positive event
information intended to make the message appear sincere and half included
negative event information to make it appear sarcastic. Scenarios were
presented in one random order or its mirror image, resulting in two
counterbalanced versions. Fourth, half of the participants were placed
under time pressure. These participants were asked to respond as quickly
as possible and were told that they would be timed out in 3 seconds if they
had not yet responded.
Fifth, and most important, participants were randomly assigned to one of
two response conditions. After reading the scenario and e-mail message,
those in the answers condition were asked to imagine that “100 people
were asked to read Tom’s e-mail message” and to estimate the “percentage
of these people who would interpret the message as sincere versus sarcas-
tic.” Responses were made on an 11-point scale ranging from 100%
sarcastic to 100% sincere, with the midpoint labeled 50/50. Participants in
the ranges condition were also asked to imagined that 100 people were
asked to read Tom’s e-mail message but to estimate upper and lower
bounds on the percentage of people who “would plausibly interpret the
message as sarcastic versus sincere.” During the two practice trials, the
experimenter told these participants that their ranges could be as wide or as
narrow as seemed appropriate for each particular item.
Results and Discussion
This study was designed to investigate two issues: (a) whether
participants under normal circumstances stop adjusting once they
reach a plausible estimate and (b) whether time pressure influences
the threshold for accepting a value as a plausible estimate or shifts
the range of estimates that appear plausible while under time
pressure. Because these are distinct research questions, we report
analyses of the control participants first and present the results of
the time pressure manipulation second. To ease interpretation of all
analyses, the dependent measures are reported in terms of the
percentage of participants who would interpret a message as
sincere.
Answers versus ranges. As in the previous studies, partici-
pants’ assessments in the control condition were egocentrically
biased. Participants estimated that a smaller percentage of their
peers would interpret a message as sincere when it was preceded
by negative event information (M ⫽ 60.78%) than when it was
preceded by positive event information (M ⫽ 78.11%), paired
t(17) ⫽ 3.70, p ⬍ .01.
Our main prediction, however, was that control participants’
point estimates in the answers condition would be skewed toward
the egocentric end of the ranges provided by participants in the
ranges condition, consistent with a process of adjustment from an
egocentric anchor that terminates once a plausible estimate is
reached. As can be seen in Table 1, this prediction was strongly
confirmed. When considering negative event information, partic-
ipants’ answers were skewed toward a sarcastic response within
the ranges provided by participants in the range condition. That is,
answers did not differ from the lower boundary closest to the
sarcastic end of the scale, t(35) ⫽ 0.29, p ⫽ .77, but they were
significantly lower than the upper boundary, t(35) ⫽⫺2.53, p ⬍
.05. The opposite occurred when participants were considering
positive event information: Answers did not differ from the upper
boundary closest to the sincere end of the scale, t(35) ⫽ –0.05, p ⫽
.96, but were significantly higher than the lower boundary, t(35) ⫽
3.18, p ⬍ .01.
These results suggest that people stop adjusting once they reach
an estimate that appears plausible. Because this estimate termi-
nates at the egocentric end of the range of plausible values,
adjustments tend to be insufficient. Note that the overlap in the
ranges provided by participants in our ranges condition is substan-
tial, being shifted over by a mere 5%. This indicates that simply
adjusting further into the range of plausible values would yield
more accurate estimates and mitigate the pronounced egocentric
biases in perspective taking.
6
This between-participants manipulation of answers versus ranges
(rather than within-participants) was used to avoid the possibility that
generating an answer would influence the generation of a plausible range,
or vice versa.
335
PERSPECTIVE TAKING AS ADJUSTMENT

Time pressure. Our secondary goal in this experiment was to
study the impact of time pressure on participants’ answers and
plausible ranges.
7
Because participants in the answers condition
provided only one response, whereas those in the ranges condition
provided two, responses from these two conditions are analyzed
separately and compared with t tests wherever appropriate. As can
be seen in Table 1, time pressure produced stronger egocentric
biases among participants in the answers condition, replicating the
results of Study 2. A 2 (time pressure: yes vs. no) ⫻ 2 (background
information: positive versus negative) ANOVA with repeated
measures on the last factor yielded a main effect for background
information, F(1, 38) ⫽ 58.84, p ⬍ .01, qualified by the predicted
interaction, F(1, 38) ⫽ 8.60, p ⬍ .01. When considering negative
background information, participants under time pressure thought
a smaller percentage of participants would interpret the message as
sincere than participants not under time pressure (Ms ⫽ 44.30 and
60.77, respectively), t(38) ⫽⫺2.46, p ⬍ .05. When considering
positive background information, in contrast, participants under
time pressure thought a larger number of participants would inter-
pret a message as sincere than participants not under time pressure
(Ms ⫽ 83.10 and 78.11, respectively), although this latter differ-
ence was not significant, t(38) ⫽ 1.44, p ⫽ .16.
Interestingly, the impact of time pressure on participants’ an-
swers was not matched by an influence on the ranges provided by
participants in the range condition (see Table 1). A 2 (time pres-
sure vs. no time pressure) ⫻ 2 (background information: positive
versus negative) ⫻ 2 (boundary estimate: low versus high)
ANOVA with repeated measures on the last two factors yielded
only two main effects. The first significant main effect was for
background information, F(1, 36) ⫽ 9.67, p ⬍ .01, indicating that
participants’ ranges were still somewhat egocentrically biased. The
second significant main effect was for boundary estimate, F(1,
36) ⫽ 237.29, p ⬍ .01, indicating that the lower boundary of these
ranges was—no surprise—lower than the higher boundary. Neither
the main effect for time pressure nor any interactions approached
significance. The differential impact of time pressure in the an-
swers versus ranges conditions suggests the operation of very
different mental processes in these two conditions. Those search-
ing for a single answer appeared to be engaging in serial adjust-
ment whereas those generating a range of plausible values were
not.
The differential impact of time pressure on responses in the
answers and ranges conditions also resulted in answer participants
in the hurried condition providing answers that fell outside their
counterparts’ range of plausible values. This may have occurred
because adjustment from an egocentric anchor requires partici-
pants to consider the plausibility of highly egocentric responses,
whereas generating a range of plausible values does not. Those
generating an answer under time pressure may therefore have
seized on an egocentric estimate considered early in the course of
adjustment that those generating a range of plausible values under
time pressure did not even consider because they had approached
their task by invoking rather different mental processes. Regard-
less of the exact cause, these results make it clear that time
pressure influences the threshold for accepting a value as a plau-
sible estimate while engaging in serial adjustment, but it does not
cause a shift in the range of values considered plausible.
In addition to providing further evidence for our anchoring and
adjustment account of adult perspective taking, these results are
also at variance with an alternative interpretation of our data based
on an inappropriate weighting or integration of information (e.g.,
Trope & Gaunt, 2000). That is, in each of the preceding studies,
people had two pieces of information to think about when adopting
others’ perspective—their own perception of a stimulus and infor-
mation that others would likely perceive the stimulus differently
than they did. Egocentric biases would emerge if participants gave
the second piece of information less weight than the first. Notice,
however, that such a weighted integration model does not predict
the results of this last study. Instead, inappropriate weighting
should produce estimates roughly centered within participants’
range of plausible values because whatever weighting formula is
followed should influence estimates of the upper and lower bounds
in the same way it influences the point estimates themselves. The
skewed estimates observed in this study, in contrast, are both
predicted and readily explained by a process of serial adjustment
that stops once a plausible estimate is reached. People tend not to
adjust far enough, it appears, because they stop once they reach an
estimate that is simply good enough.
General Discussion
Perspective taking is central to social interaction. Teachers,
lovers, parents, and nearly everyone involved in social relation-
ships base their actions, at least in part, on their understanding of
others’ perspective. But an accurate understanding of others’ per-
spective can be hard to achieve. Everyday experience and psycho-
logical research demonstrates that people tend to overestimate the
extent to which others will perceive the world as they do—that
teachers often overestimate the clarity of their lectures, lovers the
clarity of their emotions, and parents the clarity of their instruc-
tions. In short, perspective taking is often egocentrically biased.
We sought to understand the processes responsible for these
egocentric biases in perspective taking. The results of five studies
support our contention that people adopt others’ perspectives by
initially anchoring on their own perspective and subsequently
adjusting to account for differences between themselves and oth-
ers. Consistent with this anchoring and adjustment process, par-
ticipants’ judgments about others’ perceptions of an ambiguous
7
All participants in the time pressure condition responded within the
3-second response window.
Table 1
Average Point Estimates and Ranges Provided by Participants
in the Control and Time Pressure Conditions (Study 5)
Condition
Answer
estimate
Range
Low High
Control
Negative event information 60.77
a
59.26
a
73.16
b
Positive event information 78.11
a
64.10
b
78.32
a
Time pressure
Negative event information 44.30
a
58.63
b
73.05
c
Positive event information 83.10
a
64.95
b
78.21
a
Note. Means that do not share the same subscript within rows differ at
p ⬍ .05.
336
EPLEY, KEYSAR, VAN BOVEN, AND GILOVICH

stimulus were correlated with, but were more moderate than, their
own perceptions (Study 1). Because adjustment requires both time
and mental effort, participants became more egocentric when
under time pressure (Study 2) and less egocentric when provided
incentives for accuracy (Study 3). Participants were also more
egocentric when simultaneously performing actions that induce
acceptance (nodding their heads) than when performing actions
that induce rejection (shaking their heads; Study 4), indicating that
head nodding led participants to accept hypotheses encountered
early in the adjustment process, whereas head shaking led them to
reject those hypotheses and adjust further (see also Epley &
Gilovich, 2001; Forster & Strack, 1997; Wells & Petty, 1980).
Finally, participants’ estimates of others’ perceptions fell on the
outskirts of the range of plausible perceptions—too close to an
egocentric anchor—as implied by our thesis that participants adopt
a satisficing strategy when serially adjusting from their own per-
spective, stopping when they reach a plausible value (Study 5).
More generally, these results suggest that perspective taking
operates through a multistage process. Initially, people anchor on
their own perspective, presumably because it is often highly ac-
cessible. In the absence of clear evidence that another person will
see the world differently, perspective taking ends there with people
assuming that others will perceive the world as they do (Ross &
Ward, 1996). However, when others are known to be in different
situations, from different backgrounds, or in possession of differ-
ent knowledge, such naive realism is simply untenable. In these
cases, as in the experiments reported in this article, it is clear from
the outset that one’s own perspective is not shared and adjustment
is required. Although this subsequent adjustment process may take
several forms, the data we report here and elsewhere suggest that
it operates through a series of hypothesis tests ever further from the
original anchor value. The original egocentric anchor is rejected,
some “jump” is made to a nearby hypothesis, and its plausibility
assessed. If the new hypothesis seems too egocentric, another jump
is made, another value assessed, and so on until an acceptable
hypothesis is reached. The results of each experiment reported in
this article are consistent with this model, and Studies 4 and 5 in
particular highlight the serial nature of adjustment in perspective
taking.
Although such adjustment is helpful—inducing less egocen-
trism than one would see without it—it is not perfect. We observed
egocentrically biased judgments—suggesting insufficient adjust-
ment—in all but two conditions of our five experiments: when
participants were encouraged to take their time and respond at their
leisure when estimating an uninformed person’s interpretation of
an ambiguous communication (Study 2) and when participants
were offered a financial incentive for accurately estimating the
percentage of uninformed people who could taste a difference
between Coke and Pepsi (Study 3).
Insufficient adjustments, and resulting egocentric biases, are
likely to occur for at least two reasons (Epley & Gilovich, 2004a,
2004b). First, in the absence of a strong motivation for accuracy,
many judgments likely terminate as soon as an acceptable response
is generated. This satisficing leads people to accept one of the early
(and, in this case, egocentric) plausible values they encounter in
the course of adjustment. Because adjustments tend to stop as
participants reach the edge of their range of plausible values, they
tend to be insufficient (Mussweiler & Strack, 2001; Quattrone,
1982). Study 5 tested this hypothesis directly and demonstrated
that participants’ estimates of others’ perceptions are not centered
within a range of plausible values but rather skewed toward an
egocentric response. Second, serial adjustment requires cognitive
resources that are often in short supply, generally leading people to
accept hypotheses they might reject under leisurely or unencum-
bered circumstances. Together, satisficing and limited cognitive
resources make insufficient adjustment the rule rather than the
exception, causing perspective taking to be egocentrically biased.
One may wonder, however, whether the tendency to adopt
another’s perspective by adjusting from one’s own is limited to the
particular kinds of judgments or domains sampled in our experi-
ments. For example, most judgments in these experiments were
made along some continuous scale (e.g., the likelihood of a par-
ticular interpretation or the percentage of peers who would identify
an ambiguous phrase). Perhaps serial adjustment of the kind we
have identified only occurs along such ordinal dimensions and
does not characterize dichotomous or multicategorical judgments
where serial adjustment may be less natural and harder to employ.
“Do they understand?”“Do they know how I feel?”“Are they
noticing me?” To be sure, it makes little sense to talk about serial
adjustment along continuous dimensions that do not exist. How-
ever, notice that many of these multicategorical or dichotomous
responses are built on underlying continuous assessments (Tversky
& Koehler, 1994). How likely is it that they understand? How well
do they understand how I feel? How many people are noticing me?
Indeed, this is precisely what we contend happened with the
dichotomous and multicategorical judgments rendered in Studies 1
and 2. Although the final judgment in these studies may have
involved categorical distinctions, we suspect they rest on the same
sort of continuous assessments rendered in Studies 3, 4, and 5. The
anchoring and adjustment model we are proposing, then, may
apply much more broadly than it might initially appear.
This does not mean, of course, that there are no boundary
conditions to this model of perspective taking. For example, it is
reasonable to assume that people engage in such anchoring and
adjustment only when it is clear that one’s self is a reasonable
starting place in estimating others’ perspectives. In situations
where one’s own perspective is clearly irrelevant, quite different
processes may be involved. For example, when loggers assess the
perspective of environmentalists, and vice versa, they may not
anchor on their own perspective because it is likely to be seen as
irrelevant. Instead, each may generate an assessment of the other
person’s views based on some stored representation (e.g., a pro-
totype) of their ideologies and values (for a similar suggestion, see
Ames, in press; Karniol, 2003). Indeed, partisans in disputes tend
not to be egocentric in their assessments of the “other side” but to
overestimate the dissimilarity between themselves and their oppo-
nents—more akin to a contrast effect than the assimilation that
results from insufficient adjustment (Robinson & Keltner, 1996;
Robinson et al., 1995).
On a theoretical level, our results have obvious implications for
research on “theory of mind”—humans’ ability to attribute mental
states such as belief and desire to others. Since Piaget, the exact
nature of this “folk psychology” has been hotly debated. One view
is that children make sense of other minds by learning and apply-
ing a general theory of the way minds work—the so-called
“theory-theory” (Gopnik & Wellman, 1992, 1994). A contrasting
337
PERSPECTIVE TAKING AS ADJUSTMENT

view is that children use themselves as a source model, predicting
others’ thoughts and feelings by imagining themselves in the other
person’s situation—the “simulation theory” (Goldman, 1992; Gor-
don, 1992). Our results suggest that by adulthood, individuals’
attempts at perspective taking are often something of an integra-
tion of theory and simulation. Adults’ use of their own perspective
as an anchor is similar to using one’s self as a source model for
predicting others. Additionally, adults’ adjustment from that an-
chor is likely guided by their theories about how different perspec-
tives and psychological states influence judgment and perception.
Thus, neither theory may be strictly right or wrong, and both may
play a role in different components of a complicated psychological
process. Maturity may not cause people to abandon their youthful
egocentrism but rather increase the tendency to adjust away from
it (Epley, Morewedge, & Keysar, in press).
On a more practical level, the anchoring and adjustment model
we have proposed has important implications for understanding
(and resolving) social conflict. When people adjust their own
perceptions to understand another’s, they do so to accommodate
discrepancies that would otherwise distort their perception of some
target. Sellers, when estimating the interest of potential buyers,
adjust for the fact that they own an object that buyers do not (Van
Boven et al., 2000). Relationship partners adjust for the fact that
their emotions are internal and thus may not be as transparent to
others as they are to themselves (Gilovich et al., 1998). Teachers
also adjust for the fact that they know the material better than their
students when determining the level at which to present an idea.
Because people adjust to accommodate apparent external or situ-
ational influences on perception and judgment, any residual dis-
crepancy between predicted and actual beliefs is likely to be
attributed to a dispositional feature of the target—a buyer’s greed,
a student’s laziness or deficient aptitude, or a partner’s insensitiv-
ity (Pronin, Gilovich, & Ross, in press). Such inferences are likely
to be problematic not only because they are likely to be inaccurate
but also because they will be seen as unwarranted and unfair—and
thus a source of further conflict—by the target of judgment.
Insufficient adjustment when adopting another’s perspective may
thus contribute to the all-too-frequently observed spiral of conflict
and discord. The results of several of our experiments, however,
suggest that such conflict may be alleviated by encouraging people
to take their time and to think carefully about another’s perspec-
tive. Indeed, actively instructing people to do so is associated with
a variety of desirable effects, from inducing more favorable judg-
ments of others (Epley, Savitsky, & Gilovich, 2002), to increasing
altruism and helping (Batson et al., 1991), to reducing stereotyping
and prejudice (Galinsky & Moskowitz, 2000).
Of course, that people fail to understand the workings of others’
minds perfectly is hardly surprising. The human brain is the most
complicated piece of hardware in existence, and people are left to
intuit its outputs armed with nothing more than personal experi-
ence and a few principles of inference. This intuitive ability is
remarkable and appears to be matched by no other species on the
planet. The point of the experiments we have reported is not to
ridicule a shortcoming in this perspective-taking process but,
rather, to point out that this shortcoming is of a systematic variety
that provides important clues to how perspective taking works. We
hope this understanding, in turn, provides some insight into how to
make this impressive human ability even more impressive.
References
Ames, D. R. (in press). Inside the mind-reader’s toolkit: Projection and
stereotyping in mental state inference. Journal of Personality and Social
Psychology.
Baldwin, A. K., & Keysar, B. (1998, May). Perspective taking and atten-
tional resources. Paper presented at the American Psychological Soci-
ety, Washington, DC.
Batson, C. D., Batson, J. G., Slingsby, J. K., Harrell, K. L., Peekna, H. M.,
& Todd, R. M. (1991). Empathic joy and the empathy–altruism hypoth-
esis. Journal of Personality and Social Psychology, 61, 413–426.
Brinol, P., & Petty, R. E. (2003). Overt head movements and persuasion:
A self-validation analysis. Journal of Personality and Social Psychol-
ogy, 84, 1123–1139.
Chapman, G. B., & Johnson, E. J. (2002). Incorporating the irrelevant:
Anchors in judgments of belief and value. In T. Gilovich, D. Griffin, &
D. Kahneman (Eds.), Heuristics and biases: The psychology of intuitive
judgment (pp. 120–138). Cambridge, England: Cambridge University
Press.
Davis, H. L., Hoch, S. J., & Ragsdale, E. E. (1986). An anchoring and
adjustment model of spousal predictions. Journal of Consumer Re-
search, 13, 25–37.
Epley, N. (2004). A tale of tuned decks? Anchoring as accessibility and
anchoring as adjustment. In D. Koehler & N. Harvey (Eds.), The
Blackwell handbook of judgment and decision making (pp. 240–257).
Malden, MA: Blackwell.
Epley, N., & Gilovich, T. (2001). Putting adjustment back in the anchoring
and adjustment heuristic: Differential processing of self-generated and
experimenter-provided anchors. Psychological Science, 12, 391–396.
Epley, N., & Gilovich, T. (2004a). Are adjustments insufficient? Person-
ality and Social Psychology Bulletin, 30, 447–460.
Epley, N., & Gilovich, T. (2004b). Why adjustments are insufficient and
how to fix them. Manuscript submitted for publication.
Epley, N., Morewedge, C., & Keysar, B. (in press). Perspective taking in
children and adults: Equivalent egocentrism but differential correction.
Journal of Experimental Social Psychology.
Epley, N., Savitsky, K., & Gilovich, T. (2002). Empathy neglect: Recon-
ciling the spotlight effect and the correspondence bias. Journal of
Personality and Social Psychology, 83, 300–312.
Forster, J., & Strack, F. (1996). Influence of overt head movements on
memory for valenced words: A case of conceptual–motor compatibility.
Journal of Personality and Social Psychology, 71, 421–430.
Forster, J., & Strack, F. (1997). Motor actions in retrieval of valenced
information: A motor congruence effect. Perceptual and Motor Skills,
85, 1419–1427.
Galinsky, A. D., & Moskowitz, G. B. (2000). Perspective-taking: Decreas-
ing stereotype expression, stereotype accessibility, and in-group favor-
itism. Journal of Personality and Social Psychology, 78, 708–724.
Gilbert, D. T. (1989). Thinking lightly about others: Automatic compo-
nents of the social inference process. In J. S. Uleman & J. A. Bargh
(Eds.), Unintended thought (pp. 189–211). New York: Guilford Press.
Gilbert, D. T. (2002). Inferential correction. In T. Gilovich, D. Griffin, &
D. Kahneman (Eds.), Heuristics and biases: The psychology of intuitive
judgment (pp. 167–184). Cambridge, England: Cambridge University
Press.
Gilbert, D. T., & Gill, M. J. (2000). The momentary realist. Psychological
Science, 11, 394–398.
Gilbert, D. T., & Malone, P. S. (1995). The correspondence bias. Psycho-
logical Bulletin, 117, 21–38.
Gilovich, T., Medvec, V. H., & Savitsky, K. (2000). The spotlight effect in
social judgment: An egocentric bias in estimates of the salience of one’s
own actions and appearance. Journal of Personality and Social Psychol-
ogy, 78, 211–222.
Gilovich, T., Savitsky, K., & Medvec, V. H. (1998). The illusion of
338
EPLEY, KEYSAR, VAN BOVEN, AND GILOVICH

transparency: Biased assessments of others’ ability to read our emotional
states. Journal of Personality and Social Psychology, 75, 332–346.
Goldman, A. I. (1992). In defense of the simulation theory. Mind &
Language, 7, 104–119.
Gopnik, A., & Wellman, H. M. (1992). Why the child’s theory of mind
really is a theory. Mind & Language, 7, 145–171.
Gopnik, A., & Wellman, H. M. (1994). The theory theory. In L. A.
Hirschfeld & S. A. Gelman (Eds.), Mapping the mind: Domain speci-
ficity in cognition and culture (pp. 257–293). New York: Cambridge
University Press.
Gordon, R. M. (1992). The simulation theory: Objections and misconcep-
tions. Mind and Language, 7, 11–34.
Karniol, R. (2003). Egocentrism versus protocentrism: The status of self in
social prediction. Psychological Review, 110, 564–580.
Keysar, B. (1994). The illusory transparency of intention: Linguistic per-
spective taking in text. Cognitive Psychology, 23, 165–208.
Keysar, B. (1998). Language users as problem solvers: Just what ambiguity
problem do they solve? In S. R. Fussell & R. J. Kreuz (Eds.), Social and
cognitive psychological approaches to interpersonal communication
(pp. 175–200). Hillsdale, NJ: Erlbaum.
Keysar, B., & Barr, D. J. (2002). Self-anchoring in conversation: Why
language users don’t do what they “should.” In T. Gilovich, D. Griffin,
& D. Kahneman (Eds.), Heuristics and biases: The psychology of
intuitive judgment (pp. 150–166). Cambridge, England: Cambridge Uni-
versity Press.
Keysar, B., Barr, D. J., Balin, J. A., & Brauner, J. S. (2000). Taking
perspective in conversation: The role of mutual knowledge in compre-
hension. Psychological Science, 11, 32–39.
Krueger, J., & Clement, R. W. (1994). The truly false consensus effect: An
ineradicable and egocentric bias in social perception. Journal of Per-
sonality and Social Psychology, 67, 596–610.
Mussweiler, T., & Strack, F. (2001). “Considering the impossible”: Ex-
plaining the effects of implausible anchors. Social Cognition, 19, 145–
160.
Nickerson, R. S. (1999). How we know—and sometimes misjudge—what
others know: Imputing one’s own knowledge to others. Psychological
Bulletin, 125, 737–759.
Piaget, J. (1959). Judgment and reasoning in the child. Paterson, NJ:
Littlefield, Adams, & Co.
Pronin, E., Gilovich, T., & Ross, L. (in press). Objectivity in the eye of the
beholder: Divergent perceptions of bias in self versus others. Psycho-
logical Review.
Quattrone, G. A. (1982). Overattribution and unit formation: When behav-
ior engulfs the person. Journal of Personality and Social Psychology, 42,
593–607.
Queen. (1991). Another one bites the dust. On The Game [CD]. Holly-
wood, CA: Hollywood records. (Original release 1980)
Robinson, R. J., & Keltner, D. (1996). Much ado about nothing? Revi-
sionists and traditionalists choose an introductory English syllabus.
Psychological Science, 7, 18–24.
Robinson, R. J., Keltner, D., Ward, A., & Ross, L. (1995). Actual versus
assumed differences in construal: “Naive realism” in intergroup percep-
tion and conflict. Journal of Personality and Social Psychology, 68,
404–417.
Rock, I., & Nijhawan, R. (1989). Regression to egocentrically determined
description of form under conditions of inattention. Journal of Experi-
mental Psychology: Human Perception and Performance, 15, 259–272.
Ross, L., & Ward, A. (1996). Naive realism in everyday life: Implications
for social conflict and misunderstanding. In T. Brown, E. Reed, & E.
Turiel (Eds.), Values and knowledge (pp. 103–135). Hillsdale, NJ:
Erlbaum.
Trope, Y., & Gaunt, R. (2000). Processing alternative explanations of
behavior: Correction or integration? Journal of Personality and Social
Psychology, 79, 344–354.
Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heu-
ristics and biases. Science, 185, 1124–1130.
Tversky, A., & Koehler, D. J. (1994). Support theory: A nonextensional
representation of subjective probability. Psychological Review, 101,
547–567.
Van Boven, L., Dunning, D., & Loewenstein, G. (2000). Egocentric
empathy gaps between owners and buyers: Misperceptions of the en-
dowment effect. Journal of Personality and Social Psychology, 79,
66–76.
Vokey, J. R., & Read, J. D. (1985). Subliminal messages: Between the
devil and the media. American Psychologist, 40, 1231–1239.
Wegener, D. T., & Petty, R. E. (1995). Flexible correction processes in
social judgment: The role of naı¨ve theories in corrections for perceived
bias. Journal of Personality and Social Psychology, 68, 36–51.
Weingartner, K. M., & Klin, C. M. (2001, November). Readers’ use of
privileged information in estimating a character’s knowledge. Paper
presented at the 41st Annual Meeting of the Psychonomic Society,
Orlando, FL.
Wells, G. L., & Petty, R. E. (1980). The effects of overt head movements
on persuasion: Compatibility and incompatibility of responses. Basic
and Applied Social Psychology, 1, 219–230.
Wilson, T. D., Houston, C., Etling, K. M., & Brekke, N. (1996). A new
look at anchoring effects: Basic anchoring and its antecedents. Journal
of Experimental Psychology: General, 4, 387–402.
Wright, W. F., & Anderson, U. (1989). Effects of situation familiarity and
incentives on use of the anchoring and adjustment heuristic for proba-
bility assessment. Organizational Behavior and Human Decision Pro-
cesses, 44, 68–82.
Received January 8, 2004
Revision received April 19, 2004
Accepted April 20, 2004 䡲
339
PERSPECTIVE TAKING AS ADJUSTMENT