Egocentrism Over E-Mail: Can We Communicate as Well as We Think?
New York University, Stern School of Business
University of Chicago, Graduate School of Business
Jason Parker and Zhi-Wen Ng
University of Illinois at Urbana–Champaign
Without the benefit of paralinguistic cues such as gesture, emphasis, and intonation, it can be difficult to
convey emotion and tone over electronic mail (e-mail). Five experiments suggest that this limitation is
often underappreciated, such that people tend to believe that they can communicate over e-mail more
effectively than they actually can. Studies 4 and 5 further suggest that this overconfidence is born of
egocentrism, the inherent difficulty of detaching oneself from one’s own perspective when evaluating the
perspective of someone else. Because e-mail communicators “hear” a statement differently depending on
whether they intend to be, say, sarcastic or funny, it can be difficult to appreciate that their electronic
audience may not.
Keywords: e-mail, egocentrism, overconfidence, miscommunication, nonverbal behavior
Social judgment is inherently egocentric. When people try to
imagine the perspective, thoughts, or feelings of someone else, a
growing body of evidence suggests that they use themselves as an
anchor or reference point. Although precisely why this occurs—
whether the result of an overlearned and generally valid heuristic,
the residual byproduct of an earlier stage of childhood egocen-
trism, or the inevitable consequence of an effortful cognitive
process such as anchoring and adjustment—is a matter of some
debate, the fact remains that the assessment of another’s perspec-
tives is influenced, at least in part, by one’s own (Camerer,
Loewenstein, & Weber, 1989; Epley, Keysar, Van Boven, &
Gilovich, 2004; Fischhoff, 1975; Flavell, 1977; Fussell & Krauss,
1991; Gilovich, Medvec, & Savitsky, 2000; Gilovich, Savitsky, &
Medvec, 1998; Hoch, 1987; Inhelder & Piaget, 1958; Kelley &
Jacoby, 1996; Keysar, Barr, & Horton, 1998; Keysar & Bly, 1995;
Nickerson, 1999, 2001; Ross & Ward, 1996).
Nowhere is this tendency more apparent than in the music
tapping study conducted by Elizabeth Newton (1990). Participants
in her study were asked to tap the rhythm of a well-known song to
a listener and then assess the likelihood that the listener would
correctly identify the song. The results were striking: Tappers
estimated that approximately 50% of listeners would correctly
identify the song, compared with an actual accuracy rate of 3%.
What accounts for this dramatic overestimation? The answer
becomes immediately apparent when one contrasts the perspec-
tives of tappers and listeners, as Ross and Ward (1996) invited
their readers to do when describing Newton’s results. Whereas
tappers could inevitably “hear” the tune and even the words to the
song (perhaps even a “full orchestration, complete with rich har-
monies between string, winds, brass, and human voice”), the
listeners were limited to “an aperiodic series of taps” (Ross &
Ward (1996, p. 114). Indeed, it was difficult from the listener’s
perspective to even tell “whether the brief, irregular moments of
silence between taps should be construed as sustained notes, as
musical “rests” between notes, or as mere interruptions as the
tapper contemplates the “music” to come next” (p. 114). So rich
was the phenomenology of the tappers, however, that it was
difficult for them to set it aside when assessing the objective
stimuli available to listeners. As a result, tappers assumed that
what was obvious to them (the identity of the song) would be
obvious to their audience.
Of course, everyday communication is far less constrained than
in the music tapping study. Seldom are we required to tap a course
lecture or describe our research via Morse code (although it can
An analogous example comes from a study by Keysar and Henly
(2002) in which participants read aloud several ambiguous sentences (such
as “Angela killed the man with the gun”) to another study participant.
Speakers read the statement after reading a scenario that resolved the
ambiguity of the sentence (e.g., indicated whether the gun was the murder
weapon or a possession of the victim), a scenario that was unavailable to
listeners. As in the case of the tapping study, speakers assumed that what
was obvious to them (i.e., the meaning of the sentence) would be obvious
to the listener. Consistent with the speculation of Newton (1990) and Ross
and Ward (1996), Keysar and Henly found that the overestimation was due,
at least in part, to participants underestimating the ambiguity of their own
Justin Kruger, New York University, Leonard N. Stern School of
Business; Nicholas Epley, University of Chicago, Graduate School of
Business; Jason Parker and Zhi-Wen Ng, Department of Psychology,
University of Illinois at Urbana–Champaign.
This research was supported by Research Grant 1-2– 69853 from the
University of Illinois Board of Trustees awarded to Justin Kruger and by
National Science Foundation Grant SES-0241544 awarded to Nicholas
Epley. We thank Tom Gilovich and Ken Savitsky for their helpful sug-
gestions throughout this research.
Correspondence concerning this article should be addressed to Justin
Kruger, New York University, Leonard N. Stern School of Business, 40
West 4th Street, Suite 816, New York, NY 10012. E-mail:
Journal of Personality and Social Psychology Copyright 2005 by the American Psychological Association
2005, Vol. 89, No. 6, 925–936 0022-3514/05/$12.00 DOI: 10.1037/0022-3522.214.171.1245
occasionally seem that way). However, we believe that the lessons
learned from Newton’s research are nevertheless applicable to
everyday communication and one facet of everyday communica-
tion in particular: electronic mail (e-mail).
E-mail is one of the most successful computer applications yet
devised (Dimmick, Kline, & Stafford, 2000; Marold & Larson,
1999; Wittaker & Sidner, 1997). According to the U.S. Depart-
ment of Commerce, nearly half of the U.S. population currently
uses e-mail, and the number of e-mail messages currently outnum-
ber letters sent by the U.S. postal service (Thompson, 2001). Some
have even claimed that e-mail, along with the invention of writing,
printing, and telegraphy, represents the only truly revolutionary
change in communication technology (de la Sola Pool, 1984).
Although text-based communication is nothing new (people have
been writing letters to each other for centuries), its ubiquity is:
Whereas letters were at best a monthly or weekly affair, people use
e-mail on a daily— even hourly—basis (Gatz & Hirt, 2000). And
for good reason: E-mail is a quick and convenient method for
people to conduct business, stay in touch with friends and family,
and even collect data.
But there is something missing from e-mail as well. As psy-
chologists and laypeople know, much of communication is non-
verbal (Archer & Akert, 1977; Argyle, 1970; DePaulo & Fried-
man, 1998). Although the value of nonverbal communication is
sometimes overstated (DePaulo, 1992; Krauss et al., 1981; Rime,
1982), the fact remains that nonverbal information is an important
cue to the speaker’s meaning, particularly when the literal content
of the message is ambiguous (Allbritton, McKoon, & Ratcliff,
1996; Price, Ostendorf, Shattuck-Hufnagel, & Fong, 1991). After
all, the same statement can, depending on tone, emphasis, and
expression, be either sarcastic or serious, disrespectful or deferen-
tial, sanguine or somber (Abrahams, 1962; Clark, 1996; Drew,
1987; Goffman, 1959). Similarly, people use inflection and gesture
to soften the blow of negative communication, to literally tone
down bad news or mute unfavorable feedback. Whereas speech
conveys not only what is said but also how it is said, e-mail is
limited to the former. As such, e-mail is an inherently more
impoverished communication medium than voice or face-to-face
communication (Kiesler, Siegel, & McGuire, 1984; Sproull &
Kiesler, 1986; Thompson & Nadler, 2002).
This limitation is likely to be fertile ground for miscommuni-
cation and, in particular, a lack of awareness of that miscommu-
nication. E-mail communicators, after all, are well aware of the
precise message that they intend to convey. Over e-mail, we know
that we are being sarcastic when referring to the comic brilliance
of, say, Adam Sandler, just as we know that we are not when
making the same statement of John Cleese. Note, however, that
what is obvious to us may be considerably less so to the person on
the other end of the computer. Whereas we “hear” a statement
differently depending on whether we are speaking sarcastically or
seriously, our e-mail audience, without the paralinguistic cues that
in voice communication flag sarcasm, may not. And because it can
be difficult to separate one’s own experience of a stimulus from
the stimulus available to one’s audience, one’s e-mail message
may be more ambiguous than one realizes.
To be sure, to help resolve such ambiguities, the tech-savvy
occasionally make use of symbolic so-called emoticons, such as
the smiley face “:-),” the winking smiley face “;-),” or the dreaded
“:-(.” This is not such a clear remedy, however, as many emoticons
are themselves ambiguous. Emoticonuniverse.com, for instance,
lists no fewer than 300 emoticons, some of which are far from
intuitive, including “;-?,” “%-(,” and “:⬃/.”
often neglect to use emoticons altogether, particularly when the
sarcasm or humor is “obvious.” We are a reminded of an e-mail a
well-known psychologist sent to her colleagues announcing a
dinner reception in honor of a job candidate. After the usual
promise of free food, good drink, and stimulating conversation, she
sarcastically pointed out that “talking to the candidate is not
required; just don’t embarrass us.” Much to her surprise, some
colleagues took offense at the comment, thinking that she was
genuinely concerned about the embarrassment that her boorish
coworkers might cause. Apparently her colleagues did not realize
that she was being sarcastic, and apparently she did not realize that
her sarcasm was unclear.
We suspect such misunderstandings are common. People rou-
tinely overestimate how well they can communicate over e-mail,
we offer, particularly when the meaning of the message is ambig-
uous. We further argue that this overestimation is caused, at least
in part, by egocentrism, the inherent difficulty of moving beyond
one’s subjective experience of a stimulus and imagining how the
stimulus might be evaluated by someone who does not share one’s
Given the growing popularity of e-mail and the fact that suc-
cessful communication depends partly on the ability to anticipate
miscommunication (Keysar & Henly, 2002), these predictions are
of both practical and theoretical importance. Despite this, our
hypothesis has yet to be investigated in either the social judgment
literature or in the rapidly emerging literature on computer-
mediated communication. Several researchers have investigated
the ability of individuals to communicate over e-mail (e.g., Dennis
& Kinney, 1998; Hebert & Vorauer, 2003; Thompson & Nadler,
2002; Walther, 1993, 1995; Walther, Loh, & Granka, 2005), but no
researchers have investigated the perceived ability of individuals
to do so, nor have they contrasted the two. The one published
exception of which we are aware is a study by Sherman et al.
(2001). These researchers found that home page creators predicted
that others would form a more positive impression of them than
was actually the case, and that this was less true when participants
communicated face to face. If one assumes that the home page
creators intended to portray a positive impression of themselves,
then these data could be interpreted as evidence of overestimation
of communication effectiveness. However, note that these re-
searchers did not investigate e-mail, nor did they investigate the
cause of this overestimation.
The present research was designed to more directly examine
overconfidence over e-mail as well as the mechanism thought to
According to the Web site, these emoticons imply that the e-mailer
“speaks with a forked tongue,” is “confused,” and is “unsure,” respectively.
The common usage of the term sarcasm in the United States differs
from most dictionary definitions. Webster’s, for instance, offers the exam-
ple of “My, you’re early” being said to a latecomer as an example not of
sarcasm, but of irony (Agnes, 1999, p. 1272), whereas most Americans
would say that the opposite is true. In the present article, we follow the
convention set in past research (Keysar, 1994; Keysar et al., 1998) and
colloquial use by referring to instances in which one’s intended meaning
differs from the literal meaning as sarcasm.
KRUGER, EPLEY, PARKER, AND NG
underlie it. In five studies, we compared the perceived and actual
ability of participants to communicate over e-mail. In each study,
we hypothesized that participants’ predicted ability would exceed
their actual ability. Our last two studies tested the egocentrism
account of this overconfidence by experimentally manipulating the
phenomenological experience of e-mailers.
We also compared overconfidence over e-mail with more tra-
ditional modes of communication. Whereas our egocentrism anal-
ysis suggests that e-mail should be associated with greater over-
confidence than, say, voice communication, this is by no means a
forgone conclusion. Prior work has found that for all the value of
nonverbal information, people have occasional difficulty interpret-
ing nonverbal information (DePaulo, 1992; Keysar & Henly, 2002;
Lanzetta & Kleck, 1970). What is more, communicators are often
insensitive to these difficulties, believing their nonverbal cues are
clearer to others than they really are (Keysar & Henly, 2002)—
even when they do not wish them to be (Gilovich et al., 1998;
Vorauer & Claude, 1998). All of this suggests that voice commu-
nication might be associated with more, rather than less, overcon-
fidence—the opposite of what we predict. To help shed light on
this issue, Studies 2 and 3 compared overconfidence in e-mail
communication with that of voice communication.
Participants in Study 1 were given a list of 10 topics and asked
to write two statements about each one. Half of the statements
were to be serious and the other half sarcastic. Senders e-mailed
the statements to another participant, who attempted to identify
which sentences were intended to be sarcastic and which were not.
The senders then predicted the receivers’ accuracy. We predicted
that senders would overestimate their ability to communicate sar-
casm to the receiver.
Participants. Twelve Cornell University students participated in ex-
change for extra credit in an introductory psychology or human develop-
ment course. All participants had their own personal e-mail account and at
least some experience using e-mail, a characteristic of all participants in
Procedure. Participants were recruited in pairs for a study of sarcasm.
Each participant was given a list of 10 topics and asked to e-mail two
one-sentence statements about each one. The topics varied: Within each
pair, one participant was asked to write about parties, art, dating, Califor-
nia, sports, TV, food, cars, literature, and life in Ithaca, New York (the
town in which the study took place), and the other participant was asked to
write about dorm life, movies, computers, romance, Manhattan, athletics,
family, music, politics, and the Cornell Greek system. According to a
predetermined random order, participants were instructed to make half of
the sentences serious, such as “I do not like first dates,” and the other half
sarcastic, such as “I really enjoy dating because I like feeling as self-
conscious and inadequate as possible.” To ensure that participants under-
stood what we meant by sarcasm, we presented several examples designed to
convey the colloquial meaning of the term: a statement in which one’s
intended meaning is the exact opposite of the literal meaning. Apart from the
examples, participants were given no specific instructions regarding how to
accomplish their goal, except being told to avoid using emoticons such as “;-).”
Participants then anticipated how the receiver would interpret their
statements. Specifically, participants were told (correctly) that the receiver
would attempt to identify which statements were intended to be sarcastic
and which were not. Participants then indicated, for each topic, whether
they thought the receiver would be able to correctly identify the nature
(sarcastic or nonsarcastic) of the two statements by checking a box marked
“yes” or “no.” Each participant then read the 20 statements that had been
e-mailed by the other participant and indicated on a questionnaire which
statements they believed were sarcastic. The design was thus fully within-
participants, with each participant serving as both a sender and receiver.
Results and Discussion
We expected participants to overestimate their ability to com-
municate sarcasm. To test this hypothesis, senders’ predictions of
the receivers’ accuracy were compared with the receivers’ actual
accuracy. Because the data for each pair are interdependent, the
data were analyzed at the level of the dyad. Specifically, we
averaged each person’s estimate of the number of topics (out of
10) that they expected the other person to successfully decode and
compared that number with the number of topics actually decoded.
As expected, participants were overconfident: On average, partici-
pants expected 97% of their topics to be correctly decoded, compared
with the 84% that actually were, t(5) ⫽ 3.23, p ⫽ .023,d⫽ 1.32.
We attribute these results to egocentrism. Because senders
knew, for example, that the statement “Blues Brothers, 2000—now
that’s a sequel,” was meant to be sarcastic, they egocentrically
assumed that their audience would as well. They presumably did
not realize how ambiguous the statement really is without verbal
emphasis on the word “that’s,” a facial gesture such as an eye roll,
or some background information about the communicator (such as
his or her taste in films).
Note, however, that despite reliable overconfidence, accuracy
rates were quite high (84%). It would therefore be misleading to
suggest from these data that people are poor at communicating
sarcasm over e-mail. These data do suggest, however, that how-
ever able people are, they are not as able as they believe.
One limitation of Study 1 was that although participants were
overconfident in their ability to communicate over e-mail, it was
unclear whether this overconfidence had anything to do with
e-mail per se. After all, numerous studies attest to the general
tendency of individuals to be overconfident in their endeavors
(e.g., Dunning, Griffin, Milojkovic, & Ross, 1990; Keren, 1987;
Lichtenstein, Fischoff, & Phillips, 1982; Oskamp, 1965; Vallone,
Griffin, Lin, & Ross, 1990; Wright, Rowe, Boger, & Gammack,
1994). The previous study may thus have simply been yet another
instantiation of that general tendency. To resolve this ambiguity, in
Study 2 we contrasted the overconfidence people display when
communicating via e-mail with the overconfidence they display
when communicating with their voice. If our egocentrism analysis
is correct, then overconfidence should be greater over e-mail than
over voice—not because of differences in the perceived ability to
communicate but because of differences in the actual ability to
Study 2 also went beyond Study 1 by assessing confidence
among both senders and receivers. Although overconfidence in the
ability to communicate a message can be problematic, it can be
doubly troublesome when accompanied by overconfidence in the
ability to interpret that message. We investigated both kinds of
overconfidence in Study 2.
EGOCENTRISM OVER E-MAIL
Participants. Sixty Cornell University students participated in ex-
change for extra credit in an introductory psychology or human develop-
Procedure. Participants completed the experiment in pairs. On arrival
to the lab, each member of the pair was escorted to a private room and
given a questionnaire. The questionnaire explained that the study con-
cerned how people detect and communicate sarcasm, and that the first part
of the experiment involved selecting a series of statements that the other
person in the experiment would later be asked to identify as either sarcastic
or serious. Participants were next given a list of 20 statements about a
number of topics, such as food, Greek life, and Ithaca weather. Half were
identified (correctly) as being intended by the original author of the
statement to be sarcastic and the other half nonsarcastic. Participants within
each pair received a different list so that they would not be sending each
other the same statements.
Participants then selected 10 of the statements to communicate to the
other person in the experiment. They were told to select the statements that
they believed would be easiest for the other person to identify as sarcastic
or serious. The number of serious versus sarcastic statements varied ran-
domly across participants, with a mean of five statements per category per
participant. Up to this point in the experiment, no mention was made of
exactly how the statements would be transmitted to the other participant
(i.e., via e-mail, voice, or smoke signal) to ensure that participants in the
e-mail and voice conditions did not select systematically different
Meanwhile, the experimenter randomly assigned one member of the
dyad to the e-mail condition and the other member to the voice condition.
Participants in the e-mail condition were escorted to a computer and asked
to type each statement they had selected into the computer exactly as
written. Participants in the voice condition, in contrast, were escorted to a
tape recorder and asked to read each statement into the tape recorder
exactly as written. Both participants were told that the other participant
would later attempt to identify the nature (sarcastic or serious) of each
Once they had finished recording their statements, participants predicted
how many statements the other participant would be able to decode
correctly. Specifically, participants checked a box marked “yes” or “no”
for each statement to indicate whether they thought the other person in the
experiment would be able to correctly identify the true nature (sarcastic or
nonsarcastic) of the statement.
Finally, participants listened to (or read) the statements selected by the
other participant. Participants then indicated whether they thought each
statement was intended to be sarcastic or nonsarcastic as well as whether
they thought they had correctly identified the statement (yes or no).
participants were then thanked, debriefed, and dismissed.
Results and Discussion
Study 2 was a 2 (condition: e-mail vs. voice) ⫻ 2 (accuracy:
anticipated vs. actual) fully within-group factorial, with the dyad
as the level of analysis. Because participants communicated dif-
ferent numbers of sarcastic statements, perceived and actual accu-
racy were converted to a percentage. Responses from one group
were over 3 SDs away from the mean on several dependent
variables and were excluded from the analysis, yielding a final
sample size of 29 dyads.
Not surprisingly, participants in the voice condition communi-
cated more effectively than those in the e-mail condition. As
shown in Figure 1, participants who listened to the statements
decoded nearly three-quarters of them, compared with an accuracy
rate indistinguishable from chance (50%) among participants who
read them on e-mail. But more important, Figure 1 also shows that
e-mailers failed to anticipate this difference. Although partici-
pants’ actual ability to communicate sarcasm varied considerably
depending on whether they used e-mail or their voice, t(28) ⫽
2.53, p ⫽ .017, d ⫽ 0.47, their confidence in their ability did not
(t ⬍ 1). This between-condition difference in overconfidence was
confirmed by a 2 (e-mail vs. voice) ⫻ 2 (predicted vs. actual) fully
within-group analysis of variance (ANOVA), which yielded only
the predicted interaction, F(1, 28) ⫽ 5.20, p ⫽ .030,
Participants were overconfident when they communicated over
e-mail, t(28) ⫽ 3.40, p ⫽ .002, d ⫽ .61, but not when they
communicated over voice, t ⬍ 1.
An interesting finding was that an analogous pattern of over-
confidence emerged among participants’ ability to detect sarcasm.
Recall that after participants attempted to identify the nature (sar-
castic or serious) of each statement, they also indicated whether
they thought they had been successful. Although participants in the
e-mail condition were less able than those in the voice condition to
decode sarcasm, they were no less confident (Ms ⫽ 89% & 91%,
respectively), t ⬍ 1. A 2 ⫻ 2 ANOVA comparing receivers’
perceived and actual ability to detect sarcasm over e-mail and
voice revealed that overall, participants were overconfident, F(1,
28) ⫽ 33.75, p ⬍ .001,
⫽ 0.55, but this was particularly true for
receivers in the e-mail condition, F(1, 28) ⫽ 3.68, p ⫽ .065,
0.12. In sum, participants in this experiment were overconfident in
their ability to both convey and detect sarcasm over e-mail, but
they were considerably less overconfident over voice.
Taken together, the results of Studies 1 and 2 suggest that people
are overconfident in their ability to communicate sarcasm over
e-mail. Just as people have difficulty divorcing themselves from
Decision theorists may point out that an answer of “no” to this question
is irrational because it implies that the person believes that his or her
answer to the question is less than 50% likely to be correct, and if so, the
participant should change it (there being only one other possible response).
Given the frequency of “no” responses, we suspect that participants inter-
preted the question differently than decision theorists.
Figure 1. Anticipated and actual ability to communicate by condition
KRUGER, EPLEY, PARKER, AND NG
the melody that inevitably accompanies the act of tapping a song,
so, too, do people who send a sarcastic e-mail have difficulty
divorcing themselves from the sarcastic tone that inevitably ac-
companies the act of typing a sarcastic statement.
There are, however, several limitations of the data presented
thus far. First, whereas Studies 1 and 2 focused exclusively on the
communication of sarcasm (or the lack thereof), note that the
egocentrism analysis we have offered ought to apply to any subtle
form of communication that relies on paralinguistic cues. Thus, in
Study 3, we expanded our empirical focus to include not only
sarcasm and seriousness, but sadness and anger as well.
Second, whereas participants in Studies 1 and 2 were strangers,
note that the majority of (nonspam) e-mails in daily life are
between acquaintances. This presents a potential threat to external
validity. After all, it stands to reason that people are generally
better at communicating with people they know than with people
they do not, which, all else equal, should be associated with
decreased, perhaps even eliminated, overconfidence. To address
this possibility, Study 3 compared overconfidence among strangers
versus friends. In addition to addressing external validity, note that
this comparison also enabled us to examine the more general
impact of familiarity on overconfidence in communication.
Finally, whereas participants in the voice condition of Study 2
were unable to see one another, this is hardly characteristic of most
voice interactions, which are face-to-face (Panko & Kinney, 1995).
As such, a more ecologically valid test of the comparative advan-
tages and disadvantages of e-mail versus voice requires a condition
in which participants communicate face-to-face. Study 3 included
just such a condition.
In addition to increasing external validity, note that including
three levels of media richness allowed us to examine precisely
what it is about e-mail that engenders increased overconfidence.
Because we observed greater overconfidence in e-mail than in
voice-only communication in Study 2, the lack of intonation alone
appears sufficient to increase overconfidence, but gesture and
facial expression may also play a role (Daft & Lengel, 1986; Daft
& Trevino, 1987). By comparing perceived and actual ability to
communicate in e-mail, voice-only, and face-to-face communica-
tion, we could find out.
Participants. One hundred fifty-four pairs of University of Illinois
students participated on a volunteer basis.
Procedure. Each pair of participants was recruited door-to-door by one
of 77 separate experimenters following a standardized script. Depending on
condition, the experimenter was instructed to recruit a pair who identified
themselves either as friends or strangers. The experiment took place in one
of the participants’ residences, which the experimenter verified had Inter-
Once each member of the pair agreed to participate, the experimenter
explained that the study would involve constructing, and then communi-
cating, a series of five statements, each of which would be paired with one
of four emotions or tones. Their job, they were told, would be to success-
fully communicate each specific emotion or tone to the other participant.
The medium with which the statements would be communicated, however,
varied by condition: Approximately one third of participants were told that
they were to e-mail the statements to one another, and the remaining two
thirds were told that they were to speak the statements to one another.
Within the latter group, half were further told that they would do so either
with or without being able to see one another. That is, instead of commu-
nicating face-to-face, they would speak with their backs to one another.
Once the pair was told the medium in which they were to communicate,
each member was given a list of five topics (music, Illini football, dating,
dorm food, and dorm life). Each topic was matched to one of four emotions
or tones (sarcasm, seriousness, anger, or sadness). For each topic, the
participant’s job was to construct a statement that would successfully
convey the associated emotion or tone to the other participant when either
typed or spoken but without stating the emotion or tone explicitly (e.g., “I
am being sarcastic”). Participants also indicated, for each topic, whether
they expected the other participant to successfully guess the emotion or
tone (on a dichotomous “yes” or “no” scale). Special care was taken to
ensure that (a) participants were aware of the communication medium prior
to making their predictions (e.g., whether they would be communicating
via e-mail, with their voice only, or face-to-face), and that (b) participants
understood that the other participant would be picking from a list of the
four possible emotions or tones for each topic. As in Study 1, participants
in the e-mail condition were restricted from using emoticons.
Next, participants communicated the five statements to one another, one
at a time, as instructed (i.e., via e-mail, voice-only, or face-to-face).
Specifically, one member of the pair (the sender) spoke or typed all five of
his or her statements to the other participant (the receiver), and then the pair
switched roles. This was done one sentence at a time, which in the e-mail
condition meant that the statements were e-mailed one at a time rather than
in a single e-mail. After each statement, the receiver attempted to guess the
emotion or tone the sender was attempting to convey (from a list of four)
and indicated his or her confidence in that guess by checking a box labeled
“yes” or “no.” Once this was complete, all participants were then thanked
Results and Discussion
Our primary prediction was that overconfidence would be
greater when participants communicated over e-mail than when
participants communicated with their voice. To test this prediction,
we conducted a 2 (accuracy: anticipated vs. actual) ⫻ 2 (order:
Round 1 vs. Round 2) ⫻ 2 (acquaintanceship: stranger vs.
friend) ⫻ 3 (medium: e-mail vs. voice-only vs. face-to-face)
mixed-model ANOVA with the dyad as the level of analysis. The
first two factors in this design were within-participants variables,
and the second two were between-participants variables.
This analysis revealed several significant effects. First, we ob-
served a significant main effect for communication medium, F(1,
148) ⫽ 3.67, p ⫽ .028,
⫽ .05, indicating that on average, both
predicted and actual accuracy for both friends and strangers were
higher in the voice conditions than in the e-mail conditions. We
also observed a main effect for order, F(1, 148) ⫽ 11.67, p ⫽ .001,
⫽ .07, indicating that on average, predicted and actual accuracy
for both friends and strangers were higher during the second round
of communication than the first. We also observed an unexpected
Accuracy ⫻ Order ⫻ Acquaintanceship interaction, F(1, 48) ⫽
4.17, p ⫽ .043,
⫽ .03. On average, overconfidence tended to
decrease from Round 1 to Round 2 for strangers but tended to
increase for friends.
No other main effects or interactions were significant, except for
the following two exceptions. First, we observed the predicted
main effect for accuracy: Participants predicted that they would
successfully communicate more emotions or tones (M ⫽ 88.8%)
We included this prerequisite regardless of condition in order to
minimize sampling bias.
EGOCENTRISM OVER E-MAIL
than they actually did (M ⫽ 70.4%), F(1, 148) ⫽ 118.26, p ⬍ .001,
⫽ .44. Second, and more important, we observed a significant
Accuracy ⫻ Medium interaction. As Figure 2 shows, participants
were more overconfident when they communicated over e-mail
than when they communicated with their voice, F(2, 148) ⫽ 3.31,
p ⫽ .039,
⫽ 0.04. As in Study 2, participants’ ability to
communicate varied considerably depending on whether they used
e-mail or their voice, F(2, 151) ⫽ 4.68, p ⫽ .011,
⫽ 0.06, but
their confidence in their ability did not, F ⬍ 1, ns.
An interesting result was that familiarity also had no influence
on the results, Fs ⬍ 1, ns. This null effect is important for two
reasons. First, it suggests increased familiarity does not necessarily
translate into increased communication accuracy, contrary to what
one might expect. Second, and perhaps more important, it makes
it clear that the overconfidence observed in Studies 1 and 2 did not
emerge simply because participants were strangers rather than
There was also no difference in either predicted or actual com-
munication accuracy between the face-to-face and voice-only con-
ditions, Fs ⬍ 1, ns. These data, together with the data presented in
Study 2, suggest that the increased overconfidence in e-mail is due
to the lack of intonation and vocalization rather than the lack of
gesture and expression. That said, this null result is not without
alternative interpretations. For instance, whereas participants in
Studies 1 and 2 predicted their accuracy after communicating over
their assigned medium, participants in Study 3 did so before
communicating. Although all participants in Study 3 were well
aware of the medium in which they were about to communicate
their message, and although the results of Study 3 perfectly dove-
tail with the results of Studies 1 and 2, it could be argued that
collecting predicted accuracy before direct experience with the
communication medium diminishes the latter’s impact on the
former. This methodological feature was necessary in Study 3 to
avoid introducing a procedural confound, as only participants in
the face-to-face condition would have had access to a listener’s
nonverbal response to their communication when predicting their
communication accuracy. Nevertheless, future research that effec-
tively counterbalances predicted accuracy and actual communica-
tion is necessary to draw definitive conclusions about differences
in overconfidence between voice and face-to-face interactions.
Our final set of analyses focused on overconfidence in partici-
pants’ ability to comprehend subtleties in communication. Recall
that participants in Study 2 were not only overconfident in their
ability to convey sarcastic messages, but were also overconfident
in their ability to detect sarcastic messages. Study 3 showed a
similar pattern. Overall, participants believed that they understood
more emotions and tones (M ⫽ 89.4%) than they in fact did (M ⫽
70.4%), F(1, 148) ⫽ 114.07, p ⬍ .001,
⫽ 0.44. This effect was
bigger, however, in the e-mail condition (M
⫽ 89.3% vs.
⫽ 62.8%) than in either the face-to-face (M
89.4% vs. M
⫽ 73.9%) or voice-only (M
⫽ 89.4% vs.
⫽ 73.3%) conditions, F(2, 148) ⫽ 3.66, p ⫽ .028,
0.05. Here, too, this effect was uninfluenced by acquaintanceship:
Participants were just as overconfident when communicating with
a stranger as when they were communicating with a friend, F ⬍ 1,
Why do people overestimate their ability to communicate over
e-mail? Our thesis is that people are biased by their own phenom-
enology when trying to imagine the perspective of someone else.
Because people know when they are trying to be sarcastic, for
instance, they egocentrically assume that their audience will know
Study 4 was the first of two studies designed to test this
egocentrism explanation directly. Participants communicated sar-
casm to one another in a manner similar to Study 2, with one major
exception: Before predicting whether the other participant would
interpret the statements correctly, participants were asked to read
each statement out loud into a tape recorder. Half of the partici-
pants were asked to intonate in a manner consistent with the
intended meaning (i.e., to read the sarcastic messages sarcastically)
and the other half in a manner inconsistent with the intended
meaning (i.e., to read the sarcastic messages seriously).
Our reasoning was simple. If people are overconfident in their
ability to communicate over e-mail partly because of the difficulty
of moving beyond their own perspective, then forcing people to
adopt a perspective different from their own ought to reduce this
overconfidence. As a result, participants who vocalized the mes-
sages in a manner inconsistent with the intended meaning should
be less overconfident than those who vocalized the messages in a
manner consistent with the intended meaning.
Participants. Fifty-four University of Illinois students participated in
pairs in exchange for course credit in an introductory psychology course.
Procedure. As in Study 2, participants were given one of two lists of
20 statements, half of which were sarcastic and half of which were not.
Participants were told to select 10 statements that they felt would be
particularly easy for the other participant to identify as either sarcastic or
serious and to type each one into a computer for the other participant to
read later in the experiment. Up to this point, the procedure mirrored Study
2 except that there was no voice condition (i.e., all participants communi-
cated with one another via e-mail), and participants were asked to select
five sarcastic and five serious statements (instead of a varying number as
in Study 2).
When they finished typing, participants were asked to record each
statement into a tape recorder, ostensibly for “later analysis.” Half of the
Figure 2. Anticipated and actual ability to communicate by condition
KRUGER, EPLEY, PARKER, AND NG
participants were asked to vocalize each statement just as intended, that is,
to speak in a sarcastic tone when reading the sarcastic statements and in a
serious tone when reading the nonsarcastic sentences. The other half of
participants were asked to record each statement in the opposite manner,
that is, to say the sarcastic statements seriously and the serious statements
sarcastically. After recording each sentence, participants in both conditions
rated how sarcastic or nonsarcastic the sentence sounded to them on a scale
from 1 (very nonsarcastic)to11(very sarcastic) and then predicted
whether the other participant in the experiment would be able to correctly
identify the statement.
Finally, participants read the statements that had been e-mailed by the
other participant in the experiment. As in Study 2, participants indicated
whether they thought the statement was intended to be sarcastic or non-
sarcastic as well as whether they thought they had guessed correctly or not.
All participants were then thanked, debriefed, and dismissed.
Results and Discussion
Our primary prediction was that asking participants to read their
messages aloud in a manner inconsistent with their intended mean-
ing would change their phenomenological experience of the stim-
uli—that is, the sarcastic statements would no longer “sound” quite
so sarcastic, nor the serious statements quite so serious—and as a
consequence, would reduce overconfidence. Consistent with the
first part of this prediction, participants in the opposite-
phenomenology condition indicated that the sarcastic statements
sounded less sarcastic (M ⫽ 4.31) than those in the same-
phenomenology condition (M ⫽ 7.79), t(26) ⫽ 7.78, p ⬍ .001, d ⫽
1.50. Similarly, the serious statements sounded less serious to
participants in the opposite-phenomenology condition (M ⫽ 7.04)
than they did to participants in the same-phenomenology condition
(M ⫽ 3.27), t(26) ⫽ 8.77, p ⬍ .001, d ⫽ 1.69. Because participants
were assigned to conditions after selecting their statements, these
differences suggest that the manipulation was successful in chang-
ing senders’ own subjective interpretation of the stimuli.
To investigate whether these differences influenced overconfi-
dence, predicted and actual communication were compared in a 2
(phenomenology: same vs. opposite) ⫻ 2 (accuracy: anticipated
vs. actual) fully within-group ANOVA, with accuracy operation-
alized as the percentage of statements correctly identified as either
sarcastic or nonsarcastic.
This analysis revealed two interesting
effects. First, there was a main effect of accuracy: As in the
previous three studies, anticipated accuracy (M ⫽ 72.3%) ex-
ceeded actual accuracy (M ⫽ 62.8%), F(1, 26) ⫽ 6.13, p ⫽ .020,
⫽ 0.19. Second, and more important, this main effect was
qualified by a significant interaction, F(1, 26) ⫽ 8.44, p ⫽ .007,
⫽ 0.25. As shown in Figure 3
, the phenomenology manipula-
tion completely erased participants’ overconfidence. Whereas par-
ticipants in the same-phenomenology condition overestimated the
number of sentences their partner in the experiment would be able
to correctly identify, t(26) ⫽ 4.89, p ⬍ .001, d ⫽ 0.94, there was
no difference in anticipated and actual communication among
participants in the opposite-phenomenology condition, t ⬍ 1.
As in the previous two studies, participants also were overcon-
fident in their ability to detect sarcasm. On average, participants
believed they correctly identified 88.2% of the statements, com-
pared with the actual accuracy rate of 62.8%, F(1, 26) ⫽ 59.09,
p ⬍ .001,
⫽ 0.70. This difference was uninfluenced by condi
tion, F ⬍ 1, which is hardly surprising given that we manipulated
the sender’s phenomenology, not the receiver’s.
Overall, participants were once again overconfident in their
ability to communicate over e-mail. However, this overconfidence
was reduced—indeed, eliminated—once participants were forced
to adopt a phenomenology that differed from their own. Specifi-
cally, asking participants to vocalize their statements in a manner
inconsistent with their intended meaning altered their own subjec-
tive perceptions of those statements, making the sarcastic state-
ments sounded less sarcastic and the serious statements less seri-
ous. As a result, overconfidence disappeared. These findings
further implicate egocentrism as a source of overconfidence in
Studies 1 through 3 demonstrated that people are overconfident
in their ability to communicate sarcasm, seriousness, anger, and
sadness over e-mail, and Study 4 implicated egocentrism as a
probable source of this overconfidence. Study 5 was intended to
supplement these results by exploring what may be an even more
common variety of miscommunication over e-mail: humor. Infor-
mal observation suggests that attempts at humor are often less
We included nonsarcastic as well as sarcastic statements in our mea
sures of perceived and actual accuracy because we expected our manipu-
lation to change not only the phenomenological experience of the sarcastic
statements, but the phenomenological experience of the nonsarcastic state-
ments as well.
Notice that the videotape manipulation in this experiment is also
relatively immune to an alternative interpretation of Study 4 based on
experimenter demand. It is possible, after all, that deliberately asking
participants to read their statements in a manner different from that orig-
inally intended may have led them to infer that the purpose of the manip-
ulation was to reduce overconfidence and to change their responses in an
effort to be consistent with this hypothesis. Although this and other
demand-related interpretations are perhaps unlikely given that we found
evidence for our proposed mediating variable (i.e., the phenomenological
experience of the stimuli covaried by condition), Study 5 diminishes this
concern even further because the videotape manipulation is a much more
subtle manipulation of participants’ phenomenology and presumably con-
tains no experimental demand.
Figure 3. Anticipated and actual ability to communicate by condition
EGOCENTRISM OVER E-MAIL
successful over e-mail than one would think. A “funny” joke,
“amusing” anecdote, or “hilarious” tale often does not pack the
intended punch over e-mail, likely for many of the same reasons
attempts to convey emotions and tones over e-mail can fail us.
Without the verbal channel, nuances of expression, timing, and
emphasis are lost, and our electronic attempts at humor can fall
The present experiment was designed to test this hypothesis as
well as to provide converging evidence for the egocentrism ac-
count of e-mail overconfidence. Participants e-mailed a series of
humorous “deep thoughts” by Jack Handey (a pseudonym of
comedian Al Franken) to another study participant. These included
observations such as the following:
I guess of all my uncles, I liked Uncle Caveman the best. We called
him Uncle Caveman because he lived in cave, and because sometimes
he’d eat one of us. Later on we found out he was a bear. (Handey,
Half of the participants simply read the jokes and then e-mailed
them, whereas the other half first watched a videotape of the jokes
being read on the TV show Saturday Night Live (SNL). As SNL
fans know, part of what makes these jokes funny (and all jokes, for
that matter) is not merely the observations themselves, but the
nuances of timing and delivery. Thus, we expected participants in
the videotape condition to find the observations funnier than would
participants in the control condition. But more than that, we
expected this difference to translate into participants’ predictions
of how the jokes would be perceived by the other person in the
experiment. Although the actual content of the e-mail would be
constant across conditions, we expected that participants would
have a hard time distinguishing their subjective experience of the
joke from its objective properties available to the person at the
other end of the computer. As a result, participants in the videotape
condition should expect their e-mail recipient to find the “deep
thoughts” funnier than should participants in the control condition.
Thus, whereas Study 4 sought to decrease the rift between percep-
tion and reality, the present study sought to increase it.
Participants. Fifty-eight University of Illinois students participated in
exchange for course credit in an introductory psychology course.
Procedure. Participants in the first phase of the experiment (n ⫽ 29)
were recruited in groups of 4 to 5. On arrival to the lab, participants were
told that the experiment concerned everyday communication and, in par-
ticular, how people communicate humor. They were then given a list of 10
“Deep Thoughts by Jack Handey.”
Next, participants selected the five “deep thoughts” that they thought
were the funniest, which they were told (correctly) would be e-mailed to
another participant later in the experiment. Participants in the videotape
condition (n ⫽ 19) next watched a video compilation of all the “deep
thoughts” as being performed on SNL, whereas participants in the control
condition (n ⫽ 20) did not. All participants then rated how funny they
thought each “deep thought” was as well as how funny they thought the
person on the other end of the e-mail would think it was. Each rating was
made on a separate 1 (not at all funny)to11(very funny) scale. The order
in which participants provided their own humor rating and their anticipated
humor rating was counterbalanced.
In the second phase of the experiment, each participant’s selections were
e-mailed to a yoked participant (n ⫽ 29) who was told to simply read and
rate the humor of the five “deep thoughts” using the same scale as above.
The order in which Phase 1 participants made their ratings did
not influence the results and is not discussed further.
Our primary predictions were that participants in the SNL con-
dition would find the “deep thoughts” more humorous than those
in the control condition, and that this difference would translate
into participants’ predictions about how the jokes would be eval-
uated by the person at the other end of the computer. As shown in
Figure 4, our predictions were confirmed. Participants in the
videotape condition thought that the “deep thoughts” were funnier
than did participants in the control condition, t(27) ⫽ 7.82, p ⫽
.009, d ⫽ 0.98, and the same was true of participants’ predictions
of the recipients’ evaluation of the jokes, t(27) ⫽ 7.97, p ⫽ .009,
d ⫽ 0.99. Also as shown in Figure 4, this was true despite
equivalent actual audience humor ratings in the two conditions, t ⬍
1. As a result, participants overestimated how funny their e-mail
recipient would find the jokes, F(1, 27) ⫽ 19.55, p ⬍ .001,
0.42. However, a 2 (condition: videotape vs. control) ⫻ 2 (audi-
ence humor rating: anticipated vs. actual) ANOVA revealed a
significant interaction, F(1, 27) ⫽ 3.58, p ⫽ .069,
indicating that this was especially true among participants in the
These findings further implicate the role of egocentrism in
e-mail (mis)communication. As in Study 4, participants’ assess-
ments of the stimuli available to their e-mail recipient were influ-
enced by their own phenomenological experience of the stimuli.
As a result, participants overestimated the extent to which humor
would be conveyed, but this was particularly true when partici-
pants’ own phenomenological experience of the stimuli was espe-
cially rich. These results are analogous with those of Vorauer and
Figure 4. Own, predicted audience, and actual audience humor ratings by
condition (Study 5). Humor ratings were on a scale ranging from 1 (not at
all funny)to11(very funny).
KRUGER, EPLEY, PARKER, AND NG
Ross (1999), who found that people overestimated the extent to
which their personality traits were transparent to an interaction
partner, especially when self-knowledge was salient to them.
Note, however, that participants were hardly unaware that their
audience’s perspective would differ from their own. As Figure 4
shows, participants recognized that others might not find the jokes
quite as funny as they did (the difference between the dashed and
solid-black lines). However, as Figure 4 also shows, this difference
was not only underestimated, but was completely insensitive to the
actual difference in humor. As can be seen, the difference between
e-mailers’ own impression of the stimuli and their predictions of
the e-mail recipients’ impression of stimuli was the same regard-
less of the extremity of the participants’ own perspective. As a
result, the greater the difference between e-mailers’ own perspec-
tive of the stimuli and that of their audience, the greater the
miscalibration in e-mailers’ predictions.
If comprehending human communication consisted merely of
translating sentences and syntax into thoughts and ideas, there
would be no room for misunderstanding. But it does not, and so
there is. People convey meaning not only with what they say, but
also with how they say it. Gesture, voice, expression, context—all
are important paralinguistic cues that can disambiguate ambiguous
messages (Archer & Akert, 1977; Argyle, 1970; DePaulo & Fried-
man, 1998). Indeed, it is not uncommon for paralinguistic infor-
mation to more than merely supplement linguistic information, but
to alter it completely. The sarcastic observation that “Blues
Brother, 2000—now that’s a sequel” may imply one thing in the
presence of paralinguistic cues but quite the opposite in the ab-
sence of them.
The research presented here tested the implications of these
observations for the rapidly escalating technology of e-mail, a
communication medium largely lacking in paralinguistic informa-
tion. We predicted that because of this limitation subtle forms of
communication such as sarcasm and humor, would be difficult to
convey. But more than that, we predicted that e-mail communica-
tors would be largely unaware of this limitation. Because partici-
pants knew what they intended to communicate, we expected them
to assume that their audience would as well.
Five studies confirmed these predictions. In each one, partici-
pants overestimated their ability to communicate over e-mail. This
was true regardless of whether participants were trying to com-
municate sarcasm (Studies 1 through 3), humor (Study 5), or some
other emotion or tone (Study 4), and regardless of whether partic-
ipants were free to craft their own communication (Studies 1 and
3) or were constrained by the experimenter (Studies 2, 4, and 5).
Studies 2 through 5 also shed light on one cause of this mis-
calibration. We reasoned that when people try to anticipate the
perspective of their e-mail audience, they focus excessively on
their own phenomenology or experience and insufficiently con-
sider the audience’s perspective. This implies that the greater the
difference between the communicator’s own interpretation of the
stimuli and the stimuli available to the audience, the greater the
miscalibration. Consistent with this account, we found in Studies 2
and 3 that overconfidence was greater when participants commu-
nicated over e-mail than by voice, presumably because they failed
to consider the extent to which their audience’s perspective was
different from their own. As shown in Figures 1 and 2, whereas the
ability to communicate varied considerably depending on whether
participants communicated by e-mail or voice, confidence in that
ability did not. To their credit, participants did not assume that
their audience would identify all of their sarcastic statements and
thus seemed to realize that their audience did not share their own
“privileged” perspective. But nor did they take adequate account of
the gulf between their own perspective and their audience’s.
More direct evidence came from Studies 4 and 5, in which
manipulations of e-mailers’ own phenomenologies produced cor-
responding changes in their estimates of what they had communi-
cated to their audience, with overconfidence reduced when the
e-mailers’ phenomenology was brought into line with their audi-
ence’s (Study 4) and increased when it was made even more
discrepant (Study 5). In other words, whereas Studies 2 and 3 held
constant the phenomenology of the communicator and manipu-
lated the richness of the stimuli available to the audience, Studies
4 and 5 held constant the stimuli available to the audience and
manipulated the phenomenology of the communicator. In each
case, however, overconfidence varied as a function of the differ-
ence between the e-mailer’s impression of the stimuli and the
stimuli available to the e-mail recipient: The bigger the rift be-
tween the two, the greater the overconfidence.
Along the way, we also found that participants overestimated
their ability to interpret e-mail. In Studies 2 through 4, participants
overestimated their ability to discern emotions and tones, and
Studies 2 and 3 revealed that this was particularly true over e-mail.
Although not central to our hypotheses, we cannot help but note
that egocentrism likely plays a role here as well. Once a statement
is interpreted as, say, sarcastic, it may be difficult to “hear” the
statement any other way, leading people to believe they understood
their partner’s communication better than they actually did (cf.
Ross & Ward, 1996).
There are, however, some alternative interpretations of these
results that deserve mention. For example, it is possible (albeit
unlikely) that participants were simply unfamiliar with e-mail and
thus were unaware of its limitations. If so, then the observed
overconfidence may have been a function not of egocentrism, but
of the novelty of e-mail. Although the background of our partic-
ipants limits this possibility (most used e-mail regularly, if not
obsessively, and all had a personal e-mail account), we suspect that
experience with e-mail is an important moderator worth investi-
gating in future research. A reasonable hypothesis is that the more
experience people have communicating over e-mail, the less over-
confident they are likely to be. Whether overconfidence disap-
pears, however, remains to be seen.
Another possibility is that these results were produced by a
misunderstanding of statistical regression. Whenever two variables
are imperfectly correlated, extreme values on one variable are
associated with less-extreme values on the other. Because com-
municators’ own impressions of the stimuli were likely imper-
fectly correlated with their audience’s, and because participants
selected statements that were (to them) particularly funny or sar-
castic, they may have failed to recognize that their audience’s
impressions were likely to be less extreme than their own by
chance alone (cf. Kahneman & Tversky, 1973). Note, however,
EGOCENTRISM OVER E-MAIL
that this possibility cannot account for the interactions observed in
Studies 2 through 5 and thus cannot account for our overall pattern
It is also possible that there were important differences between
the way participants communicated with one another in our studies
and the way people communicate with one another in everyday life
that may have influenced the results. For instance, the voice
conditions in our studies may have differed from everyday voice
communication on factors such as time to plan remarks and level
of distraction, which may have influenced calibration. As well,
note that we explicitly prohibited participants in the e-mail condi-
tions from using emoticons, which is hardly a feature of most
e-mail programs. Although this design feature was necessary to
maximize internal validity, it does raise questions about external
validity. After all, emoticons are specifically designed to reduce
the ambiguity of e-mail communication; thus, prohibiting partici-
pants from using them may have artificially reduced accuracy and
Despite the surface plausibility of this caveat, there are several
reasons to doubt it. First, note that although emoticons can help
disambiguate messages, many emoticons are themselves ambigu-
ous (does that “;-)” after “I’m really looking forward to seeing
you” mean that she is flirting— or kidding?). Second, note that
people typically use disambiguating emoticons only when they
feel that there is an ambiguity that needs disambiguating, an event
that the present research suggests people are likely to
Further reason to doubt this alternative comes from a follow-up
study. Participants e-mailed a series of statements to one another in
a manner similar to that of Study 3. In particular, participants were
given a list of 10 topics (e.g., music, politics, dorm life) and for
each one were asked to construct a sentence that would effectively
communicate a specified emotion or tone (such as sarcasm or
anger). They were told (correctly) that each statement would be
e-mailed to another participant in the experiment, the latter of
whom would then try to guess the intended emotion or tone (from
a provided list of 20). We predicted, as in the previous studies, that
senders would overestimate the success rate of receivers. Unlike
the previous studies, however, we experimentally manipulated
whether participants were able to use emoticons. We found that
overconfidence did not differ between those who were using
emoticons and those who were not. Thus, it does not appear that
our results can be attributed to the restriction against the use of
Nor can the results be attributed to the lack of acquaintanceship
of our participants. We explicitly examined this issue in Study 3
and found that overconfidence was independent of whether par-
ticipants were acquainted with their communication partner. At
first blush, this would seem counterintuitive. After all, it stands to
reason that people are generally better at communicating with
people they know than with people they do not, which, all else
equal, should be associated with decreased, perhaps even elimi-
nated, overestimation of that accuracy. But all, we suspect, is not
equal. First, note that even if familiarity was associated with
improved communication, it would also be likely to be associated
with improved confidence, which might result in increased, rather
than decreased, overconfidence (cf. Gill, Swann, & Silvera, 1998;
Swann & Gill, 1997; Van Boven et al., 2000). Second, it is not
clear that acquaintanceship ought to enhance accuracy to begin
with: Prosody is shared, and it is unclear whether familiarity with
one’s speaker or addressee would convey much additional benefit.
Indeed, in Study 3, we found no evidence of increased communi-
cation effectiveness among acquaintances.
Although our focus has been on e-mail miscalibration, we
believe that the overconfidence we have documented here likely
characterizes a wide range of rapidly emerging media types. Chat
rooms, instant messaging, text-based gaming environments—all
have been touted for their superiority to asynchronous text media
such as e-mail because of the dynamic nature of the discourse and
ability to provide rapid feedback. But because these synchronous
media are largely text-based, there likely remains a rift between the
subjective stimuli available to the communicator and the objective
stimuli available to the audience that communicators may fail to
fully appreciate. In fact, we suspect the synchronous and rapid
nature of these mediums may actually increase the rift between
senders and receivers. Compared with synchronous media, asyn-
chronous text media such as e-mail more readily allow for reflec-
tion and reconsideration of one’s communication before transmis-
sion. As we saw in Study 4, such a moment of reconsideration may
allow communicators to recognize the ambiguity inherent in their
messages and thereby increase communication calibration. Of
course, future research is necessary in order to test these
To be sure, we do not mean to suggest that egocentrism in
communication is always undesirable. As others have pointed out,
using one’s own perspective as an indicator of another’s is a
generally valid, useful heuristic— one whose absence would ren-
der effective communication overwhelmingly difficult (Davidson,
1982; Hoch, 1987; Nickerson, 2001). But in a world with as many
diverging perspectives as our own, the downside to egocentrism
seems quite clear as well. To the extent that successful communi-
cation depends on an accurate assessment of one’s clarity (Keysar
& Henley, 2002), overconfidence of that clarity reduces the quality
of communication. Specifically, overestimating the obviousness of
one’s intentions can lead to insufficient allowances for ambiguities
in communication—with occasionally destructive results (Kruger,
Gordon, & Kuban, in press).
Consider, for example, the phenomenon known in electronic
bulletin boards and e-mail groups as “flaming”; the cascade of
openly hostile e-mail that follows a violation of electronic eti-
quette, or “netiquette” (Sproull, Kiesler, & Zubrow, 1984). Al-
though this phenomenon was originally attributed to the disinhi-
bition and deindividuation afforded by the anonymity of the
Internet (Kiesler et al., 1984; Siegel, Dubrovsky, Kiesler, &
McGuire, 1986; Sproull & Kiesler, 1991), evidence for this inter-
pretation is at best mixed (Lea, O’Shea, Fung, & Spears, 1992; Lea
& Spears, 1991). Instead, we suspect that the phenomenon can be
traced, at least in part, to egocentrism. To the extent that people
overestimate the obviousness of the fact that they are “just kid-
ding” when they poke fun or criticize, they may unwittingly
Abrahams, R. D. (1962). Playing the dozens. Journal of American Folk-
lore, 75, 209 –220.
KRUGER, EPLEY, PARKER, AND NG
Agnes, M. (Ed.). (1999). Webster’s new world dictionary (4th ed.). New
Allbritton, D. W., McKoon, G., & Ratcliff, R. (1996). Reliability of
prosodic cues for resolving syntactic ambiguity. Journal of Experimental
Psychology, 22, 714 –735.
Archer, D., & Akert, R. M. (1977). Words and everything else: Verbal and
nonverbal cues in social interpretation. Journal of Personality and Social
Psychology, 35, 443– 449.
Argyle, M. (1970). Social interaction. New York: Atherton Press.
Camerer, C., Loewenstein, G., & Weber, M. (1989). The curse of knowl-
edge in economic settings: An experimental analysis. Journal of Polit-
ical Economy, 97, 1232–1254.
Clark, H. H. (1996). Using language. Cambridge, England: Cambridge
Daft, R. L., & Lengel, R. H. (1986). Organizational information require-
ments, media richness, and structural design. Management Science, 32,
Daft, R. L., & Trevino, L. (1987). Message equivocality, media selection,
and manager performance. MIS Quarterly, 11, 355–366.
Davidson, D. (1982). Paradoxes of irrationality. In R. Wollheim & J.
Hopkins (Eds.), Philosophical essays on Freud (pp. 289–305). Cam-
bridge, England: Cambridge University Press.
de la Sola Pool, I. (1984). Communication flows: A census in the United
States and Japan. Amsterdam: University of Tokyo Press.
Dennis, A. R., & Kinney, S. T. (1998). Testing media richness theory in the
new media: The effects of cues, feedback, and task equivocality. Infor-
mation Systems Research, 9, 256–274.
DePaulo, B. M. (1992). Nonverbal behavior and self-presentation. Psycho-
logical Bulletin, 111, 203–243.
DePaulo, B. M., & Friedman, H. S. (1998). Nonverbal communication. In
D. T. Gilbert, S. T. Fiske, & G. Lindsey (Eds.), The handbook of social
psychology: Vol. 2 (4th ed., pp. 3– 40). New York: McGraw-Hill.
Dimmick, J., Kline, S., & Stafford, L. (2000). The gratification niches of
personal e-mail and the telephone: Competition, displacement, and
complementarily. Communication Research, 27, 227–248.
Drew, P. (1987). Po-faced receipts of teases. Linguistics, 25, 219 –253.
Dunning, D., Griffin, D. W., Milojkovic, J. D., & Ross, L. (1990). The
overconfidence effect in social prediction. Journal of Personality and
Social Psychology, 58, 568 –581.
Epley, N., Keysar, B., Van Boven, L., & Gilovich, T. (2004). Perspective
taking as egocentric anchoring and adjustment. Journal of Personality
and Social Psychology, 87, 327–339.
Fischhoff, B. (1975). Hindsight ⫽ ⫽ foresight: The effects of outcome
knowledge on judgment under uncertainty. Journal of Experimental
Psychology: Human Perception and Performance, 1, 288 –299.
Flavell, J. (1977). Cognitive development. Englewood Cliffs, NJ: Prentice
Fussell, S. R., & Krauss, R. M. (1991). Accuracy and bias in estimates of
others’ knowledge. European Journal of Social Psychology, 21, 445–
Gatz, L. B., & Hirt, J. B. (2000). Academic and social integration in
cyberspace: Students and e-mail. Review of Higher Education, 23,
Gill, M. J., Swann, W. B., & Silvera, D. H. (1998). On the genesis of
confidence. Journal of Personality and Social Psychology, 75, 1101–
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
transparency: Biased assessments of others’ ability to read one’s emo-
tional states. Journal of Personality and Social Psychology, 75, 332–
Goffman, E. (1959). The presentation of self in everyday life. Garden City,
Handey, J. (1992). Deep thoughts. New York: Penguin.
Hebert, B. G., & Vorauer, J. D. (2003). Seeing through the screen: Is
evaluative feedback communicated more effectively in face-to-face or
computer-mediated exchanges? Computers in Human Behavior, 19,
Hoch, S. (1987). Perceived consensus and predictive accuracy: The pros
and cons of projection. Journal of Personality and Social Psychology,
Inhelder, B., & Piaget, J. (1958). The growth of logical thinking from
childhood to adolescence: An essay on the construction of formal
operational structures. New York: Basic Books.
Kahneman, D., & Tversky, A. (1973). On the psychology of prediction.
Psychological Review, 80, 237–251.
Kelley, C. M., & Jacoby, L. L. (1996). Adult egocentrism: Subjective
experience versus analytic bases for judgment. Journal of Memory and
Language, 35, 157–175.
Keren, G. (1987). Facing uncertainty in the game of bridge: A calibration
study. Organizational Behavior and Human Decision Processes, 39,
Keysar, B. (1994). The illusory transparency of intention: Linguistic per-
spective taking in text. Cognitive Psychology, 26, 165–208.
Keysar, B., Barr, D. J., & Horton, W. S. (1998). The egocentric basis of
language use: Insights from a processing approach. Current Directions
in Psychological Science, 7, 46–50.
Keysar, B., & Bly, B. (1995). Intuitions of the transparency of idioms: Can
one keep a secret by spilling the beans? Journal of Memory and Lan-
guage, 34, 89 –109.
Keysar, B., & Henly, A. S. (2002). Speakers’ overestimation of their
effectiveness. Psychological Science, 13, 207–212.
Kiesler, S., Siegel, J., & McGuire, T. W. (1984). Social psychological
aspects of computer-mediated communication. American Psychologist,
Krauss, R. M., et al. (1981). Verbal, vocal, and visible factors in judgments
of another’s affect. Journal of Personality and Social Psychology, 40,
Kruger, J., Gordon, C., & Kuban, J. (in press). Intentions in teasing: When
“just kidding” just isn’t good enough. Journal of Personality and Social
Lanzetta, J. T., & Kleck, R. E. (1970). Encoding and decoding of nonverbal
affect in humans. Journal of Personality and Social Psychology, 16,
Lea, M., O’Shea, T., Fung, P., & Spears, R. (1992). “Flaming” in
computer-mediated communication: Observations, explanations, impli-
cations. In M. Lea (Ed.), Contexts of computer-mediated communication
(pp. 89 –112). New York: Harvester Wheatsheaf.
Lea, M., & Spears, R. (1991). Computer-mediated communication, de-
individuation and group decision-making. International Journal of Man-
Machine Studies, 34, 283–301.
Lichtenstein, S., Fischhoff, B., & Phillips, L. D. (1982). Calibration of
probabilities: The state of the art to 1980. In D. Kahneman, P. Slovic, &
A. Tversky (Eds.), Judgment under uncertainty: Heuristics and biases
(pp. 306 –334). Cambridge, England: Cambridge University Press.
Marold, K. A., & Larson, G. (1999). Is the range war over? An investiga-
tion into preferences for e-mail and v-mail. Social Science Computer
Review, 17, 466 – 471.
Newton, L. (1990). Overconfidence in the communication of intent: Heard
and unheard melodies. Unpublished doctoral dissertation, Stanford Uni-
versity, Stanford, CA.
Nickerson, R. (1999). How we know—and sometimes misjudge—what
others know: Imputing one’s own knowledge to others. Psychological
Bulletin, 125, 737–759.
Nickerson, R. S. (2001). The projective way of knowing: A useful heuristic
EGOCENTRISM OVER E-MAIL
that sometimes misleads. Current Directions in Psychological Science,
10, 168 –172.
Oskamp, S. (1965). Overconfidence in case-study judgments. Journal of
Consulting Psychology, 29, 261–265.
Panko, R., & Kinney, S. T. (1995). Meeting profiles: Size, duration, and
location. Proceedings of the Twenty-Eighth Annual Hawaii Interna-
tional Conference on System Sciences, Vol. 4, 1002–1011.
Price, P. J., Ostendorf, M., Shattuck-Hufnagel, S., & Fung, C. (1991). The
use of prosody in syntactic disambiguation. Journal of the Acoustical
Society of America, 90, 2956–2970.
Rime, B. (1982). The elimination of visible behaviour from social inter-
actions: Effects on verbal, nonverbal and interpersonal variables. Euro-
pean Journal of Social Psychology, 12, 113–129.
Ross, L., & Ward, A. (1996). Naive realism in everyday life: Implications
for social conflict and misunderstanding. In E. Reed, E. Turiel, & T.
Brown (Eds.), Social cognition: The Ontario Symposium (pp. 305–321).
Hillsdale, NJ: Erlbaum.
Sherman, R. C., End, C., Kraan, E., Cole, A., Campbell, J., Klausner, J., &
Birchmeier, Z. (2001). Metaperception in cyberspace. Cyberpsychology
and Behavior, 4, 123–129.
Siegel, J., Dubrovsky, V., Kiesler, S., & McGuire, T. W. (1986). Group
processes in computer-mediated communication. Organizational Behav-
ior and Human Decision Processes, 37, 157–187.
Sproull, L., & Kiesler, S. (1986). Reducing social context cues: Electronic
mail in organizational communication. Management Science, 32, 1492–
Sproull, L., & Kiesler, S. (1991). Connections: New ways of working in the
networked organization. Cambridge, MA: MIT Press.
Sproull, L., Kiesler, S., & Zubrow, D. (1984). Encountering an alien
culture. Journal of Social Issues, 40, 31– 48.
Swann, W. B., Jr., & Gill, M. J. (1997). Confidence and accuracy in person
perception: Do we know what we think we know about our relationship
partners? Journal of Personality and Social Psychology, 73, 747–757.
Thompson, L., & Nadler, J. (2002). Negotiating via information technol-
ogy: Theory and application. Journal of Social Issues, 58, 109 –124.
Thompson, L. L. (2001). The mind and heart of the negotiator (2nd ed.).
Upper Saddle River, NJ: Prentice Hall.
Vallone, R. P., Griffin, D. W., Lin, S., & Ross, L. (1990). Overconfident
predictions of future actions and outcomes by self and others. Journal of
Personality and Social Psychology, 58, 582–592.
Van Boven, L., Kruger, J., Savitsky, K., & Gilovich, T. (2000). When
social worlds collide: Overconfidence in the multiple audience problem.
Personality and Social Psychology Bulletin, 26, 619 –628.
Vorauer, J. D., & Claude, S. D. (1998). Perceived versus actual transpar-
ency of goals in negotiation. Personality and Social Psychology Bulletin,
Vorauer, J. D., & Ross, M. (1999). Self-awareness and feeling transparent:
Failing to suppress one’s self. Journal of Experimental Social Psychol-
ogy, 35, 415– 440.
Walther, J. B. (1993). Impression development in computer-mediated
interaction. Western Journal of Communication, 57, 381–398.
Walther, J. B. (1995). Relational aspects of computer-mediated communi-
cation: Experimental observations over time. Organization Science, 6,
Walther, J. B., Loh, T., & Granka, L. (2005). Let me count the ways: The
interchange of verbal and nonverbal cues in computer-mediated and
face-to-face affinity. Journal of Language and Social Psychology, 24,
Wittaker, S., & Sidner, C. (1997). E-mail overload: Exploring personal
information management of e-mail. In S. Kiesler (Ed.), Culture of the
Internet (pp. 277–295). Mahwah, NJ: Erlbaum.
Wright, G., Rowe, G., Boger, F., & Gammack, J. (1994). Coherence,
calibration, and expertise in judgmental probability forecasting. Orga-
nizational Behavior and Human Decision Processes, 57, 1–25.
Received March 24, 2004
Revision received August 18, 2004
Accepted August 29, 2004 䡲
KRUGER, EPLEY, PARKER, AND NG