ArticlePDF Available

Social Comparison in Everyday Life

Authors:

Abstract

94 college students recorded details of their social comparisons over 2 wks using a new instrument, the Rochester Social Comparison Record. Major results were (1) comparison direction varied with relationship with the target; (2) precomparison negative mood led more often to upward comparison than to downward comparison, supporting a selective affect–cognition priming model in which dysphoria primes negative thoughts about the self (G. H. Bower, 1991; J. P. Forgas et al, 1990) rather than a motivational self-enhancement model (T. A. Wills, 1981, 1991); (3) upward comparison decreased subjective well-being, whereas downward comparison increased it; and (4) high self-esteem individuals engaged in more self-enhancing comparison. (PsycINFO Database Record (c) 2012 APA, all rights reserved)
INTERPERSONAL RELATIONS AND GROUP
PROCESSES
Social Comparison in Everyday Life
Ladd Wheeler and Kunitate Miyake
University of Rochester
Ninety-four college students recorded details of their social comparisons over
2
weeks using a new
instrument, the Rochester Social Comparison Record. Major results
were
(a) comparison direction
varied with relationship with the target; (b) precomparison negative mood led more often to
upward comparison than to downward comparison, supporting a selective affect-cognition prim-
ing model in which dysphoria primes negative thoughts about the self (Bower,
1991;
Forgas,
Bower,
&
Moylan,
1990)
rather than a motivational self-enhancement model
(Wills,
1981,1991);
(c)
upward
comparison decreased subjective well-being, whereas downward comparison increased it; and (d)
high self-esteem individuals engaged in more self-enhancing comparison.
Festinger's theory of social comparison processes (Festinger,
1954)
continues to
be
an active arena for theory and research. A
new edited book (Suls
&
Wills,
1991), a symposium at the 1990
meeting of the Society for Experimental Social Psychology
(SESP), and frequent journal articles attest to this vigor. Yet
underlying the vitality
is
a feeling of uneasiness, obvious in the
comments occasioned by the SESP symposium, in which a
group of extraordinarily knowledgeable participants showed
little agreement about such apparently basic questions as "Do
people compare at all (or very much)? When do people com-
pare?
How
do people balance upward and downward compari-
sons? How much does similarity count in comparison? Do peo-
ple compare with actual targets, or are all comparisons con-
structed in people's heads? The problem is that there are many
measures of social comparison, and they do not agree well with
one
another,
leading
to
theoretical proliferation lacking a coher-
ent empirical base. These measures may not agree with one
another because of difficulties with the measures themselves
(Wood, 1991), because they measure different motives for so-
cial
comparison (cf.
Wood
& Taylor,
1991),
or because they have
been used in different
contexts.
The one thing on which there is
general agreement is that social comparison is a wonderfully
flexible process that can best be studied under naturalistic
Portions of the article were presented at the meeting of the Ameri-
can Psychological Society, Dallas, Texas, June 1990. Preparation of
this article was supported by the University of Rochester.
We
are very grateful to Miron Zuckerman for his support and sug-
gestions throughout this project and to Rachel Lewis, Carol Cutalo,
and Sharon Lazarow for their assistance in data
collection.
John Arro-
wood, Emory Cowen, Larry Gruder, Richard Koestner, and Joanne
Wood commented on earlier drafts. Gordon Bower provided a crucial
interpretation.
Correspondence concerning this article should be addressed to
Ladd Wheeler, Department of Psychology, Meliora
Hall,
University of
Rochester, Rochester, New \brk
14627.
Electronic mail may be sent to
wheelei@psych.rochester.edu.
conditions lacking constraints imposed by the experimenter
(Ross,
Eyman, & Kishchuk, 1986; Ruble & Frey,
1991;
Suls &
Wills,
1991;
Wills,
1991;
Wood,
1991).
Suls
and
Wills
noted that
existing methods for studying naturalistic social comparison
can obtain only retrospective accounts and that it would be an
important step to use some sort of experience sampling tech-
nique to measure social comparison as it
occurs.
The purpose
of this article is to describe such a method and to present the
first data produced by using it.
There are three basic methods for the self-recording of daily
life events (Wheeler & Reis, 1991):
1.
Interval-contingent self-recording, in which respondents
report on their experiences at some regular, predetermined in-
terval. The reports may refer to what has happened since the
previous interval or to what one
is
experiencing
at
the moment.
Much of the stress research uses this technique (cf. Delongis,
Folkman, & Lazarus, 1988).
2.
Signal-contingent self-recording, in which respondents
report on their experiences whenever signaled by the re-
searcher. These are often referred to as
beeper studies
(cf. Csik-
szentmihalyi, Larson, & Prescott, 1977; Hurlburt, 1979;
Klinger, 1978).
3.
Event-contingent self-recording, in which respondents re-
port on a denned class of events whenever one of the events
occurs. Examples are the Rochester Interaction Record (cf.
Wheeler
&
Nezlek, 1977; Wheeler, Reis, & Bond,
1989;
Reis
&
Wheeler, 1991) and the Iowa Communication Record (Duck,
Rutt, Hurst,
&
Strejc, 1991).
The most likely technique for a representative and accurate
study of
social
comparison
is the
event-contingent method. The
interval-contingent method would
be
retrospective, and the
sig-
nal-contingent method would be likely to miss many of the
comparisons. For example, one subject in a beeper study said
that he thought about sex 40% of
his
waking life, but that the
signal never caught him at it (Hurlburt, 1979). Nevertheless,
signal-contingent self-recording is excellent for certain pur-
poses (Wheeler
&
Reis, 1991).
Journal of Personality and Social Psychology,
1992,
Vol.
62,
No. 5, 760-773
Copyright 1992 by the American Psychological Association, Inc. 0022-3514/92/S3.00
760
SOCIAL COMPARISON761
Thus,
the method used here, which we call the Rochester
Social Comparison Record (RSCR, reproduced in the Appen-
dix),
is
simply to ask respondents to keep a record of all of their
social comparisons
as
they
occur.
Completing the RSCR is con-
tingent on a comparison having taken place. In addition to the
comparison dimension and relationship to the comparison
others, respondents
are
asked to indicate their mood before and
after comparison and their similarity to the comparison other
on the comparison dimension—from inferior through similar
to superior. This method should provide the first response to
Wills's
(1991,
p.
73) call
for "knowledge about
the
typical
mix of
upward and downward comparisons in prevailing naturalistic
conditions."
How does this method articulate with the group of methods
critically discussed by Wood
(1991)?
First, it
is
an information-
seeking1 measure only to the extent that respondents seek com-
parisons; in real life "comparison...
is
often effectively forced
upon the individual
by his social
environment, even if
he
would
prefer
to avoid
comparing"
(Mettee
&
Riskind,
1974,
p.
348;
see
also Wood, 1989). The RSCR does not at this time distinguish
between sought and forced comparisons, although it could be
modified to do so. Second, it is not an affiliation2 measure,
because comparison is not inferred from affiliative choice.
Third, it is an effects-of-comparison measure in the sense that
momentary mood change as a result of comparison is mea-
sured. However, the evidence for
a
comparison does not rest on
these effects. Fourth, it
is
a comparative rating method, because
respondents rate themselves relative to the comparison other.
However, it is different from the typical comparative rating
situation in that respondents
rate
themselves
relative to a
partic-
ular other
with
whom they
have
compared.
Remarking on com-
parative
ratings,
Ross et
al.
(1986,
p.
82)
wrote,
"These compari-
sons are fairly easy to make. But are they the ones people make
in their daily lives?" In the case of the RSCR, the answer is
strongly affirmative. Fifth, the RSCR is not a free-response
method, which measures comparison conclusions that people
express spontaneously in conversation or in interviews. How-
ever, it
has
many of the same advantages in that respondents are
free to compare or not on any dimension with
anyone.
It
has
the
additional important advantage of measuring comparison be-
haviors rather than comparison conclusions.
In summary, the RSCR is primarily a combination of the
information-seeking (or information-exposure) and compara-
tive ratings methods, with additional information from the ef-
fects-of-comparison method. It was designed recognizing that
in real life, comparison selection, comparative judgment, and
reaction to comparison are essentially simultaneous and insepa-
rable.
Because this method has not been used before and because
the attempt
is
to a large extent exploratory, we present a limited
set of hypotheses.
Hypothesis
1.
Similar comparisons3 will be more frequent
with close friends, and dissimilar comparisons (upward and
downward) will be more
frequent
in more distant relationships.
Not only should close friends be objectively more similar, but
one would
also
want
to avoid
dissimilar comparisons
with
close
friends. According to Brickman and Bulman
(1977,
p.
152),
"If
two people compare themselves on a valued dimension, the
chances are that one will be superior and one will be inferior.
Someone will feel bad, and both parties or
the
collectivity must
be concerned with coping with these negative feelings." Al-
though Brickman and Bulman did not make it explicit, close
friends should be particularly concerned with avoiding such
negative feelings. Similarly, according to Tesser
(1988),
upward
comparison with close friends on dimensions relevant to the
self
is
particularly threatening.
Hypothesis
2. Comparison direction will be related to pre-
comparison affect. There are opposing predictions about the
nature of this relationship. Downward comparison theory
(Wills, 1981,1991) predicts that when a person
is
experiencing
low subjective well-being (in the present study, feeling discour-
aged and unhappy), comparison will be downward
(with
some-
one worse off). The individual is motivated to increase his or
her subjective well-being and attempts
to do this by
focusing on
his
or her superiority
to
another.
In
contrast
to
this motivational
theory is a selective affect-cognition priming model in which
dysphoria primes
negative
thoughts about
the
self (Bower,
1981,
1991;
Forgas et al., 1990) that have been associated in the past
with negative feelings
(e.g.,
"Others are better than
I
am").
If one
has the thought of being inferior, one will tend to
notice,
select,
perceive, and interpret others as superior, leading to upward
comparison under conditions of low subjective well-being.
Thus,
both theories predict
a
relationship
between precompar-
ison affect and direction of comparison, but in different direc-
tions.
Hypothesis
3.
Downward comparison will enhance subjec-
tive well-being (Affleck & Tennen, 1991; Crocker & Gallo,
1985;
Gibbons,
1986;
Wills, 1981,1991). This is one of the two
major propositions of downward comparison theory, the other
being that, as stated in Hypothesis
2,
individuals will compare
downward when self-esteem is threatened. However, Wills
(1991,
p.
60) suggested that there are limiting conditions on the
enhancement of well-being through downward
comparison:
"If
the comparison situation conveys information suggesting that
the probability of becoming like the target person
is
high, pre-
sumably there would be negative change
(i.e.,
anxiety would be
increased)." Major, Testa, and Bylsma (1991) argued similarly
that downward comparison will enhance well-being primarily
when the comparison is esteem-relevant and when individuals
perceive their superior standing
as
stable or
as
under their per-
1 The information-seeking measure
is
exemplified by the rank order
paradigm, in which subjects are asked to select
a
person above or below
them in a rank order for purposes of comparing their scores. The
method has typically produced upward comparison. See Gruder
(1977) for a review.
2 The affiliation measure is exemplified by Schachter's (1959) stud-
ies of anxiety and affiliation, and the focus typically has been on the
emotional and situational similarity of the comparison target. See
Wheeler (1974) and Cottrell and Epley (1977) for reviews. See Taylor
and Lobel (1989) for a discussion of more recent work.
3
By
similar
comparisons,
we
mean similarity in performance rather
than similarity on related attributes (Goethals & Darley, 1977). The
RSCR contained a question (not
shown),
"How similar are you to the
person on other dimensions related to the comparison dimension?"
This was an attempt to measure related attributes. Not surprising (in
retrospect), subjects did not understand the question. Perhaps the ef-
fect of related attributes is best studied in the laboratory, or perhaps
there is a less abstruse way to ask about them.
762LADD WHEELER AND KUNITATE MIYAK.E
sonal control; otherwise, downward comparison might make
them feel vulnerable. These interesting theoretical qualifica-
tions probably do not apply to everyday social comparisons in
normal populations (barring strong manipulations), because
people don't feel that bad things will happen to them (Taylor
&
Brown, 1988); hence, we make the directional prediction of
downward comparison leading to more positive affect.
We predict that upward comparison will lead to negative
affective reactions. The literature on relative deprivation
(Crosby, 1976; Martin, 1986), inequity (Walster, Walster, &
Berscheid, 1978), jealousy (Salovey & Rodin, 1984), perfor-
mance comparison (Tesser,
1988),
and satisfaction with life do-
mains (Emmons
&
Diener, 1985) supports such a prediction.
Hypothesis
4.
Self-esteem will be related to the direction of
comparison. Once again there are opposing predictions about
the nature of the relationship. Downward comparison theory
predicts that individuals with low self-esteem, particularly
when self-esteem
is
threatened,
will
compare downward. These
are the people in the greatest need for self-enhancement.
Usually cited
as
supporting evidence are studies using the rank
order paradigm (Friend & Gilbert,
1973;
Smith & Insko, 1987;
Wilson &Benner, 1971).
On the other hand, there is a body of evidence using the
comparative rating method that high self-esteem individuals
rate themselves higher than others to a greater extent than do
low self-esteem individuals (Brown, 1986; Campbell, 1986;
Crocker
&
Schwartz, 1985) and do so more when self-esteem is
threatened (Crocker, Thompson, McGraw,
&
Ingerman, 1987).
Both Wood and Taylor (1991) and Wills (1991) explained the
discrepancy between the two sets of studies by noting that al-
though it is difficult for
low
self-esteem people to claim superi-
ority on a comparative rating (because they do not perceive
themselves as superior), it is less direct and therefore possible
for them to select downward comparison
targets.
As
the RSCR
has aspects of both the information-seeking and comparative
rating methods, we do not make a directional prediction.
Self-esteem can also be related to the responses to compari-
son, as discussed in Hypothesis 3. Downward comparison
theory (Wills, 1991) would argue that low self-esteem individ-
uals will respond more positively than high self-esteem individ-
uals to downward comparison, as they are in the most need of
psychic reassurance. On the other hand, Taylor and Brown
(1988)
concluded that high self-esteem individuals are the most
likely to engage in self-enhancing illusions, such
as
minimizing
negative feedback and maximizing positive feedback.
Method
Subjects
Subjects were 39 men and 55 women enrolled in the University of
Rochester. Eighty-four percent were
first-semester
freshmen. They par-
ticipated in this study to fulfill part of a course requirement.
Materials
The RSCR, reproduced in the Appendix, includes information
about the circumstances of the comparison (social interaction, visual,
etc.),
the dimension of the comparison (academics, personality, etc.),
the type of relationship to the comparison other(close friend, ordinary
friend, etc.), the comparison similarity to the other (from inferior to
superior), affective ratings
(pre-
and postcomparison ratings on affect),
and an open-ended description of the comparison. Subjects were in-
structed to
fill
out the RSCR for each social comparison they made for
2 weeks.
Procedure
At
small group organizational meetings, concrete examples of social
comparisons were provided and discussed. It was suggested that
merely noticing a similarity with or a difference from another person
would not necessarily be called a comparison unless accompanied by
some psychological reactions.
Subjects were asked to keep records of social comparisons they
made in daily life for 2 weeks. It was emphasized that it was better to
record right after a comparison was
made.
If, for some reason, compari-
sons
could not be recorded immediately, subjects
were
instructed to fill
out the social comparison record at a regular time each
day,
such as just
before
bedtime.
An
abbreviated scratch sheet
was
provided
to
facilitate
memory. Subjects
were
assured that on one
day
they might have a lot of
comparisons but on another day they might not, and there might be
individual differences in the number of social comparisons people
make. To eliminate misunderstanding, mistakes, ambiguities, and so
forth, subjects were asked to turn in their records every day at the
beginning of the
2-week
period but were allowed to stretch this
to 2
or
3
days as the
2-week
period progressed. This procedure was used to
promote regularity of record keeping and also to distribute new blank
forms as subjects needed them. After the
2-week
record keeping pe-
riod, a postexperimental questionnaire was administered to examine
any potential sources of inaccuracy in the comparison records. It was
emphasized that the promised participation credit would be given
regardless of the answers. Subjects were told that they were coinvesti-
gators rather than typical subjects in a psychological experiment and
were treated accordingly throughout the study.
Personality Scales
During the self-observation period, subjects completed several per-
sonality measures. The only one we are concerned with here, because
of its theoretical importance in the social comparison literature,
is
the
Rosenberg Self-Esteem Scale (Rosenberg, 1965).
Results
Postexperimental Questionnaire
Overall ratings on the postexperimental questionnaire were
as follows:
1.
How difficult was it to record comparisons
(1
=
very
diffi-
cult,
4 =
neutral,
and 7 = not at
all
difficult)!
Overall = 4.10.
2.
How accurate do you perceive your comparison records
were
(1
=
very
inaccurate,
4 =
neutral,
and 7 =
very
accurate)!
Overall = 4.69.
3.
How much did keeping the Comparison Record interfere
with your daily life
(1
=
very
little,
4 =
neutral,
and 7 = very
much)?
Overall = 3.78.
4.
This study made me more sensitive to comparisons so that
I changed the number of comparisons I usually make
(1
= very
little, 4 =
neutral,
and 7 = very much). Overall = 3.97.
5.
As for other factors which contribute to inaccuracy (1 =
very little, 4 =
neutral,
and 7 = very
much):
(a) forgetting, over-
all = 4.39; (b) difficulty of being attentive, overall = 4.67; (c)
lack of clarity in instructions, overall
=
2.29;
(d) I
felt an obliga-
SOCIAL COMPARISON763
tion to make comparisons, overall =
3.83;
(e)
I hesitate to report
comparisons due to guilt feelings, overall = 2.67; and (f) com-
parison was too subtle to be noticed, overall = 3.83.
Reliability
The number of social comparisons for the
first
5 days
and the
last
5
days correlated
.74.
This method simulates test-retest or
split-half reliability.
Method of Analysis
The unit of analysis throughout this study
was
the individual
subject who reported multiple social comparisons. Conse-
quently, all statistics were computed for each subject and then
were used for analysis. When degrees of freedom seem too
large, it is because of the repeated measures, not because we
have departed from using the individual as the unit of analysis.
Descriptive Statistics
Table
1
shows percentage of comparisons
by
target
sex,
situa-
tion, dimension, relationship, and direction. Percentages are
presented for men and women, separately and combined.
Male and female subjects showed
a
strikingly similar pattern
for most of the
categories.
Only a few
of the
sex
differences were
significant, and these were in relatively infrequent categories.
Men were more likely to make comparisons while daydream-
ing, t(92) = 2.18, p < .05. Women were more likely to make
comparisons on eating
habits,
t(92) =
2.08,
p
<
.05,
and appear-
ance,
t(92)
= 1.88, p < .10, whereas men were more likely to
make opinion comparisons, t(92) = 2.47, p < .05. Men were
more likely to compare with famous people,
t(92) =
2.51,
p <
.05,
whereas women were more likely to compare with family
members,
t(92)
= 2.39, p
<
.05.
Factor Analysis
of Dimensions
To simplify the data, we performed a principal-components
factor analysis using varimax rotation on the number of com-
parisons on all the dimensions containing at least 5% of the
comparisons,
thus
excluding
wealth,
opinions,
and "other"
(e.g.,
relationships and eating habits).4 A two-factor structure ap-
peared. The first factor contained academic matters, lifestyle,
and personality (eigenvalue
=
1.9; factor loadings
=
.59,
.87,
and
.76,
respectively). The second factor contained abilities, social
skills,
and appearance (eigenvalue = 1.2; factor loadings = .63,
.71,
and .69, respectively). Cross-factor loadings ranged from
.00 to
.23,
averaging
.07.
With
some
trepidation,
as
factor names
tend to take on a life of their own, we named the first factor
Lifestyle
and the second factor, Assets. The Lifestyle factor per-
haps represents what college students are—their lifestyle is re-
lated to their personality and determines the academic ap-
proach and outcome, whether they are serious students,
leaders, athletes, partiers, or some combination of these things
and
others.
The Assets factor perhaps represents the raw mate-
rials that determine how gracefully and powerfully one can
carry out a lifestyle. We are better able to characterize these
factors as we describe how they differ in the comparison pro-
cess.
Number of Comparisons
We performed an analysis of variance (ANOV\) on number
of same-sex comparisons per
10
days5 with the following vari-
ables:
dimension—Lifestyle and Assets; relationship—close
friend, ordinary friend, acquaintance, and stranger; direction
—upward, same level, downward; and sex (the first three vari-
ables were repeated measures). Upward comparisons were
those on which the subject indicated that
he
or
she was
inferior,
poor, or undesirable relative to the comparison other
(-1
to -3
on Item
5
of the
RSCR);
downward comparisons
were
those on
which the subject indicated superiority
to the
comparison other
(1 to 3). Target sex was included in a preliminary analysis and
merely muddled the
results,
which
was
not surprising in
view of
the great predominance of same-sex comparisons. There were
too few opposite-sex comparisons to be meaningfully broken
down across the other variables. That comparisons are largely
with same-sex others
has
been shown
with
other methods (Feld-
man & Ruble,
1981;
Major & Forcey,
1985;
Suls, Gaes, & Gas-
torf,
1979; Zanna, Goethals, & Hill, 1975) and serves as a par-
tial validity check of the RSCR. There were no significant ef-
fects for sex of subject. The Dimension X Relationship X
Direction means for same-sex comparisons are shown in Ta-
ble 2.
The main effect for dimension was that there were more
Lifestyle than Assets comparisons, F(l, 92) =
4.1,
p
<
.05.
The
main effect of relationship was that there were more compari-
sons with close friends than with ordinary friends and more
comparisons with ordinary friends than with acquaintances or
strangers, F(3,276) =
18.1,
p
<
.001.
(The linear contrast from
close friend through stranger produced F[l, 276] = 50.97, p <
.001.) The main effect of direction was that there were more
downward than same-level or upward comparisons, F(2,184)
=
10.9,
p<.
001.
Hypothesis
1
was that similar comparisons would be more
frequent with close friends, predicting the Direction X Rela-
tionship interaction, F(6,552)
=
5.2, p
<
.001,
shown in Figure
1.6
The close friend versus all others contrast of same versus
4 If a subject checked more than one comparison dimension, one of
us consulted the open-ended description at the end of the RSCR to
determine the most appropriate dimension. For example, if a subject
compared herself to a prettier woman in an expensive dress and
checked both physical appearance and wealth as comparison dimen-
sions,
and then went on to describe her or the other's
figure,
hair, smile,
and so on, it was coded as a physical appearance comparison. As we
stressed using only one category, fewer than 5% of the comparisons
presented this problem.
5 Although subjects were asked to use the RSCR for
14
days, there
was
a minor variation in the number of
days.
Accordingly,
we
equated
subjects on the number of days and also standardized to a 10-day pe-
riod with the following formula:
(no.
of comparisons/no, of
days)
X10.
Thus,
no data were discarded.
6 Interactions are presented as interaction residuals (Rosnow
&
Ro-
senthal, 1989) because they show the interaction effects uncontami-
nated by main effects (and all of our main effects are significant).
Residuals shown are what remains when
row,
column, and grand mean
effects are removed. Thus, a positive residual indicates a relatively
stronger effect than would be expected from the marginals. The raw
means can be obtained from Table
2
for those who want to know what
the interactions look like in combination with the main effects.
764
LADD WHEELER AND KUNITATE MIYAKE
upward and downward comparisons produced F(l, 552) =
16.61,
p < .001, supporting Hypothesis 1. However, the great
frequency of downward comparisons with ordinary friends was
completely unexpected. The logic of Hypothesis
1
would not
predict any abrupt difference between ordinary friends and
others, yet the ordinary friend versus all others contrast of
downward comparison versus same and upward7 yielded F(\,
552) =12.74,
p<.001.
These results were qualified by the three-way interaction,
Direction
X
Relationship
X
Dimension (not
shown),
F(6,552) =
3.9, p <
.001.
We examine the two strong contrasts, described
in the previous paragraph, as a function of the dimension. The
close friend versus all others contrast of same-level versus up-
Table
1
Percentage
of
Social Comparisons
for
Men,
Women,
and
Men
and
Women Combined
Table 2
Number of
Comparisons
by
Dimension,
Relationship,
and
Direction
VariableMen Women
(n
=
39) (n
= 55)
Combined
(N
=
94)
Target sex
Same sex
Opposite sex
Mixed sex
Unknown
Situation
Social interaction
Visual
Daythought
Brief contact
Telephone
Dimension
Academic
Personality
Physical appearance
Lifestyle
Abilities
Social skills
Wealth
Opinion
Other
Relationships
Eating habits
Uncategorized
Relationship
Close friend
Ordinary friend
Acquaintance
Stranger
Family member
Imaginary person
Boyfriend or girlfriend
Famous person
Other
Similarity level
1-target
superior
2
3
4-target same
5
6
7-target
inferior
71.08
22.50
4.10
2.32
45.83
20.79
17.31
8.89
7.18
24.99
13.56
11.68
14.03
13.95
7.74
4.70
3.88
0.81
0.00
4.49
31.34
27.14
19.23
15.45
0.86
3.46
0.61
1.57
0.18
4.84
12.66
12.40
25.45
18.63
19.33
6.69
68.24
24.02
6.57
1.18
51.46
21.61
10.56**
9.07
7.30
27.24
14.94
15.89*
10.88
12.37
6.10
3.74
1.63**
1.57
1.53"
4.11
31.44
25.69
18.82
14.52
3.94**
1.67
1.77
0.47**
1.68
3.69
13.59
13.43
27.12
17.59
19.14
5.43
69.41
23.39
5.55
1.65
49.12
21.27
13.36
9.00
7.25
26.31
14.37
14.14
12.19
13.03
6.78
4.14
2.56
1.26
0.89
4.27
31.40
26.29
18.99
14.91
2.66
2.41
1.29
0.93
1.08
4.17
13.20
13.00
26.43
18.02
19.22
5.96
Relationship
Dimension-
direction
Lifestyle
Upward
Same level
Downward
Assets
Upward
Same level
Downward
Close
friend
0.621
1.031
0.872
0.478
0.408
0.551
Ordinary
friend
0.395
0.418
0.967
0.282
0.25
0.448
Acquaintance
0.257
0.156
0.463
0.45
0.213
0.262
Stranger
0.126
0.057
0.202
0.445
0.142
0.554
Note. Entries are the number of comparisons adjusted to 10 days.
ward and downward comparisons was stronger on the Lifestyle
than
on the
Assets dimension, F(l, 552)
=
6.86,
p <
.01.
Simi-
larly, the ordinary friend versus all others contrast of downward
comparison versus same level
and
upward was relatively more
pronounced on the Lifestyle than on the Assets dimension,
F(\,
552)
=
8.52,
p <
.01.
In
short, both tendencies—to make simi-
lar comparisons with close friends and to make downward com-
parisons with ordinary friends—were stronger on the Lifestyle
dimension.8
Precomparison Affect Ratings and Direction of
Comparison
Precomparison and postcomparison states were rated on
two
7-point
bipolar adjective pairs:
happy-depressed
and dis-
couraged-encouraged,
r(94) = -.81. The happy-depressed
scale was reversed, and the two scales were averaged to create
an affect measure with higher numbers indicating more posi-
tive affect.
Hypothesis 2 predicted that comparison direction would be
related to precomparison affect. To test this hypothesis, the
average precomparison ratings were categorized into negative Q
to
3.5),
neutrally),
and positive
(4.5
to
7)
states.
The analysis was
a 3 (precomparison states)
X
3 (direction)
X
2 (dimension) AN-
OVA
on the number of social comparisons per
10
days,
in which
Note. Percentages are based on 23.5 comparisons per subject over
13.1
days.
The p
values
refer to significant
sex
differences.
*p<A0. **p<.05.
7 We realize that this contrast is completely post hoc and probably
indefensible;
we
include
it
to draw attention to the serendipitous result.
8 Two other interactions were significant. They are not germane
to
any of the hypotheses, but they do provide information about the na-
ture
of
the dimensions: (a) The interaction
of
Dimension
X
Relation-
ship,
F(3, 276)
=
18.2,
p <
.001, was due
to
Lifestyle comparisons
decreasing steadily from close friend
to
stranger, relative
to
Assets
comparisons. The Lifestyle-Assets contrast by the linear main effect
reported earlier for relationship produced F(l, 276) = 59.98,
p
<
.001.
(b)
The interaction of Dimension
X
Direction,
F(2,184) =
7.5,
p
<
.001,
was due to upward comparisons being relatively more frequent on the
Assets dimension than on the Lifestyle dimension. The Lifestyle-As-
sets
X
Direction (linear) contrast produced F(l, 184) =
11.76,
p
<
.001.
In short, Assets comparisons are more likely
to be
upward and with
less close friends.
SOCIAL COMPARISON765
11*
I Same
I Down
Number of
comparisons:
Residual
means
Figure
1.
RelationshipX Direction interaction for number of compar-
isons:
residual means. (Close = close friend; ordinary = ordinary
friend; acquaint = acquaintance.)
all three variables were repeated measures. Our primary inter-
est was determining whether there was a Direction
X
Precom-
parison Affect interaction and whether this interaction was the
same for the two dimensions.
The Direction X Precomparison Affect interaction, F(4,
368)= 5.17, p < .001, was that downward comparisons were
associated with precomparison positive affect rather than with
precomparison negative affect. The raw means and the interac-
tion residuals
are
shown in
Table
3.
The Linear (affect)
X
Linear
(direction) contrast on the residuals yielded F(l, 368) =17.16,
p <
.001.
The better subjects felt, the more likely they were to
compare downward. There were no other significant interac-
tions.
Effects
of
Comparison
on
Affect
Hypothesis 3 predicted that downward comparison would
enhance subjective well-being and that upward comparison
would decrease it. To test this hypothesis, the affect measure
was analyzed in a 2 (precomparison vs. postcomparison)
X
3
(direction) ANOV\, in which both variables were repeated
measures. The
analysis was
performed separately for each of the
relationships (close friend, ordinary friend, acquaintance, and
stranger) and dimensions (Lifestyle and
Assets),
because
loss
of
subjects would have been too great otherwise. (The repeated
measures analysis required that a subject have upward, same-
level, and downward comparisons, unlike the test of Hypothe-
ses 1
and
2
in which
a zero
comparison
was a
legitimate datum.)
The results of the series of Comparison
X
Direction analyses
on the happy-encouraged measure can be seen in Table 4.
These analyses are close friend, ordinary friend, acquaintance,
stranger, lifestyle, assets, and
overall.
The left third of the table
presents the mean
pre-
and postcomparison
rating.
The middle
third of the table
presents
the mean
change
scores.
The Fs at
the
far right column are the Comparison
X
Direction interaction,
equivalent to a one-way ANOV\ on the change scores.
Regardless of the relationship or the dimension, subjects felt
significantly better after downward comparisons and signifi-
cantly worse after upward comparisons. Overall, they also felt
somewhat better after a same-level comparison.
Self-Esteem
Hypothesis 4 predicted that self-esteem would be related to
direction of comparison, but there were opposing predictions
about how it should be related. Downward comparison theory
(Wills, 1991) predicts that low self-esteem will be related to
downward comparison (particularly under conditions of
threat).
However, there is a considerable body of evidence that
high self-esteem individuals are more likely to engage in
self-
enhancing strategies (see Taylor
&
Brown, 1988, for a review).
We correlated the self-esteem score with the number of up-
ward, same-level, and downward comparisons, separately for
the Lifestyle and Assets dimensions. The left side of Table 5
presents these results.
Self-esteem was positively related to the number of down-
ward comparisons on the Lifestyle dimension; there were no
significant correlations on the
Assets
dimension. Thus, the pre-
diction from downward comparison theory
was
not supported.
Rather, it was high self-esteem individuals who were more
likely to make downward comparisons.
For a finer analysis, self-esteem scores were trichotimized
and added
to the
Precomparison Affect
X
Direction of Compar-
ison analysis shown in Table
3.
Downward comparison theory
Table 3
Number
of
Comparisons in
Each
Direction
as a
Function
of
Precomparison Affect
Happy-encouraged
adjective prerating
Negative
M
Interaction residual
Neutral
M
Interaction residual
Positive
M
Interaction residual
Upward
0.56
1.34
0.61
0.04
0.93
-1.38
Direction of comparison
Same level
0.45
1.00
0.53
-0.09
0.90
-0.91
Downward
0.49
-2.33
0.91
0.04
1.59
2.29
Note. Means are the number of comparisons adjusted to
10
days.
The
number of comparisons is slightly less than shown in Table 2 because
of failures to complete one or more of the precomparison affect scales.
Interaction residuals were multiplied by
10
to avoid rounding errors.
766LADD WHEELER AND KUNITATE MIYAKE
Table 4
Effects
of Comparison
Variable
on
Happy-Encouraged Affect
Upward Same levelDownwardUpward
Change score
Same levelDownwardSignificance
Relationship
Close friend (n = 39)
Precomparison
Postcomparison
Ordinary friend (n = 30)
Precomparison
Postcomparison
Acquaintance in = 23)
Precomparison
Postcomparison
Stranger (« = 12)
Precomparison
Postcomparison
Dimension
Lifestyle (« = 65)
Precomparison
Postcomparison
Assets (n
=
46)
Precomparison
Postcomparison
Overall (n
=
75)
Precomparison
Postcomparison
4.41,
3.69,
4.17,
3.63,
4.21.
3.43,
4.16.
3.19,
4.11.
3-32,
4.45.
3.67,
4.22.
3.52,
4.12.
4.82,
4.43.
4.73,
4.09
4.32
4.90
4.96
4.16.
4.72,
4.47
4.68
4.27.
4.64,
4.35.
4.81,
4.48.
5.02,
4.46.
5.02,
4.65.
5.63,
4.32.
5.03,
4.55.
5.33,
4.42.
5.04,
72.
54.
78.
97.
79.
78.
70.
.70,
•30,
.23,
.06,
56,
•21,
.37,
.46,
•54,
.56,
.98,
•71,
.78C
.62,
F(2,
74) = 14.79**
F(2,
56) = 19.79**
F(2,
42) = 6.78*
F(2,
20) =
10.63**
F(2,
126) =
49.13**
F(2,
88) = 31.49**
F(2,
146) = 50.68**
Note. For the precomparison and postcomparison columns, within each directional choice column, the mean scores that do not share the same
subscript significantly differed from each other. For the change score rows, the mean scores that do not share the same subscript significantly
differed from each other. In all cases, p < .05 using TukeyB. Sample sizes are reduced because of failures to complete one or more of the affect
scales.
*p<.01.
**p<.001.
would predict a three-way interaction such that those low in
self-esteem experiencing negative precomparison affect should
compare downward. However, there were no significant inter-
actions involving self-esteem. Again, there was no support for
the downward comparison theory prediction.
Next we examined how self-esteem related to reactions to
comparisons. We regressed the postcomparison affect scores
on the precomparison affect scores, self-esteem, and the inter-
action between precomparison scores and self-esteem,9 sepa-
rately for
each
direction of comparison on each dimension. The
right side of Table
5
presents the correlations between the aver-
age postcomparison affect rating and self-esteem, with pre-
comparison affect partialed
out,
within each comparison direc-
tion. Higher
scores
on the affect measure indicate more happy-
encouraged responses.
High self-esteem subjects relative to low self-esteem subjects
responded more positively to same-level and upward compari-
son on the affect measures, but only on the Assets dimension.
Remember that the affective response to upward comparison
was generally quite negative (Table
4),
so these correlations do
not mean that higher self-esteem subjects felt good about up-
ward comparisons, but rather that they felt less bad about
them. (An analysis by tertiles of self-esteem confirmed that
even the highest self-esteem subjects responded negatively to
upward comparison.)
The interaction between self-esteem and precomparison af-
fect was included in the regression analysis predicting post-
comparison affect, because downward comparison theory pre-
dicts that
low
self-esteem individuals have a particular need for
enhancement when self-esteem is threatened. In contrast,
Crocker
et
al.
(1987) found that high self-esteem individuals are
more likely to self-enhance when threatened. In fact, however,
the interaction term was never a significant predictor of post-
comparison affect. The effects of self-esteem were independent
of the level of threat.
In summary, high self-esteem was related to
(a)
a greater fre-
quency of downward comparisons on the Lifestyle dimension
and (b) more positive reactions to same-level and upward com-
parisons on the Assets dimension.
Discussion
We
discuss the limitations of the research first, because they
have an impact on what we can say about the meaning of our
results. Although it is common to use first-semester college
freshmen as subjects, ours
is
a
situation in which the fact of their
being in a new and challenging environment may have had a
9 The
change score could be used
as
an index of
reactions,
but it is
correlated to both pre- and postcomparison affect and tends to be
unreliable (Cohen &
Cohen,
1983).
One
could also correlate
the
change
score to self-esteem with precomparison affect partialed
out.
We
felt
uncertain about
such a
roundabout
procedure.
The
procedure
we used
was
more
straightforward and should yield unambiguous results.
SOCIAL COMPARISON767
Table 5
Self-Esteem Correlations
With Number of
Comparisons
and
With
Affective Reactions
to
Comparisons
Dimension
Lifestyle
Upward
Same level
Downward
Assets
Upward
Same level
Downward
Number of
comparisons*
.02
.15
.26*
-.04
.12
.08
Happy-
encouraged11
.01
.11
.12
.25*
.36**
.02
* A
positive correlation indicates that the higher the self-esteem, the
greater the number of comparisons. b A positive correlation indi-
cates that the higher the self-esteem, the more positive the postcom-
parison affect, with precomparison affect partialed out.
* p
<
.05,
two-tailed.
**
p
<
.01,
two-tailed.
strong impact on their
responses.
We
are not certain, for exam-
ple,
that comparison dimensions later in
college
would produce
the same factors. Nor are we convinced that relationship and
direction of comparison would necessarily interact in the same
way. The fascinating result that comparisons with ordinary
friends were predominantly downward might not occur among
respondents who had more time to sort out their friendships.
The freshman might call some dormmates "ordinary friends"
not out of positive sentiment but out of frequent (and perhaps
situationally constrained) interaction. In short, this research
should be seen as bearing on social comparison at a particular
life-transition stage (Deutsch, Ruble, Fleming, Brooks-Gunn,
&Stangor, 1988).
Our method of data collection is a second limitation, as well
as being a strength. We have no illusions that respondents re-
ported every social comparison. Comparisons are often subtle
and fleeting. We do believe, however, that reported compari-
sons were really comparisons and that they were the most sa-
lient comparisons to the respondents. In addition to not having
data on all comparisons, the fact that all the data about a com-
parison were recorded at the same time is problematical, be-
cause it could be argued that precomparison affect does not
really reflect a precomparison appraisal but instead a post-
comparison appraisal of precomparison affect.
We
discuss this
somewhat more fully later in this article.
Comparison Dimensions
The two comparisons dimensions revealed through factor
analysis, Lifestyle and Assets, related to the comparison pro-
cess differently in several ways. Assets comparisons were rela-
tively more likely with acquaintances and strangers and were
more likely to be upward (see Footnote 7). The tendencies to
make similar comparisons with close friends and downward
comparisons with ordinary friends were both less pronounced
on the Assets dimension. Last, the direction of comparison on
the Assets dimension was not related to self-esteem. These dif-
ferences suggest to us that Assets comparisons are more likely
to be forced than are Lifestyle comparisons. Assets (ability, ap-
pearance, and social skills) can be seen fairly directly, even in
strangers, whereas Lifestyle (personality, lifestyle, and aca-
demic matters) are more likely to be revealed through interac-
tion and conversation, over which an individual has more con-
trol. For example, a person does not have to ask a friend how
prepared the friend is for a physics exam if
the
person doesn't
want to be reminded of
his
or her own lack of preparation. On
the other hand, one has less choice in at least noticing differ-
ences in ability or physical appearance. This
was
shown vividly
by a female subject's description of an upward physical appear-
ance comparison with an acquaintance:
I drove up to a building in [dorm complex] and saw this girl I had
met before. She is very pretty and she
was
getting out of a [expen-
sive car] and dressed very nicely. This usually doesn't affect me
but there is something about this
girl
that really
gets
me.
I
felt very
down, jealous, and envious for a while afterwards and
keep
think-
ing about it. I hate the fact that it bothers me like this!
Let us stress that we are not arguing that Lifestyle compari-
sons are totally sought and Assets comparisons are totally
forced; a friend may volunteer that he or she is well- or ill-pre-
pared for an exam, and we may selectively fail to perceive dif-
ferences in ability or appearance. However, there does seem to
be a difference in the degree to which comparisons are sought
or forced. Let us also note that the several internally consistent
and meaningful differences between the two dimensions lends
plausibility and validity to the RSCR.
Direction ofComparison and Relationship With the
Target
As predicted, same-level comparisons were relatively more
frequent with close friends than with
others,
consistent with the
position of Brickman and Bulman (1977) that dissimilar com-
parisons would be painful to one party or the other and consis-
tent with the expectation that close friends ought to be similar
in many ways. The totally unexpected result was that down-
ward comparisons were relatively more frequent with ordinary
friends than with anyone else. We noted earlier that ordinary
friends during the first semester of college may not be like
ordinary friends in more settled circumstances, and so we
would
like to see
these results replicated with
a
different popula-
tion. However, it
is
entirely possible that the phenomenon will
replicate—that one of the functions of ordinary friends is to
afford downward comparison. Downward comparison with
close
friends could threaten the relationship, and
one is likely
to
have some criticisms of ordinary friends that keep them out of
the close friend category.
Precomparison Affect and Direction ofComparison
Probably our most controversial result was that respondents
compared upward when they felt bad and downward when they
felt
good.
The
result is controversial because downward compar-
ison theory
(Wills,
1981,1991) predicts
the
opposite.
According
to this theory, when a person is experiencing low subjective
well-being, she or he will compare downward for purposes of
self-enhancement. There are at least two ways to approach this
inconsistency.
First, the method of data collection may have produced mis-
leading results. Subjects indicated their affect before compari-
768LADD WHEELER AND KUNITATE MIYAKE
son, the direction of
the
comparison, and their affect after com-
parison on the same questionnaire, perhaps obscuring the di-
rection of
causation.
Thus,
feeling bad after
an
upward compari-
son may have led subjects to infer that they felt bad before the
comparison, and feeling good after a downward comparison
may have led them to infer that they felt good before the com-
parison, leading
to
an artifactual relationship between precom-
parison affect and direction of comparison. This problem
might be particularly acute when there was a long period of
time between
the
occurrence of the comparison and
the
record-
ing of it. If subjects recorded a comparison immediately after it
occurred, we would expect them to be aware of their precom-
parison affect. If, however, the comparison occurred hours be-
fore it was recorded, memory of precomparison affect might
not be clear, but memory of postcomparison affect would be
good, because recording the comparison would revivify reac-
tions to it. Under these circumstances, inferring precompari-
son affect from postcomparison affect would seem to be more
likely. Unfortunately, we did not collect data about the times at
which comparisons occurred and when they were recorded.
This should certainly be done in future research.
The major argument against this affect-inference explana-
tion is that subjects were able to distinguish between pre- and
postaffect, as shown by the strong affect changes as a result of
comparison direction (Table
4).
The affect-inference argument
would
have
to get very tortuous to explain how preaffect can be
inferred from postaffect when the two differ systematically
as
a
function of comparison direction.
The second possibility is that downward comparison theory
is wrong about low subjective well-being leading to downward
comparison. Most of the recent supporting evidence comes
from studies of victimized individuals, such
as
cancer patients.
The widely accepted reasoning about these studies
is as
follows:
(a) Victimized individuals are threatened; (b) victimized indi-
viduals make downward
comparisons;
and
(c)
therefore, threat-
ened people make downward comparisons. However, the
various studies of victimized individuals have not found a rela-
tionship between severity of threat and downward comparison,
as is implied by downward comparison theory (cf. Wood &
Taylor, 1991, pp. 43-45). Moreover, the second assumption,
that victimized individuals make downward comparisons, can
be questioned; whether they do appears to depend on the
method used for measuring comparisons. For
example,
Taylor,
Wood, and Lichtman (1983) found that, when asked directly
how they were coping in comparison to
others,
only
2
out of 78
cancer patients said that they were coping worse than other
cancer
patients.
In addition, free responses made
by
these same
patients in interviews (Wood, Taylor, & Lichtman, 1985) indi-
cated a strong preponderance of downward comparison. In
contrast, using retrospective questions about the direction of
comparison and responses to it, Buunk, Collins, Taylor, \an-
Yperen, and Dakof (1990) found that cancer patients from a
very similar population were only slightly less likely to have
made upward comparisons than to
have made downward
com-
parisons. Furthermore, Taylor and Lobel (1989)
reviewed
the
literature on cancer patients and concluded that these patients
preferred contact with and information about
others "who have
either overcome their threatening circumstances or adjusted
well to them and that they avoid exposing themselves to those
who are doing poorly" (p. 571).
Thus,
whether one finds that victimized individuals make
downward comparisons seems to depend on what one accepts
as a measure of comparison. Conclusions about comparative
status,
whether indicated on rating scales or by free response in
interviews, point to downward comparison. Preferences for in-
formation and contact point to upward comparison. Memory
of comparisons made points about equally to upward and
downward comparisons.
What of the Experimental
Evidence
That
Threat
Leads to
Downward Comparison?
Most of the experimental evidence comes from studies using
the rank order paradigm, which in the vast majority of cases
has shown upward comparison.
The study most frequently cited in support of the hypothesis
that threat leads to downward comparison is that by Hakmiller
(1966).
In this
study,
the subject
was
told that, on the basis of the
Minnesota Multiphasic Personality Inventory (MMPI), she
should belong to
a
subgroup of three people
(out
of
six)
with low
scores on the personality trait "hostility toward one's own par-
ents,"
described very
negatively,
as
leading
to
personality deteri-
oration (high
threat),
or somewhat
positively,
as
reflecting matu-
rity (low threat). However, a second test, which was said to
relate to the MMPI with better than 50-50 accuracy in rank
ordering the entire group, showed the subject to have a quite
high score, either
60
or
80
on a scale of
100.
Subjects in the high
threat condition were more likely to choose to see the highest
score (most hostility) in the group than subjects in the low
threat condition, who showed no preference for any particular
score.
Hakmiller himself recognized the possibility of an alterna-
tive "maudlin [sic] curiosity" explanation, namely, that the
highest score "might have been viewed as less 'pathological'
and therefore, less interesting to the Low-Threat subjects"
(Hakmiller, 1966,
p.
38). In short, given the descriptions of the
trait, subjects in the two conditions were not faced with dimen-
sions
of equal drama and interest. In the much cleaner manipu-
lation of threat, there was no difference between subjects re-
ceiving a score of 60 and those receiving a score of
80,
although
the latter group was predicted to be more threatened than the
former in the high-threat condition. Hakmiller's study has
never been replicated, and many rank order studies
have
shown
choice of
the
highest score in the group even when the trait to
be evaluated
is
a positive
one.
(For explanations, see Arrowood
&
Friend,
1969;
Gruder, Korth, Dichtel,
&
Glos,
1975;
Wheeler
etal., 1969.)
In fact, two of the other most frequently cited studies used
avoidance of the highest (most positive) score as evidence for
defensive comparison under threat (Smith
&
Insko,
1987; Wil-
son & Benner, 1971). The Smith and
Insko study had
the clear-
est results, showing that the highest ranked score was chosen
less frequently when