Demonstrations of Implicit Anti-Fat Bias:
The Impact of Providing Causal Information and Evoking Empathy
Bethany A. Teachman
University of Virginia
Kathrine D. Gapinski, Kelly D. Brownell, and
Three studies investigated implicit biases, and their modifiability, against overweight persons. In Study 1
(N ⫽ 144), the authors demonstrated strong implicit anti-fat attitudes and stereotypes using the Implicit
Association Test, despite no explicit anti-fat bias. When participants were informed that obesity is caused
predominantly by overeating and lack of exercise, higher implicit bias relative to controls was produced;
informing participants that obesity is mainly due to genetic factors did not result in lower bias. In Studies
2A (N ⫽ 90) and 2B (N ⫽ 63), participants read stories of discrimination against obese persons to evoke
empathy. This did not lead to lower bias compared with controls but did produce diminished implicit bias
among overweight participants, suggesting an in-group bias.
Key words: obesity, anti-fat, bias, stereotype, attitude, implicit
Negative attitudes toward obese persons are pervasive, and the
effects of anti-fat discrimination are evident across key domains of
life, from employment to education to health care (Puhl &
Brownell, 2001). Unlike stigma encountered by most other mar-
ginalized groups, the stigma of obesity is somewhat unique in that
both obese and average-weight people report similar levels of
dislike toward overweight persons as a group, suggesting no pro-
tective in-group bias (Crandall, 1994). It is not surprising, given
the pervasive negative messages, that obesity among women is
associated with higher probability for depressive episodes and a
marked increase in suicidal ideation and attempts (Carpenter,
Hasin, Allison, & Faith, 2000). Further, the experience of stigma
is associated with negative health consequences (Guyll, Matthews,
& Bromberger, 2001; Krieger, 1999). In fact, research shows that
poor health, diminished quality of life, and lowered access to
health services have all been related to discrimination based on
age, gender, and race (Forster, 1993; Williams, 1999). The same
consequences may follow from weight discrimination. These con-
sequences may be particularly important in the obesity domain,
given that obese persons are already at heightened risk because of
compromised physical health associated with obesity. Therefore,
evaluating anti-fat attitudes and discrimination may be critical
from both a social and a health perspective.
Anti-fat attitudes can be expressed both explicitly, through
conscious self-report, and implicitly, when an evaluation occurs
outside of awareness or conscious control. Most research on anti-
fat bias has used explicit measures of attitudes and stereotypes, but
new evidence suggests that anti-fat bias can be activated without
conscious awareness or intention and can even differ in important
ways from conscious views. Bessenoff and Sherman (2000) used
a lexical decision task to demonstrate that implicit anti-fat evalu-
ations predicted how far participants chose to sit from an over-
weight woman, whereas explicit attitudes did not. Teachman and
Brownell (2001) found strong implicit bias even among health
professionals who specialized in obesity treatment and who did not
report negative attitudes toward overweight persons.
It is clear that the relationship among explicit and implicit
evaluations is variable. The literature often reveals little associa-
tion between implicit and explicit stereotypes (see Dovidio,
Kawakami, Johnson, Johnson, & Howard, 1997; Greenwald &
Banaji, 1995). This incongruence can occur because negative
implicit responses to marginalized groups can happen outside of
awareness or because individuals are motivated to deny these
responses, perhaps to appear or be fair minded. We assume that
implicit and explicit attitude measures are both valid assessments
of a given person’s evaluation but that they reflect different com-
Bethany A. Teachman, Department of Psychology, University of Vir-
ginia; Kathrine D. Gapinski, Kelly D. Brownell, and Melissa Rawlins,
Department of Psychology, Yale University; Subathra Jeyaram, Depart-
ment of Psychology, Oxford University, Oxford, England.
We thank Rebecca Puhl, Marlene Schwartz, Brian Nosek, and other
members of the Yale University Eating Disorders and Implicit Social
Cognition research groups for their helpful advice and feedback on this
project. In addition, we are grateful to the Rudd Foundation for their
support of this project.
Correspondence concerning this article should be addressed to Bethany
A. Teachman, Department of Psychology, University of Virginia, 102
Gilmer Hall, P.O. Box 400400, Charlottesville, Virginia 22904-4400.
Health Psychology Copyright 2003 by the American Psychological Association, Inc.
2003, Vol. 22, No. 1, 68–78 0278-6133/03/$12.00 DOI: 10.1037/0278-6188.8.131.52
ponents of the attitude response, which is why they can be variably
Little is known about the pervasiveness of implicit anti-fat
biases in the general population or about the relationship among
implicit and explicit anti-fat attitudes and stereotypes. Even less is
known about changing these implicit attitudes in the service of
reducing bias, although there is now mounting evidence that some
implicit biases can be modified (at least temporarily; e.g., race
bias; Dasgupta & Greenwald, 2001). This article presents the
results of three studies that evaluated implicit and explicit anti-fat
biases among general and student populations and investigated two
theoretically derived approaches to changing attitudes. The litera-
ture suggests that anti-fat bias is especially strong because being
overweight is deemed blameworthy (e.g., Crandall, 1994). Further,
the fact that weight is thought to be controllable is likely to reduce
empathy for obese persons. Consequently, we hypothesized that
manipulations that reduced blame (by manipulating the controlla-
bility of obesity) or increased empathy could shift anti-fat atti-
tudes. On the basis of societal messages that being thin is beautiful
and being fat is abhorrent, we expected that the general population
would demonstrate strong implicit biases against obese persons.
However, given social norms to appear tolerant, we expected little
explicit report of anti-fat bias.
Study 1 was designed to investigate implicit attitudes and ste-
reotypes toward obese persons in the general population, to deter-
mine how these attitudes relate to explicit bias, and to evaluate
whether information about the causes of obesity would influence
the expression of implicit and explicit biases.
Evaluating perceived controllability as a bias-reduction strategy
derives from work by Crandall and colleagues, who suggested that
anti-fat attitudes result from holding individuals responsible for
their condition (Crandall, 1994; Crandall & Biernat, 1990). Several
studies have indicated that reducing the perceived control over-
weight persons are judged to have over their weight may decrease
explicit anti-fat bias (e.g., DeJong, 1980, 1993). For example,
Crandall (1994) was able to change the common opinion that
overweight people lack willpower by educating participants about
the contributions of genetics and physiology in obesity (thereby
reducing blame). It is not known whether implicit bias can be
changed. Given this evidence that causal attributions can influence
bias and research suggesting that the general public is misinformed
about the true causes of obesity, Study 1 examined whether pro-
viding information about the causes of obesity could affect both
implicit and explicit anti-fat biases. We hypothesized that bias
would be lower when participants were informed that the primary
cause of obesity was genetics (i.e., a cause outside of the person’s
control) and higher when the principal causes were indicated as
overeating and lack of exercise (i.e., causes perceived to lie within
the person’s control).
A large tent was set up at a Connecticut beach, and participants who
approached the tent or were walking by were invited to take part in a short
survey about their attitudes toward different groups in exchange for either
$1, a lottery ticket, or a beverage. The refusal rate was less than 5% after
people approached the tent. This locale was used because it is an environ-
ment where weight and shape are salient, increasing the accessibility of
Further, other studies conducted at the beach using
implicit attitude measurement found comparable effects for attitudes mea-
sured in the laboratory on political preferences and sexual orientation
(B. A. Nosek, personal communication, August 5, 2001). Individuals who
approached the tent were screened to determine that they were over 18
years old and could read English fluently, and then they completed the
study either alone or in a group of up to three people. The sample (N ⫽
144) was approximately equal for gender (54% female), had a mean age
of 35 years (SD ⫽ 13.99; range ⫽ 18–78 years), was predominantly
Caucasian (89%), and had a mean body mass index (BMI) of 26.37
(SD ⫽ 4.71) and 24.44 (SD ⫽ 4.61) for men and women, respectively.
Cause of obesity primes. Participants were assigned to one of three
prime conditions. Either they received no prime (no-prime condition, N ⫽
48) or they were asked to read a “recently published news article” that
reported the results of a research study indicating that the primary cause of
obesity was genetics (genetics condition, N ⫽ 48) or the primary causes of
obesity were overeating and lack of exercise (behavior condition, N ⫽ 48).
This design therefore manipulated whether obesity was perceived to be
predominantly within or outside personal control. To make the manipula-
tion believable, we designed the prime to indicate that 80% of the cause of
obesity could be explained by either genetics or the behavioral factors, and
the remaining 20% was due to the other factor (i.e., genetics or behavior,
whichever was not stated as the primary cause).
Implicit bias. The Implicit Association Test (IAT; Greenwald,
McGhee, & Schwartz, 1998) is a measure that has been used to reflect
implicit attitudes primarily related to social prejudice, such as gender
stereotypes (e.g., Rudman, Greenwald, & McGhee, 1996) and racial eval-
uations (e.g., Dasgupta, McGhee, Greenwald, & Banaji, 2000), and has
only recently been applied to clinical research (e.g., Teachman, Gregg, &
Woody, 2001). The IAT uses reaction time to measure implicit memory-
based associations without requiring conscious introspection. Processing
speed is assumed to be an indirect measure of the individual’s degree of
association between two concepts, and the degree of implicit association is
interpreted as an index of a person’s unconscious or implicit attitude. The
measure is implicit in the sense that it measures evaluations that occur
outside conscious control and, at times, outside conscious awareness. The
IAT instructs participants to classify words into superordinate categories
(categories that are at a more general level). For example, in the practice
task for the current study, participants decided whether words such as
daisies, tulips, bugs, and mosquitoes belonged to the superordinate cate-
gory flowers or insects. Simultaneously, they classified words associated
with the descriptive categories good and bad. Participants classified words
from the four categories under two different conditions. In one condition,
the category labels flowers plus good versus insects plus bad were paired;
in the other condition, the labels were switched, so flowers plus bad versus
insects plus good were paired together. Participants generally categorized
stimuli faster when the paired categories matched the way they
Supporting this assertion, we found higher effect sizes at the beach on
our measures of implicit anti-fat bias relative to other implicit attitude
measures gathered at the same site and time (e.g., political attitudes: d ⫽
.8 and attitudes toward sexual orientation: d ⫽ .9; B. A. Nosek, personal
communication, August 5, 2001).
IMPLICIT ANTI-FAT BIAS
implicitly associated or evaluated those categories in memory (e.g., flowers
as good and insects as bad) than when they were mismatched.
To evaluate implicit anti-fat bias, we asked participants to classify
stimuli while associating fat and thin people with positive and negative
attributes. Because of negative social attitudes about weight, we expected
stimuli to be classified more easily when the category pairings reflected
negative associations toward obesity (e.g., fat people with bad) versus
category pairings that reflected positive associations (e.g., fat people with
good). Implicit associations to one target category were assessed relative to
a participant’s associations to the other target category (associations with
fat people were measured relative to implicit associations with thin people).
Since we were interested in both implicit attitudes and stereotypes toward
overweight individuals, we had participants complete two different IAT
tasks. To measure attitudes, or simple valence evaluation, participants
completed an IAT in which they were asked to associate the target
categories fat people and thin people with the attribute categories good and
bad. To measure an implicit stereotype about overweight individuals,
participants completed an IAT in which they were asked to associate the
target categories fat people and thin people with the attribute categories
motivated and lazy.
Each IAT task consisted of two pages (the order was counterbalanced
across participants). On one page, the target and attribute categories were
paired on either side of a column in a way expected to match negative
implicit associations with overweight (e.g., fat people with bad heading up
one side of the column and thin people with good heading up the other side
of the column; see sample IAT page in Teachman & Brownell, 2001, p.
1531). On the other page, the target and attribute categories were paired to
contradict expected negative associations with overweight (e.g., thin peo-
ple was paired with bad on one side and fat people was paired with good
on the other side). Participants were given 20 s to classify as many words
as possible on one page; then they were given 20 s to classify words on the
second page, on which the category pairings were switched. The variable
of interest was the difference in the number of correctly classified items
under the two different category pairings (incorrectly classified items were
not counted in the difference score because errors can indicate misunder-
standing of the instructions or an effort to increase speed at the cost of
accuracy, thus distorting the actual implicit evaluation).
Participants were asked to work as quickly and accurately as possible,
and they were told to try to avoid making mistakes (i.e., misclassifying a
word) but to continue without stopping should this occur. Further, they
were told not to skip items. Given the novelty of the task, we asked all
participants initially to complete an unrelated practice IAT task to famil-
iarize them with the procedure. Three items for each target and attribute
category were selected based on participants’ expected ease of categoriza-
tion and familiar usage. These stimuli were approximately matched for
length, and the ease of categorization was evaluated during pretesting.
Explicit bias. Participants completed an established measure of stereo-
types about overweight individuals. The Fat Phobia Scale (FPS; Robinson,
Bacon, & O’Reilly, 1993) is a 50-item, semantic differential scale in which
participants rate their feelings about what “fat people are like” on a series
of different opposing dimensions (e.g., smart vs. stupid). It parallels the
IAT design in that it compares both ends of an attribute dimension
(whether fat people are smart relative to stupid) but is dissimilar in that it
does not evaluate attitudes toward fat persons relative to thin persons
(asking only for judgments about overweight individuals).
Manipulation check. To evaluate the effectiveness of the prime ma-
nipulation about the causes of obesity, we asked participants, “What do you
feel is the primary cause of obesity?” as the final question of the study. We
then coded their open-ended answers to indicate whether they reported a
factor that lay within or outside personal control (e.g., overeating vs.
genetics) to determine whether participants remembered the news article
they had read.
Participants first completed a general demographics questionnaire and
then received instructions on how to complete the IAT practice task.
Participants completed one of four versions of the practice task (which was
counterbalanced to control for order effects and placement of category
labels on either the left or right column), and then they underwent the prime
manipulation. Participants in the no-prime condition simply completed the
IAT tasks, whereas participants in the genetics and behavior conditions
carefully read the “recently published news article” before completing the
two IAT tasks. The order of the implicit attitude and stereotype tasks was
counterbalanced, and, within each IAT task, the order in which matched
versus mismatched category pairings appeared was counterbalanced. Par-
ticipants then completed the FPS and the manipulation check.
Data were checked to confirm that all participants completed a
minimum of five items on each IAT page as evidence that they
understood and attended to the task (one lazy–motivated IAT score
was deleted based on this criterion). Next, IAT pages with high
error rates (i.e., ⱖ35% incorrectly classified items) were omitted
because this may indicate distraction or lack of understanding.
These corrections resulted in deleting 17 participants’ bad–good
and/or lazy–motivated IAT scores from IAT analyses (their ex-
plicit data were retained). In addition, 1 participant’s IAT effect
was more than 3 SDs above the mean. However, when data were
analyzed both with and without this participant’s data, the pattern
of results was the same, so we report the full sample here.
A logistic regression to determine whether prime condition
would predict whether a factor that lay within or outside personal
control was stated as the primary reason for obesity was signifi-
(1, N ⫽ 133) ⫽ 4.86, p ⫽ .03, B ⫽⫺.53, SE ⫽ .24.
Fifty-nine percent of the no-prime group reported an internal cause
such as overeating (suggesting that this is the baseline response),
whereas 81% of the behavior prime group indicated an internal
cause, compared with only 65% of the genetics prime group.
Follow-up t tests between the behavior and genetics prime groups
(the two main groups of interest) indicated significant differences
in the reporting of internal versus external causal factors for
obesity, t(88) ⫽ 3.53, p ⫽ .001, d ⫽ .79, and similar differences
were found between the behavior and no-prime groups, t(86) ⫽
4.55, p ⬍ .0001, d ⫽ .98.
There was no significant difference between the genetics and
no-prime groups, t(89) ⫽ 1.15, p ⬎ 1.0, which somewhat limits the
conclusions that can be drawn from the genetics manipulation, but
the nature of the study design required that the behavior and
genetics research study manipulations be identical to make the test
fair. Consequently, the overall difference across groups and the
difference between the behavior and genetics prime groups were
the critical factors in establishing the validity of the test. The
genetics manipulation is congruent with previous research by
The materials used across the three studies are available from Bethany
A. Teachman on request.
TEACHMAN, GAPINSKI, BROWNELL, RAWLINS, AND JEYARAM
Chlouverakis (1975), who found that many people considered
obesity to be due to overeating despite expert opinion about the
role of genetic factors. These studies suggest that blaming attitudes
toward obese persons are easier to exacerbate than to diminish.
Because the IAT is a relative measure, evidence of an implicit
effect can be interpreted as both pro-thin and anti-fat biases. IAT
effects were calculated by creating difference scores where the
number of items correctly classified in the mismatched condition
(e.g., fat people ⫹ motivated) was subtracted from that in the
matched condition (e.g., fat people ⫹ lazy). This resulted in a
positive score for most participants but a negative score for indi-
viduals who classified more items when fat people was paired with
positive attributes (only 4% of the sample indicated such a pro-
fat/anti-thin bias). To better control for individual differences in
the number of items completed and to maximize the reliability of
correlations, we inserted the difference score into the following
algorithm: max/min ⫺ 1 ⴱ
max ⫺ min
, where maximum (max)
and minimum (min), respectively, represent the category pairings
where the highest versus the lowest number of items was correctly
classified. This composite scoring of the IAT is based on simula-
tions run by Nosek and Lane (1999).
To establish evidence of implicit bias, we conducted t tests to
show that the IAT effects differed from zero. More items were
correctly classified when fat people was paired with negative
attributes, lazy: t(110) ⫽ 9.68, p ⬍ .0001, d ⫽ 1.85; bad:
t(109) ⫽ 10.37, p ⬍ .0001, d ⫽ 1.99, than with positive attributes.
The implicit attitude (bad–good) and stereotype (lazy–motivated)
IAT measures were not significantly different from one another,
t(93) ⫽ 1.11, p ⬎ .10, d ⫽ .23, and were moderately positively
correlated (r ⫽ .33, p ⫽ .001).
In contrast to the strong evidence for anti-fat/pro-thin implicit
bias, the total score for the explicit FPS indicated a slight pro-fat
bias. The average item score (M ⫽ 2.81, SD ⫽ .44) was signifi-
cantly different from 3 (the neutral point on the 5-point semantic
differential scale), t(143) ⫽ 5.29, p ⬍ .0001, d ⫽ .88, and all six
factors of the scale indicated a similar pattern.
The bad–good IAT was not significantly related to the explicit
FPS (r ⫽ .03, p ⬎ .10), but the lazy–motivated IAT showed a
positive correlation with the scale (r ⫽ .29, p ⫽ .002). Thus, the
implicit and explicit stereotype measures were related, but the
attitude measures were not. These variable relations are not sur-
prising given the relative nature of the implicit measures (fat vs.
thin people) compared with the explicit scale and the likelihood
that explicit responses are more vulnerable to social desirability.
Additionally, there were neither significant sex differences on the
bias measures nor significant correlations with age or education.
Interestingly, BMI for women (but not men) was negatively related
to the lazy–motivated IAT (r ⫽⫺.34, p ⫽ .008), indicating that
women with higher body mass tend to have weaker stereotypes of
fat people as lazy. This finding suggests that an in-group bias
among women may be present (e.g., higher weight persons show
less negative beliefs about overweight people).
Influence of Manipulating the Causes of Obesity
on Anti-Fat Bias
To increase reliability, we averaged the bad–good and lazy–
motivated IATs, and then we used this total IAT and the total
explicit scale as the dependent measures for planned contrasts
comparing the three prime conditions. As expected, telling partic-
ipants that obesity was primarily due to overeating and lack of
exercise resulted in higher implicit bias, t(124) ⫽⫺2.22, p ⫽ .03,
d ⫽ .40, and a trend toward higher explicit bias, t(141) ⫽⫺1.91,
p ⫽ .06, d ⫽ .32, relative to the no-prime control group and
marginally significantly higher implicit bias than the genetics
prime condition, t(124) ⫽⫺1.69, p ⫽ .10, d ⫽ .30. In contrast,
telling participants that obesity was primarily due to genetics did
not result in a lower bias than the no-prime condition for either the
implicit or explicit measures (both ts ⬍ .60, p ⬎ 1.0). For easier
visual inspection, Figure 1 indicates the z scores (with a meaning-
ful zero point) and standard error bars for the implicit and explicit
bias measures for each prime condition.
Additional Explicit Stereotype Data
To address the absence of data regarding explicit stereotypes
about thin people, we asked a second sample of participants (N ⫽
Figure 1. Implicit and explicit anti-fat/pro-thin bias by cause of obesity
prime condition in Study 1. To facilitate comparison of the magnitude of
implicit and explicit biases, both measures were transformed into z scores
(making the scales equivalent), and then a meaningful zero point was
created by subtracting the mean and standard deviation from each z score.
Thus, the zero point indicates no preference for fat or thin people, a
positive value indicates anti-fat/pro-thin bias (as evident on the implicit
measure, which is the average of the Implicit Association Test [IAT]
tasks), and a negative value indicates pro-fat bias (as evident on the explicit
measure, which is the Fat Phobia Scale [FPS]). Standard error bars are
shown to indicate the variance for each group, taking into account the
IMPLICIT ANTI-FAT BIAS
25) from the same beach location to complete the FPS, rating what
thin people were like. Although this did not allow direct compar-
ison of attitudes toward fat versus thin people (because the samples
were independent), it did permit some comparison, given the
parallel nature of the FPS and the Thin Phobia Scale (TPS) and the
assumed similarity of the samples.
As expected, participants rated
thin people slightly more positively than negatively on the total
scale (reflecting the 50 semantic differential adjective pairs),
t(24) ⫽ 3.94, p ⫽ .001, d ⫽ 1.61, indicating that the average item
score (M ⫽ 2.69, SD ⫽ .40) was significantly different from 3, the
neutral point on the 5-point semantic differential scale. Impor-
tantly, average item scores did not differ across the FPS and the
TPS, t(167) ⫽ 1.12, p ⬍ .20, suggesting no reported explicit bias
toward fat people.
Study 1 provided strong support for the presence of anti-fat/pro-
thin implicit attitudes and stereotypes but a relative absence of
explicit anti-fat bias. The manipulation of information about the
cause of obesity indicates that the bias can be made more negative
by promoting personally controlled attributions of cause but can-
not easily be made more positive, given the difficulty of convinc-
ing the general public that obesity is largely due to genetic factors.
In Study 2A, we used a new approach to bias reduction. Rather
than providing statistical information about the causes of obesity,
we tried a more affective manipulation in the hopes of changing
feelings about overweight persons. The rationale for trying this
new approach came in part from theoretical and empirical work
demonstrating that evoking empathy to a member of a stigmatized
group can reduce bias and also from our Study 1, which pointed to
the difficulty of changing attitudes through an information
Empathy is similar to sympathy but has a stronger component of
relating to another person and taking his or her perspective. Batson
et al. (1997) induced more positive attitudes toward stigmatized
groups, such as AIDS victims, homeless people, and convicted
murderers, by evoking empathy for members of these groups. Few
studies have used empathy to reduce anti-fat stigma. However, an
intervention by Wiese, Wilson, Jones, and Neises (1992) with
medical students included (among other strategies) a video of an
obese woman talking about discrimination she faced and role-
playing exercises in which students took the perspective of an
obese person. These components of the intervention likely induced
empathy, and this study did find reduced bias against obese per-
sons. On the basis of these suggestive findings, we explored
whether empathy induction by means of written stories of weight
discrimination could lead to lower anti-fat bias. We hypothesized
that bias would be lower relative to controls because the discrim-
ination story would evoke empathy for the overweight protagonist
of the story, which would then generalize to the larger group of
Participants completed explicit and implicit measures of anti-fat
bias after reading story primes that either (a) evoked empathy
toward an obese person, (b) evoked empathy toward a person in a
wheelchair (an alternate stigmatized group), or (c) were neutral in
valence (and were not expected to evoke empathy). This study was
conducted in the laboratory to provide better control, and a com-
puterized form of the IAT was used that allowed us to evaluate a
range of implicit bias measures. In addition, in Study 2A we used
a different approach to assess explicit stereotypes about fat people
relative to thin people to make the implicit and explicit measures
more comparable, thereby addressing this weakness from Study 1.
Female participants were recruited from the Yale University community,
either through introductory psychology classes or from signs posted around
campus. Only women were used in this study because of the need to reduce
the confound of mismatched gender of participant to gender depicted in the
stimuli (i.e., stories and pictures). They completed the study in exchange
for $7 or course credit. The sample (N ⫽ 90) had a mean age of 21 years
(SD ⫽ 3.87; range ⫽ 18–44 years); the reported ethnicity was 56%
Caucasian, 21% Asian/Pacific Islander, 10% Black, 8% Hispanic, and 5%
other; and the mean BMI was 21.48 (SD ⫽ 2.63).
Story primes. Empathy was evoked by having participants read a
first-person account of the troubles experienced by a member of a stigma-
tized group (an approach well established in previous research; e.g., Bat-
son, Klein, Highberger, & Shaw, 1995; Batson et al., 1997) and then
having them think about the feelings of that individual and complete a
writing exercise to further elaborate their feelings. Depending on prede-
termined counterbalancing, participants were assigned to one of three
conditions: reading about a person who is obese, reading about a person
who is in a wheelchair, or reading a neutral story about a nonstigmatized
person. The wheelchair condition was included to determine the effects of
evoking empathy toward a nontarget group to see whether the effect would
generalize to influence anti-fat/pro-thin bias. In each condition, participants
read two narratives that were three to four pages in length, followed by a
short writing task. The setting of one story was always a student’s first day
at college, and the other was a 15-year-old’s first day at sleep-away camp
(order of the story primes was counterbalanced).
In the obesity-empathy condition, the story primes gave first-person
accounts of an extremely overweight woman experiencing prejudiced jokes
and social rejection because of her weight, with the protagonist describing
her sadness and hurt feelings. The story primes for the wheelchair-empathy
condition used identical wording, except that instead of being overweight,
the main character was in a wheelchair. Finally, story primes in the neutral
condition used the same college and camp settings and also gave first-
person accounts, but these accounts were from the perspective of a non-
stigmatized woman who described activities that were neutral in valence,
without referencing emotions or feelings.
Manipulation check. To check the effectiveness of the empathy ma-
nipulation, we asked participants to complete a short questionnaire that
rated their feelings toward the main character in the story on a 7-point
Likert scale for each of six characteristics: sympathetic, warm, compas-
sionate, softhearted, tender, and moved. This empathy index was devel-
oped by Batson and colleagues and has been used to measure empathy in
a number of previous studies (e.g., Batson, 1991; Batson et al., 1995).
Implicit bias. The computerized version of the IAT is the more com-
mon form of the test and follows logic identical to that of the paper-and-
pencil version already discussed. Participants were asked to classify stimuli
Demographic information was not gathered for this second sample;
however, given the identical location and recruitment procedures, the
samples were expected to be comparable.
TEACHMAN, GAPINSKI, BROWNELL, RAWLINS, AND JEYARAM
into categories as quickly and accurately as possible to index the strength
of implicit associations between two paired categories (relative to a dif-
ferent pairing). The dependent variable was the difference in average
response latency across all classification trials for one category pairing
versus the other pairing.
In the computer IAT, picture and word stimuli appeared in random order
in the center of a PC screen from any of the four categories being
associated (e.g., fat people, thin people, good, bad), while two of the
category labels were always paired visibly near the top of one side of the
screen (such as fat people paired with bad) and the other two category
labels were paired on the other side of the screen (in this case, thin people
paired with good). After participants categorized stimuli into their respec-
tive sides of the screen on the basis of the paired category labels, the
category pairings were switched (i.e., switching the example above, fat
people was paired with good and thin people was paired with bad). Again,
participants were asked to categorize stimuli from the four categories into
the new category pairings (see Greenwald et al., 1998, for a more detailed
description). Participants completed an unrelated practice IAT task to
initially familiarize themselves with the procedure and a series of 20
practice trials classifying the stimuli for each new category pairing before
the 36 experimental classification trials were conducted. Stimuli included
words to depict each category label, along with pictures of overweight and
Participants completed two sets of IAT tasks, with the order of the IAT
tasks and category pairings randomized within each task (there was a break
between the two sets to reduce fatigue and to reinforce the empathy
manipulation). In the first set of IAT tasks, the strength of associations
between three pairs of category labels was evaluated. Two parallel attitude
tasks were included, comparing associations toward fat people and thin
people as good versus bad. In one case, word stimuli were used to depict
all four categories (referred to as Fat/Bad Words IAT), and in the other
case, pictures were used to depict fat and thin people (referred to as
Fat/Bad Pictures IAT). The third task in this set looked at stereotypes of fat
people and thin people (depicted using pictorial stimuli) as valuable versus
worthless (depicted using word stimuli; referred to as Fat/Worthless IAT).
The second set of IAT tasks looked at another bad–good attitude measure
but varied the category labels, using the terms overweight people and
underweight people (referred to as Overweight/Bad IAT). In the final task,
implicit evaluations of one’s own weight were determined on the basis of
associations between the categories fat and thin with me and not me
(referred to as Others/Fat IAT).
Explicit bias. Participants were asked to rate their feelings, on a
7-point semantic differential scale, from 1 (bad/negative)to7(good/
positive), describing how they feel fat people and thin people are. Attitudes
toward normal weight people were also evaluated to allow a different
relative comparison to fat and thin people. Using analogous scales, partic-
ipants were asked to rate how worthless to valuable they felt fat people and
thin people were. In addition, participants used comparable semantic
differential scales to rate how underweight or overweight they felt them-
selves to be, as well as how underweight or overweight they felt others to
be on average. These explicit rating scales mirrored the relative associa-
tions assessed by the implicit association tests, so in each case a difference
score between attitudes toward fat versus thin people, and the self versus
others, was obtained.
Participants were told that the purpose of the study was to investigate
attitudes and feelings in novel situations and that they would be reading
two true stories. Following informed consent, the participants then read the
first story (either obesity empathy, wheelchair empathy, or neutral, de-
pending on their condition) and completed the associated writing exercise
and manipulation check. Participants then completed the first set of IAT
tasks. To reinforce the empathy manipulation, participants then read the
second story prime, with the associated writing exercise and manipulation
check, followed by the second set of IAT tasks and a demographics
questionnaire. The placement of the explicit bias measure was counterbal-
anced so that half of the participants completed it before the first set of IAT
tasks and half completed it at the end of the study with the demographics
questionnaire. Finally, participants were fully debriefed and thanked for
Prior to conducting the planned analyses, IAT response latencies
less than 300 ms or greater than 3,000 ms were counted as errors
and recoded as 300 or 3,000 ms, respectively. In addition, data
were examined for error rates (i.e., percentage of stimuli classified
incorrectly) on the critical IAT blocks. All participants’ total error
rates were less than 25%, though 1 participant had one IAT task
score removed because of an error rate on this task greater than
40%. Additionally, 1 participant’s IAT data were excluded be-
cause this individual’s average response time was over 1,400 ms
on the IAT tasks. This slow responding suggests strategic, rather
than implicit, responding. Further details on these data manage-
ment procedures are provided in Greenwald et al. (1998). Finally,
the split-half reliability of the IAT data was calculated across IAT
tasks (average r ⫽ .67; range r ⫽ .52–.83), suggesting good
psychometric properties relative to other reaction time measures.
To establish the effectiveness of the manipulation, we compared
the total score on Batson’s empathy index (e.g., Batson, 1991)
across the three prime conditions using an analysis of variance
(ANOVA). As expected, results indicated a significant difference
across conditions, F(2, 86) ⫽ 24.54, p ⬍ .0001, f ⫽ .39, and
follow-up tests confirmed higher reported feelings of empathy
toward the main character in the obesity condition, t(56) ⫽ 6.27,
p ⬍ .0001, d ⫽ 1.68, and the wheelchair condition, t(58) ⫽ 6.57,
p ⬍ .0001, d ⫽ 1.73, relative to the neutral story condition, and no
difference between the obesity and wheelchair conditions, t(58) ⫽
.48, p ⬎ .10. (The effect size f was described in Rosenthal and
Rosnow, 1991, and is commonly used for ANOVAs to index the
magnitude of an effect independent of sample size. As recom-
mended by Cohen, 1988, a magnitude between .1 and .25 reflects
a small effect, between .25 and .4 reflects a medium effect, and
above .4 reflects a large effect.) On the 7-point Likert scale (with 7
indicating high empathy), the average item scores for the condi-
tions were as follows: obesity (M ⫽ 5.53, SD ⫽ .87), wheelchair
(M ⫽ 5.64, SD ⫽ 0.93), and neutral (M ⫽ 4.30, SD ⫽ 0.61).
Pictures were used in some tasks to explore the generalizability of the
implicit anti-fat findings and to encourage participants to focus on over-
weight and underweight people rather than simply on overweight or
underweight as a concept. All pictures were full-body, realistic computer-
animated models, developed from the Landsend.com Web site, where they
are used to model clothing. Models were created that varied in hair and skin
color, but all were dressed in identical black jumpsuits. For each variation
of hair and skin model, matching overweight and underweight counterparts
were developed so that the only difference in appearance was body size.
IMPLICIT ANTI-FAT BIAS
We used t tests to evaluate anti-fat/pro-thin bias on the IATs, on
the basis of whether the relative response time was significantly
different from zero. Main effects for all five IAT tasks confirmed
our hypothesis of implicit bias: Fat/Bad Pictures IAT,
t(88) ⫽ 2.45, p ⫽ .02, d ⫽ .52, Fat/Bad Words IAT, t(88) ⫽ 4.40,
p ⬍ .0001, d ⫽ .94, Overweight/Bad IAT, t(87) ⫽ 3.38, p ⫽ .001,
d ⫽ .72, Fat/Worthless IAT, t(88) ⫽ 3.15, p ⫽ .002, d ⫽ .67, and
Others/Fat IAT, t(88) ⫽ 6.55, p ⬍ .0001, d ⫽ 1.40.
We conducted t tests to determine whether the explicit bias
difference scores were significantly different from zero. Despite
the evidence of strong implicit anti-fat/pro-thin bias, there was no
evidence of explicit fat–thin bias on the Bad/Good scale,
t(88) ⫽ 1.43, p ⬎ .10, d ⫽ .30, on the Worthless/Valuable scale,
t(87) ⫽ 1.25, p ⬎ .10, d ⫽ .27, or on the explicit evaluations of
others as fat relative to the self, t(87) ⫽ .43, p ⬎ .10, d ⫽ .09.
Interestingly, the only evidence of explicit bias emerged when
attitudes toward fat people as good versus bad were compared
relative to attitudes toward normal weight people, t(87) ⫽ 7.14,
p ⬍ .0001, d ⫽ 1.53, with means indicating more positive evalu-
ations of normal weight people (M ⫽ 5.01, SD ⫽ 1.21) than either
thin people (M ⫽ 4.18, SD ⫽ 1.12) or fat people (M ⫽ 4.01,
SD ⫽ 1.04), with 4 indicating the neutral point on the scale.
When collapsing across conditions and looking at the full sam-
ple, a pattern of small positive relationships between the various
measures of implicit bias was found (though correlations ranged
from r ⫽ .04 to r ⫽ .39, indicating wide variability), with the
expected significant positive relations among the three implicit
measures of attitude (Fat/Bad Words and Pictures IATs, r ⫽ .24,
p ⫽ .03; Overweight/Bad IAT with the Fat/Bad Words IAT, r ⫽
.30, p ⫽ .005, and with the Fat/Bad Pictures IAT, r ⫽ .37, p ⫽
To look at the relationship between implicit and explicit anti-
fat/pro-thin biases, we created composite measures of implicit and
explicit biases to simplify analyses given the large number of
variables and to increase power and reliability of the measures.
The five IAT tasks were averaged to create the total implicit bias
measure, and the three comparable explicit difference scores were
averaged to create the total explicit bias measure (the scale com-
paring evaluations of fat people relative to normal weight people
was excluded given the incongruent relative comparison to the
There was a moderate positive correlation between
the total implicit and explicit bias measures (r ⫽ .36, p ⫽ .0004).
Interestingly, this relationship was driven by the strong correlation
between the implicit and self-reported biases in the obesity con-
dition (r ⫽ .52, p ⫽ .004), as the implicit and explicit measures
were not significantly related in the other conditions (wheelchair:
r ⫽ .28, p ⬎ 1.0; neutral: r ⫽ .11, p ⬎ 1.0). BMI was strongly
related to explicit ratings of the self as fat relative to others (r ⫽
.55, p ⬍ .0001) but not to implicit ratings (r ⫽⫺.14, p ⬎ 1.0).
Influence of Manipulating Empathy on Anti-fat Bias
ANOVAs conducted for the total bias measures (with prime
condition as a three-level between-subjects variable) indicated no
significant differences on either the total implicit bias measure,
F(2, 86) ⫽ 1.36, p ⬎ .10, or the total explicit bias measure, F(2,
86) ⫽ .41, p ⬎ .10.
However, when looking at the bias as a
function of weight status, the empathy manipulation had been
effective for overweight participants (based on the standard cutoff
of a BMI over 25), with lower bias expressed when empathy was
evoked. Specifically, a significant interaction was found across the
prime conditions, with BMI as a two-level between-subjects vari-
able (BMI ⬎ or ⬍ 25), on the total implicit bias measure, F(2,
86) ⫽ 6.63, p ⫽ .002, f ⫽ .40.
This interaction is depicted in Figure 2, with the means and
standard error bars for each weight group and prime condition
indicated. Follow-up tests looking at each weight group separately
indicated that there was no difference across empathy conditions
for normal and underweight participants, but there was a difference
for overweight participants, F(2, 7) ⫽ 4.05, p ⫽ .07, f ⫽ 1.08. The
interaction with BMI for the total explicit bias measure was not
significant, F(2, 86) ⫽ 1.03, p ⬎ .10, f ⫽ .16, though the pattern
of results for the two BMI groups across empathy conditions was
similar for the implicit and explicit bias measures.
Study 2B pulled data from a larger study of implicit anti-fat bias
(see Teachman, Gapinski, & Brownell, 2002) to replicate the
findings from Study 2A that hearing about discrimination against
obese persons has a differential impact on implicit anti-fat/pro-thin
bias depending on weight status. Given the small number of
overweight women in Study 2A (typical of a college-age female
population), a sample from the general population was used to
include a larger proportion of overweight participants and a mixed-
Participants were recruited from the same location and in the same
manner as in Study 1. This sample (N ⫽ 63) was approximately equal for
gender (51% female), the mean age was 42 years (SD ⫽ 16.51, range ⫽
18–80 years), and the sample was predominantly Caucasian (79%). BMI
was 25.94 (SD ⫽ 6.25) for men and 25.64 (SD ⫽ 3.64) for women.
Materials and Procedure
Story prime. In this study, we used a novel approach to evoke empathy.
Participants read a story of severe discrimination that was modified from a
true news story (with identifying information changed for ethical reasons),
which described an accomplished, obese young woman who was sent to
“fat camp,” where she died after being verbally abused and forced to
exercise excessively in the hot sun.
The views of self and others as fat were included with the attitude and
stereotype measures given the expected consistency between views of
one’s own identity and views of out-groups.
Interestingly, when the wheelchair and neutral conditions were com
pared, there was a marginally significant trend indicating lower implicit
anti-fat/pro-thin bias when empathy toward a disabled person was evoked,
t(58) ⫽ 1.69, p ⬍ .10, d ⫽ .75, suggesting that evoking empathy toward
one marginalized group effectively minimized implicit bias to a different
TEACHMAN, GAPINSKI, BROWNELL, RAWLINS, AND JEYARAM
Implicit bias. Study 2B also extended the evidence of implicit anti-fat
bias to another stereotype domain. The lazy–motivated IAT task was
included along with an IAT that paired the categories fat people and thin
people with the categories stupid and smart (using stimuli like dumb and
intelligent). This task was selected because explicit stereotypes about obese
persons’ incompetence have been documented, but this domain has no
grounding in health outcomes. The design, administration, and method of
scoring of these IAT tasks were identical to those in Study 1 (further details
about this sample, materials used, or procedures are available from Be-
thany A. Teachman on request).
Once again, t test results indicated strong implicit anti-fat/pro-
thin associations with lazy relative to motivated, t(49) ⫽ 6.39, p ⬍
.0001, d ⫽ 1.83, and similar support was found for the smart–
stupid stereotype, t(49) ⫽ 5.27, p ⬍ .0001, d ⫽ 1.51. To look at
the influence of the discrimination story manipulation (half of the
participants read the story of discrimination, half did not), we
conducted t tests for the implicit and explicit measures to deter-
mine whether the no-prime group was significantly different from
the discrimination prime group. Similar to Study 2A, when the
sample was analyzed as a whole, there was no evidence of a
reduction in either implicit or explicit biases following the dis-
crimination prime (all ts ⬍ 1.5, ps ⬎ 1.0). However, replicating the
findings from Study 2A, the manipulation was effective for over-
weight participants (based on the standard cutoff of a BMI over
25). A significant interaction was found indicating that overweight
participants showed less implicit bias if they had read the discrim-
ination prime, but if participants’ BMIs were below 25, the ma-
nipulation had no effect: combined IAT measure, F(1, 40) ⫽ 6.09,
p ⫽ .02, f ⫽ .40 (see Figure 2).
Across these studies, strong evidence of implicit anti-fat/pro-
thin attitudes was found, regardless of the category labels used to
portray overweight and underweight people, the use of pictorial or
word stimuli, or the form of the implicit association test used
(paper–pencil or computer-based). Implicit stereotypes of fat peo-
ple (relative to thin people) as lazy versus motivated, stupid versus
smart, and worthless versus valuable were demonstrated, further
establishing the pervasiveness of the implicit bias. These stereo-
types, which include attributes clearly unrelated to health, were
exhibited in both the general population and among college stu-
dents. The consistency of the implicit bias across the different
samples and methodologies stood in contrast to the lack of explicit
anti-fat/pro-thin bias. Interestingly, the only evidence of self-
reported anti-fat attitudes emerged when evaluations of fat people
as good versus bad were compared relative to evaluations of
normal weight people. These studies support the pervasiveness of
implicit bias against overweight people and the unique demonstra-
tions of bias that can be learned from implicit measures. Given the
relative nature of the measures used, the results can be interpreted
as both anti-fat and pro-thin biases. On the basis of popular
messages in the media that promote fat jokes and the idolization of
thin models, it seems likely that both pressures are active in
Across the studies, a surprising pattern of results emerged from
the different approaches to modifying anti-fat/pro-thin biases. Al-
though providing information to participants that obesity is pre-
dominantly caused by behavioral factors, such as overeating and
lack of exercise, led to higher bias (compared with the other
groups), giving comparable information that obesity is mainly due
to genetics did not result in lower implicit or explicit biases. Our
attempts to evoke empathy through stories of discrimination
against an overweight young woman did not produce lower bias
across the whole sample, further demonstrating the durability of
the anti-fat judgments. However, across two studies, the manipu-
lation did lead to lower implicit bias for overweight participants.
This may be important given that self-blame and internalizing of
negative social messages are common in overweight people.
Figure 2. Interaction between body mass index (BMI) and empathy
prime conditions for implicit anti-fat/pro-thin bias in Studies 2A and 2B. A:
Interaction between BMI and empathy condition (obesity–wheelchair–
neutral) for implicit anti-fat/pro-thin bias. Zero point indicates no differ-
ence in response time across Implicit Association Test (IAT) conditions,
whereas positive numbers reflect an anti-fat/pro-thin bias and negative
numbers reflect an anti-thin/pro-fat bias. B: Interaction between BMI and
empathy condition (discrimination story–no prime) for implicit anti-fat/
pro-thin bias. Zero point indicates no difference in number of correctly
classified items across IAT conditions, whereas positive numbers reflect an
anti-fat/pro-thin bias and negative numbers would reflect an anti-thin/pro-
IMPLICIT ANTI-FAT BIAS
In general, variable relations were found between implicit and
explicit bias measures, though across two studies in this article
(and in Teachman & Brownell, 2001), there was a tendency to find
small positive correlations between stereotypes of implicit and
explicit beliefs about fat people being relatively lazy. It may be
more socially acceptable to acknowledge beliefs that overweight
people are lazy (given cultural values that obese people are re-
sponsible for their condition) than it is to directly evaluate an
overweight person as bad. It is interesting to note that in Study 2A,
there was a strong relationship between implicit and explicit biases
when people were primed with empathy toward an obese person,
compared with the other prime conditions. Reading about obesity
may lead participants to think about their own views of obese
persons, and this elaborative process may encourage more consis-
tency across implicit and self-report measures (support for this link
between elaboration and attitude consistency comes from Nosek,
Perceived Controllability as a Bias Reduction Strategy
The finding in Study 1 that giving participants information
about the behavioral causes of obesity led to higher bias (than the
other groups), but giving participants information about genetic
contributions did not lead to lower bias is consistent with Bell and
Morgan (2000). However, the finding differs from other studies
that found an obese person was viewed less negatively when
participants were informed that the obesity was due to a medical
condition, such as a thyroid problem (e.g., DeJong, 1980, 1993;
Rodin, Price, Sanchez, & McElligot, 1989; Weiner, Perry, &
Magnusson, 1988). One possible explanation is that our research
study manipulation was more credible when the behavioral causes,
rather than the genetic causes, were cited, even though the manip-
ulation check indicated overall differences across groups. The
absence of a difference between the no-prime control group and
the genetics prime group on the manipulation check supports this
idea, and Chlouverakis (1975) has also found that many people
consider obesity to be caused by overeating, despite expert opinion
about the role of genetic factors. Consequently, participants’
strong prior beliefs may have led them to discount the research
study. Future work that uses a more extensive manipulation em-
phasizing the role of genetics and biology in weight control could
address this issue.
An alternative explanation is that the studies cited above noted
a very specific disease that caused the obesity rather than just
referring to more general biological or genetic factors. This greater
causal specificity may have been more credible. Further, our study
asked people to make judgments toward obese persons as a group
rather than to a particular obese individual (as in many of the
previous studies). It may be easier to accept that a given person is
blameless for his or her condition but more threatening to one’s
worldview to believe that obesity in general is blameless.
The fact that negative attitudes toward obese persons could be
so easily exacerbated suggests that fat jokes, teasing, derogatory
portrayals of obese persons in the media, and other social phe-
nomena might intensify bias. Greenberg, Eastin, Hofshire, Lachlin,
and Brownell (2001), for instance, found that overweight charac-
ters on entertainment television were imbued with many more
negative characteristics than were other characters and were less
likely to be shown in positive interactions (e.g., romantic relation-
ships). Given the results of Study 1, one might argue that negative
portrayals, particularly those that support blaming attributions,
make anti-fat attitudes worse.
Empathy as a Bias Reduction Strategy
The finding across two studies, that evoking empathy did not
result in lower bias in general (compared with a control group) but
did diminish bias for overweight participants, is intriguing. These
results suggest that it may be premature to propose that “there is
simply no evidence to suggest that fat people display an in-group
bias when it comes to expressing prejudice about fat people”
(Crandall, 1994, p. 890). Rather, it may be that although both
overweight and normal weight people endorse approximately the
same amounts of anti-fat dislike in general (Crandall, 1994), there
is an in-group bias such that overweight persons are more respon-
sive to contextual factors that can mitigate anti-fat views. Remind-
ing overweight persons about anti-fat discrimination may be one
such factor that would promote in-group support, perhaps because
obese people tend to view their status in the group as temporary
(Quinn & Crocker, 1998). Thus, without the reminder of the shared
prejudice obese persons have experienced, overweight people may
be less inclined to develop the group consciousness that could lead
to an in-group bias.
It is also possible that the empathy stories did not produce lower
bias for the full sample because in addition to evoking empathy,
the stories also reminded participants of the negative stereotypes
associated with obesity. There is evidence that the strength of
implicit bias toward a stigmatized out-group is increased after
viewing images of an out-group member (relative to viewing the
face of an in-group member; Dovidio et al., 1997; Fazio, Jackson,
Dunton, & Williams, 1995). In the present study, participants read
stories describing negative evaluations of an obese person. It is
possible that this strengthened previously established anti-fat bias,
which countered the empathic feelings produced by the manipu-
lation. Future research might investigate the effects of priming
participants with positive attitudes and stereotypes toward obese
persons so that new pro-fat associations can be formed, allowing
for reconstruction of the current negative biases. This approach has
shown promise in the area of race bias, in which priming partici-
pants with positive exemplars of African Americans led to a
reduction in bias (Dasgupta & Greenwald, 2001).
The current line of studies firmly establishes the presence of
implicit anti-fat/pro-thin biases, but many open questions remain
about what factors are needed to effectively modify these biases.
These studies are limited by the cross-sectional design and brevity
of the manipulations used relative to the pervasive anti-fat mes-
sages in our culture, making it difficult to determine whether null
results occurred because of the weakness of the prime or because
the intervention does not reduce bias. Also, it is possible that the
location of Studies 1 and 2B (a beach) is a unique environment that
may actually strengthen anti-fat attitudes, making it more difficult
for our primes to be effective. The revealing clothing typically
worn at the beach may intensify body focus, causing more negative
feelings about fat in general. However, the similar findings for
Study 2A, which was conducted in the laboratory, suggest the
setting played a limited role in the results. A further limitation of
TEACHMAN, GAPINSKI, BROWNELL, RAWLINS, AND JEYARAM
these studies is that our samples are relatively homogenous (all
data were collected in Connecticut and participants were pre-
dominantly Caucasian), potentially limiting the generalizability of
our results. It would be interesting to examine implicit anti-fat
bias among people of different ethnic and racial backgrounds,
given findings of different levels of explicit anti-fat bias across
The strength of the implicit biases observed in these studies is
alarming, as is their resistance to modification. The effect sizes
found in this study are at least as strong as, and often stronger than,
the effects for implicit biases found against other stigmatized
groups (e.g., race bias: ds ⫽ .71 and .88; age bias: ds ⫽ 1.42 and
.99; and gender stereotypes: d ⫽ .72; Nosek, Banaji, & Greenwald,
2002). Further, the relative absence of self-reported anti-fat views
indicates to us the importance of alternative methods of assessing
the anti-fat biases that are clearly so insidious and pervasive in our
culture. These studies establish implicit anti-fat stereotypes across
a range of domains that have not previously been explored, dem-
onstrate the implicit bias in the general population for the first
time, and represent some of the first attempts to alter implicit
anti-fat bias. Although the strength of the implicit biases in these
studies was disappointing, we remain hopeful that these implicit
anti-fat biases can be effectively reduced. Change in this domain
may be even more challenging than changing other stereotyped
beliefs, given the relative acceptability of anti-fat attitudes (e.g.,
Kilbourne, 1994), meaning that not everyone is motivated to
change these beliefs. This may help us understand why overweight
people in our studies showed reduced implicit bias given their
likely motivation to control negative beliefs against their own
in-group. This finding is hopeful because of the psychological and
health consequences that follow the experience of discrimination
from others and self-blame. The challenge is set to continue
exposing implicit biases and to identify ways to counter them to
ultimately promote equitable treatment of our increasingly over-
Batson, C. D. (1991). The altruism question: Toward a social–
psychological answer. Hillsdale, NJ: Erlbaum.
Batson, C. D., Klein, T. R., Highberger, L., & Shaw, L. L. (1995).
Immorality from empathy-induced altruism: When compassion and jus-
tice conflict. Journal of Personality and Social Psychology, 68, 1042–
Batson, C. D., Polycarpou, M. P., Harmon-Jones, E., Imhoff, H. J., Mitch-
ener, E. C., Bednar, L. L., et al. (1997). Empathy and attitudes: Can
feeling for a member of a stigmatized group improve feelings toward the
group? Journal of Personality and Social Psychology, 72, 105–118.
Bell, S. K., & Morgan, S. B. (2000). Children’s attitudes and behavioral
intentions toward a peer presented as obese: Does a medical explanation
for the obesity make a difference? Journal of Pediatric Psychology, 25,
Bessenoff, G. R., & Sherman, J. W. (2000). Automatic and controlled
components of prejudice toward fat people: Evaluation versus stereotype
activation. Social Cognition, 18, 329–353.
Carpenter, K. M., Hasin, D. S., Allison, D. B., & Faith, M. S. (2000).
Relationships between obesity and DSM–IV major depressive disorder,
suicide ideation, and suicide attempts: Results from a general population
study. American Journal of Public Health, 90, 251–257.
Chlouverakis, C. S. (1975). Controversies in medicine (II)—Nature and
nurture in human obesity. In B. Q. Hagen (Ed.), Overweight and obesity:
Causes, fallacies, treatment (pp. 36–39). Provo, UT: Brigham Young
Cohen, J. (1988). Statistical power analysis for the behavioral sciences
(2nd ed.). Hillsdale, NJ: Erlbaum.
Crandall, C. S. (1994). Prejudice against fat people: Ideology and self-
interest. Journal of Personality and Social Psychology, 66, 882–894.
Crandall, C. S., & Biernat, M. (1990). The ideology of anti-fat attitudes.
Journal of Applied Social Psychology, 20, 227–243.
Dasgupta, N., & Greenwald, A. G. (2001). On the malleability of automatic
attitudes: Combating automatic prejudice with images of admired and
disliked individuals. Journal of Personality and Social Psychology, 81,
Dasgupta, N., McGhee, D. E., Greenwald, A. G., & Banaji, M. R. (2000).
Automatic preference for White Americans: Eliminating the familiarity
explanation. Journal of Experimental Social Psychology, 36, 316–328.
DeJong, W. (1980). The stigma of obesity: The consequences of naive
assumptions concerning the causes of physical deviance. Journal of
Health and Social Behavior, 21, 75–87.
DeJong, W. (1993). Obesity as a characterological stigma: The issue of
responsibility and judgments of task performance. Psychological Re-
ports, 73, 963–970.
Dovidio, J., Kawakami, K., Johnson, C., Johnson, B., & Howard, A.
(1997). On the nature of prejudice: Implicit and controlled processes.
Journal of Experimental Social Psychology, 33, 510–540.
Fazio, R. H., Jackson, J. R., Dunton, B. C., & Williams, C. J. (1995).
Variability in automatic and controlled components of racial prejudice.
Journal of Experimental Social Psychology, 33, 451–470.
Forster, P. (1993). The forty something barrier: Medicine and age discrim-
ination. British Medical Journal, 306, 637–639.
Greenberg, B. S., Eastin, M., Hofshire, L., Lachlin, K., & Brownell, K. D.
(2001, October). The portrayal of overweight and obese persons in
commercial television. Paper presented at the meeting of the North
American Association for the Study of Obesity, Quebec, Canada.
Greenwald, A. G., & Banaji, M. (1995). Implicit social cognition: Atti-
tudes, self-esteem, and stereotypes. Psychological Review, 105, 4–27.
Greenwald, A. G., McGhee, D. E., & Schwartz, J. L. K. (1998). Measuring
individual differences in implicit cognition: The implicit association test.
Journal of Personality and Social Psychology, 74, 1464–1480.
Guyll, M., Matthews, K. A., & Bromberger, J. T. (2001). Discrimination
and unfair treatment: Relationship to cardiovascular reactivity among
African American and European American women. Health Psychol-
ogy, 20, 315–325.
Kilbourne, J. (1994). Still killing us softly: Advertising and the obsession
with thinness. In P. Fallon, M. A. Katzman, & S. C. Wooley (Eds.),
Feminist perspectives on eating disorders (pp. 395–418). New York:
Krieger, N. (1999). Embodying inequality: A review of concepts, mea-
sures, and methods for studying health consequences of discrimination.
International Journal of Health Services, 29, 295–352.
Nosek, B. A. (2002). Moderators of the relationship between implicit and
explicit attitudes. Unpublished doctoral dissertation, Yale University.
Nosek, B. A., Banaji, M. R., & Greenwald, A. G. (2002). Harvesting
implicit group attitudes and beliefs from a demonstration website. Group
Dynamics, 6, 101–115.
Nosek, B. A., & Lane, K. (1999). Analyzing paper–pencil IAT data.
Unpublished manuscript, Yale University.
Puhl, R., & Brownell, K. D. (2001). Bias, discrimination, and obesity.
Obesity Research, 9, 788–805.
Quinn, D. M., & Crocker, J. (1998). When ideology hurts: Effects of belief
in the Protestant ethic and feeling overweight on the psychological
well-being of women. Journal of Personality and Social Psychology, 77,
IMPLICIT ANTI-FAT BIAS
Robinson, B. E., Bacon, J. G., & O’Reilly, J. (1993). Fat phobia: Measur-
ing, understanding, and changing anti-fat attitudes. International Jour-
nal of Eating Disorders, 14, 467–480.
Rodin, M., Price, J., Sanchez, F., & McElligot, S. (1989). Derogation,
exclusion, and unfair treatment of persons with social flaws: Controlla-
bility of stigma and the attribution of prejudice. Personality and Social
Psychology Bulletin, 15, 439–451.
Rosenthal, R., & Rosnow, R. L. (1991). Essentials of behavioral research:
Methods and data analysis (2nd ed.). New York: McGraw-Hill.
Rudman, L., Greenwald, A. G., & McGhee, D. E. (1996, October). Pow-
erful women, warm men? Implicit associations among gender, potency,
and nurturance. Paper presented at the meeting of the Society of
Experimental Social Psychology, Sturbridge, MA.
Teachman, B., & Brownell, K. (2001). Implicit associations toward obese
people among treatment specialists: Is anyone immune? International
Journal of Obesity, 25, 1–7.
Teachman, B., Gapinski, K., & Brownell, K. (2002, February). Implicit
anti-fat/pro-thin stereotypes: Not just a matter of health. Poster session
presented at the third annual meeting of the Society for Personality and
Social Psychology, Savannah, GA.
Teachman, B., Gregg, A., & Woody, S. (2001). Implicit attitudes toward
fear-relevant stimuli in individuals with snake and spider fears. Journal
of Abnormal Psychology, 110, 226–235.
Weiner, B., Perry, R. P., & Magnusson, J. (1988). An attributional analysis
of reactions to stigmas. Journal of Personality and Social Psychol-
ogy, 55, 738–748.
Wiese, H. J. C., Wilson, J. F., Jones, R. A., & Neises, M. (1992). Obesity
stigma reduction in medical students. International Journal of Obe-
sity, 16, 859–868.
Williams, D. R. (1999). Race, socioeconomic status, and health: The added
effects of racism and discrimination. Annals of the New York Academy
of Sciences, 896, 173–188.
TEACHMAN, GAPINSKI, BROWNELL, RAWLINS, AND JEYARAM