The Influence of Sociocultural Factors on Body Image:

A Meta-Analysis
Guy Cafri, Yuko Yamamiya, Michael Brannick, and J. Kevin Thompson, University of South Florida

Various factors have been implicated in the development The impact of a ‘‘thin ideal’’ of feminine beauty on body
of body image dissatisfaction. Especially important are image has received significant research attention because
three constructs: awareness of a thin ideal, internalization of its pervasive influence on people in Western cultures
of a thin ideal, and perceived pressures to be thin. Using and its central role in eating disorders (Thompson,
meta-analysis, we calculated the strength of the relation- Heinberg, Altabe & Tantleff-Dunn, 1999). An important
ships between each of these constructs and body image,
area of research that has emerged in recent years is the
assessment of factors that contribute to the development
and we evaluated the differences in magnitude across the
of body image problems. Sociocultural factors have re-
average effect sizes. We also tested the moderating
ceived the most attention in this regard because inter-
effects of age and ethnicity, and we compared the average
personal and media influences are widely regarded as the
effect sizes with those from meta-analyses of prospective
source from which body image attitudes emanate (e.g.,
and experimental studies in order to determine whether Bordo, 1993; Fallon, 1990; Striegel-Moore, Silberstein, &
the effect sizes differed by study design. The results Rodin, 1986). In fact, empirical evidence supports socio-
indicated the following: all three sociocultural factors had cultural variables as important risk factors for the
statistically significant relationships with body image; development of body image dissatisfaction, which is in
internalization and perceived pressures have a signifi- turn a casual risk factor for eating pathology (Stice, 2002).
cantly stronger relationship to body image than does Three constructs related to perceived influence of
awareness; the effect sizes from cross-sectional studies social and cultural factors have received particular
were significantly larger than those of both longitudinal attention with respect to their relationship with body
and experimental studies; and neither age nor ethnicity image attitudes: awareness of a thin ideal in the media,
was a statistically significant moderator of the relation-
internalization of a thin ideal, and perceived pressures to
be thin (e.g., Stice 2002; Thompson & Stice, 2001). The
ship between awareness and body image or that between
relationships that we assessed in this meta-analysis are
internalization and body image. In this article, we discuss
between these three sociocultural constructs and body
the limitations and implications of the findings on future
image. Not only did we determine the strength of each
research, theory, and clinical application.
relationship and whether it was significantly different
Key words: internalization, media, sociocultural, body from zero, but we also compared the magnitudes of the
image, body dissatisfaction, eating disorder, eating pa- obtained effect sizes relative to one another, determined
thology, awareness, pressures. [Clin Psychol Sci Prac 12: if age and ethnicity were moderators of these effect sizes,
421–433, 2005] and examined whether the effect sizes were smaller or
larger than those obtained in experimental and pro-
Address correspondence to J. Kevin Thompson, Department spective studies.
of Psychology, University of South Florida, Tampa, FL 33620- Historically, ideals of beauty have often fluctuated.
8200. E-mail: [email protected]. For instance, in the period between the 1400s and 1700s

doi:10.1093/clipsy/bpi053
Ó The Author 2005. Published by Oxford University Press on behalf of the American Psychological Association D12.
All rights reserved. For permissions, please e-mail: [email protected]. 421
the beauty ideal was a woman who was plump, big quently, Heinberg, Thompson, and Stormer (1995)
breasted, and generally maternal in appearance (Fallon, developed a measure based on exploratory factor
1990). This standard can be contrasted with the analytic methods that distinguished between awareness
dominant thin ideal that has been pervasive over the and internalization of sociocultural ideals, what the
past thirty years (Thompson et al., 1999). Supporting authors called the Sociocultural Attitudes Towards
the dominance of the thin ideal and its role in Appearance Questionnaire (SATAQ). Awareness of
producing and perpetuating body dissatisfaction are the thin ideal has been defined as the simple knowledge
content analyses of media such as magazines and that a standard exists, as opposed to the internalization
television, correlational investigations of the relation of the thin ideal, which is a profound incorporation or
between exposure to the thin ideal and body dissatis- acceptance of the value, to the point that the ideal
faction, and experimental studies based on thin indi- affects one’s attitudes (body image) or personal be-
viduals’ exposure to images (for reviews, see Levine & havior (dieting) (Thompson et al., 1999; Thompson &
Harrison, 2004; Thompson et al., 1999). Stice, 2001). The distinction between awareness and
The use of randomized experiments to examine the internalization is apparent if one compares a typical
effect of media images on psychological processes has awareness item, ‘‘People think that the thinner you are,
increased dramatically in recent years. In these studies the better you look in clothes,’’ to an internalization
participants are typically presented with various media item, ‘‘I wish I looked like a swimsuit model’’
images of thin bodies that are contrasted with individ- (Heinberg et al., 1995). Results of exploratory factor
uals of ‘‘average’’ weight. The results of numerous such analyses support the distinction between awareness and
experiments have led to the robust finding that exposure internalization (Heinberg et al. 1995; Thompson, van
to thin images elicits body dissatisfaction (Groesz, den Berg, Roehrig, Guarda, & Heinberg, 2004),
Levine, & Murnen, 2002). Some researchers have although the labels assigned to the factors can be
attempted to assess the relationship between level of misleading. Specifically, the evolution of the terms
media exposure and relevant variables such as body awareness and internalization within psychodynamic
image dissatisfaction and eating pathology (e.g., Stice, theory may lead to the belief that the awareness factor
Schupak-Neuberg, Shaw, & Stein, 1994). Although is simply assessing processes that are in conscious
assessment of media exposure was initially in widespread awareness whereas internalization is assessing some
use among researchers, it has generally been abandoned more latent self-schema. Given an absence of empirical
in favor of more subjective indicators of media in- methods to distinguish between these cognitive pro-
fluence, such as measures of internalizing a media cesses, one finds it necessary to abide by a strict
influence, which appear to be more directly related to interpretation of the item content within these scales—
body image dissatisfaction (Thompson et al., 1999). that is, the difference is better articulated as passive
Internalization, awareness, and perceived pressures endorsement of a thin ideal versus active cognitions/
are the three factors that have evolved as the most attitudes related to it. Although it might be more
frequently assessed sociocultural factors related to a thin appropriate to call awareness ‘‘passive endorsement of
ideal of beauty. To inform the meaning of these a thin ideal’’ and internalization ‘‘active endorsement of
constructs as well as the scales that are used to assess the thin ideal,’’ we continue to refer to these factors as
them, a review is of the studies that developed these awareness and internalization because these are the labels
constructs in order. used in the studies reviewed in this meta-analysis.
It is important first that we clarify the distinction Another important development in assessing socio-
between awareness and internalization. Initially, Stice cultural factors was examining perceived pressures to
and colleagues (1994) assessed personal stereotypes be thin. Stice, Nemeroff, and Shaw (1996) developed
associated with having a thin body using a scale called the Perceived Sociocultural Pressure Scale, which
the Ideal-Body Stereotype Internalization, which, as assessed perceived pressures to be thin from family,
the name implies, purports to measure an internaliza- friends, dating partners, and the media (e.g., ‘‘I’ve
tion component of sociocultural influence. Subse- perceived a strong message from my family to have

CLINICAL PSYCHOLOGY: SCIENCE AND PRACTICE  V12 N4, WINTER 2005 422
a slender figure’’). Thompson and colleagues have (2004) argued that the IBSS–R was a measure of
developed similar scales that assess pressures, which can awareness because when items were factor-analyzed
be found in various adaptations of the original from both the IBSS–R and previous versions of the
SATAQ; however, these scales focus exclusively on SATAQ, almost all the IBSS–R items loaded on an
media pressures (e.g., Cusumano & Thompson, 2000; awareness factor. The concern related to whether the
Thompson et al., 2004). IBSS–R assessed awareness or internalization was driven
The methods used in research to evaluate socio- by the results of several studies showing that in-
cultural influence have advanced in two different direc- ternalization has a stronger association and is a better
tions: one based on the original Ideal-Body Stereotype predictor of body image dissatisfaction (e.g., Heinberg
Internalization and Perceived Sociocultural Pressure et al., 1995).
scales or their modifications and the other using the A central assumption in this discussion and area of
original SATAQ or its adaptations. The initial Ideal- research is that internalization is more strongly related to
Body Stereotype Internalization Scale, developed by body image than is awareness. This hypothesis is
Stice et al. (1994), underwent a revision in an effort to effectively testable by comparing average effect sizes
make the items subtler, resulting in what is now called obtained using meta-analytic methods. The importance
the Ideal-Body Stereotype Scale–Revised (IBSS–R; of testing this hypothesis is multifaceted. The results
Stice, Ziemba, Margolis, & Flick, 1996), which has been would inform variable selection in research, provide
used in several subsequent studies (e.g., Stice & Agras, empirical support for a particular sociocultural model of
1998).1 The original Perceived Sociocultural Pressure influence, and inform the design of prevention and
Scale has remained the same and has been used in sev- intervention strategies. Therefore, a central aim was to
eral studies (e.g., Stice, 2001). The SATAQ and its compare the effect sizes obtained from the relationship
adaptations have a slightly more complicated history. between internalization and body image and that
Shortly after the development of the original SATAQ, between awareness and body image. Another goal of
a revision was developed that modified and updated this meta-analysis was obtaining an average effect size
the items (SATAQ–R; Cusumano & Thompson, 1997). for the relationship between perceived pressures and
Thereafter, a scale was developed specifically for ado- body image and comparing this value to the effects
lescents, named the Multidimensional Media Influence found for internalization and awareness, given that this
Scale (MMIS), consisting of subscales that measured factor has not received as much research attention.
awareness, internalization, and pressures (Cusumano & Moreover, previous research suggests that age and
Thompson, 2000). The Sociocultural Internalization ethnicity are important demographic characteris-
of Appearance Questionnaire–Adolescents (SIAQ–A) tics related to body image, with ethnic minorities and
is another scale that was developed, but it exclu- those of older ages least strongly influenced by the mass
sively targets an internalization component of appear- media (Cashel, Cunningham, Landeros, Cokley, &
ance among adolescents (Keery, Shroff, Thompson, Muhammad, 2003; Ewell, Smith, Karmel, & Hart,
Wertheim, & Smolak, 2004). Finally, a scale called the 1996; Warren, Gleaves, Cepeda-Benito, Fernandez,
SATAQ–3 was developed, which assesses internalization Rodiguez-Ruiz, in press). Therefore, we hypothesized
(thin ideal and athleticism), pressures, and information that age and ethnic composition of the sample would
(the awareness subscale was dropped from the final moderate the relationship between sociocultural factors
version; Thompson et al., 2004). and body image such that samples with lower
An important question that arose in the last few years percentages of ethnic minorities and older ages would
regarding assessment of sociocultural factors is whether exhibit smaller associations between sociocultural factors
the IBSS–R assesses internalization or awareness. The and body image. Finally, we were interested in
reason for raising this point is that, on face value, the determining whether the effect sizes in this meta-analysis
items of the IBSS–R seem to target an awareness were larger or smaller than those of meta-analyses
construct (e.g., ‘‘Slender women are more attractive’’; focused on prospective and experimental studies in this
Stice, Ziemba, et al., 1996). Thompson and colleagues subject area.

SOCIOCULTURAL FACTORS  CAFRI ET AL. 423

ME THOD tency or test–retest reliability equal to or exceeding
Data Collection .7 (Nunnally, 1978). For instance, the Eating Disorders
To locate studies for the present meta-analysis, we Inventory–Body Dissatisfaction subscale (EDI–BD;
employed several procedures. First, we conducted Garner & Olmsted, 1984) met our criteria for docu-
a computer-based search on PsychInfo and Medline, mented evidence of reliability and validity. With respect
using the key words internalization, media, sociocultural, to measurements of the three sociocultural factors of
SATAQ, body image, body dissatisfaction, eating disorder, interest (viz., internalization, awareness, and perceived
eating pathology, IBSS, MMIS, awareness, and pressures. pressures), we used a strict inclusion criterion. For inter-
Second, we performed a manual search of the table of nalization, measures were limited to the appropriate
contents of two journals that frequently publish articles subscale on the SATAQ (Heinberg et al., 1995) or one
in this area—International Journal of Eating Disorders and of its many adaptations (e.g., SIAQ–A; Keery et al.,
Eating Disorders: Journal of Treatment and Prevention— 2004). For awareness, the measures included the aware-
from 1990 to the present. Third, we examined the ness subscale from the SATAQ or an adaptation, or the
reference sections of relevant articles. Last, we contacted IBSS–R (Stice, Ziemba, et al., 1996). Measures for per-
several researchers who were well known in this area of ceived pressures were limited to the appropriate subscale
study and requested copies of unpublished data. of the SATAQ or an adaptation, or the Perceived Socio-
cultural Pressure Scale (Stice, Nemeroff, et al., 1996).
Inclusion/Exclusion Criteria Finally, we decided to include only correlations ob-
We included only the studies that were written in tained from cross-sectional data. The decision to limit
English, though we did not restrict countries in which our data to cross-sectional designs was based on the fact
studies were conducted, resulting in the inclusion of that most studies in this area were cross-sectional.
samples from eight countries (United States, England, Moreover, the relatively few studies that used longitu-
New Zealand, Australia, India, Italy, Spain, and dinal designs would probably have cross-sectional data,
Poland). Ages of the participants and years of the whereas the converse would not be true. Data taken from
publication were not restricted in the study either. We the initial period of prospective studies were included.
did however choose to exclude studies that used Correlation coefficients (r) had to be reported in
specialized samples (e.g., eating disordered and cosmetic a study. When the correlation coefficients were not
surgery patients) because they would likely exhibit provided, the primary investigator of the study was
either direct or indirect range restriction on the contacted, and the appropriate correlations were re-
examined variables, which would reduce the size of quested. To ensure independence, we included (a) only
the examined associations (e.g., Calogero, Davis, & the findings of one article if multiple articles reported
Thompson, 2004; Didie & Sarwer, 2003). Moreover, findings from the same sample and (b) only a single
only data on female participants were included because effect size to characterize the relationship between two
methods of assessing male-appropriate sociocultural constructs in any given article. We were able to locate
influence and body image (i.e., measures assessing a total of 22 studies that assessed the relationship between
a muscularity dimension) have not been widely used the sociocultural factors and body image, with 31 effect
(Cafri & Thompson, 2004). sizes for internalization, 25 effect sizes for awareness, and
We included only studies that used measures of body 7 effect sizes for perceived pressures (some studies
image and sociocultural influence that have demon- included multiple samples; see Table 1).
strated adequate validity and reliability. For a scale to
demonstrate adequate validity, it had to be created via Coding Studies
factor analysis and/or exhibit appropriate convergent or The first and second authors coded 15 of the 22 studies
discriminant validity (i.e., statistically significant corre- independently (7 studies were coded by both the first
lations for convergence and nonsignificant correlations author and the second). Intraclass correlations (3) were
for divergence). With respect to evidence of reliability, used to calculate reliabilities of data entries (i.e., effect
the scale needed to demonstrate either internal consis- sizes and sample sizes) made by the first and second

CLINICAL PSYCHOLOGY: SCIENCE AND PRACTICE  V12 N4, WINTER 2005 424
Table 1. Study characteristics

Dependent Independent
N M age Ethnicity Country measure measure

Cash, Melnyk, & Hrabosky (2004) 464 20.40 C 5 58%; AA 5 26%; A 5 6%; U.S. BIQ SATAQ–3
H 5 4%; O 5 6%
Cashel, Cunningham, Landeros, 150 20.86 C 5 100% U.S. EDI–BD SATAQ
Cokley, & Muhammad (2003)–1
Cashel et al. (2003)–2 44 20.86 AA 5 100% U.S. EDI–BD SATAQ
Cashel et al. (2003)–3 31 20.86 H 5 100% U.S. EDI–BD SATAQ
Cusumano & Thompson (1997) 175 24.00 C 5 75%; AA 5 7%; H 5 10%; U.S. EDI–BD SATAQ–R
A 5 7%; O 5 1%
Cusumano & Thompson (2000) 107 10.25 C 5 65%; AA 5 21%; H 5 1.9%; U.S. EDI–BD MMIS
A 5 1.9%; O 5 9.3%
Forbes, Doroszewicz, Card, & 111 19.60 C 5 100% Poland BCS (adapted SATAQ
Adams-Curtis (2004)–1 version)
Forbes et al. (2004)–2 85 20.30 C 5 100% U.S. BCS (adapted SATAQ
version)
Heinberg, Thompson, & 191 26.50 None given U.S. EDI–BD SATAQ
Stormer (1995)
Keery, Shroff, Thompson, 462 11.80 C 5 73%; AA 5 3%; H 5 10%; U.S. EDI–BD SIAQ–A
Wertheim, & Smolak (2004)–1 A 5 2%; NA 5 2 %; O 5 8.5%
Keery et al. (2004)–2 384 12.60 C 5 74%; AA 5 6%; H 5 11%; U.S. EDI–BD SIAQ–A
A 5 2%; NA 5 1%; O 5 6%
Keery et al. (2004)–3 433 12.60 C 5 85%; AA 5 3%; H 5 5%; U.S. EDI–BD SIAQ–A
A 5 1%; NA 5 2%; O 5 4%
Keery et al. (2004)–4 352 12.90 None given U.S. EDI–BD SIAQ–A
Keery et al. (2004)–5 484 13.30 None given Australia EDI–BD SIAQ–A
Keery et al. (2004)–6 94 11.70 None given India EDI–BD SIAQ–A
Low et al. (2003) 72 19.50 C 5 93%; AA 5 1.4%; H 5 1.4%; U.S. EDI–BD SATAQ
A 5 2.8%; O 5 1.4%
McClintok (2003) 190 15.41 Pakeha 5 74%; Maori 5 2.3%; New Zealand EDI–BD MMIS
Pacific 5 2.3%; South African 5
1.2%; A 5 6.9%; O 5 2%
Ruggiero, Hannover, 223 17.02 None given Italy (North) EDI–BD SATAQ
Mantero, & Papa (2000)–1
Ruggiero et al. (2000)–2 108 17.10 None given Italy (South) EDI–BD SATAQ
Sands & Wardle (2003) 356 10.86 None given England Collins’ (1991) SATAQ
silhouette
Sarwer et al. (in press) 555 20.50 C 5 65%; AA 5 10%; H 5 9%; U.S. MBSRQ MMIS
A 5 9%; O 5 6%
Smolak, Levine, & Thompson (2001) 175 — None given U.S. BES SATAQ
Steinberg (2004) 633–652 20.20 C 5 66%; AA 5 14%; H 5 11%; U.S. EDI–BD SATAQ–R
A 5 4%; O 5 5%
Stice (2001) 231 14.90 C 5 65%; AA 5 4%; H 5 2%; U.S. SBPS (adapted) PSPS
A 5 20%; O 5 8%
Stice & Agras (1998) 218 17.00 C 5 78%; AA 5 2%; H 5 10%; U.S. SBPS (adapted) PSPS, IBSS–R
A 5 4%; NA 5 1%; O 5 4%
Stice, Schupak-Neuberg, 238 20.00 None given U.S. EDI–BD IBSS
Shaw, & Stein (1994)
Stice & Whitenton (2002) 496 13.01 C 5 68%; AA 5 7%; H 5 18%; U.S. SBPS (adapted) PSPS
A 5 2%; NA 5 1%; O 5 4%
Thompson, van den Berg, Roehrig, 172 21.00 C 5 65%; AA 5 10%; H 5 15%; U.S. EDI–BD SATAQ–3,
Guarda, & Heinberg (2004)–1 A 5 2%; O 5 7%; IBSS–R
Thompson et al. (2004)–2 195 20.00 C 5 58%; AA 5 15%; H 5 12%; U.S. EDI–BD SATAQ–3
A 5 5%; O 5 8%;
Thompson-Leonardelli (2003)–1 131 19.20 AA 5 100% U.S. BES SATAQ
Thompson-Leonardelli (2003)–2 165 18.68 C 5 100% U.S. BES SATAQ
Twamley & Davis (1999) 249 20.00 C 5 77% U.S. BES SATAQ, IBSS–R
Warren, Gleaves, Cepeda-Benito, 100 18.90 C 5 100% U.S. BSQ SATAQ–R
Fernandez, & Rodriquez-Ruiz
(in press)–1
Warren et al. (in press)–2 100 19.80 Mexican American 5 100% U.S. BSQ SATAQ–R
Warren et al. (in press)–3 100 18.70 Spanish 5 100% Spain BSQ SATAQ–R

Ethnicity: A 5 Asian, AA 5 African American, C 5 Caucasian, H 5 Hispanic, NA 5 Native American, O 5 other. Dependent measure: BCS 5 Body Cathexis
Scale, BES 5 Body Esteem Scale, BIQ 5 Body-Image Ideals Questionnaire, BSQ 5 Body Shape Questionnaire, EDI–BD 5 Eating Disorders Inventory–
Body Dissatisfaction, MBSRQ 5 Multidimensional Body Self Relations Questionnaire, SBPS 5 Satisfaction with Body Parts Scale. Independent measure:
IBSS–R 5 Ideal-Body Stereotype Scale–Revised, MMIS 5 Multidimensional Media Influence Scale, PSPS 5 Perceived Sociocultural Pressures Scale, SATAQ 5
Sociocultural Attitudes Towards Appearance Questionnaire, SATAQ–R 5 Sociocultural Attitudes Towards Appearance Questionnaire–Revised, SATAQ–3 5
Sociocultural Attitudes Towards Appearance Questionnaire (also measuring internalization, pressures, and information), SIAQ–A 5 Sociocultural Internalization
of Appearance Questionnaire–Adolescents.

SOCIOCULTURAL FACTORS  CAFRI ET AL. 425

authors (Shrout & Fleiss, 1979). The interrater reliabil- using only ethnicity as a moderator, and finally in-
ities ranged from 0.99 to 1.00, indicating an acceptable cluding both moderators.
agreement level between the raters. The two discrep- The computations followed recommendations by
ancies that were found were resolved through consensus. Lipsey and Wilson (2001). All correlations were trans-
The average age of participants in the studies ranged formed using a Fisher’s r to z transformation, and
from 10.25 to 26.50. inverse variance weights were used to compute a
weighted average effect size. Inverse variance weights
Data Analyses were also used in the weighted regression analyses to test
Our initial model for data analysis was a type of age as a moderator. The random-effects model was used
random-effects model called a mixed-effects model to compute mean effect sizes and confidence intervals for
(Lipsey & Wilson, 2001; Overton, 1998).2 In a mixed- each of the effect size distributions. Average effect sizes
effects model, the observed variance in effect sizes is and confidence intervals were transformed back to r
assumed to be due to sampling error, moderators using the inverse of the r to z transformation.
included in the study (e.g., age and ethnicity of A series of tests were used to evaluate whether there
participants), and unexplained variance (random-effects were statistically significant differences between aver-
variance). A sequence of tests and parameter estimates aged correlation coefficients (Chen & Popovich, 2002).
are computed in such a model. First, a study character- First, these tests were used to compare the magnitude of
istic is tested to determine if it moderates the obtained average correlation coefficients obtained in this meta-
effect sizes; more explicitly, it determines whether analysis: internalization–body image, awareness–body
systematic sources of variability influence the relation- image, and pressures–body image. Significance tests for
ship between two existing variables in such a way that the difference between dependent correlations were
effect sizes differ at different levels of the moderator. conducted because many of the correlations used to
Second, the residual variance after accounting for the calculate the average r values came from the same
moderator and sampling error is then assumed to be studies. Second, we tested whether the averaged cor-
random and is incorporated into the estimate of the relation coefficients in this study were significantly
average effect sizes (maximum likelihood estimates of different from those obtained in meta-analyses of
the random-effects variance component were used; prospective (Stice, 2002) and experimental studies
Lipsey & Wilson, 2001; Overton, 1998). The moderators (Groesz et al., 2002), which provides a test of whether
of interest in this meta-analysis were the average age and study design moderates the magnitude of obtained effect
percentage of ethnic minority participants in each study. sizes. Significance tests for the difference between
Both age and percentage of ethnic minorities were independent correlations were conducted.
treated as continuous predictors, whose effects were
tested using a weighted least-squares regression approach R E S UL T S
(Lipsey & Wilson, 2001; Steel & Kammeyer-Mueller, Internalization–Body Image
2002).3 Information regarding ethnicity could not be The distribution of effect sizes for the relationship
coded for every study because either the information in between body image and internalization consisted of 31
the study was not available or it was unclear how effect sizes (see Figure 1) from 18 studies and 7,079
ethnicity should be coded, particularly in the interna- participant responses. The results of the weighted regres-
tional samples (e.g., McClintock, 2003; Sample 3 of sion computed with age as a moderator indicated a small
Warren et al., in press). Given the absence of in- and nonsignificant effect, QB 5 1.20, p 5 .27, R2 5 .03.
formation on ethnicity, running a regression with both When percentage of ethnic minorities was used as a
moderators entered simultaneously would reduce the moderator, there was a statistically nonsignificant effect,
total number of studies included in the analysis and but a moderate proportion of variance was accounted
hence reduce the power of our statistical tests. Therefore, for, QB 5 2.15, p 5 .14, R2 5 .08. When both
each time that we tested for moderators, we ran the moderators were entered simultaneously, the result was
analysis by first using only age as a moderator, then a statistically nonsignificant effect, QB 5 2.69, p 5 .26,

CLINICAL PSYCHOLOGY: SCIENCE AND PRACTICE  V12 N4, WINTER 2005 426
Figure 1. Dot plot of internalization–body image relationships for all the studies.

R2 5 .10. Subsequently, an overall Q test was com- as a moderator indicated a small and nonsignificant
puted to determine whether there was significant effect, QB 5 .464, p 5 .50, R2 5 .02. As with the
variability among the effect sizes. The Q test was association between internalization and body image,
significant, Qtotal 5 96.88, p , .05, indicating signifi- when the percentage of ethnic minorities was used as
cant variability. Moreover, the size of the random- a moderator, there was a statistically nonsignificant
effects variance component was moderate, REVC 5 effect, but a moderate proportion of variance was
.01. The significant overall Q test and moderate size of accounted for, QB 5 1.34, p 5 .25, R2 5 .07. When
the random-effects variance component indicated that both moderators were entered simultaneously, the result
a random-effects model was appropriate. Applying a was a statistically nonsignificant effect, QB 5 1.46, p 5
random-effects model to calculate the average effect size .48, R2 5 .07. An overall Q test was computed to deter-
and confidence interval resulted in a significant average mine whether there was significant variability among
z value, z 5 .55, CI 5 .50, .59. Applying a z to r the effect sizes. The Q test was significant, Qtotal 5
transformation to the mean and confidence interval for 58.57, p , .05, suggesting significant variability. The
the result yielded r 5 .50, CI 5 .46, .53. Note that the size of the random-effects variance component was
lower bound of the confidence interval is nowhere near moderate, REVC 5 .0079. Based on the results of the
zero, indicating that the overall mean is significantly significant overall Q test and moderate random-effects
different from zero. variance component, a random-effects model was chosen.
Applying a random-effects model to calculate the
Awareness–Body Image average effect size and confidence interval resulted in
The distribution for the relationship between body a significant average z value, z 5 .30, CI 5 .26, .35, p ,
image and awareness consisted of 25 effect sizes (see .05. Applying a z to r transformation to the mean and
Figure 2) from 18 studies and 4,742 participant confidence interval for the result yielded r 5 .29, CI 5
responses. As before, the weighted regression with age .25, .34, p , .05.

SOCIOCULTURAL FACTORS  CAFRI ET AL. 427

Figure 2. Dot plot of awareness–body image relationships for all the studies.

Pressures–Body Image there were differences between the average r values

The distribution for the relationship between body for internalization–body image, awareness–body
image and internalization consisted of seven effect sizes image, and pressures–body image. The correlation for
(see Figure 3) from six studies and 1,998 participant internalization–body image (average r 5 .50) was
responses. The overall Q test was not significant, significantly larger than that of awareness–body image
Qtotal 5 11.12, p 5 .08. Although nonsignificance (average r 5 .29), t (2726) 5 7.23, p , .05, but not
usually suggests that there is not significant variability significantly different from that of pressures–body
among the effect sizes, in this case the small number of image (average r 5 .48), t (367) 5 .254, p . .05. The
studies may have limited the power of this test. The correlation for pressures–body image was significantly
random-effects variance component was also rather larger than that for awareness–body image, t (621) 5
small, REVC 5 .0032. Nevertheless, the average z value 4.19, p , .05.4
based on a random-effects model was significant, z 5
.52, CI 5 .46, .59, p , .05. Applying a z to r
transformation to the mean and confidence interval for Differences Across Study Designs
the result yielded r 5 .48, CI 5 .43, .53, p , .05. The averaged cross-sectional correlation coefficients in
this study were compared to those in other meta-
analyses of prospective and experimental studies using
Differences Between Internalization, Awareness, and Pressures significance tests for the difference between independent
Significance tests for the difference between depen- correlations. In a recent meta-analysis the average
dent correlations were conducted to determine whether prospective relationship between awareness and body

Figure 3. Dot plot of pressures–body image relationships for all the studies.

CLINICAL PSYCHOLOGY: SCIENCE AND PRACTICE  V12 N4, WINTER 2005 428
image was r 5 .18 and between pressures and body significantly larger than the association between aware-
image, r 5 .09 (Stice, 2002).5 The average r values in the ness and body image suggests that in a research context
prospective studies were significantly smaller than those internalization rather than awareness should be assessed
in the cross-sectional studies for the relationship between as a predictor of body image attitudes because it can
awareness and body image, z 5 2.92, p , .05, and potentially account for a greater proportion of variance
pressures and body image, z 5 16.68, p , .05.6 In in body image attitudes (i.e., if the decision needs to be
a recent meta-analysis of experimental studies exposing made between choosing between a measure of in-
participants to thin media images (Groesz et al., 2002), ternalization and one of awareness). Perhaps more
significant increases in body dissatisfaction with a small important are the theoretical implications, which are
average effect size were demonstrated, d 5 .31, r 5 .15 that it is not simply being aware of a thin ideal that is
(the d value was converted to the correlation coefficient r important but rather incorporating such a standard into
using a formula provided by Hunter & Schmidt, 1990). how a person thinks he or she should look. This
This average r value found in experimental studies theoretical point has clinical implications for the design
was significantly smaller than the association between of prevention and intervention programs, such as the
internalization–body image, z 5 16.56, p , .05, dissonance-based eating-disorders prevention program
awareness–body image, z 5 5.79, p , .05, and developed by Stice and colleagues (e.g., Stice, Trost, &
pressures–body image, z 5 12.14, p , .05. Chase, 2003), suggesting that they should not only at-
tend to challenging existing cognition regarding what is
D IS C U S S I O N considered ideal but perhaps focus more so on cognition
The average effect sizes for the relationship between that relates one’s self to that ideal via comparison and
internalization, awareness, perceived pressures, and setting personal body modification goals.
body image were assessed. The relationships between Another important finding of this meta-analysis is
each of the examined sociocultural factors and body that perceived pressures have a large association with
image were significantly greater than zero, with the body image, one that is comparable in magnitude to the
effects for internalization and pressures both significantly relationship between internalization and body image.
larger than those for awareness but not significantly However, the small number of effect sizes for perceived
different from those of one another. Moreover, the pressures limits our confidence that the average effect
magnitude of the difference between internalization and size calculated in this meta-analysis is representative of
awareness, as well as pressures and awareness, was the population effect size. The lack of research into
substantively meaningful, approximately the difference perceived pressures highlights the need for future studies
between a large and medium effect size according to to make a more concerted effort to incorporate measures
Cohen’s designations (1988). Neither age nor ethnicity of this construct.
was a statistically significant moderator of the relation- Although previous research investigating the in-
ship between body image and either internalization or fluence of sociocultural factors on body image using
awareness. Finally, comparing the effect sizes in this longitudinal and experimental designs provides evidence
meta-analysis with those of previous meta-analyses in for their role as causal risk factors for body image
this area suggests that cross-sectional designs demon- dissatisfaction (Groesz et al., 2002; Stice, 2002), the
strated consistently larger effects than those of both findings of this meta-analysis suggest that effect sizes
prospective and experimental studies. obtained in cross-sectional studies are, on average,
The overall finding that sociocultural factors have significantly larger than those obtained in experimental
medium-to-large associations with body image high- and longitudinal studies. This finding indicates that the
lights their clinical significance, with the differences magnitude of effects expected in experimental and
across the average effect sizes in this meta-analysis having prospective studies should be lower than those obtained
particularly important implications for research, theory, in cross-sectional studies, which is an issue that should be
and clinical practice. For instance, the finding that the addressed in future reviews of the literature as well as in
association between internalization and body image is the planning phase of research studies.

SOCIOCULTURAL FACTORS  CAFRI ET AL. 429

 It is also important that neither age nor ethnicity and awareness is moderate (r range 5 .32–.38; Stice,
was found to be a statistically significant moderator of 2001; Stice & Agras 1998; Thompson et al., 2004), as is
the internalization–body image or awareness–body the case for the relationship between awareness and
image relationship. Finding that age is not a significant internalization (average r 5 .42).7 Further studies are
moderator runs somewhat contrary to theories that needed to determine whether different sociocultural
adolescence is a critical period of development in which measures are predicting unique variance in body image
sociocultural influences are at their apex (e.g., Ewell dissatisfaction.
et al., 1996) because it would be expected that as age
increases, the association between each of the socio- N O T ES
cultural factors and body image would decrease. A 1. Both the acronyms IBIS and IBSS have been used to
limitation to consider is range restriction due to an describe the same measure, but for the sake of consistency,
absence of studies with average ages greater than 26.5, IBSS is used throughout this article.
which could have ultimately led to an attenuated 2. A critical decision in meta-analysis is model selection.
moderator effect. Another point is that age varies within Specifically, researchers generally have the option of using
either a fixed-effects model or a random-effects model. A
each sample so that the mean age used as the moderator
fixed-effects model assumes that effect sizes obtained from
at the study level may lead to an underestimate of the
studies correspond to a population effect with the only source
actual moderating influence of age at the raw data level.
of variability stemming from subject-level sampling error.
With respect to ethnicity, although statistically signifi- That is, the sample of studies obtained by the meta-analyst
cant moderating effects were not found, this finding represents a random sample from a single population. In
may have been due to low power stemming from contrast, a random-effects model assumes that not only
a small sample of studies, placing ethnic minorities into a subject-level sampling error but also some other additional
a heterogeneous group rather than conducting analyses variance that explains variability across the obtained effect sizes
separately for each ethnic group, and a loss of infor- that is random and cannot be accounted for. Oftentimes, it is
mation due to the analyses taking place at the study level not entirely clear whether a fixed-effects or random-effects
rather than the raw data level. model is more appropriate, given that the variety of
The results of this meta-analysis suggest that future considerations (viz., inferential, practical, and empirical) that
should be taken into account when selecting a model in any
studies should examine the extent to which different
particular context. The general consensus, however, seems to
sociocultural factors identified in this article are
be that random-effects models are more appropriate (e.g.,
empirically distinct or independent from one another.
Hedges & Vevea, 1998).
For instance, Thompson and colleagues’ results (2004) 3. Age can be conceived as a categorical variable if we
support the contention that perceived pressures and separate those below and above the age of 18. Using the age of
internalization are not statistically distinct. In their study, 18 to dichotomize groups is meaningful because it marks
the constructs had a correlation of .72 in one of the a transition point between being an adolescent and being
samples, and when internalization (thin ideal combined a young adult and is often accompanied by changes in one’s
with athlete dimensions), awareness (as measured by the environment (living with parents versus going to college),
IBSS–R), pressures, and information were entered into which can affect the extent that sociocultural factors are
a regression with body dissatisfaction as the dependent associated with body image. When we analyzed the data using
variable, only pressures predicted a statistically signifi- a mixed-model design but treated the moderator as a categor-
ical rather than continuous variable (i.e., the analog to a one-
cant proportion of variance over and above the other
way analysis of variance), we got the same results as when age
factors. Notably though, in the other sample of the
was treated continuously. There was a nonsignificant result
Thompson study, a more moderate correlation of .54
when age was treated as a categorical moderator for the
was observed between pressures and internalization. internalization–body image relationship, QB 5 .40, p 5 .53,
Along the same lines it could be argued that pressures and the awareness–body image relationship, QB 5 1.65,
and awareness are not statistically distinct or, more so, p 5 .20.
that awareness and internalization are not distinct. 4. To calculate the statistical test for the difference between
Indeed, in some studies the correlation between pressures correlations, we found it necessary to enter a value for the

CLINICAL PSYCHOLOGY: SCIENCE AND PRACTICE  V12 N4, WINTER 2005 430
correlation between the sociocultural constructs that were Cafri, G., & Thompson, J. K. (2004). Measuring male body
being compared. Although the correlation between these image: A review of the current methodology. Psychology of
constructs was not reported in this meta-analysis (with one Men and Masculinity, 5(1), 18–29.
exception; see note 6), all available data were collected from Calogero, R. M., Davis, W. M., & Thompson, J. K. (2004).
the primary studies in this meta-analysis and averaged. The The Sociocultural Attitudes Toward Appearance Ques-
sample size of the average correlation between the socio- tionnaire (SATAQ-3): Reliability and normative compar-
cultural constructs was used as the sample size value entered isons of eating disorders patients. Body Image, 1, 193–198.
in the equation and in computing the degrees of freedom for *Cash, T. F., Melnyk, S. E., & Hrabosky, J. I. (2004). The
the test. assessment of body image investment: An extensive
5. Two studies (Stice, 2001; Stice & Whitenton, 2002) were revision of the Appearance Schemas Inventory. International
used to calculate both the cross-sectional and prospective r Journal of Eating Disorders, 35, 305–316.
values. Consequently, to eliminate this dependency and thus *Cashel, M. L., Cunningham, D., Landeros, C., Cokley,
use a test of difference between independent correlations, we K. O., & Muhammad, G. (2003). Sociocultural attitudes
calculated a revised estimate of the cross-sectional effect sizes by and symptoms of bulimia: Evaluating the SATAQ with
removing these two studies. diverse college groups. Journal of Counseling Psychology,
6. In Stice’s meta-analysis (2002) a total of five studies were 50(3), 287–296.
included to examine the prospective relationship between Chen, P. Y., & Popvich, P. M. (2002). Correlation: Parametric
perceived pressures and body image dissatisfaction. For three of and nonparametric measures (Sage University Papers Series on
these five studies, however, the items/scales would not have Quantitative Applications in the Social Sciences No. 07–
met the inclusion criteria of this study because of limited 139). Thousand Oaks, CA: Sage.
information regarding validity and reliability. In one study Cohen, J. (1988). Statistical power analysis for the behavioral
only a single item, one that had no demonstrated validity or sciences (2nd ed.). Hillsdale, NJ: Erlbaum.
reliability, was used as a measure of pressures (‘‘Has anyone Collins, M. E. (1991). Body figure perception and preferences
told you to go on a diet?’’ Byely, 2000). Another study assessed
among preadolescent children. International Journal of Eating
experience being teasing, which is perhaps a specific form of
Disorders, 10, 199–208.
pressure and not a more general measure of perceived pressures
*Cusumano, D. L., & Thompson, J. K. (1997). Body image
(Cattarin & Thompson, 1994). The third study assessed
and body shape ideal in magazines: Exposure, awareness
perceived importance of thinness from friends and family
and internalization. Sex Roles: A Journal of Research, 37,
and the experience of being teased, which again may not
701–724.
necessarily fall within the domain of pressures (Field et al.,
*Cusumano, D. L., & Thompson, J. K. (2000). Media influence
2001). Arguably, the lack of methodological rigor in these
and body image in 8–11 year old boys and girls: A
studies may have brought down the average effect size. If only
preliminary report on the Multidimensional Media In-
the most methodologically rigorous studies are included (Stice,
fluence Scale. International Journal of Eating Disorders, 29,
2001; Stice & Whitenton, 2003), the average effect size would
37–44.
be r 5 .17. This value is still significantly different from the
Didie, E. R., & Sarwer, D. B. (2003). Factors that influence the
average cross-sectional r value, z 5 7.5, p , .05.
decision to undergo cosmetic breast augmentation surgery.
7. We were able to identify 14 studies in this meta-analysis
Journal of Women’s Health, 12, 241–253.
that assessed the correlation between awareness and internal-
Ewell, F., Smith, S., Karmel, M. P., & Hart, D. (1996). The
ization. Significant heterogeneity among the effect sizes, Q 5
76.17, p , .05, combined with a moderate random-effects sense of self and its development: A framework for
variance component, REVC 5 .026, suggested that a random- understanding eating disorders. In L. Smolak, M. P.
effects model was most appropriate. The average effect size was Levine, & R. Striegel-Moore (Eds.), The developmental
r 5 .42, CI 5 .32, .52. psychology of eating disorders: Implications for research, pre-
vention, and treatment (pp. 107–133). Mahwah, NJ: Erlbaum.
REFERENCES Fallon, A. (1990). Culture in the mirror: Sociocultural
Note: References marked with an asterisk indicate studies determinants of body image. In T. F. Cash & T. Pruzinsky
included in the meta-analysis. (Eds.), Body images: Development, deviance and change (pp.
80–109). New York: Guilford Press.
Bordo, S. (1993). Unbearable weight: Feminism, Western culture, *Forbes, G. B., Doroszewicz, K., Card, K., & Adams-Curtis, L.
and the body. Berkeley: University of California Press. (2004). Association of the thin body ideal, ambivalent

SOCIOCULTURAL FACTORS  CAFRI ET AL. 431

 sexism, and self esteem with body acceptance and the *Sands, E. R., & Wardle, J. (2003). Internalization of body
preferred body size of college women in Poland and the ideal shapes in 9–12 year old girls. International Journal of
United States. Sex Roles, 50, 331–346. Eating Disorders, 33, 193–204.
Garner, D. M., & Olmsted, M. P. (1984). The Eating Disorder *Sarwer, D. B., Cash, T. F., Magee, L., Williams, E. F.,
Inventory manual. Odessa, FL: Psychological Assessment Thompson, J. K., Roehrig, M., et al. (in press). College
Resources. students and cosmetic surgery: An investigation of
Groesz, L. M., Levine, M. P., & Murnen, S. K. (2002). The experiences, attitudes, and body image. Plastic and Re-
effect of experimental presentation of thin media images on constructive Surgery.
body satisfaction: A meta-analytic review. International Shrout, P. E., & Fleiss, J. L. (1979). Intraclass correlations:
Journal of Eating Disorders, 31, 1–16. Uses in assessing rater reliability. Psychological Bulletin, 2,
Hedges, L. V., & Vevea, J. L. (1998). Fixed- and random- 420–428.
effects models in meta-analysis. Psychological Methods, 3, *Smolak, L., Levine, M. P., & Thompson, J. K. (2001). The
486–504. use of the Sociocultural Attitudes Towards Appearance
*Heinberg, L. J., Thompson, J. K., & Stormer, S. (1995). Questionnaire with middle school boys and girls. In-
Development and validation of the Sociocultural Attitude ternational Journal of Eating Disorders, 29, 216–223.
Towards Appearance Questionnaire (SATAQ). Interna- Steel, P. D., & Kammeyer-Mueller, J. D. (2002). Compar-
tional Journal of Eating Disorders, 17, 81–89. ing meta-analytic moderator estimation techniques under
Hunter, J. E., & Schmidt, F. L. (1990). Methods of meta-analysis: realistic conditions. Journal of Applied Psychology, 87,
Correcting error and bias in research findings. Newbury Park, 96–111.
NJ: Sage. *Steinberg, A. (2004). College students’ beliefs regarding weight
*Keery, H., Shroff, H., Thompson, J. K., Wertheim, E., & and body shape. Unpublished doctoral dissertation, Tampa,
Smolak, L. (2004). The Sociocultural Internalization of University of South Florida.
Appearance Scale—Adolescent version. Eating and Weight *Stice, E. (2001). A prospective test of the dual pathway model
Disorders, 9, 56–61. of bulimic pathology: Mediating effects of dieting and
Levine, M., & Harrison, K. (2004). Media’s role in the negative affect. Journal of Abnormal Psychology, 110, 124–
perpetuation and prevention of negative body image and 135.
disordered eating. In J. K. Thompson (Ed.), Handbook of Stice, E. (2002). Risk and maintenance factors for eating
eating disorders and obesity (pp. 695–717). New York: Wiley. pathology: A meta-analytic review. Psychological Bulletin,
Lipsey, M. W., & Wilson, D. B. (2001). Practical meta-analysis. 128, 825–848.
Thousand Oaks, CA: Sage. *Stice, E., & Agras, W. S. (1998). Predicting onset and
*Low, K. G., Charanaomboon, S., Brown, C., Hiltunen, G., cessation of bulimic behaviors during adolescence: A
Long, K., Reinhalter, K., et al. (2003). Internalization of the longitudinal grouping analyses. Behavior Therapy, 29,
thin ideal, weight and body image concerns. Social Behavior 257–276.
and Personality, 31, 81–90. Stice, E., Nemeroff, C., & Shaw, H. (1996). A test of the
*McClintok, J. M. (2003). The influence of sociocultural and dual pathway model of bulimia nervosa: Evidence for
interpersonal factors on body image disturbance and unhealthy restrained-eating and affect regulation mechanisms. Journal
dieting in female adolescents. Unpublished doctoral disserta- of Social and Clinical Psychology, 15, 340–363.
tion, University of Waikato, Hamilton, New Zeland. *Stice, E., Schupak-Neuberg, E., Shaw, H. E., & Stein, R. I.
Nunnally, J. (1978). Psychometric theory (2nd ed.). New York: (1994). Relation of media exposure to eating disorder
McGraw-Hill. symptomatology: An examination of mediating mecha-
Overton, R. C. (1998). A comparison of fixed-effects and nisms. Journal of Abnormal Psychology, 103, 836–840.
mixed (random-effects) models for meta-analysis tests of Stice, E., Trost, A., & Chase, A. (2003). Healthy weight
moderator variable effects. Psychological Methods, 3, 354– control and dissonance-based eating disorder prevention
379. programs: Results from a controlled trial. International
*Ruggiero, G. M., Hannover, W., Mantero, M., & Papa, R. Journal of Eating Disorders, 33, 10–21.
(2000). Body acceptance and culture: A study in northern *Stice, E., & Whitenton, K. (2002). Risk factors for body
and southern Italy. European Eating Disorders Review, 8, dissatisfaction in adolescent girls: A longitudinal investiga-
40–50. tion. Developmental Psychology, 38, 669–678.

CLINICAL PSYCHOLOGY: SCIENCE AND PRACTICE  V12 N4, WINTER 2005 432
Stice, E., Ziemba, C., Margolis, J., & Flick, P. (1996). The dual *Thompson-Leonardelli, K. I. (2003). The influence of socio-
pathway model differentiates bulimics, subclinical bulimics, cultural variables on the development of body image and
and controls: Testing the continuity hypothesis. Behavior disordered eating in African and European American
Therapy, 27, 531–549. college women. Dissertation Abstracts International, 63,
Striegel-Moore, R., Silberstein, L., & Rodin, J. (1986). 3486.
Toward an understanding of risk factors for bulimia. *Twamley, E. W., & Davis, M. C. (1999). The sociocultural
American Psychologist, 41, 246–263.
model of eating disturbance in young women: Effects of
Thompson, J. K., Heinberg, L., Altabe, M., & Tantleff-Dunn, S.
personal attributes and family environment. Journal of Social
(1999). Exacting beauty. Washington, DC: American
Psychological Association. and Clinical Psychology, 18, 467–489.
Thompson, J. K., & Stice, E. (2001). Thin ideal internalization: *Warren, C. S., Gleaves, D. H., Cepeda-Benito, A.,
Mounting evidence for a new risk factor for body image Fernandez, M. D. C., & Rodriguez-Ruiz, S. (in press).
disturbance and eating pathology. Current Directions in Ethnicity as a protective factor against internalization of
Psychological Science, 10, 181–183. a thin-ideal and body dissatisfaction. International Journal of
*Thompson, J. K., van den Berg, P., Roehrig, M., Guarda, A., & Eating Disorders.
Heinberg, L. J. (2004). The Sociocultural Attitudes
Towards Appearance Scale-3. International Journal of Eating Received July 29, 2004; revised January 6 and April 25, 2005;
Disorders, 35, 293–304. accepted May 2, 2005.

SOCIOCULTURAL FACTORS  CAFRI ET AL. 433