F(

Aa

Cb

Wc

od

e

P

ARRA

KEAAEEG

vdrnLBortTs

caP

(

1h

Body Image 10 (2013) 399– 405

Contents lists available at SciVerse ScienceDirect

Body Image

j o ur nal ho me page: www.elsev ier .com/ locate /bodyimage

actor structure of the Eating Disorder Examination QuestionnaireEDE-Q) in male and female college athletes

lison M. Darcya, Kristina K. Hardyb,c, Ross D. Crosbyd, James Locka, Rebecka Peeblese,∗

Child and Adolescent Psychiatry, Stanford University School of Medicine, Department of Psychiatry and Behavioral Sciences, 401 Quarry Road, Stanford,A 94305, United StatesNeuropsychology Division, Children’s National Medical Center, Department of Pediatrics, The George Washington University School of Medicine,ashington, DC, United States

Neuropsychology Division, Children’s National Medical Center, Department of Psychiatry & Behavioral Science, The George Washington University Schoolf Medicine, Washington, DC, United StatesNeuropsychiatric Research Institute, Fargo, ND, United StatesThe Craig Dalsimer Division of Adolescent Medicine, The Children’s Hospital of Philadelphia, Perelman School of Medicine at The University ofennsylvania, Philadelphia, PA 19104, United States

a r t i c l e i n f o

rticle history:eceived 25 July 2012eceived in revised form 25 January 2013ccepted 28 January 2013

a b s t r a c t

The study explored the psychometric properties of the Eating Disorder Examination Questionnaire (EDE-Q) among 1637 university students. Participants were divided into male (n = 432) and female (n = 544)competitive athletes, and male (n = 229) and female (n = 429) comparison groups comprised of individualswho had not engaged in competitive sports for at least one year. All groups were subjected to confirma-tory factor analysis (CFA) to test the fit of the published factor structure in this population, and then

eywords:ating disordersthletesssessmentating Disordersxamination-Questionnaireender

exploratory FA (EFA). A three-factor solution was the best fit for three out of four groups, with a two-factor solution providing best fit for the male comparison group. The first factor for all groups resembleda combined Shape and Weight Concern subscale. The factor structure among male and female com-petitive athletes was remarkably similar; however, non-competitive athletic/low activity males appearqualitatively different from other groups.

© 2013 Elsevier Ltd. All rights reserved.

Introduction

The Eating Disorder Examination (EDE) structured clinical inter-iew has long been considered the measure of choice in eatingisorders (ED) research (Wilson, 1993). It is a well-validated andeliable instrument (Fairburn & Cooper, 1993) with publishedorms for many clinical and non-clinical groups (e.g., Byrne, Allen,ampard, Dove, & Fursland, 2010; Fairburn & Cooper, 1993; Wade,yrne, & Bryant-Waugh, 2008). A self-report questionnaire versionf the measure exists with many studies demonstrating its internaleliablility (e.g., Luce & Crowther, 1999; Peterson et al., 2007) and

emporal stabilitity (Mond, Hay, Rodgers, Owen, & Beumont, 2004).he EDE-Questionnaire (EDE-Q) is based on the same four-subscaletructure as the EDE, which were chosen on “rational rather than

∗ Corresponding author. Present address: The Craig Dalsimer Division of Adoles-ent Medicine, The Children’s Hospital of Philadelphia, Perelman School of Medicinet The University of Pennsylvania, 11 NW Room 19, 34th and Civic Center Boulevard,hiladelphia, PA 19104, United States. Tel.: +1 215 590 6864; fax: +1 215 590 4708.

E-mail addresses: peeblesr@email.chop.edu, cheekc@email.chop.eduR. Peebles).

740-1445/$ – see front matter © 2013 Elsevier Ltd. All rights reserved.ttp://dx.doi.org/10.1016/j.bodyim.2013.01.008

empirical grounds” on the basis of conceptual similarity of content(Mond et al., 2004). Many studies have compared the EDE and theEDE-Q, generally finding support for its convergent validity amongclinical and community samples (Binford, Le Grange, & Jellar, 2005;Grilo, Masheb, & Wilson, 2001; Mond et al., 2004; Passi, Bryson, &Lock, 2002; Wilfley, Schwartz, Spurrell, & Fairburn, 1997; Wolk,Loeb, & Walsh, 2005) with the exception of ratings of objectivebinge eating episodes, which tend to be overestimated on the EDE-Q among clinical samples (Bohn, Doll, Cooper, O’Connor, Palmer, &Fairburn, 2008; Wolk et al., 2005). The EDE-Q is increasingly beingused in eating disorders research as both an outcome measure andas a descriptive adjunct (Anderson & Williamson, 2002). In addi-tion, Mond and colleagues have investigated and acknowledgedits acceptability as a screening instrument in prospective epidemi-ologic research (Mond et al., 2004). Despite its wide application,there is not yet a consensus on the factor structure of the EDE-Q inclinical or community samples (Peterson et al., 2007).

There have been four studies investigating the factor structure of

the EDE and the EDE-Q. Using principal components analysis withvarimax rotation, Mannucci, Rica, Di Bernardo, Moretti, Cabras, andRotella (1997) found support for a two-component structure of theEDE among obese men (n = 23) and women (n = 92). Eating, Weight

4 y Imag

CRe(yAst4

ogTtgBrfiWERidchtwg

tgEt(mabptpwi

vawpAtwSiiatni

Qataattw

00 A.M. Darcy et al. / Bod

oncern, and Shape Concern comprised the first component, andestraint comprised the second. Wade et al. (2008) conducted anxploratory factor analysis (EFA) of the child version of the EDEwith non-orthogonal Promax rotation) among 699 female 12–15ear old twin sets resulting in an unstable factor structure (twin set

resulted in a 5 factor structure and twin set B resulted in 4 factortructure). However, the first factor was comprised of 8 items fromhe Shape and Weight Concern subscales and accounted for over0% of the variance for both groups (Wade et al., 2008).

The first factor analysis of the EDE-Q was of the Dutch versionf the measure by Pennings and Wojciechowski (2004) among aroup of women with anorexia nervosa (AN) and healthy controls.he authors found that a one-factor solution provided best fit forhe data. Peterson and colleagues (Peterson et al., 2007) investi-ated the factor structure of the EDE-Q among 203 females withN symptoms using principal components analysis with Promaxotation. The data indicated a 4 component solution was the bestt, with the first component consisting of most of the Shape andeight Concern items, the second component loosely fitting the

ating Concern Factor, the third comprising four items from theestraint subscale, and importance of weight and shape compris-

ng the fourth component. Thus, each investigation has supportediffering structures, with the exception that Weight and Shape Con-ern tends to cluster together. Unfortunately, most of these studiesad insufficent numbers to allow for conducting a confirmatory fac-or analysis (CFA), the only procedure that can specifically answerhether a given model (i.e., the published 4-factor structure) is a

ood fit or not.One recent study (Byrne et al., 2010) investigated 5 differing fac-

or structures described in the literature to directly compare theiroodness of fit. Byrne and colleagues found that none of the testedDE models provided a good fit to their data from ED (n = 158);reatment seeking obese (n = 170) and non-ED community-basedn = 329) participants, with the exception of a brief one-factor

odel consistent with the findings of Wade and colleagues. Theuthors interpreted this finding as being related to the fact thatoth were clinical samples. Since it was an assessment of previouslyublished structures, this study did not include an exploratory fac-or analysis (EFA). In spite of the lack of consensus of psychometricroperties among community samples, the EDE-Q is one of the mostidely used instruments in the field of ED and is used as a screening

nstrument among groups considered high risk for ED.Both male and female elite athletes can be considered at an ele-

ated risk for developing EDs. Between 15 and 78% of female elitethletes demonstrate disordered eating (Byrne & McLean, 2001),hich, alongside menstrual irregularity and osteoporosis, com-rises an area of specific health detriment known as the Femalethlete Triad. Young adult males participating in athletic activi-

ies, especially those that include specific requirements to “makeeight” (e.g., wrestling), may be at risk for developing EDs (Braun,

unday, Huang, & Halm, 1999). Given the potential utility of thenstrument in screening studies and as an outcome measure, it ismperative that the psychometric properties of the EDE-Q be evalu-ted among high-risk groups such as university-aged athletes, andhat norms be available for these groups. Furthermore, virtuallyothing is known about the factor structure of the EDE or the EDE-Q

n males.This study aimed to explore the factor structure of the EDE-

among a large group of competitive athletic men and women,s well as a comparison group of men and women to establishhe contribution of competitive athleticism. We aimed to evalu-te comprehensively the factor structure by first conducting a CFA,

nd then an EFA, using procedures designed to specifically test forhe appropriate number of factors that should be extracted fromhe data. Given previous findings, we hypothesized that shape andeight concern would load on the same factor.

e 10 (2013) 399– 405

Method

Participants and recruitment

Participants (n = 1637) were aged 18–26 and affiliated with oneof ten top-ranked National College Athletics Association (NCAA)schools in the United States. Participants were recruited via a pop-ular social networking website and invited to complete an onlinesurvey examining their athletic participation, eating behaviors, andhealth. Interested participants gave consent online and completedan online survey consisting of a total of 295 questions. Partici-pants only completed questions that were relevant to them basedon involvement with specific sports and could skip non-relevantquestions. The survey took approximately 30 min to complete. Thisstudy was conducted in compliance with the Stanford UniversityPanel on Medical Research in Human Subjects and the NCAA.

Measures

The Eating Disorders Examination Questionnaire (EDE-Q;Fairburn & Beglin, 2008) is a 28-item self-report measure of eatingdisorder psychopathology based on the EDE interview (Fairburn &Cooper, 1993). The EDE-Q asks participants to rate on a 0–6 scale thefrequency or severity of core symptoms and related psychopatho-logical behaviors and beliefs over the past 28 days. The measurecomprises four subscales: Restraint, Eating Concern, Shape Con-cern, and Weight Concern. Participants were asked to complete theEDE-Q as part of the larger survey.

Data Analysis

Given that the psychometric properties of the measure amongathletes are unknown, confirmatory factor analysis (CFA) esti-mated by robust maximum likelihood method (MLM; Bentler,1995; Satorra & Bentler, 1994) was used to assess the fit of the pub-lished four-factor model of the EDE-Q representing the subscalesRestraint; Eating Concern; Shape Concern and Weight Concern.Analyses were conducted separately for male competitive ath-letes (n = 432); and female competitive athletes (n = 544); and male(n = 229) and female (n = 429) comparison groups, using LISREL 8.8.A multiple imputation (MI) procedure was used to handle missingdata.

Preliminary diagnostics indicated that multivariate distributionof the data was non-normal. Given the moderate size of each group,the MLM method was chosen over the more popular maximumlikelihood (ML), or alternative non-normal equivalent, the WeightLeast Squares (WLS) method, as it is a better behaved estimatoracross different levels of non-normality, model complexity, andsample size (Brown, 2006).

Several widely used models of goodness-of-fit were employed.Due to violations of normality, the Satorra–Bentler scaled chi-square test statistic was employed (Satorra & Bentler, 1994). Otherindicators of good fit included Bentler’s Comparative Fit Index (CFI)and Tucker–Lewis Index (TLI; Tucker & Lewis, 1973) with val-ues of ≥0.96 indicative of good fit (Hu & Bentler, 1999). Finally,standardized residuals and modification indices were examined toindicate the presence of areas of localized stress and provide overallinterpretability of the usefulness of the model (Brown, 2006). Stan-dardized residuals greater than 2.58 are indicative of significantunder-estimation (in the case of positive values) or over-estimation(in the case of negative values). Modification Indices can be con-ceptualized as the point difference in the chi-square statistic with 1

degree of freedom if the parameter was freely estimated. Thus, goodfitting models should produce modification indexes that are small,with indexes >3.84 indicative of a model that could be significantlyimproved (p < .05) (Brown, 2006).

A.M. Darcy et al. / Body Image 10 (2013) 399– 405 401

Table 1Demographic information and EDE-Q Subscale scores for the samples.

Competitive athletes Comparison Total

Male (n = 432) Female (n = 544) Male (n = 229) Female (n = 429) (N = 1634)Mean (SD) Mean (SD) Mean (SD) Mean (SD) Mean (SD)

Age 21.03 (1.77) 20.63 (1.48) 20.90 (1.71) 21.01 (1.70) 20.87 (1.66)BMI 24.30 (3.31) 22.84 (2.92) 24.09 (3.95) 22.52 (3.48) 23.32 (3.41)RES 1.02 (1.18) 1.68 (1.49) 1.03 (1.21) 1.65 (1.48) 1.41 (1.41)EC .27 (0.64) .85 (1.17) .33 (0.74) .96 (1.22) .65 (1.06)SC 1.08 (1.12) 2.39 (1.57) 1.42 (1.38) 2.41 (1.59) 1.94 (1.56)WC .77 (1.01) 1.94 (1.54) 1.01 (1.26) 2.11 (1.53) 1.55 (1.50)

S ; SC:

lrsayFStptturtvrotcadwim

P

t7a8a

(C5I(sg

aOttdie

and none of the lower CI values were below this cut off. CFI valueswere above 0.96 for all groups with the exception of the femalecomparison group, and TLI values were not above this cut off forthe female comparison group nor for male competitive athletes.

Table 2CFA Global Fit Indexes for the four groups.

Fit index Competitive athletes Comparison

Male(n = 432)

Female(n = 544)

Male(n = 229)

Female(n = 429)

Chi-square 750.77 1157.74 529.02 1185.82Chi-square df 202 202 202 202p <.001 <.001 <.001 <.001

SRMR 0.1081 0.0670 0.0861 0.0711RMSEA 0.0793 0.0933 0.0842 0.1067

90% lower CI 0.0733 0.0881 0.0755 0.100890% upper CI 0.0855 0.0986 0.0930 0.1126

CFI 0.9627 0.9794 0.9694 0.9561

Global .80 (0.85) 1.77 (1.32)

D: standard deviation; BMI: body mass index; RES: Restriction; EC: Eating Concern

Data that demonstrated poor fit were submitted to maximumikelihood exploratory factor analysis (EFA) with promax (oblique)otation to identify latent constructs. Given the lack of consen-us of number of factors, Horn’s parallel analysis (PA; Horn, 1965)nd Velicer’s minimum average partial (MAP; Velicer, 1976) anal-sis were used to determine the number of factors to extract.ollowing the practice of other investigators (Hayton, Allen, &carpello, 2004), these methods were chosen over more conven-ional approaches (e.g., Kaiser’s eigenvalue >1 rule, Cattell’s screelot test) given the evidence to suggest the tendency of the moreraditional approaches to overestimate or inconsistently identifyhe number of factors. Syntax specified by O’Connor (2000) wassed to identify the number of factors. Briefly, the PA generates aandom set of the same number of observations and variables as inhe observed dataset. Using principal components analysis, eigen-alues are extracted from the random data, and this extraction isepeated for 100 iterations. Eigenvalues from the 95th percentilef the set are then compared to those from the observed data, andhose with values greater than the random data are retained. Prin-ipal axis factoring (EPA) was chosen over principal componentsnalysis because the aim was to identify the latent structure of theata rather than reduce the number of variables. All EFA analysesere conducted in SPSS version 17.0. We used a cut-off of .4 to guide

tem factor loading, corresponding to the lowest typically observedagnitudes in the social sciences (Costello & Osborne, 2005).

Results

articipants

A total of 1688 participants responded to the invitation. Out ofhis group, 51 participants were excluded for the following reasons:

participants were outside of the desired age range; 14 were notffiliated with target universities; 18 had very few or no responses;

divulged that they had not been honest in completing the survey,nd 4 represented duplicate entries.

Of 1637 participants, 59.5% (n = 977) were female and 40.5%n = 661) were male. The ethic breakdown was as follows: 74.2%aucasian (n = 1210); 11.2% Asian (n = 183); 6.1% Hispanic (n = 99);.4% African American (n = 88); 0.1% Native Hawaiian/Pacific

slander (n = 2); 1.6% biracial (2 non-Caucasian) or multiracialn = 2); 1.0% other (n = 17). The overwhelming majority of theample (94.0%) was full-time with 91.4% of the sample being under-raduates.

Out of the entire sample, 976 were competitive athletes defineds participating in competitive sports currently or in the past year.ur comparison group (n = 671) had not engaged in any competi-

ive athletics for at least one year. This group was heterogeneous in

erms of activity level, with 22% (n = 671) considered low activity,efined as not partaking in any regular exercise, and 78% engag-

ng in varying amounts of recreational though non-competitivexercise.

.95 (1.02) 1.70 (1.29) 1.40 (1.24)

Shape Concern; WC: Weight Concern.

The competitive athletic group represented all levels of com-petition with 34.4% competing at national, international and elitelevels; 46.1% practiced at least 45 weeks/year. In terms of total timespent exercising, 33% of the competitive athletic group practiced atleast 19 h per week and 7.7% trained at least 25 h per week.

The competitive athletic groups represented a broad range ofsport types both “lean” (i.e., sports in which weight and bodytype offer a competitive advantage) and “non-lean” sports. Leansports that were practiced by this group were swimming; div-ing; cheerleading; dance; running; triathlon; gymnastics; rowing;wrestling; cycling; equestrian; figure skating; synchronized swim-ming (n = 494; n = 189 males, n = 305 females). Non-lean sports thatwere practiced included soccer; water polo; lacrosse; rugby; ulti-mate Frisbee; volleyball; softball/baseball; football; field hockey;ice hockey; basketball fencing; golf; power lifting; squash; martialarts; field; tennis (n = 467; n = 236 males, n = 231 females). Manyparticipants did multiple sports from both categories so there wassubstantial over-lap.

Additional demographics and EDE-Q subscale scores by groupare presented in Table 1.

Confirmatory Factor Analyses

Results of the confirmatory factor analyses can be seen inTable 2. Satorra–Bentler chi-square statistics were significant forall models, though this was expected given the size of the sam-ples (Satorra & Bentler, 1994). While SRMR values were below 0.08for both the female groups (competitive athletes and comparison),they were larger than 0.08 for both male groups. RMSEA valueswere higher than the 0.06 cut off indicative of good fit for all groups

TLI 0.9574 0.9764 0.9650 0.9497

Notes: df: degrees of freedom; SRMR: standardized root mean square residual;RMSEA: root mean square errors of approximation; CI: confidence interval; CFI:Comparative Fit Index; TLI: Tucker–Lewis Index.

402 A.M. Darcy et al. / Body Image 10 (2013) 399– 405

Table 3Summary of EFA results for male and female the competitive athletic groups.

Item Male competitive athletes Female competitive athletes

Factor 1 Factor 2 Factor 3 Factor 1 Factor 2 Factor 3

Dietary Restraint Restraint over eating .017 .046 .702 .024 .027 .788Avoidance of eating −.146 .762 −.002 (−.078) (.379) (.263)Food avoidance −.011 −.090 .815 −.062 −.037 .857Dietary rules −.036 .023 .686 −.104 .076 .794Empty stomach −.227 .645 .180 (.004) (.377) (.364)

Eating Concern Preoccupation with food −.023 .526 .034 −.109 .932 .079Fear of losing control over eating .044 .720 .045 .115 .683 −.005Social eating .099 .781 −.137 .242 .675 −.126Eating in secret −.062 .821 −.113 −.052 .738 −.038Guilt about eating .277 .573 .023 .279 .446 .154

Shape Concern Flat stomach (.249) (−.017) (.396) .296 −.058 .413Importance of shape .595 −.041 .137 .494 .226 .123Preoccupation with shape/weight .114 .609 .002 −.030 .890 .060Dissatisfaction with shape 1.020 −.261 −.010 .943 −.045 −.026Fear of weight gain .231 .476 .123 (.334) (.225) (.302)Discomfort seeing body .826 −.004 −.047 .889 .045 −.089Discomfort exposing body .860 −.035 −.146 .879 .044 −.141Feelings of fat .514 .182 .142 .747 −.030 .150

Weight Concern Importance of weight .600 .060 .111 .559 .228 .105Reaction to prescribed weighing .418 .380 −.206 .458 .288 −.101Dissatisfaction with weight .748 .053 −.027 .933 −.013 −.036Desire to lose weight .415 .204 .291 .695 −.149 .366

Eigenvalue 8.861 1.873 1.768 10.215 1.683 1.394% variance 42.193 8.919 8.417 53.763 8.859 7.334

BN strate

Ogmtl

saupcteSTiatbEsF

scWwFaiaieDuIaE

olded values indicate highest loading.ote: Italicized items in parentheses are loadings from the last solution that demon

verall, while the picture was somewhat mixed, global indices ofoodness of fit were not indicative of particularly good fit for anyodel, particularly for the male groups. However, in evaluating

he model, it is important to consider the presence of any areas ofocalized strain since this can influence other indices.

Examination of standardized residuals for all groups revealedpecific areas of ill fit. Female competitive athletes showed manyreas of strain, with many specific areas of overestimation andnderestimation. Among 18 areas of overestimation, particularlyroblematic indicators were Social Eating, Eating in Secret, Dis-omfort Seeing Body, Discomfort Exposing Body, and Reactiono Prescribed Weighing. Among an equivalent number of under-stimated areas, particularly troublesome indicators were Flattomach, Fear of Weight Gain, and Reaction to Prescribed Weighing.he model yielded somewhat fewer areas of strain for compet-tive athletic men than women; although there were still manyreas of overestimation, and there were fewer areas of underes-imation. In particular, the model overestimated the relationshipetween many indicators and the following items: Social Eating,ating in Secret, Discomfort Exposing Body, and Reaction to Pre-cribed Weighing. The model underestimated relationships withlat Stomach and Fear of Weight Gain.

For the female comparison group, there were a number ofpecific areas of ill fit. Among several areas of overestimation, Dis-omfort Exposing Body; Dissatisfaction with Weight; Desire to Lose

eight; Dissatisfaction with Shape; and Preoccupation with Foodere particularly problematic. Among areas of underestimation,

ear of Weight Gain; Flat Stomach; Importance of Shape; Guiltbout Eating; Importance of Weight Reaction to Prescribed Weigh-ng and Empty Stomach were problematic. The 4 factor modelppeared to be a better fit for comparison males, with only 3ndicators showing significant overestimation with model param-ters (Dissatisfaction with Shape, Dissatisfaction with Weight, andesire to Lose Weight); though more indicators showed significant

nderestimation with model parameters (Desire to Lose Weight,

mportance of Weight, Food Avoidance, Dietary Rules, Flat Stom-ch, Importance of Shape, Dissatisfaction with Shape, Discomfortxposing Body, Importance of Weight).

d poor loading and were therefore excluded from the analyses.

The substantial list of underestimated and overestimated rela-tionships in this model suggests that the published 4-factor modelis problematic for all groups, providing best fit for our compari-son males. While there were many large modification indexes forall but the male comparison group, we examined the completelystandardized expected parameter change (EPC) values given oursample sizes and there were no indications for respecification ofany parameters. While the picture was mixed, we feel that takentogether, the four-factor model failed to demonstrate good fit forany group.

Exploratory Factor Analysis

Given that the data failed to provide sufficient evidence of goodfit for any of the groups, all groups were submitted to EFA. Table 3shows the solutions for male and female competitive athletes. Thetable shows initial factor loadings for items that did not sufficientlyload on any factor and were subsequently removed before re-running the analyses without these items. The MAP and PA tests forboth of these groups suggested that a 3-factor model would be mostappropriate for both sets of data. For males, the first factor was com-prised of most of the items from both the Shape and Weight Concernsubscales, with the exception of the desire for a flat stomach itemwhich was removed because it failed to load adequately on any fac-tor and Fear of Weight Gain as well as Preoccupation with Shapeand Weight which both loaded on the second factor as well as all ofthe items from the Eating Concern subscale and the Empty Stomachitem from Restraint. The third factor was comprised of 3 items fromthe restraint subscale – Restraint over Eating, Food Avoidance, andDietary Rules. Reaction to Prescribed Weighing cross-loaded on thesecond factor indicating that this item may capture a more complexdimension in this group. Since this may represent a salient aspectof body image for males involved in lean sports, it was decided to

retain this item in the first factor. In the original solution, 58.75% ofthe variance was accounted for. Re-running the analyses after theremoval of the Desire for a Flat Stomach item, led to a model thataccounted for 59.53% of the variance.

A.M. Darcy et al. / Body Image 10 (2013) 399– 405 403

Table 4Summary of EFA results for male and female comparison groups.

Item Male Female

Factor 1 Factor 2 Factor 1 Factor 2 Factor 3

Dietary Restraint Restraint over eating .646 .069 −.083 −.012 .927Avoidance of eating −.026 .785 (−.155) (.354) (.296)Food avoidance .459 .012 −.005 −.090 .871Dietary rules (.241) (.351) −.112 .167 .736Empty stomach .041 .770 (.085) (.358) (.189)

Eating Concern Preoccupation with food −.070 .884 −.136 .891 −.022Fear of losing control over eating .356 .412 .003 .731 .028Social eating (.286) (.394) .195 .631 −.091Eating in secret −.141 .411 .106 .560 −.054Guilt about eating (.368) (.375) .137 .524 .227

Shape Concern Flat stomach .637 −.012 (.263) (−.039) (.345)Importance of shape .799 −.075 .516 .326 .045Preoccupation with shape/weight .066 .860 −.082 .834 .067Dissatisfaction with shape .943 −.130 .902 −.105 .078Fear of weight gain .594 .318 (.263) (.227) (.380)Discomfort seeing body .827 .014 .922 −.009 −.055Discomfort exposing body .776 −.065 .933 −.088 −.160Feelings of fat .749 .045 .590 .118 .177

Weight Concern Importance of weight .801 −.004 .510 .377 .041Reaction to prescribed weighing .530 .141 .565 .185 −.194Dissatisfaction with weight .841 −.046 .814 −.050 .095Desire to lose weight .862 −.090 .514 −.063 .448

Eigenvalue 9.35 2.05 9.53 1.60 1.34% variance 49.23 10.83 52.98 8.92 7.45

BN strate

rscFafoata1

pttfaswfcDlat

tcpaai7RIEa

olded values indicate highest loading.ote: Italicized items in parentheses are loadings from the last solution that demon

The factor structure for competitive athletic women wasemarkably similar to that of the men, though it was a 19-itemolution rather than a 21-item solution. The first factor was mainlyomprised of Shape and Weight Concerns, with two exceptions;ear of Weight Gain was removed and Preoccupation with Shapend Weight was loaded on the second factor with all of the itemsrom Eating Concern, as for the men. The third factor was comprisedf 3 items from the Restraint subscale–Restraint, Food Avoid-nce, and Dietary Rules, as well as the Flat Stomach item. Fromhe initial analysis running all 22 items, the percentage of vari-nce increased from 66.10% to 69.96% running the analysis with9 items.

Table 4 shows the factor structure for the male and female com-arison groups. For the male comparison group, the MAP and PAests indicated that a 2 factor model best described the data, andhe initial solution accounted for 57.25% of the variance. The firstactor was comprised of Restraint Over Eating and Food Avoidance,ll of the items from the Weight Concern and Shape Concern sub-cale, with the exception of Preoccupation with Shape and Weighthich loaded on the second factor along with Empty Stomach item

rom the Restraint subscale; and all items from the Eating Con-ern subscale with the exception of those that were removed. Theietary Rules, Social Eating and Guilt About Eating items did not

oad adequately (>.4) on any factor and so were removed from thenalysis leading to a 19-item model that accounted for 60.06% ofhe variance.

In contrast to the males, both the MAP and PA test indicatedhat a three-factor structure would best fit the data for the femaleomparison group. The first factor was very similar to the counter-art male group with the exception that the Fear of Weight Gainnd Desire for a Flat Stomach items did not adequately load ontony factor and were removed. Factor 2 corresponded to the orig-nal Eating Concern subscale and the third factor, accounting for.45% of the variance, was comprised of 3 items from the Dietary

estraint Subscale; Restraint, Food Avoidance, and Dietary Rules.

n addition to Fear of Weight Gain and Flat Stomach, Avoidance ofating and Empty Stomach did not adequately load to any subscalend were removed from the analyses. When the model was run

d poor loading and were therefore excluded from the analyses.

again without these items, the resulting 18-item model accountedfor 69.35% (an improvement from 63.23%).

Discussion

The study aimed to provide a full exploration of the factor struc-ture of the EDE-Q in men and women who participate at a highlevel of competitive sports, and those who do not. We aimed toanswer specifically whether the current published 4-factor modelof the instrument is appropriate for these groups and explore thecontribution of competitive athleticism. In line with the most sta-ble finding from previous research (Byrne et al., 2010; Petersonet al., 2007; Wade et al., 2008), we hypothesized that the Weightand Shape Concern subscale would load on one factor for allgroups.

Our findings suggest that in all groups apart from the male com-parison group, a three-factor solution was the best fit for the data. Ashypothesized, Weight and Shape Concern tended to load togetherfor all four groups. This is largely consistent with the findings ofother EFA studies, and the only other CFA study, which demon-strated that a brief, 8-item one-factor model provided best fit to thedata. However, findings from previous studies are not directly com-parable for various reasons. First, this is the first sample of athletes,and to our knowledge, the largest community sample of adults toprovide EDE-Q data to date. Second, this is the first factor analy-sis among a male sample large enough to facilitate such analyses.Third, this study recruited participants on the Internet, which is stilla relatively novel procedure. Finally, this is the first study to employCFA and then sequentially extend the findings to an EFA, employingprocedures specifically designed to extract the appropriate num-ber of factors. It is noteworthy that in both the male groups, thenumber of factors with eigenvalues >1 was overestimated by onemore factor than the PA and MAP analysis suggested were appro-priate (4 for competitive athletic men and 3 for male comparisons).

Thus applying the conventional method, Kaiser’s eigenvalue >1rule, without the MAP and PA procedure would have lead to themistaken impression that the factor structure for males is morecomplex than for equivalent female samples. Thus, employment of

4 y Imag

tfian

toaaenpwhwmbHocIwwDmfapms

trtgtc

eswcpmcrsttma1

wreitlppeimcn

04 A.M. Darcy et al. / Bod

hese methods represents a significant strength of the study. Thendings provide more evidence that the EDE-Q should be vieweds a 3-factor instrument when used with male elite athletes oron-competitive athletic/low activity university-aged women.

The male comparison group differed from the other groups inhat items from the Dietary Restraint subscale loaded onto the sec-nd factor, whereas in the other groups, these items comprised

third factor. The difference in structure between competitivethletic and comparison males seems to suggest a qualitative differ-nce in the salience or function of certain items for males who areot involved in competitive sport. For example, for the male com-arison group, Desire for a Flat Stomach loaded on the first factor,hereas this factor failed to load significantly on any factor for theighly athletic males. We may consider, for example, that malesho are involved in competitive sport may have more pressure toanipulate diet with a view to changing shape, and in particular

uild muscle, and they would therefore want a “6-pack” stomach.owever, among the non-competitive group, a significant portionf whom were low-activity, desired body changes may be less spe-ific, and staged, and so a flat stomach may be desirable at first.ndeed, this item tended not to be endorsed by a group of boys

ith anorexia nervosa in a recent comparison of EDE responsesith age- and percent body weight-matched females (Darcy, Celiooyle, Lock, Peebles, Doyle, & Le Grange, 2012), suggesting that itay be a less useful item for more symptomatic males, or those

or whom muscle building is valued. Dietary Rules, Social Eating,nd Guilt about Eating also did not load adequately for male com-arison subjects, suggesting that these symptoms do not pertain toales who are not competitive athletes, and probably reflects less

ocialization in general around eating for males.The three remaining groups were remarkably similar with a

hree-factor structure providing best fit, with the items that wereemoved from the data from competitive athletic women amonghe poorer loading items for the equivalent male group. This sug-ests that elite athletes, whether male or female, may be subjecto similar behaviors and cognitions around shape and weight con-erns, and to a lesser extent, dietary concern and eating concern.

Since the comparison male group is probably closer to the gen-ral male population, we suggest that the EDE-Q in its current formhould at least be used in conjunction with additional measureshen attempting to characterize ED symptoms in males. Clini-

ians and researchers may consider using additional measures thatertain to the male experience in conjunction with (a) a 19-itemeasure comprised of two subscales including the above 13 items

orresponding to weight and shape concerns, and 6 items cor-esponding to avoidance/preoccupation items or (b) the 13-itemubscale corresponding to weight and shape concerns in conjunc-ion with additional measures. Examples of additional measureshat may be a useful adjunct in a screening or evaluation of males

ight be the Drive for Muscularity Scale (McCreary & Sasse, 2000)nd the Obligatory Exercise Questionnaire (Pasman & Thompson,998).

A limitation of the study was the heterogeneity of the groupsith respect to activity level, and within the athletic groups, with

espect to type of sport. It is suggested that type of sport – forxample, those for which weight or shape has competitive signif-cance – may significantly influence ED symptoms. Since many ofhe participants in this study practiced multiple sports from bothean and non-lean categories, the resultant substantial over-laprecluded separate analysis according to this division. An exam-le of the potential influence of involvement in lean sports mayxplain the cross-loading of the Reaction to Prescribed Weighing

tem for both athletic groups. Cross-loading suggests that this is a

ore complex item, related to more than one construct, thus forompetitive athletes this item may have a double-salience that isot present for individuals not actively competing at this level. This

e 10 (2013) 399– 405

study attempted to control for the influence of current involvementat competitive and elite levels since to divide by sport would nothave yielded the numbers sufficient to conduct the analysis. Fur-ther research could investigate the performance of this measureamong those who are immersed in single specific sport types suchas competitive gymnastics or wrestling.

In summary, we present a sequential CFA and EFA of the EDE-Qin the largest non-clinical sample of adults to date, providing a sub-stantial sample of both males and of competitive athletes. For allgroups, weight and shape concerns showed significant overlap inconstruct. We found that a three-factor solution was the best fit forthe data in all groups other than comparison males. These malesmay be best served by a different measure, and future researchshould consider both gender and activity level when examining eat-ing disorder behaviors and screening tools in university students.

References

Anderson, D. A., & Williamson, D. A. (2002). Outcome measurement in eating disor-ders. In W. W. IsHak & T. Burt, et al. (Eds.), Outcome measurement in psychiatry: Acritical review (pp. 289–301). Washington, DC: American Psychiatric PublishingInc.

Bentler, P. M. (1995). EQS structural equations program manual. Encino, CA: Multi-variate Software.

Binford, R. B., Le Grange, D., & Jellar, C. C. (2005). Eating Disorders Examinationversus Eating Disorders Examination Questionnaire in adolescents with full andpartial syndrome bulimia nervosa and anorexia nervosa. International Journal ofEating Disorders, 37, 44–49.

Bohn, K., Doll, H. A., Cooper, Z., O’Connor, M. E., Palmer, R. L., & Fairburn, B. (2008). Themeasurement of impairment due to eating disorder psychopathology. BehaviorResearch and Therapy, 46, 1105–1110.

Braun, D., Sunday, S., Huang, A., & Halmi, K. (1999). More males seek treatment foreating disorders. International Journal of Eating Disorders, 25, 415–424.

Brown, T. A. (2006). Confirmatory factor analysis for applied research. New York: TheGuilford Press.

Byrne, S. M., Allen, K. L., Lampard, A. M., Dove, E. R., & Fursland, A. (2010). Thefactor structure of the Eating Disorder Examination in clinical and communitysamples. International Journal of Eating Disorders, 43, 260–265.

Byrne, S., & McLean, N. (2001). Eating disorders in athletes: A review of the literature.Journal of Science and Medicine in Sport, 4, 145–159.

Costello, A. B., & Osborne, J. W. (2005). Best practices in exploratory factor anal-ysis: Four recommendations for getting the most from your analysis. PracticalAssessment, Research & Evaluation, 10, 1–9.

Darcy, A. M., Celio Doyle, A., Lock, J., Peebles, R., Doyle, P., & Le Grange, D. (2012).The Eating Disorders Examination in adolescent males: How does it compare toadolescent females? International Journal of Eating Disorders, 45, 110–114.

Fairburn, C. G., & Beglin, S. J. (2008). Eating Disorder Examination Questionnaire(EDE-Q 6.0). In C. G. Fairburn (Ed.), Cognitive behavior therapy and eating disorders.New York: Guilford Press.

Fairburn, C. G., & Cooper, P. J. (1993). The Eating Disorder Examination. In C. G.Fairburn & G. T. Wilson (Eds.), Binge eating: Nature, assessment and treatment(12th ed., pp. 317–360). New York: Guilford Press.

Grilo, C. M., Masheb, R. M., & Wilson, G. T. (2001). A comparison of different meth-ods for assessing the features of eating disorders in patients with binge eatingdisorder. Journal of Consulting and Clinical Psychology, 317–322.

Hayton, J. C., Allen, D. G., & Scarpello, V. (2004). Factor retention decisionin exploratory factor analysis: A tutorial on parallel analysis. OrganizationalResearch Methods, 7, 191–205.

Horn, J. L. (1965). A rationale and test for the number of factors in factor analysis.Psychometrika, 30, 179–185.

Hu, L., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covarience struc-ture analysis. Conventional criteria versus new alternatives. Structural EquationModelling, 6, 1–55.

Luce, K. H., & Crowther, J. H. (1999). The reliability of the Eating DisorderExamination-Self-Report Questionnaire Version (EDE-Q). International Journalof Eating Disorders, 25, 349–351.

Mannucci, E., Rica, V., Di Bernardo, M., Moretti, S., Cabras, P. L., & Rotella, C. M. (1997).Psychometric properties of the EDE 12.0D in obese adult patients without bingeeating disorder. Eating and Weight Disorders, 2, 144–149.

McCreary, D. R., & Sasse, D. K. (2000). An exploration of the drive for muscularity inadolescent boys and girls. Journal of American College Health, 48, 297–304.

Mond, J. M., Hay, P. J., Rodgers, B., & Owen, C. (2004). Validity of the Eating Dis-order Examination Questionnaire (EDE-Q) in screening for eating disorders incommunity samples. Behavior Research and Therapy, 42, 551–567.

O’Connor, B. P. (2000). SPSS and SAS programs for determining the number of

components using parallel analysis and Velicer’s MAP test. Behaviour ResearchMethods, Instrument Computers, 32, 396–402.

Pasman, L., & Thompson, J. K. (1998). Body image and eating disturbance in oblig-atory runners, obligatory weightlifters, and sedentary individuals. InternationalJournal of Eating Disorders, 7, 759–769.

y Imag

P

P

P

S

T

A.M. Darcy et al. / Bod

assi, V. A., Bryson, S. W., & Lock, J. (2002). Assessment of eating disorders inadolescents with anorexia nervosa: Self-report questionnaire versus interview.International Journal of Eating Disorders, 33, 45–54.

ennings, C., & Wojciechowski, F. L. (2004). The Eating Disorder Questionnaire(EDE-Q): Netherlands normative scores for anorexic patients and a non-eatingdisorder control group. Gedragstherapie, 37, 293–301.

eterson, C. B., Crosby, R. D., Wonderlich, S. A., Joiner, T., Crow, S. J., & Mitchell, J. E.(2007). Psychometric properties of the eating disodrer examination question-naire: Factor structure and internal consistency. International Journal of EatingDisorders, 40, 386–389.

atorra, A., & Bentler, P. M. (1994). Corrections to test statistics and standard errors

in covariance structure analysis. In A. von Eye & C. C. Clogg (Eds.), Latent variableanalysis: Applications for developmental research (pp. 399–419). Thousand Oaks,CA: Sage.

ucker, L. R., & Lewis, C. (1973). A reliability coefficient for maximum likelihoodfactor analysis. Psychometrika, 38, 1–10.

e 10 (2013) 399– 405 405

Velicer, W. F. (1976). Determining the number of compoonents from the matrix ofpartial correlations. Psychometrika, 41, 321–327.

Wade, T. D., Byrne, S., & Bryant-Waugh, R. (2008). The Eating Disor-der Examination: Norms and construct validity with young and mid-dle adolescent girls. International Journal of Eating Disorders, 41, 551–558.

Wilfley, D. E., Schwartz, M. B., Spurrell, E. B., & Fairburn, C. G. (1997).Assessing the speific psychopathology of binge eating disorder patients.Interview or self-report? International Journal of Eating Disorders, 35, 1151–1159.

Wilson, G. T. (1993). Assessment of binge eating. In C. G. Fairburn & G. T. Wilson

(Eds.), Binge eating: Nature assessment and treatment (pp. 227–249). New York:Guilford Press.

Wolk, S. L., Loeb, K. L., & Walsh, B. T. (2005). Assessment of patients with anorexianervosa: Interview versus self-report. International Journal of Eating Disorders,37, 92–99.