The Estimation of Body Mass Index and Physical Attractiveness is Dependent on the Observer's Own Body Mass Index

Author(s): Martin J. Tovee, Joanne L. Emery, Esther M. Cohen-Tovee Reviewed work(s): Source: Proceedings: Biological Sciences, Vol. 267, No. 1456 (Oct. 7, 2000), pp. 1987-1997 Published by: The Royal Society Stable URL: http://www.jstor.org/stable/2665687 . Accessed: 18/04/2012 00:32
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

The Royal Society is collaborating with JSTOR to digitize, preserve and extend access to Proceedings: Biological Sciences.

http://www.jstor.org

THE ROYAL iJS OCIETY

doi10.1098/rspb.2000.1240

The estimation of body mass index and physical attractiveness is dependent on the observer's own body mass index
I, Martin J. Tove&e Joanne L. Emeryl and Esther M. Cohen-Tovee12
University Newcastle, of of Ridley Building, Newcastle NE] Upon 'Department Psychology, T7yne 7RU, UK MentalHealthTrust, George's St of and Northumberland Hospital, Morpeth, 2Department Psychological Therapies Research, UK Northumberland 2NLU, NE61 A disturbance in the evaluation of personal body mass and shape is a key featureof both anorexia and bulimia nervosa. However, it is uncertainwhetheroverestimation a causal factorin the developmentof is these eating disorders or is merely a secondary effectof having a low body mass. Moreover, does this overestimation extend to the perception of other people's bodies? Since body mass is an importantfactor in the perception of physical attractiveness, wanted to determinewhetherthis putative overestimation we of self body mass extended to include the perceived attractivenessof others. We asked 204 female observers (31 anorexic, 30 bulimic and 143 control) to estimatethe body mass and rate the attractiveness of a set of 25 photographicimages showing people of varyingbody mass index (BMI). BMI is a measure of weightscaled forheight (kg m-2). The observersalso estimatedtheirown BMI. Anorexic and bulimic observerssystematically overestimatedthe body mass of both theirown and other people's bodies, relative When the degree of to controls,and they rated a significantly lower body mass to be optimallyattractive. overestimationis plotted against the BMI of the observer there is a strongcorrelation.Taken across all our observers, as the BMI of the observer declines, the overestimationof body mass increases. One possible explanation for this resultis that the overestimationis a secondary effect caused by weight loss. Moreover, if the degree of body mass overestimationis taken into account, then there are no significant differencesin the perceptions of attractivenessbetween anorexic and bulimic observers and control observers. Our results suggest a significantperceptual overestimation of BMI that is based on the observer's own BMI and not correlated with cognitive factors,and suggeststhat this overestimationin regimes. eating-disordered patientsmustbe addressed directlyin treatment Keywords: anorexia nervosa; bulimia nervosa; body-mass estimation; body image; attractiveness; body mass index 1. INTRODUCTION Eating disordersare an increasing problem in the female population. The proportion of women who sufferfrom these conditions continuesto rise and currenttherapeutic regimes have only a limited success in treating these conditions (e.g. Garner & Garfinkel1997; Fairburn et al. 1999), particularly in anorexia nervosa where the longterm mortalityrate is more than 10% (DSM-IV, American Psychiatric Association 1994). To be able to treat these conditions more effectively need a better underwe standing of their central features. A key feature of the diagnostic criteriaforboth anorexia nervosa and bulimia nervosa seems to be a distorted evaluation of personal body mass and size (DSM-IV, American Psychiatric Association 1994). Many researchers have suggested that there are two components of body-image dysfunction: a perceptual body-size distortionand a'cognitive-evaluative' dysfunction (e.g. Cash & Deagle 1997; Gardner 1996; Slade 1988; Cash & Brown 1987). The perceptual distortionis defined whenan observeris unable to gauge herbody size accurately and the cognitive distortion is when an observer can accurately estimate her size but may be dissatisfiedwith her size, shape or some other aspect of her appearance
*Authorforcorrespondence (mj.tovee(ncl.ac.uk).

(Cash & Deagle 1997; Gardner 1996). In this study we focus primarily on the perceptual distortion,which has to been difficult quantifyreliably (e.g. Slade 1988, 1994). It is uncertain whether overestimationis a causal factor in the developmentof these eating disordersor is merelya secondary effectof having a low body mass. If overestiof mation is a secondary effect weight loss, then it might be expected to develop as body mass declines. If women of a range of body-mass values (fromemaciated to obese) are tested forbody-mass estimation,it should be possible to determine whether or not the degree of estimation is linked to the body mass of the subject. we Additionally, will determinewhetherthe overestimation of body size is specificto the observer'sown body or whetherit extends to estimatingthe size of other women's is a bodies. If the overestimation purelya perceptualdeficit, problem with making the fine within-categoryjudgements necessary to assign a body to a particular bodymass value, then one might expect this overestimationto extend to the perceptionof other women'sbodies. of The putative overestimation body mass has implications for the perception of physical attractiveness.Body mass is usually measured in termsof the body mass index (BMI), which is a measure of weight scaled for height with units of kg m-2 (Bray 1978). Most studies suggest in thatthe BMI ofa subjectis an importantfactor how both men and women perceive female physical attractiveness
?) 2000 The Royal Society

Proc. Soc.Lond. (2000) 267, 1987-1997 R. B

Received May 2000 8 13 Accepted June 2000

1987

1988

M. J.Tovee and others Estimation BMI depends observer's BMI of on own

Table

The observers estimated bodymass on a marked linearscale to withtherelevant corresponding theBMI scale and annotated 2. METHODS labels: emaciated, underweight, and normal,overweight obese. The position the markwas thenmeasured the scale and on Observers(31 femaleanorexic,30 femalebulimicand 143 of femalecontrol)ratedthe size and attractiveness a set of 25 as BMI (BMIeSt)allowingready comparexpressed estimated of colour photographic to BMI images of women.The eating-disordered isonwiththe actualvalues.We askedobservers estimate becausewe wantedto use a measure thansimpleweight patientswere recruited fromthe Eating Disorder In-Patient rather Serviceat the RoyalVictoriaInfirmary, Newcastle Upon Tyne, ofbodymassscaledto thesize ofthebody.One can be tall and and were diagnosedon the basis of DSM-IV (AmericanPsy- thinor shortand fat,and stillhave the same weight. felt, We that of therefore, BMI is a better chiatric Association were age matchedand representation thedegreeoffat 1994).The controls drawn fromGP lists in the same geographical depositionon a body. Previous studies have suggestedthat areas as our female observersshould be very accurate in judging other eating- disorderedsubjects.The controlshad no historyof women'sBMI and thereare severalputative visual cues they and eatingdisorders scoredwithin normalrangeon a setof the screening questionnaires the including Beck DepressionInven- could use (Toveeetal. 1998,1999;Tovee & Cornelissen 2000). is the ratio(which correlated with tory(BDI), the Beck AnxietyInventory (BAI), the Rosenberg These include perimeter-area Self-EsteemScale (RSE), the Body Shape Questionnaire BMI at betterthan 0.97) or just simple relativelower-body widths (Tovee et al. 1999; Tovee & Cornelissen 2000). For Beliefs (BSQ), theEatingDisorder Questionnaire (EDBQ) and waistwidthis correlated withBMI at better the Eating Disorders Examination Questionnaire(EDEQ) example,relative than0.95. So usingthesesimple visualcues it shouldbe possible (Becketal. 1961; Rosenberg 1965;Cooperetal. 1987;Cooperetal. & and 1997;Fairburn Beglin1994).A summary thedata is shown tojudge BMI accurately, as can be seen from of figure the 2, in table 1. All observers matchvery closelythe testedwithinthe normal range for BMIest values fromcontrolobservers visualacuity actual BMI valuesoftheimages.Observers ratedattractiveness usingtheSnellenchart. To generate on a scale of0 (leastattractive) 9 (mostattractive). order The the images forrating, to womenwere consenting of and subjects videotaped standingin a set pose at a standarddistance, of presentation the 25 imageswas randomized withthe entire twice.The first through run set wearingclose-fitting leotardsand leggings. grey Images were werepresented was used to makeobservers thenframe-grabbed storedas 24-bitcolour pictures. aware ofthe rangeofvariability and of The in 11.4 imagesvariedin BMI from to 34.3. Fiveoftheimagescame bodyfeatures represented theimages.Onlyon thesecondrun each oftheBMI categories from wereobservers askedto ratethem. through (Bray1978):emaciated(below 15),underweight (15-20), normal (20-25), overweight (25-30) and obese (above 30). Examplesof theseimagesare shownin 3. RESULTS 1. figure The waist-hipratio (WHR) of the imageswas kept a within comparatively (a) The estimation of other women's BMI narrow 0.74 to 0.93,witha range(from mean of 0.82, s.d.= 0.05) to minimizeany potentialeffect Most control observers were able accurately and of WHR on the size or attractiveness The faces of the consistentlyto estimate the BMI of the women in the ratings. imageswereobscured. additionto our set of 25 images,the In 25 pictures (figure 2). When estimating the BMI of the observers estimated body mass of an image of themselves. 25 images, the estimates show only slightly reduced the This allowsus to makea comparison between estimation the accuracy at the extremes of the BMJ range (i. e. the of observers were generally good at estimating the BMJ of the observer's own image and the estimation otherwomen' of female images across the range of BMJ values). As the bodies.
Proc.R. Soc. Lond. B (2000)

(e.g. Singh 1993, 1994; Henss 1995; Furnham et al. 1997; Tovee et al. 1998, 1999, 2000; Tovee & Cornelissen 1999, 2000). Thereforethe appraisal of another person's attractiveness requires an observer to estimate accurately the BMI of a subject. An overestimationof another person's BMI would systematically shiftan observer's perception of that person'sbody attractiveness, just as an overestimation of the observer's own BMI would shifther perception of her own attractiveness. For example, if an observer has a BMI of 20, which is regarded as more attractive than a BMI of 22, then an overestimationby two BMI units could result in the observer believing herselfto be more unattractive than she is. This could produce a strong pressure on the observer to reduce her BMI throughdieting to increase her own perceived attractiveness. To explore these questions we have asked 204 female observers (including anorexic and bulimic observers), whose body mass ranged fromemaciated to obese, to rate a set of 25 picturesof women forBMI and attractiveness. They also estimated their own BMI from a picture of themselves (they did not rate their own attractivenessas this was considered too stressful the eating-disordered for observers).

? 1. Descriptive statistics (givenas mean s.d.) for the three observer groups

(BMI, body mass index; BDI, Beck depressioninventory; BAI, Beck anxietyinventory; RSE, Rosenberg self-esteem; EDBQ, eating disordersbeliefsquestionnaire;BSQ, body shape, EDEQ-WC, the weightconcernsubscaleof the eating disordersexaminationquestionnaire;EDEQ-SC, the shape concern subscale of the eating disorders examination
questionnaire.)

controls age BMI BDI BAI RSE EDBQ BSQ EDEQ-WC EDEQ-SC age ofonset duration illness of 28.3 ? 9.2 22.5+?3.7 5.8 + 6.4 8.1 + 7.6 30.5+?4.8 18.2+ 15.4 79.4+?29.1 8.9 + 8.9 12.9+ 9.8

anorexics

bulimics

27.2 ? 8.0 27.5 + 7.3 14.9+ 1.9 23.5+?5.7 28.4 + 13.8 22.2+ 12.3 21.9 + 13.4 20.5 + 14.4 17.9+?4.7 21.9+?4.7 65.4+ 19.7 58.6I21.1 133.4+?34.7 147.5+?34.1 23.9 + 7.6 26.5 + 7.1 32.9 + 7.8 33.4 + 6.9 18.7+ 6.1 17.7 + 4.0 8.5 + 6.6 9.8 + 6.4

Estimation of BMIdepends observer's BMI on own

M. J. Toveeand others 1989

1. one The imageon thefarleft Figure Examplesoftheimagesused in thisstudy.Each image is from ofthefiveBMI categories. *. IS~~~~~~~~~~~~~~~~ 0 isfrom emaciatedcategory the and has a BMI of 13.52. The nextimageis from underweight the and has a BMI of category 17.34. The nextimage is from normalrangeand has a BMI of22.07. The fourth the the and image is from overweight category hasa BMI of28.01. The finalimage is from obese category the and has a BMI of34.05.
.L,.,

~
32

(a)

'.

(b)

+'

U~~~~~~~~~~~~~~~

20

12
8

...

8. (d) 32 28
24
U

(c)

.
0

20

..~

____bulimics

16
12

10

14

18

22 actual

26 BMI

30

34

38

10

14

18

22 actual

26 BMI

30

34

38

Figure Estimated BMI ofeach image plottedagainsttheactual BMI oftheimagefor(a) control 2. observers, bulimic (b) The dottedline indicatestheequalityline (i.e. theline upon whichestimations observers (c) anorexicobservers. and wouldfall ifthey werecompletely we The accurate). (d) For comparison purposes, showtheplotsofall threeobserver groupstogether. estimates theanorexicobservers represented filled are circlesand a dottedline,thebulimicobservers open squaresand by by by a solidline,and thecontrols crosses by and a dashedline.

estimates the observers all threegroupsseem to by in three groups,the dip at the extremes should average show same pattern(i.e. a slight the 'dip' at the extremes out. The control in observers very were oftheBMI rangeof the images),whenwe comparethe accurate their estimate averageestimations the BMI by observersin the of theBMI (BMIest)of the 25 images,overestimating of on
Proc. Soc.Lond. (2000) R. B

1990

of on own M. J. Tovee and others Estimation BMI depends observer's BMI

5 -(a) 4 -(c
0
0

(b)

3-

0i

(b

~
25 30
0

~
1S

'

I'

'I'

-2-

0 0 1S 20

0o

35 3304 4010

25

~ ~~ ~ ~ ~

~~?
+

+~~~~~~~~~~~~

~~~1
a

~
0.

~~

~0

00

-2

0~~~~~~~~~~~~~~~~~
ano .............. rexi...b
15 20

b -3ic (b) ulim

e ......................................... 25
3 0 lte agis gis lte fosres 5 h h 0 beve' bevrsB ahpitrpeet 5

f e aver ge .o e ....................th.th .2
30 35 4

103c4

(b 4blii obeves

Figure3.Th avrg+vrsiaino vrsiaino vrg 3. Figure Th + (c)

h M fte2 iue iue fte2 M h anorexicssresad()altregop

M Ifr()controloseres o a controloseres h vrg fte2

judgements made by a particular observer. As can be seen, there is a continuum of error in body-mass estimates with changiug observer BMI.

average by only 0.35 BMIest units. However, the anorexic observers overestimatedthe BMIest of the images by an average of 2.29BMIest units and the bulimic observers overestimatedby 0.99BMIest units. These values are all significantly different fromeach other (one-wayANOVA, = 15.92, p < 0.0001; Post Hoc test, Games-Howell F2,201 p < 0.001). However, the overestimation does not represent a qualitative difference between anorexic and bulimic observers and control observers. Instead the degree of overestimationseems to be inverselyproportional to the BMI of the observer.This can be illustratedby plotting the average overestimationby each observer against the observer's own BMI (figure 3). There is a significant and the observer's correlationbetween the overestimation own BMI (Pearson correlation, d.f.=202, r=-0.59, p <0.0001). As the BMI of the observer declines, the overestimateof the BMI of another person rises.The estimates by the anorexic and the bulimic observers can be
Proc.R. Soc. Lond. B (2000)

seen as part of a continuum based on observer BMI, rather than as qualitativelyseparate estimates.This relationship can also be seen within the individual observer correlationbetween the estigroups.There is a significant mated BMI and the observer'sown BMI forboth control and bulimic observers (controls, d.f.=141, r=-0.591, p<0.0001). bulimics, d.f.=28, r=-0.586, p<0.0001; The data from the anorexic observers show the same trendbut the correlationdoes not reach statisticalsignificance (d.f.= 29, r= - 0.203, p = 0.273). The fact that the correlation does not reach significance in the anorexic group may be partially due to the fact that these observers all fall within a narrow range of BMI values, compared to the BMI values of the other observersin the controland bulimic groups. (b) The estimation ofpersonal BMI We also asked our observers to estimate the BMI of their own bodies. The controls overestimated by

Estimation BMI depends observer's BMI of on own

1.42BMIest units, the bulimics by 2.43BMIest units and

M. J.Tovee and others 1991

the anorexics by 4.28BMIest units. The overestimations by the three groups are all significantly different(oneway ANOVA, F2201= 33.98, p < 0.0001; Post Hoc test, Games-Howell show that anorexics and bulimics are differentfrom controls at the level of p < 0.0001 and anorexics are different frombulimics at p < 0.05). These overestimatesof own body are all significantly higher for each observer group than the average overestimatesfor other women's bodies (independent-samples t-test assuming unequal variances, control observers, t=4.73, d.f.=279, p < 0.0001; bulimic observers, t= 2.09, d.f.=56, p < 0.05; anorexic observers, t=3.14, d.f.=56, p < 0.005). Again these overestimatesmay not actually representqualitative differences between control,bulimic and anorexic observers. Instead, if one correlates the overestimates against the BMI of the observers in all three observer groups, then the degree of overestimation inverselycorrelated with the BMI of the is significantly observer (Pearson correlation, d.f.=202, r=-0.629, p < 0.0001). If one looks at this relationship in each observer group separately, degree of overestimation is inverselycorrelated with the observer BMI for both the control and bulimic observers (controls, d.f.= 141, r= -0.697, p < 0.0001; bulimics, d.f.=28, r= -0.807, the overestimap < 0.0001) (figure 4a,b). Interestingly, tion is actually positivelycorrelated with observer BMI in the anorexic group (anorexics, d.f.= 29, r= 0.496,

no obvious separation into subgroups based on overestimationof other women's bodies. The degrees of overestimation in the two subgroups in this case are not significantly different (BMI > 14 subgroup overestimated by 2.49 BMIest units, the BMI < 14 subgroup by 2.18BMIest units; independent-samples t-test, d.f.=29, t= 0.864, p = 0.416). The two subgroups are also not significantly different(at the level of p < 0.05) on any other variable such as the questionnaire results (BSQ, BDI, EDBQ, RSE, EDEQ shape-concern scale, EDEQ weight-concern scale, EDEQ eating-concern scale), average age, age of onset or duration of the condition. (c) The corrected estimate of personal BMI To separate a general overestimationfroman overestimation specific to the observer's own body for each observer we decided to subtractthe their overestimateof other women's bodies fromtheir overestimateof personal BMI. As there is a'dip' in accuracy in the estimation of BMI values at the extremeends of the range, we decided not to subtractthe average error in estimationfor all 25 images. Instead, we subtracted the average for the five women in the same BMI category as each observer. For example, for an observer with a BMI in the emaciated range we subtractedtheir average estimation for the five images in the emaciated range, and so on. This produced values of 0.99BMIest units for controls, 1.12BMIest units for bulimics and 2.24BMIest units for anorexics. These values are again significantlycorrelated with personal BMI across all our observer groups (Pearson correlation, d.f.= 202, r -0.481, p < 0.0001) and show a broad continuum of overestimation with changing BMI. However, looking at each observergroup separatelyshows that although thisrelationshipholds forthe bulimic obserand controls vers (d.f.=28, r=-0.701, p <0.0001) (d.f.=143, r=-0.610, p <0.0001), there is a positive relationshipbetween overestimationand observer's BMI foranorexic observers(d.f.= 29, r= 0.573,p < 0.001). For the two anorexic subgroups, the pattern is unchanged from the unadjusted values. The BMI > 14 subgroup overestimateson average by 3.8BMIest units, and the BMI < 14 subgroupis more accurate, theyslightly underestimate by on average 1.0BMIest unit. The two in different the accuracy of subgroups are significantly their estimation (independent-samples t-test, d.f.= 29, t= -4.92, p < 0.0001). The correlation between the accuracy of estimation and the observer's BMI is not for significant the BMI > 14 subgroup, (d.f.= 19,r= 0.105, p = 0.649) or forthe BMI < 14 subgroup,although thereis a strong trend toward a negative correlation (d.f.=8, r= -0.539, p = 0.108). (d) Perceptual and cognitive measures In addition to estimatesof BMI, our subjectscompleted several questionnaires designed to measure dissatisfaction with personal body size or shape. The results of these measures do not seem to be correlated with the error in estimatingBMI. First, we correlated the accuracy of the estimates of other women's BMI by our three observer groups with the BSQ, the EDBQ and all seven subscales of the EDEQ (including the shape- and weight-concern subscales). None of the correlationsreached the p < 0.05 level of significance.We then repeated the process, but

p=

0.005). In figure 2, we have already illustrated that there is a 'dip' in the estimation of other women's BMI at the extreme ends of the BMI range (i.e. an observer tends to underestimate the BMI of very low or very high BMI subjects). So one might expect anorexic subjects (who have very low BMI values) to underestimate their own BMI values. However, although there is a 'dip' in the overestimation of personal BMI at low BMI values, most of the anorexic observers still overestimate their own BMI (figure 4c). Only four out of 30 anorexic observers actually underestimated their own BMI. Careful inspection of the estimates by anorexic observers of their own BMI suggests the existence of two distinct data clouds (i.e. two separate subgroups). One suibgroup makes higher overestimates (observers whose BMI falls between 14 and 18) and one subgroup makes more accurate estimates (observers whose BMI is below 14). The latter group of 21 observers overestimate BMI, on

average, by 1.1BMIest units and the formergroup of ten observers overestimate by 5.3 BMIest units. The two estimatesare significantly different (independent-samples t-test, d.f.=29, t= -4.268, p < 0.0001). If one treats these two subgroups as a single population, then there is a positive correlationbetween overestimationof personal BMI and the observer'sBMI, as the BMI < 14 subgroup is significantlymore accurate. However, this does not representa smooth change in accuracy of personal BMI estimationacross the anorexic population as a whole, but a sharp discontinuity between the two subgroups. If one examines the correlationbetween observer'sBMI and the accuracy of estimation within these two subgroups separately,then there is no significant correlation for the BMI > 14 group (d.f. = 19, r= 0.069, p = 0.766) but there group (d.f.=8, r= -0.706, p < 0.05). However, there is
Proc.R. Soc. Lond. B (2000)

is a significant negative correlation for the BMI < 14

1992 M. J.Tovee and others Estimation BMI depends observer's BMI of on own

(a) 80 0 0Q9

(b)

a~~~~~~~~
w 0

\'~
0

f,

0~~~~~~~
?

-8

(c) 8 -4 4
(bo)U
+c

+
+ +0+

(d)
*d
+

+
K ?

+ anorexics
?

+ control+ o +
+ ++ +

as
a
0

-8

Ocontrols s bulimics

ei
+

+ s w
-W
+ + +

+ +0

0 0

~~~~~~+
+ + 4 -

~ ~ 00 0 ~~~~~~~~~~~~~~0

o~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
0 0

-8
0

10

15

20

25 30 BMI observer

35

40

10

15

20

30 25 observer BMI

35

40

Figure 4. The overestimation of the observer's own BMI plotted against the observer's actual BMI for (a) control observers, (b) bulimic observers, (c) anorexic observers and (d) all three observer groups. Again, there seems to be a continuum of error in observer BMI. body-mass estimates with changZingZ

correlatingthe questionnaire resultsagainst the accuracy of estimatingpersonal BMI foreach of our three observer groups. Once again, none of the correlationsreached the p < 0.05 level of significance. (e) Accuracy of estimation and the mood of the observer It is possible that the apparent change in the accuracy estimationof both personal BMI and other women's BMI with the observer'sown BMI is due to other factors,such as changing mood (e.g. Cohen-Tovee 1993). For example, if an observeris depressed,her performanceat estimating BMI may suffer.However, the results of the questionnaires designed to assess mood do not correlate with the error in estimating BMI. First, we correlated the accuracy of the estimatesof other women's BMI by our three observer groups with the BDI, the BAI and the RSE. None of the correlations reached the p < 0.05 level of XVe significance. then repeated the process, but correlating the questionnaire results against the accuracy of estimating personal BMI for each of our three observer
Proc.R. Soc. Lond. B (2000)

groups. Once again, none of the correlationsreached the p < 0.05 level of significance. (f) A neural substratefor the overestimation? of It is possible that the increasingoverestimation BMI with decreasing observerBMI is linked to poor nutrition. The poor nutritional intake that leads to a lower BMI may also lead to the damage of cortical neurons and a lack of neurotransmitter substrates.Structuralimaging of anorexic and bulimic subjects has shown irreversible brain shrinkage,which would be consistentwith neuronal cell death (e.g. Lankenau et al. 1985; Krieg et al. 1989; Lauer et al. 1990; Lambe et al. 1997). Given that up to 50% of the human cortex is involved to some degree in the processing and analysis of visual information (e.g. Drury et al. 1996), it would be expected that the neural changes documented in low-BMI subjectswould manifest themselvesin their perceptual analysis of the visual cues to a body's BMI. If this is true, then one mightexpect the duration of the disease in anorexics or bulimics to be correlated with the magnitude of overestimation.

Estimation BMI depends observer's BMI of on own However, all the correlations between illness duration and the overestimation of other women's BMI and personal BMI failed to reach the p < 0.05 level of significance forboth anorexic and bulimic observers.

M. J.Tovee and others 1993

curvaceous body being regarded as more attractive.The range of body shapes in this study was comparatively narrow compared to the range of BMI values for the body images. However, there was some variation in shape. The WHR varied from0.74 to 0.93, the waist-bust ratio (WBR) (a measure of upper-body shape) varied (g) Attractiveness and BMI BMI is an important component in the attractiveness from 0.83 to 1.16 and the bust-hip ratio (BHR) (a of the female body (e.g. Tovee et al. 1998, 1999, 2000; measure of whether the body is an 'hour glass' shape or not) varied from 0.79 to 0.96. Although none of these Tovee & Cornelissen 1999, 2000) and inaccuracies in the estimation of BMI have serious implications for the features reached significancein a multiple regression to determine their relative importance for ratings of attracperception of attractiveness. If the attractiveness of a body is dependent to some degree on an observer's estitiveness, theyare all significantly correlatedwiththeratings. mate of her BMI, then a systematicoverestimation of This is true of all three groups of observers.There is a BMI will in turn systematicallyshiftthis perception of significant correlation of attractiveness with WHR (Pearson correlations, controls, d.f.=141, r=-0.302, attractiveness. This is illustratedby the observers' attracp <0.0001; p < 0.0001; bulimics, d.f.= 28, r=-0.335, tivenessratingsfor our set of 25 images. If one plots the attractiveness ratingsagainst the real BMI of the images, anorexics, d.f.= 28, r= - 0.349, p < 0.0001), with WBR the shape of the functionis the same for all observers. (controls, d.f.=141, r=-0.326, p < 0.0001; bulimics, d.f.=28, r=-0.377, However, the position of the peak of the curve (i.e. the p <0.0001; anorexics, d.f.=28, most 'attractive' BMI) differsbetween individuals. To d.f. r=-0.443,p < 0.0001) andwithBHR (controls, = 141, r= 0.129, p < 0.001; bulimics, d.f.= 28, r= 0.178, we polynomial quantifythis difference, used a third-order regressionto estimate the optimally attractive,or 'peak' p < 0.0001; anorexics, d.f.= 28, r= 0.282, p < 0.0001). BMI, for each subject's ratings (Tovee et al. 1999). If one This suggeststhat all three groups prefera more curvafirstcompares the average peak values for the anorexic, ceous body shape. To explore further whether there are differences bulimic and control observers,the values are significantly groups we used different(one-way ANOVA, F2,200 = 46.923, p < 0.0001). between observers in the three different The peak for the control observersis at 20.62 BMI units, multiple regressionto carry out three pairwise comparisons between groups. In the models the group factorwas which is significantlyhigher than that for both the bulimic observers at 19.84 BMI units and the anorexic classifiedusing dummy variables. There was no effectof group forWHR, WBR or BHR at the p < 0.05 level of observers at 17.98BMI units (Post Hoc tests, GamesHowell and Tamhane, p < 0.001). The peak values forthe significance. These results would be consistent with a anorexic and bulimic observers are also significantly common internal representationof what constitutesan attractivebody across all threegroups. It is only the overdifferent from each other (PostHoc tests, Games-Howell and Tamhane, p < 0.001) (figure 5a,b). These differences estimation of body mass that shiftsthe apparent preferin attractiveness ences of anorexic and bulimic observers. seem to be based on the BMI of the individual observerswithinthe groups, ratherthan a qualitative difference between groups. If one plots the position of 4. DISCUSSION the peak of the curve against the BMI of each observer, of The resultsshow a clear patternof overestimation the then the two are correlated (Pearson correlation, BMI of otherwomen'sbodies by all threeobservergroups. d.f.= 201, r=0.550, p < 0.0001) (figure 6a). If one now This overestimation inversely is proportionalto the obserplots the attractiveness ratingsfor the images not against ver's own BMI. So, although both anorexic and bulimic their real BMI, but against the BMI estimated by each observers significantlyoverestimated compared to the observer, the differencesin the positions of the curves a controlobservers,thismay not represent primaryfeature disappear (figure 5c,d). The peak values are now no of the anorexic and bulimic observers'clinical condition. longer significantly different controls (21.17),bulimics for Instead, it may be an extensionofa trendtowardsoveresti(21.00) and anorexics (21.05) (one-way ANOVA, = 0.487, p = 0.615). If one now plots the position of mation of BMI with declining personal body mass that is F2,200 the peak of the curve based on estimated BMI against already evidentwithinthe controlgroup. In bulimic and control observers,the magnitude of the each observer's BMI, then the two are no longer correoverestimationof personal BMI is also inversely correlated (Pearson correlation,d.f.= 201, r= 0.078, p = 0.272) lated with the observer'sown BMI. The situation is more (figure6b); i.e. the BMI of the observerno longer predicts the BMI they find most attractive.This suggeststhat all complex in anorexic observers.The average magnitude of of overestimation by anorexic observers is significantly three groups of observersagree on the attractiveness a seem to arise from higher than for bulimic or control observers, which is given BMI, the apparent differences consistent with previous studies (Slade 1988, 1994; errorsin estimatingthe BMI of the women in our set of Garner & Garfinkel 1997). However, across the whole 25 images. Once the errors in estimation have been in anorexic observer group, the trend of overestimationis controlled for,there are no longer any differences the downward with decreasing BMI, rather than upward as attractiveness preferences. is found in the other observergroups. This relationshipis true forboth forthe 'raw' estimationsand forthe estima(h) Attractiveness and shape tions corrected for the observer's estimationof five other Body shape is also regarded as a significantcue to attractiveness (e.g. Singh 1993, 1994; Henss 1995; bodies in the same BMI category (i.e. corrected for the Furnham al. 1997; et Toveeet al. 1997,1998,1999): a more accuracy of their general estimation of people with the
Proc.R. Soc. Lond. B (2000)

1994

M. J.Tovieeand others Estimation BMI depends observer's BMI of on own

(a)

(b)
0
+

7
60

a 51 a 3
2

~~~0
0
0

++

l~~~~~~~
4

~ ~0 ~ ~ ~ ~ ~ 0 ~ ~~~~~~~~~~+4 ~ ~ ~
*

0~~~ 0~~~~

og

0'

*
0

10
(c)
7-

15

20

BMI

25

30

35

10 (d)

15

20

BMI

25

30

35

Fiur

4.

0trcivns

0 asafnto0fteata

~~~0

fec

fte2

iae,a

ae

y()aoei

bevroe

0~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~.

C,)

a
0

15

20

001

circles) and control observers (filled circles), and (b) bulimic observers (crosses) and control observers (filled circles). Each point representsthe average rating by the observers in each observer group. A regressionline for each functionis superimposed. Attractiveness as a functionof the estimated BMI of each of the 25 images, as rated by (c) anorexic observers (open circles) and control observers (filled circles), and (d) bulimic observers (crosses) and control observers (filled circles). Each point represents the average rating by the observers in each observer group. A regressionline for each functionis superimposed.

same BMI). So the change in the pattern of overestimation in anorexic observers (relativeto bulimic and control observers)does not seem to be due to changes in the accuracy of estimatinglow-BMI bodies relativeto higher-BMI bodies. However, this pattern may be due to the putative existenceof two subgroupswithinthe anorexic observers: those with a BMI < 14 and those with a BMI > 14. The former subgroupis significantly moreaccurate in estimating personal BMI than the latter subgroup. Several studies have suggested that those patients diagnosed with anorexia nervosa may not be a homogeneous group, but may in display differences aetiology and symptoms(e.g. Welch et al. 1990; DaCosta & Halmi 1992). However, the two in subgroupsin our studydo not differ significantly any of the questionnaire measures used to explore attitudes to food, body shape and size, or those assessing mood and
Proc.R. Soc. Lond. B (2000)

in anxiety. An alternativeexplanation for the differences estimation could be that the BMI < 14 subgroup are all hospitalized, whereas the BMI > 14 subgroup are mainly likely The BMI < 14 subgroup are therefore out-patients. to be receivingconsiderablymore psychiatricand psychoThis treatlogical treatmentand nutritionalintervention. aimed at addressing the ment would include interventions of overestimation body size. Thus, the increased accuracy in personal body-size estimation in this subgroup may in treatment an in-patient the just reflect impact ofintensive service. However, if this explanation is correct,one might in differences our questionnairemeasures expect significant it as well, which is not the case. Alternatively, may be incorrectto propose the existenceof two subgroupsin our anorexic observer group, and the pattern of estimates in our comparatively small sample size may have led to a

own of Estimation BMIdependsonobserver's BMI

and others 1995 M. J.Toviee

26 24 -

(a)

22 S S

~~~~~~

Og

20 -

,< 21

go

24 - t 214 16

^ .
S

19-

18 10

, 20

, 35

, 40

15

25 30 observerBMI

Figure 6. (a) The position of the peak of each attractiveness n tur of your prceptin BMI of curve based on the actual our ow the BMI,which images plotted against correlated each observer's BMI; the two are significantly (Pearson correlation, r=0. 550, p < 0.000 1). (b) The position of the peak of each attractivenesscurve based on the estimated facor be plotted BMI mantanigagainst each observer's BMI; the two are no longer correlated (Pearson correlation, r = 0.078, p = 0.272) .

false impression that would be dispelled with a larger sample size. of The negative correlationbetween the overestimation BMI and observer's BMI in the control and personal bulimic observersmightprovide a mechanism that acts to reinforce weight-loss behaviours in a positive feedback loop (i.e. weight loss causes you to overestimate BMI, your perception of what is an attractivesize and shifting

It is interestingto note that as an observer's BMI increases much beyond the upper limits of the normal range (i.e. above a BMI of 24units), she tends to underestimatethe BMI of both her own and other bodies. This resultis consistentwith a previous reportof underestimation of body weightby overweightwomen (Klesges 1983). This underestimationcould potentiallyplay a role in the process by which a person's body mass increases beyond the normal range, as a subject may underestimate any weightgain. The degree of overestimation with changing BMI should not be overstated; it modulates the accuracy of BMI estimation rather than producing large shifts in accuracy. In the estimationof other women's BMI, if one looks within the 'normal' BMI range (20-25 BMI units), then the shift in estimation is less than 2.0 BMI units. Similarly,in the estimationof personal BMI, the shiftin estimation is less than 1.0BMI unit. It is only when one looks at the extremesof the BMI range in our observers (from a BMI of 11.53 to 41.12) that this change in observer BMI of 29.59 units produces dramatic changes the accuracy of estimation. For other women's BMI, the very low BMI observersoverestimateby up to 4.31BMIest units, whereas the higher BMI observers underestimate by up to 2.69 BMIestunits (a range of 7.0 BMIeStunits).For of theestimation personal BMI, theverylow BMI observers by overestimate up to 10.3BMIestunits,whereas the higher BMI observersunderestimateby up to 9.5 BMIest units (a of range of 19.8BMIest units). So normally the effect BMI on estimation is a modulating factor, rather than a Within the range of BMI values found controllingfactor. in most people in the general population, the shiftin estiin mation will not produce great differences the perception of size and attractiveness.It is only when there is a steady and sustained drop in personal BMI that the misperceptionof body size becomes a potentially signifibehaviour. cant factorin maintainingdietary-control It is possible that the basis for the overestimationof BMI may lie in part with the neural changes associated with changes in body mass. Structuralimaging studies of eating disordered patients have shown irreversiblebrain shrinkage,consistentwith cell death, and this is likely to of alter the functioning the visual system (e.g. Lankenau et al. 1985; Krieg et al. 1989; Lauer et al. 1990; Lambe et al. 1997). So it is possible that the neural changes documented in low-BMI subjects would manifest themselves in the detection and analysis of the visual cues to the BMI of a body. However, the changes in perception cannot be easily explained by these neural changes. First, in the case of anorexic and bulimic observers, if the overestimation is a reflection of neural damage caused by dietary abnormalities, then the magnitude of the overestimationshould correlate with the duration of the eating disorder (i.e. the length of time there has been severe dietary abnormalities). However, it does not. Second, for the estimation of other women's BMI, there is a continuum of misjudgementfrom overestimationin low-BMI observers through to underestimation in higher-BMI observers. Finally, for personal BMI, the degree of overestimation actually falls with decreasing BMI in anorexic observers. This suggests any explanation based on neural damage through malnutritioncan only be part of the story.

Proc.R. Soc. Lond. B (2000)

1996 M. J.Tovee and others Estimation BMIdependsonobserver' BMI of own As we have said earlier, it has been argued that there may be two components to body-image distortion: a perceptual distortion and a cognitive dissatisfaction (Cash & Brown 1987; Slade 1988, 1994; Cash & Deagle 1997). The putative cognitive component is usually measured by questionnaires,such as we have used, which with her assess the degree to which a subject is dissatisfied size, shape or some other aspect of her appearance. Consistent with our own findings,previous studies have suggested the two components are only weakly correlated, or not significantly correlated at all (e.g. Cash & Green 1986; Ben-Tovim et al. 1990; Hsu & Sobkiewicz 1991; Keeton et al. 1990). Additionally, although some aspects of the cognitive dissatisfactionseem to be linked to mood (e.g. Cohen-Tovee 1993), this is not the case for the overestimation reported here. This suggests that a can arise independentlyfromany perceptual dysfunction The fact that the overestimation cognitive dysfunction. extends to other women's bodies is consistent with it being a purely perceptual error. If it were a cognitive deficit, perhaps a miscalibrationof an internalrepresentation of personal body mass and size, then it would be unlikelyto extend to an overestimationof other women's a bodies. However, if it were a perceptual deficit, problem with making the fine within-category judgements necessary to assign a body to a particular body-mass value, then one might expect this overestimationto extend to the perceptionof other women'sbodies, as we findhere. Our resultssuggesta significant perceptual component in body-image dysfunction, overestimawith a significant tion of both personal BMI and other women's BMI. The primary treatment of body-image disturbance in both anorexia nervosa and bulimia nervosa is through cognitive behavioural therapy (e.g. Rosen et al. 1989; KearneyCooke & Striegel-Moore 1997; Fairburn et al. 1999). It is possible that this treatmentfor eating disorders could be enhanced by addressing the perceptual problem directly, perhaps by giving a client feedback regardingher pattern of overestimation, via a BMI-estimation task such as that used here. The results show that although anorexic and bulimic observers apparently find bodies with a significantly lower BMI more attractive than controls, these preferences seem to be by-productsof their overestimationof BMI. Their internal representationof an attractiveBMI seems to be the same as that of a control observer, but their overestimationsystematically shiftsthis representation relativeto the real BMI of a particular body. Our results suggest that people of a similar BMI are likelyto findeach other attractive,and so there may be a positive assortmentfor BMI in human mate selection. A number of studies have investigatedthis area and they have indeed reported a weak to moderate correlation between intra-pair body mass of between 0.1 and 0.25 (for reviews, see Spuhler 1968; Allison et al. 1996). This correlationcannot be simplyexplained by factorssuch as similar environmentalconditions or partner influence,as a recent study compared intra-pair body mass prior to marriage and cohabitation (Allison et al. 1996). After correctingfor age, this study found a weak, but statistically significant,correlation of 0.13. Taken as a whole, these studiessuggesta weak, but significant, of effiect positive assortment BMI in human mate selection. for
Proc.R. Soc. Lond. B (2000)

The modulation of preferredBMI by personal BMI suggestsan intriguinganswer to the reportsof differences in body-mass preferences for differentcultures (e.g. Furnham & Nordling 1998; Craig et al. 1996; Yu & Shepard 1998, 1999; Wetsman & Marlowe 1999). Are these differencescultural? Or are they actually merely in reflecting physicaldifferences the subjectstested? If the observers in one cultural group have an average BMI differentto the observers in the comparison cultural group, then differences in preferred BMI would be expected. For example, Polynesianshave been reportedto find optimally attractive a body mass heavier than comparable westernpopulations (Craig et al. 1996). They have also been reported to have heavier personal BMI values. So it is possible that what is reportedas a cultural between differenceis related to the physical differences the cultures, since within a single cultural group (as reported here), people with a larger BMI preferred images of women with a larger BMI. Without correcting for the BMI of the observer, it may be impossible to in determine whether differences body-mass preferences between two populations are derived fromcultural differences, or merely differencesin the body-mass of the observersin the two groups. anafor We thankDr PiersCornelissen his advice on statistical lysisof the data and Dr Bruce Charltonfor a criticaland We constructive readingof the manuscript. also thankDr Sara McCluskey for allowing us to recruither patientsfor this study. REFERENCES V Allison, D. B., Neale, M. C., Kezis, M. I., Alfonso, C., mating S. Heshka, S. & Heymsfield, B. 1996 Assortative Behav.Genet. 26, for relativeweight:geneticimplications. 103-111. and Association1994 Diagnostic statistical AmericanPsychiatric DC: 4th manual mental of disorders, edn.Washington, American Psychiatric Association. Beck,A. T., Ward,C. H., Medelson,M., Mock,J. & Erbaugh, Arch.Gen. for depression. J. 1961 An inventory measuring Psychiatry 4, 561-571. H. M. Ben-Tovim, I., Walker, K., Murray, & Chin,G. 1990 D. measures? body image or body attitude Body size estimates: . 9, _nt.Eat. Disord. 57-67. of and measurement, classification Bray, A. 1978Definition, G. thesyndromes obesity. _.Obes. 99-112. of Int. 2, Cash, T. E & Brown,T. A. 1987 Body image in anorexia nervosa and bulimia nervosa: a review of the literature. Behav. Modif. 487-521. 11, Cash, T. E & Deagle, E. A. 1997 The natureand extentof in disturbances anorexia nervosa and bulimia body-image 22, nervosa:a meta-analysis. . Eat. Disord. 107-125. Int. and body image Cash,T. F. & Green,G. K. 1986 Body weight and affect. I. cognition among college women: perception, Pers. Assess. 290-301. 50, E. Cohen-Tovee, M. 1993 Depressedmood and concernwith and shapein normal youngwomen.Int._.Eat. Disord. weight 14,223-227. C. M. Cooper,P. J., Taylor, J.,Cooper,Z. & Fairburn, G. 1987 and of The development validation thebodyshape questionJ. 6, naire.Int. Eat. Disord. 485-494. E. Cooper,M., Cohen-Tovee, M., Todd,G., Wells,A. & Tovee, beliefquestionnaire: preliM. J. 1997 The eatingdisorder Behav. Ther. 381-388. Res. 35, minary development.

Estimation BMI depends observer's BMI of on own

Craig, P. L., Swinburn, B. A., Matenga-Smith, T., Matangi, H. & Vaughn, G. 1996 Do Polynesians still believe that big is beautiful? Comparison of body size perceptions and preferences of Cook Island Maori and Australians. NZ Med. Y. 109, 200-203. DaCosta, M. & Halmi, K. A. 1992 Classifications of anorexia nervosa: question of subtypes.Int.Y. Eat. Disord.11, 305-313. Drury, H. A., Van Essen, D. C., Anderson, C. H., Lee, C. W, Coogan, T. A. & Lewis, J. W. 1996 Computerized mappings of the cerebral cortex: a multi-resolution method and a flattening surface-basedcoordinate system.Y. Cogn. 8, Neurosci. 1-28. Fairburn,C. G. & Beglin, S. J. 1994 Assessmentof eating disorders: interventionor self report questionnaire. Int. Y. Eat. Disord.16, 363-370. Fairburn, C. G., Shafran, R. & Cooper, Z. 1999 A cognitive behavioural theory of anorexia nervosa. Behav. Res. Ther. 37, 1-13. in Furnham, A. & Nordling, R. 1998 Cross-cultural differences preferencesfor specific male and female body shapes. Pers. Individ. 25, Diff/ 635-648. Furnham, A., Tan, T. & McManus, C. 1997 Waist-to-hip ratio and preferences body shape: a replication and extension. for Pers.Individ. Diff 22, 539-549. Gardner, R. M. 1996 Methodological issues in assessmentof the perceptual component of body image disturbance. Br. . 87, Psychol. 327-337. Garner, D. M. & Garfinkel,P. E. 1997 Handbook treatmentfor of eating disorders. New York: Guilford Press. Henss, R. 1995 Waist-to-hipratio and attractiveness.A replication and extension.Pers.Individ. Diff 19, 479-488. Hsu, L. K. & Sobkiewicz, T. A. 1991 Body image disturbance: time to abandon the concept for eating disorders? Int. Y. Eat. Disord.10, 15-30. Kearney-Cooke, A. & Striegel-Moore, R. 1997 The etiology of and treatment body image disturbance. In Handbook treatof ment eating disorders for (ed. D. M. Garner & P. E. Garfinkel), pp. 295-306. New York: Guilford Press. Keeton, W P., Cash, T. F. & Brown,T. A. 1990 Body image or body images? Comparative, multidimensional assessment Pers.Assess. among college students.J. 54, 213-230. Klesges, R. C. 1983 An analysis of body image distortionsin a non-patientpopulation. Int.Y. Eat. Disord.2, 35-41. Krieg, J. C., Lauer, C. & Pirke, K. M. 1989 Structural brain abnormalities in patients with bulimia nervosa. Psychiatry Res. 27, 377-387. Lambe, E. K., Katzman, D. K., Mikulis, D. J., Kenndy, S. H. & Zipursky,R. B. 1997 Cerebral gray matter volume deficits after weight recovery from anorexia nervosa. Arch. Gen. 54, Psychiatry 537-542. Lankenau, H., Swigar, M. E., Bhimani, S., Luchins, D. & Quinlan, D. M. 1985 Cranial CT scans in eating disordered patientsand controls.Comp. 26, Psychiatgy 136-147.

M. J.Tovee and others 1997

Lauer, C. J., Lassle, R. G., Fichter, M. M., Pirke, K.-M. & Krieg, J.-C. 1990 Structural brain alterations and binging vomiting behaviour in eating disordered patients. Int. Y. Eat. Disord.9, 161-166. Rosen, J. C., Salzberg, E. & Srebnik, D. 1989 Cognitivebehavior therapy for negative body image. Behav. Ther. 20, 393-404. self-image. Princeton Rosenberg, M. 1965 Society theadolescent and UniversityPress. Singh, D. 1993 Adaptive significanceof female physical attractiveness: role of waist-to-hip ratio. Y. Pers. Soc. Psychol.65, 293-307. Singh, D. 1994 Ideal female body shape: the role of body weight and waist-to-hipratio. Int.Y. Eat. Disord.16, 283-288. Slade, P. D. 1988 Body image in anorexia nervosa. Br. . 153(Suppl. 2), 20-22. Psychiatry Slade, P. D. 1994 What is body image? Behav. Res. Ther. 32, 497-502. Spuhler, J.N. 1968 Assortative mating with respect to physical characteristics. Eugen. Q 15, 128-140. Tovee, M. J.& Cornelissen, P. L. 1999 The mysteryof female beauty. Nature 399, 215-216. & Tovee, M. J. Cornelissen, P. L. 2000 Female and male percepand profile.Br. . tions of female attractivenessin front-view (In Psychol. the press.) Tovee, M. J.,Mason, S., Emery, J.L., McClusky, S. E. & Cohen-Tovee, E. M. 1997 Super models: stick insects or hour glasses. Lancet350, 1474-1475. Tovee, M. J., Reinhardt, S., Emery,J.L. & Cornelissen, P. L. 1998 Optimal BMI and maximum sexual attractiveness. Lancet352, 548. Tovee, M. J., Maisey, D. S., Emery,J.L. & Cornelissen, P. L. Proc.R. Soc. 1999 Visual cues to female physical attractiveness. Lond. B 266, 1-8. Tovee, M. J., Tasker, K. & Benson, P. J.2000 Is symmetrya visual cue to attractivenessin the human female body? Evol. Hum. Behav.21, 191-200. Welch, G., Hall, A. & Renner, R. 1990 Patient subgrouping in anorexia nervosa using psychologically based classification. Wetsman,A. & Marlowe, F. 1999 How universal are preferences for female waist-to-hip ratios? Evidence from the Hadza of Tanzania. Evol. Hum. Behav.20, 219-228. Yu, D. W. & Shepard, G. H. 1998 Is beauty in the eye of the beholder? Nature 396, 321-322. Yu, D. W & Shepard, G. H. 1999 The mysteryof female beauty reply.Nature 399, 216.

Eat. Disord. 311-322. Int.J. 19,

As this paper exceeds the maximum length normally permitted, the authors have agreed to contributeto productioncosts.

Proc.R. Soc. Lond. B (2000)