Introduction

Anorexia nervosa (AN) is an eating disorder with serious psychological and physical ramifications. Some of the characteristics of AN include a refusal to maintain normal body weight, an intense fear of gaining weight, and a disturbance in perception of body size or shape [1].

Body image is a multidimensional psychological concept that includes perceptual, cognitive, attitudinal, and behavioral components. Body image disturbance (BID) refers to both body size distortion and body dissatisfaction. BID occurs when there are dysfunctions in one or both of these components. BID is a core diagnostic feature of AN, and has been identified as a risk factor for eating pathology [2]. Body size distortion is the perceptual component of body size overestimation and has been frequently reported in AN patients. (see reviews in [3–5]). However, the extent or even existence of body size overestimation in AN is controversial. A recent review of contemporary studies conducted between 2003 and 2013 by Gardner and Brown [5] found greater consistency in the overestimation of body size by individuals with AN as compared to healthy controls.

Distortion of body size estimation involves both perceptual and attitudinal components [6]. The perceptual component of body size estimation refers to one’s ability to accurately judge one’s body size. The attitudinal component refers to one’s subjective feelings about weight and body shape (i.e., dissatisfaction). These two components have been found to be largely independent of one another. Two reviews of earlier studies on BID in AN both concluded that attitudinal effects are more prominent than perceptual disturbance in individuals with eating disorders [6, 7].

Recent advances in both technology and methodology have helped clarify the role of body size estimation in eating disorders. Gardner and Bokenkamp [8] utilized an advanced psychophysical methodology called adaptive probit estimation (APE) to investigate the perceptual and attitudinal roles of body size estimation. They compared 35 female patients with either AN or bulimia nervosa (BN) to a control group of 15 male and 19 female undergraduate students. Results revealed that the eating disorder patients overestimated their body size to a greater extent than controls, and patients with AN overestimated more than patients with BN. Importantly, there were no differences between groups in the ability to detect changes in body size, indicating that the overestimation was due to the attitudinal or affective component. Several investigators reported findings suggesting overestimation of one’s body size may not be characteristic of AN itself, but rather could be a function of low body weight [3, 9–12].

The primary goals of this study were to explore two questions: (1) do participants with AN overestimate body size as compared to controls, and (2) does the size of one’s body co-vary with accuracy of body size judgments. This study utilizes two proven psychophysical methodologies to examine both perceptual and attitudinal components of body size distortion in a group of female subjects with AN and a control group. We hypothesized that subjects with AN would significantly overestimate their body size when compared to non-eating-disordered controls, with the use of well-established psychophysical methodologies in conjunction with the digital photography video distortion methodology. Based on previous work from our group [13, 14] we also hypothesized that the accuracy of body size estimation would be independent of an individual’s BMI. We predicted that participants with AN would overestimate their body size more than controls and that the difference in overestimation would be related to their psychopathology, not their BMI. In the video distortion methodology, subjects view a computerized static picture of themselves and either make judgments about whether it is distorted in size or adjust their image wider or thinner. In addition, measurements of body dissatisfaction, anxiety level, and drive for thinness were obtained to examine their influence on estimating body size.

Materials and methods

The study was approved by the Colorado Multiple Institutional Review Board (COMIRB). A total of 74 females with a primary diagnosis of AN participated. AN diagnosis was based on established criteria in the DSM IV [1]. Eleven female controls between the ages of 12 and 21 participated in the study. All participants with AN were actively undergoing treatment in the Children’s Hospital Colorado eating disorders program in Aurora, Colorado or the Eating Disorder Center of Denver. Controls were recruited through local advertisements and flyers. All control participants had a current and lifetime body weight between 90 and 110 % of ideal body weight. Controls were excluded if they were taking psychiatric medication, had a DSM-IV diagnosis, had any eating disorder symptoms, or had first degree relatives with current or history of eating disorders. No attempt was made to measure or control for possible visual perceptual differences in the two samples. Demographic characteristics of both samples are shown in Table 1.

Table 1 Means/(standard deviations) and group comparisons for age, body mass index (BMI), point of subjective equality (PSE), difference limen (DL), body size estimation with method of adjustment (MOA), and body dissatisfaction measures and anxiety questionnaire responses for 74 anorexic and 11 control female participants
Full size table

A computer software program designed by Gardner and Boice [15] was used to measure body size distortion and body dissatisfaction. The method of adjustment (MOA) and adaptive probit estimation (APE) are two psychophysical methodologies used in the software program. In the MOA, a static digital picture of the participant’s frontal profile is randomly distorted between 20 and 30 % too wide or too thin. Ten trials are presented, including five wherein the initial picture is too thin (ascending trials) and five where the initial picture is too wide (descending trials). Using buttons on a mouse, the participant adjusts the displayed picture until the figure matches the size the participants perceive themselves to be as well as the size they would like to be ideally. The percentage discrepancy between an individual’s actual and perceived size represents a measure of body size distortion, and the percentage difference between an individual’s perceived and ideal size represents body dissatisfaction.

APE is an advanced psychophysical procedure that has the capability to measure separately the perceptual and attitudinal components of body size estimation, as discussed by Gardner [16]. The details of this psychophysical method are beyond the scope of this paper and can be found in Gardner and Boice [15]. Briefly, on a given trial, a static digitized picture of the frontal profile of a participant is distorted either too wide or too thin and the participant is asked to judge whether the picture is wider or thinner than they actually are. A total of 320 images (8 blocks of 40 images each) with varying percentages of distortion are presented. Following data collection, the program calculates the percent of body size distortion (the difference between the participant’s actual and perceived size) and the point of subjective equality (PSE). The PSE is the point at which a participant reports that their image is subjectively equal to their perceived size. It is calculated as the body size a participant would report to be too wide 50 % of the time and too thin 50 % of the time. The PSE represents the attitudinal or affective component of body size estimation. The percent of change that is necessary to be detectable by the participant is represented by the difference limen (DL) also referred to as the just noticeable difference. The DL is the percent of change in body size that must occur in order for the participant to detect a change 50 % of the time. The DL is dependent on one’s sensory capabilities and represents the perceptual component in body size estimation.

The Eating Disorder Inventory 2 (EDI-2; 16) contains 91 items used to assess 11 attitudinal and behavioral traits common in patients with eating disorders. The two subscales used in the present study were “drive for thinness” and “body dissatisfaction.” Good test–retest reliability coefficients for both subscales have been reported [17, 18]. Most subscales have internal consistency levels above 0.70 and all subscales discriminated between eating disorder patients and normal controls [19].

Participants also completed the Multidimensional Anxiety Scale for Children (MASC) by March [20] which assesses various anxiety symptoms present in children and adolescents. The measure consists of 39 statements rated on a four-point Likert scale. Good test–retest reliability and convergent validity have been reported [20–23].

Procedure

In order to complete the body image assessment computer program, each participant was photographed with a digital camera against a white backdrop. Participants stood with their feet hip distance apart and their arms outstretched roughly at a 45 % degree angle. The color digital picture was transferred to the computer and participants viewed distorted pictures of themselves sitting approximately two feet from the monitor. Following completion of the method of adjustment and APE computer tasks, participants completed two self-report questionnaires in a comfortable, private location.

Statistical analysis

For body size estimation, the proprietary body image software [15] described above was used. The AN and control groups were compared using t tests and effect sizes reported using Cohen’s d. Intraclass correlation coefficients were calculated and compared between actual and estimated BMI for control and AN patients. The within-group correlations were reported as Pearson correlation coefficients.

Results

Accuracy of body size estimations were calculated by both the MOA and by APE. For the MOA an average was found between the ascending and descending series of trials. For the APE methodology the PSE was calculated. Table 1 shows that participants with AN overestimated their body size significantly more than controls both with the MOA and with the APE methodology, with large effect sizes in both instances. The correlation between body size estimations obtained with the two psychophysical methods was high, r = 0.76, t (83) = 10.65, p < 0.0001. Subsequently, an average of the two values was used as a measure of size estimation. As shown in Table 1, control participants slightly underestimated their perfectly accurate body size estimation by −1.62 % (SD = 3.33), a difference that was not significant from zero. Participants with AN overestimated their body size by 8.93 % (SD = 10.46), a difference significantly greater than zero distortion, t (73) = 7.34, p < 0.0001. As hypothesized, AN participants significantly overestimated their body size compared to controls, with a difference between the two groups of 10.55 %. The difference is highly significant (p < 0.0001) and has a large effect size (Cohen’s d = 1.36). Participants with AN were significantly more variable in their size estimations as compared to controls (F max = 9.87, p < 0.0005). As shown in Table 1, the ability of the AN and control groups to reliably detect changes in their body size (DL) was similar with participants with AN having a mean DL value of 2.27 and controls 2.72, a difference that was not significant. Within the context of the present study, and as noted earlier, the DL reflects the sensory aspects of the perceptual component in body size estimation.

Numerous studies have reported that individuals with AN have greater body dissatisfaction than non-eating-disordered subjects. We measured body dissatisfaction two ways, including the actual/ideal size discrepancy in the MOA and the EDI subscale for body dissatisfaction (EDI-BD). For the MOA, ideal size judgments were collected on 53 of the 74 participants with AN, as this test was added to the protocol later in the study As shown in Table 1, participants with AN had significantly greater body dissatisfaction than controls on both measures, with large effect sizes for both measures. We also explored correlations between BMI and the clinical rating scales. For the AN subjects, the self-rated scales for body dissatisfaction were significantly correlated with BMI (EDI-BD r = 0.30, p = < 0.05). The EDI subscale for Drive for Thinness and the MASC, a measure of anxiety, were not significantly correlated with BMI (EDI-DT r = 0.16; MASC r = −0.12). There were no significant differences for the same correlations in the control group, although this lack of significance is tempered by the relatively small number of control participants.

To test whether body size affects body size perception accuracy, we conducted a regression analysis predicting estimated BMI as a function of participant’s actual BMI. Estimated BMI is calculated by applying the percentage of body size estimation distortion to a participant’s actual BMI. For example, a participant with an actual BMI of 20 who overestimates his/her body size by 10 % has an estimated BMI of 22 %. Because BMI values represent ratio scale values this conversion to percentage values is appropriate and has been used by previous researchers [3].Separate regression analyses were conducted for AN and control participants.

Figure 1 illustrates the relationship for both groups. If participants were perfectly accurate in estimating their BMI, then data points would fall on the 45-degree diagonal line extending from the intersection of the vertical and horizontal axes, with a slope of 1.0. Data points falling above this line reflect overestimations of BMI and points below the line underestimations. Figure 1 shows that more participants with AN overestimated their BMI and that the slope of the regression line slightly increases as actual BMI increases. However, the slope of 1.28 (standard error = 0.16) is not significantly greater than 1.0, t (71) = 1.75, p > 0.05. As noted earlier, and is shown in Table 1, overall body size overestimation of participants with AN was significantly greater than control participants.

Fig. 1
figure 1

Regression lines for estimated BMI for controls and participants with anorexia. The dashed diagonal line represents accurately estimated BMI. Points above this dashed line represent overestimation of body size while points below represent underestimation

Full size image

Figure 1 also shows that the control subjects were significantly more accurate in predicting their estimated BMI, irrespective of their actual BMI. The slope of the regression line for controls is 1.0 (standard error = 0.08) and is only slightly and non-significantly below the diagonal line reflecting perfect estimations of BMI. Our data suggest that most participants with AN are less accurate in estimating their BMI as compared to the control participants. In the AN group, overestimation of BMI occurs at nearly all BMI values and increases non-significantly with increasing actual BMI values.

Given the limitations of the small control group in the present study (n = 11), we retrospectively examined data from two previous studies with healthy control participants using the same MOA and APE methodology used in the present study [13, 14]. For both studies, we again conducted regression analyses comparing accuracy of body size estimations across a range of BMI values in participants without eating disorders. The healthy control participants were again accurate in their judgments irrespective of their actual BMI.

Although not a primary hypothesis of the study, we also examined the relationship between the variables of anxiety (as measured by the MASC), drive for thinness, and body dissatisfaction (as measured by the EDI), and body size estimation. Table 1 shows that participants with AN had significantly higher levels of drive for thinness, body dissatisfaction, and anxiety. For participants with AN there was a significant positive correlation between drive for thinness and body size estimation, with higher levels of drive for thinness associated with greater body size overestimations [r (69) = 0.44, p < 0.01]. For participants with AN there were also significant positive correlations between both measures of body dissatisfaction and body size estimation, including the EDI-2 subscale for body dissatisfaction, r (69) = 0.54, p < 0.01 and the actual/ideal body size discrepancy, r (51) = 0.78, p < 0.01. On both measures, greater body size dissatisfaction was associated with greater overestimation of body size. For participants with AN, there was no significant correlation between anxiety and body size estimation. For control participants, no significant correlations were found between measures of body size estimation and drive for thinness, body dissatisfaction or anxiety. It should be noted, however, that the small sample size of the control group required a very high critical value of r for reaching significance.

Discussion

This study presents data evaluating body size estimation from a large sample of females with anorexia nervosa. Our findings clearly reflect a significant difference in the accuracy between AN and control participants in judging their body size. These findings are in agreement with other studies comparing these two groups that used psychophysical techniques to separate the affective from the perceptual factors in body size estimation [8, 24, 25]. In all three of these previous studies, the overestimation of body size in AN was attributable to the affective component, and not to any perceptual or sensory deficit inherent in AN psychopathology. As concluded by Thompson and Gardner [26], “it is one’s beliefs, attitudes, and thoughts that influence willingness to perceive body image as either normal or distorted, rather than any sensory deficit” (p. 139). When measuring body size estimation with the video distortion technique and using established psychophysical procedures that include methods that separate the perceptual (i.e., sensory) and affective components, participants with AN were more likely to overestimate body size more than those without such eating disorder pathology and the difference can be shown to be due to the affective component. In agreement with previous findings [8, 24, 25], AN subjects in this study were significantly more variable in their size estimations as compared to controls.

We investigated a limited number of factors that could be related to this affective component. Of the three that were measured (body dissatisfaction, drive for thinness, and anxiety) both body dissatisfaction and drive for thinness were strongly correlated with overestimation of body size, but importantly only in participants with AN. Controls had relatively low levels of body dissatisfaction and drive for thinness and a much smaller range of body size estimations, both of which likely contributed to the non-significant correlations found between these factors and body size estimation in this group. Anxiety did not correlate significantly to body size estimation in either the AN or control groups. The relatively small number of participants in the control group tempers our findings, and future studies should strive for a larger comparison control sample.

One of the major aims of this study was to examine whether body size of participants co-varies with size estimation, independently of eating disorder psychopathology. Our data clearly indicate that the two are independent. Controls are accurate in estimating their body size across a range of BMI values. This was shown not only with our current data but also in two previous studies using the same psychophysical procedures. This methodology also consistently demonstrates that participants with AN overestimate their body size at a range of BMI values. In our study, accuracy of size perception did not vary with BMI for the control group, who more accurately estimated their body size perception independent of BMI.

Our finding that AN subjects were not significantly different than the control group in reliably detecting changes in their body size (DL) supports earlier findings that AN and control participants do not differ in their sensory sensitivity for detecting changes in their body size, which reflects the perceptual aspect of body size estimation [8, 24, 25].

The whole body techniques employed in this study, either the MOA or APE, clearly delineated the accuracy of body size estimates between AN and controls and the results strongly indicate that these estimates are not significantly related to BMI. The validity of these methodologies for individuals without eating disorders was demonstrated by the consistent accuracy of their size estimates, irrespective of their BMI. Our findings are consistent with the conclusion from the meta-analysis by Smeets et al. [28] that “These findings lend credibility to the conclusion that overestimation of body size in anorexic patients is indeed a consequence of their eating disorder” (p. 274).

It appears that attitudinal issues related to the psychopathology of AN, specifically body dissatisfaction and drive for thinness, are related to the ability to correctly estimate body size. We found significant correlations between the combined measure of body size estimation and all measures of body dissatisfaction and drive for thinness. This is a topic that has not been extensively researched and deserves further investigation. We recommend that future studies evaluate changes in body size estimation, body dissatisfaction, and drive for thinness as individuals progress through weight restoration and again when they have been weight restored for intervals of time. It would also be interesting to explore whether individuals with AN have a difference in treatment outcome related to their degree of body size overestimation. In addition, future studies should continue to use established psychophysical methodologies, including those that allow for the separation of the affective and perceptual components.

Limitations of this study include a relatively small number of participants in the control group. We underestimated the amount of time and effort that would be necessary to recruit age appropriate control participants within the required weight ranges during the period approved for the study. Using data from previous studies with healthy controls to make comparisons to the present data can also be problematic. We recommend that future studies strive for greater equality of participants in the AN and control groups including males and a wider age range of both patients (including children and older adults) and controls to more fully evaluate the factors related the perception of body image in AN and the impact on outcomes and interventions. Body image disturbance should also be evaluated in other eating disorder populations, such as bulimia nervosa, binge eating disorder, and eating disorder not otherwise specified (EDNOS).

Our results suggest that treatment interventions for patients with AN should focus on the attitudinal and behavioral components of body image. Evidence-based research is needed to develop therapeutic approaches for body image distortion in patients with AN. It would be interesting to explore use of the Body Image Software program [15] combined with a cognitive behavioral therapy approach as an intervention for body image distortion in patients with AN. The vast majority of published studies on interventions for body image are based on non-clinical samples of primarily female college students [28, 29]. Research on the efficacy and impact of interventions for AN, including weight restoration and psychotherapeutic approaches should include evaluation of body image and body size estimation in addition to changes in weight and medical stability. It is not clear if improvement in attitudes about body image such as drive for thinness and body dissatisfaction is correlated with improvement in accuracy of body size perception or vice versa. Anorexia nervosa remains a complex and often treatment refractory mental illness. Research on factors such as body image distortion, which contribute to persistence of the illness, remains critical to understanding the disease process and improving treatment outcomes.