Factor analysis of the adolescent version of the Eating Disorders Examination Questionnaire (EDE-Q): results from Swedish general population and clinical samples

The Eating Disorder Examination Questionnaire (EDE-Q) [1] is perhaps the single most widely used self-report measure of eating disorder (ED) symptoms. Nevertheless, important questions remain about its validity and applicability in relation to particular groups of individuals, especially adolescents of both genders. The EDE-Q comprises four subscales of postulated underlying psychopathology (Restraint, Eating Concern, Shape Concern, and Weight Concern¹) and measures the occurrence and frequencies of key ED behaviours. The adult version of the instrument has been the subject of numerous studies investigating its reliability and validity in different populations [1–11]. Test-retest reliability is generally reported as good, ranging from 0.66 to 0.94 for subscales and from 0.51 to 0.92 for the key ED behaviour items [7, 10]. The subscales demonstrate acceptable internal consistency with alphas ranging from .70 to .91 [3, 7, 8]. Studies of convergent validity comparing the EDE-Q with its interview equivalent (the EDE) have generally demonstrated good agreement between measures [1, 4, 11].

Despite extensive psychometric work on the adult version of the EDE-Q and a positive picture of reliability and internal consistency, factor analytic studies are highly equivocal. The postulated four-factor structure thought to characterise the instrument’s subscales has never been replicated. Since the subscales were originally based on theoretical assumptions, rather than being empirically derived, this may not be surprising. Using exploratory factor analysis (EFA) Peterson and co-workers [7] found a three-factor model, where Shape and Weight Concerns merged into one factor, and Restraint and Eating Concern remained fairly distinct. In a study of the general population using EFA, Becker and colleagues [5] found support for a two-factor model where Eating, Shape, and Weight Concern collapsed into one factor, while Restraint remained a stand-alone factor. Darcy and co-workers [12] carried out both EFA and confirmatory factor analysis (CFA) on EDE-Q data from male and female college athletes and a control group. For the male control group a two-factor solution provided the best fit, whereas for the other three groups a three-factor solution fit best. Work by Allen and colleagues [2] with CFA in both clinical and community samples only managed to find support for a one-factor model consisting of eight Shape and Weight items. An EFA on an adult clinical Portuguese sample found three factors, roughly representing Shape/Weight concern, Restraint, and Eating concern [13]. Other studies that were unable to support the four-factor structure in adults, have found support for a single factor or a briefer questionnaire in relation to bariatric surgery candidates [14], Chinese and US students [15, 16], as well as Italian in- and outpatients [17]. This broad spectrum of factor analytic findings is a problem, and shows consistency only in not replicating the purported structure of the instrument. It appears impossible to conclude that the EDE-Q as currently used reflects a consistent and uniform underlying structure of psychopathology.

Another important problem is that there is little systematic knowledge about the EDE-Q in relation to adolescent girls and boys. Since adolescents comprise a significant proportion of ED patients it is important to know how well the instrument performs in relation to younger individuals from both clinical and non-clinical samples. Its factor structure has been investigated using British [18] and Mexican [19] adolescents, with both studies failing to find support for the four-factor model using CFA. However, both of these studies used the adult version of the instrument. There is a distinct version of the EDE-Q for use with younger subjects [20]. It diverges from the adult version, focusing on the past 14 rather than 28 days, and uses simplified language more suitable for younger individuals. Unfortunately, research on the adolescent version of the EDEQ has been extremely limited. Only two studies present normative data [20, 21]. Factor analytic studies of the adolescent version of the EDE-Q are needed in order to investigate its underlying structure and whether it shows similar properties to the adult version. This is important since the hypothetical four-factor structure of the EDE-Q is commonly accepted and used in both clinical assessment and for screening purposes in adolescents. Knowledge is also needed about the adolescent version in relation to both normal girls and boys, and in relation to clinical samples. A psychometrically sound adolescent version of the EDE-Q would aid researchers in the estimation of risk and identification of early onset, and would help clinicians to measure key symptoms and monitor treatment results in individual cases.

Given the limited knowledge of the adolescent version of the EDE-Q the present study aimed to investigate its factor structure in Swedish girls and boys from the general population and girls with a diagnosed eating disorder. Since there is no previous study investigating the factor structure of this particular version of the instrument we first used confirmatory factor analysis (CFA) to test the four-factor model, followed by exploratory factor analysis (EFA) if the four-factor model was not supported.

The present study aimed to investigate the validity of the adolescent version of the EDE-Q by examining its factor structure in samples of population-based and clinical Swedish young adolescents. Although there has been considerable factor analytic work on the adult version of EDE-Q and a small number of studies examining it in relation to adolescents, the present study is unique in that it tests the adolescent version of the EDE-Q in large samples of both girls and boys from the general population, and girls with a diagnosed eating disorder. This increases the generalizability of the study, and makes for a better evidence base for conclusions about underlying dimensions of eating pathology in population-based and clinical samples of adolescents.

Results of CFA failed to support the four-factor model of the EDE-Q. These findings are in keeping with a growing body of evidence in relation to the adult version of the EDE-Q that has failed to confirm the four-factor model of the questionnaire [2, 12, 28, 35, 36]. Using PA to determine the optimal number of factors in each sample, our results suggest that in both samples of adolescent girls a single factor centring on Dissatisfaction with Shape and Weight is sufficient to explain well over half the variance in the adolescent version of the EDE-Q. Studies of the adult version of the instrument in university students have tended to find three factors (e.g. Darcy et al) [12]. White and colleagues [18] who studied a notably young sample between 14 and 18 years of age also found a three-factor structure in adolescents. Both these studies utilised PAF, as was done in the present study. However, only the study by Darcy and colleagues utilised PA to determine the number of factors to retain for further analysis. White and colleagues used traditional Kaiser criteria and retained factors with eigenvalues greater than 1. Given that such methods may overestimate or inconsistently identify the number of factors in a data set [33], White and colleagues [18] may have inadvertently retained too many factors. Interestingly, the PA used in the present study closely followed the procedures adopted by Darcy et al. [12], which raises the question of why we found only one factor in our samples of girls.

One possible reason for the discrepancy between our results for girls and those found by Darcy’s [12] group could be the age of the two samples. Mean age in our data was 13.5 years for both samples of girls, while mean age in Darcy’s [12] samples were all around 21 years. It may be that issues relating to shape, weight and eating, although important to young teenage girls, are less differentiated during these earlier years. In adolescence, restrictive eating pathology is most common, with bulimic and other ED symptoms becoming more prevalent in young adulthood [37]; hence both underlying issues and actual ED behaviours are possibly more one-dimensional in younger girls. Our results imply that it may be important to take developmental factors into account when considering problematic aspects of shape, eating and weight in girls. During earlier years these concerns may tend to be global in nature. As these girls grow older the same concerns might tend to differentiate and develop into the underlying structures that have been found in studies of the EDE-Q focusing on three factors. Such a developmental perspective may also have implications for prevention, and suggests that early interventions in girls might usefully target self-esteem in relation to general ideas about shape and weight.

Boys from the general population scored by far the lowest on the all EDE-Q subscales and had the most diverse factor structure, which suggests some considerable gender differences from early teenage years. When boys develop issues related to disordered eating at an early age these may tend to centre on weight concern, discomfort in relation to their bodies and restrained eating. These differences imply a complexity of weight-related concerns in relation to gender. For normal boys, eating disorder related concerns may be more narrow and specific, focusing on dissatisfaction with perceived over- or even underweight (as in Factor 1) and eating in a more restrained manner (Factor 2), whereas for girls, body, weight and eating may relate more generally to who they experience themselves to be as individuals, i.e. more global self-worth. In other research, we have shown that the association between self-image and eating disorder symptoms is several times stronger in girls as compared to boys [21].

When assessing weight-related issues, it may be valuable to keep these gender differences in mind. Clinically, it may be important for therapists to be alert to young teenage girls having undifferentiated beliefs about the central importance of shape and weight for self-esteem, while in boys these concerns may be more domain-specific and less impactful. When considering gender differences in young adolescents it is important to note that one third of the items had to be deleted in boys in order to attain a stable factor structure. In both samples of girls only one item (eating in secret) was deleted. This raises the possibility that the EDE-Q is of limited utility in relation to young teenage boys.

It is interesting that the item concerning secretive eating was the only one to be excluded from the factor analysis of all three samples, with markedly low communalities. Although this item appears to have little relevance to the underlying dimensions of disordered eating at a young age, it raises the possibility that secretive eating might be useful as an early warning sign of a potential eating disorder, given that the item is included in factor analyses of older and clinical samples. This could be tested in future longitudinal research. Future studies could also examine factors that may mediate or moderate the early form of shape and weight concern found in adolescents. Given the universal relevance of the Dissatisfaction with Shape and Weight factor for both samples of girls, it may be important to pay it closer attention in community-based research endeavouring to chart risk factors longitudinally. For example, future studies could examine factors that may mediate or moderate shape and weight concern over time in different groups of adolescents.

Although the present study has the advantage of testing the adolescent version of the EDE-Q in relatively large samples of girls and boys from the general population, along with cases of diagnosed eating disorders in girls, it also has weaknesses. The factor structures we found through EFA may be sample-specific, hence not generalizable to other samples. Nevertheless, our study is a starting point for evaluating the factor structure of the adolescent version of the EDE-Q, and we encourage other researchers to conduct EFA in other samples of adolescents. Another weakness of the present study concerns the sample of clinical boys, which was too small to include in the present analyses. It therefore remains to be seen if clinical boys show similar patterns to other groups (and in particular boys from the general population). The present study could also have been affected by the relatively high degree of non-attendance at school on the day the EDE-Q was administered. It has been found that body image dissatisfaction may lead to absence from school [38], which could mean that some pupils with higher levels of eating psychopathology were underrepresented. Our work, nevertheless, widens the evidence base, arguing against using the EDE-Q as a four-subscale measure, provides indicators for future research, and raises some important clinical questions about gender and age differences in underlying ED psychopathology.

The factor structure of the adolescent version of EDE-Q in young teenage girls centres on a single underlying dimension of dissatisfaction with shape and weight. This appears to be the case in both clinical and general population samples. Over time, this general factor may become increasingly differentiated, giving way to the more complex patterns found in older girls and women. There appear to be important gender-based differences in the dimensions underlying the EDE-Q in young adolescents. Therapists should be alert to beliefs that girls have about the importance of shape and weight for their self-image at an early age. Weight-related concerns in young teenage boys may not have the same overall impact on self-image, but therapists should nevertheless be prepared to address these directly in treatment if and when they arise.