The purpose of this study was to examine the psychometric properties of the 26-item Eating Attitudes Test (EAT-26) using Rasch Analysis in adult men and women from 1) a college/university and 2) individuals with overweight and obesity enrolled a behavioral weight management program. These two samples were combined in the current study to evaluate the psychometric properties of the EAT-26 and to determine if the instrument functions differently based on subgroup affiliations. Model data fit indicated a total of 7 misfit items which were removed from the final analysis based on unacceptable infit and outfit mean square residual values. Of the 7 items removed from analysis, the majority were in the oral control factor (items 8: “Feel that others would prefer if I ate more,” 13: “Other people think that I am too thin,” 15: “Take longer than others to eat my meal,” and 19: “Display self-control around food”). The oral control factor pertains to self-control of eating and perceived pressure from others to gain weight (5). Item-difficulty identified item 20: “I feel that others pressure me to eat” as the most difficult to agree (logit=1.03), while the least difficult item to agree was item 1: “I am terrified about being overweight” (logit=-0.72). Given the current study sample average BMI was 33.87 and an estimated 58% had obesity based on BMI ≥ 30, the majority of participants in the sample rated the fears of being overweight item consistently “always,” or “often.” It is well-established that people with overweight and obesity experience bias, discrimination, and ridicule based on body shape and size (53, 54). Internalized weight bias, also known as weight self-stigma, occurs when an individual with overweight or obesity internalizes negative beliefs and stereotypes about people in larger bodies (55). It is possible that fear of being overweight was rated “always” or “often” across this sample, in part, because of internalized weight bias from those with overweight and obesity. After removal of items 8, 9, 13, 15, 19, 25, and 26, the remaining 19 items exhibited fit appropriate to the expectations of the model.
The item-person map shows that the items differentiated participants with moderate to high levels of eating disorder risk and did not differentiate between those with low risk for having an eating disorder (< -1 logits). This allows the positions of item difficulties and person abilities screened to be easily examined visually and to note any gaps. Identification of gaps in item distribution could be used to help guide the development of new items and remove the duplicated items. No items discriminated people at lower risk of disordered eating. It is important to create additional items that will differentiate participants who are considered “low risk” for an eating disorder. The addition of these items is especially critical if future research were to longitudinally examine disordered eating over time using the EAT-26. It is important for the EAT-26 to adequately identify participants who are at low risk of disordered eating to accurately assess disordered eating change over time. For example, Richardson and colleagues (2015) examined the relationship between financial difficulties and eating attitudes in university students and observed higher eating attitudes scores at baseline significantly predicted greater financial difficulties at 3-4 months (56). However, in light of the current study findings, these findings may be due to the psychometric limitations of the instrument in that it did not adequately discriminate participants at lower levels of eating disorder risk. Additionally, items 10: “Feel extremely guilty after eating,” 16: “Avoid foods with sugar in them,” and 18: “Feel that food controls my life,” all fell on the same difficulty level of the item-person map. Items 17: “Eat diet foods,” 21: “Give too much time and thought to food,” 22: “Feel uncomfortable after eating sweets,” 5: “Cut my food into small pieces,” and 7: “Particularly avoid food with a high carbohydrate content” all loaded on the same level of the item-person map. Participant burden could be reduced by removing some of these items from the same factor that fall on the same difficulty level.
DIF analysis indicated that roughly 25% of the items retained in the EAT-26 differed based on having a BMI of ≥ 30. Items 3 (“I find myself preoccupied with food”), 12 (“I think about burning up calories when I exercise”), 14 (“I am preoccupied with the thought of having fat on my body”), 18 (“I feel that food controls my life”), and 20 (“I feel that others pressure me to eat”) functioned differently as a result of obesity status (obese vs. non-obese). Despite having the same risk for disordered eating, items 12, 14, and 20 were more difficult for individuals with obesity to rate “always” or “usually” than individuals who were not obese. Additionally, individuals with obesity were more likely to agree with items 3 and 18 than individuals without obesity. These findings are consistent with the literature on obesity, namely that preoccupations with food as associated with EDs are a strong predictor of dieting (57). Furthermore, dieters have shown to report higher food and dieting-related thoughts than non-dieters (58). The tendency of food preoccupation and the belief that food controls life within individuals with obesity is similar to findings reported in Desai and colleagues (2008), namely that participants with overweight were more likely than their normal weight peers to be preoccupied with food (25).
No items functioned differently based on sex. This finding is promising given the ED treatment issues pertinent to men, which includes: stereotypes of EDs as a “woman’s issue,” muscularity-oriented disordered eating as distinct from how EDs present in female populations, and inadequate health literacy among health practitioners in the field of EDs (32, 33, 59). Schaefer and colleagues (2019) reported no evidence of clinically significant differential item functioning in the EAT-26 in undergraduate men and women (21). Unlike the present findings related to differential functioning and sex with the EAT-26, the 12-item short form of the Eating Disorder Examination Questionnaire (EDE-QS) was examined using a Rasch analysis and differential item functioning was observed across sex groups (60). It is imperative to continue to assess sex differential functioning for EDs in other eating disorder instruments since EDs in men are underdiagnosed and many men fail to recognize disordered patterns of behavior because of cultural influences of EDs as a woman’s illness (61, 62).
As shown in Table 2, the six-category Likert-type rating scale did not function well. Thresholds did not advance in order with category 1 = none, category 2 = -0.74, category 3 = -0.66, category 4 = .60, category 5 = .19, category 6 = .61. This indicates that a different response format is warranted. In the current sample, 31.56% of people answered “Never,” 22.65% answered “Rarely,” 23.33% answered “Sometimes,” 9.5% answered “Often,” 7.71% answered “Usually,” and 5.25% answered “Always.” Of the six-categories, there are several infrequently selected (“Always,” “Usually,” and “Often”) and one category used more frequently than other (“Never”). These findings are consistent with other work that found lower endorsements on other ED self-report measures, particularly for restrained eating-items for people with overweight and obesity (63). The problems inherent in the six-category Likert-type rating scale functioning of the EAT-26 may partially explain why previous studies have observed insufficient sensitivity to detect a full or partial ED and why use of the EAT-26 within samples with overweight recommend a cut-off score of 11 instead of the originally proposed cut-off score of 20 (25, 64).
Strengths of this study include an adequate sample that consists of both adult men and women and participants from different BMI categories. Additionally, the utilization of Rasch analysis overcomes several limitations of traditional methods based on classical test theory. The current study contributes to what is known about EDs, particularly in identifying differences in the EAT-26 measure amongst people with obesity. Finally, the generalizability of these findings is appropriate for others who may want to incorporate an ED questionnaire into a program or clinical trial to screen for participant eligibility or monitor disordered eating development for those enrolled in a weight-focused intervention. Because our sample includes college students and adults enrolled in a behavioral weight loss program, these findings are applicable to those working in obesity and weight loss interventions.
There are several limitations of this study. First, the EAT-26 yields a referral index that is based on three criteria: 1) the total score of the EAT-26, 2) participant responses to behavioral questions around eating symptoms and weight loss, and 3) the individual’s BMI. This study only examined the 26 items of the EAT-26 and did not take into account the other two criteria for referring respondents to a qualified professional. Previous findings show higher BMI is positively correlated with EAT-26 scores (65). These findings contrast the recommendation of the tool using BMI, which encourages professional evaluation for a possible eating disorder if a person falls in the “extremely underweight” category compared to age-matched population norms (eat-26.com). Future work should compare differences in behavioral questions of the EAT-26 to determine if there are any problematic differences between people of different BMIs. For example, if the behavioral questions do not detect disordered eating in people at elevated BMIs because of the way the questions are designed and not because of the individual’s disordered eating behaviors, then this would further add support for the development of a separate tool that assesses disordered eating in people in larger bodies.
Social desirability is a possibility when it comes to self-report measures, especially those on sensitive topics such as EDs. It is possible participants may have responded to questions on the EAT-26 in order to present themselves in a positive light. This has been observed in other work, particularly that social desirability affects the assessment of eating behavior (66). Future work should investigate the psychometric quality of eating disorder instruments and include a measure to account for social desirability, such as the Marlow-Crowne Social Desirability Scale (67). The EAT-26 was included as an eligibility screening tool for a behavioral weight loss intervention for some adults in the current sample. It is possible that participants were aware of eligibility requirements specified in the informed consent and failed to report disordered eating patterns within the EAT-26 with the intention of enrolling in the program. Chandler and Paolacci (2017) noted the pitfalls of relying on participant self-report to determine eligibility and show participants can misrepresent relevant study inclusion criterion (68). Another limitation of this study is the lack of confirmed cases of EDs in this sample. If the current study had a record of diagnosed EDs, this would allow study of the efficacy of the EAT-26 to detect ED risk. Additionally, the Rasch analysis could be used to understand differential functioning of the EAT-26 based on eating disorder type (e.g., AN, BN, and binge eating).
As previously noted in the methods, the current study combined a sample of university students with adults enrolled in a behavioral weight loss program. Eligibility criteria were different for the two studies, where university students were required to be 18 and older while those enrolled in the behavioral weight loss program had more stringent inclusion and exclusion criteria based on previous health history and chronic medical conditions. For adults enrolled in the behavioral weight loss program, one exclusionary criterion was disordered eating based on EAT-26 scores. However, only two prospective participants were excluded from participation (group 2) based on EAT-26 scores in the original study. As a result, it is not believed the current study is reflective of selection bias as it pertains to the EAT-26 scores available and analyzed.