A Single Summative Global Scale of Disordered Eating Attitudes and Behaviors: findings from Project EAT, a 15-year Longitudinal Population-based Study

Introduction

Disordered eating attitudes and behaviors include, but are not limited to, excessive perceived importance of weight and shape, unhealthy weight control behaviors, and binge eating. Although disordered eating attitudes and behaviors have often been individually examined in population-based studies (1–12), a large proportion of the individuals engage simultaneously in several disordered eating attitudes and behaviors (13–15). From the perspective of public health, it is therefore of interest to examine whether these interrelated disordered eating attitudes and behaviors could be assessed efficiently using a single summative scale of disordered eating attitudes and behaviors. The appropriateness of such a scale depends on whether a single latent construct is a good fit to the data, and whether engaging in a greater number of disordered eating attitudes and behaviors is associated with more severe outcomes than engaging in fewer of them. Creating a summative scale may contribute to population-based surveillance and research, especially to clarify the prevalence and different levels of disordered eating attitudes and behaviors.

To screen disordered eating attitudes and behaviors in the general population, several instruments and questionnaires have been developed. These instrument and questionnaires include SCOFF, the Minnesota Eating Behavior Survey (MEBS), the Eating Pathology Symptoms Inventory (EPSI), the Disordered Eating Questionnaire (DEQ), the Ottawa Disordered Eating Screen for Youth (ODES-Y), the Risk Behaviour Related to Eating Disorders (RiBED-8) questionnaire, and the Disordered Eating Attitude Scale (DEAS) (16–25). Of the instruments and questionnaires, DEQ, RiBED-8, and DEAS have been further developed into scales for use in the general population (21,23–25).

However, these disordered eating behavior scales are limited in certain aspects. First, some of the scales are lengthy, including eight (23) to 25 items (24,25) in each scale, which limits their use as efficient tools for including in epidemiological studies that may assess multiple variables and be limited in space for items assessing disordered eating attitudes and behaviors. Additionally, these scales often include subscales (24,25), which limit the use as a single summative scale and the use as an indicator to measure the severity of disordered eating attitudes and behaviors. Moreover, most studies examining psychometric properties of the scales have been limited to adolescents (21,23) or women (24,25), which has resulted in far less knowledge of psychometric properties of the scales in men. Lastly, a majority of the existing disordered eating behaviors scales have been utilized in cross-sectional studies (21,24,25) and thus have limited data on their stability over time. Only one screening instrument has been created to identify disordered eating behavior among adolescents in Denmark, and its reliability was examined over a three-year period (23). To the best of our knowledge, no large population-based study of adolescents in the United States has examined the appropriateness of creating a summative scale of disordered eating attitudes and behaviors, nor examined the stability and validity of the scale over a 15-year period of time crossing important stages of development from adolescence to adulthood.

Therefore, to address that gap, the primary aim of our study was to longitudinally examine the appropriateness of creating a single summative scale, derived from five interrelated disordered eating attitudes and behaviors - perceived importance of weight and shape; extreme unhealthy weight control behaviors (i.e., self-induced vomiting); less extreme unhealthy weight control behaviors (i.e., meal skipping); chronic dieting; and binge eating impairment. These attitudes and behaviors, which represent three domains (i.e., concerns about weight and shape, unhealthy weight control behaviors, and binge eating and associated symptoms) have been- found to be associated with body satisfaction, self-esteem, depressive symptoms, and body mass index (BMI) (5,6,13–15,26–30).

We hypothesized that the interrelated disordered eating attitudes and behaviors among adolescents would load onto a single latent factor at baseline and at two additional follow-up time points. Given the known associations of the individual disordered eating attitudes and behaviors with body satisfaction, self-esteem, depressive symptoms, and body mass index (5,6,13–15,26–30), we hypothesized that higher disordered eating attitude and behavior scores on our scale would be associated with lower body satisfaction and self-esteem, as well as higher depressive symptomatology and BMI.

Methods

Study Design

Project EAT (Eating and Activity in Teens and Young Adults) , a population-based longitudinal study of weight-related health among adolescents, was developed as a cross-sectional survey in 1998–1999 (31). At baseline (EAT-I, 1998–1999), 4746 adolescents attending either middle or high school in Minneapolis-St. Paul, Minnesota USA participated and completed the survey. Given growing research interest in the eating and weight-related health of young people, Project EAT was followed-up at five-year intervals with participants who had provided sufficient contact information at baseline (N=3672 of 4746). Follow-up mailed/online assessments were conducted in 2003–2004 (EAT-II) and 2008–2009 (EAT-III) as participants progressed through adolescence and emerging adulthood (7,32). In 2015–2016, 2270 participants who previously responded to either EAT-I or EAT-II, I were contacted for a follow-up (EAT-IV). Out of 2,270 participants, 1830 completed the follow-up survey at EAT-IV online, by mail, or by phone (33,34). Data from participants who responded at all three waves (EAT-I, II, and IV) were used for the current analysis (N=1492). The University of Minnesota’s Institutional Review Board Human Subjects Committee approved all protocols.

Because attrition did not occur completely at random, and the majority of missing data were due to loss to follow-up, response probability weighting was used to account for missing data and potential response bias (35,36). Weights were derived as the inverse of the estimated probability that an individual responded at all three-time points based on several baseline covariates, including demographics, overweight status, parental living situation, and grade in school. After weighting, there were no significant differences between the original sample at EAT-I and our analytic sample. Sociodemographic characteristics of the 1492 adolescents at EAT-I is provided in Table 1.

Table 1.

Sociodemographic characteristics of adolescents at EAT-I (1998–1999, N=1492)

	Adolescents at EAT 1 (1998–1999)
Age (Mean, sd)	14.9 (1.6)
Female (%, n)	48.1% (858)
Race/Ethnicity (%, n)
Asian	14.0 (207)
Black or African American	7.4 (109)
Hawaiian or Pacific Islander	0.3 (5)
Hispanic / Latinx	3.2 (48)
Native American or Alaska Native	2.1 (31)
White	71.2 (1054)
Mixed race/ethnicity	1.8 (26)
SES
Low	9.7 (143)
Low-medium	14.6 (216)
Medium	23.5 (347)
Medium-high	32.1 (474)
High	20.1 (297)
BMI (sd)	22.1 (4.1)
Self-esteem (sd)	18.2 (3.7)
Body satisfaction (sd)	34.1 (9.3)
Depressive symptoms (sd)	11.9 (3.2)

Percents are weighted by non-response propensity to reflect the original EAT-I sample population, Ns are unweighted.

Survey Development

To allow longitudinal comparison and examination of secular trends, key items were included across all EAT surveys included in this analysis (EAT-I, II, and IV). Decisions to retain or drop items were based on their relevance to the study aims, their use in earlier analyses, and the performance of constructs in the literature (37). Scale psychometric properties were examined in the full EAT-IV survey sample. The test-retest reliability of BMI was assessed in EAT-III over a 2-week period in a subgroup of 125 participants. All other items test-retest reliability was assessed in a subgroup of 103 participants who completed the EAT-IV survey twice within one to four weeks.

Disordered eating attitudes and behaviors

The Project EAT Disordered Eating Attitudes and Behaviors (DEAB) scale replicates the Problematic Relationship to Eating and Food (PREF) scale created in the Coronary Artery Risk Development in Young Adults (CARDIA) study (14). For the Project EAT DEAB scale in this study, the following five items were included: perceived importance of weight and shape, extreme unhealthy weight control behaviors, less extreme unhealthy weight control behaviors, chronic dieting, and binge eating impairment, which together represent three domains: concerns about weight and shape, unhealthy dieting or weight control behaviors, and binge eating and associated symptoms. These disordered eating attitudes and behaviors were assessed at baseline (EAT-I) and re-assessed at two later time points: EAT- II and EAT-IV. Each disordered eating attitude and behavior was dichotomized following the cutoff points in the Diagnostic and Statistical Manual of Mental Disorders used to diagnose eating disorders (38). A single, global summative DEAB score was created by counting the number of disordered eating attitudes and behaviors reported. Details of the disordered eating attitudes and behaviors are shown in Table 2.

Table 2.

Model fit statistics for tests of measurement invariance in the global disordered eating attitudes and behavior scale at EAT I, EAT II, and EAT IV

Model	df	x2	CFI	TLI	RMSEA (CIa)
Congeneric	87	146.50	.98	.98	.01 (.02 – .03)
Weak Invariance	95	154.94	.98	.98	.02 (.01 – .03)
Strong Invariance	NA	NA	NA	NA	NA
Adjusted strong invariance model	94	149.29	.98	.98	.02 (.01–.03)

Note.

^a90% confidence interval limits for testing RMSEA. All estimates were significant at p < .01; df = degrees of freedom; χ² = chi-square fit statistic; CFI = Comparative Fit Index; TLI = Tucker-Lewis Index; RMSEA = Root Mean Square Error of Approximation.

Body satisfaction, self-esteem, depressive symptoms, and BMI used to assess the convergent validity of Project EAT Disordered Eating Attitudes and Behaviors (DEAB) scale

Body satisfaction, self-esteem, depressive symptoms, and BMI were assessed at EAT-I, II, and IV. Verbatim questions, points assigned, and psychometric properties are provided in Appendix Table 1.

Demographics

Age, ethnicity/race, gender, and socioeconomic status were self-reported at baseline. Ethnicity/race was assessed from the question “Do you think of yourself as (a) white, (b) black or African American, (c) Hispanic or Latino, (d) Asian American, (e) Hawaiian or Pacific Islander, (f) American Indian or Native American.” Socioeconomic status was derived from parental education level, family eligibility for public assistance, eligibility for free or reduced prices school lunch, and parental employment status (7,39).

Statistical analysis

To assess whether the Project EAT DEAB scale was an appropriate representation of the disordered eating attitudes and behaviors, confirmatory factor analysis was performed to examine the fit of the five disordered eating attitudes and behaviors to a single-factor model. Confirmatory factor analysis (CFA) with weighted least squares means and variance adjusted (WLSMV) estimators for categorical indicators (40) was applied to test the extent to which five disordered eating attitudes and behavior items loaded onto a single underlying latent construct. Building upon studies suggesting disordered eating attitudes and behaviors persist over time (7,8,41), the fit of the data to this single-factor model was examined over three-time points, at baseline (EAT-I), 5-year (EAT-II), and 15-year (EAT-IV) follow-ups.

Several model fit indices were examined to determine the adequacy of the one-factor model, including the chi-square test (χ²), Comparative Fit Index (CFI), and Tucker Lewis Index (TLI), and the Root Mean Square Error of Approximation (RMSEA). Smaller χ² values indicate better model fit. Ideally, the chi-square test for model fit would reveal a non-significant p-value (p>.05), though this is less common for analyses with large sample sizes. CFI and TLI values larger than 0.95 provide support for a well-fitting model, and RMSEA less than or equal to 0.06 are ideal with values less than 0.08 indicating adequate fit (42).

To examine measurement invariance over time, or the extent to which the latent factor measures the same construct at different time points, three models were tested: congeneric, weak invariance, and strong invariance models. In the congeneric model, the same factor structure was tested at each time point without any constraints. The weak invariance model tested the same factor structure with equivalent factor loadings across time. The strong invariance model tested the same factor structure across each time point with equivalent factor loadings and indicator residual variances at each time point. A chi-square difference test evaluated differences in model fit across levels of measurement invariance.

Lastly, the convergent validity of the scale was examined by assessing its bivariate correlation with body satisfaction, self-esteem, depressive symptoms, and BMI. Bivariate correlations between the scale and each construct were assessed at all three measurement occasions (EAT-I, EAT-II, and EAT-IV).

All analysis was performed using Mplus statistical software (43). Pairwise deletion was used to produce less biased estimates (44) and to handle missing data with WLSMV estimation(43–45).

Results

Assessment of the Single Factor Model of Disordered Eating Attitudes and Behaviors

As hypothesized, a one-factor model fit the data well at baseline (χ²(5) = 9.578, p = 0.088; CFI = 1.00; TLI = 1.00, RMSEA = 0.02). Strong factor loadings (0.60–0.87) were reported for each of the five disordered eating attitudes and behaviors indicators, which provides support for a global assessment of disordered eating attitudes and behaviors as a single construct (Figure 1).

Figure 1:

Factor analysis of the five-disordered eating attitudes and behavior variables at baseline (1998–1999), EAT II (2003–2004), and EAT IV (2015–2016)

Assessment of the Project EAT Disordered Eating Attitudes and Behaviors (DEAB) Scale Over Time

The one-factor model fit the data well in EAT-II (χ²(5) = 10.960, p = 0.052; CFI = 1.00; TLI = 0.99, RMSEA = 0.03), and EAT-IV (χ²(5) = 41.235, p < 0.001; CFI = 0.96; TLI = 0.91, RMSEA = 0.07). Strong standardized factor loadings were obtained at baseline (0.67 – 0.89), EAT-II (0.69 – 0.87) and EAT-IV (0.59 – 0.77). Model fit from tests of measurement invariance is presented in Table 3. The results from the congeneric and weak invariance models fit the data well, suggesting that item thresholds are stable across measurement occasions (i.e., between EAT-I and EAT-II). Strong invariance is a measure of the extent to which mean DEAB scores can be compared across measurement occasions. The strong invariance model did not fit the data well, suggesting that the mean DEAB score could not be compared across measurement occasions. Modification indices revealed that the source of the model misfit was due to invariance in the binge impairment item variance at different measurement occasions. To address this issue, an adjusted strong invariance model was run wherein the residual variances for each factor indicator, except the binge impairment item, were set to equal across measurement occasions. The adjusted strong invariance model converged, suggesting the increases in binge impairment across measurement occasions differentially related to the overall severity of disordered eating attitudes and behaviors at different time points. For example, at EAT-I and EAT-II, the threshold associated with having a greater than 50% model-estimated chance of a score of 1 on the binge impairment item as compared to a score of 0 was higher than the threshold during EAT-IV. This suggests that a score of 1 point at EAT-IV would be associated with a lower severity on the disordered eating attitudes and behaviors latent construct (Mean=0.96) than participants scoring 1 point at either EAT-I (Mean=1.65) and EAT-II (Mean=1.79). This pattern was consistent across categorical threshold boundaries (i.e., from a score of 2 to 3 and 3 to 4 points on the binge impairment item). A chi-square difference test was conducted to determine whether the most parsimonious model (adjusted strong invariance) or a more relaxed model (i.e., congeneric) fit the data best. The results from the chi-square difference test revealed that the fit of the two models did not significantly differ (Δχ²(7) = 2.79, p > 0.05) (Figure 1). This suggests that the five disordered eating attitudes and behaviors reflect a common latent construct over time, wherein the role of the binge impairment item relative to the overarching latent variable changes in young adulthood.

Table 3.

Correlations between global disordered eating attitudes and behavior scores and hypothesized five correlated constructs at each time point

	DEAB at EAT I	DEAB at EAT II	DEAB at EAT IV
Body satisfaction (total score)	−0.63	−0.60	−0.61
Self-esteem (total score)	−0.60	−0.41	−0.44
Depressive symptoms (total score)	0.44	0.41	0.45
BMI (kg/m2)	0.28	0.31	0.44

Note. All correlations are significant at p < .001

Convergent validity of the Project EAT Disordered Eating Attitudes and Behaviors (DEAB) scale

To test the convergent validity of the DEAB scale, bivariate correlations between the latent factor that the scale is designed to represent and constructs known to be associated with disordered eating attitudes and behaviors, namely body satisfaction, self-esteem, depressive symptoms, and BMI were examined at each measurement occasion (EAT-I, EAT-II, and EAT-IV). The DEAB scale demonstrated adequate convergent validity. The latent factor was negatively correlated with body satisfaction (r = −0.60 - −0.63) and self-esteem (r = −0.41 - −0.60) and positively correlated with depressive symptoms (r = 0.41 – 0.45) and BMI (r = 0.28 −0.44) across time points (Table 4). Descriptive characteristics of constructs related to the DEAB scale at each time point (EAT-I, II, and IV) are provided in Table 4.

Table 4.

Descriptive characteristics of constructs related to the disordered eating attitudes and behaviors scale at each time point

Descriptive statistics, M (SD)	EAT I	EAT II	EAT IV	Range
Body satisfaction (total score)	34.1 (9.3)	33. (9.3)	32.0 (8.9)	10–50
Self-esteem (total score)	18.2 (3.7)	18.4 (3.6)	19.2 (3.4)	6–24
Depressive symptoms (total score)	10.4 (2.8)	11.1 (2.9)	10.5 (2.9)	6–18
BMI (kg/m2)	22.1 (4.1)	24.5 (5.1)	27.9 (6.4)	12–63

Discussion

The purpose of our study was to examine whether a set of five disordered eating attitudes and behaviors that interrelate would load onto a single factor with a reasonable fit. The fit of the one-factor structure to the epidemiological survey data of adolescents confirmed that the five disordered eating attitudes and behaviors can be integrated into a single summative scale. Therefore, our summative Project EAT DEAB scale may reflect the severity of these attitudes and behaviors among adolescents in the general population as well as in research settings. Our finding of a good fit of the one-factor structure of disordered eating attitudes and behaviors is consistent with studies conducted in Italy (21) and Denmark (23). The study conducted in Italy assessed 24 items (e.g., “Reduce food intake to lose weight,” “Ruminate about weight or body shapes,” and “Worry about weight or body shape”) from the DEQ among adolescents and confirmed a single latent factor of the attitudes and behaviors using principal component analysis (21). The other study conducted in Denmark demonstrated a single underlying latent construct of the eight disordered eating behaviors from RiBED-8, including items such as “I throw up to get rid of what I have eaten,” “I am on a diet,” and “I have a bad conscience when I eat sweets.” (23) However, findings from our study and others (21,23) do not indicate that the one-factor structure is the only structure that can explain disordered eating attitudes and behaviors. In fact, other studies that took an exploratory approach aimed at identifying the most optimal underlying structure of various disordered eating attitude and behaviors report mixed results including that two or more factors made for an ideal factor structure(19,20,22,24,25). However, those studies differ from ours, where in our study, we intended to test the a priori hypothesis that a one-factor structure fits the data well and support that such behaviors can be viewed on a single dimension and measured with a single scale.

The proposed Project EAT DEAB scale appears to be reliable across different stages of development, as evidenced by the factor structure’s stability from adolescence (EAT-II) to young adulthood (EAT-IV). However, our study also revealed that the association between the indicator for binge impairment and the latent construct for disordered eating attitudes and behaviors differed across time. More severe binge impairment was associated with a greater severity of disordered eating attitudes and behaviors at EAT-I and EAT-II, than at the other time points. These finding indicate that more severe binge eating at those stages of life (i.e., adolescence and young adulthood) was associated with more concurrent disordered eating attitudes and behaviors than observed at other times (i.e., adulthood). Higher binge impairment at EAT-IV was, however, associated with less severe disordered eating attitudes and behaviors, relative to earlier measurement occasions, which suggests that the relationship between binge impairment and the overall Project EAT DEAB scale score differs by a function of developmental period (i.e., from adolescence to adulthood). Although that finding has no bearing on the utility of the Project EAT DEAB scale over time, it does imply that binge impairment may be more associated with concurrent disordered eating attitudes and behaviors in adolescence, whereas binge impairment may occur in isolation in adulthood.

The significant correlations between the Project EAT DEAB scale score and body satisfaction, self-esteem, depressive symptoms, and BMI suggest the good convergent validity of the scale in light of other studies (28–30) that have identified relationships between dimensions of eating pathology and psychological outcomes. In particular, the significant correlation between the composite scale and psychological outcomes suggests that the composite score reflects the underlying severity of disordered eating attitudes and behaviors in ways that are correlated as predicted with psychological and physical health in a research setting.

Overall, the results of our study suggest that the Project EAT DEAB scale is a reliable and valid tool that may serve as an efficient proxy for the severity of disordered eating attitudes and behaviors and be useful in public health settings, including for the surveillance of such behaviors from adolescence into adulthood and in population studies. Indeed, in longitudinal studies, including Project EAT (2,5–11,26), the Avon Longitudinal Study of Parents and Children (12), and the Danish National Birth Cohort, individual disordered eating behaviors, such as restrictive dieting behaviors (7,9), unhealthy weight control practices(3,7–10,12), and binge eating (3,12) are often assessed and examined (3). That said, a single summary measure of disordered eating attitudes and behaviors may be useful in these cohorts to capture a global picture of the scope of the problem and to assess the correlations of disordered eating attitudes and behaviors overall with adverse health outcomes. However, in the future, inclusion of additional domains and items to the Project EAT DEAB scale such as- fear of weight gain, emotional eating, concerns about eating, eating patterns, and body appreciation may help to broaden its breadth and afford a more global picture of disordered eating attitudes and behaviors. A strength of our study is that we had a large sample of adolescents who were racially and ethnically diverse. Using such a relatively large sample allowed us to test the empirically-driven models. In addition, the measurement of disordered eating attitudes and behaviors at baseline and during two additional follow-up periods allowed us to examine whether the structure held over time. In addition to these strengths, a few limitations of our study should be noted. First, we were unable to validate the Project EAT DEAB scale against a gold standard of eating disorder measurements. Second, although our study included men and participants from diverse ethnic and racial backgrounds, the limited sample size in those sub-populations restricted us from evaluating whether the factor structure differed by sociodemographic characteristics. Thus, our findings should be replicated in those sub-populations in order to verify the scale’s applicability across different groups. Lastly, significant attrition occurred in the sample over time; however, the response propensity method used in the analyses was designed to provide unbiased estimates for the demographic make-up of the original school-based sample, thereby allowing the results to be more fully generalizable to the population of young people in the Minneapolis-St. Paul metropolitan area.

Conclusion

Findings from this community-based study of adolescents indicate that five common disordered eating attitudes and behaviors can be usefully combined into a single measure, supporting the utility of this composite scale as an index of the severity of disordered eating attitudes and behaviors in public health research settings. Future work that examines a similar approach to measurement of disordered eating attitudes and behaviors in other age, ethnic/racial, geographic regions, and countries is necessary to confirm the utility of this brief assessment tool in other populations. Given the known associations of disordered eating attitudes and behaviors with a variety of outcomes of major public health importance, efficient assessments that can be used across different settings are important for advancing research on prevention and treatment in the population.

Abbreviation

DEAB

Disordered Eating Attitudes and Behaviors