(PS7-55) Item Response Theory Analysis of the Broad Anxiety Survey

Components of anxiety can serve as transdiagnostic constructs that help with understanding of a variety of psychopathologies. However, fear and broader components of anxiety are often confounded in self-report measures of eating disorders. The recently developed Broad Anxiety Survey (BAS; Hildebrandt et al, under review) overcomes some of these confounds by distinguishing between distinct anxious states (Fear, Worry, Disgust) and the threat characteristics that promote them (Uncertainty, Proximity, Location). The scale elicits subjective responses along each dimension in 10 common threatening eating or body image related situations. Initial evidence indicates the 10-item subscales are a result of single dimensions with second order factors for affect and threat respectively.  

To further understand these factors, in a separate sample of adults (n = 324, 52 ED; % Female =73.1, Age = 27.2, SD = 11.6, % Non-White = 24.4, BMI = 25.9 , SD = 6.83), we applied item response theory to the scales to examine item properties of the BAS scales. We applied 2 parameter item response model to each of the 7 domain scales.  Individual item thresholds and information criteria were estimated using theta parameterization.  Data were collected on sex, diagnosis, and race/ethnicity and models were also estimated separately by group for comparison.   

The results indicate that items relating to clothing feeling tight after a meal share the most information about threat and harm on the higher end of the latent trait. Additionally, eating without knowing the number of calories or nutrients shared more information about the latent trait in terms of the dimensions of proximity and location. Items were more heavily weighted towards the severe end of the the latent trait. The location, worry and disgust dimensions had more coverage of the latent trait by the 10 items. The items shared more information about the fear dimension on the higher end of the latent trait.  

Additionally, the item properties varied substantially when calculated separately by group. When specific to ED vs. Community sample, items generally provided more information across the latent trait, which is likely a result of scaling differences. For example, clothing feeling tight after a meal shared a high level of information across levels of the latent trait in terms of the dimension of threat in the eating disorder group. However, information shared by this same item in the community sample was greater on the high end of the latent trait than the low end of the latent trait. 

IRT modeling indicated that the selected items provided excellent coverage of the latent traits under investigation. Psychometric properties in a larger sample capture a full range of variability which provides desirable measurement qualities in mixed sample populations. When used clinically with ED patients, the subscales may provide higher precision in measuring food and body related anxiety, which could help with experimental designs where small differences in anxiety may be meaningful to both theory and outcome. Future research plans include formal testing of differential item functioning and applications of IRT models to multivariate IRT to examine the threat and affect domains of the measure in more detail.