Estimating Individual-Level Risk of Eating Disorders Using Item-Level Screening Scores and the Bayesian Graded Response Model

Likert-style screening tools are commonly used for early detection of eating disorders. We show that the Bayesian Graded Response Model, fitted to item-level screening responses, can provide a 0-100% risk estimate for each individual. Our approach provides better accuracy and a better characterization of individual differences than aggregated scores.