A comparison of polytomous IRT models in the investigation of the psychometric properties of PHQ-9 among Italian early adults

Over the past decades, Item Response Theory (IRT) measurement models and related methodologies have gained increasing attention in various research contexts. One of them is the clinical assessment field, since IRT models allows to relate the performance of a test to an unobservable latent trait (i.e., disorder). Here, trait level estimates depend on both individual responses and on the properties of each item of the item set, which allows - for example - to appropriately measure the clinical condition handling on a Likert point scale. A number of IRT models have the ability to handle polytomous response options, and the Rating Scale Model is considered the appropriate solution when items are intended to share the same response scale with ordered categories. Despite so, the adoption of an IRT model in the psychometric validation of clinical tests is often aprioristic, with little consideration about the nature of the response scale. The present study provides a comparison between three different polytomous IRT models (Rating Scale Model, Graded Response Model, Partial Credit Model) in the description of the psychometric properties of the Patient Health Questionnaire (PHQ-9) on a large representative sample of Italian early adults (18-26 years). Specifically, we propose a multilateral perspective in the psychometric evaluation of PHQ-9, consistently with classical methodologies as well as parametric and nonparametric IRT techniques. Importantly, we emphasize the statistical and theoretical impact of selecting different IRT models for the investigation of polytomous response scales.