Commentary: Explore Conditional Dependencies in Item Response Tree Data.

Abstract

Item response tree (IRTree) models are widely used in various applications for their ability to differentiate sets of sub-responses from polytomous item response data based on a pre-specified tree structure. Lyu et al. (Psychometrika) article highlighted that item slopes are often lower for later nodes than earlier nodes in IRTree applications. Lyu et al. argued that this phenomenon might signal the presence of item-specific factors across nodes. In this commentary, I present a different perspective that conditional dependencies in IRTree data could explain the phenomenon more generally. I illustrate my point with an empirical example, utilizing the latent space item response model that visualizes conditional dependencies in IRTree data. I conclude the commentary with a discussion on the potential of exploring conditional dependencies in IRTree data that goes beyond identifying the sources of conditional dependencies.