Join the Quantitative Psychology area for a talk by Dr. Sally Paganin (Assistant Professor, Department of Statistics, The Ohio State University)!
Title: Beyond normality: Bayesian extensions of Item Response Theory Models
Abstract: Item Response Theory (IRT) models are a widely used class of statistical models for describing latent traits—such as ability, attitude, or health status—of individuals responding to a series of items. Standard IRT formulations typically assume that latent traits are continuous, normally distributed, and independent across individuals. However, in many modern survey and testing contexts, these assumptions are violated, as surveys are administered to populations different from those for which they were originally designed.
In this talk, I will present Bayesian extensions of IRT models that relax these assumptions. First, I will address non-normal latent traits. Semiparametric extensions based on Dirichlet process mixtures allow the latent distribution to be represented as a flexible mixture of normal components without pre-specifying the number of subgroups. I will discuss implications for model identifiability and estimation strategies. Second, I will consider dependence among latent traits, showing how interactions among individuals—such as peer effects in university exam settings—can be incorporated into the model.
The models presented can be implemented using NIMBLE, an R package for probabilistic programming with hierarchical models.
About Sally Paganin: Sally Paganin is an Assistant Professor of Statistics at The Ohio State University. She earned her PhD in Statistics from University of Padova (Italy) in 2019. Her research focuses on Bayesian methodology, with particular emphasis on nonparametric priors, latent variable modeling, and computational aspects. She is also interested in the development of open-source statistical software, and she is an active collaborator to the NIMBLE project, a flexible R-based software for hierarchical models.