Title: 
Model comparison in item response theory

Speaker: 
Ezgi Aytürk, Ph.D.Fordham University

Abstract: 
Item response theory (IRT) is a set of latent variabletechniques for modeling responses to psychometric tools such as testsand questionnaires. It has several advantages over the classicalanalysis, including better characterization of the measurement error andreliability, greater control on item quality, and sample-free estimationof the item and person parameters. Each IRT model has a unique set ofassumptions regarding how the item properties (e.g., difficulty,discrimination, response scale) and latent trait interact to determineitem responses. Consequently, the scoring of respondents on the latenttrait depends on the specific IRT model chosen. I will present a studywhere we used IRT to (a) modify an existing depression scale for a moreaccurate depression screening among cancer patients and (b) select ameasurement model that best represents the responses to this revisedscale. In presenting this study, I will focus on the practice of usinggoodness-of-fit indices in IRT model comparison. I will then present myrecent work on the shortcoming of IRT goodness-of-fit indices incontrolling for model complexity and discuss future research direction. 

Zoom link: https://boun-edu-tr.zoom.us/j/95248088564?pwd=VjlMWmRWWGUxVlZqUzQ0TXROcWRMQT09
Meeting ID: 952 4808 8564
Passcode: 802119