Analyzing Cognitive Testing Data with Extensions of Item Response Theory Models

• Activity Number: 360
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Jonathan Gruhl*+ and Elena A. Erosheva and Paul K. Crane
• Companies: University of Washington and University of Washington and University of Washington
• Keywords: Item Response Theory ; Latent Variable Models ; Mixed Outcomes ; Bayesian Estimation ; Cognitive Functioning ; Psychometrics

Abstract

Cognitive tests are important and commonly used instruments in studies of brain disorders. Analogously to data from educational tests, one could use extensions of Item Response Theory (IRT) models to analyze cognitive testing outcomes. We demonstrate this on the analysis of data from the Subcortical Ischemic Vascular Disease (SIVD) study. Drawing on recent developments in IRT modeling, we extend IRT models to allow for not only a mixture of binary and ordinal outcomes but also count, continuous, and right censored outcomes. We use a generalized linear latent variable approach for model formulation and estimate model parameters in a Bayesian framework via Metropolis-Hastings and Gibbs sampling. We compare our results to those from a simpler IRT approach that treats all outcomes as ordinal as well as to results from a more general approach that relies on extended rank likelihood methods.