A Semiparametric Approach Towards Diagnostic Classification Models

• Activity Number: 152
• Type: Invited
• Date/Time: Monday, August 10, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Mental Health Statistics Section
• Author(s): Jingchen Liu*
• Companies: Columbia University
• Keywords: Diagnostic classification models ; latent variables ; regularization

Abstract

A serious issue in the applications of diagnostic classification models is the model lack of fit indicating that the model structure is not flexible enough to account for the data patterns. For instance, the DINA model does not pass model checks for the fraction subtraction data even when the Q-matrix is fitted based on the data. On the other hand, a complex model that fits the data well typically are not as interpretable as the simple ones. In this talk, we try to obtain a compromise between the two by starting with a fully nonparametric model with discrete latent factors. Such a model imposes minimal assumptions on the data structure besides the discreteness of the latent variables. We reduce the model complexity by imposing penalties on the parameters to reach better interpretability.