10/25/2013

Post date: Oct 26, 2013 5:36:27 PM

Speaker: Gary Chan, University of Washington, Seattle

Topic: Marginalizable mixed models for categorical and survival data

Abstract:

We introduce a class of parametric random effect models for correlated binary, ordinal and survival data, with parameters that can be directly interpreted as marginal effects. Flexible correlation structure can be modeled by correlated random effects. For categorical outcomes, this leads to a robust estimation procedure with an optimal weighting matrix being the inverse of a genuine model-based covariance matrix, and correlation parameters can be estimated by solving a composite likelihood score function. For survival outcomes, a composite likelihood function is maximized and that leads to an EM-type algorithm.

*This research is the dissertation work of Miss Rui Zhang under Dr. Chan's supervision.