B1521
Title: Comparative performance of a semi-parametric generalized linear model in selected analysis settings
Authors: Paul Rathouz - University of Texas at Austin (United States) [presenting]
Abstract: A semi-parametric extension of the generalized linear model family has been previously introduced in which the mean model is specified as in any quasi-likelihood formulation, and the response (reference) distribution is fully specified, albeit non-parametrically. The dimension of the non-parametric component in the semi-parametric generalized linear model (SPGLM) is of the same order as the cardinality of the support space of the response. These models have been applied and are often well-suited for settings ranging from ordinal data with finite support to continuous data. After introducing the SPGLM and addressing some important computational advances over the past several years, we will more formally address two important aspects of this modeling framework. First, using available data, we will conduct a comparative analysis between the SPLGM and the popular proportional odds model, focusing on goodness-of-fit and model interpretation. Then, we will show how the fitted SPGLM model can be used to estimate and make inferences on the CDF as a function of covariates, and, time-permitting, how this can be extended to obtain estimates of quantiles as a function of covariates.
