Modeling Adverse Birth Outcomes Through a Factor Model for Bayesian Quantile Regression

• Activity Number: 32
• Type: Contributed
• Date/Time: Sunday, August 1, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Lane F. Burgette*+ and Jerome P. Reiter
• Companies: Duke University and Duke University
• Keywords: Quantile regression ; factor model ; low birth weight ; Bayesian computation

Abstract

In this talk, we consider modeling birthweight and gestational age with data from a cohort study of pregnant women. Here, we are interested in the effects of covariates on the lower quantiles (rather than the mean) of the response distributions. Quantile regression techniques have been proposed to model these non-central aspects of the response distribution. However, standard quantile regression proves inadequate for this analysis because of a low signal-to-noise ratio and a large number of predictor variables. To address these challenges, we develop a Bayesian quantile regression factor model which allows us to model the response quantiles as a function of a small number of latent factors. We then extend this framework to model the lower quantiles of a latent ``gestational development'' factor, of which gestational age and birthweight are indicators.