Bayesian Lasso Regression for Zero-Inflated Spatial Data

• Activity Number: 448
• Type: Topic Contributed
• Date/Time: Wednesday, August 1, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Rajib Paul*+ and Magdalena Niewiadomska-Bugaj and Amy B. Curtis and Catherine L. Kothari
• Companies: Western Michigan University and Western Michigan University and Western Michigan University and Western Michigan University
• Keywords: conditional autoregressive model ; diabetes rate ; logistic regression ; Poisson Distribution

Abstract

In regression analysis for zero inflated data, two quantities are modeled in terms of linear functions of covariates - the probability of observing zeros, which is usually modeled through logistic regression and the nonzero mean function. We consider multivariable spatially correlated response variables, where the spatial dependence is modeled through multivariate conditional autoregressive (MCAR) models. Bayesian LASSO receives a great deal of importance for its ability in variable selection. We develop lasso-type priors for regression coefficients, which enables us to identify important covariates in a pool. We apply our approach to Michigan health policy data on diabetes rates and all our inferences are based on a Markov Chain Monte Carlo algorithm.