Rajib Paul, Magdalena Niewiadomska-Bugaj, Amy B. Curtis, and Catherine L. Kothari
Western Michigan University
In two-part regression analysis for zero-inflated count data, two quantities are usually modelled in terms of linear functions of covariates: the probability of zero is modelled through logistic regression, and the log-transformation of nonzero mean is modelled by Poisson regression. This talk considers spatially correlated response variables where the spatial dependence is modelled through conditional autoregressive (CAR) models. Bayesian LASSO receives a great deal of importance for its ability in variable selection. Lasso-type priors are developed for regression coefficients, which enable us to identify important covariates from a pool. The approach is applied to Michigan health policy data on diabetes rates and all our inferences are based on a Markov chain Monte Carlo algorithm.
All statistics students are expected to attend.