报告题目： Variableselection for random effects two-part models
报 告 人： 刘磊教授（美国华盛顿大学）
报告时间 ：2018年6月26日 上午9:00-10:00
Random effectstwo-part models have been applied to longitudinal studies for zero-inflated (orsemi-continuous) data, characterized by a large portion of zero values andcontinuous non-zero (positive) values. Examples include monthly medical costs,daily alcohol drinks, relative abundance of microbiome, etc. With the advanceof information technology for data collection and storage, the number ofvariables available to researchers can be rather large in such studies. Toavoid curse of dimensionality and facilitate decision making, it is criticallyimportant to select covariates that are truly related to the outcome. However,owing to its intricate nature, there is not yet a satisfactory variableselection method available for such sophisticated models. In this paper, weseek a feasible way of conducting variable selection for random effectstwo-part models on the basis of the recently proposed “minimum informationcriterion" (MIC) method. We demonstrate that the MIC formulation leads toa reasonable formulation of sparse estimation, which can be conveniently solvedwith SAS Proc NLMIXED. The performance of our approach is evaluated throughsimulation, and an application to a longitudinal alcohol dependence study isprovided.