Statistics Seminar（2015-09）

Topic: Model Heterogeneity in Data Analysis: Detecting Clusters and Outliers via Cross-Validating Predictive Distributions

Speaker: George C. Tiao, University of Chicago

Time：Tuesday, 4th June, 14:00-15:00

Location：Room 217, Guanghua Building 2

Abstract: This talk presents a procedure for detecting heterogeneity in a sample with respect to a given model. It can be applied to find if a univariate sample or a multivariate sample has been generated by different distributions, or if a regression equation is really a mixture of different regression lines. Based on some special features of cross-validating predictive distributions, the idea of the procedure is first to split the sample into more homogeneous groups and second to recombine the observations in order to form homogeneous clusters. These two phases, splitting and recombining, form the core of the procedure. The proposed procedure is exploratory and can be applied to find heterogeneity in any statistical model. The performance of the procedure is illustrated in univariate, multivariate and linear regression problems.

Your participation is warmly welcomed!