A Test of Homogeneity for Two-Sample, Two-Component Mixture Models

• Activity Number: 616 - New Advances in Semiparametric Modeling and Testing for Complex Data
• Type: Topic Contributed
• Date/Time: Thursday, August 2, 2018 : 8:30 AM to 10:20 AM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Ollivier Hyrien*
• Companies: Fred Hutchinson Cancer Research Center
• Keywords: finite mixture models; hypothesis testing; EM algorithm

Abstract

We consider a two-sample, two-component mixture model, and propose a new procedure for testing whether the two samples are identically distributed. The test is built by parametrically specifying the component densities, and involves fitting the model using an EM algorithm. We show that the procedure remains valid when the component densities are misspecified, and if the EM algorithm is prematurely stopped. We evaluate its finite sample performances using simulations. The method is applied to real data.