Student Seminar Series - November 15, 2005
University of Minnesota
School of Statistics
College of Liberal Arts

Hypothesis Tests with Multiple Linear Inequality Constraints


Yumin Huang


Tuesday, November 15, 2005
2:00 PM, 300 Ford Hall
Minneapolis, East Bank Campus

Refreshments at 1:30 PM
300 Ford Hall


Abstract

We consider the problem of testing the null hypothesis that a multivariate normal mean vector is constrained to lie in a set satisfying multiple inequality constraints, which can also be described as testing multivariate one-sided null hypotheses. Lehmann (1952) showed unbiasedness does not hold for such tests. Related but different problems of testing one-sided hypotheses have been investigated by Liu and Berger (1995), Perlman and Wu (2004). Our null hypothesis is a polyhedral convex subset of a Euclidean space. Our alternative hypothesis is the complement of the null. We give examples showing the biasedness of several traditional tests and present a new test having better properties. Our new test consists of two stages: pretest and posttest. In the pretest, a face of the null hypothesis is selected according to the value of PIC (pretest information criterion). In the posttest, the null hypothesis is rejected if the likelihood ratio statistic is greater than a critical value that depends on the face chosen in the pretest. The new test is constructed such to be pointwise asymptotic unbiased. We also discuss computational methods for finding the critical values of the test.