Testing for Equality of Means with Incomplete Data on One Variable: A Monte Carlo Study

Abstract

This article is concerned with the relative merit of the procedures proposed by Lin [2], Metha and Gurland [3], and Morrison [5] for testing the equality of means of a bivariate normal distribution with correlation ρ and common variance σ² when some observations on one variable are missing. A Monte Carlo study of the powers and level of significance is conducted. Comparisons among these tests and the paired t test are made. The preferred test is established for each combination of ρ and difference of means.