HOMOGENEITY FOR MULTINOMIAL POPULATIONS BASED ON φ-DIVERGENCES

Abstract

A test of homogeneity of independent populations is introduced in terms of the φ-divergence. The asymptotic power of the test is determined for local and nonlocal alternative hypotheses. A power approximation is also proposed, based on a normalizing transformation of the test statistics for testing homogeneity. The behaviour of the proposed approximations of the power function is numerically compared. A new statistic is obtained and the respective test is more powerful, in some cases, than the existing tests for testing homogeneity of multinomial populations.