2-SAMPLE TESTS WITH MULTINOMIAL OR GROUPED FAILURE TIME DATA

We study families of two-sample tests with grouped or multinomial data. The tests are based on models for which some function of hazard or continuation ratio probabilities in two distributions differ in some systematic way and include well-known tests such as the discrete Mantel-Haenszel and a Pearson two-sample test. Asymptotic power formulas and Pitman asymptotic relative efficiencies for pairs of tests are provided and simulation is used to study the adequacy of asymptotic power formulas for finite-sample calculations. Results in the paper are used to help design experiments for comparing lumber preservatives.