Abstracts presented at the 1998 Annual Meeting of the International Biometric Society, Japan Region, jointly held with the Japanese Society of Applied Statistics at Chuo University, Tokyo (May 12-13, 1998)

s presented at the 1998 Annual Meeting of the International Biometric Society, Japan Region, jointly held with the Japanese Society of Applied Statistics at Chuo University, Tokyo (May 12-13, 1998). ASSESSMENT OF INFLUENCE OF HETEROSCEDASTICITY IN MULTIPLE COMPARISON PROCEDURES Kenji Abe and Manabu Iwasaki Department of Industrial Engineering, Seikei University: [email protected] Many multiple comparison procedures for the one-way ANOVA model assume homoscedasticity in different treatment groups. In practical toxicological studies and also in clinical trials, however, we often encounter experiments with heteroscedasticity in groups. In this paper, for frequently-used multiple comparison procedures such as Dunnett's test and detection of dose-response relationships, some of the consequences of the heteroscedasticity are assessed by theoretical consideration and also by computer simulations. It is shown that p-values for some particular contrasts tend to be much smaller than expected. In order to avoid such difficulty, we propose certain procedures which take the heteroscedasticity into account, and also show that such procedures have desirable properties in comparison of means between different groups. COVERAGE PROBABILITIES OF CONFIDENCE INTERVALS FOR THE DIFFERENCE IN PROPOTIONAS FOR THE PAIRED-SAMPLE DESIGN