[Statistical analysis of pharmacological data: problem of multiple comparison].

Problems of multiple comparison were discussed without assuming technical knowledge of statistics. For the first question concerning why to use multiple comparison procedures, theoretical bases of statistical inference and multiple comparison (including type I error rate and familywise error rate) were briefly outlined. For the second question concerning how to properly use multiple comparison procedures, multiple comparison procedures were introduced, and their characteristics were compared. Families of comparisons are different among Dunnett's, Tukey's and Scheffe's tests. Assumptions of dose-response relationship are different among Dunnett's, Williams' tests and linear regression analysis. Duncan's test does not control familywise error rate at a fixed level. For the last question concerning what is remarked for multiple comparison, approaches to several problems such as abnormality and heteroscedasticity were provided. Philosophy and strategy to multiple comparison problems were discussed.