1993 年 14 巻 1.2 号 p. 17-30
This paper critically revicws two important concepts which should be considered even in the analysis of clinical trials. These are interaction and confounding. Both must often be a great burden for the interpretation. It may be more difficult to detect the existence of confounding than interaction. It is noticed that testing for the imbalance of baselinc factors is a part of the check for confounding. It illustrates a realistic situation in which both confounding and interaction are involved. Tests for interaction are shown as the basis of odds ratio, risk ratio and risk difference, with an empirical comparison. Finally, testing for qualitative interactions is discussed for a possiblc application.