要因計画の分散分析において単純主効果検定に使用する誤差項の選択について

抄録

This article discusses the appropriate choice of error terms for testing simple main erects as post-hoc analysis when the interaction was significant in the analysis of variance. More specifically, the problem of choice of error terms arises when we are interested in the simple main effects of the nonrepeated factor (A) in the mixed design. One of the procedures which have been frequently used is to pool the error term for A and the error term for the interaction, A by the repeated factor (B). However, this procedure assumes that the within-cell mean square is homogeneous at all levels of B. As E. F. Lind-quist (1953) and G. Keppel (1973) pointed out, such assumption is not reasonable and contradicts the observations in many cases. An alternative procedure is to compute the within-cell mean square at each level of B. These individual or separate error terms are more appropriate because they don't need the above homogenity assumption.