交互作用和間接效果

Abstract

將一個迴歸函數看成因果關係的果(effect)而進一步的剖析是否有不同性質的效果是許多不同領域共同興趣。受最多關注的是是否存在交互作用的問題。而交互作用通稱為解釋變數因相互關係而產生的結果。另外，尤其在社會學、政治學與流行病學裡特別想剖析的是迴歸函數是否存在由一個解釋變數經由其他解釋變數再來影響反應變數的效果。此一效果稱之為間接效果(indirect effect)。我們將分析當所有解釋變數的間接效果加起來是否等於交互作用的問題。

Considering a regression function as a causality relationship and attempting to decompose this regression function into effects of various types is a topic very popular for many years. Among the various effect decomposition, the interaction assessment has been received very much attention. On the other hand, decomposition of the effect of one explanatory variable into effect without mediation by other explanatory variable and effect with mediation by on specific explanatory variable is useful in sociology, political and epidemiology. The aim of this paper is to analyze the relationship between these two important topics.