效果解構：從單變量到多變量

Abstract

這篇文章提出對效果解構的進一步研究。對Baron and Kenny的三步驟檢定及由高子瑗提出的兩步驟檢定做了模擬分析研究。這篇文章的一個重要貢獻是提出一個具有數個媒介量(mediators)的效果解構分析。在這個多變量的問題裡，我們推導出Baron and Kenny的效果解構以及不具爭議的效果解構。這些直接、間接及全部效果均可以寫成分配參數的函數或是多元迴歸之係數的函數。最後我們進行了資料分析並呈現其結果。

This paper provides a further research on effect decomposition. We re-prove the theory of effect identity applied by Baron and Kenny (1986) to define direct, indirect and total effects. We also conduct a simulation to compare the powers between tests of Baron and Kenny’s three steps and the new two steps introduced by高子瑗 (2015). We introduce effect decomposition when there are multiple mediators. This allows us to present Baron and Kenny’s effects and new effects without ambiguity in this multivariate version. Formulations of direct, indirect and total effects in terms of distributional parameters and coefficients of multiple linear regressions are introduced.