潛在調節徑路模型的模型設定

Model Specification of Latent Moderated Path Model

中介與調節效果是社會與行為科學中，用以設想現象的分析架構，Edwards與Lambert（2007）提出了調節路徑分析的一般性架構，同時考量中介與調節效果，而提出八種可能的模型。由於該架構基於迴歸分析，在測量誤差較大時可能影響係數估計，本研究建議將此架構推展至潛在變項的版本，而能利用結構方程模型處理測量誤差。八種模型中，基礎中介模型設定相當簡易，本研究則推導其中五種模型之模型設定與所需限制式，並提供包括基礎中介模型之六種模型之LISREL程式供使用，LISREL程式正確性則以模擬資料加以確認。 Mediation and moderation are two popular analytic frameworks for social and behavioral sciences. Edwards and Lambert (2007) proposed the moderated path analysis framework, which involves eight models, to incorporate mediation and moderation into a comprehensive framework. However, when there are considerable measurement errors, the above approach may lead to biased estimates since it is based on regression analysis. This study aims to extend the moderated path analysis framework into its latent variable version which is reformulated with structural equation models. By doing so, measurement errors can be taken into account. Besides the basic mediated model, model specification and associated constraints of five models are derived from the study. LISREL codes for the six models are also provided and their accuracy is demonstrated by simulated datasets.