兩種多群體階層線性模式的研究

The Study of Two Types of Multi-group Hierarchical Linear Modeling

本研究延伸因素分析與結構方程模式的多群體分析概念應用到多層次研究，透過分群變項將樣本資料區分為兩個不同群體進行多層次分析的多群體比較，以檢視多層次模式上的各個研究變項對結果變項的影響是否會因不同群體而異。由於多層次資料存在兩個階層結構的特性，本研究根據分群變項來自個體與組織兩種不同層次，提出兩種類型的多群體階層線性模式，分別以方程式與圖示說明這兩種類型的差異，並透過24 家公司664 位員工的實徵資料，分別以員工性別與公司的產業特徵製造業與否分成兩個群體，進行組織層次變項組織創新氣氛與個體層次變項組織承諾對員工工作滿意的多群體多層次分析示範，分別檢視員工性別與公司產業特徵對多層次研究架構的調節效果。本研究所提出的兩種類型階層線性模式不僅可以同時進行兩群體迴歸係數的估計外，亦可以進行不同群體間迴歸係數的差異檢定進行對等性的比較。最後，提出本研究多群體階層線性模式的相關應用與限制。

The study extends the concepts of multiple-group analysis of traditional confirmatory factor analysis and structural equation modeling to the multilevel research. Based on the grouping variables, we separate the whole two-level structure data into two sub-group samples and apply the multiple-group analysis of multilevel modeling to examine the moderation effect of the grouping variables of the impacts of the explanatory variables on the outcome variable in the multilevel research. Because of the two-level structure property of the multilevel modeling, the grouping variables maybe belong to the individual level or the organizational level. Based on the different level of grouping variables, we create two types of multiple-group hierarchical linear modeling and introduce their rationales, equations and diagrams to compare their differences of the structure relationships between two groups. In addition, we provide the empirical data to demonstrate how to analyze the multiplegroup analysis of the multilevel modeling. Through the organizational variable: organizational innovative climate and individual variable: organizational commitment and outcome variable: employees’ satisfactory of 664 employees from 24 organizations, we separate the sample into two groups based on the employee’s gender and industrial characteristic of the company respectively and run HLM software to illustrate the parameters estimations and the contrast tests of the two-group regression coefficients on these two types of multiple-group hierarchical linear modeling. Finally, we conclude their extensions and limitations.