结构方程模型统计检验力分析:原理与方法

Power analysis in structural equation modeling:Principles and methods

结构方程模型是心理学、管理学、社会学等学科中重要的统计工具之一。然而,大量使用结构方程模型的研究忽视了对该方法的统计检验力进行必要的分析和报告,在一定程度上降低了这些研究的结果的证明效力。结构方程模型的统计检验力分析方法主要有Satorra-Saris法、MacCallum法与Monte Carlo法三类。其中Satorra-Saris法适用于备择模型清晰、检验对象相对简单、检验方法基于χ2分布的情形;MacCallum法适用于基于χ2分布的模型拟合检验且备择模型不明的情形;Monte Carlo法适用于检验对象相对复杂、采用模拟或重抽样方法进行检验的情形。在实际应用中,研究者应当首先判断检验的目的、方法以及是否有明确的备择模型,并根据这些信息选择具体的分析方法。

Structural equation modeling(SEM)is an important statistical tool in psychology,management,and sociology.However,the lack of analyses and reports of statistical power in many studies using SEM has reduced the probative power of the study results.There are three main types of statistical power analysis methods for SEM:the Satorra-Saris,MacCallum,and Monte Carlo methods.The Satorra-Saris method applies to cases where the alternative model is clear,analysis object is simple,and the statistic to be examined is based on theχ2 distribution.The MacCallum method is applicable to the case ofχ2-based overall model fit tests with unknown alternative models.Further,the Monte Carlo method applies to cases where the analysis object is complex,or the test is performed using simulation or resampling methods.In practice,researchers should clarify the purposes and methods of the test as well as the availability of alternative models as a priority.The research method can then be determined on the basis of the above information.