在实证研究中,构念和观测变量之间的关系经常被忽视.本文详细区分了反映式测量模型和构成式测量模型,通过分析我国管理学领域的三本学术刊物2002年到2007年以结构方程模型(SEM)为数据分析工具的实证研究论文,发现我国管理学研究中测量模型误设的情况普遍存在.蒙特卡罗仿真分析表明,测量模型的误设将导致相关路径系数的被显著扩大或缩小,而且可能导致I型错误和Ⅱ型错误.由于现有的SEM软件不能处理构成式测量模型,本文提出了模型细化法和模型分解法,能够将构成式测量模型转换成反映式测量模型.%Structural Equation Modeling (SEM) is a prevalent research method in the Chinese management study. SEM can be divided into measurement models and structural models. A well-specified measurement model is a prerequisite to analyzing the structural model. Reflective and formative measurement models are two different measurement models. Measurement model miaspecification has been a serious problem in the fields of Marketing, Organizational Behavior, and Management Information Systems. Measurement model misspecification can lead to invalid research results. Many approaches can facilitate the internationalization and normalization of SEM in Chinese management studies. These approaches include the discussion of the difference between reflective and formative measurement models, analysis of measurement model misspecification problems, and investigation of the consequence of measurement model miaspecification problems. In the first part, we discuss major differences between reflective and formative measurement models. Measurement models include reflective and formative measurement models. In the reflective measurement model, constructs are treated as causes of measures, and the measures are reflective manifestations of underlying constructs. In the formative measurement model, measures are specified as causes of constructs, and the measures represent underlying latent variables. Reflective and formative measurement model can be further distinguished by various features. We discussed these features from the perspectives of hierarchism, causality, correlation between measures, deletion of measures, and measurement errors.In the second part, the current status of measurement model misspecification problems in Chinese management studies was analyzed. We chose 315 published papers from three top Chinese management journals between 2002 and 2007. All these studies used SEM as their analytic tool.A total of 315 measurement models, consisting 314 reflect measurement models and 1 formative measurement model, were identified. Among all measurement models, 111 measurement models were not specified correcfiy because these models should be specified as formative measurement models. The average ratio of measurement model miaspecification problems is 35. 24 percent of all papers published in top three Chinese management studies from 2002 to 2007.In the third part, Monte Carlo simulation is conducted to investigate the negative effect of measurement model misspecifications.We conduct 12 simulations using EQS 6. 1 on the position of misspecification measurement model, sample size, and the correlation between formative measures. The simulation results indicated that measurement model misspecification may lead to estimation bias of path coefficients. Measurement model misspecifications deflated the path coefficients between the misspecified constructs and their antecedents, inflated the path coefficients between the misspecified constructs and their dependent variables, and have no effect on path coefficients between latent variables. The results also indicated that measurement model misspecifications may lead to Type I or Type Ⅱ errors.In thefourth part, two methods are used to transform formative measurement model to reflective measurement model. The transformation method shows that measurement model misspecifications have negative influence on the validity of structural models and newly-built theories. Maximum likelihood-based covariance SEM cannot be applied directly to the transformation of formativemeasurement models. This study adopts the model refining and model decomposition methods to transform formative measurement models to reflective measurement models.In summary, this paper clarifies the difference between reflective and formative measurement models and analyzes the current status of measurement model misspecification problems that exist in Chinese management studies. Our analyses of these problems indicate that estimation bias of path coefficients can be resulted from improper measurement model specifications. Researchers interested in the Chinese management study should pay attention to problems associated with measurement model miaspecifications. A careful specification of measurement models can help derive reliable and valid theoretical results and provide useful practical implications.
展开▼
