You probably don't want to include two variables with a bi-variate correlation of .7 or more in the same analysis. If you find yourself in this situation, you may need to consider omitting one of the variables. 请参考-------Pallant, J. (2007). SPSS survival manual: A step-by-step guide to data analysis using SPSS version 4, p. 140, Maidenhead, Berkshire, England: McGraw-Hill Education.
One of these problems is multicollinearity, i.e., high correlations among the latent exogenous constructs. Mason and Perreault (1991) have documented the conditions under which multicollinearity may pose problems in regression. They show that multicollinearity leads to inaccurate estimates of coefficients and standard errors as well as inference errors, but they also argue that the problem should not be viewed in isolation, and that a high R2 and large sample size can offset the problems caused by multicollinearity. Although these factors should also be relevant in the context of SEM, their work is silent about certain issues that are specific to SEM, most notably, measurement error. The ability of SEM to incorporate measurement error makes it difficult to assess the impact of multicollinearity on parameter estimates (Bollen 1989). Mason and Perreault (1991) show that increasing explained variance in the dependent variable mitigates the effects of multicollinearity. Removing measurement error should increase the amount of variance explained by the structural model and, by extension, mitigate multicollinearity. However, measurement error also attenuates correlations among the exogenous variables. The presence of measurement error is likely to mask the true correlation among latent exogenous constructs. Thus, controlling for measurement error should result in higher correlations among the exogenous constructs than not controlling for measurement error. 请参考--------Grewal, R., Cote, J. A., & Baumgartner, H. (2004). Multicollinearity and measurement error in structural equation models: Implications for theory testing. Marketing Science, 23(4), 519-529.
|