Moderator Analysis Based on Subgrouping: Problems Arising from the Use of Standardized Variables

The use of parameters based on standardized variables, such as correlation coefficients or beta-weights, in subgrouping moderator analysis is addressed. It is suggested that a comparison of correlation coefficients or beta-weights across different populations may lead to results that are erroneous with respect to true causal processes. Moreover, it is proposed that even in descriptive analyses parameters based on standardized data may vary across populations simply because of population specific idiosyncracies. It is recommended that subgrouping moderator analysis be based on unstandardized data and that more attention be given to causal considerations. (Author)