Hello,
I am using the Chin, Marcolin and Newsted (1996, 2003) method for modelling interaction in a model with formative indicators. This works fine, however I noticed something unexpected: mediation relationships that were not significant initially are now significant.
For example, in the original model IV -> DV and M -> DV would both be significant. Adding the path IV -> M would result in both paths above to lose significance. My interpretation is that M does not mediate IV -> DV. When the latent variable data are used with the same model, all paths are significant in the mediated relationship. The r-squared value for DV does not change however between the mediated and non-mediated scenarios.
Perhaps I am misunderstanding something, however I thought that the structural model would behave the same way regardless of whether the measurement model was calculated ‘on-the-fly’ or provided in advance as per Chin’s method.
Regards,
Michael