I guess it depends on the degree of collinearity. Can you be more specific?
Christian Geiser Thank you for responding to my question. I want to describe my situation. I did a linear regression analysis and there was an observed relationship between the variables, but after a moderation analysis in AMOS, the relationship between the independent variables and the dependent variable disappeared, but a strong relationship between the interaction variable and the dependent variable was observed. To explain why this was the case, I carried out a hierarchical regression analysis and there was no multicollinearity between the variables without the interaction variable, but when the interaction variable was added to the analysis, the multicollinearity for one independent variable was 6.497, for the other independent variable was 7.565, and for the interaction variable was 17.852.
It is impossible to know what happened without knowing more about your exact moderation analysis (sample size, types of variables, equations, centering, etc.).
Did you center the predictors before creating the interaction term? Centering takes away part of the collinearity. (Collinearity is not necessarily problematic unless it leads to strongly inflated standard errors and/or other obvious problems.) Also, the interpretation of the lower-order regression coefficients changes when you add the interaction term (relative to an analysis without the interaction), and the interpretation of the regression coefficients for lower-order terms is different for a centered vs. uncentered analysis. Therefore, in order to comment on this, one would have to know more about exactly how you specified your moderation model. See
Aiken, L. S., & West, S. G. (1991). Multiple regression: Testing and interpreting interactions. Newbury Park: Sage.
Cohen, J., Cohen, P., West, S. G. & Aiken, L. S. (2003). Applied multiple regression/correlation analysis for the behavioral sciences. Mahwah, NJ: Erlbaum. (Chapters 7 and 9)
Hayes, A. F. (2018). Introduction to mediation, moderation, and conditional process analysis (2nded.). New York: Guilford Press.
Christian Geiser My sample is 241 subjects. I have standardised and centred the kinematics, multicollinearity remains unchanged.
Check at https://stats4edu.com/services/structural-equation-modeling/
- Badges
- Science topic
- Similar topics
- Psychology