In a typical moderation model with one moderator, is it a problem if the moderator and the dependent variable show a strong correlation (say 0.85)? What if a strong correlation exists between the moderator and independent (predictor) variable?
Another related question to moderation is that: Why sometimes when we switch the dependent and independent variables in a moderation model with the same moderator variable, the significance of the moderation changes? For example, when a model with say rumination as the independent variable, depression as the dependent variable, and gender as the moderator variable is significant; why it become non-significant when we switch the location of dependent and independent variables, put depression as independent and rumination as dependent (the example is not real, just to convey the pint).
Please let me know if any further information is required to make it clearer. Thanks in advance for your help :)
Q1. I don't see why that would be a problem but it should be fairly easy to check.
Q2. Well if you switch x,y the problem is no longer regression but is often called calibration. Because the RHS and LHS are switched the probability status of the x and y terms has changed. Ott and Longnecker's Introduction to Statistical Methods and Data Analysis have a nice introduction to calibration. Best, David Booth
The relationship is no problem at all for the method, in the PROCESS analysis is based on regression, it depends on how you catalog the study variables as well.
I agree with both David Eugene Booth and Cristian Ramos-Vera
Dear Mahdi Mazidi
As mentioned above, strong correlations between study variables (including moderator) is not an issue if the correlation values do not cross multicollinearity threshold (mostly vary from 0.80 to more than 0.90). In other words, general redundancy rule applies for correlation values.
In this respect, you need to differentiate between continuous (for example age in years) and categorical (gender) nature of the moderator because both are dealt differently in moderation analysis. For continuous moderator most of the time we need to mean center both independent and moderating variables and calculate their interaction term to regress with the dependent variable.
IV*M = Interaction Term (only if the interaction term is significant then we can move further to the interaction plot to look for the impact on two or three different levels of the moderator)
As for switching IV and DV it changes the whole theoretical foundation as well as the interaction term and so obviously output of the moderation also changes.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics