I received the reviewer comment about “testing an interaction (Step 3) based on non-significant main effect of moderating variable on dependent variable (Step 2)”. I ran a hierarchical regression model to test interaction terms. Is a significant direct effect of moderator on dependent variable required to test an interaction term? If yes, how can I deal with this issue? Thank you!
It is correct there must be a significant effect between the moderator and the ID, but the most important is the significant effect of the moderator on the interaction of IV-ID
Hi Hj Cho ,
no that's incorrect. Neither the IV nor the moderator must have a main effect. Both can be completely unrelated to the DV and yet there is a moderator effect.
Can be easily proven with a simulation if you like :)
Best,
Holger
Dr. Holger Steinmetz is right It is not necessary to have a significant direct effect of a moderator on DV. It seems that the reviewer may know the interaction effect quite well. Be nice to the reviewer and provide some references for your argument. The following book would be a starting point.
Cohen, J., Cohen, P., West, S. G., & Aiken, L. S. (2003). Applied multiple regression/correlation analysis for the behavioral sciences (3rd ed.). Mahwah, NJ: Erlbaum.
Dear Hj Cho, in this article we show that the effects of the first-order terms can be transformed so that their conditional effect is zero, a transformation that does not influence the moderator effect:
Moosbrugger, H., Schermelleh-Engel, K., & Klein, A. (1997). Methodological problems of estimating latent interaction effects. Methods of Psychological Research Online, 2(2), 95-111.
By the way, there does not exist any main effects, instead, a linear effect is the conditional effect of the first independent variable when the second independent variable is fixed to zero.
HTH
Karin
Not that it is necessary, but one more echo, no is the answer. This is an issue with which my students often become entangled, until they realize that the effects are independently estimated.