There is a question What if interaction wipes out my direct effects in regression?. An answer was given that the true main effects are in the model without the interaction.

I have the opposite situation. Main effects are not significant in step 1. A main effect is significant in step 2 after adding the interaction, but the interaction is not significant. Can I rightfully assume this is not a true main effect?

DV: guilt
Step 1: Moral conviction (MC) + Speaking out (SO) 
Step 2: MC + SO + MCxSO
  • $\begingroup$ "Can I rightfully assume this is not a true main effect?" Yes. $\endgroup$ – Jake Westfall May 11 '13 at 2:55
  • 3
    $\begingroup$ The difference between significance and non-significance is not necessarily significant. Are you confusing small changes in p-values across the .05 border with practical importance? or has he size of the main effect changed in a theoretically meaningful way? $\endgroup$ – Jeromy Anglim May 11 '13 at 3:55
  • 1
    $\begingroup$ It depends on your definition of "true main effect" and it is, I think, not useful to get hung up on that; instead, present both models (with the coefficients) and describe what is going on. $\endgroup$ – Peter Flom - Reinstate Monica May 11 '13 at 12:16
  • $\begingroup$ Thank you for your responses. In this case the p-values across the border go from .090 to .047. The "what to report" becomes an issue because I have over 100 regressions (my dissertation chair had me run all these extra analyses) to report and I'm trying to summarize results appropriately. $\endgroup$ – Lisa May 11 '13 at 15:10
  • $\begingroup$ Possible responses for your chair: 1. Should the concepts of parsimony or penalization play any role in this work? 2. At what point would I need to start accounting for the multiple comparison problem? 3. Why stop at about 100? $\endgroup$ – rolando2 May 11 '13 at 19:03

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.