Explaining Non-Significant Moderations

Schools often teach us how to conduct and interpret Moderations. What they don't teach us is how to explain why a moderation didn't work out for statistical/methodological reasons.

Assuming that the proposed moderation is theoretically sound, what are the possible statistical or methodological reasons for its non-significance?

1 Answer

The reasons an interaction term or the main effect variable (a.k.a. 'moderator') included in an interaction term would be non-significant are largely the same as the reasons any other variable can be non-significant:

• there isn't really an effect of that variable
• the $N$ is too small for the size of the effect / the effect size is too small for the $N$

When referring only to the significance of the main effect, note that the meaning of a main effect variable, when an interaction term is included, is the effect of that variable when the other interacting variable is $0$. So again, there may not be an effect at that point, or it may be too small given your $N$. Also be aware that typically the main effect term is correlated with the interaction term, which expands their standard errors and reduces you power.

• Hmm my situation is a lil odd in that sense. My N = 300, which already has a few potential type 1 errors (some of the 95% CIs are 3 decimal points above zero). If this provides enough power, would I hence be able to conclude that a significant effect does not exist? Oct 4, 2014 at 2:40
• Oh yes, I have strong skews on those variables too. How would this factor into the situation? Oct 4, 2014 at 2:41
• The information in your comments doesn't change anything about the answer. It is the same as what I've already given. Regarding whether you can conclude that the effect does not exist at all, you may want to read my answer here. Oct 4, 2014 at 3:25