Wed, 02/20/2013 - 08:30

Hi.

I am a little bit confused here. I am running a univariate model including GxE interaction in it. As a result, I am getting that c path is very close to zero and I can omit it. However, its moderation is statistically significant and I have to keep it in the model. How would one interpret that?

On one hand, I understand that when we don't take a moderator into account (or M=0), we calculate paths for all persons in the data set despite their difference in the measured environment. And in this case, the c path on average might be zero, but it does change with different values of the moderator.

On the other hand, I am a bit worried about the model. How can moderation of a path be estimated if this path does not exist on the first place?

Anyone who can disentangle this for me, please? Thank you in advance.

If I understand correctly, it's just like moderation in regression. The main effect might be zero (and not statistically significant), but still have a non-zero interaction. I think you're on the right track when you say "in this case, the c path on average might be zero, but it does change with different values of the moderator.".

I suspect the formula for the path looks like this:

cTotal = c0 + mod * c1

Therefore, for values of mod not equal to zero, cTotal is not equal to zero.

I thought so too, but then got confused at some point and worried that the modelling might go wrong if I try to estimate the moderation of something that does not exist. But it is indeed just like in a regression model: the intercept might be zero while the interaction is significant.

Thank you!