The difference between interaction and moderation effects

noetsi

Fortran must die
#1
I don't really understand the difference between these. Say that you think the impact of X1 on Y is influenced by X2 (with the X1 and X2 being predictors). When do you have moderation effect and when interaction?

How do you model moderation?
 

spunky

Doesn't actually exist
#2
Moderation and interaction effects are the same thing. "Moderation" is more used in social-science land. "Interaction" is what everybody else uses. But it's the same thing.
 

hlsmith

Not a robit
#3
In the epidemiologic literature, there is a distinction based on, I believe, your ability to modify the variables or not. I can't remember which one needs to be mutable? But analytically they are pretty much synonyms.
 

spunky

Doesn't actually exist
#4
We have the same thing in social-science-land where we borrow the language from ANOVA and call a "main effect" the predictor we truly care about and the "moderator" whichever one's levels, we believe, would yield different values for the main effect. But as far as the regression part goes, both are multiplicative interactions so I always end up thinking that really only comes down to semantics so... ¯\_(ツ)_/¯
 

hlsmith

Not a robit
#5
Agreed, but will note additive interaction terms can be used and are favored in epi due to interpretation of a policy change. So kind of like the relative risk versus the risk difference debate.
 

noetsi

Fortran must die
#6
Thanks. Different nomenclature for the same thing always confuses me.

In structural equation models they have indirect effects A>B>C where A has an indirect effect on C through B (as well as a direct effect commonly). I am not sure if that is moderation or interaction or neither.