Interaction concept put into plain English


TS Contributor
Interaction concept in plain English

as you may know, I am not a statistician, rather a 'consumer' of statistics. When I study something new, or when I present statistics to a non-stats audience, I usually try to put stats in layman terms, trying to get/give an idea at a commonsense level before going into some specifics.

That said, as per title, I would appreciate if anyone here could explain in layman terms the idea of interaction between predictors in the context of binary Logistic Regression. A practical example (or reference to an understandable worked example) would be extremely appreciated as well.



Omega Contributor
Well there is effect modification, so the relationship between two variables influences their effects on the outcome. The relationship can result in synergistic or antagonistic changes in their effects. So when both are present, the odds for the outcome are greater (or worse) than would be expected given your knowledge of their independent effects. Yeah, I don't think I did a good job dumbing it down, but I typically use the following example, sorry I do have the real numbers at hand. Also, interaction can be on the multiplicative or additive scale, but that concept may just muddy the waters.

Smokers are at risk for lung cancer (binary outcome)
Asbestos workers are at risk for lung cancer.
Asbestos worker who smoke have a risk for lung cancer that is greater than just adding the two independent risks factors together.