Should we winsorize the interaction variables ?

[Lenny Tran]

I have two variable, a and b
a is a normal vari
b is a dummy ( 0 and 1)
c = a*b
c is for interation

I winsor2 the variable a and b, but i dont winsor2 the variable c
is that ok ?

[Joseph Coveney]

Copacetic here, but how do you winsor2 b?

[Joseph Coveney]

OK.

Why are you winsorizing what look like predictors? Leverage problem?

[Phil Bromiley]

Winsorizing like most approaches to outliers has little in the way of theoretical justification. That doesn't mean you don't do it, but it does mean that a good answer about how to apply it is hard to answer. Commentators on this listserve have differing opinions on winsorizing (and other outlier management techniques). One big question about what to winsorize and at what level is whether reviewers will think you've messed with winsoring to manipulate findings.