11-30-2016 01:43 PM
I am trying to build an uplift model for the first time. I found a paper on uplift modeling by Victor Lo (see attached).
I am following the procedure mentioned in the paper. But I am not sure how to handle the interaction effects between the treatment variable and the independent variables. When I run a model with variables (both main and interaction effects) whose interactions with the treatment variable are NOT significant (p>0.05) I am getting, as expected, a declining response rate in the deciles (by differential score: treatment - control) and a better uplift than when I run it with variables having a significant (p<0.05) interaction with treatment. Shouldn't it be the other way around? Am I missing something here?