I am using RCT data to estimate the impact of a program. After doing the straightforward analysis, I decided to estimate the program impact by subgroups (treatment status*subgroup). The subgroups were created using post-program data, which is potentially endogenous. The reviewer recommends Heckman sample selection correction, which I suspect is not appropriate. Would you have other recommendations? Is it possible to do subgroup analysis for characteristics that are determined after treatment exposure? Thank you!!