assign some covariates a treatment effect but others not

[mgiangreco]

In the case where we have k categorical covariates, is it possible to have only a subset n of these k covariates generate a treatment effect, while the remaining k-n covariates exhibit no treatment effect at all? I see that we can set treatment_option_weights, but these weights seem to applied evenly across all categories.

[mgiangreco]

see https://github.com/mgiangreco/heterogeneous-effects/blob/master/notebooks/problem_framing.ipynb

[mgiangreco]

ideally, for some subgroups the treatment effect would be 0 for assignment = 0 and for assignment = 1, whereas in other subgroups the treatment effect would be 0 for assignment = 0 but something other than zero for assignment = 1 -- i.e., there is some true effect of treatment, but it only applies to certain subgroups.