ENH: GLM outcome models in TreatmentEffect (part of #8987)

Is your feature request related to a problem? Please describe

I work with binary endpoints and would like to use TreatmentEffect with a logistic outcome model. Currently, the outcome model is limited to OLS the docstring says Logit and Poisson "will become available in future", and #8987 tracks this

Describe the solution you'd like

Being able to pass a GLM as the outcome model, e.g.

TreatmentEffect(GLM(y, x, family=Binomial()), treat, results_select=res_probit)

and have all five estimators (ra, ipw, aipw, aipw_wls, ipw_ra) work with it.

Describe alternatives you have considered

No response

References

• Stata: teffects ra/aipw/ipwra ..., omodel(logit|probit|poisson) ([TE] teffects)
• @pzivich's comment in #8987 suggesting generic GLM support; zEpid's AIPW handles binary outcomes this way
• R marginaleffects (g-computation on GLMs)

Additional context

No response