Suppose that one has a binary treatment $Z$, and assume that $Z=1|X=x \sim Bern\left(e(x)\right)$. Furthermore, suppose I want to estimate the propensity score by a random forest. Are there consistency and asymptotic results in this case and if so, what are the regularity conditions that are needed to achieve them?

  • $\begingroup$ Well usually people just use boosted models like the ones from the R-package twang, but following their documentation I can't even find anything about specific conditions for it to function, except checking convergence afterwards. See the Vignettes here: cran.r-project.org/web/packages/twang/index.html $\endgroup$ Commented Jul 8, 2023 at 11:09
  • $\begingroup$ If you have a parametric model like this, is there really a need to use a random forest? $\endgroup$
    – Scriddie
    Commented Jul 10, 2023 at 8:28


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.