:package: :game_die: R/txshift: Efficient Estimation of the Causal Effects of Stochastic Interventions, with Corrections for Outcome-Dependent Sampling
Currently, in generating the auxiliary covariate for the efficient influence function, a bounding-type procedure is implemented for the post-intervention (counterfactual) conditional density ratio, where non-finite values are set to 1. This should be extended to the case of natural/observed conditional densities, since poor estimates of such could lead to numerical instability in downstream steps of the procedure. See https://github.com/nhejazi/txshift/blob/master/R/fit_mechanisms.R#L443-L448. Suggested by @jeremyrcoyle.
nhejazi
commented
5 years ago
Currently, in generating the auxiliary covariate for the efficient influence function, a bounding-type procedure is implemented for the post-intervention (counterfactual) conditional density ratio, where non-finite values are set to 1. This should be extended to the case of natural/observed conditional densities, since poor estimates of such could lead to numerical instability in downstream steps of the procedure. See https://github.com/nhejazi/txshift/blob/master/R/fit_mechanisms.R#L443-L448. Suggested by @jeremyrcoyle.