Functionality for true continuous treatment

[blind-contours]

Hey Nima,

Is there work underway to make this package work for continuous treatment? I've tested it under simple continuous A situations and get errors related to eif_component_names for onestep estimation and estimator_args[["max_iter"]] error for tmle.

[nhejazi]

thanks for trying out the package for this, David, and sorry you ran into some trouble with it. apparently, I made a poor design decision, since the wrapper function (medshift()) should have immediately stopped you from trying this with a non-binary treatment variable. currently, the package support Kennedy's incremental propensity score intervention (which is described for mediation in Diaz + H 2020), which applies to binary variables, but does not implement the modified treatment policy (MTP) shift intervention A + delta. this has to do with the resultant estimators being challenging to implement in that setting, as (if I recall rightly) they require evaluating expectations wrt a post-intervention mediator density e(Z + delta | A, W), which cannot be reparameterized in a convenient way, unlike the non-mediational case. I'll try to implement a check to restrict to binary A sometime soon and leave this open in case others run into this issue.