Support for Categorical D in PLIV

[royzawadzki]

I'm having some trouble with the PLIV on R. It doesn't appear to support binary treatments as it doesn't let you do a classifier for ml_r. Am I doing something wrong here? Thanks

[MalteKurz]

You are not doing anything wrong! The feature you are trying to use is just not yet implemented. Some details and alternatives:

1. To use a classifier for ml_r you need to ensure that the treatment variable is binary with values 0 and 1. Furthermore, the predictions need to be probabilities and not labels. Only then, we have E(D|X) = P(D=1|X) and can plug-in the predicted probabilities for P(D=1|X) into the Neyman orthogonal score. At the moment this is implemented for the PLR but not yet for the PLIV. I will put your feature request on our list but can't predict when it will be picked up. If you like to contribute it, feel free to file a pull-request.
2. For binary treatment variables the IRM and IIVM (interactive regression models, see https://docs.doubleml.org/stable/guide/models.html#interactive-regression-model-irm) could be interesting alternatives. They allow for heterogeneous treatment effects. Whether they are suitable for you primarily depends on what your target parameter is. If your intrumental variable is also binary, you may want to estimate a local average treatment effect (LATE) and then I believe the Interactive IV regression model (IIVM) is a good alternative to the PLIV with classifier for ml_r.