Score function "IV-type" for PLIV with two IVs

[AdamBaumann]

When trying to compare the result of using the "partialling out" and "IV-type" score functions for PLIV with two IVs there is seemingly only the option of using "partialling out". Is it possible to implement the IV-type score function here in a similar fashion as for the PLIV with one IV as stated in the documentation? Below is the exact message regarding this:

Error in private$check_score(self$score) :
Assertion on 'score' failed: Must be element of set {'partialling out'}, but is 'IV-type'.

[PhilippBach]

Hi @AdamBaumann ,

thanks for opening this discussion. Indeed, we are only supporting the 'partialling out' score for PLIV with multiple instruments $Z$ at the moment. Right now, the implementations of more complex IV models is still experimental, so when you use such a model, please make sure you double check with a paper reference and verify the validity in a simulation.

Basically, adapting the score in a way you proposed would require to adjust the lines below, I guess

doubleml-for-r/R/double_ml_pliv.R

Lines 693 to 714 in c9a67bc

	r_hat = dml_cv_predict(self$learner$ml_r,
	c(
	self$data$x_cols,
	self$data$other_treat_cols,
	self$data$z_cols),
	self$data$treat_col,
	self$data$data_model,
	nuisance_id = "nuis_r",
	smpls = smpls,
	est_params = self$get_params("ml_r"),
	return_train_preds = FALSE,
	task_type = private$task_type$ml_r,
	fold_specific_params = private$fold_specific_params)
	d = self$data$data_model[[self$data$treat_col]]
	y = self$data$data_model[[self$data$y_col]]
	if (is.character(self$score)) {
	if (self$score == "partialling out") {
	psi_a = -r_hat$preds * d
	psi_b = r_hat$preds * y
	}

As far as I see, it should not be too much effort, so feel free to implement it yourself and open a pull request (see our Contributing guidelines).

Extending IV models is definitely on the list of the next features we want to include in DoubleML. We are currently working on some other topics, so we haven't completed the work on this yet.

[PhilippBach]

In case you need help or guidance when contributing to our package, feel free to contact us at anytime!