G-estimator for the linear IV model for trials with non-compliance, with treatment effect heterogeneity, using Super learner for nuisance fits
This is an R function that implements the g-estimator for a linear IV model with treatment effect heterogeneity as detailed in DiazOrdaz K., Daniel R., Kreif, N. (2018) Data-adaptive doubly robust instrumental variable methods for treatment effect heterogeneity. arXiv e-print (arXiv:1802.02821)
The user can specify whether the non-adherence is onesided or two-sided, treatment effect modification is of interest or not and whether parametric or data-adaptive fits (using the Super Learner SL) are to be used.
If SL is used, the user must specify separately a library of learners suitable for the type of exposure and instrument for the SL to use.
In the paper, we assume both exposure and instrument are binary, but if the user is happy to assume linearity on the exposure holds for the IV model, the methods extends to continuous exposure. See Vanteelandt and Didelez Robustness and efficiency of covariate adjusted linear instrumental variable estimators eprint arXiv:1510.01770