ARE, AIE and MIG estimators for ITR evaluation from observation data

This repository reproduces results from the paper A Comprehensive Framework for the Evaluation of Individual Treatment Rules From Observational Data. The code implements the average rule effect (ARE), average implementation effect (AIE) and maximal implementation gain (MIG) estimators as proposed by Grolleau, Petit and Porcher (2022).

Authors

This package is written and maintained by François Grolleau ([email protected]).

Reproducibility

• The new_itr_situation folder contains the following files.

figure_2.R implements the toy example given in the paper and repoduces Figure 2.

boot_func_new_itr.R contains the bootstrap functions used for the new ITR situation application.

mimic_new_itr.R reproduces Figure 5 for the new ITR situation application.

• The partially_implemented_itr_situation folder contains the following files.

algo1.R implements the EM algorithm from the paper and returns ARE, AIE and MIG estimates. ARE, AIE and MIG estimates along their bootstrap standard errors can be obtained in one line of code. An example is given at the end of the file.

simulations.R reproduces the simulations given in the paper.

plot_results.R plots the results of the simulations and reproduces Figure 4 from the paper.

References

François Grolleau, François Petit and Raphaël Porcher. A Comprehensive Framework for the Evaluation of Individual Treatment Rules From Observational Data. 2022. [arxiv]