Using nonexperimental methods to address noncompliance

The analysis compares estimates of the incremental impact for those who receive HPOG with a program enhancement to the standard HPOG program. The experimental benchmark for the incremental impact comes from two-stage least squares with random assignment as an instrumental variable for enhancement take-up. Then, ignoring the randomly assigned conditions, the analysis estimates the counterfactual for those who "take up" the enhancement using ordinary least squares and inverse propensity weighting. The analysis also tests whether adding information that is only available due to the experiment-who complied with their randomization status and who did not-improves the nonexperimental estimates. The analysis compares these estimates using statistical tests recommended by the within-study comparison literature.