The code in this repository reproduces the results in [1].

This package depends on the dare package, which implements the loss function and interfaces deeptrafo and deepregression for stochastic gradient descent via tensorflow and keras. The dare package can be installed from here.

All results can be reproduced by running make all or executing the scripts in ./inst/code/ manually following the order in the Makefile. The results can also be reproduced in parts. For the 401k application, make 401k; for the schooling application, make schooling; for the simulation results, make run-simulations vis-simulations; Figure 3 can be reproduced with make loss-landscape; and for all other figures use make figures.

[1] Kook, L., & Pfister, N. (2024). Instrumental Variable Estimation of Distributional Causal Effects. arXiv preprint arXiv:2406.19986. doi:10.48550/arXiv.2406.19986.