Fair and Robust Estimation of Heterogeneous Treatment Effects for Policy Learning

Authors: Kwangho Kim, Jose R Zubizarreta

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate the methods in a simulation study and illustrate them in a real-world case study.
Researcher Affiliation Academia 1Department of Statistics, Korea University, Seoul, South Korea 2Department of Health Care Policy, Harvard Medical School, MA, USA 3Departments of Biostatistics and Statistics, Harvard University, MA, USA.
Pseudocode No The paper describes the estimation steps in prose and mathematical formulations but does not include a clearly labeled pseudocode or algorithm block.
Open Source Code No The paper does not provide any statement about releasing the source code for the methodology or a link to a code repository.
Open Datasets Yes We next illustrate our methods on the COMPAS dataset originally gathered to assess the risk of recidivism (Angwin et al., 2016).
Dataset Splits Yes We use roughly two-thirds of the data to estimate bg, and the rest to estimate the welfare and unfairness using the same setting as in the preceding subsection.
Hardware Specification No The paper does not provide specific details about the hardware used to run the experiments.
Software Dependencies No All nuisance functions are estimated using the cross-validation super learner ensemble implemented in the Super Learner R package to combine generalized additive models, adaptive regression splines, and random forests. (Specific version numbers for the R package or other libraries are not provided).
Experiment Setup Yes b(W) consists of the polynomial terms W, W 2, W 3 and {Wj Wk Ws}j,k,s to form the third-order Taylor expansion. All nuisance functions are estimated using the cross-validation super learner ensemble implemented in the Super Learner R package to combine generalized additive models, adaptive regression splines, and random forests. ... estimate bτ using (b P) with K = 2 splits, under δ = (i.e., with no fairness constraints) and δ = 0.