A New Paradigm for Counterfactual Reasoning in Fairness and Recourse

Authors: Lucius E.J. Bynum, Joshua R. Loftus, Julia Stoyanovich

IJCAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 5 Experiments In this section, we demonstrate our criteria computationally on real and simulated data.
Researcher Affiliation Academia Lucius E.J. Bynum1 , Joshua R. Loftus2 and Julia Stoyanovich1,3 1New York University, Center for Data Science 2London School of Economics, Department of Statistics 3New York University, Tandon School of Engineering lucius@nyu.edu, J.R.Loftus@lse.ac.uk, stoyanovich@nyu.edu
Pseudocode Yes Algorithm 1 Backtracking Counterfactual Sampling
Open Source Code Yes Code to reproduce all our experiments is available online.5 Our code repository is located on Git Hub at: https://github.com/ lbynum/backtracking-counterfactual-fairness-and-recourse.
Open Datasets Yes Used to illustrate counterfactual fairness in [Kusner et al., 2017], the law school dataset from [Wightman, 1998] contains information on law students demographics and academic performance compiled from a 1998 Law School Admission Council survey.
Dataset Splits No The paper mentions 'datasets of size 500' and 'random sample of 5000 observations' but does not specify explicit training, validation, or test dataset splits.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper does not list specific software dependencies with version numbers (e.g., programming language versions, library versions, solver versions) used in the experiments.
Experiment Setup Yes We fit several predictors on datasets of size 500 from the above equations, each a linear regression using some subset of the available covariates (excluding A)." and "We generate data for Scenario 2 with a small modification to Scenario 1: A Bern(0.5), ZA N(A/2, 1), ZA N(0, 1), X1 N(2ZA + ZA , 1), X2 N(3ZA , 1), and Y N(X1 + X2 + 2ZA + ZA 2, 1).