Code for Counterfactual Situation Testing: Uncovering Discrimination under Fairness given the Difference. ACM Conference on Equity and Access in Algorithms, Mechanisms, and Optimization (EAAMO), pages 1-11, 2023.
All datasets are contained within the data/ folder. The scripts get_data_< > (also within the data/ folder) prepare each dataset for the counterfactual situation testing pipeline. The rest of the scripts are within the src/ folder.
The scripts get_cf_data_< > generate the counterfactual dataset for the < > case. In the case for the law school (real) data, a second get_cf_data_ script is used for the intersectional discrimination case (denoted by _int). For this dataset in particular, we generate the counterfactual dataset in R to be able to compare the more 'frequentist' and 'Bayesian' approaches for the abduction step. The latter approach uses RStan to run a MCMC.
The scripts run_exp_< > run the experiments -- standard situation testing, counterfactual situation testing with/without centers, and counterfactual fairness -- for the < > case.
Finally, the scripts analysis_< > contain the additional analysis (e.g., relevant figures) performed on the < > case. Results fall within the results/ folder. Different runs (in terms of parameter runs) are stored in specific folders.
Notes:
-
See the eeamo2023 branch for the code specific to the conference paper; new experiments for a future journal version will start soon.
-
The folder scm_models contains an abstract class for structural causal models as well as an example class for the loan application scenario; ignore this folder as it is not used in the current implementation.