2. Fairness Extensions in Optimal Policy#
The OptimalPolicy class in mcf includes experimental capabilities for fairness adjustments, available through its solvefair method.
These adjustments aim to improve fairness in policy allocations with respect to specified protected variables, following the methodology introduced in the forthcoming work of Bearth, Lechner, Mareckova, and Muny (2025).
The approach is based on variable preprocessing. The guiding principle is straightforward: if the variables used to construct the decision rule are independent of certain protected features, then the resulting decisions will also be independent of those features. In practice, this means that individuals in each protected group should have an equal probability of being assigned to any given treatment.
To transform a variable \(X\) with respect to a protected feature \(S\), the method proceeds in two steps:
Conditioning: Compute the cumulative distribution function (CDF) of \(X\) given \(S\),
\[u = F_{X \mid S}(X \mid S)\]so that \(u\) is uniformly distributed within each level of \(S\).
Marginal Mapping: Apply the inverse marginal CDF of \(X\),
\[\tilde{X} = F_X^{-1}(u)\]to restore the original marginal distribution.
The transformed \(\tilde{X}\) can be interpreted as a fairness-adjusted version of \(X\): it preserves the original distribution of \(X\) but eliminates its statistical dependence on the protected feature \(S\).
The adjustment can be applied to the decision variables, the policy scores, or both. Note that the fairness adjustment implemented here may slightly differ from the R implementation used in Bearth, Lechner, Mareckova, and Muny (2025).
2.1. Example#
To use the fairness adjustments, configure the OptimalPolicy class with the appropriate parameters and call the solvefair instead of the solve method to build the decision rule.
import os
from mcf.example_data_functions import example_data
from mcf.optpolicy_main import OptimalPolicy
# Generate data
training_df, prediction_df, name_dict = example_data(
obs_y_d_x_iate=1000,
obs_x_iate=1000,
no_features=5,
no_treatments=2,
seed=12345,
type_of_heterogeneity='WagerAthey',
descr_stats=True
)
# Define parameters for OptimalPolicy
params = {
'var_d_name': 'treat',
'var_polscore_name': ('y_pot0', 'y_pot1'),
'var_x_name_ord': ('x_ord0', 'x_ord1'),
'var_protected_name_ord': ('x_ord0'), # Specify at least one protected feature
'gen_outfiletext': "OptPolicy_Simple_Example",
'gen_outpath': os.getcwd() + '/outputOPT'
}
# Initialize and adjust fairness scores
myoptp_fair = OptimalPolicy(**params)
training_fair_df, fairscore_names, _, _ = myoptp_fair.fairscores(training_df.copy(), data_title='training')
# Update params to use the fair scores
params['var_polscore_name'] = tuple(fairscore_names)
# Solve for optimal allocation rule
alloc_train_df, _, _ = myoptp_fair.solve(training_fair_df.copy(), data_title='training fair')
# Evaluate the allocation
results_eva_train, _ = myoptp_fair.evaluate(alloc_train_df, training_fair_df.copy(), data_title='training fair')