Fix Representation (Optimally) Before Fairness: Finite-Sample Shrinkage Population Correction and the True Price of Fairness Under Subpopulation Shift
This work addresses the issue of spurious fairness-accuracy tradeoffs for machine learning practitioners, providing an actionable evaluation protocol, though it is incremental in refining existing fairness methods.
The paper tackles the problem of misrepresenting subgroup proportions in training data under subpopulation shift, showing that optimal finite-sample correction involves shrinkage reweighting and that apparent fairness-accuracy tradeoffs can be artifacts of improper baselines. Experiments on benchmarks like Adult and COMPAS validate the protocol, revealing the true fairness-utility frontier.
Machine learning practitioners frequently observe tension between predictive accuracy and group fairness constraints -- yet sometimes fairness interventions appear to improve accuracy. We show that both phenomena can be artifacts of training data that misrepresents subgroup proportions. Under subpopulation shift (stable within-group distributions, shifted group proportions), we establish: (i) full importance-weighted correction is asymptotically unbiased but finite-sample suboptimal; (ii) the optimal finite-sample correction is a shrinkage reweighting that interpolates between target and training mixtures; (iii) apparent "fairness helps accuracy" can arise from comparing fairness methods to an improperly-weighted baseline. We provide an actionable evaluation protocol: fix representation (optimally) before fairness -- compare fairness interventions against a shrinkage-corrected baseline to isolate the true, irreducible price of fairness. Experiments on synthetic and real-world benchmarks (Adult, COMPAS) validate our theoretical predictions and demonstrate that this protocol eliminates spurious tradeoffs, revealing the genuine fairness-utility frontier.