🤖 AI Summary
This work addresses the apparent trade-off between prediction accuracy and group fairness commonly observed in existing literature, which often stems from distorted subgroup proportions in training data and leads to misleading assessments of fairness interventions. The authors propose a two-stage paradigm under settings where subgroup distributions are stable but their prevalence is shifted: first, apply a finite-sample optimal shrinkage reweighting method to correct representation bias, then evaluate fairness interventions. This approach integrates importance weighting with shrinkage estimation to decouple spurious trade-offs from genuine ones. Experiments on real-world benchmarks such as Adult and COMPAS, as well as synthetic datasets, validate the theoretical claims and accurately recover the true Pareto frontier between fairness and utility.
📝 Abstract
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.