# Fair and Calibrated Toxicity Detection with Robust Training and Abstention

arXiv: https://arxiv.org/abs/2605.14074 · HTML: https://mlsanity.com/paper/2605.14074/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 9 of its 11 references that are in the map.
- First posted on arXiv 13 May 2026.

## Builds on (most cited first, 9 of 9)

- [On Calibration of Modern Neural Networks](https://mlsanity.com/paper/1706.04599/) (2017)
- [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://mlsanity.com/paper/1910.01108/) (2019)
- [Distributionally Robust Neural Networks for Group Shifts: On the Importance of Regularization for Worst-Case Generalization](https://mlsanity.com/paper/1911.08731/) (2019)
- [Selective Classification for Deep Neural Networks](https://mlsanity.com/paper/1705.08500/) (2017)
- [On Fairness and Calibration](https://mlsanity.com/paper/1709.02012/) (2017)
- [Nuanced Metrics for Measuring Unintended Bias with Real Data for Text Classification](https://mlsanity.com/paper/1903.04561/) (2019)
- [An Investigation of Why Overparameterization Exacerbates Spurious Correlations](https://mlsanity.com/paper/2005.04345/) (2020)
- [Simple data balancing achieves competitive worst-group-accuracy](https://mlsanity.com/paper/2110.14503/) (2021)
- [Counterfactual Fairness in Text Classification through Robustness](https://mlsanity.com/paper/1809.10610/) (2018)

## Abstract (arXiv)

Fairness in toxicity classification involves three integrated axes: ranking, calibration, and abstention. Training-time interventions and post-hoc safety mechanisms cannot be evaluated independently because the former determines the efficacy of the latter. We compare Empirical Risk Minimization (ERM), instance-level reweighting, and Group DRO across these axes, combined with temperature scaling, confidence-based abstention, and per-identity threshold optimization. Evaluation uses subgroup AUC, BPSN/BNSP AUC, error gaps, and per-subgroup Expected Calibration Error (ECE) with bootstrap CIs (n = 1000). We report four findings. (1) Calibration disparity is a hidden fairness violation. ERM has near-perfect aggregate calibration (0.013) but is significantly miscalibrated across all identity subgroups (+0.029 to +0.134). (2) Training interventions reshape rather than eliminate disparity. Reweighted ERM improves ranking (BPSN AUC +0.06 to +0.12) but worsens the calibration-fairness gap by up to +0.232. Group DRO eliminates calibration disparity but only by becoming uniformly miscalibrated globally (ECE 0.118). (3) Post-hoc methods inherit training failure modes. Temperature scaling fails because miscalibration is non-uniform. Confidence-based abstention works under ERM but breaks under DRO, where the risk-coverage curve rises with deferral. (4) Abstention itself is unfair. Confidence-based deferral helps background content far more than identity-mentioning content. We argue that SRAI fairness requires a multi-axis framework: methods that differ only in aggregate ranking can differ sharply in failure modes that determine real-world harm.