Fairer and more accurate, but for whom?

By Alexandra Chouldechova, M. G'Sell. 2017

66 citations recorded by the source corpus.

Complex statistical machine learning models are increasingly being used or considered for use in high-stakes decision-making pipelines in domains such as financial services, health care, criminal justice and human services. These models are often investigated as possible improvements over more classical tools such as regression models or human judgement. While the modeling approach may be new, the practice of using some form of risk assessment to inform decisions is not. When determining whether a new model should be adopted, it is therefore essential to be able to compare the proposed model to the existing approach across a range of task-relevant accuracy and fairness metrics. Looking at overall performance metrics, however, may be misleading. Even when two models have comparable overall performance, they may nevertheless disagree in their classifications on a considerable fraction of cases. In this paper we introduce a model comparison framework for automatically identifying subgroups in which the differences between models are most pronounced. Our primary focus is on identifying subgroups where the models differ in terms of fairness-related quantities such as racial or gender disparities. We present experimental results from a recidivism prediction task and a hypothetical lending example.

Subfields: Evaluation & Benchmarks, AI Ethics, Fairness & Society

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Evaluation & Benchmarks

In AI Ethics, Fairness & Society

Citations in the corpus

Prediction-Based Decisions and Fairness: A Catalogue of Choices, Assumptions, and Definitions (2018, background): “For example, it has been shown that different functional forms can differ in their estimates for individuals and disparities between groups of individuals [14].”
Achieving Fairness through Adversarial Learning: an Application to Recidivism Prediction (2018, background): “Other researchers investigate COMPAS in relation to priors since black inmates have more priors (Chouldechova & G’Sell, 2017).”
Characterizing Fairness Over the Set of Good Models Under Selective Labels (2021, background): “While many models may perform similarly overall, they may perform quite differently over various subgroups in the population, and therefore have drastically different predictive fairness properties [11].”
Calibration for the (Computationally-Identifiable) Masses (2017, positive): “This result adds to the growing literature on fairness-accuracy trade-oﬀs [FKL16,BHJ + 17,CG17].”
Multiaccuracy: Black-Box Post-Processing for Fairness in Classification (2018, background): “2017; Kim, Reingold, and Rothblum 2018; Rothblum and Yona 2018), discussions of the utility-fairness tradeoffs in fair classification (Angwin et al. 2016; Kleinberg, Mullainathan, and Raghavan 2017; Chouldechova 2017; Chouldechova and G’Sell 2016; Corbett-Davies et al. 2017; Pleiss et al. 2017).”
What’s in a Name? Reducing Bias in Bios without Access to Protected Attributes (2019, background): “(Chouldechova and G’Sell, 2017).”
Multicalibration: Calibration for the (Computationally-Identifiable) Masses (2018, result): “This result adds to the growing literature on fairness-accuracy trade-offs (Fish et al., 2016; Berk et al., 2017; Chouldechova & G’Sell, 2017).”
What is AI Literacy? Competencies and Design Considerations (2020, background): “Lack of workforce diversity can affect who systems are developed for [33]—a significant issue in AI, where biased algorithms can have pronounced adverse effects on marginalized subgroups [32].”
Repairing without Retraining: Avoiding Disparate Impact with Counterfactual Distributions (2019, background): “This may provide an interesting alternative to identify subgroups in which discrimination is most pronounced (see e.g, Chouldechova and G’Sell, 2017; Zhang and Neill, 2016).”
Reasons, Values, Stakeholders: A Philosophical Framework for Explainable Artificial Intelligence (2021, background): “1 More concretely, the development of explainable AI systems is motivated by several social and ethical reasons (see, for example, [10, 11, 12, 13, 14, 15, 16, 17, 18, 19]): increasing societal acceptance of prediction-based decisions, establishing trust in the results of these decisions, making…”
A case study of algorithm-assisted decision making in child maltreatment hotline screening decisions (2018, background): “This issue was recently explored in Chouldechova and G’Sell (2017), who showed that model disagreement may be highly pronounced on particular salient subgroups, and hence the decision to adopt one model rather than another may come down to what those subgroups look like and how they are affected.”
Fairness in Machine Learning: A Survey (2020, methodology): “…(e.g., References [9, 43, 56, 90, 148, 263, 283, 307]), i.e., “correct” training data and eliminate biases [148]; and (2) to identify groups (or subsamples) of the data that are significantly disadvantaged by a classifier, i.e., as a means to evaluate a model (e.g., References [2, 70, 315]).”
Algorithmic Fairness: Choices, Assumptions, and Definitions (2021, background): “For example, it has been shown that different functional forms can differ in their estimates for individuals and disparities between groups of individuals (Chouldechova & G’Sell 2017).”
Auditing Black-Box Models Using Transparent Model Distillation With Side Information (2017, methodology): “They were used explicitly in [21] and more recently in [11].”