COMPAS Risk Scales : Demonstrating Accuracy Equity and Predictive Parity Performance of the COMPAS Risk Scales in Broward County

2016

Subfields: AI Ethics, Fairness & Society

Citations in the corpus

• Model Cards for Model Reporting (2018, background): “Researchers have discovered systematic biases in commercial machine learning models used for face detection and tracking [4, 9, 49], attribute detection [5], criminal justice [10], toxic comment detection [11], and other applications.”
• Inherent Trade-Offs in the Fair Determination of Risk Scores (2016, background): “Much of the applied work on risk scores, as noted above, focuses on calibration as a central goal [8, 10, 13].”
• The accuracy, fairness, and limits of predicting recidivism (2018, methodology): “In response to this analysis, Northpointe argued that the ProPublica analysis overlooked other more standard measures of fairness that the COMPAS score satisfies ( 3 ) [see also the studies of Flores et al .”
• Prediction-Based Decisions and Fairness: A Catalogue of Choices, Assumptions, and Definitions (2018, background): “These techniques are already common in lending [52, 84, 42], hiring [88, 89, 55], and online advertising [119], and increasingly figure into decisions regarding pretrial detention [3, 80, 27], immigration detention [77], child maltreatment screening [122, 34, 13], public health [95, 105], and welfar”
• Interpretations are useful: penalizing explanations to align neural networks with prior knowledge (2019, background): “We aim to keep this rate low and relatively even across races, similar to the common “equalized odds” notion of fairness (Dieterich et al., 2016); note that a full investigation of fairness and its most appropriate definition is beyond the scope of the work here.”
• Achieving Fairness through Adversarial Learning: an Application to Recidivism Prediction (2018, methodology): “Northpointe responded to the ProPublica analysis by arguing that COMPAS satisfies the fairness metric of calibration (Dieterich et al., 2016).”
• Distill-and-Compare: Auditing Black-Box Models Using Transparent Model Distillation (2017, background): “COMPAS, a proprietary score developed to predict recidivism risk, has been the subject of scrutiny for racial bias [5, 8, 11, 13, 16, 25].”
• Fair Prediction with Disparate Impact: A Study of Bias in Recidivism Prediction Instruments (2016, neutral): “Northpointe in its response 7 argues for a still different approach that checks for a fairness criterion termed predictive parity , which it demonstrates COMPAS also satisfies.”
• Wasserstein Fair Classification (2019, background): “…and society, such as health care, criminal risk assessment, social services, hiring, financial lending, and online advertising (De Fauw et al., 2018; Dieterich et al., 2016; Eubanks, 2018; Hoffman et al., 2018; Malekipir-bazari and Aksakalli, 2015; Perlich et al., 2014), is raising concern that…”
• Fair Regression: Quantitative Definitions and Reduction-based Algorithms (2019, background): “Unlike overall accuracy equality in classification (Dieterich et al., 2016), which requires the losses on all groups to be equal, BGL does not force an artificial decrease in performance on every group just to match the hardest-to-predict group.”
• Fairness and Abstraction in Sociotechnical Systems (2019, neutral): “at ProPublica who demonstrated that a certain measure of fairness (equality of error rates) was violated [7], against Northpointe, the creators of COMPAS, who argued that it was fair because the test accuracy was equalized across groups [19].”
• 50 Years of Test (Un)fairness: Lessons for Machine Learning (2018, background): “The rise of interest in fairness today has corresponded with public interest in the use of machine learning in criminal sentencing and predictive policing, including discussions around compas [16, 20, 42] and PredPol [25, 49].”
• How Do Fairness Definitions Fare?: Examining Public Attitudes Towards Algorithmic Definitions of Fairness (2018, background): “Equivant (formerly called Northpointe), the company that developed the COMPAS tool, on the other hand, focused on positive predictive value, which is similar for whites and blacks (Dieterich et al., 2016).”
• A Causal Bayesian Networks Viewpoint on Fairness (2018, neutral): “This evidence led ProPublica to conclude that COMPAS had a disparate impact on black defendants, leading to public outcry over potential biases in RAIs and machine learning writ large.”
• Path-Specific Counterfactual Fairness (2018, background): “Machine learning is increasingly being used to take decisions that can severely affect people’s lives, e.g. in policing, education, hiring (Hoffman et al., 2015), lending, and criminal risk assessment (Dieterich et al., 2016).”
• Fairness in Machine Learning: Lessons from Political Philosophy (2017, background): “…accuracy equity’, which considers the accuracy of a predictive model for each group, con-c (cid:13) 2018 R. Binns. ditional on their predicted class Dieterich et al. (2016); ‘equality of opportunity’, which considers whether each group is equally likely to be predicted a desirable outcome…”
• On Fairness and Calibration (2017, methodology): “It is commonly accepted amongst practitioners that both classifiers h1 and h2 should be calibrated with respect to groups G1 and G2 to prevent discrimination [4, 10, 12, 16].”
• Fairness in Criminal Justice Risk Assessments: The State of the Art (2017, background): “Conditional use accuracy equality has also been a common concern in criminal justice risk assessments (Dieterich et al., 2016).”
• The age of secrecy and unfairness in recidivism prediction (2018, neutral): “(20) based on their definition of fairness.”
• An algorithm for removing sensitive information: Application to race-independent recidivism prediction (2017, methodology): “For example, Dieterich et al. [2016] argue that fairness is defined by similar accuracy and positive predictive value by class.”