The accuracy, fairness, and limits of predicting recidivism
By Julia Dressel, Hany Farid. 2018
1,076 citations recorded by the source corpus.
Should we trust computers to make life-altering decisions in the criminal justice system? Algorithms for predicting recidivism are commonly used to assess a criminal defendant’s likelihood of committing a crime. These predictions are used in pretrial, parole, and sentencing decisions. Proponents of these systems argue that big data and advanced machine learning make these analyses more accurate and less biased than humans. We show, however, that the widely used commercial risk assessment software COMPAS is no more accurate or fair than predictions made by people with little or no criminal justice expertise. In addition, despite COMPAS’s collection of 137 features, the same accuracy can be achieved with a simple linear classifier with only two features.
Subfields: Clinical Psychology, Sociology and Political Science, Social Sciences & Humanities
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Clinical Psychology
- Yiling Chen
- Vivian Lai
- Ramaravind Kommiya Mothilal
- Nina Grgic-Hlaca
- Eduardo Almeida Soares
- Sam Corbett-Davies
- Sharad Goel
- Amit Sharma
In Sociology and Political Science
- Krishna P. Gummadi
- Adrian Weller
- Ben Green
- Sam Corbett-Davies
- Sharad Goel
- Nina Grgic-Hlaca
- Nitin Kohli
- Kristina Lerman
In Social Sciences & Humanities
- Sam Corbett-Davies
- Krishna P. Gummadi
- Nathan Kallus
- Kristina Lerman
- Sharad Goel
- Aram Galstyan
- Fred Morstatter
- Nripsuta Ani Saxena
Citations in the corpus
- Predictability and Surprise in Large Generative Models (2022, methodology): “13 Here, we use the same prompts outlined in [20] but instead ask language models [3] instead of people to predict recidivism.”
- Explainable Artificial Intelligence (XAI): Concepts, Taxonomies, Opportunities and Challenges toward Responsible AI (2019, positive): “The proposal then includes an algorithm that both generates explanations and reduces bias, as it is demonstrated for the use case of recidivism using the Correctional Offender Management Profiling for Alternative Sanctions (COMPAS) dataset [400] .”
- A Survey on Bias and Fairness in Machine Learning (2019, background): “n example is COMPAS, which is a widely used commercial risk assessment software that was compared to normal human judgment in a study and was later discovered to be not any better than a normal human [41]. It is also interesting to note that although COMPAS uses 137 features, only 7 of those were pr”
- Fair-by-design explainable models for prediction of recidivism (2019, methodology): “In a study conducted by ProPublica [3] that analyzed the efficacy of COMPAS on more than 7000 individuals arrested in Broward County, Florida between 2013 and 2014 it was found that the likelihood of a non-recidivating black defendant being assessed as high risk is nearly twice that of white…”
- Fairness Under Unawareness: Assessing Disparity When Protected Class Is Unobserved (2018, background): “e a lack of discrimination with respect to protected classes [2, 26]. Examples of such decisions include employment and compensation [14, 20], university admissions [4], and sentence and bail setting [3, 7, 16]. Another example relevant to the financial services industry is credit decisioning [6], w”
- Learning Certifiably Optimal Rule Lists (2017, neutral): “cy that is again comparable to COMPAS. CORELS identies the same rule list across all 10 folds of 10-fold cross-validation experiments (Figure 13). In work subsequent to ours (Angelino et al., 2017), Dressel and Farid (2018) conrmed this result, in the sense that they used logistic regression to cons”
- Ethical principles in machine learning and artificial intelligence: cases from the field and possible ways forward (2020, methodology): “Dressel and Farid (2018) achieved this result by using a linear predictor-logistic regressor that made use of only two variables (age and total number of previous convic-tions of the subject).”
- Getting a CLUE: A Method for Explaining Uncertainty Estimates (2020, methodology): “Class predictions only depend on 2 of this dataset’s input features (Age and Previous Convictions) [34].”
- Neural Additive Models: Interpretable Machine Learning with Neural Nets (2020, background): “COMPAS is a proprietary score developed to predict recidivism risk, which is used to inform bail, sentencing and parole decisions and has been the subject of scrutiny for racial bias [1, 8, 42].”
- Should We Trust Algorithms? (2020, methodology): “…public were as good as COMPAS (65% accuracy), and that COMPAS performance could be matched by a simple rule-based classifier (Angelino, Larus-Stone, Alabi, Seltzer, & Rudin, 2017), and even a regression model with only two predictors (age and total previous convictions) (Dressel & Farid, 2018).”
- Explaining machine learning classifiers through diverse counterfactual explanations (2019, methodology): “taset into 80%-20% train and test sets, and use TensorFlow library to train a neural network model. For COMPAS, we obtain an accuracy of 0.67, comparable to the best accuracy reported on this dataset [3, 12]. Similarly, for Adult and LendingClub datasets, we obtain accuracies of 0.83 and 0.66 respec”
- WILDS: A Benchmark of in-the-Wild Distribution Shifts (2020, neutral): “On the COMPAS dataset, logistic regression on a small number of features performs comparably to a black-box commercial algorithm (Jung et al., 2020; Dressel and Farid, 2018).”
- A Framework for Understanding Sources of Harm throughout the Machine Learning Life Cycle (2019, background): “, “rearrest” to measure “recidivism” [15]) were also differentially measured proxies.”
- A Framework for Understanding Unintended Consequences of Machine Learning (2019, background): “It is worth noting that such an evaluation is further complicated by the proxy label “rearrest” used to measure “recidivism” (Dressel and Farid, 2018).”
- Current Challenges and Future Opportunities for XAI in Machine Learning-Based Clinical Decision Support Systems: A Systematic Review (2021, methodology): “For example, the AI-based system COMPAS (Correctional Offender Management Profiling for Alternative Sanctions) has been widely criticized for being unreliable and racially biased [52].”