Predictive policing algorithms reinforce existing racial disparities
Predictive policing systems trained on historical crime data reproduce and amplify existing racial disparities in enforcement: because the data records where police made arrests rather than where crime actually occurred, the algorithms direct more patrols to already over-policed minority neighborhoods, generating more recorded incidents there and feeding the disparity back into the next round of predictions.
The feedback-loop mechanism is formally demonstrated and the input-data problem is thoroughly documented: arrest and incident data measure enforcement activity, not underlying crime, and simulations show place-based predictive systems concentrate patrols in minority neighborhoods far beyond what victimization data would justify. Direct field measurement of deployed systems' disparate impact is thinner — departments rarely release the data needed — which is the main limitation keeping this from a stronger verdict.
This claim analysis is fresh and accurate as of 2026-07-07
Premise Assessment
Is the claim as stated true? Four dimensions, each 0–25, sum to 100. The verdict label is derived from this score. Full rubric →
Quality and quantity of direct evidence for or against the claim — RCTs, systematic reviews, natural experiments, large cohort studies.
Lum & Isaac's simulation applying PredPol's published algorithm to Oakland drug-crime data showed predicted hotspots concentrated in Black neighborhoods at roughly twice the rate of white neighborhoods despite survey data indicating drug use is roughly uniform across groups; Richardson et al. documented deployed systems ingesting data from police departments under federal investigation for unconstitutional practices.
Whether the proposed mechanism is valid and established — does the how make sense, or are there fundamental flaws in the causal logic?
The feedback mechanism is unusually well-specified: Ensign et al. formally proved that predictive systems retrained on discovery data generated by their own patrol allocations produce runaway concentration on initially over-sampled neighborhoods — a mathematical result about the algorithm class, not merely an empirical correlation.
Degree of agreement among domain experts and relevant scientific or policy bodies — depth and quality of consensus, not just majority opinion.
Algorithmic fairness researchers and a substantial share of criminologists agree the enforcement-data problem is real and serious; some criminologists counter that place-based prediction using victim-reported crimes (rather than arrests) partially mitigates the bias, so consensus is strong on the mechanism but weaker on how much deployed systems are affected in practice.
Whether findings hold across independent studies, populations, and contexts — resistance to p-hacking and publication bias.
The simulation and formal results have been reproduced and extended by independent teams, but direct field audits of deployed systems' racial impact remain rare because departments and vendors rarely release deployment data — several cities (Los Angeles, Chicago) discontinued programs after critical internal reviews rather than publishing rigorous impact evaluations.
Individual vs. Structural
How much of the outcome is explained by structural forces versus individual agency? Four dimensions, each 0–25. Higher scores indicate stronger structural causation. Full rubric →
Score component breakdown not yet available for this entry.