Predictive Policing and Feedback Loops

Predictive-policing systems forecast places or people for police attention. Place-based tools use recorded incident data to highlight map areas for patrol; person-based tools rank people by estimated risk of involvement in an event. A forecast is not a finding that a crime occurred or that a person is dangerous. It shapes an institutional response, and that response becomes part of the environment being measured. Lum and Isaac’s 2016 “To Predict and Serve?” analysis applied a PredPol-style approach to Oakland drug-arrest data and compared the resulting patrol predictions with public-health survey estimates of drug use. Because arrests reflect enforcement as well as offending, using arrest records as a proxy for crime can direct patrol toward places that already received more police attention. New stops and arrests can then add observations to the same data stream, producing a feedback loop. The study illustrates a mechanism and a specific case, not proof that every predictive system has the same effect. Chicago’s Strategic Subject List (SSL) is a case. The City of Chicago’s historical data page says the program ended in 2019 and describes a score estimating the probability an individual would be involved in a shooting as victim or offender. A 2016 NIJ-sponsored quasi-experimental evaluation found people on the list were not more or less likely to become victims of homicide or shooting than a comparison group; city-level analysis supported that result. The listed group was more likely to be arrested for a shooting, and the authors discussed whether officers used scores as investigative leads. The study concerns one pilot, not every predictive-policing program. Evaluation should separate forecasts from operational policies, record where patrols were sent, compare with alternatives, and measure outcomes that do not simply restate police activity. Community impact, false positives, feedback, and due-process protections matter. A model can prioritize attention, but it cannot turn recorded data into an objective measure of true crime without examining how those records were generated.