Hunt, P., Saunders, J., & Hollywood, J. S. (2014). Evaluation of the Shreveport Predictive Policing Experiment. Rand Corporation.
Location in the Matrix; Methodological Rigor; Outcome:
Neighborhoods, Focused, Proactive; Rigorous; No evidence of an effect
What police practice or strategy was examined?
The study examined the Predictive Intelligence Led Operational Targeting (PILOT) program, a predictive policing strategy using special operations resources focused on narrow locations more likely to incur property crimes. The PILOT consisted primarily of two types of activities: (1) identify on a monthly basis the small areas at increased risk of property crime using historical property crime data, juvenile arrests records, disorder calls, and seasonal variation pattern information, and (2) employ overtime police labor and conduct various policing interventions in the identified areas to prevent crime from increasing. Spots (grid cells with 400 feet on a side) with a predicted probability of over 40% of at least one property crime over the next month were identified as predicted hot spots and received directed patrol and other strategies by overtime police officers. Monthly deployment meetings were held to discuss and direct intelligence-gathering activities, and daily roll call (including daily maps highlighting recent crimes and daily field interview logs) was used for updating and disseminating information gathered from field.
How was the intervention evaluated?
Four high crime districts in Shreveport, Louisiana were randomized into treatment and control groups (two each), and two middle-level crime districts were also randomized into treatment and control groups. Thus, PILOT was implemented in three districts with three matched controls; no significant differences in pre-treatment crime levels for the experimental or control group were detected. Control sites also received overtime funding to conduct special operations on property crimes, but instead of using PILOT or predictive policing, hot spots were identified using conventional crime mapping techniques identifying clusters of property crimes. Changes in police-recorded property crime rates from the seven months before the program to the seven months during the program were compared between the experimental and the control districts to determine the treatment effect.
What were the key findings?
No statistical difference in property crime was found between the experimental districts and the control districts.
What were the implications for law enforcement?
The authors suggest that the program may have failed to show effects in part because the predictive model did not provide enough additional information over conventional analysis and because the preventive strategies tended to be vague. Further experimentation is required to understand whether identifying hot spots based on a predictive model is more effective than identifying them based on prior crime locations.
Where can I find more information about this intervention, similar types of intervention, or related studies?