Micro Places – Ratcliffe et al. (2021)

Study Reference:

Ratcliffe, J., Taylor, R., Askey, A., Thomas, K., Grasso, J., Bethel, K., Fisher, R., & Koehnlein, J. (2021). The Philadelphia predictive policing experiment. Journal of Experimental Criminology.

Location in the Matrix; Methodological Rigor; Outcome:

Micro-places; Focused; Proactive; Rigorous; No evidence of an effect

What police practice or strategy was examined?

This study examines the impact of different patrol strategies on property and violent crime using predictive policing software in Philadelphia, PA. The software – HunchLab – identified micro-level grids where crime was likely to occur on a particular day. Police were then given varying levels of information or assignments, including being aware of the hot spots in their district, being deployed in marked police vehicles, and being deployed in plainclothes and unmarked police vehicles.

How was the intervention evaluated?

A randomized block design assigned 20 Philadelphia police districts into one of the four experimental conditions resulting in five districts in each condition. Blocks were based on socioeconomic status, percentage of white non-Hispanic, and crime scores. Assigned conditions included: (1) an “awareness” group where officers received information about where the hot spots were located in their district and were given no further direction, (2) a “marked” patrol group with dedicated marked vehicles sent to the hot spots, (3) an “unmarked patrol” group that spent their time on the hot spots in unmarked vehicles and wearing plainclothes, and (4) a control group that did not have access to the crime predicting software and conducted “business as usual.” Three hot spots (which included 500 ft x 500 ft mission grids plus the surrounding street blocks) per district were predicted each day, and these places could vary daily. The intervention was divided into two phases to target specific crime types: the first took place for 90 days between 8:00am and 4:00pm and focused on property crime outcomes, while the second phase resumed two months later and lasted for 92 days between 6:00pm and 2:00am. The second phase focused on violent crime outcomes. The impact of the interventions on property and violent crime were measured using the police department’s geocoded incident data.

What were the key findings?

When examining targeted microgrid cells and their surrounding grid cells during the treatment shifts, marked patrol cars resulted in a 31% decrease in expected property crime counts. Marked patrol cars in the targeted microgrid cells also produced a temporal diffusion of benefits to the subsequent 8-hour shift, resulting in a 40% decrease in expected crime counts. However, these declines were not statistically significant because the overall counts were low. No other strategy produced meaningful reductions in crime. Violent crime was an even rarer event; compared to the control condition, violent crime counts increased in all three experimental conditions, but these results, along with violent crime changes in the 8 hours post-shift, were not statistically significant.

What were the implications for law enforcement?

While statistically significant reductions in property or violent crime were not observed in this study, the results suggest that a dedicated and focused marked patrol car operating in a hot spot may reduce property and violent crime. Agencies should weigh the costs and benefits of such programs to decide if the limited crime control benefits are worthwhile.

Where can I find more information about this intervention, similar types of intervention, or related studies?