Study
Robin and colleagues (2020a) conducted a quasi-experimental design (using propensity score matching) to evaluate the impact of Milwaukee’s closed-circuit television (CCTV) camera program on improving surveillance and investigative capacity of the existing public surveillance system, during a 12-month intervention period (January through December 2018) in high-crime, high-traffic intersections.
During the study period, the study authors worked closely with the Milwaukee Police Department to acquire and install a total of 24 new panoramic cameras, 12 new pan, tilt zoom cameras, and 9 new automatic license-plate-recognition cameras to add to the existing surveillance system (which at the time had 42 cameras at 40 intersections across the city). Because some intersections were already equipped with older cameras (which were antiquated, had poor image quality, and offered limited operational support), the study authors conducted an in-depth analysis of violent and property crimes, gunshot detection alerts, stolen and recovered vehicle data, and information from specialized Milwaukee Police Department staff, to identify all high-crime, high-traffic intersections that would be best suited for the installation of new CCTV cameras. However, due to the limited number of new high-definition CCTV cameras, the police department expanded its surveillance network to include two specific geographic areas: 1) the Center Street Corridor (on the north side of the city) and 2) the Muskego Way neighborhood (on the south side). The addition of these two intersections expanded surveillance across 42 intersections, of which 32 intersections had a new CCTV camera installed (i.e., the intervention intersections, also referred to as group 1).
Intervention intersections consisted of two area types: 1) intersection areas receiving new CCTV cameras that did not have a CCTV camera previously installed (group 2; n = 18), and 2) intersection areas receiving new CCTV cameras that were installed alongside an existing CCTV camera (group 3; n = 14). Intersections that had existing cameras but did not receive a new CCTV camera were excluded from the analysis; thus, all 32 intervention intersections were analyzed as a single treatment group, compared with matched comparison intersections.
Matched comparison intersections were identified using propensity score matching, based on block group characteristics (such as crime and arrest trends, percentage of female-headed households, and percentage of people on public assistance). Crime and arrest trends were obtained from census and administrative data. Administrative data were provided by the Milwaukee Police Department and included geographic identifiers that allowed the study authors to pinpoint exact crime and arrest locations. Using ArcMap geographic information system software and a combination of shapefiles from census Tiger files, the study authors identified all intersections in the city, and then spatially linked intersections to a file with camera locations and information on the types of cameras. After removing intersections within 500 feet of intervention intersections, there were 8,245 intersections included in the pool of eligible comparison intersections for propensity score matching. From that sample of intersections, there were 32 matched comparison intersections (which were also split into the two area subgroups based on intervention intersections).
Of the residents in intervention intersections, 56.4 percent were Black, and 23.6 percent were Hispanic, with 27.7 percent of people under the age of 18, 8.4 percent unemployed, and 33.6 percent living under the poverty line. In comparison intersections, 53.2 percent of residents were Black, and 23.1 percent were Hispanic, with 26.4 percent of people under 18, 8.7 percent unemployed, and 28.2 percent living under the poverty line. At baseline, there were statistically significant differences between intervention intersections compared with comparison intersections. Specifically, intervention intersections had a higher percentage of residents under the poverty line and on public assistance, and lower levels of residential mobility (i.e., people who had moved in the past year). Thus, a difference-in-differences analysis was used to control for group differences, using certain covariates (i.e., race, percent renting, and crime and clearance rates for all crime types such as violent crimes, drug crimes, and simple assault clearances).
Data were collected following the intervention for four quarters in 2018 (i.e., post-intervention). Outcome of interests in the study were the number of crimes and crime clearances compared with pre-intervention (i.e., four quarters in of trends for any crime type, such as violent crimes, property crimes, simple assault crimes, drug crimes or group B offenses [i.e., disorderly conduct, drunkenness, and loitering]). The CrimeSolutions review of this study focused on the number of crime clearances for all crime types, at post-intervention. Crime clearances (defined as an arrest linked to the location where a crime occurred, regardless of where the arrests occurred) were measured by an arrest for any crime type. Negative binomial and Poisson panel regression models were used to determine differences between crime clearances in intervention and comparison intersections. The study authors conducted subgroup analyses within the intervention and comparison intersections to determine whether CCTV camera assignment areas (intersection areas that received only new CCTV cameras versus intersection areas that received new CCTV cameras alongside existing cameras) affected crime and crime clearance outcomes.
