This study sought to quantify the size and geographic extent of the impact CCTV cameras have on crime rates.
The study was conducted in Philadelphia, Pennsylvania and uses as a foundation the 115 cameras installed as of the summer of 2011. The investigators constructed viewsheds to represent the areas visible by the cameras.
The viewsheds were drawn by considering what could be seen from the camera and from where on the street that the camera could be seen. Members of the research team went to the Video Monitoring Unit (later the Real Time Crime Center) at the Philadelphia Police Department and viewed camera feeds in all directions, and they noted the farthest point they could clearly read street signs (or equivalent). To obtain information about the visibility of the camera, a research team member was sent to each camera site and by looking from the street toward the camera location determined from what points on the street the camera could be seen. Using Philadelphia streets and parcel lines, the locations gathered from field work were digitized on a map.
Much of the quantitative data was obtained from the Philadelphia Police Department's (PPD) INCT (incident) database. The INCT database used in this study records incidents that have been confirmed by an officer as founded. It is therefore a list of confirmed crime events. Because it is completed by the reporting officer, the location in the INCT is where the incident actually occurred rather than the location of the original 911 call. Researchers also collected data from the Philadelphia Police Department's arrest database, the Preliminary Arraignment Reporting System (PARS). This database contains a record for each arrest made by police officers in the city. Records in the arrest database can be linked to INCT using a unique identifier for each INCT record, known as a DC number. At the times when multiple persons are arrested for the same incident, the PARS database will record the details of all offenders arrested.
The investigators examined detectives' perceptions of the effectiveness of CCTV for investigations through interviews with just under 100 investigators and detectives who requested CCTV camera footage from the Philadelphia Police Department video monitoring unit. Field interviews were conducted with 100 randomly selected members of the public half a block from a CCTV camera, with 25 people interviewed at each of four sites. Interviewees were asked a variety of questions to elicit their understanding of how far cameras can see, who operates and views the cameras and camera footage, and whether the presence of camera influences their feelings of public safety.
For the interrupted time series analyses, crime (and arrest) data were drawn from the viewsheds of a subset of the cameras. Fieldwork undertaken during the summer of 2010 visited all of the installed and functioning camera sites and recorded basic descriptive data of the camera environment. The investigators recorded details of the local environment out to approximately one half-block from the camera location. At the time of visiting the sites (August 2010), there were 194 cameras positioned within the city.Across the city some cameras were installed in isolation, while in other parts of the city cameras were installed within the viewshed of another camera, or they were installed such that their viewshed overlapped another camera or they were installed within a couple of blocks of another camera's viewshed. As a result, the city CCTV camera system was comprised of either individual cameras or clusters of cameras. Researchers defined clusters based on whether camera viewsheds overlapped one another, or a camera was positioned within 1,200 feet of another camera's viewshed (a distance of about three city blocks). Of the 115 cameras or so that form the overall core of this research project, 86 camera viewsheds were included as part of 13 clusters.
The propensity score matching was conducted using 108 unique viewsheds. The varying shapes and sizes of the actual viewsheds made selecting matched control areas analytically intractable. Researchers used a conservative measure of half a block in all directions to represent the camera viewsheds. A set of polygons was created for the entire city of Philadelphia. Each corner was represented by a polygon that extended half the distance to the next intersection. Three different types of data were used to describe corner geographies in this analysis: 1) facility types that could act as crime generators and attractors; 2) demographic and economic characteristics and 3) crime incident data. Corners were matched on demographic, economic and opportunity factors (i.e., facilities).
Interviews were also conducted with just under 100 investigators and detectives who requested CCTV camera footage from the Philadelphia Police Department video monitoring unit and 100 randomly selected members of the public half a block from a CCTV camera, with 25 people interviewed at each of four sites.
A spatial polygon file of displacement areas was used to create an estimated displacement area around the 13 clusters. In this way, there was a customized displacement space that was tailored to the spatial characteristics of the cluster areas. The control area used for these analyses consisted of the remainder of the city that was at least one mile from the viewshed of any CCTV camera in the city. Control variables and time-varying covariates were used to model crime counts over time, separately, for each of the five crime categories. Those same models were then applied to the crime counts located within CCTV viewsheds if those camera areas were in the 13 clusters.
Time Series: Discrete
CCTV camera viewshed areas and members of the public around four CCTV camera sites in Philadelphia, Pennsylvania.
Philadelphia Police Department's (PPD) INCT (incident) database
Video Monitoring Unit at the Philadelphia Police Department
Philadelphia Police Department's Preliminary Arraignment Reporting System (PARS)
administrative records data
geographic information system (GIS) data
coded on-site observation
This study contains four Stata data sets and two GIS shapefiles.
CCTV_public_perception_survey.dta: This file includes 100 cases and 16 variables regarding public perception of CCTV cameras, such as: if respondent thought there were any around, if cameras could see them, how far down the street the camera could see, whether the cameras were recording and/or being monitored and if so by whom, concern about being filmed, how safe respondent feels in the area, and familiarity with and time spent in the area. Demographic information includes age, race/ethnicity, and gender.
Multilevel_modeling_data.dta: This file includes 1560 cases and 18 variables regarding monthly frequencies of various crime types (burglary, disorder, narcotics, vehicle, and violent street felonies) in 13 CCTV camera clusters. The data also include time-varying covariates (centered linear, centered quadratic, and centered cubic), month and year information, number of operational cameras in the cluster, and area of the cluster.
Propensity_score_matching_data.dta: This file contains 20987 cases and 28 variables, and links to the Thiessens4PSM shapefile via the variable Input_FID and contains attribute data of the Thiessen polygons. Variables include measures for shape area, population density, household information (such as linguistically isolated, below poverty level, female headed, and owner or renter occupied), and number of employed males and females. The file also includes camera information, such as location, ID, installation date, and date camera was live. Also included is propensity score matching data for various crimes (burglary, disorder, narcotics, vehicle, and violent), as well as total numbers of bars, schools, stations and check cashing.
Time_series_analysis_data.dta: This file includes 120 cases and 33 variables regarding monthly frequencies of and number of arrests for various crime types (burglary, disorder, narcotics, vehicle, and violent street felonies) in the target, control, and displacement areas within selected CCTV camera viewshed areas. The data also include time-varying covariates (centered linear, centered quadratic, and centered cubic), mean monthly temperature, month, and year.
Thiessens4PSM.shp: This file contains a set of Thiessen polygons that are based on the locations of the street intersections in Philadelphia. Each polygon represents the area closer to it than any other intersection in the City of Philadelphia.
Control_areas_more_than_one_mile_from_any_viewshed.shp: This file contains areas of the City of Philadelphia at least one mile from any public CCTV camera viewshed based on the viewsheds estimated during the summer of 2011.
The response rate was 100 percent for the public perception survey.
Some Likert-type scales were used in the public perception survey.