Program Profile: License Plate Recognition Technology (as a Crime Deterrent)

This program consists of the use of a vehicle-scanning device deployed by law enforcement to detect vehicles that have been reported stolen or missing. The program is rated Ineffective. There were no statistically significant program effects on general crime, auto-related crime, vehicle-theft calls for service, or auto-theft.

An Ineffective rating implies that implementing the program is unlikely to result in the intended outcome(s).

This program's rating is based on evidence that includes at least one high-quality randomized controlled trial.

This program's rating is based on evidence that includes either 1) one study conducted in multiple sites; or 2) two or three studies, each conducted at a different site. Learn about how we make the multisite determination.

Program Goals

License plate recognition (LPR) technology is a vehicle-scanning device deployed by law enforcement to detect wanted vehicles (such as those reported stolen or missing). Originating in the United Kingdom in the 1980s, LPR has been used predominantly to combat auto theft, although it has also been used to detect wanted vehicles associated with other types of crime, such as known vehicles of at-large suspects. Because LPR can collect and store the dates, times, locations, and plate names of cars in its range, it can also be used to locate individuals or suspects based on previously collected data as well as confirm a suspect’s alibi or whereabouts at a particular place and time.

Prior to the advent of LPR,  police officers had to go through the lengthy process of determining whether a vehicle was suspicious based on their discretion, calling the dispatcher, and then waiting to see if the license plate matched one of the wanted vehicles in the database. LPR automates this manual approach and also reduces the use of officers’ discretion in deciding which tags to run since all tags within view are scanned. Some LPR systems can scan up to four lanes of traffic and read 8,000-10,000 plates in just one shift.

Program Components

LPR consists of infrared cameras that are normally mounted on police cruisers (although they can also be attached to a fixed location, such as a toll plaza). As civilian cars pass, the cameras capture the unique reflective material of the license plates in a photo. The photos are then run through character-recognition software to determine the exact numbers and letters of the plates. Nearly simultaneously, the plates are checked against a “hot list” of vehicles in the state; a signal to law enforcement is triggered when a car is of interest. The officer then verifies the accuracy of the hit by looking at the numbers and letters on the plate before taking any action.

Study

Lum and colleagues (2011) used a place-based block randomized design to evaluate the deterrent effect of license plate recognition (LPR) technology on general crime—and specifically, auto theft—in Northern Virginia. Two officers from both the Fairfax County Police Department (FCPD) and the Alexandria Police Department (APD) participated in the study for 30 consecutive days when their schedules permitted. Due to limited departmental resources, not all officers were available at the same time, so the experiment began on February 22, 2010, and did not conclude until June 1, 2010. The officers were instructed to use the “sweep and sit” method of patrolling (drive around the area and then park), and additionally were told to patrol in 30-minute increments. The brief duration of these increments derived from the so-called “Koper Curve,” which suggests that officers should not stay in hot spots for long periods of time but rather move from hot spot to hot spot, staying for short periods of time (ideally no longer than 12-15 minutes) (Koper 1995).

Based on strong previous research indicating that “hot spots” are more effective on the micro level (Lum et al. 2011; Weisburd and Eck 2004), 30 small auto theft locations were deemed eligible for study in eastern Fairfax County and all of Alexandria, except for its eastern portion. In addition to using APD and FCPD supervisors’ input, hot spots were determined statistically through kernel density analysis, followed by confirmation of each location through nine CrimeStat spatial and temporal analysis of crime (STAC) simulations.

Of the 30 hot spots, 15 were randomly assigned to receive the LPR intervention while the other 15 spots received “business as usual policing” (meaning no change in the existing police activities in that area). The hot spots were block-randomized so that the random selection of seven hot spots from the APD jurisdiction and eight hot spots from FCPD’s jurisdiction were chosen to receive the LPR treatment.

In addition to ongoing feedback from their respective supervisors during the experiment, the four officers in the treatment group received training on how to use the LPR system prior to the experiment. The experiment lasted 30 officer working days for each officer. Each day of consecutive work, supervisors gave their officers an envelope containing the five random hot spots they would be patrolling that day, and provided a log sheet to record their findings. The outcomes of the LPR technology intervention were compared with the “business as usual” policing post-intervention.

Outcomes included both the offense-specific and generalized deterrent effects of LPR deployment. Inferences about the offense-specific deterrent effect of LPR were based on auto-related offenses, including auto theft, theft from auto, and other auto-related crimes (such as driving under the influence and reckless driving). Inferences about the generalized deterrent effect were based on counts of reported crimes and disorders, including crimes against person and property (which include auto-related crimes), weapon-related crimes, disorder offenses, and drug activity. Data was collected during five time periods: (1) preintervention period, (2) intervention period, (3) postintervention period (30 days after the intervention stopped in each jurisdiction), (4) seasonal-lag of intervention period (crime counts in the same period of the intervention from the previous year), and (5) seasonal postintervention period (crime counts for the same period of the postintervention, but for the previous year). A negative binomial generalized linear model was used to estimate the effect of LPR on these outcome measures.

There were some potential limitations to the randomized design. Notably, the FCPD deployed a detective and a patrol officer from a marked auto-theft specialized unit, whereas the APD used two patrol officers from a marked patrol unit. Additionally, the FCPD officers overlapped the days they worked, so it is possible they patrolled the same days and times. Due to limited resources, the APD only deployed one officer at a time, but the officer nearly always worked during daylight. No subgroup analyses were conducted.

Study

Taylor, Koper, and Woods (2011) used a randomized controlled experiment in Mesa, Arizona, between 2008 and 2009 to study the effects of LPR devices on auto theft. The research team focused on “journey after crime” routes rather than “hot spots” for their analysis. With input from the Mesa Police Department (MPD) and geographic analysis of auto theft and auto recovery locations, 117 main transit routes in Mesa where criminals commonly drive stolen vehicles were identified. Each transit route, referred to as “hot routes,” was roughly 1 mile in length, located in both residential and business areas. Individual routes often wove through different types of roads, including interstates, highways, and residential streets.

The LPR intervention was implemented for 30 weeks, which were divided into 15 biweekly periods. The 117 transit routes were randomly assigned to receive one of three conditions. Whereas 45 hot routes were randomly assigned to the intervention group, and received an auto-theft unit to patrol the streets equipped with the LPR technology, another 45 routes were assigned to a comparison group that received an auto-theft unit to patrol the streets but were not equipped with the LPR technology (officers would do a manual check through mounted computer terminals in their cars). The remaining 27 routes were assigned to the other comparison group that received neither the specialized auto-theft unit nor the LPR, but just standard patrol. Both the treatment and comparison groups were instructed to patrol daily three LPR check routes and three manually checked routes for an hour each. In total, the officers had to patrol each route for 32 hours with its respective assigned intervention. Both the biweekly treatment period and time of day patrolled were randomly determined, so that the places and times worked with and without LPR were comparable. The routes assigned to be patrolled by the vehicle-theft unit (routes in the LPR intervention group and routes in the manual check comparison group) were randomly assigned to receive treatment during one of these biweekly periods (the officers worked 10-hour shifts 4 days a week, resulting in 8 days of treatment for each route). The unit worked three LPR routes and three manual check routes, each of which was patrolled daily for approximately an hour (each route received a minimum 8-hour intervention time by four officers, or 32 officer hours). The time of day which the unit patrolled each route varied according to a preset schedule so that the unit did not work the same routes at the same time of the day. Therefore, both the biweekly treatment period and the time of day patrolled were determined randomly for each route. The CrimeSolutions review of this study focused on the differences between the treatment group with the LPR technology and the comparison group that received an auto-theft unit but did not have the LPR technology and conducted manual checks.

Outcome measures included the number of arrests for auto theft (including stolen plates and theft of property from vehicles) and the number of recoveries of stolen vehicles. Auto-theft data was based on calls for service to the police (911 calls) for vehicle theft, and incidents of auto theft were based on the Uniform Crime Report. Multivariate models, including negative binomial (count model) regression, were used to examine the data. No subgroup analyses were conducted.

These sources were used in the development of the program profile:

These sources were used in the development of the program profile: