Study 1
Weisburd and colleagues (2015) used a block-randomized experimental design in their evaluation of the Automated Vehicle Locator (AVL) technology. The experiment took place in Dallas, Texas, which is the third-largest city in the state and ninth largest in the country.
Study participants included members of the Dallas Police Department (DPD). The city is divided into seven patrol divisions across a 385-square-mile area and has a population of 1.2 million, the majority of which (51 percent) are white. A total of 232 beats were used in the study: 21 very low crime beats, 94 low crime beats, 100 medium crime beats, and 17 high crime beats. A beat ranges in size from 0.13 to 8.84 square miles, but on average is about 1.40 square miles. A total of 873 vehicles were equipped with AVL technology.
Trajectory analysis was used to identify beats with similar crime trajectories. Each beat within a trajectory was randomly assigned to a treatment or control group. There were 116 beats in the treatment group and 116 beats in the control group. The treatment included weekly dissemination of AVL-collected deployment information, at both the beat level and the hot spot level, to DPD division commanders. AVL deployment information was not collected for the control condition beats and hotspots, and police patrolled these areas as usual. The experiment lasted a total of 13 weeks.
Data were collected at both the beat level and hot spot level. Beat-level data were collected using a web-based, intranet application only accessible internally by DPD. Personnel entered the intended deployment information based on available resources. This information was compared with actual deployment dosage measured via AVL data collected from police vehicles. Hot spot-level data were collected via weekly Compstat meetings hosted by division commanders. A form was developed and used to capture high-crime locations that would be focused on that week. Actual patrol dosage in hot spots was measured via AVL data collected from the vehicles. AVL data were measured for each cell within a grid consisting of quarter-mile cells that spanned the whole city. The block randomized design of the study helped to account for the large variation in crime rates across beats.
Two categories of outcomes were measured; the first dealt with the impact of AVL information on police patrols and the second with the impact on crime via the altered patrol assignments. The police patrol outcomes were measured by investigating if AVL information led to an increase in patrol expected by commanders, and if those expectations were then actualized by police. Patrol dosage was measured based on the number of hours of patrol. The crime measure included homicide, rape, robbery, aggravated assault, burglary, theft, unauthorized use of a motor vehicle, auto theft, burglary of a motor vehicle, narcotics/drugs, vandalism/criminal mischief, and assault. No subgroup analyses were conducted.