Sherman and Weisburd (1995) conducted a randomized field trial to evaluate the impact on crime of increased police presence in specified "hot spots" in Minneapolis. First, they identified the hot spots of crime by analyzing calls made to police by citizens from June 6, 1987, to June 5, 1988, and compiled clusters of addresses that produced the highest volume of calls. After careful analysis, 110 sites were identified. The sites were then randomly assigned to either the control group or the experimental group, each consisting of 55 sites.
The experiment lasted 1 year, from Dec. 1, 1988, to Nov. 30, 1989. During this period, police patrol was intensified in the treatment hot spots (though presence was intermittent rather than constant). For 7 days a week, police officers patrolled the hot spots for 3 hours at a time during 2 dosage periods: 11 a.m. to 7 p.m., and 7 p.m. to 3 a.m. These were the times identified as “hottest,” as determined by the number of calls made to police, thus the experiment was limited to the period between 11 a.m. and 3 a.m.
To evaluate the impact of the police presence on crime, the researchers first analyzed phone calls made to police by citizens (as distinguished from calls made from police to dispatchers). Using information obtained from police logs, the researchers compared phone calls in the year preceding the experiment (Dec. 1, 1987, to Nov. 30, 1988) to calls made during the 12 months of the experiment (Dec. 1, 1988, to Nov. 30, 1989). Sherman and Weisburd distinguished between calls from the public regarding “hard crime” and “soft crime,” with calls about “hard crime” including more serious crimes.
They also employed 16 observers and 3 supervisors to monitor crime and disorder in the 50 most-active hot spots in both the experimental and the control group, coming to a total of 100 observed sites. All observations were made between 7 p.m. and 3 a.m. The two outcomes measured in the study were number of citizen calls made to police and observed disorder.
To ensure compatibility with the research protocol, unit-minute ratios were computed; these are a measure of the number of minutes police spent in a location to all observed minutes of the group. The mean unit-minutes across hot spots within experimental groups were generally consistent, but the combined ratio between treatment groups and control groups varied greatly over calendar months. Data from May to August varied to a great extent because of peaks in call loads for these months and vacation time for officers. Because of these factors, the differences in outcomes between experimental and control groups for these months were substantial enough to disrupt the experiment. This left only 6½ months of data of the fully implemented experiment. In addition, a new computer-aided dispatch system was implemented from Oct. 1 through Nov. 10, 1989, which caused errors in reporting data. Thus, the researchers proposed four time periods for proper analysis of the data: 1) Dec. 1, 1988, through June 15, 1989; 2) Dec. 1, 1988, through July 31, 1989; 3) Dec. 1, 1988, through Nov. 30, 1989; and 4) Dec. 1, 1988 through Nov. 30, 1989, excluding Oct. 1 through Nov. 10. They conducted analyses for all 4 periods but decided the period with the July 31 cutoff date was most suitable to test the hypothesis.