Offender-Focused Policing

Study 1

Summary

Groff et al. (2015) conducted a randomized control trial in which 81 violent hot spots in Philadelphia, Pennsylvania were randomly assigned to one of three treatments (foot patrol, problem-oriented policing, or offender-focused policing) or one control condition. Control received "business-as-usual" policing consisting of random patrol between calls for service. Outcomes included violent crime (homicide, robberies, aggravated assaults, or simple non-felony assaults) and violent felonies (except simple assaults) assessed at baseline (June 7, 2010) with follow-up data collected across 19 bi-weekly observation periods that ended on February 27, 2011 (or 8.5 months after baseline).

Groff et al. (2015) found, compared to the control area, the treatment area with offender-focused policing demonstrated significantly reduced:

• incidences of violent crime

Evaluation Methodology

Design:

Recruitment: To identify hot spots, first, violent crime hotspots in Philadelphia were mapped using 2009 incident data and spatial statistics. Initial analyses identified 818 street corners with high violent crime counts adjacent to at least one other street corner with a high violent crime count. Many of these 818 high-crime street corners were clustered, and consequently, target areas often included several of them. Another analysis was performed on a subset of violence crime events involving the use or threat of lethal force, including: 1) homicide, 2) armed robbery (either person or carjacking), and 3) aggravated assault. Robberies that did not include the use of a deadly weapon and simple assaults were excluded so that the identified hot spots would represent the most serious violent crime hot spots. From these analyses, 81 potential deployment areas were delineated from the map of hot spots by district captains (i.e., local police commanders). The commanders used their operational knowledge to delineate the deployment area boundaries. The command team was asked to identify 27 areas suitable for each of the three treatment modalities. The deployment area boundaries were revised to balance police operations with research priorities (i.e., achieving geographic separation of the target areas to examine spatial displacement and diffusion effects). The final 81 hot spots recruited for the study contained an average of 3 miles of streets, .044 square miles, and 23.5 intersections, making them larger than the size of hot spots in other experiments.

Assignment: The experiment used a stratified randomized design with an unequal randomization ratio of 3:1. This design accommodated the Philadelphia Police Department's desire to address 60 hot spot areas (20 per treatment type) concurrently to have the largest possible impact on violent crime. Police commanders classified the 81 hot spots into three groups based on their qualitative assessments of the hot spots' amenability to the three different tactics. Authors reported they expected that hot spots appropriate for each tactic might be more similar to one another than to places selected for the other tactics, and thus, these groups were used as qualitative strata in the experimental design. Randomization was performed separately for each stratum (foot patrol, problem-oriented policing, and offender-focused policing), producing 20 areas assigned to each treatment and 7 areas being assigned to control in each stratum (resulting in 21 total hot spots assigned to control). The district captains responsible for implementing the treatments did not participate in the randomization process, and the captains were not informed of the location of the control area locations.

Attrition: It appeared that no hot spots dropped out of the study so there was no attrition.

Sample: Demographic data for the hot spots were not reported.

Measures: Crime rates were collected from the Philadelphia Police Department crime incident database and were measured using two outcome variables:

• All violent crime, which included homicides, robberies, aggravated assaults and simple (nonfelony) assaults.
• Violent street felonies, which included homicides, robberies and aggravated assaults.

Analysis: A repeated-measures multilevel analysis was conducted in which bi-weekly outcome observations were nested within treatment and control areas. Three treatment effect contrast codes were created to provide a comparison between treatment and control areas only within each of the three randomization strata and only during the biweekly observations when the treatments were implemented (i.e., 20 treatment vs. 7 control areas for each stratum). In addition, three sets of control variables were included in the models to adjust in various ways for the different implementation timing of the three programs Mixed-effects negative binomial models generated model estimates. The exposure variable was geographic area in square miles to control for differences in the hot spot sizes. Incident rate ratios (IRRs) were calculated to report the expected percentage change in the predicted count of the outcome variable, and a Bonferroni correction was applied to reduce the chance of making a type I error for multiple outcomes.

Intent-to-Treat: Not discussed. However, outcomes were analyzed according to the condition in which each hot spot was assigned, which is in line with intent-to-treat protocol.

Outcomes

Implementation Fidelity: Fidelity was measured separately for each treatment. For the foot-patrol treatment, authors checked daily logs, reviewed incident databases, and interviewed at least one foot patrol officer from each area to ensure compliance. The interviews revealed that approximately one-third of the foot patrol officers coordinated with other agencies to solve problems and some mentioned focusing on problem people.

In the problem-oriented policing treatment, fidelity was evaluated through: 1) a review of the action plans submitted, 2) interviews of the headquarters personnel and the program director, and 3) field visits to the sites. Each district captain was required to submit and continuously update an action plan documenting the progress in each area. Fieldwork and fidelity surveys showed that even with the additional support of a mentor from headquarters, problem-oriented policing teams undertook relatively "shallow" analyses. As a result of analysis, at least eight problem-oriented policing areas switched to a focus on nonviolent crime and quality-of-life issues. In the end, just over half of the problem-oriented policing sites remained consistently targeted to violent crime problems.

For offender-focused policing, treatment fidelity was measured via reports filed by officers working in treatment areas. The reports provided lists of the individuals they were targeting and the number of times the targeted individuals were questioned. According to self-report data from the offender-focus team members and patrol officers, officers made frequent contact with these prolific offenders ranging from making small talk with a known offender to serving arrest warrants for a recently committed offense. The most frequent tactic used was surveillance followed by aggressive patrol and the formation of partnerships with beat officers.

Baseline Equivalence: Baseline measures found all treatment and control groups to be statistically equivalent on three geographic compositions (geographic area, street network length and intersection count) and the two outcome measures (90-day all violent crime counts and 90-day violent felony counts).

Differential Attrition: Not applicable as it appeared that no attrition (or drop-outs of hot spots included at random assignment) occurred.

Posttest: The offender-focused treatment reduced expected "all violent crime counts" by roughly 42% related to the offender-focused control areas. There were no differences in violent crime between problem-oriented policing treatment and control hot spots, nor were there significant differences between foot patrol policing treatment and control conditions.

Similarly, the expected violent felony counts in the offender-focused treatment areas were approximately 50% lower than the expected violent felony counts observed in the offender-focused control hot spots. Again, no differences in violent felony counts were found in the problem-oriented and foot patrol policing treatment hot spots compared to control hot spots.

Long-Term: Not conducted.

Study 2

Summary

Ratcliffe et al. (2015) employed a quasi-experimental design in a complementary study to Study 1 (Groff et al., 2015) by using the same region, data, and time frame, but combining the control group areas into one area for comparison with treatment areas. Residents who lived in the hot spot areas were surveyed for their perceptions of crime, safety, and police services at baseline and posttest.

Ratcliffe et al. (2015) did not measure a Blueprints behavioral outcome.

Evaluation Methodology

Design:

Recruitment: After crime hot spots in the city of Philadelphia were identified as described in Study 1, researchers used a geographic information system to identify residences that fell within the borders of each hot spot.

Assignment: From lists of taxable properties in the randomized intervention or control areas, addresses were randomly sampled within each area, such that the sample of households included 1,830 in the offender-focused area, 1,860 in the foot patrol area, 1,830 in the problem-solving area, and 1,855 in the control area. The study deviated from the original randomization by combining the area for the offender-focused control group with the control group areas for the other treatments. Thus, the authors referred to the design as quasi-experimental.

Assessments and Attrition: Repeated cross-sectional surveys were mailed to sampled addresses within the identified study areas at baseline and posttest. The respondents were not necessarily the same for each wave; thus, attrition was not measured. There was an overall 8.5% (628) return rate at baseline and an 8.8% (647) return rate at posttest. Missing data on the measures ranged from 3.2% to 12.7%.

Sample:

The sample of residents who returned surveys at baseline was predominantly female (66.1%), just over half Black (50.9%), age 50 or older (57.0%), and had at least a high school diploma (83.2%).

Measures:

All outcome measures focused on satisfaction or perceptions and came from self-reported survey responses of neighborhood residents. The perception of violent crime scale assessed whether participants felt that violent crime was a "big problem," by asking about specific types of crime (robbery and muggings, murder, gang violence, gun violence, and others), α=.94. The satisfaction with police services was assessed with a scale ranking police responses in their neighborhood from "very poor" to "very good," α=.89. Residents' perceptions of property crime were gauged with three questions about people breaking into homes, people breaking into or stealing cars, and vandalism of homes, buildings or properties, α=.82. Residents' perceptions of physical disorder asked about abandoned houses and cars, litter, and homeowners allowing their property to become "run down," α=.70. Residents' perceptions of social disorder were assessed with a scale asking about excessive noise, the homeless population, disruptive teenagers, drugs, and truancy, α=.84. The scale for perception of safety comprised two questions about feeling safe/unsafe in the neighborhood, α=.71. The final scale, perceptions of procedural justice, asked opinions on the honesty, courtesy, fairness, and good judgment of the police, α=.84.

Analysis:

The effect of the intervention on residents' perceptions of safety and order within their communities was evaluated with linear regression, controlling for demographic variables. A contrast coding scheme was used which allowed estimation of the impact of each of the interventions relative to control locations after the treatments were administered while controlling for pre-treatment responses. The analysis of individuals did not adjust for clustering within sites, the unit of assignment.

Intent-to-Treat: The analysis used all cases with posttest data, and multiple imputation accounted for missing data among returned surveys, but only 9% of those selected returned the surveys.

Outcomes

Implementation Fidelity: Police officers' compliance with the offender-focused policing program was assessed using reports that listed the targeted offenders and detailed the frequency and nature of each interaction with them.

Baseline Equivalence: Not reported.

Differential Attrition: Not applicable; sample Ns were not tracked.

Posttest: There were no significant results. The authors noted more positively that perceptions of police did not worsen as a result of increased presence in the area.

Long-Term: No follow-up assessments.

Study 3

Summary

Santos and Santos (2016) used a randomized controlled field experiment of 48 hot spots in one suburban city with average crime levels. Blocked into three groups based on crime frequency (low, medium, high), the identified hot spots were randomly assigned to intervention and control groups. For the intervention, detectives regularly visited previous offenders at home over a 9-month period (October 2013-June 2014). Data were collected on four measures during October 2012-June 2013 (a "pre-test period") and compared to data collected during the 9-month intervention period.

Santos and Santos (2016) found, compared to the control area, the treatment area with offender-focused policing demonstrated significantly reduced:

• repeat arrests

Evaluation Methodology

Design:

Recruitment: In this study of Port St. Lucie, Florida, hot spots were identified based on the makeup of the neighborhoods in the city and the number of reported crimes in the hot spots. Clusters of residentially zoned census blocks were merged together so the hot spots were consistent in square mileage and numbers of reported crimes. Criteria for selecting the hot spots included: at least 15 crimes, around 0.60 square miles, residential areas, with identifiable boundaries and neighborhoods. The process identified 48 hot spots.

Assignment: The 48 hot spots were assigned through a partially blocked randomization design. The blocks included: low, medium and high crime per offender rates for residential burglary or theft from vehicle. Equal numbers of hot spots in each block were randomly assigned as either treatment or control areas, which resulted in six high crime per offender hot spots, 13 medium crime per offender hot spots, and five low crime per offender hot spots, for a total of 24 in each group.

The study identified 151 targeted offenders who resided in the intervention hot spots. For the intervention group, each offender's home address was verified through official databases or visits from detectives. The study lacked the time and resources to do the same for offenders in control hot spots. The verification process also might have compromised the integrity of the control group by alerting the offenders to extra police attention.

Attrition: Data were collected during October 2012-June 2013 (a "pre-test period") and during October 2013-June 2014 (the 9-month intervention period). All 48 hot spots were used in both periods

Sample:

Port St. Lucie had a population in 2014 of about 170,000 and a crime rate below average for the US. The hot spots averaged .58-.73 square miles and had populations of 3,026-3,471.

Of the 151 targeted offenders residing in the intervention hot spots, 70% were White, 27% were Black, and 3% were Hispanic. Most of the targeted offenders were between 18 and 35 years of age (70%), adults (88%) and male (88%). No data on offenders residing in the control hot spots group were available.

Measures:

The authors assessed the behavioral impact of the hot spots program through four measures. The measures were based on reported crimes and arrests, and the authors noted the limitations of such data. Three of the four measures were collected at the hot spot level:

(1) The count of reported crimes for residential burglary and theft from vehicle in each hot spot. Data were obtained from the police agency's records management system.

(2) The count of burglary, theft, and drug offense arrests in each hot spot for individuals who live in the hot spots. Data came from arrest data.

(3) The ratio of burglary, theft, and drug offense arrests per individuals arrested who live in the hot spot. Data came from arrest data.

The authors also collected data at the individual level, but only for offenders living in hot spots assigned to the intervention group:

(4) The count of all arrests for each targeted offender. Data came from arrest data.

Analysis:

Negative binomial and ordinary least squares (OLS) regression were used to test the effect of the presence of intervention and its dosage on crime and offender recidivism with pretest period measures included as a control. Additionally, average treatment effects were calculated for each hot spot for each outcome variable; this analysis used t-tests to see if the mean difference in outcomes between the pretest and posttest periods were significantly different.

Intent-to-Treat: No hot spots were dropped.

Outcomes

Implementation Fidelity: The number of offenders in a hot spot ranged from 1 to 13 (average = 6), with the number of contacts with police ranging from 1 to 42 (average 7-8 contacts). Detectives were successful in making contact with targeted offenders approximately 80% of the time.

Baseline Equivalence: Census blocks provided information about population that was useful for the equivalency analysis of the treatment and control group assignments. Independent t- test results comparing the intervention and control hot spot area means of crimes, arrests, repeat arrests, and geographic size as well as population and housing density found no statistically significant differences.

Differential Attrition: No attrition.

Posttest: At the end of the 9-month treatment, regression analyses showed the program had no statistically significant effects on three of the measures at the hot spot level (reported crime, arrests, repeat arrests).

Average treatment effect analysis found no significant differences on reported crime or arrest counts between hot spots in the intervention group and those in the control. For repeat arrests, the OLS models in Table 9 showed a marginally significant effect of the intervention, while the comparison of the change within conditions in Table 10 showed a significantly smaller increase due to the intervention.

For the treatment hot spots only, individual level comparison of targeted offenders' arrests during the intervention and during the same time period one year prior showed a significant 68% reduction in average arrests per targeted offender.

Long-Term: Long-term effects (> 1 year) were not examined.