Study 1
Esbensen and colleagues (2012) used an experimental, multisite, longitudinal (5 years) panel design across seven cities in the continental United States to study the impact of Gang Resistance and Education Training (G.R.E.A.T.) on gang membership, delinquency, and attitudes toward the police. The seven cities included in the evaluation were Albuquerque, N.M.; Chicago, Ill.; Dallas–Fort Worth, Texas area district; Greeley, Colo.; Nashville, Tenn.; Philadelphia, Pa.; and Portland, Ore. Several factors were taken into consideration when choosing evaluation sites, such as school district size, length of program history at a site, degree of program implementation, race/ethnicity composition of the area, police department characteristics (e.g., department size and organizational structure), and the amount of youth crime and gang activity in the area. At each site, between four and six schools were identified for study participation. The goal behind school selection was to choose schools that would represent their respective districts. The final sample included 31 schools and 195 sixth and seventh grade classrooms (102 received G.R.E.A.T., and 93 served as controls). Classrooms were randomly chosen to receive the intervention or serve as part of the control group. Active parental consent and student assent were obtained for everyone included in the study.
Of the 4,905 students enrolled in the 195 classrooms, 4,372 students (89.1 percent) returned completed consent forms, with 77.9 percent of parents/guardians (3,820) allowing their child to participate. The student sample was evenly split between males and females. Just over half (55 percent) resided with both biological parents, and the large majority (88 percent) were born in the United States. The sample was racially/ethnically diverse, with Hispanic youths (37 percent), white youths (27 percent), and African American youths (17 percent) accounting for 81 percent of the sample. Sixty-one percent were age 11 or younger at the pretest/baseline assessment.
The researchers reported that the random assignment was moderately successful and produced comparable treatment and control groups. There were three statistically significant differences that slightly favored the treatment group. The treatment group had a higher awareness of services, more negative attitudes about gangs, and a little less frequency of delinquent acts. However, these slight differences were controlled for in the analyses and had a negligible impact on the size or significance of group differences in the outcomes.
Outcomes measured were gang membership, delinquency, violent offending, and attitudes toward police officers. Self-report student surveys were used to gather data on each of these outcomes. The survey instrument was built specifically for this study, but the researchers drew from several other existing surveys of youth to inform their questions and scales. Gang membership was measured by self-nomination (e.g., “Are you now in a gang?”), which is the same method that many police departments use throughout the country. Delinquency and violent offending were assessed with a 15-item delinquency inventory that included both the prevalence and frequency of criminal activity during the past 6 months. Lastly, attitudes toward the police were measured by eight questions, six of which dealt with global attitudes toward police officers and law enforcement in general and two that asked what students thought of police officers as teachers. A baseline/pretest assessment was completed by all students and followed up with annual surveys for the next 5 years.
Because of the highly nested nature of the research design—observations within students within classrooms within cities—a multilevel analysis was required. As there were only a small number of cities of this level, the analysis was treated as a fixed effect through a series of dummy variables (values of 1 or 0), but all other levels of the model included random effects. A logistic model was applied to the gang membership measure, a negative binomial model was used for the counts of general delinquency and violent offending, and a linear model was used for attitudes toward the police. The study authors did not conduct subgroup analyses.