The analysis that estimated impacts after 3 years of the program used data from the full 8th grade student body, including those who entered the study schools as 6th-graders or 7th-graders and remained as 8th-graders and those who were new in spring 2009 (third follow-up). Nineteen percent of the students in the third follow-up sample were not in the original sample at baseline, while 81 percent were in both samples. These data were used to answer questions about the effects of the intervention across the general student population after 3 years of program implementation. Sixty one percent (5,854) of students enrolled in the study schools completed a survey at the third follow up that provided data for the impact analysis after 3 years of the program. Student demographic characteristics at baseline for the 36 schools that participated for all 3 years of program delivery indicate that minority students composed 72 percent of the sample in intervention schools and 61 percent of the sample in control schools. Forty-nine percent of the students in each group were male, and 60 percent in each group lived in single-adult households. A two-tailed t-test, obtained from multilevel regression models using the school as the unit of analysis, indicated that none of the mean demographic characteristics was statistically different between students attending intervention schools and those attending control schools.
To address the research question regarding how the program impacts students at high risk for violent behaviors, we identified a subset of students at high risk for violent and aggressive behaviors, based on student responses to the fall 2006 baseline survey. Student self-report was considered critical for this study because many of the behaviors of concern often take place outside of the classroom and may not be reported to the school administrators. In addition, students have the best knowledge of their own behaviors and experiences. On the other hand, some students may hesitate to disclose information about their own behaviors, especially if they have concerns about the confidentiality of the data. The survey administration procedures used in this study were designed to make students comfortable with providing self-report data (see details in the "Data Collection and Outcome Measures" section below). A large study of the test-retest reliability of the Youth Risk Behavior Survey, which asks students about numerous risky behaviors, found that three of eight violence and victimization items had at least "substantial" reliability (kappa statistic ≥ 61 percent) and that five had at least "moderate" reliability (kappa statistic ≥ 41 percent) (Brener et al. 2002).
The high-risk sample was divided into two subgroups. High-risk perpetrators were defined as students who had self-reported perpetration of any one of eight serious acts of violence at school at least once in the past 30 days.1 Approximately 25 percent of the total treatment and control group sample, or 1,923 students, were assigned to the high-risk perpetrator group at baseline. High-risk nonperpetrators were defined as students who self-reported endorsing violence or reacting inappropriately to anger but who did not self-report any of eight serious acts of violence. Approximately 3 percent of the total treatment and control group sample, or 230 students, were assigned to the high-risk nonperpetrator group.
Unlike the remainder of the students in the full sample, the high-risk samples were to be followed longitudinally; the research team attempted to survey the students in the high-risk subgroups even if they left one of the study schools and were attending another school in the same district at each follow-up. If a student went to a different district, the student was not followed. Seventy-one percent of the students identified as high risk at baseline in the 36 schools that participated for 3 years completed a survey at the third follow-up (N = 1,357).