Kellam and colleagues (1994) evaluated the effects of the Good Behavior Game (GBG) on students at a 6-year follow-up period. The researchers began the initial evaluation of GBG in the 1985–86 school year by selecting five urban areas within Baltimore City, Md. From a total of 19 schools, the three or four most similar schools were identified within each of the five urban areas and randomly assigned to one of three conditions were 1) GBG, 2) the Mastery Learning (ML) program (designed to improve poor reading achievement), and 3) an external control condition with no experimental intervention. Both interventions were administered at the classroom level. In order to avoid confounding intervention status with school effects, classrooms in the intervention schools (GBG and ML schools) were divided into two types: intervention classes (GBG or ML) and internal control classes (classes in the same school which did not receive either intervention). Individual first-grade classrooms were randomly assigned to intervention or internal control groups within the intervention schools.
This multilevel design originally included 19 schools, 41 classrooms, and 1,196 students within five urban areas. The initial sample of students was 49 percent male, and 65 percent African American, 31 percent white, and 4 percent other ethnic groups. Children ranged in age from 5.0 to 9.7 years, with an average age of 6.6 years. There were no significant differences between the intervention groups on baseline characteristics of teacher ratings of aggressive, disruptive behavior, fall-of-first-grade achievement, or free/reduced-price lunch. However, there were significant differences between GBG students and internal control students on measures of depressive symptoms.
The final sample at the 6-year follow-up included 693 students who were in the same intervention condition for 2 years. There were 153 students from eight classrooms who received the GBG intervention and 86 students from six classrooms who were in the GBG internal control condition (students who were in the same school but did not receive the GBG intervention). The sample of the external control condition included 157 students from 11 classrooms. Finally, there were 163 students from nine classrooms who received the ML condition and 134 students from seven classrooms who were in the ML internal control condition (students from the same school who did not receive the ML intervention). No information was provided on the comparability of the intervention groups included in the follow-up sample. Most of the students lost to follow-up had moved out of the system. A small number of others refused to participate in the study.
The primary outcome measures were the Teacher Observation of Classroom Adaptation—Revised, which measured aggressive behavior, and the Diagnostic Interview Schedule for Children, which identified conduct disorder. Latent growth modeling and hierarchical logistic regression models were used to adjust for variations at baseline.
Witvliet and colleagues (2009) used a randomized control trial to evaluate the GBG on children’s externalizing behavior development. Study participants included 825 kindergarten children from 47 classrooms in 30 elementary schools in two urban areas and one rural area in the Netherlands. Intervention began in first grade for 425 children in the GBG group and 220 children in the comparison group. The average age of study participants was 6 years at the end of kindergarten. Of the children, 56 percent were from a Dutch/white background, 10 percent were Moroccan, 10 percent were Turkish, 6 percent were Surinamese, 5 percent were from the Netherlands Antilles, and 13 percent were from other ethnic backgrounds. Also, 36 percent were from low socioeconomic-status (SES) families. The groups were not significantly different on baseline characteristics, except that control students had lower SES scores. However, 113 students dropped out of the student cohort between first and second grade because of grade retention or because they moved to another school. These students were not significantly different from the students who remained in the study, except that they were significantly more likely to come from low–SES families.
The study coordinator randomly assigned participating classes to one of three conditions: a control condition, a GBG–only condition, or a GBG and parent intervention condition. Intervention began in first grade and continued for 2 years. Since the parenting component began after the 2 years of GBG, the effects are not analyzed in this study and data for the two intervention groups is analyzed together in this study.
Measures of teacher ratings included a) the Problem Behavior at School Interview used to measure externalizing behavior (the sum of the oppositional and conduct problems scales) and prosocial behavior in kindergarten (assessed by four items) and b) the 11-item Social Problems scale of Achenbach’s Teacher’s Report Form. Peer acceptance, the number of mutual friends, and proximity to others were measured through peer nominations. Descriptive statistics and latent growth modeling were used in the analysis of data.