Effects of a universal classroom behavior management program in first and second grades on young adult behavioral, psychiatric, and social outcomes

19 schools, 41 classrooms, 1,196 children in the first-grade cohorts; the cohort-1 analytic sample was 922 (238 GBG, 169 internal controls, 515 additional controls); 689 of 922 (75%) interviewed at ages 19-21

Baltimore City public schools, five urban areas; children entering first grade in 1985-86 (cohort 1) and 1986-87 (cohort 2), followed to ages 19-21.

Classroom-level randomisation within schools, with an internal control condition in the same schools. The long-run outcome data come from a 90-minute telephone interview at ages 19-21 using the CIDI-UM instrument coded to DSM-IV criteria — self-report, not administrative records. The Johns Hopkins Prevention Research Center ran both the trial and the follow-up and has been the principal promoter of the GBG in the United States, so the evaluation is developer-involved rather than independent. The second cohort is the study's own internal replication and is the part most often left out of citations.

Cohort 1 males: drug abuse/dependence disorder 19% (GBG) vs 38% (internal control), p = 0.01; among high-aggressive males 29% vs 83%, p = 0.02. Regular smoking in males 6% vs 19%, p = 0.03 (high-aggressive males 0% vs 40%, p = 0.008). Antisocial personality disorder 17% vs 25%, p = 0.07 overall (high-aggressive males 38% vs 80%, p = 0.10). High-school graduation in males 68% vs 54%, p = 0.09 (high-aggressive males 75% vs 20%, p = 0.03). Alcohol abuse/dependence 13% vs 20%, p = 0.08. Females were null throughout: drug abuse/dependence 8% vs 7%, p = 0.93, and no main effect on smoking. COHORT 2, whose GBG teachers received little retraining or mentoring, attenuated badly: no significant effect on alcohol abuse/dependence (contradicting cohort 1), no significant effect on ASPD, and the drug-abuse benefit appeared among LOW-aggressive rather than high-aggressive youth — the opposite of the cohort-1 moderator that the programme's theory rests on.

Low by design — classroom randomisation. The residual concern is that the headline results are subgroup estimates in small cells (the high-aggressive male comparisons rest on dozens of children, not hundreds), which is a power and multiplicity problem rather than a heredity problem.

Cited as the strongest long-run evidence in school behaviour management, and it is — but the study contains its own partial replication failure. Cohort 2 lost the alcohol and ASPD effects and reversed the direction of the baseline-aggression moderator. The independent English replication (Humphrey et al. 2018, 77 schools) found nothing at ITT.