Good Behaviour Game: Evaluation Report and Executive Summary

77 primary schools randomised (38 intervention, 39 control); target cohort N = 3,084 Year 3 pupils (aged 7-8 at baseline); 19% of starters missing from the final analysis

England, three regions; universal delivery to whole classes by class teachers over two school years (September 2015 to July 2017).

The single most important test of the Good Behavior Game outside the United States, and the design the archive most wants: a preregistered (ISRCTN 64152096), two-arm cluster-randomised efficacy trial, delivered by the programme owner (Mentor UK, supported by the American Institutes for Research) but evaluated by a wholly independent team at the University of Manchester led by Neil Humphrey. EEF rated the findings "high security" — well-powered (MDES 0.15-0.20), balanced on prior attainment — with the only demerit being 19% missing at follow-up. The primary outcome was reading on the Hodder Group Reading Test, an independent standardised measure; behaviour outcomes were teacher ratings. Results were insensitive to multiple imputation and to every modelling variation attempted.

Null on everything at intention-to-treat. Reading ES = 0.03 (95% CI -0.08 to 0.16); concentration problems ES = 0.03 (-0.15 to 0.21); disruptive behaviour ES = 0.06 (-0.09 to 0.22); pro-social behaviour ES = -0.13 (-0.36 to 0.11). No differential benefit for FSM-eligible pupils. The pre-specified "at-risk boys" subgroup showed marginal, non-significant favourable trends on concentration problems (-0.29, -0.66 to 0.08) and disruptive behaviour (-0.30, -0.68 to 0.07), which did not survive multiple imputation in one case. No effect on teacher stress, classroom-management self-efficacy or retention. The implementation record is the other headline: gameplay never reached developer-specified frequency or duration (averaging about 15 minutes), and ONE QUARTER of intervention schools abandoned the game before the trial ended, citing set-up cost, poor fit with curriculum content, and the rule forbidding the teacher to interact with pupils during play. Cost GBP 4,000 per school start-up, GBP 37.04 per pupil.

Low. School-level randomisation with balanced prior attainment; genes cannot differ between arms.

This IS the replication test. The Baltimore trials (Kellam and colleagues) reported substantial long-run effects on young-adult psychiatric and substance outcomes in high-aggressive boys; transported to England, evaluated independently, and measured on attainment and teacher-rated behaviour, the programme produced nothing at ITT. The at-risk-boys trend is the only echo of the Baltimore subgroup result and it is not statistically significant.