Thirty-five studies with 106 separate driver improvement programmes were included (more than 1,300,000 people in the intervention groups).
In terms of study quality, the majority (92%) of studies used random or 'close to random' allocation to treatment groups. Most studies showed good or very good baseline comparability of the treatment groups (72%), and were rated as having very good or good validity (65%).
Driver improvement programmes were associated with a small but statistically significant reduction in crashes (d=0.030, 95% CI: 0.027, 0.034, P<0.05) and violations (d=0.061, (95% CI: 0.057, 0.064, P<0.05). Significant heterogeneity was detected for both meta-analyses (P<0.05).
Licence suspension or revocation was the most effective intervention for reducing crashes (d=0.113, 95% CI: 0.101, 0.126, P<0.05) and violations (d=0.190, 95% CI: 0.177, 0.202, P<0.05). Significant heterogeneity was detected for both meta-analyses (P<0.05).
Reductions in crashes and violations were also reported for warning letters, group meetings and individual hearings. Educational and informational material interventions were not associated with a reduction in either outcome.
Interventions associated with reductions in violations were only weakly associated with reductions in crashes (correlation, r=0.30).
Statistical heterogeneity remained even after stratification by a number of modifying factors.
The results of all subgroup analyses were reported.