Twenty studies were included, with 30,229 participants. Five studies used multilevel models, five used general linear models and the others used descriptive statistics. No other details of study design were reported. The duration of follow-up ranged from none to 10 years.
Drug use: The unweighted mean effect size for drug use was 0.05 (range -0.08 to 0.36). This effect size was less than small according to Cohen’s classification. No significant statistical heterogeneity was found.
Psychosocial behaviour: The unweighted mean effect size for psychosocial behaviour was 0.10 (range -0.09 to 0.38). This effect size was less than small according to Cohen’s classification and significant statistical heterogeneity was found (p<0.0001).
The regression model: Found that five characteristics accounted for most of the heterogeneity. Longer follow-up time (p<0.02) and more rigorous statistical methods (p<0.001) were associated with a reduced intervention effect. Later D.A.R.E. year (p<0.02) and more white (p<0.002) and more black participants (p<0.03) were associated with an increased intervention effect.