Ninety-four studies (with 136 treatment control comparisons) were included. The number of participants was not reported.
Characteristics of the students.
The ES was similar for programmes targeting the general student population compared with high-risk students; the mean ESs (11 high-risk student studies) were 0.05 and 0.07, respectively, (P=/<0.05). Cognitive-behavioural programmes had a higher ES when targeting high-risk youths in comparison with the general student population, but the difference was not statistically significant: ES 0.20 (5 studies) versus ES 0.05 (39 studies).
Programmes targeting middle-junior high-school students showed higher ESs than programmes targeting late elementary or senior students, but the differences were not statistically significant; the mean ESs were 0.09 (95% confidence interval, CI: 0.05, 0.14) for middle-junior students, 0.05 (95% CI: 0.00, 0.10) for late elementary students, and 0.04 (95% CI: -0.05, 0.14) for senior students. However, studies of late elementary school students had, on average, longer follow-up periods (2 years) than studies of middle-junior high students (1 year) or senior students (6 months to 1 year), and this might have confounded comparisons. The duration of follow-up did not appear to be associated with ES: the ESs were similar when using data from the longest and shortest follow-up time periods, and the regression analysis showed no relationship between duration of follow-up and ES.
Characteristics of the programme.
Programme length did not appear to be related to the ES (regression analysis, P=0.06). The scatter plot showed that the positive relationship was due to one outlier and, after excluding this outlying study, the relationship was not significant.
The studies suggested that programmes delivered by peers were effective, but they showed no effect for programmes delivered by peers and teachers together. However, the authors stated that the results might have been confounded by other unmeasured characteristics of the programmes. The multivariate regression analysis reported a main effect of peer involvement (P=0.05), no main effect of teacher involvement (P=0.71), and an interaction between peer and teacher involvement (P=0.02). The ES was 0.20 (95% CI: 0.09, 0.30) for peers without teachers, 0.04 (95% CI: -0.03, 0.11) for peers and teachers together, and 0.05 (95% CI: 0.02, 0.08) for no peer involvement.
Methodology.
None of the methodological criteria were significantly related to the ES (the results were presented).
There were no significant differences in ES between studies published in peer-reviewed journals (91 studies; mean ES 0.05), book chapters (6 studies; mean ES 0.08), or unpublished studies (39 studies; mean ES 0.12).
The funnel plot showed no evidence of publication bias.