Effectiveness of the intervention on self-reported consumption and harms.

aAverage number of respondents at-risk in each community (n) and the proportion averaged across experimental (n = 10) and control (n = 10) communities.bMean for linear regression outcome.cStandard error for linear regression outcome.dMean difference for linear regression outcome.eStandard error obtained using jack-knife estimate because of the small number of clusters.fAdjusted for age and sex distributions, and pre-intervention values. The interaction term tests whether the post-intervention OR for experimental communities is statistically significantly different to that of control communities, after adjusting for pre-intervention ratios.gUnadjusted analysis, controlling for clustering only.hStatistically significant at the 5% level.iAdjusted for age and sex distributions, and pre-intervention values. The interaction term tests whether the post-intervention mean for experimental communities is statistically significantly different to control communities, after adjusting for pre-intervention means.