Question 1	To understand the extent to which the level of four dimensions of coalition functioning varied across the 12 CTC coalitions, four unconditional multilevel models (one for each dimension of functioning) were estimated to determine the proportion of variance attributed to differences among members and/or differences among CTC coalitions (Muthén and Muthén 2010). The 2-level hierarchical linear model with random intercept nested coalition members (level 1) within coalitions (level 2): Yij=β0j+rij; whereas β0j =γ00+µ0j.
Question 2	Coalition functioning factor scores (based on coalition member reports) were then mean aggregated at the coalition level for a series of meta-regressions (Hedges and Olkin 1985) that could relate coalition functioning in the CTC communities to the effect size differences in the adoption of science-based prevention (based on community leader reports) calculated for each of the 12 matched community pairs. This meta-analytic approach, described elsewhere in detail (Shapiro et al. 2013a), permits examination of the impact of coalition functioning, which was observed only in the 12 intervention communities, on the effect size difference in community leader reports of adoption of science-based prevention between matched pairs of intervention and control communities (Shapiro et al. 2015). The outcome variable used in each meta-regression was the unbiased effect size estimate for each pair, and the predictor in each model was one of the dimensions of coalition functioning hypothesized to influence the size of the intervention effect. Fixed effect meta-regressions were estimated in SPSS (v.19) with the METAREG.SPS macro (Wilson 2005), assuming that the variability in the effect size differences between pairs of communities is systematic and exclusively due to variation in the dimension of coalition functioning under examination. In a meta-regression, the sum of squares for the regression model (QR) is analogous to an F test in traditional regression analysis. If significant, the predictor is significantly associated with heterogeneity in the effect size difference in adoption of science-based prevention. If the sum of squares of the residual (QE) is significant, variability in the effect size difference in adoption of science-based prevention remains to be explained. In cases where QR and QE for the fixed effect model are both significant, the model is re-specified as a mixed effect model to consider sources of variance in addition to the moderator (Lipsey and Wilson 2001; Raudenbush 1994).
Question 3	Eight multilevel models (one for each of the four dimensions of functioning as related to each of the two coalition capacities) were estimated to determine the extent to which each dimension of functioning predicted levels of capacities (new skills and organizational linkages). The four dimensions were modeled independently because high intercorrelations would obscure relationships (Allen 2005). These dimensions have moderate shared variance (average of 64 %), but modeling them independently speaks to the practice literature that seeks to enhance these dimensions of functioning separately (Shapiro et al. 2013b).