The Construct of Community Readiness
We used structural equation modeling (SEM) with AMOS, Version 4 (Arbuckle, 1999
) with maximum likelihood estimation to test the fit of our hypothesized second-order factor model. This analysis had two steps. First, the hypothesized model fit was assessed in comparison to an alternative independence model that estimated means and variances of the observed variables (see ). Second, the fit of an alternative one-factor readiness model was assessed in comparison to the same independence model. Both models were identified by setting factor means to zero and factor variances to one.
Hypothesized second-order factor model with standardized factor loadings.
See for a summary of the fit indices. As outlined by Hu and Bentler (1995)
, we used multiple fit indices (the chi-square, Tucker Lewis index [TLI], comparative fit index [CFI], relative noncentrality index [RNI], root mean square error of approximation [RMSEA], and the change in chi-square to change in degrees of freedom ratio for the nested models) to assess model fit. Based on the combination of several factors, our hypothesized four-factor model was deemed to have the best fit. This model reported a nonsignificant chi-square value χ2
= 183) = 108.57, p
> 0.05. In addition, in comparison to the baseline model, the relative fit indices were above 0.90 (0.947 TLI/0.957 CFI/0.957 RNI), and the RMSEA was within the acceptable range (0.038). In addition, all regression weights (factors to items) were high, generally similar, and significantly predicted by their respective factors (see ).
Fit Statistics for Latent Variable Readiness Analyses
In contrast, the alternative one-factor readiness model was shown to fit poorly. The chi-square value χ2 (90, N = 183) = 182.70, p < 0.001 was significant, and all relative fit indices and the RMSEA were below acceptable values. In addition, the change in chi-square and change in degrees of freedom χ2 (4, N = 183) = 74.13, p < 0.001 between our hypothesized model and this alternative model was significant. One similarity remained between both models: All regression weights (factors to items) remained significantly predicted by the single readiness factor.
Relation of Community Readiness with Individual, School, and Community Factors
Two sets of analyses examined the relation of community readiness with individual and community characteristics. First, the individual, school, community, and workplace characteristics were added to the SEM model as predictors of perceptions of community readiness (correlations between the predictor variables were estimated). Model fit, the significance of regression weights, and the r-squared were assessed. Second, at the level of the community, hierarchical regressions2
were run in order to assess the amount of shared variance between community readiness and the independent variables, along with significant variance of each domain.
The fit of the community readiness model predicted by individual and community characteristics was deemed to be acceptable. Although the chi-square value was significant χ2 (198, N = 183) = 260.52, p = 0.002, all relative fit indices were around 0.90 (TLI = 0.887/CFI = 0.912/RNI = 0.912), and the RMSEA was within acceptable range (0.042).
In assessing the value of individual, school, community, and workplace characteristics in predicting community readiness, team members’ history of collaborative experience (β = 0.28, p < 0.001) and perceptions of school proactiveness (β = 0.25, p < 0.01), school problems (β = −0.19, p < 0.05), and community substance use norms (β = −0.27, p < 0.001) were all significant predictors of perceptions of perceived community readiness; that is, individuals with more experience in collaboration, and those who perceived better functioning schools and less acceptance of adolescent substance use, rated their communities as more ready. This model accounted for a significant (34%) amount of variance in readiness F(8, 145) = 11.07, p < 0.001.
See for results of the community-level analysis. The community-level analysis showed similar findings. The final five-step model including community demographics, team member characteristics, characteristics of school functioning, community substance use atmosphere, and workplace characteristics was significant, F(10, 17) = 4.94, p < 0.01. Individual characteristics, perceptions of school functioning, and the community substance use atmosphere each added significant additional variance in predicting ratings of community readiness. That is, teams with older team members and those with more experience in collaboration rated their communities as more ready; teams that reported their school districts as better functioning rated their communities as more ready; and teams that reported their communities’ norms were more accepting of adolescent substance use rated their communities as less ready.
Community Readiness Predicted by Individual and Community Characteristics at the Team Level
Level of Within-Community and Cross-Respondent Agreement
Two sets of intraclass correlations (ICCs) were calculated in order to assess the degree to which respondents within communities agree on the level of readiness in their community. The ICC is a useful tool to use to answer this question because it is based on a reparame-terization of total variance into the variance component due to variability between communities and the variance component due to variability within communities. The ICC represents the variance between communities divided by the total variance, and ranges from 0 to 1. High values indicate that a large proportion of total variance is due to variability between communities, not to within-community variability. If respondents within each community agree on the readiness of their community, and there is a fair degree of variability across communities in the level of readiness, then the ICC should be relatively large.
ICCs were first calculated for the readiness measure and each subdomain with the full community sample to assess the agreement of all individuals within the same community. We then calculated these ICCs for the same measures separately by reporter (i.e., TMs, ADs).
See for the ICC values and significance levels. The full sample analysis demonstrated significant ICCs (p < 0.05) for the initiative, efficacy, and readiness scales, while the attachment subscale value approached significance (p < 0.10). For the ICCs by respondent type, the ICC was significant for the initiative, efficacy, and community readiness scales. The ICC for community agency directors was significant for the efficacy scale.
Community Readiness and Subscale ICC Values for TM, AD, and Full Sample
Two analyses were run to assess the agreement between the TMs and ADs. First, simple correlations of the subscales were run for each sample to assess how the construct operated within each sample. Second, simple correlations were run across average AD and TM reports for each community. This analysis included creating mean scale scores that were aggregated by respondent category (AD vs. TM) within community for the community readiness scale and each subdomain.3
The simple correlations for the readiness subscales were quite similar for both TMs and ADs. All subscale correlations ranged from r = 0.52 to 0.73, with the exception of the correlation between the attachment and leadership subscales (the correlation was about 0.30). Hence, the construct seems to operate similarly within each sample.
Cross-respondent type correlations are presented in . Ratings of readiness and each subscale ranged from small to moderate (r = 0.15 to 0.31, p < 0.15); that is, respondents at different organizational levels from the same community show low levels of agreement about the level of readiness in their communities.
TM Report of Community Readiness with AD Report of Community Readiness, n = 27