The challenge of the family-system outcome mediator: Using multiple family-member perspectives in a confirmatory evaluation model of a family-based juvenile court diversion program

A review of existing evaluations of successful family-based programs revealed that few documented evaluations of family-based programs have demonstrated both improved family functioning and family contribution to reduced problem behaviors. These results can lead program developers to believe that resources focused on family-based interventions are better spent elsewhere.;Insignificant family functioning effect may be due, not to programming, but instead to the method of measurement and data analysis. Family researchers traditionally have utilized models that include either single-perspective variables or single variables calculated from multiple perspectives. Newer technology in the form of easily accessed computer programs for analyzing structural equation models (Cook & Goldstein, 1993; Kenny & Berman, 1980) has created a potentially viable and more accurate alternative in the form of latent variable models with correlated rater error.;The present study explored the potential advantage of utilizing such a model for evaluating a family-based program. Using existing data from a confirmatory family-based program evaluation (Reynolds, 1998), the study compared a model with both multiple perspectives and correlated rater error to more standard models seeking to determine if the latent variable model would more explicitly reveal family functioning as a mediator of longer term outcomes. Three models were tested against the latent variable model: one with adolescent perspective only, another with parent perspective only, and the third with averaged perspectives.;Results demonstrated that the latent variable model revealed the family functioning variable (perception of family goal agreement) as a mediator far more strongly than the model with adolescent perspective alone and somewhat more strongly than the models with parent or averaged perspectives. The averaged-perspective model worked to establish the mediator only slightly better than the parent-only model. Correlated rater error somewhat tempered the mediating effect of the latent variable. In sum, with this diversion program data, analysis without the latent variable would lead to the conclusion that change in perception of goal agreement was insignificant to program effect. With the latent variable included, perception of goal agreement would emerge as a powerful change agent.;Despite non-parametric data that precluded the use of significance testing, this study demonstrated the importance of utilizing latent-variable models when evaluating family-based programs.