Analysis approaches to the evaluation of community interventions must be sensitive to a wide variety of analytic contaminants that may bias the statistical assessment of changes in outcome measures. These contaminants include model misspecifications related to failures to control for community-specific time trends, temporal autocorrelated errors in equations, spatial autocorrelated errors among geographic units, and other failures of unit independence otherwise indexed by estimated intraclass correlations. Although an enormous amount of progress has been made toward the solution of many of these analytic problems over the past year, the contemporary evaluator of community interventions is left with a number of unenviable design and analysis choices; choices that inevitable force an assessment of the relative threats of different sources of error to be internal and external validity of the evaluation. This article describes the choices made for the evaluation of the Community Trial Project outcome data. (A)
Abstract