The overall goal of this proposal is to improve the methodology for evaluation of community-based cancer control programs. The research will address: (1) When should community surveys employ cohort samples rather than repeated cross-sectional samples? (2) When should matching be used in allocating communities to treatment groups? (3) How large are community- level sources of variation, and how should they best be used in determining sample-size requirements and performing power calculations? (4) What measures of the community environment can be used to assess proximate program effects or to circumvent the need for costly surveys of individuals, and how well do they perform? The project will involve both theoretical and empirical work, drawing on data resources and collaborative opportunities provided by several completed, ongoing, and proposed studies. Theoretical work on cohort vs. cross-sectional samples will include extension of a model that seeks to explain when and why the two sampling approaches yield different results and to provide a basis for incorporating attrition, cost, selection and measurement bias, and statistical power considerations into a choice of sampling plan. Analytic methods will also be developed and tested for a hybrid sampling design involving replacement of lost cohort members by other similar individuals from the community. Available data will be used to estimate parameters of the theoretical model, including intertemporal correlations, and to test alternative analytic methods with real data. Theoretical work on matching of communities will include examination of power trade-offs for matched vs. unmatched allocation schemes when non- parametric tests of program effect are used, pros and cons of stratification or covariate adjustment instead of matching, and handling of unequal sample sizes. Available data will be used to examine the community factors that are sufficiently correlated with measures of program outcome to be candidate matching factors or covariates. For community-level variance, parametric approach to obtaining the variance and distribution of estimates will be explored, and new methods for binary outcomes will be studied. Existing data will be used for empirical estimation of community-level variance and to identify correlates of high and low estimates. Diffusion theory, borrowed from the social sciences, will be examined as a possible basis for explaining community level variation in adoption of cancer prevention modalities. New environmental indicators will be developed in concert with community- based interventions on smoking cessation in the workplace.
Showing the most recent 10 out of 15 publications
