The proposed project seeks to advance knowledge about the design, conduct, analysis, and interpretation of community-based interpretation studies for cancer control. In these studies: (1) aggregates of individuals, such as entire communities, are allocated en bloc to intervention or control groups; (2) a relatively small number of such aggregates are studied; (3) there is heavy reliance on survey methods to assess behavioral change outcomes; and (4) blinding of individual participants to their treatment group membership is usually impossible. These features pose special methodological challenges. Several issues concerning experimental design and data analysis will be addressed. First, statistical methods and computer software will be developed for calculation of sample size, power, and smallest detectable program effect in relation to various design parameters, such as the number of communities studied and the number and timing of surveys. Second, empirical estimates of parameters needed to apply such methods will be obtained from many existing sources. Third, methods for allocation of communities to treatment groups will be compared. Fourth, the relative merits of cohort and repeated cross-sectional surveys will be compared, partly through analyses of data from ongoing and completed studies. Fifth, methods for analysis of unbalanced designs will be reviewed and compared when applied to real data. In addition, two broad measurement problems arising in such studies will be investigated. First, we will attempt to conceptualize and to develop community-level outcome measures for use in such studies, and pilot test candidates in the field. Second, the validity of survey-based behavioral measures will be evaluated in the context of studies which do not permit blinding, chiefly by examining their relationship to physiologic """"""""gold standards"""""""" across intervention and control groups.
Showing the most recent 10 out of 15 publications
