Master of Science
Epidemiology and Biostatistics
Dr. Neil Klar/Dr. Gerarda Darlington
In this thesis we investigate methods for assessing the intervention effect in completely randomized, cluster randomization trials where participants are measured prior to random assignment and again after completion of the intervention, i.e. a pretest-posttest design. Attention is further limited to binary outcomes. We compare the performance of six test statistics used to test the intervention effect. Test statistics are obtained from cluster-specific and population-averaged extensions of logistic regression. A simulation study is used to estimate type I error and power for the test statistics. In addition, we examine the effect on power of correctly assuming a common pretest-posttest association. Cluster-specific models yielded satisfactory 5% type I error rates while a longitudinal approach yielded the lowest power. Assumptions about the pretest-posttest association had little effect on power. Data from a school-based randomized trial are used to illustrate results.
Borhan, ASM, "Methods for the Analysis of Pretest-Posttest Binary Outcomes from Cluster Randomization Trials" (2012). Electronic Thesis and Dissertation Repository. 825.