Date of Award
Master of Science
Epidemiology and Biostatistics
Dr. Neil Klar
The principal aim of this study is to address the problem of breaking the matching when testing the independent relationship between a continuous individual-level risk factor and a binary outcome variable. We compare the performance of five test statistics for testing the effect of individual-level risk factors. These statistics are obtained from extensions of logistic regression distinguished by whether or not matching is maintained. Type I error rate and statistical power are compared for these five test statistics through a simulation study. The results show that an unmatched analysis leads to overly liberal type I error rates when the matching correlation for both individual-level risk factor and outcome are strong. However, in the presence of large-scale community intervention trials, breaking the matching to perform an unmatched analysis will be valid and efficient. The discussion will be illustrated using data from the Greenwich Asthma trial.
Zhang, Ling, "RISK FACTOR ANALYSES IN MATCHED-PAIR CLUSTER RANDOMIZATION TRIALS" (2009). Digitized Theses. 3918.