Sample Size Analysis and Issues About No-Perfect Matched-Controls for Matched Case-Control Study
Chunyan Liu · OhioLink ETD Center (Ohio Library and Information Network) · 2006
Answering the following two questions is the major aim of this thesis: what is the appropriate sample size for a case-control study, and what should we do if the available matched-controls to a case group are not enough to give the analyses the required power?To address the first question, power simulations are used.Mantel-Haenszel and likelihood ratio methods are the best for non-matched data.For pair-matched data, Schlesselman's method without independence assumption is recommended.With regard to multiple-matched data, the method recommended for pair-matched data incorporating Ury's rule is suggested.After applying the conditional and unconditional logistic regression to the test data set, the following are the suggested alternatives to the second question: 1) bring in more cases to the case group and do M:1 matched conditional logit analyses.2) use all the controls and do the unconditional logistic regression adjusting for the original matching variables.