Equivalence test via non-zero rate difference in matched-pair studies with incomplete data

• 1、 Department of Statistics, Yunnan University, Kunming 650091

Abstract：Under the assumption of missing at random, four statistics and two test procedures including the approximate unconditional and Bootstrap-resampling procedures are proposed to test the equivalence of two correlated proportions in an incomplete matched-pair design. A computationally feasible EM algorithm is presented to evaluate the unconstrained/constrained maximum likelihood estimations of parameters. Extensive simulation studies show that the approximate unconditional and Bootstrap-resampling methods behave satisfactorily in terms ofthe type I error rates, and the approximate unconditional score test procedure is highly recommended because its type I error rate is well controlled and its computational burden is not heavy. An example is used to illustrate the proposed procedures.

keywords： Approximate unconditional test Bootstrap-resamping test Equivalence Matched-pair design Missing data

点击查看论文中文信息

点击收起