All Times ET
Thursday, February 3
Thu, Feb 3, 1:30 PM - 3:30 PM
Virtual
Virtual
Dependent Proportions and Causal Inference
On the Noninferiority McNemar Test (305273)
*Zhigang Zhang, Memorial Sloan-Kettering Cancer CenterKeywords: McNemar Test, Non-inferiority Design, Sample Size Determination, Power Calculation
The McNemar test is commonly used for comparing two dependent proportions arising from paired or matched individuals. In this study we thoroughly examine the McNemar test under the non-inferiority setting. We show that the usual conditional test bears a systematic bias while the unconditional test does not. We also derive sample size formula for practical usage when designing trials. Both theoretical considerations and numerical studies are carried out for establishing large sample properties and illustrating finite sample performance.