|
Activity Number:
|
233
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 3, 2009 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Biopharmaceutical Section
|
| Abstract - #305041 |
|
Title:
|
Sample Size and Power Investigation for Exact McNemar Test
|
|
Author(s):
|
John D.S. Hwang and James Lee*+
|
|
Companies:
|
B.R.S.I. and Daiichi Sankyo Pharma Development
|
|
Address:
|
399 Thornall St., Edison, NJ, 08837,
|
|
Keywords:
|
Exact McNemar Test ; binary response data ; repeated measurement design ; sample size ; power
|
|
Abstract:
|
For a matched pair or repeated measurements design with 2 time points, exact McNemar test may be employed for the binary response data. Determination of sample size or power provides a challenge since the binomial distribution depends on a parameter Pr = P01 / (P01 + P10). In this paper Pr is found to be a function of correlation r and odds ratio which are easier to specify. This functional relationship also helps to numerically assess conditions under which a paired design is preferable to an unpaired design, as in the Normal distribution case. The weighting function employed in evaluating unconditional power is found to be another binomial distribution whose parameter is a function of the multinomial parameters Pij. Formulae expressing Pij as function of marginal parameters P1, P2 and r are developed which enables the evaluation of unconditional power and sample size determination.
|
