Hi,
I have been plaaning for an analysis based around a simple linear regression
with one covariate (so y=mx+c), with gaussian error distribution with
variance sigma_y. The hypothesis test framework is based around testing m=0.
I have looked for references for calculating a sample size for this and
haven't been able to. I have worked it through and come up with:
N=(U_alpha/2+U_beta)^2 * sigma_y
-------------------------------------------------
theta_r^2 * sigma_x
where: U_alpha/2 and U_beta are the standard normal statistics for the size
and power of the test
sigma_y is the response variance, as above
theta_r is the 'clinically relevant' gradient to be detected
sigma_x is the sample variance in the covariate
Has anyone come across this before; is it correct?; or does anyone know of
any references for this type of question.
Thanks,
Jim Price.
___________________________________________________________________________
AstraZeneca R&D Charnwood
Clinical Science
Bakewell Road, Loughborough, Leics LE11 5RH, England
Tel: +44 (0)1509 645298 Fax: +44 (0)1509 645591
[log in to unmask]
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|