Hi friends,
We are planning a study in which we do expect some effect to be present quite strongly, i.e. we do not only expect some difference to significantly deviate from 0 but from some number which is itself larger than 0. So our primary hypothesis is, that this effect is larger than A, where A is some justified choosen amount. However, it may occur unfortunately, that despite our measures, our sample size calculation and so on, the result in the obtained sample does not show that difference significantly. In that case we would like to be sure to show that the difference is at least significant from 0. So our secondary hypothesis is that the effect is larger than 0. This effect does of course not need to be proven if the first hypothesis turns out to be accepted.
My first Q is, whether this is a common approach to testing for differences, or, if not, what other alternatives there are to show some size of an effect.
My second Q is, in case this is a justified approach, possibly needing two very much related tests on the same data, whether such a specific situation needs to partition the alpha risk, where it is not known in advance, whether both tests will be performed, and where in a usual case only one of the tests would be done without any alpha adjustment.
Regards - Jim.
Y. (Jim) Groeneveld MSc
Biostatistician
Vitatron B.V.
Meander 1051
6825 MJ Arnhem
The Netherlands
+31/0 26 376 7365; fax 7305
[log in to unmask]
www.vitatron.com
I wish I had a webcam to show you the boring view from my window.
[common disclaimer]
|