I need to find out how many patients I need to power sufficiently my
primary analysis, which would be a Dunnett's test comparing means of
two active trt groups versus a placebo.
Does one know whether proc power or proc glmpower can do this in SAS ?
I did look into the on-line doc. The contrast statement allows for
testing general null hypothesis like L x Beta = 0 versus L x Beta
different from 0, Beta being the vector of parameters in the model Y =
X Beta + epsilon, and L being the matrix of contrasts I want tested.
The example 34.1 in PROC GLMPOWER shows a sample size determination
based on a one-way ANOVA model. My concerns are that:
- It tested a null hypothesis of 4 comparisons, with a comparisonwise
alpha level of 0.025 and this suggests that there is no adjustment for
multiplicity, while the Dunnett's test does protect the familywise
error rate on the family of all tested contrasts.
- The contrast matrix I would need to perform my multiple comparisons
to control Dunnett's test would be like:
1 -1 0
1 0 -1
and it's not of full rank. Does the proc glmpower works in that case ?
- Even if the procedure works with the above contrast matrix, does
that mean it follows Dunnet's methodology for sample size
calculations, whichy is documented in Jason Hsu's book: "Multiple
Comparisons: Theory and Methods" (I could not find that reference in
those listed in proc glmpower documentation).
Any help on how to calculate the sample size for this test is more
Thanks for your help.
||5/4/2010 3:25:34 PM