Hi Andreas,
This is a beta option to support multivariate data in randomise. We've tested it internally but it has never been used 'in anger'.
With the --permuteBlocks option specified the blocks -e option define multivariate observations and then all permutation operations will keep those sets of observations together and in the same order. As with traditional multivariate analyses, there is no support for any missing data (i.e. the exchangeability blocks specified with the -e option must be perfectly balanced... same number of observations in each multivariate observation.)
The motivation was for a 2-group longitudinal design where there were exactly K>2 measures for each subject, and n1 and n2 subjects in each group (n1 & n2 not necessarily equal), and the interest was in comparing whether the temporal profile of the 2 groups was identical. The test for this effect is a F-test, comprised of either K t contrasts, testing whether the profile is different.
E.g. for K=3, n1=2, n2=3 the exchangeability block file and design matrix and the would be
1 1 0 0 0 0 0
1 0 1 0 0 0 0
1 0 0 1 0 0 0
2 1 0 0 0 0 0
2 0 1 0 0 0 0
3 0 0 0 1 0 0
3 0 0 0 0 1 0
4 0 0 0 1 0 0
4 0 0 0 0 1 0
5 0 0 0 1 0 0
5 0 0 0 0 1 0
and the F-contrast would be
-1 0 0 1 0 0
0 -1 0 0 1 0
0 0 -1 0 0 1
With the --permuteBlocks option, there are (n1+n2)-choose-n1 possible permutations; here 5-choose-3 = 5! / 2!3! = 20. Note that I haven't included the usual set of (here 5) subject indicator EV's. The reason is that including them won't change the estimates of any of the 3 COPEs that are part of the F-test, and amount they would reduce the residual sums of squares would be identical for every permutation, and thus won't alter permutation inferences.
If anyone wants to give it a spin I'd be interested to see how they fare.
-Tom
On Fri, Dec 23, 2011 at 12:21 AM, Andreas Bartsch
<[log in to unmask]> wrote:
what is the beta option of
--permuteBlocks permute exchangeability blocks
in randomise for?
Cheers,
Andreas
--
__________________________________________________________
Thomas Nichols, PhD
Principal Research Fellow, Head of Neuroimaging Statistics
Department of Statistics & Warwick Manufacturing Group
University of Warwick, Coventry CV4 7AL, United Kingdom
