Dear Nick,
So sorry I didn't see this... March was rough and it got lost in the pile.
While it is not documented anywhere, 'exact permutation tests' are not possible with even a single nuisance regressor. There have been a number of approximate methods proposed (I'm almost ready to submit a paper summarizing them) and SnPM implements one such approximate method. So, the answer is: SnPM won't stop you from including an arbitrary number of nuisance regressors (as long as you have 1 or more DF), but the validity of the test will degrade with more and more regressors. I'd base it on DF... as long as you have 20 or more DF, you can 'take liberties', but if you don't have much data, I'd use the fewest number of nuisance regressors you can get away with.
-Tom
On Wed, Mar 28, 2012 at 8:42 PM, Nicholas Dufour
<[log in to unmask]> wrote:
Hello All-
Is it possible to include an arbitrary number of nuisance regressors when performing a simple regression in SnPM (as is possible when doing a multisubject T-test for instance).
Best,
Nick
--
__________________________________________________________
Thomas Nichols, PhD
Principal Research Fellow, Head of Neuroimaging Statistics
Department of Statistics & Warwick Manufacturing Group
University of Warwick, Coventry CV4 7AL, United Kingdom