On 4 Oct 2008, at 09:59, Thomas Nichols wrote: > Setting up a fourth variable would lead to a linearly-dependent > design matrix. I.e. the sum of EV2, EV3 and (your proposed) EV4 > would equal EV1. If you dropped the grand mean covariate, you could > take the approach you suggest, and, in fact, that's the dummy > variable approach taken inhttp://www.fmrib.ox.ac.uk/fsl/feat5/detail.html#FTests > . Sorry to resurrect an old thread. I'm doing similar thing as Ben at the moment and am wondering why you would choose the grand-mean approach to specifying the EVs and contrasts versus the (simpler to my mind) approach of using the dummy variable? If I under stand this correctly (assuming n=3 and there are 3 groups and no repeated measures) long mid short age 1 0 0 12 1 0 0 11 1 0 0 13 0 1 0 11 0 1 0 17 0 1 0 14 0 0 1 10 0 0 1 12 0 0 1 16 and contrasts (* indicates inclusion in an F-test) 1 0 0 0 * 0 1 0 0 * 0 0 1 0 * 1 -1 0 0 #long > mid 0 1 -1 0 #mid > short will allow me to ask the question "is there a group effect (the F- test)" and then to tweeze apart some of the possible pair-wise comparisons having accounted for variance attributable to age. Am I correct? Regards, -- Dr Colm G. Connolly Institute of Neuroscience The Lloyd Building University of Dublin Trinity College, Dublin 2, Éire Fax: +353-1-671-3183 Please note that electronic mail to, from or within the Trinity College Dublin, may be the subject of a request under the Freedom of Information Act.