I should add that I'm speaking from the general standpoint of how the GLM works (so for a FEAT GLM, R GLM, SPSS GLM). Randomise with the -D option is different as it implicitly adds the column of 1s. Cheers, Jeanette On Mon, Mar 18, 2013 at 12:09 PM, Jeanette Mumford < [log in to unmask]> wrote: > Yes, there's a huge difference, one model has 2EVs and the other has 1. > This will always indicate models are not equivalent. This is also a > different problem than what you asked originally as I thought you had a > group covariate as well. Your second model would need a column of 1s to be > equivalent to the first. > > Without the column of 1's you're assuming your two group's means are > perfectly centered about 0. > > Cheers, > Jeanette > > > > > On Mon, Mar 18, 2013 at 11:11 AM, Vincent Koppelmans <[log in to unmask]>wrote: > >> Hi Jeanette, >> >> Thanks for your help. >> >> So is there any difference in the outcomes of the following two designs? >> >> A) >> Group A = 0 1 >> Group B = 1 0 >> >> Design.mat >> 0 1 >> 0 1 >> 0 1 >> 1 0 >> 1 0 >> 1 0 >> >> Design.con >> 1 -1 >> -1 1 >> >> >> B) >> Group A = -1 >> Group B = 1 >> >> Design.mat >> -1 >> -1 >> -1 >> 1 >> 1 >> 1 >> >> Design.con >> 1 >> -1 >> >> Thanks, Vincent >> >> >> Op 18 mrt. 2013, om 12:01 heeft Jeanette Mumford < >> [log in to unmask]> het volgende geschreven: >> >> Hi, >> >> You're on the right track. Add a covariate that is 1 for left and -1 for >> right. Assuming Ev1=group1 EV2=group2 and EV3=handedness, the >> [ 0 0 1] contrast will test for when left>right >> [0 0 -1] contrast will test for when right<left >> >> No collinearity will occur if you limit yourself to these 3EVs and you >> can get at all inferences of interest. >> >> Cheers, >> Jeanette >> >> On Mon, Mar 18, 2013 at 10:20 AM, Vincent <[log in to unmask]> wrote: >> >>> Dear FSL experts, >>> >>> When you create a design matrix using for example "design_ttest2 design >>> 7 11" you will get a 2 by 18 matrix. The two columns both represent group >>> membership, which are both needed in order to find out where group1 > >>> group2 and where group2 > group1. >>> >>> However, when I enter a dichotomous covariate of no interest (e.g. >>> handedness), I do not need to create two columns. Creating one column would >>> be sufficient to model out the effect of handedness. Is that correct? >>> >>> If I would be interested for which voxels left > right, and when right > >>> left, then I should create two columns, just like the group variable >>> created by design_ttest2 (and change the the design.con file accordingly). >>> Right? >>> >>> Now if I consider the last scenario, where there are two columns for >>> handedness, but that are entered as covariates of no interest, would that >>> result in multicolinearity? How does randomise handles this? >>> >>> Thanks, >>> >>> Vincent >>> >> >> >> >