 |
 |
Hi Carsten, > I would very much appreciate comments of experienced SPM statisticians I'm not one of those, but since it appears no-one else has replied (at least not on-list) I'll say something. The design matrix columns should include either a mean (constant) column and an extra effect for one of the groups, or two separate group-mean columns. I think your ANCOVA model sounds okay in this respect, but your first sounds like it assumes that one of the groups is zero mean (apart from the covariates), which I don't think is right. I'm afraid I don't know what you mean when you refer to the error columns -- the error is what cannot be modelled by the design; it is not part of the design matrix. Perhaps this is a constant/mean term? In either case, a contrast with a 1 over your other covariate should indeed test for significant correlation of that with the images you select (e.g. GM probability). I would expect the results of the correlation model and the ANCOVA model to be the same (as long as the correlation one includes two group means or an overall mean and an additional group effect). Best, Ged. > An investigation of two groups was performed using MRI and optimized VBM > (n=26). In addition, other data were acquired in both groups. Statistically > spoken we investigate the variable gray matter probability in each voxel, > with knowledge of the categorial group factor, the continous whole brain > volume factor, and the continous other factor. > > Our first approach was to build a multiple correlation. We chose the simple > correlation (regression) model in SPM and loaded a text-file containing the > three factors group in three lines. This leads to a SPM matrix with the four > rows row 1:group (0 and 1), row 2:whole brain volume (continous), row3:other > factor (continous), and row4: error. In the results step, a contrast was > defined putting +1 on the other factor (row3). In our opinion this should > model the correlation of the other factor with gray matter probability. Is > this thought correct? > > Second there was a discussion about using an ANCOVA model. Definition of the > ANCOVA with both groups and the two covariates whole brain volume and other > factor leads to a SPM design matrix with five rows. Row 1: group 0, row > 2:group 1, row 3: error, row 4: whole brain volume (continous) and row 5: > other factor (continous). However we are not sure about the significance of > putting a contrast +1 on the other factor (row 5). Does this kind of > analysis tell us anything about the relation of the other factor with gray > matter probability, and what exactly? > > Yours > Carsten >