 |
 |
Hi Simon > Hi all, > I have a probably simple question and would be grateful for your help: > I'm analyzing a repeated measures ANOVA (15 subj, each 3 conditions) > and have a > single covariate varying over subjects and conditions. > Setting up a within-subject ANOVA design using SPM8 the covariate is > represented in column 4. > Is is right to use the t-contrast (0 0 0 1) or (0 0 0 -1) to evaluate > its significance? yep that's all good > How are 'repeated measures' taken into account? (Are subject scaled > individually? not sure what you mean by scaled but as you see in your design matrix you have 15 additional regressors oni the right weigting each condition per subject - ie the within subject is taken care off. > Are there problems, if data behaves awkwardly?) > > The questions arose when I tried to confirm my rmANOVA results > (although not quite correctly) by calculating a one sample t-test > pooling data from all subjects and conditions together using the > above mentiond regressor as covariate (contrast used: 0 1 and 0 -1). > Unexpectedly, evaluating the regressor using this simplified approach > yielded completly different results. Specifically, clusters that were > highly significant in the rmANOVA desing vanished. This seems > counterintuitive to me, since degrees of freedom are (virtually) > increased when treating images from a single subject as independent > (I originally expected increased t-values). What could this imply for > the data? well the problem here is that the error terms are completly different so it's hard to say what whent wrong = rm ANOVA in SPM follows the same rules as other stats: normallity and to some extend heteroscedasticity (accounted for by setting your independence / variance options options hope this helps cyril -- The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336.