 |
 |
Dear Tom, the contrast looks actually fine to me. But you wrote the following: > > Then I entered the subjects scans using the order: G1T1 G1T2 (conditions: > 1 1; 1 2), followed by G2T1 G2T2 (conditions: 1 1; 2 2). The second bracket should be (conditions: 2 1; 2 2). Have you specified this correctly in your code? As a quality check (or actually an alternative way to test the interaction), you could calculate the contrast [T2 - T1] for each subject on the first level, and then enter the con-images into a two-sample t-test on the second level (with the factor group). This should in any case give you a valid test of your interaction. To clarify the issue with the zeros and ones: this has to do with estimability of contrasts in overparameterized models. It is difficult to give an easy explanation, but maybe this presentation by Dr. McFarquhar can help you: http://www.sbirc.ed.ac.uk/cyril/SPM-course/Talks/2017/11-Contrasts_MMcFarquhar.pdf However, the way you specified your model circumvents this problem. (You specified the main effect for subjects, and the two-way interaction between your other factors. This is fine. Overparameterization is an issue when you would also specify the main effects... which, to the best of my knowledge, does nothing but complicate the specification of the contrasts later). Best, Lukas Am Mi., 25. März 2020 um 13:43 Uhr schrieb tom parker < [log in to unmask]>: > Dear SPMers, > > I am a newbie to SPM so I need to ask you for some advice. > I have 2 groups of subjects that have been scanned twice. > The second group should show more brain changes over time compared to the > first one (the one way ANOVA within subject also showed more changes in the > second group). > > To test this hypothesis, I have used a Flexible Factorial Design in SPM > with 3 factors: > - 1st factor: subject (independence: yes, variance: unequal) > - 2nd factor: group (independence: yes, variance: unequal) > - 3rd factor: time (independence: no, variance: unequal) > > Main effects & Interactions: > Interactions: 2 3 > Main effect: 1 > > Then I entered the subjects scans using the order: G1T1 G1T2 (conditions: > 1 1; 1 2), followed by G2T1 G2T2 (conditions: 1 1; 2 2). > I have also added 5 covariates. Please see attached a snapshot of my > design. > > To define the contrast Group*Time, I used an F-contrast and included 1 -1 > -1 1 in columns for reduced design. > I did this because whichever contrast I wrote in contrast weights matrix, > it said it was not valid. The results using this reduced contrast did not > make sense to me. > > Searching for information on how to define contrasts in this type of > design I saw that people write things like ones(1,3) or zeros but, to be > honest, I have no idea of what these ones and zeros are or how I should > define them. I have 123 subjects * 2 time points in the first group and 35 > subjects * 2 time points in the second group. > > Could someone help me to define the Group*Time contrast for my analysis? > > Thank you so much! > > -- Mag. Lukas Lengersdorff Universität Wien Fakultät für Psychologie Institut für Grundlagenforschung und Forschungsmethoden *Social, Cognitive and Affective Neuroscience Unit* [log in to unmask]