 |
 |
Hello, I have some questions about flexible factorial design specification, contrasts and interpretation of results on my project. I have two groups of healthy individuals performing a block task with 3 conditions (A, B, C) before and after administration of drug (14 subjects) or placebo (13 subjects). In a first level analysis I have created contrast images for comparisons I am interested in: A - B, B - A, B - C. My principal question is if the difference between before and after (Before - After and After - Before) is different depending on the treatment (drug or placebo) for each comparison. To do so, I have runned a second level analysis using a flexible factorial design according to this thread (https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind1405&L=spm&P=R101250&1=spm&9=A&J=on&X=63DB0725DB224B368C&Y=nandapalhano%40gmail.com&d=No+Match%3BMatch%3BMatches&z=4): Factor 1: Subjects - Independence: yes; Variance: equal Factor 2: Treatment (2 levels) - Independence: yes; Variance: unequal Factor 3: Time (2 levels) - Independence: no; Variance: equal Main effect (1), (2), (3). Interactions (2,3). I have done this only for one comparison (A-B). Now I have the following questions: 1) Can I do 3 separated models, one for each comparison I am interested in? Or, I have to include a novel factor in my previous model, e.g. Conditions (3 levels) even if I am not interested in interactions between them? 2) Considering my previous model, I have created the F contrast [ 0 0 0 0 1 -1 -1 1 zeros(1, 27)] , to investigate the interaction between treatment and time. As this contrast do not answer my question, I have created 4 additional T contrasts in order to determine the direction of the effects I observed in the interaction: - Drug Before > Drug After: 0 0 1 -1 1 -1 0 0 zeros(1,27) - Drug After > Drug Before: 0 0 -1 1 -1 1 0 0 zeros(1,27) - Placebo Before > Placebo After: 0 0 1 -1 0 0 1 -1 zeros(1,27) - Placebo After > Placebo Before: 0 0 -1 1 0 0 -1 1 zeros(1,27) Nevertheless, I am not sure how to combine the contrasts to extract the info about the direction of effects. Should I select the F contrast, and then use the other 4 as a mask? Do I have to combine the T contrasts in pairs? So, would it be 4 combinations to test? How do I know if some cluster I am observing at the interaction level is due a greater change in a group than in other? I am particularly interested in know if (DrugAfter > DrugBefore) >> (PlaceboAfter > PlaceboBefore). 3) I understand that creating the F contrast [ 0 0 0 0 1 -1 -1 1 zeros(1, 27)] , I am looking for (DrugBefore - DrugAfter) - (PlaceboBefore - PlaceboAfter). So, if I want to know if (DrugAfter - DrugBefore) - (PlaceboAfter - PlaceboBefore) I have to multiply the F contrast by -1? Is it necessary to mask with the above T contrasts? 4) Mathematically, why I have very similar but not equal result in the contrast (Drug Before > Drug After) in the flexible factorial design compared with a paired t-test? 5) Why the contrast [0 0 0 0 -1 1 0 0 zeros(1,27)] is a invalid contrast? I was wandering if this would correspond to the paired t-test. 6) I have a behavioral variable that I want to include in the model to look for possible correlations. My question is whether the condition (A - B) is correlated with this variable (an index measuring how well the subject performed the task). Is this possible in the Flexible Factorial design? Or should I run separated models, e.g. one sample t-tests for each group, contrast image (A - B) and add this variable as a covariate? Thanks very much for any help, Fernanda