 |
 |
Hi, This isn't a correct design because of the pairing, which makes things a little more difficult to think about. If you did not have pairing of each subject for pre-post then the first three EVs would tell you about (1) pre vs post, (2) sick vs non-sick and (3) interaction. With the pairing it is different, because you are effectively doing the same thing as subtracting each subject's pre-post and just using those values (as EV4-EV8 in this example completely model each subject's mean - and effectively remove it). If you think about the possible questions that you can ask when you only have the difference of pre-post for each subject, then you can see that you can only really ask two questions: (a) if the overall mean non-zero and (b) is there a difference in means between sick and non-sick. Note that in this case the "means" are the means of the differences in pre-post for each subject. Questions (a) and (b) correspond to EV1 and EV3 in your design. EV2 is trying to ask questions about the means of the pre and post rather than the difference. For example, for subject 1 you have a 1 for both pre and post in this EV, whereas if you wanted to know about only the pre-post difference then you would need one of them to have a -1, which is what you see in EV3. So EV3 is asking about the difference in sick and non-sick, averaging the difference of pre-post in each case. I hope that is clear. So for your design you just want to delete EV2 and everything should be fine. If you really wanted to know about the actual mean of pre and post (rather than the average of the differences) when you could put a contrast on EV4 to EV8 which is 1 1 1 -1 -1. However, normally this kind of question is not useful in a paired design. As for F-test - if you put one on EV1 and EV3 then you will find out if there is any difference between either sick and non-sick or pre-post. It just depends if you are interested in this. I hope this answers your questions. All the best, Mark On 10 Jan 2008, at 22:28, SUBSCRIBE FSL Kelly Brown wrote: > Hello all FSL statisticians: > > I am working with FSLVBM and would like to create a mixed model to run > in randomise. I have two groups of subjects (sick and not sick) > that each > undergo two sessions (pre and post treatment). Does this look like the > correct design and contrast files for this type of analysis? ( have > cut the > number of subjects down in each group for ease in viewing...) > > > Grp EV1 EV2 EV3 EV4 EV5 EV6 EV7 EV8 > 1 1 1 1 1 0 0 0 0 > 1 1 1 1 0 1 0 0 0 > 1 1 1 1 0 0 1 0 0 > 2 1 -1 -1 0 0 0 1 0 > 2 1 -1 -1 0 0 0 0 1 > 1 -1 1 -1 1 0 0 0 0 > 1 -1 1 -1 0 1 0 0 0 > 1 -1 1 -1 0 0 1 0 0 > 2 -1 -1 1 0 0 0 1 0 > 2 -1 -1 1 0 0 0 0 1 > > > > 1 2 3 4 5 6 7 8 > C1 1 0 0 0 0 0 0 0 > C2 0 1 0 0 0 0 0 0 > C3 0 0 1 0 0 0 0 0 > > > Because I get an error when I view the design and the Efficiency in > the Glm gui: > > Problem with processing the model: Warning: at least one EV is > (close to) a > linear combination of the others. You probably need to alter your > design. > (Design matrix is rank deficient- ratio of min:max eigenvalues in > SVD of > matrix is 2.30159e-17) > > Warning: design matrix uses different groups (for different > variances), but > these do not contain "separable" EVs for the different groups (it is > necessary that, for each EV, only one of the groups has non-zero > values) > > Also, do you recommend then completing an F-test for main effect? > If so, how > do I put that in my contrast file? > > Thanks so much for your help! > > Cheers, > Kelly Brown > UCHSC Research Fellow >