Dear Eugene,
Thanks so much for your message. This seems clear, I just have 2 questions.
For the 22 columns that you suggested where I would compare subject means, I just want to clarify: the first one would have 1 for the first (top) box, -1 for the second (below), and 0s for all the other boxes in that column? And the second would have 1 for the first, 0 for the second, and -1 for the third box, and then all 0s? Then the same pattern for the second 11, except starting in the lower hald (#37) for group number 2? (I have 24 subjects, 2 groups, and 3 scans each, so there are 72 first-level feat entrys.) Just wanted to clarify this.
Secondly, for the interaction column, I don't understand how to set that up, where to the 1s and -1s go? Just one column to determine interaction between scan and group?
Thanks so much for your help,
Katie
------------------------------
Date: Mon, 14 Mar 2011 16:59:46 +0000
From: Eugene Duff <[log in to unmask]>
Subject: Re: repeated measures anova in feat
Hi Katie,
Actually it's a bit more complex than that, as you have a within-subject and an across-subject factor. Closest example is 2-factors 2-levels, but you need to model subject means rather than one overall mean. One way to do it:
1 column modelling difference between groups
2 columns modelling differences between session 1&2 and 1&3
11 columns modelling differences between subject 1 and each of the other subjects in group 1
11 columns modelling differences between subject 12 and each of the other subject in group 2 and 1 column for an interaction if you want.
F-test on column 1 contrasts for group effect
F-test across column 2&3 contrasts for session effect
You need to do it this way because you can't model subject means directly because this will also fit any group differences.
If you want the F-tests there's no way of getting away from all those 0s and 1s I'm afraid! I think you could get just the differential contrasts by modelling mean and session effects in subject level models.
Let me know if this is unclear.
Cheers
Eugene
original post:
Hello, I would like to set up a feat analysis and I can not find an example that quite fits. I have 2 groups (12 in each group) and 3 separate scans. I have conducted inidividual first-level analyses on all 72 scans, and would like to evaluate for (a) group effect, (b) scan effect, and (c) group by scan effect. As I look at the examples, it seems I may need to have a separate EV for each person, so it seems I will have 72 rows, and columns = 24 people + 2 group EVs + 3 scan ID EVs = 29 columns. Does that seem right? I haven't figured out the group by scan interaction EVs yet. Is there a simpler way to set this up, by chance? It will be quite diffucult to manually fill in all these 0s and 1s.
Thanks so much for your help, Katie
|