Number of items falling into each condition
Posted on 03/07/10 16:42:04
Number of posts:
Dear PLS experts,
I have used PLS for E.R.FMRI data for which three conditions A,B,C with different number of items 12, 14, 50 (in average between subjects) for each condition were specified.I am wondering whether number of items fall into each condition might cause one of the significant LVs . When I look at significant LVs, non of the them reflect a patern like C vs. A&B which could be in line with the number of items. Is this the right way of checking the potential effect of number of items? Does that mean that number of items do not have any effects at all?
Does PLS as robust as conventional method (SPM) handelling different number of items fall into different conditions?
Interestingly, when I reduce the number of items associated to Condition C to something around A,B (e.g.17), I got roughly the same pattern as it was before. Can anybody explain this in more detail?
Sincerely,
Kambiz Rakhshan
PHD
University of Oslo
email:kambiz_rakhshan@yahoo.com
Untitled Post
rmcintosh
Posted on 03/08/10 08:45:58
Number of posts: 394
Certainly having a wide range of event numbers per condition will cause some problems in the estimation in the sense that the conditions with more events will have more reliable means. If you get no reliable patterns that reflect the difference in event numbers then you can get assured that it is a minimal effect. You can test it directly by using a non-rotated PLS and putting in the explicit contrast, which in your case would be something like -1 -1 2 for C vs A+B.