I am trying to determine the optimum FLAME design for analyzing a
multi-subject, multi-session (e.g. pre & post treatment) multi-group (e.g.
treatment 1, treatment 2, control) & multivariate (difference between
treatment 1, 2 & control; and difference between treatment success) data
set. For example, all subjects were scanned under identical conditions
prior to and following treatment. A small portion of each group was
responsive to treatment while the rest were not. I have attached a text
file containing a condensed example of the current design I am using for
your reference.
There are only a limited number of successful responders in each group (e.g.
Cont only has 1 successful responder). It appears from the results that
each group is being weighted equally, which is causing the smaller groups
(e.g. N=1 for Cont-success) to over influence the results.
Is there a different design that will account for this error? Or, is there a
way to assess each variate independently (i.e. effect of treatment type;
effect of success) while controlling for the effects of the alternate variate?
Thanks for the help
Chris
EV1 EV2 EV3 EV4 EV5 EV6 EV7 EV8 EV9 EV10 EV11 EV11...
Pre Tmt 1/success 1 1
Pre Tmt 2/success 1 1
Pre Cont/success 1 1
Post Tmt 1/success -1 1
Post Tmt 2/success -1 1
Post Cont/success -1 1
Pre Tmt 1/no success 1 1
Pre Tmt 2/no success 1 1
Pre Cont/no success 1 1
Post Tmt 1/no success -1 1
Post Tmt 2/no success -1 1
Post Cont/no success -1 1
Contrasts EV1 EV2 EV3 EV4 EV5 EV6
success pre-post 1 1 1 0 0 0
success post-pre -1 -1 -1 0 0 0
no success pre-post 0 0 0 1 1 1
no success post-pre 0 0 0 -1 -1 -1
success - no (pre-post) 1 1 1 -1 -1 -1
no - success (pre-post) -1 -1 -1 1 1 1
Tmt1 pre-post 1 0 0 1 0 0
Tmt2 pre-post 0 1 0 0 1 0
Cont pre-post 0 0 1 0 0 1
Tmt1-Cont 1 0 -1 1 0 -1
Cont-Tmt1 -1 0 1 -1 0 1
Tmt2-Cont 0 1 -1 0 1 -1
Cont-Tmt2 0 -1 1 0 -1 1
...
|