Hi Roberto
> Dear all,
> this must be a basic question for must of you, but I need your qualified
> opinion.
well let see
> I am running an spm analysis with a cannonical 2 by 3 design, and I wonder
> which of these two approaches is the appropriate:
> After specifying the matrix design with six conditions (i.e., A1, A2, A3, B1,
> B2, B3) I might run for the first level analysis:
> A1: 1 0 0 0 0 0
> A2 0 1 0 0 0 0
> . 0 0 1 0 0 0
> . 0 0 0 1 0 0
> . 0 0 0 0 1 0
> A6 0 0 0 0 0 1
> which represents the averages on each condition per subject
> at the second level, then, I might contrast, for example, A1>B1 with the *com
> images in a full factorial anova
> the second approach would be doing direct contrasts at the first level:
> E.g, A1>B1 1 0 0 -1 0 0 per subject, and then, performing
> per-specific-contrast one sample t tests at the second level with all the
> respective subject's com images.
> which of these two approaches is the most appropriate? I think the second one
> is more sensitive per specific comparisons while keeping the advantages of
> the random effects test. is not it?
both approaches are valid and correspond to e RFX analysis, now you may gain
some sensitivity with approach 2 but loose flexibility - to my opinion the best
is to set up the full factorial anova as you can then look at A (111) B
(000111) obvisously A vs B, at all simple effects A1 vs B1, A2 vs B2 and B3 vs
B3 but more importantly at the interactions between these factors eg A1>B1 but
A2<B2 .. (1-10-110) - I guess it really depends on what question you want to
ask - if your interest is always in A>B then yes compute for each subject
A1>B1, A2>B2 and A3>B3 then enter those 3 contrasts in an ANOVA ...
hope this helps
cyril
------------------------------------------------------------------
Message sent via Psychology Dept. WEB-Mail Gateway.
http://www.psy.gla.ac.uk/
------------------------------------------------------------------
|