On Tue, 12 Aug 2008 18:57:54 +0100, Armin Blickenstorfer
<[log in to unmask]> wrote:
>Dear community
>
>I have 2-factorial design Desgin:
>1st factor 'Group' (1. TREAT, 2. CTRL) and 2nd factor 'Cond' (1. FES, 2. VOL).
>
>My goal is to compare the contrast of FES vs VOL in TREAT vs. FES vs VOL in
>CTRL. So basically compare the two differences.
In statistical nomenclature, this is the _interaction_ of the two factors.
>How do I do that? Is it legitimate to compute 2 t-contrasts (one for each
>group) and then use 2 contrasts as input in another t-test?
Best way is to directly make a contrast. That is, have one analysis with all
four (4 = 2*2) combinations of factor levels. Then the interaction would
simply be contrast like
1 -1 -1 1
It is important, however, to be careful about certain issues. For example,
is "condition" a within-subject factor? I.e., does every subject have both FES
and VOL? If so, then you need to take that into account.
>Thank you and best regards,
>
>Armin Blickenstorfer, MSc UZH
>PhD Student
>
>Institute of Neuroradiology
>University Hopital Zurich
>Frauenklinikstrasse 10
>CH - 8091 Zurich
>
>Tel: +41 (0) 44 255 5603
>Fax: +41 (0) 44 255 4504
|