Dear Kevin,
So you have a mixed design with one between subject factor with two
levels (group) and one within subject factor with two levels (time). I
would recommend you look at earlier posts from this list that explain
how to define partitioned error models to test for main effects and
interactions and will be as simple as one or two sample t-tests.
Concerning your specification of a Flexible Factorial design, the
'Conditions' entry will be [1 1;1 2] for a subject in group 1 and [2 1;
2 2] for a subject in group 2. For the 'Main effects & Interactions'
entry, you should select: main effect of factor 1 (subject) and
interaction of factors 2 and 3 (group and time). The design matrix will
have four columns for the four cells of your 2x2 factorial design
followed by N columns modelling subject effects.
Then contrasts are as follow:
main effect of group: 1 1 -1 -1
main effect of time: 1 -1 1 -1
group x time interaction: 1 -1 -1 1
(but you shouldn't put subjects effects when testing for group; look at
partitioned error models where you will first create average and
difference along time for each subject and enter these in separate
second level models).
Best regards,
Guillaume.
On 14/12/16 05:17, Kangik Kevin Cho wrote:
> Dear SPM experts,
>
> I have subjects divided into two groups, who are scanned twice (two time points).
> I'm interested in the group, time and interaction effect.
> (2 x 2 within-subjects ANOVA)
>
> I’ve got the idea from links below, to use 'Flexible Factorial’.
> However, I have not been successful in finding clear information about the 2x2 within-subject contrast in the Jiscmail.
>
> - https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind0712&L=spm&D=0&1=spm&9=A&J=on&d=No+Match%3BMatch%3BMatches&z=4&P=340817
> - https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind0712&L=spm&D=0&1=spm&9=A&J=on&d=No+Match%3BMatch%3BMatches&z=4&P=369331
> - https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind0801&L=spm&D=0&1=spm&9=A&J=on&d=No+Match%3BMatch%3BMatches&z=4&P=157697
>
>
> The design could be summarized as below. (name / dependency / variance / grand mean scaling (0) / ancova (0))
> subject / independent / equal
> group / independent / unequal
> time / dependent / equal
>
> The two scans from two time points for each subjects are entered as below
> eg)
> Scans : subj1_pre, subj1_post
> Conditions : 1 1; 1 2 (first column represents group, second column represents the timepoint)
>
> For the main effects & interactions
> - 1, 2, 3 are given separately as the main effect factor number
> - [2 3] is given as the interaction factor numbers
>
> AND For the contrast :
> (group1 subjects are entered first)
> - N1 : Number of subjects in group 1
> - N2 : Number of subjects in group2
>
> Group effect : 1 -1 0 0 1/2 1/2 -1/2 -1/2 repmat(1/N1, 1, N1) repmat(-1/N2, 1, N2)
> Time effect : 0 0 1 -1 N1/(N1+N2) -N1/(N1+N2) N2/(N1+N2) -N2/(N1+N2) zeros(1,N1+N2)
> Interaction : 0 0 0 0 1 -1 -1 1 zeros(1,N1+N2)
>
> It would be highly appreciated, if anyone with experience in within-subject ANOVA could help confirm or correct above contrast.
>
> Regards,
> Kevin
>
--
Guillaume Flandin, PhD
Wellcome Trust Centre for Neuroimaging
University College London
12 Queen Square
London WC1N 3BG
|