Dear Bertram and Rik,
I have never used nor read about the "flexible factorial model" in
SPM, so I apologize if my comment is silly.
Since the design matrix Bertram presented is not of full column rank,
I presume the model must include constraints (specifically for this
model, that the sum of the coefficients for columns 9 and 10 equals
zero, and the the sum of the coefficients for columns 11 and 12 equals
zero). If so, then if one wishes to compare the two groups (and by
this I presume Bertram means to compare the averages over conditions
of the two groups, and not the difference between conditions of the
two groups), then the contrast weightings on the coeffcients of
columns 9-12 are superfluous (again, unless I am simply wrong here in
my assumptions about how the flexible factorial model is working). It
seems to me that the correct contrast to compare the groups (and I am
assuming the model knows which error terms to use for between versus
within subject contrasts) would be what Bertram stated but with zeros
for entries 9 - 12.
As an aside, strictly speaking the weights in entries 9 - 12 of
Betrams's contrast weights vector should contribute zero to the value
of the contrast (because of the constraints), but perhaps the
numerical algorithm ends up with problems in computing the variance of
such a contrast.
Sincerely,
Eric
Quoting Rik Henson <[log in to unmask]>:
> Apologies, but I don't know precisely why you have such strange
> results. Are you sure you entered the scans correctly (ie didn't select
> repeat scans by mistake)? BTW, you also have a very weak design, with
> only 16 scans and hence 6dfs (rank(x)=10). Did you try to esimate
> nonsphericity? If not, I think there must be another mistake somewhere,
> because I have never seen such errors.
>
> BTW, I may not be the best person to help, because I don't use SPM5's
> 2nd-level GUI (if you want batch scripts for general within/between
> ANOVAs, try this
> http://www.mrc-cbu.cam.ac.uk:8080/people/rik.henson/personal/batch_spm_anova.m), and haven't yet matured to SPM8, so cannot answer your question about what
> has
> changed.
>
> Rik
>
> Bertram Walter wrote:
>> Dear SPMers, dear Rik,
>>
>> I have used flexible factorial for a 2 (between subjects) by 2
>> (within subjects) factorial design. Each group consisted of 4
>> subjects. This resulted in the following design:
>>
>> First 8 columns are subject's constants, last 4 columns regressors
>> for the repeated measures in the two groups.
>>
>> 1 0 0 0 0 0 0 0 1 0 0 0
>> 1 0 0 0 0 0 0 0 0 1 0 0
>> 0 1 0 0 0 0 0 0 1 0 0 0
>> 0 1 0 0 0 0 0 0 0 1 0 0
>> 0 0 1 0 0 0 0 0 1 0 0 0
>> 0 0 1 0 0 0 0 0 0 1 0 0
>> 0 0 0 1 0 0 0 0 1 0 0 0
>> 0 0 0 1 0 0 0 0 0 1 0 0
>> 0 0 0 0 1 0 0 0 0 0 1 0
>> 0 0 0 0 1 0 0 0 0 0 0 1
>> 0 0 0 0 0 1 0 0 0 0 1 0
>> 0 0 0 0 0 1 0 0 0 0 0 1
>> 0 0 0 0 0 0 1 0 0 0 1 0
>> 0 0 0 0 0 0 1 0 0 0 0 1
>> 0 0 0 0 0 0 0 1 0 0 1 0
>> 0 0 0 0 0 0 0 1 0 0 0 1
>>
>> The contrast for the comparison of groups is
>> 0.5 0.5 0.5 0.5 -0.5 -0.5 -0.5 -0.5 1.0
>> 1.0 -1.0 -1.0
>>
>> Using an F test for this contrast gave an spmF image with negative F values.
>>
>> As the test compares only two groups I calculated also a T test for
>> this contrast. The resulting spmT image contained only zeros.
>>
>> What was wrong? Using spm8b for the same design and tests gave
>> reasonable results.
>>
>> BTW: The beta maps for the first 8 regressors (subjects) were
>> slightly different in spm5 and spm8b. The remaining 4 beta maps
>> seem to be indentical. What has been changed in the model
>> estimation process?
>>
>> Cheers and
>> thanks in advance
>>
>> Bertram
>>
>
> --
>
> -------------------------------------------------------
> Dr Richard Henson
> MRC Cognition & Brain Sciences Unit
> 15 Chaucer Road
> Cambridge
> CB2 7EF, UK
>
> Office: +44 (0)1223 355 294 x522
> Mob: +44 (0)794 1377 345
> Fax: +44 (0)1223 359 062
>
> http://www.mrc-cbu.cam.ac.uk/people/rik.henson/personal
> -------------------------------------------------------
|