Hi again,
You are right that the results look more meanigful if I treat the
factors as fixed effects. However, I assumed that my factors (A and B)
were random effects, because all four levels (A1, A2, B1, B2) were
measured in each subject. Is that not correct?
Regards,
Hanne
Steve Smith wrote:
> This depends - it is more likely that your factors are "fixed effect
> factors" (which is a completely separate issue to the choice of mixed
> effects cross-subject variance modelling that you correctly chose in
> the FEAT GUI). If this is the case then you don't need to take those
> ratios - the 3 F stats and derived Zfstat images that FEAT gives you
> are already what you want. Maybe this will then give you results which
> you find more meaningful.
>
> Cheers.
>
>
>
> On 14 Jan 2009, at 13:16, Hanne Lehn wrote:
>
>> Hi,
>>
>> I have a question about ANOVA analysis in FEAT.
>>
>> I have just run a 2x2 ANOVA on my data (two factors, both repeated
>> measures and with two levels). The inputs were 76 cope files (19
>> subjects x 4
>> levels), derived from the first and second level GLM (within
>> subjects). I
>> specified the design matrix as suggested in the 2x2 example on the
>> website,
>> but replaced EV4 with 19 subject-specific EVs. This gave me a total
>> 22 EVs. I
>> chose mixed effects modelling and requested three F-tests, for the
>> two main
>> effects and the interaction effect. As suggested in the web example,
>> I divided
>> the output fstat images of the two main effects by the fstat image of
>> the
>> interaction effect, and then converted these to zfstat images.
>>
>> Is this the correct way to do it?
>>
>> The reason I ask is that the results are quite different from what I
>> expected,
>> based on the standard GLM analysis (pairwise comparisons). For
>> example, the
>> standard analysis shows a large difference between the two levels of
>> factor A
>> (A1>A2), but the ANOVA shows no main effect of factor A. The two zfstat
>> images look a bit odd, they contain scatters of single voxels, and
>> clusters
>> appear only with z 0-1.5. The original zfstat images look more
>> "normal", and in
>> line with my predictions, as does the zfstat image of the interaction
>> effect.
>>
>> If anyone knows whether and where I went wrong, please let me know.
>> Thanks in advance!
>>
>> Hanne
>>
>
>
> ---------------------------------------------------------------------------
>
> Stephen M. Smith, Professor of Biomedical Engineering
> Associate Director, Oxford University FMRIB Centre
>
> FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
> +44 (0) 1865 222726 (fax 222717)
> [log in to unmask] http://www.fmrib.ox.ac.uk/~steve
> ---------------------------------------------------------------------------
>
--
Hanne Lehn
PhD student Neuroscience
MR-Centre, St. Olav's Hospital
7006 Trondheim, Norway
Phone: (+47) 73 59 88 04
Fax: (+47) 73 86 77 08
|