Hi
>
> 1. Can I do a mixed effect analysis using fsl_glm ( I guess no). I
> can use fsl_glm to conduct one sample t-test of the parameter
> estimates (betas) obtained from the regression (fsl_glm) of subject
> specific design matrix with a components of interest. However, this
> only gives me a fixed effect result right?
>
Not quite, you're right in that there is only a single between-
subjects variance included in the analysis and no combination of lower-
level (within session) and between session /subject variance.
Therefore, as you say, this is not a mixed-effects analysis. However,
the variance that is accounted for is the between session/subject
variance - this therefore corresponds to a simple random effects
analysis.
>
> 2. Whe using fsl_glm for regressing against design matrix, do I use
> variance normaised ICA time course from 't...txt' or raw timecourse
> from 'melodic_mix'.
This does not matter as soon as you convert our between subject
estimates to Z-stats (using the --out_z)
> Is it valid to normalise model time course from design matrix to
> unit variance and regress it with unit variance normalised ICA time
> course?
Absoutely, the parameter estimate then is the correlation.
>
> 3. If I regress my design matrix with timecourse of a ICA component
> of interesst from Melodic_Mix using fsl_glm, I get large beta values
> (>200). If I normalise variance of both timecourse to unit variance,
> I get low beta value (>0<1).
Yes, prior to normalising this parameter estimate is in terms of raw
image intensities. If you divide by the mean intensity of the voxels
within the associated spatial maps this then becomes a %BOLD estimate.
In the second case you get a correlation score
> Don't know if both design timecourse and ICA timecourse need to be
> normalised to unit variance before regressing using fsl_glm.
>
Both is valid, as soon as you perform some significance test (e.g.
test for difference between groups or test for significant average
deviation from 0) everything should come out in a wash and you should
be getting the same result.
hth
Christian
> Regards,
> Govinda
>
>
>
> On Sat, Sep 27, 2008 at 10:40 AM, Christian F. Beckmann <[log in to unmask]
> > wrote:
> Hi
>
> you could try running fsl_glm on these parameter estimates - the
> software also does accept ascii text matrices as input and does not
> only worl on nifti files
> cheers
> Christian
>
>
>
> On 22 Sep 2008, at 00:38, Govinda Poudel wrote:
>
> Hi,
>
> I have regressed subject specific design matrix with subject ICA
> time course (from melodic mix) and obtained parameter estimates for
> individual subjects. I was wondering what the best option is to
> obtain group stat from these individual subject parameter estimates.
>
> Regards,
> Govinda Poudel
>
>
|