Thanks, Jeanette. This makes sense.
Since the highest (3rd) level feat deals with heteroscedasticity across subjects. I will be using mixed effects for this level. I am assuming dealing with heteroscedasticity is handled in ME GLM.
Thanks again,
- BettyAnn
> Hi,
>
> It is a weighted linear regression, where the weights are the lower level varcopes.
> So the second level cope (average) is a weighted average. If s_i is the sqrt of
> the first level varcope for each run, the second level variance (varcope) for the
> weighted average is:
>
> 1/ SUM_i (1/s_i)
>
> Where SUM_i is the summation over each run, i.
>
> I wouldn't toss a subject for one bad run, just omit the bad run(s) and use the
> good ones. The higher level (3rd level) feat analysis is designed to handle
> heteroscedasticity (unequal variances across subjects).
>
> Cheers,
> Jeanette
>>
>> On Tue, Feb 5, 2013 at 10:52 PM, bettyann <[log in to unmask]> wrote:
>> Hello,
>>
>> I am trying to find and understand how FSL/feat computes fixed effects GLM.
>> I am currently reading:
>>
>> Comparing groups of subjects in fMRI studies: a review of the GLM approach.
>> http://www.fmrib.ox.ac.uk/analysis/techrep/tr00dl1/tr00dl1.pdf
>>
>> Are there other technical reports or papers I should read?
>>
>> What brought me to this is as follows:
>> Our subjects do four runs of the same task from which I create a mean using FE
>> GLM. If a subject has 'too much motion' in any one run of the four, data from
>> all four runs are discarded.
>>
>> I have been discarding all four runs because I was concerned that the SNR would
>> be different depending on how many runs go into the FE GLM.
>>
>> Which led me to wonder how the betas (copes) and their variances (varcopes) are
>> computed from the FE GLM. Are the variances from the FE based on the variances
>> from the inputs, ie, the individual four runs (since each run has its own beta
>> and variance)? Or is the variance from the FE the variance of the beta inputs
>> to the FE GLM?
>>
>> Can anyone point me toward any information to understand how the FE is done?
>>
>> Thanks,
>> - BettyAnn
|