Thank you very much Donald. 

Do you think this command line is correct?: 
 fsl_glm -i GM_mod_merg_s3 -d design.mat -c design.con -m GM_mask --out_t=t_scores

This gives me a single image for the whole design though (which includes 6 contrasts). My question is how can I get the t-scores for only one contrast ?

Sinan

 
On Wed, 25 Jan 2012 23:24:12 -0500, MCLAREN, Donald <[log in to unmask]> wrote:

>See inline responses.
>
>On Wed, Jan 25, 2012 at 9:51 AM, Sinan Dirik <[log in to unmask]> wrote:
>
>> Dear FSL users and experts,
>>
>>    Sorry for posting a similar question in a short interval, but I am
>> running out of time and couldn't find a solution yet.
>>
>>   We compared the whole-brain GM  between 3 groups of subjects in FSL-VBM,
>> in an ANOVA design, with age and gender as covariates. TFCE was used to
>> display results. While we had some positive results, no differences were
>> found between 2 of these 3 groups.
>>
>> We reported our findings and we were asked 'to test formally if the
>> negative results were due to statistitcal power' by a reviewer.
>>
>>  Thanks to previous posts in this email list and inputs in response to my
>> previous question, I am aware that post-hoc power analyses may be
>> meaningless, however my plan is to include a power analysis (because we
>> were asked to) and point out this controversy in the response, which I
>> think is a safer way to deal with this request.
>>
>> So my questions start here:
>>
>> 1. To calculate the power I know that I need the effect size (cohen's d)
>> and sample sizes for a regular t-test. Is that the same for FSL-VBM
>> analysis ?
>>
>
>Effect size is effect size irrespective of the measure. Their are equations
>to calculate effect sizes from both t- and F-tests.
>
>
>>
>> 2. If so, what is the best way to get the effect size for a VBM analysis ?
>>
>> -I can think of two ways to do this: by using fsl_glm i) to get PEs or ii)
>> t-scores and calculate effect size by using t-scores with sample sizes. Do
>> you think these are correct ?
>>
>
>Yes. Use t-scores.
>
>
>>
>> -I tried: fsl_glm -i GM_mod_merg_s3 -d design.mat -c design.con -m GM_mask
>> -o PE --out_t=t_scores
>>  which I think gave me a single PE image and t-score image for the model?
>>
>
>I would pick a specific region. Effect sizes and power differ across the
>brain.
>
>
>>
>>  --But my design includes 6 contrasts and I am interested in the effect
>> size for only 2 of those contrasts. How can I get the PEs and the t-scores
>> for these two contrasts separately ?
>>
>
>Each contrast has a t-score. It is computed automatically.
>
>
>>
>>  --And what is the relation of the scores on the PE image with cohen's d ?
>>
>>
>It depends on the variance, which is why you need t-statistic.
>
>
>
>>  I would very much appreciate your inputs. Please try to be basic while
>> guiding me since I am a novice both in statistics and FSL.
>>
>
>There are a number of pages that describe the relationship between cohen's
>d and the t-test.
>
>
>>
>> Many thanks,
>>
>> Sinan
>>
>