Dear Bonnie.
I would have expected the combined results to be slightly more sensitive as there are more degrees of freedom available to get an estimate of the residual noise. This normally gives a tighter estimate of the noise and improves the overall statistics. However, you may have been in the situation where the noise was not the same in the different groups and so some of the pairings gave lower average noise estimates, and hence higher t-statistics. It is always difficult to compare statistics by looking at how many values are above a threshold, as this can reflect tiny changes in values around the threshold. I think that your results are probably just the kind of statistical fluctuations that inevitably occur due to the random nature of the noise. I certainly wouldn't expect that on average you would be better off doing individual pairs. If your results are similar and show the same regions to be affected then this also doesn't change the underlying biological story that you are finding, and so I would just stick with the combined GLM.
I hope this helps.
All the best,
Mark
On 26 Nov 2012, at 07:53, Bonnie Y K Lam <[log in to unmask]> wrote:
> Dear FSL experts,
>
> I am comparing the patterns of white matter atrophy in 3 disease groups to control.
>
> I have tried a 4-EVs GLM contrast (combining the 3 disease groups and control in one contrast) and also did separate 2-EVs t-test for each disease group and control.
>
> I found the locations of white matter change resulted from the different GLM contrasts to be similar but the number of voxels change differ by 10%.
>
> The separate t-test seems to give more voxel change than the combined analysis.
>
> Is the reason for detecting less white matter change in the combined analysis due to each individual is compared to the mean FA skeleton of all subjects (white matter loss being less sensitive due to the averaging process)?
>
> I noticed a lot of literature used combined multiple disease groups in one analysis despite separating them might be more sensitive.
> I would like to know which is a preferred contrast in this case (combined or separate)?
>
>
> Thanks,
> Bonnie
>
|