Dear SPM-list,
does anyone know if there is a way to run a one sample t test against a constant other than zero?
More specifically, I have maps of longitudinal GM change (GMmap at BL - GMmap at FU; Intra- and Inter-individually DARTEL registered,...) and I am using a simple one sample t-test (contrast: 1) to assess regions that show significant volume loss over the follow-up period in one group. Given that global GM loss over FU is already highly significant, voxel-wise effects at p(FDR)<0.05 include most parts of the brain.
In order to get more regionally specific effects I wonder if it is possible to run the one-sample t-test against the average signal reduction per GM voxel, for example calculated as group average global GM loss in mm^3 divided by total volume of the GM mask in mm^3 (i.e. number of GM voxels). This would ask the question: "Where in the brain is the GM change over follow-up higher than global GM loss, i.e. where are the "hot spots" of longitudinal GM loss" and in my opinion this would be a preferable approach over increasing the statistical threshold (e.g. to p(FWE)<0.05) to obtain regional specificity.
A similar approach is often employed in cross-sectional studies of volumetric differences between groups by covarying for global GM instead of TIV. However, a similar ANCOVA approach using global GM loss over FU as covariate does not change the results in the one-sample t test setting I am using here (and I donĀ“t really understand why...).
An obvious solution for the test I am proposing is to substract the group average signal reduction per GM voxel (calculated as described above) from each map of GM change and run the one-sample t-test on these data. However, the best way would be to specify this parameter directly in the stats setup if such an option exists.
Does anyone know of such an option? Does the analysis I am proposing sound reasonable? If I had to go for the subtraction-option, would it be more correct to subtract from each individual map the average change per GM voxel of this subject instead of the group average change per GM voxel?
Any advice will be highly appreciated.
Best,
Michel
|