Dear list,
I am not sure if you got my mail yesterday- here it is again - even if it is a really basic question I would be glad to get some help.. Thanks in advance!
"Dear fsl experts,
I know this will be a very basic question, but I want to be sure (although I read all the manuals) to have understood the statistical principles of my analysis correctly:
After having performed a two sample ttest in randomise I correlated the data in the significant voxels with some behavioral measures. Before I established my design matrix and contrasts using the glm GUI.
My command: randomise -i all_L1_skeletonised -o my_correlation_output.nii.gz -m significant_results_mask.nii -d measure_demeaned.mat -t measure_demeaned.con -n 5000 --T2 -V -D
My first question: can I get wrong results because I wrote in "all_L1_skeletonised" instead of "all_L1_skeletonised.nii.gz"?
Is it true that the correlation analysis using randomise
1. is a non-parametric test design (based on a permutation test)
2. does permutations for each voxel
3. because there are two contrasts (one for positive and one for negative correlation) the results are only one-sided?
If the results are one-sided and I want to report results for uncorrected p-values (I don't get results for tfce_corrp) - is a p-value of 0.01 then equivalent to a two-sided p-value of 0.02?
Thank you very much in advance,
Hanna Gärtner"
|