help > Questions about statistical analyses performed by NiiStat
Showing 1-1 of 1 posts
Display:
Results per page:
Jan 21, 2021  03:01 PM | anzianom
Questions about statistical analyses performed by NiiStat
Hello,
I am performing on Niistat an unimodal VLSM analysis with stroke lesion masks and a single behavioural continuous variable. I am documenting myself in order to find the right a priori analytical strategy for a good Power- false positives trade-off, and I have some questions:

1) a- Is the statistical computation performed based on a nonparametric Brunner-Munzel rank order test (BM)?
b- A specific problem of my dataset is that I probably will be obliged to probably reduce to 5-6 the n for the minimum number of participants with the lesioned voxel (minimum overlap).
Does the last version of Niistat implement the use of permutation-derived BM in case of n<10 like in my case?
c- I know that this adaptation of the BM corrects for the erroneously inflated results of the standard BM in these situations; however, what kind of performances in terms of power and false+ should I expect in my case with the permutation-derived BM? And how are these performances influenced by the n anyway? for example is the difference between n = 7, 5 , or 3 substantial and what is the minimum n to keep this kind of analysis valid to use?

2) With regards to the normal permutation corrections (NP):
a- Is it based on a permutation-based FWER correction with a v =1 (meaning that at an α of 5% I'll have a 5% probability of having a single false+ among all my positive voxels)?
b- Is it possible to perform a continuous permutation-based FWER showing t thresholds for different values of v?
c- Is it possible to perform a minimum cluster size-based thresholding for FWER corrections?

3) Since I consider the NP with v =1 too conservative I performed an FDR correction and as a result I have z thresholds to be computed as + and -infinite in the results of Niistat.
a- I know that FDR can become surprisingly conservative in cases when the results are very weak or if n is too low, is it true?
b- Is the result I obtained a possible outcome or I probably made something wrong in the procedure?

4) In order to perform multivariate analyses:
a- Is it sufficient to organize the data in the excel file as described in the introduction page and then set the analysis on freedman-lane?

Thanks in advance for your help
Marco Anziano