NBS analysis

Submitted by cha on

Dear GraphVar users,

I would like to run an analysis with NBS in GraphVAR. I use to run NBS with the NBS1.2 toolbox in SPM8. I am a little confused with the different options in GraphVar. Could someone help me? I attach a capture of my workspace to see you the options I have chosen.

Can someone explain to me how to choose the threshold t as in NBS1.2?

Where can I choose the number of permutations? Is it in the "number of repetition" part of the workspace?

Thanks for your help,

All the best,

Cha

 

Johann Kruschwitz

Wed, 07/29/2015 - 09:51

Permalink

Hi Cha,

yes - for group comparisons "number of repetitions" is the number of permutations (for correlational analyses you could also use random networks for testing against random data). Generally, it is advisable to choose a high number of repetitions to derive accurate p-values later on (so please do not use only 3 repetitons as in your workspace). In GraphVar you can determine the set of supra-threshold links by setting the initial alpha level for each link (here we do not use t). This happens later on in the results viewer (there is a button "NBS" that will open the "network inspector" where you will have the option to choose the alpha level).
Best,
Johann

Hi Johann,

Thanks for you answer. 

The workspace I gave you as an example was ok for you? I mean I select the right options, (execpt the number of repetitions). 

For the number of repetitions, could I use 5,000 repetitions? (as I used to do in NBS1.2)

In the network inspector, I could select the alpha level. But this is not the case for t values. How can I apply the arbitraty t threshold? If it is not possible, what is the default value in the graphvar script?

Thanks for your help

Cha

 

Hi Cha,

from what I see in your worspace I think you want to compute a group comparison on the raw matrices (whereas your network is defined as starting with the superior frontal gyrus), you don´t want to apply any conversions on the data (i.e., no change of the raw values to e.g. absolute values), you probably have Z transformed values as your CorrMatrix array is called "Z" and as you don´t have selected the r to z transformation. If this is what you want to do it seems okay.

Yes you could use 5000 repetitions (I think GraphVar is currently slower as NBS1.2 though - this will change in a next release where we will vectorize the statistics).

Yes - in GraphVar we do determine the set of supra-threshold links based on their significance (i.e., alpha level of the respective test statistic) This means we have expressed the threshold as a p-value rather than a test statistic (i.e., currently there is no option to define the supra-threshold links based on an arbitrary test-statistic threshold as e.g. "t"). Please do not confuse this alpha-level with the "Significance" field under "Advanced Settings" in NBS1.2, which will specify the FWER corrected significance level of the graph component. In GraphVar you see the respective alpha of the graph component in column five in the Network Inspector (please refer to the manual page 25 where the results panel of the Network Inspector is explained in more detail).

Best,
Johann

Attachment Size
capture_errorNBS.png 155.94 KB
Forums
GraphVar