Doubt regarding requirement of number of null model network

Submitted by minihk on Thu, 09/10/2015 - 09:43

Hi all,

I am using GraphVar for deriving network measures in an rsfmri study. I would like to know the number of null model network required for a particular rewiring algorithm.

eg:- In my study I am using rewiring alogorithm 'randmio_und' and from literature review I came to know that the default value for null model network is 20 and it will vary aaccording to the algorithm you are using. Anybody have idea regarding this. If so pls help me out.

 

Thanks,

Mini

Forums:

marina_weiler

Mon, 08/15/2016 - 18:43

Hi Mini

I have the same question, did you find an answer?

thanks,

Marina

 

Johann Kruschwitz

Tue, 08/16/2016 - 09:43

Hi,

when testing against null-model networks I would suggest to compute as many random networks  as possible (at least 100 … but 1000 would be better, or even more). This idea goes back to the concept of bootstrapping, where you create random data that ultimately results in a distribution of difference under the null hypothesis. By placing the resulting metric/statistical value of your original data in the corresponding random-network/permutation derived distribution a p-value is calculated based on its percentile position (whereas the non-parametric p-value will depend on the number of null-models; e.g. 100 null-models result in a minimal p-value of 0.01). The same would count if random networks are used for normalization purposes of the respective graph metric (or computation of small-worldness). In this case you want to have an estimate of the “noise” by estimating how much of the respective graph metric is already contained in subject specific random networks. The estimation of this noise will be better the more random networks are created (c.f., flipping a coin 20 times will not give you a good estimate of the probability, whereas a 1000 flips will bring you really close to the 50% chance for heads and tails). You will find more information also in the GraphVar Manual (chapter 2.5).