Hello Andrew and all,
Thank you very much for the toolbox. I started using it today and will use it to finalize my analysis of MEG connectivity data.
The question I have is:

The distribution of my connectivity measure is not normal. Therefore I am not sure that using T-test or F-test or any parametric statistics makes sense in my case. Is there a plugin or a quick hack to use nonparametric statistics like Kruskal-Wallis or others instead of t-test?

Thank you a lot for your great work.

Adham

Hi Adham,

The NBS does not support Kruskal-Wallis. However, if you are worried about data non-normality, you could apply a power transform to your data before using the NBS. For example, use x^2 to correct for a downward skew and log(x) to correct for an upward skew. The other thing to remember is that the NBS is inherently nonparametric: the t-test and F-test in the NBS are used as a measure of variation, but not to explicitly compute a p-value. So non-normality is not as much of a concern because p-values are computed non-parametrically using permutation testing. 

So I'd suggest applying a power transform if you are particularly concerned about your data not being normal, but also remember that the NBS is inherently non-parametric.

Andrew

Originally posted by Adham Elshahabi: