Within-subject coherence statististics for virtual sources

[Jan Mathijs Schoffelen]

Dear Lorina,

You have some virtual-channel coherence data, which you put into a .powspctrm-field to fool fieldtrip's routines, correct? These coherence-spectra you obtained from a bunch of subjects, and in different conditions. Now you want to test your null-hypothesis of exchangeability between the conditions, by means of a permutation test. So far so good, because the ingredients are there. Let's go through your questions:


> Dear Robert 
& CO,

>  Thanks so much for your help. I followed your suggestions can now 
> run the scrip (as specified in your email below) and everything 
> seems to work ok. However, that I am a bit surprised by the values 
> that I get in the stat field. 
>  As suggested, I specified in my script cfg.method = ‘diff’. I was 
> expecting to be calculating the p value, for my test statistic 
> which would be the dependent samples t test. However, in the stat 
> field that I get from running the script, I have positive and 
> negative values. This makes me wonder whether cfg.statistic = 
> ‘diff’ is calculating p values, or, if not, what is it calculating? 

cfg.statistic = 'diff' means that the statistic that is computed is indeed the difference in coherence between the two conditions. It is done by calling the function statfun_diff, and you verified correctly that the function does what it claims to do. The probability which is obtained after the permutation is stored in the field .prob

>   
>  From the other statfun_xxx functions in the fieldtrip ‘private’ 
> folder, I thought that the statfun_depsamplesT would be more 
> appropriate for calculating the t statistic for my two conditions. 
> However, when I modify my script by substituting cfg.statistic = 
> ‘diff’ for cfg.statistic = ‘depsamplesT’ and run it, it crashes.

could you be a bit more specific about why it crashes? Theoretically it shouldn't crash, but I am not aware of the intricacies of the function. It might be worthwile to check whether there is a statfun_paired-tstat in the release-version, which should do the same trick. From a practical point of view it does not matter whether you use a T-statistic or just the difference (the difference is just that you divide the diff between the two conditions by the variance of the diff between the two conditions).

Yours,

Jan-Mathijs