Dear NBS experts, 

I am new to this toolbox and still trying to get a grip on the best way to use it. I am especially wondering about the primary threshold.
Although I know that there is no right or wrong threshold, I noticed that generally a range from 2 to 4/4.5 is applied (by reading the other posts on the forum and some papers that used NBS reporting on this choice). I also stumbled upon a thread here that suggested to link it to the desired effect size, by using the formula sqrt(n)*effect size. So I tried all of these options.

I am using NBS on a total of 56 connectivity matrices to find differences between OCD patients (n = 28) and healthy controls (n = 28). My parcellation scheme was quite fine-grained, so I ended up having a 250x250 connectivity matrix per subject. 
The default primary threshold of 3.1, and also values above (up to 4/4.5) yield very dense significant subnetworks, that are considerably hard to interpret as they take up almost the entire network. This is true for both contrast directions [1 -1] and [-1 1].
By using the above-mentioned formula, with a desired effect size of 0.7, I set the primary threshold to 5.23 and started to obtain more manageable results. However, I cannot help noticing that this value is way higher than what it is usually reported. 

I wonder if this has anything to do with the number of connections the NBS has to test? Is it common to see such high primary threshold values? Is there any rule or reasoning linking the choice of the threshold to the size of the connectivity matrix (hence to the number of tests being performed)? 

Thanks in advance for any help you may provide 

Best,

Hi Samantha, 

there is no default threshold.

The approach that you have taken based on a minimum effect size is very reasonable. This will ensure that you do not detect effects that are below your minimum meaningful effect size of 0.7.

A primary threshold of t=5.23 sounds very reasonable to me. Remember to explain that this threshold came about by converting your minimum effect size to a t-statistic value.

I don't think that there is a clear relationship between threshold and the size of the connectivity matrix. Perhaps effects may become weaker at high resolutions dues to lower SNR, although higher-resolution may provide better spatial specificity to localize effects. So I am not sure if there is a clear relation. 

It seems that you are on the right track.

Best wishes,
Andrew



Originally posted by Samantha Baldi: