Hi Dr Zalesky,

I have a question concern regard to the Threshold. I do not quite understand while I did not get significant graph components with low threshold (i.e 3.3), but when I changed threshold to (i.e 4.0) there would be a significant graph components. As you mentioned in the manual, any connections with test statistic higher than the predefined primary threshold, which would assemble into a set of supra-threshold followed by permutation testing. Lastly, as I experimented with my own data, its seemed that low threshold would yield more number of nodes and connections as oppose to high threshold, so what is an appropriate threshold to use?

Bests,

Thomas

Hi Thomas,
this is certainly possible and completely normal. When the threshold is increased, the null also becomes more constrained, and thus "easier" to reject.

There is no right or wrong threshold. You may want to present findings for a few different thresholds. Or you may want to set a threshold corresponding to your minimum meaningful effect size. I recommend giving consideration to the minimum effect size that is meaningful in your experiment. Once this is known, you can set the threshold to match this effect size.

Andrew


Originally posted by Thomas Tan:

Hi Dr Zalesky,

Thank you for your input regard the threshold. Since you mentioned calculation of effect size in order to set the appropriate threshold. In the case of functional connectivity, how should I calculate the effect size (i.e take the group mean of functional connectivity from adjacency matrix(164x164) for patient and control and  how about standard deviation of each group?). 

Bests,

Thomas 
Originally posted by Andrew Zalesky:

Hi Thomas,

the effect size is determined by you. This is determined by your hypothesis, experimental design, etc.

For example, if you are only interested in a Cohen's d of 0.2 or above, then the threshold for a t-test would be set to t = sqrt(n) * 0.2, where n is the number of subjects in your experiment.

Andrew


Originally posted by Thomas Tan: