help > RE: Test-statistic threshold
Aug 13, 2019  06:08 PM | Andrew Zalesky
RE: Test-statistic threshold
Hi Craig,

The Threshold Box only accepts a t-statistic value in the case of a t-test and a F-statistic in the case of an F-test. It does not accept p-values.

However, you can always convert a p-value threshold into a t-statistic threshold, if you know the degrees of freedom in your experiment. The degrees of freedom is roughly equal to the number of subjects. There are many online calculators available to convert p-values to t-statistics.

The choice of threshold for the NBS is up to the user. I suggest selecting a threshold that corresponds to your minimum meaningful effect size.

Strictly speaking, only one threshold should be tested, but I think that it is informative to understand the characteristics of your findings to different threshold choices and this should be reported.


Originally posted by Craig Peterson:
Dear Dr Zalesky, 

Thank you for this excellent resource. 

I have a question about the test-statistic threshold (primary threshold). 

My understanding from reading the NBS manual and associated paper is that this is an arbitrary threshold, in your 2010 Neuroimage paper, for instance, a t-statistic threshold of 3 is used. I have seen several papers however which use a value corresponding to a p-value, e.g. the primary threshold might be set to p < .01.

My question is how does one use a p-value, rather than a t-value as the threshold (I presume that the 'Threshold' box within the NBS interface is only for entering larger test-statistic values rather than p-values?). I tried entering smaller values values here (e.g. 0.01), but it resulted in a very dense and difficult to interpret connectivity pattern). Thank you.

Threaded View

Craig Peterson Aug 13, 2019
RE: Test-statistic threshold
Andrew Zalesky Aug 13, 2019