I am using 3dttest++ with ETAC and was wondering if someone could help me understand the sample size requirements. Here are examples of the messages I’ve gotten when attempting to run 3dttest++ -ETAC with fewer datasets than required:
You can’t use -ETAC with nval_AAA=10 < 17 in a 1-sample test
You can’t use -ETAC in a 2-sample test with nval_AAA+nval_BBB=10 < 14
You can’t use -ETAC in a paired 2-sample test with nval_AAA+nval_BBB=20 < 28
Is there a specific statistical reason for these dataset minimums?
I think nval_AAA and _BBB refer to the number of -setA/-setB elements (datasets, probably). In the first 1-sample test with only -setA, ETAC wants at least 17 datasets as input for doing the permutations. And similarly for the 2-sample or paired tests. 10 inputs seems too few for that aspect.
Thank you for your response! The error messages were helpful in understanding how many datasets are required for ETAC to run, and they did match up to the number of datasets in -setA and -setB, as you described.
However, I was wondering if there was a specific reason that 17 is the number of datasets required for a 1-sample test (and 14 for a 2-sample test; 28 for a paired 2-sample test). Is there any statistical reason that relates to the way the permutations are performed?
That was likely just based on Bob’s basic feeling for it at the time that he wrote the method. The world will surely not end going down to 16, but random permutations will not follow reasonable distributions with too few choices.