Sunday, November 17, 2019

Statistical significance under fire

Sorry, wrong number: Statistical benchmark comes under fire

Like the student T-test, the probability this would happen from a Gaussian distribution.

Not all theories produce guassian draws.  But if the data sequence is complete enough, maybe even massaged a bit, then one can compute everything numerically with convolutions, or even monte carlo.  Then the probability should be mostly correct and you can get computed histograms of the effect..

No comments: