"[F]or every type of experiment there is a numerical quantity, depending partly on the chance of the single event succeeding, and partly on the total number of the trials of it, which we may term the '. ... it gives us a measure of the frequency with which deviations of various sizes will occur."