Further significance tests

Abstract

In a previous paper (afterwards referred to as Paper I) tests have been given for the significance of some quantities found statistically. The results are given in the form P(q | θh)/P (˜ q|θh); here h denotes the previous knowledge and θ the experimental evidence used, while q is the hypothesis that all the variations outstanding can be attributed to accidental error or random variation, and ˜q the hypothesis that at least part of them is systematic. It has been supposed in the analysis that q and ˜q are equally probable on the information h; but if they are not, the only alteration is that the ratios evaluated now representIf successive batches of relevant information are available the total effect on the probability of q can therefore be got by multiplying the values ofgiven by the investigations separately. In each case the assumption that q has prior probability ½ is really a practical working rule rather than a statement of fact.