On a Significance Test for Two Poisson Variables

A further practical question deals with the extrapolation of Ractliffe's results to Poisson expectations with A < 5. In this paper, the exact distribution of the test-statistic u will be derived. It will be shown that the approach to normality is slow. Formulae will be given for numerical calculation of individual and cumulative probabilities. Particularly for a probability level of 5 per cent (double-tail), the u-test performs much better than the customary "binomial" test, with the added advantage of simplicity and non-requirement of tabulated values. The u-test may safely be applied from A = 2 5 upwards. 50