TESTING THE CONCORDANCE OF EVENTS WITH THEIR PROBABILITIES

A test due to D. R. Cox (1958a, 1958b) for the agreement between an observed set of independent events and independently prescribed probabilities is considered. Although the exact distribution can be found for small numbers of events this is often not possible. A method of modifying the scoring system slightly is shown to result in much simpler calculation with very little loss in power. The exact distribution is desirable because of slow convergence to normality. The logic of using a simulation test is considered.