Test of Significance in a Verification Program

The article is limited to consideration of a forecasting program in which a forecaster chooses as his forecast one event out of a finite group of well-defined, mutually exclusive events, for example, “rain” or “no rain.” If a table of scores is prepared, from which a forecaster's score can be obtained for each of his forecasts, the sum total of a set of scores will become normally distributed as the number of forecasts is made large. It is possible to set confidence limits on the score total or on the difference of two score totals. But it is necessary to assume that the forecasts by which a forecaster's performance is judged are independent of each other and that the days constitute a fair sample of the climatology of his station.