A COMPARISON OF SCATTER-DIAGRAM ANALYSIS WITH DISCRIMINANT ANALYSIS AND A NOTE ON MAXIMIZING THE SKILL SCORE

Abstract

TWO questions concerning statistical objective forecasting systems are investigated: (1) Does multiple discriminant analysis (MDA) yield better results than do scatter-diagram analyses made by meteorologists experienced in their use? (2) How should categorical forecasts be derived from probability estimates when it is desired to maximize the Heidke skill score? It is determined that for the particular case tested, a small sample of data and two predictor variables, a subjective analysis will usually yield a better probability forecast than MDA. A method devised by Bryan for making categorical forecasts from probability estimates appears to give better results on test data than do two other methods investigated.