PRE-RACE BEHAVIOR OF HORSES AS A PREDICTOR OF RACE FINISHING ORDER

The behaviour and appearance of 867 horses entered in 67 races at two Melbourne racecourses, Flemington and Moonee Valley, was assessed over a 20-month period, We recorded 29 variables for each horse, 19 of whi ch were behaviour/appearance variables. Fourteen variables were record ed in the parade ring/mounting yard as the horse was being led by the groom prior to the race, and five variables were scored while the hors e was on the way to the starting gate with the jockey up. Seven variab les were race-book information, and three variables were recorded from the published race results: actual weight carried, starting price; an d finishing order. The data were analysed using the techniques of univ ariate and bivariate analysis, and multivariate discriminant analysis. Univariate analysis showed that only tail elevation (P < 0.05), neck angle with the jockey up (P < 0.01), and resistance to the bit (P < 0. 05) had a significant relationship with finishing order. However, the power of these variables to discriminate finishing order was poor when compared with the traditional variables of starting price (P < 0.001) and weight carried (P < 0.001). Classification of horses into winners (finishing in the first 20-percentile) and losers (finishing in the l ast 20-percentile) showed that winners tended to be fitter and more re laxed and losers tended to be more aroused and required greater contro l. Bivariate analysis confirmed that increased elevation of the head, neck and tail were associated with increased arousal. Bivariate analys is also suggested that sweating on its own was not a reliable performa nce indicator, but in conjunction with other variables might indicate losers. Multivariate discriminant analysis was used to discriminate wi nners (horses finishing in the first 20-percentile) from other horses, and losers (last 20-percentile) from other horses. The data were sele cted for analysis on the basis of performance at Moonee Valley, and th en the computed discriminant functions were used to predict group memb ership (i.e. winners or losers) at Flemington. All variables were ente red into the analysis using a stepwise method. Six variables were requ ired for maximum discrimination of losers, in contrast with eleven var iables required for discrimination of winners. Classification results using the discriminant functions to predict winners and losers showed that 31/46 (67.4%) losing predictions were correct In contrast only 9/ 32 (28.1%) winning predictions were correct. We conclude that pre-race behaviour and appearance of horses is a valuable aid in predicting lo sing horses and that this information has potentially high economic wo rth. (C) 1996 Published by Elsevier Science B.V.