An Evaluation of Human and Computer-Based Predictions of the 2003 Rugby Union World Cup

Abstract

The purpose of the current investigation was to compare the accuracy of human and computer-based methods of predicting the 2003 Rugby Union World Cup. The computer-based methods were multiple linear regression, binary logistic regression, artificial neural networks and a simulation package. The computer-based methods used data from the previous 4 Rugby Union World Cups to develop predictive models of international rugby union results based on team strength, distance travelled to the tournament and recovery days between matches. The computer-based methods correctly predicted between 39 and 44.5 of the 48 matches which was more accurate than the 40.66 averaged by 42 individual humans. However, there was a far greater range of accuracies for the human predictors with the most successful human correctly predicting the outcomes of 46 of the 48 matches. Furthermore, an expert focus group successfully predicted 43 of the 48 matches which was better than the average computer-based method. The most successful of the computer-based methods was a simultion model. The accuracy of the artificial neural network predictions increased with the number of nodes within the middle layer. The study provided evidence that computer-based methods are more successful at predicting the outcomes of international rugby union matches than the average human, but are not as successful as human experts.