Assessing Decision Rules For Stopped Cricket Games

Abstract

In interrupted limited overs cricket games, teams do not always get a chance to finish the full game. In such cases, different methods can be used to establish which team wins. The method currently in use is called the Duckworth-Lewis method; it was introduced at the international level in 1998. The purpose of this project is to investigate and check the accuracy of Duckworth-Lewis method and make statistical comparisons with other proposed methods. The accuracy is checked via bias estimation, Cohen's Kappa and root mean square error methods; cross-validation is used to assess out-of-sample accuracy. The resource table, a summary of the expected fraction of total runs scored by a given point in the game, has missing values and monotonicity flaws. To improve the resource table, different statistical methods such as isotonic regression and Gibbs sampling, are used to construct different resource surfaces. The accuracy results show that the Duckworth-Lewis displays the lowest accuracy out of all the methods; whereas, the improved Duckworth-Lewis is more accurate at predicting the new target or results of stopped games. Thus, the accuracy of the Duckworth- Lewis can be improved by ignoring the old games and constructing the resource surface using the modern data.