Statistical analysis of diagnostic accuracy with applications to cricket

Abstract

In the sport of cricket, as with any other sport, spectators and officials would like the games to be as fair as possible. To this end, we evaluate methods used to determine the winner of interrupted games using statistical accuracy. In the traditional One Day International cricket matches, the current Duckworth–Lewis (DL) method and the discounted most productive overs (DMPO) method are each used for predicting the winner. However, with the growing popularity of shorter Twenty20 matches, a new Bhattacharya–Gill–Swartz (BGS) method has also been introduced. We created both classical and Bayesian intervals to estimate the true accuracy of each. Using past game data from 2005–2010, we compared the DL, DMPO and BGS methods using the new accuracy intervals and receiver operating characteristic (ROC) curves.