Ranking International Limited-Overs Cricket Teams using a Weighted, Heteroskedastic Logistic Regression with Beta Distributed Outcomes

Abstract

A regression-based ranking method is developed and applied to international lin1ited-overs cricket, using a database of matches played between September 1999 and December 2007. The stn1cture employed is a generalised linear model with logistic link function and beta distributed outcomes and is used to estimate team strength parameters which in turn yield ranking scores. The outcome variable for the regression is a newly proposed measure of the margin of victory based on the Duckwo1ih-Lcwis methodology. The model uses Weibull weighting to discount the impact of matches played in the past and incorporates a heteroskedastic structure to account for the potentially skewing effects of uncom1nonly large victories. Finally, the model is flexible enough to allow exan1ination of the effects of other factors such as home ground advantage.