Predictive models of the 2015 Rugby World Cup: accuracy and application

Abstract

The current investigation compared 12 models of outcomes of international rugby union matches and then used the most accurate model to investigate performances within the 2015 Rugby World Cup. The underlying linear regression models were used within a simulation package that introduced random variability about performance evidenced by the residual distribution of the regression analyses. Each model was used within 10,000 simulations of the 2015 Rugby World Cup from which match outcome and team progression statistics were recorded. The most accurate model with respect to the actual 2015 tournament was developed using data from all seven previous tournaments rather than restricting cases to the most recent three tournaments. The model was more accurate when the data used violated the assumptions of linear regression rather than transforming variables to satisfy the assumptions. The model included World ranking points as a predictor variable and was more accurate than corresponding models that represented relative home advantage as well. The most accurate model used separate models for the pool and knockout stage matches although the 9 models that separating these match types were less accurate on average than when the two match types were considered together. This model was used to investigate properties of the 2015 Rugby World Cup. The tournament disadvantaged three teams in the World’s top 5 who were drawn in the same pool. Teams ranked in the World’s top 7 did not perform as well as predicted while teams ranked 16th and below performed better than predicted suggesting that the strength in depth in international rugby union is increasing. There was a small effect of having additional recovery days from the previous match compared to the opponents which was worth 4.1 points. The information produced by this research should be considered by those design tournaments such as the Rugby World Cup.