A Bayesian Mixture Model approach to expected possession values in rugby league

Area of Science:

• Sports Analytics
• Statistical Modeling
• Rugby League Research

Background:

• Traditional rugby league analysis often relies on zonal models.
• Existing models may lack the granularity to capture nuanced spatial probabilities.
• There is a need for advanced statistical approaches to model possession outcomes.

Purpose of the Study:

• To develop a novel Bayesian Mixture Model for rugby league.
• To create a smooth Expected Possession Value (EPV) pitch surface.
• To estimate individual possession outcome probabilities across the entire pitch.

Main Methods:

• Utilized 99,966 observations from the 2021 Super League season.
• Employed a Bayesian approach with 33 defined centers for probability estimation.
• Interpolated probabilities to all pitch locations and derived an EPV measure.

Main Results:

• Generated a smooth EPV pitch surface with location-specific probabilities.
• Enabled visualization of team attacking and defensive strengths.
• Developed an actual vs. expected player rating system.

Conclusions:

• The Bayesian Mixture Model offers greater flexibility than zonal approaches for rugby league analysis.
• The model provides more insightful results and can be adapted to other sports.
• This novel approach enhances the understanding of possession value in rugby league.