Player ratings in continuous and discrete team sports

Sargent, J 2013, Player ratings in continuous and discrete team sports, Doctor of Philosophy (PhD), Mathematical and Geospatial Sciences, RMIT University.


Document type: Thesis
Collection: Theses

Attached Files
Name Description MIMEType Size
Sargent.pdf Thesis application/pdf 8.12MB
Title Player ratings in continuous and discrete team sports
Author(s) Sargent, J
Year 2013
Abstract In team sports, there is a mounting focus on the player membership of a team; how they perform as individuals, how they cooperate and how these actions contribute to the team’s success. This dissertation investigates player performance and player rating methodologies in continuous and discrete team sports, using the Australian Football League (AFL) and cricket as case studies. After a discussion on the different types of player ratings, this dissertation progresses to the different types of AFL player data collected for analysis, how such data can be aggregated to produce a match “score”, X, for each player and how these scores can be approximated with a normal distribution. An Elo-influenced adjustive player rating system (APR) is discussed where simulation techniques pit the player being rated against an opponent player in the same position. The main limitation of the APR was that the data and scoring was too player-centric, overlooking aspects of teamwork and a player’s contribution to the team. Player interaction was investigated which required transactional data to be analysed. Interactions between pairs of players followed a negative binomial distribution with parameters estimated using a Pearson chi-squared approach. Player performance in a match was quantified using eigenvector centrality, an important network statistic, indicating a player’s level of interaction with other central players. Team strength was calculated by averaging each player’s centrality in the network. The team index for any match was adequately related to the score margin for that match making it possible to observe different players’ contribution to team performance (margin) when included and excluded from a simulated network. Data on the Geelong football club was used to conceptualise the In team sports, there is a mounting focus on the player membership of a team; how they perform as individuals, how they cooperate and how these actions contribute to the team’s success. This dissertation investigates player performance and player rating methodologies in continuous and discrete team sports, using the Australian Football League (AFL) and cricket as case studies. After a discussion on the different types of player ratings, this dissertation progresses to the different types of AFL player data collected for analysis, how such data can be aggregated to produce a match “score”, X, for each player and how these scores can be approximated with a normal distribution. An Elo-influenced adjustive player rating system (APR) is discussed where simulation techniques pit the player being rated against an opponent player in the same position. The main limitation of the APR was that the data and scoring was too player-centric, overlooking aspects of teamwork and a player’s contribution to the team. Player interaction was investigated which required transactional data to be analysed. Interactions between pairs of players followed a negative binomial distribution with parameters estimated using a Pearson chi-squared approach. Player performance in a match was quantified using eigenvector centrality, an important network statistic, indicating a player’s level of interaction with other central players. Team strength was calculated by averaging each player’s centrality in the network. The team index for any match was adequately related to the score margin for that match making it possible to observe different players’ contribution to team performance (margin) when included and excluded from a simulated network. Data on the Geelong football club was used to conceptualise the methods in these chapters. For the latter section of the research, a Visual Basic program was written that called conditional probability distributions to simulate outcomes—runs and dismissals—while a limited overs cricket match was in-play. These likelihoods were conditional on the player’s order in the batting list, the delivery number—both discrete variables—and the type of batsman (fast, medium or slow scorer). The simulated batsman scores were then adjusted for team strength, innings and venue effects using multiple linear regression. This dissertation demonstrates the benefits of the model by fitting log-normal distributions to simulated innings (n=500) by Australia’s Ricky Ponting in the 2011 ODI World Cup quarter final. It was then possible to approximate how likely he was to achieve a certain score prior to the match, then at 10, 20 and 30 over intervals. It is anticipated that real-time information of a batter’s score expectations will add confidence to wagering in individual performance markets such as “highest score”, as well as live player-rating revisions.
Degree Doctor of Philosophy (PhD)
Institution RMIT University
School, Department or Centre Mathematical and Geospatial Sciences
Keyword(s) Adjustive player ratings
network analysis
negative binomial distribution
in-play simulation
Versions
Version Filter Type
Access Statistics: 416 Abstract Views, 2167 File Downloads  -  Detailed Statistics
Created: Fri, 06 Jun 2014, 09:41:32 EST by Maria Lombardo
© 2014 RMIT Research Repository • Powered by Fez SoftwareContact us