Abstract
The aim of this thesis was to create a collection of metrics that quantify the overall contribution of a rugby league player to the success of their team, to inform player recruitment and selection practices. To achieve this, five studies were undertaken: a review that evaluated the use of player performance analytics among team invasion sports; two experimental studies involving the creation of a series of metrics that evaluated the contributions of individual players to team success in men’s and women’s rugby league, which were then validated against current methods of team and player performance evaluation in rugby league; a third experimental study which developed a model to predict match outcome in men’s and women’s rugby league that incorporated player-specific metrics and lineup composition with broader team-based attributes; and lastly, the development of a decision support system for a rugby league club that can inform the selection of an optimal team lineup based on player form, using constraints such as player availability and positional eligibility.A review of the literature on performance analytics in team invasion sports identified three primary methods for assessing player contributions to team success: performance indicators, spatiotemporal data, and the performance-pay relationship. Performance indicators proved valuable for evaluating player contributions, with machine learning models such as Bayesian networks and Random Forests showing potential in predicting performance. Spatiotemporal analytics, employing tools such as positioning and sensor systems, enable detailed tracking of player movements to evaluate contributions on a play-by-play basis. Additionally, the link between player performance, market value, and salary underscored the importance of aligning payroll allocation with performance metrics. The review concluded that despite the availability of effective methods for analysing player performance, the extent to which coaches use these tools in practice remains uncertain. Rugby league in particular presents unique challenges due to the complexity of its plays and the involvement of multiple players. However, methods from other sports such as baseball and basketball offer transferable insights. For example, incorporating metrics that adjust for opponent strength or evaluating the performance of player combinations could enhance recruitment strategies. Ultimately, the review highlighted the potential for developing advanced, evidence-based tools to better assess and optimise player performance in rugby league.
The first experimental study aimed to develop objective metrics to evaluate player performance in men’s Australian elite rugby league, providing a more accurate way to measure individual contributions to team success. These metrics, known as Wins Created, Losses Created and Net Wins Added were validated by their ability to predict team wins, Dally M Team of the Year selections, and State of Origin picks, showing promise but also highlighting areas for improvement. The metrics predicted a team’s win total within a margin of two wins per season, demonstrating a strong correlation between the metrics and team performance. However, the binary nature of game outcomes (win or loss) means exceptions occur, such as defensive contributions being undervalued in certain scenarios like intercept tries. The metrics accurately predicted 54.4% of Dally M Team of the Year winners, with accuracy varying by position. Playmakers like five-eighths achieved an 80.0% success rate due to their prominent role, while locks, whose contributions span attack and defence, had a lower accuracy of 20.0%. State of Origin predictions achieved 59.7% accuracy, with better results for Queensland (69.9%) compared to New South Wales (49.4%), attributed to Queensland’s smaller, more consistent player pool and historical success.
The second experimental study addressed the lack of objective player-based metrics in women’s elite rugby league in Australia, refining methods from the first experimental study while addressing identified challenges. The metrics accurately predicted team wins, with an average margin of error of one win per season. For the Dally M top ten leaderboard, the Root Mean Square Error (RMSE) was 8.2, where average top 10 player scores were 12 points. The metrics identified 65.4% of Dally M Team of the Year recipients, reflecting ‘substantial’ agreement, though this dropped to 46.2% when strength-of-schedule adjustments were applied. Predictions for the Rugby League Players Association(RLPA) Dream Team award winners also showed less than 50% agreement, partly due to the penalising nature of ranking deviations. A key improvement was the development of a model to classify interchange players' positions with 78.6% accuracy, enhancing substitution strategies and positional analysis. This model also suggested tactical opportunities by highlighting misclassified players who might excel in other roles. The findings underscored differences between women’s and men’s rugby league, particularly regarding the roles of “spine” positions. Only five of the top twenty predicted women’s players were in “spine” roles, contrasting with 18 in men’s rugby league, suggesting that strength and speed, as seen in positions like edge back row and lock, were more critical in the women’s game. Refinements from earlier models, including the removal of counter-intuitive coefficients, improved predictive accuracy. Notably, the women’s study achieved a higher success rate (65.4%) for predicting Dally M Team of the Year recipients compared to the men’s study (54.4%).
The third experimental study built on the metrics developed in earlier studies to create a model to predict match outcomes in elite men’s (NRL) and women’s (NRLW) rugby league. By integrating player-specific metrics with team attributes like Elo ratings, the model demonstrated strong predictive power. Two testing methods were used: a standard training-testing split and an order-preserved method maintaining the chronological order of matches. For all matches in the dataset, the order-preserved method achieved 64.4% accuracy for NRL games and 55.0% for NRLW games. During the 2022-2023 NRL seasons, the model’s accuracy exceeded 69.0%, while the NRLW model reached 75.0% using the standard method and 66.7% with the order-preserved approach for Round 5–9 of the 2023 season. The NRL model surpassed previous benchmarks in the literature, which achieved 63.2% accuracy using team-based factors like location and form. The results reinforced the importance of combining team composition with broader game features (e.g., opponent strength and match location) in influencing outcomes.
Building on the success of the previous study, the final study introduced a decision support system (DSS) to assist rugby league clubs to optimise their team lineups. The system accounted for key factors such as player form, availability, and positional eligibility, aligning with the decision-making processes of coaching staff. During the 2024 NRL season, the DSS accurately predicted 91.2% of player selections (372 out of 408) for the 17-player squads. Its performance was even higher in the NRLW, correctly predicting 97.4% of selections (149 out of 153). These results demonstrate the DSS's potential as a practical and reliable tool for coaches, especially when addressing player absences due to injuries, suspensions, or representative duties. A notable feature of the DSS was the use of percentile values to contextualise player rankings. This feature allows coaches to compare a player’s performance relative to their positional peers, making it easier to identify underperforming players or those excelling in lower grades. Colour-coded indicators highlighted players ranked in the bottom 25% for their position, signalling that their selection might not be justified and prompting consideration of alternative players. Beyond team selection, the DSS offered valuable insights for recruitment and retention strategies. By evaluating whether players met performance expectations, the system could inform decisions about retaining existing players or targeting recruits to strengthen specific positions. The ability to align player performance data with strategic needs positions the DSS as a versatile tool for both match preparation and long-term planning.
This thesis has developed comprehensive metrics to quantify player contributions in men’s and women’s rugby league, aiming to enhance recruitment and selection practices. It has demonstrated that objective, data-guided methods can effectively evaluate individual performance and predict team success. Key contributions include creating new performance metrics, developing predictive models for match outcomes, and integrating these tools into a decision support system for team selection. These advancements provide a framework for clubs to improve recruitment, team selection, and strategic planning. The thesis highlights the transformative potential of data analytics in rugby league, offering clubs a competitive edge and supporting the sport’s growth and professionalisation. By laying the foundation for future advancements, this work underscores the importance of evidence-based tools in modern team management and coaching, positioning data-driven strategies as essential for the ongoing evolution of rugby league.
| Date of Award | 2025 |
|---|---|
| Original language | English |
| Supervisor | Jocelyn MARA (Supervisor) & Ibrahim RADWAN (Supervisor) |
Cite this
- Standard