Actions Speak Louder Than Goals: Valuing Player Actions in Soccer (1802.07127v2)

Abstract: Assessing the impact of the individual actions performed by soccer players during games is a crucial aspect of the player recruitment process. Unfortunately, most traditional metrics fall short in addressing this task as they either focus on rare actions like shots and goals alone or fail to account for the context in which the actions occurred. This paper introduces (1) a new language for describing individual player actions on the pitch and (2) a framework for valuing any type of player action based on its impact on the game outcome while accounting for the context in which the action happened. By aggregating soccer players' action values, their total offensive and defensive contributions to their team can be quantified. We show how our approach considers relevant contextual information that traditional player evaluation metrics ignore and present a number of use cases related to scouting and playing style characterization in the 2016/2017 and 2017/2018 seasons in Europe's top competitions.

• The paper introduces VAEP, a probabilistic framework that quantifies every on-ball soccer action by computing scoring and conceding probabilities.
• It leverages the SPADL language and CatBoost model to translate event stream data into actionable player performance metrics.
• The framework provides practical applications in scouting and recruitment, offering nuanced insights beyond traditional statistics.

An Evaluation of the "Actions Speak Louder than Goals" Paper: A Comprehensive Framework for Valuing Player Actions in Soccer

The paper "Actions Speak Louder than Goals: Valuing Player Actions in Soccer" presents a novel probabilistic framework called VAEP (Valuing Actions by Estimating Probabilities) for assessing the contribution of soccer players to their team's performance. The essence of the paper lies in its attempt to transcend traditional metrics that focus on rare events such as goals and assists by quantifying the value of all on-the-ball actions while accounting for contextual nuances and subsequent game states.

Key Contributions

1. SPADL Language: The paper introduces SPADL (Soccer Player Action Description Language), which offers a unified representation of event stream data. SPADL is thorough in its ability to capture all relevant data points, enabling accurate action analysis while being vendor agnostic.
2. Framework for Action Valuation: At the core, VAEP employs scoring and conceding probabilities to assign values to individual actions. Each action's value reflects its expected effect on the game state, thus providing an explicit metric for both offensive and defensive contributions.
3. Predictive Modeling: A probabilistic classification approach is employed using various features derived from sequences of actions to predict scoring and conceding probabilities. The paper indicates that CatBoost, a gradient boosting method, yields well-calibrated probabilities ideal for VAEP.
4. Analytical Use Cases: Presenting use cases pertinent to player acquisition and characterization, the framework's practical applicability in scouting and evaluating playing styles across top European leagues is demonstrated. This includes identifying promising talent and aiding recruitment strategies by computing player ratings from action values.
5. Supplementary Tools: An open-source Python package is provided, facilitating the conversion of event stream data into SPADL, implementing the evaluative framework, and predicting probabilities pertinent to scoring and conceding.

Implications and Future Directions

Theoretically, VAEP advances sports analytics by offering a more comprehensive evaluative metric for player actions than traditional statistics such as goals and assists. The method allows for more nuanced player comparisons across different contexts and leagues. Practically, this tool aids teams in making data-informed decisions for player recruitment and strategy formation by quantifying less tangible influences.

Future explorations could look into incorporating off-the-ball actions to provide even more comprehensive assessments of player contributions. There are challenges in ensuring trust and adoption, as scouts and clubs may resist less intuitive metrics over traditional ones. The adaptability of VAEP across varied leagues with disparate event stream data quality also warrants further research.

Overall, the paper contributes significantly to soccer analytics by offering a scalable and context-aware approach to player performance evaluation. The work not only enriches theoretical discourse but also provides practical tools that have the potential to transform player assessment modalities in the soccer industry.