ABSTRACT
Player movement metrics in football such as speed and distance are typically analysed as aggregates, sometimes outside of any specific tactical or match context. This research adds context to a player’s movement over the course of a match by analysing movement profiles s and bringing together tools from the sport science and sports analytics literature. Position-specific distributions of player movement metrics: speed, acceleration and tortuosity were compared across phases of play and in-game win probability using 25 Hz optical player tracking data from all 52 matches at the 2019 FIFA Women’s World Cup. Comparing the distributions using the Kolmogorov–Smirnov test and Wasserstein distances, differences were identified in these movement profiles across, in and out of possession phases, with small negligible overall positional trends across in-game win probabilities. In-game win probabilities are used in tandem with phases to present a player specific case study. The results demonstrate how sports analytics metrics can be used to contextualise a subset of movement metrics from sport science and provide a framework for analysis of further movement metrics and sports analytics modelling approaches.
Acknowledgements
The authors would like to acknowledge Tom Worville (The Athletic) for his help with the visualisation library ggplot2 in R and the Friends of Tracking group and particularly Laurie Shaw (Manchester City, Harvard University) for their public github with resources for using and visualising football tracking data. Whilst none of that code was used directly in this research it was a valuable benchmark.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Data availability statement
StatsPerform tracking and event data provided by FIFA, Statsbomb data available via Statsbomb open data repository: https://github.com/statsbomb/open-data.
Notes
1 Throughout this paper football is used to refer to association football and if any other football code is referenced it will be specified which code is being discussed.
2 These competitions were selected because of their availability in open or public data sets to the authors, not for any particular relevance of the matches themselves.
3 The home advantage parameter was used in training the model because the training set of games included both league and tournament matches but it is not used in any of the in-game predictions used in this paper because the data are entirely made up of tournament games from the neutral venue 2019 Women’s World Cup.
4 These thresholds are based on UEFA’s thresholds and there is some discussion about whether they translate to women’s football since they were based on the men’s version of the game (Bradley & Vescovi, Citation2015) but they are left unchanged here because they are only used as a guide to visually analyse continuous distributions