Space-Creating versus Dead Possession: An Off-Ball Possession-Quality Index for Broadcast Football
Seongjin Choi
Read on arXiv →Key claim
New metrics reveal the true value of football possession.
In plain English
Imagine you're a football coach trying to analyze your team's performance. You might look at possession statistics, but those numbers can be misleading. For instance, having 60% possession doesn't tell you if that time was spent effectively pushing the opponent back or just passing the ball around without creating any real threat. This is what's called a 'sterile possession.' Current models focus on on-ball actions, like shots or passes, but they miss the bigger picture of how those actions impact the game. They don't account for whether holding the ball actually created space or just wasted time. This paper proposes a two-layer approach to address this gap. First, it introduces a junk-possession index that evaluates each possession sequence based on its potential threat, flagging sequences that don't contribute to scoring, especially in critical match situations. The analysis of 103 World Cup matches shows that these flagged sequences correlate negatively with points and expected goals, indicating they are indeed less valuable. Second, it develops a Space-Creation Index that uses video analysis to determine whether a possession was effective in creating space on the pitch. The results reveal that a significant portion of flagged possessions were ineffective, highlighting the limitations of traditional on-ball value metrics. This new framework allows for a more nuanced understanding of possession in football, separating valuable actions from those that merely consume time without contributing to the team's offensive strategy.
The paper introduces a new framework for evaluating football possession that incorporates off-ball actions, which is a significant extension of existing models.
The findings are supported by statistical correlations and a purposive sample from World Cup matches, though further validation could strengthen the claims.
Deep reliability assessment
The methodology supports the claim that the junk-possession index provides additional outcome-relevant information beyond on-ball action-value models. However, the spatial analysis is limited to a small sample of matches, which may not generalize across all contexts.
Reproducibility
yes, the paper provides a GitHub repository for the code: https://github.com/nowayfootball/junk-possession.
Key figure
The key architectural diagram likely illustrates the two-layer framework combining the event-side junk-possession index and the spatial Space-Creation Index.
