Author(s): MA, R., BISCHOFBERGER, J., EXEL, J., BACA, A. , Institution: UNIVERSITY OF VIENNA, Country: AUSTRIA, Abstract-ID: 1074

INTRODUCTION:
Despite the growing availability of tracking data, quantifying off-ball defensive performance remains a conceptual challenge in football analytics. Passive contributions through positioning and anticipation are hard to capture in prevalent event-based frameworks. To address this gap, we propose a two-step framework that first quantifies off-ball defensive actions using a set of dedicated metrics and subsequently integrates these metrics into a network-based representation to analyse individual and collective defensive behaviour.
METHODS:
We propose a novel set of metrics that quantify defensive involvement and responsibility based on the offensive passing actions of the opposition. Involvement is based on players’ proximity to opponent passes, while responsibility reflects the expected involvement derived from roles within dynamically detected formations. Players’ defensive behaviours are classified as either faults (being outplayed) or contributions (stopping the opponent from progression) based on the value of each offensive pass (Expected Threat). The validity and robustness of the model are evaluated on a cross-gender and cross-competition data set spanning 516 matches.
Based on this involvement-responsibility framework, Defensive Valued Networks (DVNs) are created at both the player and team level. Player-level DVNs represent the passing interactions a defender is involved in, while team-level DVNs model shared defensive responsibility between players during opposition actions. A total of 64 matches from the Men’s World Cup 2022 provided by PFF FC are used. Network metrics (total degree and network density) are correlated with measures of defensive success (goals and shots against) to examine the effectiveness of our approach.
RESULTS:
All four types of metrics (involvement, responsibility, fault and contribution) show higher validity (~1SD) in a combined index of noisy proxies than traditional metrics. For example, a central defender’s market value is much more strongly associated with their fault (competition-averaged r = -0.423) than with their number of interceptions (r = 0.074).
Network metrics derived from fault-only DVNs show significant correlations with conceded shots for both player and team level network: Player’s network r = 0.233 (p < 0.01) and total degree r = 0.345 (p < 0.001); team network density r = 0.294 (p < 0.001) and total degree r = 0.311 (p < 0.001). If contribution is considered alone, significant correlations only exist in the shared team-level network, (density r = 0.364, p < 0.001; total degree r =0.388, p < 0.001).
CONCLUSION:
We propose involvement- and responsibility-based metrics to quantify off-ball defensive performance. When cast into a network representation, it can be used to quantify the value attributed to each individual and the team's defensive cooperation. These finding demonstrates that the presented metrics can help clubs determine which players contribute towards positive match outcomes through effective off-ball defending.