Miss it like Messi: Extracting value from off-target shots in soccer

Ethan Baron ¹, Nathan Sandholtz ², Devin Pleuler ³

⁰University of Toronto, Toronto, ON, Canada.

Journal of Quantitative Analysis in Sports +

|

March 13, 2024

View abstract on PubMed

Summary

This study introduces new soccer analytics metrics that value off-target shots, unlike current models. These novel metrics better measure shooting skill by analyzing shot trajectories, offering improved stability and predictive power.

Area Of Science

Sports Analytics
Football Analytics
Performance Measurement

Background

Measuring soccer shooting skill is difficult due to limited scoring data and contextual factors.
Existing advanced metrics like expected goals added and post-shot expected goals improve upon conversion rates but ignore off-target shots.
Off-target shots, comprising nearly two-thirds of all attempts, are currently assigned zero value in all developed metrics.

Purpose Of The Study

To propose novel soccer shooting skill metrics that incorporate data from off-target shots.
To develop a player-specific generative model for shot trajectories to quantify skill in off-target attempts.
To demonstrate the improved stability and predictive power of these new metrics compared to existing state-of-the-art methods.

Main Methods

Developed a player-specific generative model for soccer shot trajectories using a mixture of truncated bivariate Gaussian distributions.
Utilized the generative model to compute new shooting skill metrics that assign non-zero value to off-target shots.
Evaluated the proposed metrics against current state-of-the-art metrics for stability and predictive accuracy.

Main Results

The proposed metrics successfully incorporate the skill signal present in off-target shot trajectories.
New metrics demonstrate greater stability compared to existing advanced soccer analytics metrics.
The enhanced metrics exhibit increased predictive power in evaluating shooting performance.

Conclusions

Off-target shot trajectories contain valuable information for assessing soccer shooting skill.
The developed generative model and associated metrics offer a more comprehensive approach to measuring shooting performance.
These novel metrics provide a more accurate and stable evaluation of player shooting ability in soccer analytics.

Keywords:

Bayesian hierarchical model generative model mixture model player valuation shot trajectories spatial data