Summary of Biases in Expected Goals Models Confound Finishing Ability, by Jesse Davis and Pieter Robberechts
Biases in Expected Goals Models Confound Finishing Ability
by Jesse Davis, Pieter Robberechts
First submitted to arxiv on: 18 Jan 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Applications (stat.AP)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper investigates the limitations of using Expected Goals (xG) to evaluate finishing skill in soccer analytics. It critiques the current method of comparing a player’s cumulative xG with their actual goal output, highlighting issues such as high variance in shot outcomes and limited sample sizes. The authors propose three hypotheses: that the deviation between actual and expected goals is an inadequate metric, that including all shots in cumulative xG calculation may be inappropriate, and that xG models contain biases arising from interdependencies in the data. They find that sustained overperformance of cumulative xG requires both high shot volumes and exceptional finishing, and that there is a persistent bias that makes actual and expected goals closer for excellent finishers than it really is. To address these limitations, they develop a technique from AI fairness to learn an xG model calibrated for multiple subgroups of players. They demonstrate the effectiveness of this approach by recalculating Lionel Messi’s Goals Above Expectation (GAX) using their new model, finding that it underestimates his GAX by 17% and is even higher than typical elite high-shot-volume attackers. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper looks at how we measure a soccer player’s finishing skill. Right now, we use something called Expected Goals (xG). It compares the player’s actual goals with their expected goals based on where they shoot from. But this method has some problems. For one thing, it doesn’t take into account that sometimes shots don’t go in even when they’re good chances. And it also assumes that all players have the same chance of scoring, which isn’t true. The authors came up with three ideas to make xG better: use a different way to calculate expected goals, only include certain types of shots, and fix some biases that are built into the model. They found out that it takes both shooting a lot and being really good at finishing to do well on this metric. And they also discovered that there’s an underlying bias that makes it seem like top players are better than they actually are. |
