Advanced soccer analytics: building and applying a pitch control model in Python
youtube.com
youtube.com
Football statistics were traditionally based on specific event: passes and shots. From these you can compute certain statistics like % of Possession (contrary to what it may look, % Possession is calculated from passes, not actual possession time) and Shots on Target.
Football is notoriously a low scoring sport, and shots differ in quality quite a bit, so a measure was created to address this: Expected Goals (xG). This was around 2010, and only this season hit the mainstream as the Premier League broadcasters started to present those values (based on Opta's model).
More advanced stats, but similar in concept, were created since, like Expected Assists and xG Chain (in this case, a value is attributed to each player that participated in the possession chain).
But even shots are kind of rare (usually around 10 shots on target per match), and these stats completely disregard the defensive side of the equation, so increasingly full positional data is used in Football Analytics.
In 2018, William Spearman presented an influential paper at MIT Sloan Sports Analytics Conference called "Beyond Expected Goals" (this video is an open implementation of that paper)[1]. He was later hired by Liverpool FC as their lead Data Scientist.
You can watch a video by Spearman himself about the Pitch Control Model and recent innovations here.[2]
As you can see, this is pretty close to the state of the art in Football Analytics. It's a huge moment that very few people noticed, so I'm trying to get it out there.
[1] http://www.sloansportsconference.com/wp-content/uploads/2018...