Great blog post! I really enjoyed the historical context around Big Chances. I’ve often felt frustrated by the lack of context in some xG models and how subjective Big Chances can feel, but I hadn’t considered how that subjectivity might actually bring valuable context to the table. This idea feels genuinely innovative.
Thinking about entropy, I see parallels with situations where a player receives a great pass right in front of goal but just misses the chance (say, too late to touch) for what could have been a 0.99 xG shot. These moments wouldn’t show up in xG, but the team was incredibly close to scoring, and those situations are worth capturing. I’m not sure if metrics like xT or xA fully account for these, as the goalkeeper’s positioning plays such a critical role.
That said, subjectivity could still be a challenge. We don’t always agree on what constitutes a “Big Chance,” and it feels like it might also be player-dependent. For instance, in a 1-on-1 situation, the attacking player’s pace or technical ability might influence whether they attempt to dribble or take the shot immediately, which could significantly affect the perceived quality of the chance.
Another key question is the threshold. Because Big Chances are binary, there’s a lot of room for noise. Even though the idea intuitively feels like it could have predictive power – if measured correctly – it’s still untested. Testing this seems tricky, as I don’t think we currently have the data to define or evaluate it reliably.
Lastly, for the general audience, I think a quick, clear definition of entropy in a sentence or two could make this more accessible. Not everyone will have the background to intuitively understand its role in the context of soccer.
That said, I love the direction this idea is headed. Thanks for the blog post – it’s sparked some great thoughts!
I like the idea — I've wanted something like this for a while — but I'm not sure it gets you to repeatability, since breakaways tend to be pretty chaotic and situational too. Without off-ball data I'd imagine an entropic big chance would be pretty much any pass attempt into the box, or if you set the threshold high enough maybe passes into the box at a certain fast break velocity, but whether those attempts came from repeatable patterns of play seems like a separate question (starting with "what does repeatable even mean in soccer").
Depending on the selected xG threshold, every pass attempt into the box may indeed qualify as an EBC and I'm not sure that's inherently a bad thing. I'm pretty certain this will provide a strong signal for building a repeatable model. Ultimately, the goal here is to strip the raw xG values of some of the noise that is introduced.
Great blog post! I really enjoyed the historical context around Big Chances. I’ve often felt frustrated by the lack of context in some xG models and how subjective Big Chances can feel, but I hadn’t considered how that subjectivity might actually bring valuable context to the table. This idea feels genuinely innovative.
Thinking about entropy, I see parallels with situations where a player receives a great pass right in front of goal but just misses the chance (say, too late to touch) for what could have been a 0.99 xG shot. These moments wouldn’t show up in xG, but the team was incredibly close to scoring, and those situations are worth capturing. I’m not sure if metrics like xT or xA fully account for these, as the goalkeeper’s positioning plays such a critical role.
That said, subjectivity could still be a challenge. We don’t always agree on what constitutes a “Big Chance,” and it feels like it might also be player-dependent. For instance, in a 1-on-1 situation, the attacking player’s pace or technical ability might influence whether they attempt to dribble or take the shot immediately, which could significantly affect the perceived quality of the chance.
Another key question is the threshold. Because Big Chances are binary, there’s a lot of room for noise. Even though the idea intuitively feels like it could have predictive power – if measured correctly – it’s still untested. Testing this seems tricky, as I don’t think we currently have the data to define or evaluate it reliably.
Lastly, for the general audience, I think a quick, clear definition of entropy in a sentence or two could make this more accessible. Not everyone will have the background to intuitively understand its role in the context of soccer.
That said, I love the direction this idea is headed. Thanks for the blog post – it’s sparked some great thoughts!
I like the idea — I've wanted something like this for a while — but I'm not sure it gets you to repeatability, since breakaways tend to be pretty chaotic and situational too. Without off-ball data I'd imagine an entropic big chance would be pretty much any pass attempt into the box, or if you set the threshold high enough maybe passes into the box at a certain fast break velocity, but whether those attempts came from repeatable patterns of play seems like a separate question (starting with "what does repeatable even mean in soccer").
I like "Entropic Big Chance"!
Depending on the selected xG threshold, every pass attempt into the box may indeed qualify as an EBC and I'm not sure that's inherently a bad thing. I'm pretty certain this will provide a strong signal for building a repeatable model. Ultimately, the goal here is to strip the raw xG values of some of the noise that is introduced.