"Over the past five games Portland is posting a 1.4 goals per 90 goal scoring average, but according to FBref they have posted just 0.94 expected goals per 90 over that stretch. While xG can be noisy and sometimes not definitive"
Stat question: at what point does xG stabilize? I know it's pretty much useless as far as single games (No, really - it is! Despite everyone using it that way! Don't use it that way! Way too much noise!), and five games seems like a somewhat small sample, although better. But is xG better looked at a season at a time? A half season? I've never really known and am curious.
Just curious what you think is a better (less 'noisy') metric for chance creation in a single game. (The general idea behind xG is that, no, it doesn't take in all the variables of a single SOG during a match, but because it relies on 10k+ similar shots in its database that it has 'stabilization' built in. So it isn't perfect representation of the chance of a single shot scoring, but it does say "this % of shots from this location/shot type have been scored in the past". To me, that's helpful info to have in determining whether or not you're creating quality scoring chances in a single match.)
Honestly when looking at a single game, I think the eye test is most reliable when looking at quality of chances. Obviously that’s subjective and not something we can quantify, but xG doesn’t take into account so many different aspects of the game and I believe can be very misleading
Yep. In a single game, a shot that might otherwise carry a high xG doesn't account for unexpected defensive positioning/a mishit shot/deflection that happens in one particular situation that doesn't happen in 100 others, which is why it takes a certain amount of time to level out and become a reliable indicator of shot quality or whether the shot "should" result in a goal.
Over time, a lot of that noise quiets down and it's easier to get a picture of shot quality from a given part of the pitch, but for a single shot in a single game, it's not a great indicator.
I'm not sure there is one. I'm not trying to criticize xG or the idea of metrics like it; I just know that xG is mostly noise and no signal in a single game, and was wondering how long it took to be reliable as an indicator.
I think there's a flaw in your question. xG's aren't generated in a silo by each team, or even over a season or several seasons. It's generated by a giant database of records of 10's of thousands of matches. It only gets more accurate as more actions are added to the database, so an individual team's season is only a grain of sand in the beach of data. "Reliability" is a terribly vague complaint... What is your expectation of how xG should function?
I don't have one. I understand how it functions and how it's built. I was just asking when I can take it seriously for a team in the course of a season. I think of it in terms of baseball - it takes a certain number of at bats for rate stats in baseball to stabilize and be meaningful (depending on the stat, it can be 100-150 AB), and I was wondering what that point is for xG.
Individual shots, to use your metaphor, are absolutely grains of sand; my question is, over the course of a season, when do they form a beach. Or at least a sandbar. When can I look at xG and have it be meaningful, as opposed to in-game xG, which really is not?
Post-shot xG is a more useful small-N metric…and it’s also typically less (often MUCH less…) than the raw xG stat.
So you’re probably safe concluding that the current-to-date form the Timbers attack is well below 1G/game. Which, needless to say, is a big reason why they’re so poor, and which also matches the eye test; poor quality chances, typically either not on frame or saveable.
"Over the past five games Portland is posting a 1.4 goals per 90 goal scoring average, but according to FBref they have posted just 0.94 expected goals per 90 over that stretch. While xG can be noisy and sometimes not definitive"
Stat question: at what point does xG stabilize? I know it's pretty much useless as far as single games (No, really - it is! Despite everyone using it that way! Don't use it that way! Way too much noise!), and five games seems like a somewhat small sample, although better. But is xG better looked at a season at a time? A half season? I've never really known and am curious.
Just curious what you think is a better (less 'noisy') metric for chance creation in a single game. (The general idea behind xG is that, no, it doesn't take in all the variables of a single SOG during a match, but because it relies on 10k+ similar shots in its database that it has 'stabilization' built in. So it isn't perfect representation of the chance of a single shot scoring, but it does say "this % of shots from this location/shot type have been scored in the past". To me, that's helpful info to have in determining whether or not you're creating quality scoring chances in a single match.)
Honestly when looking at a single game, I think the eye test is most reliable when looking at quality of chances. Obviously that’s subjective and not something we can quantify, but xG doesn’t take into account so many different aspects of the game and I believe can be very misleading
Yep. In a single game, a shot that might otherwise carry a high xG doesn't account for unexpected defensive positioning/a mishit shot/deflection that happens in one particular situation that doesn't happen in 100 others, which is why it takes a certain amount of time to level out and become a reliable indicator of shot quality or whether the shot "should" result in a goal.
Over time, a lot of that noise quiets down and it's easier to get a picture of shot quality from a given part of the pitch, but for a single shot in a single game, it's not a great indicator.
I'm not sure there is one. I'm not trying to criticize xG or the idea of metrics like it; I just know that xG is mostly noise and no signal in a single game, and was wondering how long it took to be reliable as an indicator.
I think there's a flaw in your question. xG's aren't generated in a silo by each team, or even over a season or several seasons. It's generated by a giant database of records of 10's of thousands of matches. It only gets more accurate as more actions are added to the database, so an individual team's season is only a grain of sand in the beach of data. "Reliability" is a terribly vague complaint... What is your expectation of how xG should function?
I don't have one. I understand how it functions and how it's built. I was just asking when I can take it seriously for a team in the course of a season. I think of it in terms of baseball - it takes a certain number of at bats for rate stats in baseball to stabilize and be meaningful (depending on the stat, it can be 100-150 AB), and I was wondering what that point is for xG.
Individual shots, to use your metaphor, are absolutely grains of sand; my question is, over the course of a season, when do they form a beach. Or at least a sandbar. When can I look at xG and have it be meaningful, as opposed to in-game xG, which really is not?
Post-shot xG is a more useful small-N metric…and it’s also typically less (often MUCH less…) than the raw xG stat.
So you’re probably safe concluding that the current-to-date form the Timbers attack is well below 1G/game. Which, needless to say, is a big reason why they’re so poor, and which also matches the eye test; poor quality chances, typically either not on frame or saveable.