Thanks for this write up. It seems to address something I saw in 2023 that has always bothered me about xG. Houston Dynamo's Corey Baird was put in behind with a looping awkward ball on about the 12' spot. GK was able to come out and nearly get to the ball first. Very difficult situation for Baird but he managed to just get their first and steered the ball towards goal. GK blocked that shot but it deflected up and over the on-rushing GK, falling to Baird's feet now on 5' line, past the GK. No defender close. Easy tap in goal. Baird was given an xG of .7 for the first difficult deflection against a GK basically at the same spot as he, then an xG of .3 for the open goal sitter. I probably need to read this article several more times to understand but it seems to address that situation. I don't get how those xG values make any sense.
I just can't figure out how xG really adds any insight from this. Seems we are adjusting the original .3 chance to 1, because the rebound randomly fell for Baird. That happens in that situation something like 1 time out of 50 or something like that in that situation. Your basically taking a half chance and adjusting it up to 1 because a very rare rebound fell for him leading to easy goal. That doesn't make sense to me.
One approach that would make sense to me would be to adjust the original chance up to .4 or something. The second shot is a continuation of the first shot, so it modifies the first shot xG by the chance of the rebound happening and being scored. While I say that makes sense, I see all sorts problems with that approach. But a method that randomly adjusts a chance up from .3 to 1 makes no sense to me. Why would it not be a better measure of everything xG is trying to measure to just count the whole chance as 1 chance: .30 xG.
Maybe my comment was not clear. For your 0.7 and 0.3 example, you would apply the formula in the blog post which gives you a 0.79 xG total for the entire sequence.
Ah. Ok. Yeah, I missed that. So your suggesting the method used at the time would be improved the way you say. That makes sense and would be an improvement over what they did.
It should be xG1 plus the product of (1 - xG1) and xG2 right? Unless I’m being dense about something (which is likely) or it’s not displaying correctly in the mobile version
Thanks for this write up. It seems to address something I saw in 2023 that has always bothered me about xG. Houston Dynamo's Corey Baird was put in behind with a looping awkward ball on about the 12' spot. GK was able to come out and nearly get to the ball first. Very difficult situation for Baird but he managed to just get their first and steered the ball towards goal. GK blocked that shot but it deflected up and over the on-rushing GK, falling to Baird's feet now on 5' line, past the GK. No defender close. Easy tap in goal. Baird was given an xG of .7 for the first difficult deflection against a GK basically at the same spot as he, then an xG of .3 for the open goal sitter. I probably need to read this article several more times to understand but it seems to address that situation. I don't get how those xG values make any sense.
Hey Donald! It explains exactly how to address this sort of situation! Intuitively, the xG total for a sequence should be capped at 1.0.
I just can't figure out how xG really adds any insight from this. Seems we are adjusting the original .3 chance to 1, because the rebound randomly fell for Baird. That happens in that situation something like 1 time out of 50 or something like that in that situation. Your basically taking a half chance and adjusting it up to 1 because a very rare rebound fell for him leading to easy goal. That doesn't make sense to me.
One approach that would make sense to me would be to adjust the original chance up to .4 or something. The second shot is a continuation of the first shot, so it modifies the first shot xG by the chance of the rebound happening and being scored. While I say that makes sense, I see all sorts problems with that approach. But a method that randomly adjusts a chance up from .3 to 1 makes no sense to me. Why would it not be a better measure of everything xG is trying to measure to just count the whole chance as 1 chance: .30 xG.
Maybe my comment was not clear. For your 0.7 and 0.3 example, you would apply the formula in the blog post which gives you a 0.79 xG total for the entire sequence.
Ah. Ok. Yeah, I missed that. So your suggesting the method used at the time would be improved the way you say. That makes sense and would be an improvement over what they did.
Great read (typo in the formula, missing the +)
Thanks, Ron! I don't see a missing plus sign?
It should be xG1 plus the product of (1 - xG1) and xG2 right? Unless I’m being dense about something (which is likely) or it’s not displaying correctly in the mobile version
Sounds like latex isn't properly rendering on mobile.
Yeah that's weird, just saw that it's right on browser