I wonder if there is a way to factor in goals as part of the path dependence of the outcomes. an obvious instance in the actual game might be if someone misses a penalty (whatever xG that is, maybe 0.8) and someone taps it in as a follow up (lets say xG 0.9). Then that's 72% chance of two goals appearing in each simulation, even though that couldn't occur in real life. Obviously this is an issue with xG too, but maybe a weighted reduction in the simulator (increase RNG range ?) immediately following a simulated goal
Very interesting, and thanks for sharing the methodology! I like blending xg with non-shot xg for this, that feels a lot more rounded than classic xpoints. My only methodological concern would be including post-shot xg. Although in theory it makes sense to reward finishing, the problem at the moment is that it doesn't include goalkeeper shot stopping. Eg if Raya had instead saved the two west ham shots (with the same xg and post-shot xg) that doesn't move the needle on "deserves to win" as I understand your current implementation. So including psxg rewards strikers for getting things on target, but doesn't reward astonishing stops. Not sure how you'd capture both though...
Also, if you keep playing with this "toy", i would be curious how this stat from MoneyPuck's explanation comes together for Premier league or Arsenal matches: "While the Meter is not meant to be a prediction of who will actually win the game, teams that finish the game at above 50% on the meter have historically won 64% of games."
I wonder if there is a way to factor in goals as part of the path dependence of the outcomes. an obvious instance in the actual game might be if someone misses a penalty (whatever xG that is, maybe 0.8) and someone taps it in as a follow up (lets say xG 0.9). Then that's 72% chance of two goals appearing in each simulation, even though that couldn't occur in real life. Obviously this is an issue with xG too, but maybe a weighted reduction in the simulator (increase RNG range ?) immediately following a simulated goal
"All of these estimates for win, loss, and draw are then combined and then go into a final number that is presented in the ‘Deserve’ to Win-O-Meter."
How are these estimates combined? Is the average the outcome of each simulation?
yeah, no weighting put on them right now.
Thank you!
Here’s a question Scott, how many goals has Arsenal / City / Liverpools No. 9 scored this year ?
Ie for Arsenal the player played ‘up front’ either Jesus / Havertz / Leo ?
Very interesting, and thanks for sharing the methodology! I like blending xg with non-shot xg for this, that feels a lot more rounded than classic xpoints. My only methodological concern would be including post-shot xg. Although in theory it makes sense to reward finishing, the problem at the moment is that it doesn't include goalkeeper shot stopping. Eg if Raya had instead saved the two west ham shots (with the same xg and post-shot xg) that doesn't move the needle on "deserves to win" as I understand your current implementation. So including psxg rewards strikers for getting things on target, but doesn't reward astonishing stops. Not sure how you'd capture both though...
Also, if you keep playing with this "toy", i would be curious how this stat from MoneyPuck's explanation comes together for Premier league or Arsenal matches: "While the Meter is not meant to be a prediction of who will actually win the game, teams that finish the game at above 50% on the meter have historically won 64% of games."
That would be interesting to try and get a sense of
I was always curious how "10,000" simulations worked. This seems to be way less complex than my mind imagined it would be.