Elliot rejoins the analytics pod to ruin it.... erm...contribute to it with Scott. They break down the data and discuss Leicester City, Lacazette, United at the weekend, and aesthetics versus results.
Comments
Instead of re-watching just a half, maybe doing segments/highlights of the game is better. Really wanted insight on the Vardy goal.
2020-10-30 12:30:54 +0000 UTC
They do, and I confess to not knowing exactly what factor they use to blend the two, but at least that's 2 events. For this, any one of 3 shots eliminates the probability other two, so that's got to be a pretty complex blending process, and I have no idea what an appropriate factor is. Not to mention what do you do if you're trying to model the mid-post miss - do you give partial xG for the near post miss, but not the far post, as it hasn't happened? etc. Really fascinating stats question and I really liked Scott's answer that you have to reflect the value of the cross more than that of the whiffs and using buildup and non-shot stats. Things like Scott's threat graph really tell the story of the game better for me!
2020-10-29 23:46:53 +0000 UTC
Yes that’s true! As for the latter scenario, the xG models already account for similar situations I think. For example if a 0.6 shot is saved and the rebound is a 0.7 shot that scores, they don’t give 1.3. They blend it
2020-10-29 23:38:28 +0000 UTC
@Elliot, reflecting on your question of registering 'whiffed' shots, a big thing to remember is that you have to have tracking data to consistently build a match report, so you have to have a way to consistently register when a player whiffs a shot. I think that's a pretty complex ask (not impossible, but complex).
Also have to account for the whiff leading to other chances - the classic cross that flies just in front of the near post runner and penalty spot runner and far post runner all at the same time all just missing it. Each one looks to have a clear header, but a single cross can't register as "3 xG".
Enjoyed it quite a bit guys!