It’s a new season and I’ve got a fair whack of new followers on twitter over the summer that might be new to all this stuff.
With that in mind I’ll start with the Gylfi Sigurdsson miss in the first couple of minutes. The debate was sparked on my regular Everton forum with a poster stating we should have been 2-3 goals up at half time. Here’s the chance:
Watching it in real time it’s fairly difficult to judge how much time he has at point of strike (which in itself tells you my view on using words like ‘should’ when describing goalscoring chances). Here’s the still:
To me, Patrick van Aanholt (No.3) has got much of the left side of the goal as we look at it covered. Keeper Vicente Guaita is positioned nicely down the middle and set well for a shot straight at him or a dive to the gap in the right. That gap will be closed by Scott Dann (No.6) to a great extent by the time the ball’s reached the 6 yard box.
The phrase ‘should have scored’ implies to me that it’s a better than 50% chance that this ends up a goal. I posted up the still image on the forum. Only one person involved in the discussion was willing to put a % on it. They stated that penalties are only scored about 70% of the time so add in a moving ball, two defenders and a keeper and you were looking at 20-25%. Everyone else involved stuck to their guns that Sigurdsson should have buried this.
What do the expected goal models (google it) say? Understat say 37%, Infogol say 35%, Statsbomb say 13%, my model says 17%.
I’m pretty sure Understat and Infogol use Opta’s ‘big chance’ marker to beef up these type of chances. ‘Big chance’ is a subjective measure the people recording the match for the data company add in. No subjective markers in my numbers. Nor in Statsbomb’s as far as I’m aware and they also model defensive pressure and keeping position in theirs too which the others don’t.
To cap it off, the guy on the forum who got it pretty much level with the models ended up saying ‘but stats like that just make football a bit boring, like VAR does’.
I find it fascinating, I hope you do too! Everyone who models football in these kinds of ways pretty much agree that we overestimate scoring chances when watching the game.
Based on my model, Everton were most likely to score just once during the whole game, let alone by half time. The likelihood of scoring 2-3 during the whole game was still less than 30% (23.16+6.81). The game finished 0-0.
All the models agree that Everton shaded the game in terms of expected goals, did enough to score and probably win the game. Using Danny Page’s simulator my model states that Palace win this game 26% of the time, it’s a draw 33% of the time and Everton win 41% of the time.
A pretty solid start from The Toffees away from home, especially considering the model suggests that Palace at home would have been restricted to a goal or less 82% of the time.
Even if you think this is a load of wotsit, please give watching the game through this kind of prism a go. Just once. For me, like.