This is quite interesting: https://understat.com/league/EPL Its based on the expected goals measure which seems to be a statistical prediction of the odds on scoring from any attacking attempt. It seems we are pretty close to expectation being about two goals down on expectation for goals scored and two up on goals conceded. Man Utd are a massive outlier on both measures such that they should have 17 fewer points if they were bang on target on both measures. I guess that shows they have fantastic strikers and a brilliant goalkeeper and/or that defenders and attackers both panic against then when the ball is near the goal. Palace, Southampton and West Brom would have a lot more points on this measure that they've actually got. Not sure what to deduce to be honest.
Firstly : https://understat.com And then ... "Its based on the expected goals measure which seems to be a statistical prediction of the odds on scoring from any attacking attempt. ... Not sure what to deduce to be honest." That the "quality" of Spurs total chances per game results in the total goals closely aligning with the expected goals. Beyond that, nothing. You don't have the definition of the 10 "features" their ML algorithms are using to classify/weight each shot on goal.
There's two things about the modern football statto I can't stand: talking up xG as if it's the Word of God, basing all knowledge of players on football radars even though those things make absolutely no sense to anybody. Still, it could be worse - there's the ones who use a player's FIFA stats as the basis of how good they are...
For ManU it could be the statistical expression of what a few people "feel", which is that they don't seem to be doing as well as their second place would suggest. In fact based on many of their performances you might argue they are the weakest team in the top 5.
Like all statistics they probably tell you something but exactly what is hard to fathom. And confirmation bias is going to be strong in this one. For example it made me think that our finishing is bang average so our success must be due to making lots of chances.
Spurlock : if you don't want to know the result, look away now ... The supporting stats are not there. So here is a trivial example. The team has 10 shots on goal on target. Their system assigns a probability P that each shot results in a goal. They create a histogram with 10 probability bins (0,1, 0,2, ... 1.0) , and populate it with the shot probabilities. Suppose : - the shot distribution is uniform (there is one shot in each bin) . - their system has determined that a shot with P >= minimum P (MP) should result in a goal. - they set MP = 0.8 then this means we expect : - 2 goals to have been scored - the goals to have come from those shots that are in the 0.8/0.9 bins As you have neither the histogram nor MP, you have stats that you cannot debate. So until I get that info (as well as the 10 "feature" definitions their ML system uses to compute P) , I say : MEH.
