Been bugging me slightly for a while but most xG models that I have seen have Liverpool in second place behind City, which is obviously something riling Guardiola as well.
I am biased so I am not willing to accept that Liverpool have been lucky this season, nor have I ever thought that xG models were robust enough to pass off as anything more than entertainment draped in dumbed down statistical sophistication for football audiences, given that the definition of a chance is highly subjective.
There have been more than a few occasions this year when I have thought that there is something drastically wrong with how the data is being presented, or is it just that I am too biased to think rationally?
Glad to hear what those that work with football statistics have to say about it.
(all data from
https://understat.com/league/EPL)
Our xG differences in isolation aren't huge. We've scored 7 more than expected - the main beneficiary being VVD who has scored 3 but would have only been expected to score 0.59 - and conceded 3 less than expected. Leicester and Southampton have the highest swings with +18 and -13(/14) respectively.
The only games where we've had a lower xG than our opponents are:
- Southampton (A). Won 2-1 but would have expected to lose 2-1. Southampton have a net -13/14 xG across the season so that's just in keeping with their general performance at both ends of the pitch.
- Chelsea (A). Won 2-1 but would have expected a 1-1 draw. Chelsea have a net -8/9 xG across the season so likewise.
- Man City (H). Won 3-1 but would have expected a 1-1 draw. Fabinho scored a wonder goal - it happens from time to time - while half their xG came from Kyle Walker's missed header very late on in the game.
(
https://understat.com/team/Liverpool/2019)
It should also be noted we had a higher xG than Man U in the only game we dropped any points. Ignoring that game though the above three results would have cost us 7 points and taken us from 55 points to 48 points. Even had we lost them it would still be 46 points.
Where it gets screwy - more-so than xG already is anyway - is when that's extrapolated to xPTS (expected points). The league table has us on an xPTS of 39.49. That's 15.51 points less than we actually have even though we've only had those three games where we didn't have a higher xG than our opponents and also drew a game where we did have a higher xG! It just seems the model doesn't cope well when a team does enough to win without running rampant.
If you want to cut through everything though and see how flawed xG is then consider the Son goal against Burnley, the one where he picked up the ball on the edge of his area and ran through the Burnley team before slotting past Nick Pope. The definition of a wonder goal and not one I can remember anything remotely similar to (George Weah for AC Milan, maybe). xG has that as a 55% chance of resulting in a goal (a penalty is only 76%!) because of where and how the final shot was executed.