I saw this on Twitter and thought I'd put it in here. I'm not entirely sure what it all means but it seemed relevant.
Thanks for posting, I've been searching for an update just a few days ago.
Below is the table enlarged.
The most interesting columns are the 4th, 5th, 6th, 7th from the left:
expected goals for (xG), expected goals against (xGA), difference of xG and xGA (xGD expected goal difference), comparison of that expected goal difference versus actual GD.
Looking at these columns and also doing some maths not displayed in the table, according to the xG models used, Chelsea scored about 8 goals more than the model would have predicted. They also conceded 2.6 less than the model predicted. That's how their '10.5' in the xGD vs GD column result, so it's fair to say that based on those models, Chelsea overperformed in both departments, both in attack and defence, more pronounced though in attack.
Similarly, Liverpool overperformed in attack (by about 7 goals) and slightly underperformed in defence (by about 4 goals). I don't have the time now to look it up but well possible that the latter is due to Karius against Bournemouth as IIRC, you were pretty okay in defence too in early December; I may check this later.
United should have two goals more than they have and about 2 goals less. So they underperform slightly in both departments but have decreased the gap in comparison to early December.
For a differentiated view, it's also interesting to rank the xG and xGA columns:
City created more quality chances than anybody else (expressed via the number 42.7), followed by Liverpool, Spurs, Chelsea, Arsenal, United.
Defensively, Chelsea allowed the fewest quality shots (expressed via the number 15.6), followed by Liverpool (won't bode well with those Caf users who emphatically insist on how poor your/Liverpool's defence is
), United, Spurs, City, Southampton, West Brom and Arsenal.
Columns 8 and 9 I don't find particularly helpful in the current situation (xG and xGA adjusted to 90 minutes) as long as all teams have played the same amount of games and roughly the same minutes.
Columns 10 and 11 just show the number of shots taken and conceded per game without evaluating quality, columns 12 and 13 display conversion rates (most are used to goals per shot but shots required per goal gives you the same idea). The guy who's doing it like that on his blog like this also has four categories each for attack and defence depending on a team's performance, hence the next two columns.
The 4th and 3rd column from the right show expected goals for per shot taken (xGpSH) and expected goals against per shot conceded (xGAgSH), in other words, how many shots do you need to create quality shots (or allow your opponent to create quality shots).
TSR is total shots ratio which I personally find still valuable because it gives you the ratio at first glance (total shots taken / (total shots taken + total shots conceded) ) .
PDO is a funny thing and was supposed to reflect the amount of luck a team has. The concept (that luck regresses over a season) has been tossed by some stats guys (e.g. 11tegen11) but others still use it.