Some good discourse here around the merits of xG as a statistic. Good to see most are correctly understanding its limited explanatory power, and key caveats (gamestate and variance). At no point has anyone (competent) claimed it's the be all and end all.
A couple of applied points
Stranded WoodleyRoyal so humour me here, are 'expected goals' supposed to predict the amounts of goals you are going to score before the game has even kicked off?
All they show is if your strikers we're perfect in a game then given where each shot was taken from, how many of those should have resulted in a goal. So for example a penalty is "worth" 0.7 goals, as a penalty is scored 70% of the time. So 7 shots where you have just 10% chance of scoring are worth 1 penalty - so in short, it is likely the more shots you have the higher your xG even if all your shots were from distance and frankly non-threatening.
According to here:
https://experimental361.com/2020/10/28/ ... -oct-2020/we were lucky to score even 1 last night where as Blackburn should have definitely scored at least 1 given their chances.
This is a good summary of how its measured, but missing a key point on 'variance'. xG doesn't 'add' up, its multiplies (probabilistically), so it's naïve to say you would expect goals scored/conceded to equal the rounded sum of xG.
Here is a tool to play around with turning xG numbers its distributions, allowing a much more informed interpretation of the stat. Its pre-loaded with the (obviously incorrect) numbers from the Blackburn game.
(Confusingly, Red = Blackburn, Blue = RFC)
Picking up on the specific points made by Stranded:
- "we were lucky to score even 1 last night" - given the shots we had, we would have expected to score >0 goals 63% of the time, so would have been quite unlucky not to score.
- "Blackburn should have definitely scored at least 1"- given the shots they had, Blackburn would have expected to score >0 goals 83% of the time, so I think that assessment was fair.
Based on those numbers, we had a 18% probability of winning that game, so unlikely, but far from a robbery. This overly simplistic point also misses the fact that we were leading for all but 12 mins of the game; that subtlety is also missing from the xG => xP style analyses such as the Experimental 361 tables. But we also know those numbers (via Opta) are wrong. If we say Meite's open goal was actually 0.8xG (rather than the 0.27 which doesn't account for the lack of defenders or goalkeeper), we get
these distributions. In that case the probability of winning has almost doubled to 31% (vs losing at 37%, a near toss-up). Then take into account that the model missed one near certain OG (as no shot was recorded). The model quality is limited to the quality of it's input data after all
WestYorksRoyal If you replaced Aguero at City with Puscas, would it still show the same xG for chances they get? That clearly shows the fallacy of xG if so. Would you say City's or Spurs' xG figures are unsustainable because Aguero and Kane are outperforming their xG? Obviously not.
I don't see why Joao and Meite can't keep this form up based upon what I've seen, so why can't we continue outperforming xG?
WYR is right to say it seems odd that 'finishing skill' isn't accounted for, and yes the same xG would be awarded regardless of whether the player taking the shot was Aguero, Puscas, or even Rafael, which does seem bizarre on the face of it. But finishing skill doesn't seem to hold up in the long term. Very few players consistently outperform their xG numbers even though they are considered 'elite' finishers. Understat.com has xG figures going back 6 seasons for the top 5 leagues, so we can try to find this 'finishing skill'. Here we see:
-
Messi - 202 goals vs 169 xG, and outperforming in 5/6 full seasons. Truly the GOAT, with good evidence he is better-than-modelled finisher
-
Ronaldo - 189 goals v 182xG, pretty much bang on apart from one season of massive outperformance in 14/15 (he exceeded xG by 8.6). Loves a shot, but so many go high and wide...
-
Lewandowski- 172 goals vs 175 xG, a tiny underperformance despite being regarded as one of the best finishers in the world
-
Aguero 128 goals v 122xG
-
Kane- 145 goals v 121xG, outperforming in every season, potentially an elite finisher
-
Neymar- 107 goals vs 113xG
My interpretation is that its overly simplistic to distill an elite striker into an ability to finish well - ALF could finish very well, yet wasn't an 'elite' striker. Instead, the top strikers have other skillsets that allow them to consistently generate shooting opportunities, such positioning, anticipation, balance and strength. How about instead of calling Lewadowski a lethal finisher, we talk about how good he is at making space and being in the right place to consistently score so many goals?