Are you confused about the ‘xG’ (expected goals) stat flying about recently? Wolves Analytics guests posts to explain why these stats will shape the future of analytics in football.
I was listening to a recent episode of the podcast, and the conversation came up about Expected Goals.
This is something that has become more mainstream over the past 12 months, to the extent that it is now used on Match of the Day.
Many of my Twitter posts use the idea of Expected Goals (or xG, for those too lazy to write out the full thing), so I thought it was probably worth putting together some of explanation.
Perceiving The Game
First of all, let me ask you a question.
After the Arsenal game on Sunday, were you thinking a) “Wolves scored one and Arsenal one. Therefore both teams drew and nothing else mattered”, or b) “bloody hell, how didn’t we win that? I can’t believe Jota missed, Gibbs-White hit the post and they scored a complete fluke”.
If you thought the former, this article probably isn’t for you. If you thought the latter, then read on…
xG in essence measures just how good a chance it was.
Should Jota score that chance 10 times out of 10? 6 times out of 10? 4? Only once?
Alternatively, if Henrikh Mkhitaryan swings that ball in 10 times, how often does it end up in the back of the net? Goal chances are not born equal. xG measures the quality of these chances.
Don’t believe what you see
But let’s not get ahead of ourselves. At this stage of the season, teams have played a handful of games, and after 12 games the league table has a habit of lying.
It might be that teams have had two games where they have played brilliantly, but been hit by a sucker punch. Or a team has been on the wrong end of a couple of bad refereeing decisions.
It is often said that luck evens itself out over the course of a season, but it certainly is not even over a 12 game stretch.
On this date twelve months ago, Crystal Palace were rock bottom with just one point, while Brighton and Huddersfield were both flying high in the top half.
It’s fair to say that the league table at that point was not completely indicative of the way it would look at the end of season
So analysts, both within and outside clubs, have been looking for ways to measure ‘performance’ at this stage of the season, rather than points.
The most obvious way to do this is to look at the number shots that a team takes and the number of shots a team concedes.
Generally, better teams will take more shots than they give up. Worse teams will give up more shots than they take. Wolves, so far this season, feature very well:
|Team||Shots For||Shots Against||Shot Difference|
But this is not an ideal measure of performance.
Of our 173 shots, 81 have been from outside the penalty area. Ruben Neves alone has taken 31, the most in the Premier League.
Bearing in mind he’s scored one of those shots this season, could you say a Neves shot from 40 yards is as good a chance as Jota’s miss? Or Helder Costa’s chance against Tottenham? They clearly are not chances of equivalent quality.
The Vision Of xG
This is where xG comes in.
xG is a way of measuring the quality of these chance. xG models take the Jota chance, for example, and look at how often a goal has been scored from equivalent situations.
There is a really good explanation of how it is calculated here, but some of the things that will be considered are:
- Location of shot: a shot from inside the 6-yard box has a better chance of going in than one from 50 yards
- Type of shot: a header is more difficult to score than a shot with the foot
- The assist: typically, it’s easier to score from a through ball than a deep cross
- Speed of build up: if a team attacks quickly, there’s a chance defenders will be out of position. Slow build up play will often mean defenders are well placed.
There are models that are taking this even further, and consider things like the position of defenders; a striker who is under intense pressure from a defender is more likely to snatch a shot than a striker who has all the time in the world to pick their spot.
xG models take these factors, and compares them to a database of thousands of shots, looking for shots with similar characteristics.
Back To London
So let’s go back to the Arsenal game, and the chances that we had at the end.
Based on all of those factors above, InfoGol gave Jota’s chance a scoring probability of 0.52 (that is, that chance would, on average, be scored 52 times out of 100).
Comparatively, Morgan Gibbs-White effort was given a scoring probability of 0.05 – you would expect a shot in that situation to be scored once in every 20 shots.
Add up the scoring probability of the 12 shots that we had during the game (Cavaleiro’s goal had a probability of 0.47, Costa’s first half chance had a probability of 0.21 and Traore’s at the end was 0.21, plus a few pot shots with low probabilities), and you get a total xG of 1.78
When you start to add up the xG for and xG against over the course of a season, it starts to paint a picture of how well a team is performing.
At this time last season, we went through a real purple patch where everything we hit went in.
At that stage, our results (measured by goal difference) were better than the performances (measured by xG).
This generally isn’t sustainable – you can’t rely on Barry Douglas scoring a freekick every week.
Fast forward a few weeks, and the goals had dried up, but the performances (xG) remained fairly steady.
By the end of the season, the actual goals scored and conceded was pretty close to what the expected goal models say.
The Essence of xG
So, let’s look at an xG table for the Premier League this season:
This generally suggest that Wolves have done well this season, and when considering xG, we have been the fifth best team in the league, better than Arsenal and Manchester United.
This doesn’t mean that we will finish fifth; we are almost certainly too far behind in terms of actual points to have a hope of making that gap up.
But it does mean, that from an xG perspective, our performances have been far better than our current position in the table would suggest.
This has been a brief overview of xG.
If you made it this far, give yourself a pat on the back – I’m sure I lost some readers after 3 sentences.
But if this has interested you, there is a fairly well established xG community on Twitter – I’d suggest @StatsBomb, @mixedknuts, @footballfactman, @MC_of_A, @TheM_L_G and @experimental361 for starters. Oh, and of course, @WolvesAnalytics.
Listen to the podcast, every week for FREE, here.