You may hear the term xG used more and more right now, with analysts and data experts springing up all over the place.
Once upon a time, a team usually had one or two fan-driven media outlets, such as this, but now there almost always seems to be an analytics page dedicated to a club too. The Imps have one, Impanalytics and a quick visit to that page immediately brings up a discussion on xG. These accounts deal in stats, analysing the game and the numbers to produce data that interests people like me.
As you may know, I also subscribe to Wyscout, a football-industry platform which creates a huge amount of data for journalists, clubs, scouts, agents and the like to pour over. Some might say stats are rubbish (happy new year Jack) and they can only tell you what you want to see, others rely heavily on them to pinpoint weaknesses, strengths and trends.
Expected goals, almost always referred to as xG, is one such number which I know divides people. I first came across it on the NTT20 podcast, hearing it on their betting preview. It fascinated me because they rolled it off as though it was a firm value, an undisputable stats in the same vein as goals, shots and bookings. Those are a defined number, if you score two goals, the value of your goals is two. If you make 200 passes and 100 are successful, your pass completion rate is 50%. Those are stats, facts, numbers that cannot be argued against. XG is different. It also poses a problem for me – if I want to start a sentence with xG, do I capitalise the ‘x’, as it is usually lower case? Oh, the woes of xG.
For those who do not know, let us break down xG and explain exactly what it is. In simple terms, xG is the probability that a shot will result in a goal, based on elements of the shot and the events that came before it. For instance, if Jorge Grant has a shot at goal, then it will produce an xG value – how likely that effort is to go in and become an actual goal (for which there is no abbreviation, you’ll be glad to know). The sort of variable which defines a shots’ xG rating is position on the field, body part the shot is taken with, the build-up play and how the effort was directly created. If Jorge received a pass into feet 12-yards out and the keeper slipped over as it came in, that would likely get a higher xG than a scramble in the area which sees the ball fall to him, 12-yards out, but with six players between him and the goal.
What makes xG rather ambiguous at times is the fact several different models are used. Wyscout will have their own xG model, whilst Statsbomb will have another. That means that whilst xG is a useful metric, it is also subject to a degree of interpretation. One agreed xG value is for a penalty, with 0.75 the xG value. To explain that, a goal is worth 1 (obviously), and no goal is worth 0. In terms of xG, a penalty must have a value, as it is a clear effort at goal, and usually, three in four are scored. Essentially, xG of 0.5 means a player would be expected to score 50% of the time. So, if a player has an xG of two for a game, but didn’t score, it means he has not taken good chances. This is where we hear comments such as ‘they outperformed xG’ – this means a team’s collective xG is smaller than the sum of their goals. You may have heard me on the podcast say something similar, and a classic example of this is our recent win against Wimbledon. Their xG, according to Wyscout, was 1.46 – so they should have scored between one and two goals. Their key chances came in the first half, with Pigott and Palmer both having two opportunities. Incidentally, the goal they did score on 31 minutes was rated 0.18 xG – so one in five of those goes in. That was based on the fact we gave the ball away in the build-up, the cross would usually have been defended better by Eyoma and Palmer was on a tight angle.
Our own xG for the game was 0.86, meaning we outperformed our expected goals. That’s because Edun’s goal got a rating of 0.05, meaning one in 20 of those efforts goes in. The chances created by Brennan, those crosses into the area, are not counted as they didn’t result in an effort at goal. When you consider the shots ratio was 10 from us, four on target and 11 from them, two on target, the xG is backed up by the shots data. Basically, we did do well to win 2-1 and were actually more clinical than you might give us credit for.
I know xG baffles some, but I have been asked about it quite a bit recently and hope this article goes some way to helping you understand what it is, for the next time I go on about it on the podcast!
That leads me on to a recent tweet I saw about xA, or expected assists. Again, I have mentioned this before on the podcast, maybe once, but I’m sure people switched off to it. Rather than me go text-heavy as I have done here, I’ll refer you to a short video by Opta, which covers xA. It does feel a little more ambiguous than xG, but this might explain it better for you statheads.