Expected Goals, usually shortened to xG, is possibly more divisive than Steve Evans.
It might even be more divisive than left and right, and over time, I’ve found that when it comes to xG, I’m a centrist. Once upon a time, I was heavily xG driven, but now I can spot its flaws and understand better ways to use it.
As requested by a Patreon, this is an analysis of xG, with the pros and cons.
What is xG?
Expected Goals (xG) aims to describe the quality of chances created or conceded by assigning a likelihood that any given shot becomes a goal. That likelihood is drawn from historical data for similar shots, taking into account location, angle, body part and shot type. Add all those probabilities together for each chance, and you get a single match figure that reflects how many goals a team might have scored on average from the chances they created.
That is the basic description, and it leaves out some nuances, but every single shot is assigned an xG rating. If you have ten shots, all punts from 25 yards, you’ll be lucky if your xG reaches 0.5. If you have one shot, a penalty, it’s around 0.76.
It is important to understand the limits and the way different providers build their numbers. xG is not a single stat. It is a judgment based on a model or an opinion.
This is why one platform can disagree with another on the same game.
Why do different platforms disagree
Live services such as FotMob report xG as the game unfolds. With the availability of football streaming, those people assembling stats can watch in real time, and therefore get an xG rating out as the game unfolds. For the top flight, there are multiple angles, but for League One, you might only be dealing with one angle, so the figure for those sites can be lower.
Whether they have people in stadiums or analysts working remotely, they are making quick calls in real time. That is difficult. It is like trying to write a match report while the match is still going on. You will get the broad strokes right, but some details are bound to be imprecise.
Post-match services such as Wyscout rewatch the game and tag every action. They can rewind, check contact points and measure positions more carefully. With time on their side, they will often assign slightly different values for the same moments.
The result is that your Wyscout xG can be higher or lower than your FotMob xG for the very same match.
Why is xG Not a Single Method of Measuring Chances?
xG is built on shots. If no shot happens, there is no entry. That means a wicked cross that flashes through the six-yard box with nobody getting a touch does not exist in the xG total. City can send eleven crosses into the box, connect with five and still generate no xG for six of those chances. Conversely, a faint glancing header that sails over the bar will add a small amount.
This creates situations that jar with common sense. Think about the famous Euro ’96 moment at the back post for Paul Gascoigne. No shot, no xG, yet everyone recognises it as one of the biggest missed chances you will see. That is a reminder that xG is a guide to the quality of shots, not a complete record of threat.
Game state matters
Game state shapes matches, and it shapes xG. Score early, and your plan will change. If City score after 30 seconds against a side like Bolton, the sensible approach is to tighten up, take fewer risks and manage the game. Supporters might shout for a second and a third, but coaches must weigh what happens if you over-commit, concede one and hand over momentum.
It works the other way, too. 3-0 up against Plymouth with fifteen minutes left, you do not often chase a fourth. You consider resting your legs for the weekend. Meanwhile, the opposition can take more hopeful efforts from range, or bring on players to create. They attempt more, and their xG goes up as a result. If it stays 0-0, that probably doesn’t happen, so sometimes a beaten team create late xG even though the game has already gone.
In some City games recently, the bulk of opposition xG has arrived late as we hold on (Bolton and Northampton spring to mind). In many, our best xG comes when we start fast, as we have against Exeter and Stevenage. On a single-match basis, that can skew numbers.
One game, small truth. Ten to fifteen games, bigger truth
For a single afternoon, xG can be used, fairly or unfairly, as a shield. You can say, we had higher xG, therefore we deserved something. That is not what it really shows. It only shows that our collection of chances carried a greater cumulative probability than the opposition’s collection. That is all.
Stretch the view to 10, 12 or 15 games and you begin to learn more. Over time, patterns emerge. If a team keeps scoring wonder strikes beyond normal expectation, the well will run dry. In 2017/18, Plymouth had a season where they finished seventh, thanks in no small part to Graham Carey smashing worldies.
They were sixth from bottom across the season for xG. It worked for a spell, but not forever – they did not improve, xG was third from bottom in 2018/19, and they were relegated. The same applies in reverse. If your xG is consistently higher than your goals, you can expect finishing to turn at some point, as long as chance creation holds.
Consider a match where Lincoln post 1.1 xG from twelve shots. On average, that is 12 chances each worth roughly one in 10. Rotherham might score from two half-chances worth a combined 0.08. The model tells you those two go in three or four times in a hundred. On that day, they did. It can happen. That does not make xG useless. It means one game is a noisy sample.
Examples inside a match can also surprise. An Adam Jackson header valued at 0.22 might be the best chance of the game, higher than the value of any single Rotherham shot that actually went in. You might not even see that moment back on short highlights, but the model will flag it.
Players and keepers are not equal, yet the value is the same
Another limitation is that xG does not consider the shooter or the goalkeeper. A penalty is often treated at around 0.76. That works as an average, but it ignores the taker’s skill. A forward with a superb record inflates the true probability. A centre back who rarely takes penalties deflates it. The same goes for the goalkeeper. Facing a 6 ft 7 in keeper is not the same task as facing a short emergency keeper pressed into service after a red card. The model assigns the same value regardless of taker, keeper or minute.
Offside also wipes a chance off the slate. You can be a stud’s width off, put the ball in the net, and there will be no xG for that action.
Why xG is still useful
So why back it? I’ve proven it is flawed. Why mention it? Because it helps identify trends and strip away the noise of single results. It is not the be-all and end-all. It does not award points. It is one flavour in the cake. Used alongside other measures, such as box entries, crosses, and positional attacks, it gives a truer picture. If Lincoln have higher xG, higher box entries and more crosses than the opposition, that points toward a sustained attacking presence rather than one padded moment, such as a missed penalty that alone contributes 0.76.
Modern football uses data because it works across long horizons. Clubs do not employ analysts for fun. They invest because the numbers help them make better decisions more often. Recruitment teams have used chance quality and chance volume as part of player evaluation for years. That sits alongside the eye test and tactical context. It is not a single switch that guarantees success for everyone, because in any 24-team league a few clubs will always finish at the bottom. What data can do is reduce avoidable mistakes and find value before everyone else sees it.
Defence first, then creation
There is another lesson from repeated League One campaigns. Teams that win promotion tend to share a strong defensive profile, including low xG against. When City set that platform, results have followed. Concede very little, and the margins tip your way over time. That does not excuse individual errors, poor refereeing or the occasional wonder strike against you. It adds context. You can keep a game tight, concede 0.6 xG, and still lose to a clean hit.
On another day the ricochet helps you, the flag goes up, or the free kick is given.
How to talk about xG after a match
xG should not become a blunt weapon for any agenda. Some will shout that only one stat matters. Others will insist that a higher xG proves deserved victory. Both miss the point. After a defeat, there can still be positives. After a win, there can still be warnings. Use xG to inform that discussion, not to replace it.
If we have a higher xG, ask when it arrived. Was it stacked late when the other side settled on the result, or was it spread across the game while we pushed? Did we rack up a cluster from a blocked shot sequence worth 0.6 in total, which feels like a penalty on paper but was really three half-chances? Did our best chance come from a set piece rather than the flowing move everyone remembers? Add box entries, crossing volume and where the shots were taken from. That is a fuller picture.
The bottom line
xG is here to stay.
What we should do is treat it as an informed estimate rather than an absolute. Do not let a single match number dominate your view. Look for patterns across ten to fifteen games. Combine it with other metrics. Remember game state.
Accept that players and keepers differ, even when the model does not. Most of all, resist the urge to blame the tool when the problem is how it is used. With sensible interpretation, xG helps separate what is repeatable from what is random, which is exactly what supporters, coaches and recruiters need over a long season.
You must be logged in to post a comment.