Expected Goals (xG) Explained — and Where It Misleads You
What xG actually counts, how a single shot gets its value, why the metric reads a run of matches better than the scoreline does, six situations in which it will mislead you, and how to turn a goals-versus-chances gap into a priced bet.
Expected goals, written xG, is the sum of the chances a team created, where every shot is worth the share of similar shots that historically went in. A tap-in from six yards might be worth 0.42 of a goal, a strike from thirty yards 0.02. Add them up and you get a number like 1.53 — not a prediction of the score, a measurement of the chances. It is the most useful public football statistic there is, and it misleads people in six predictable ways.
This page covers point 6 of the pre-match checklist: what the metric measures, how one shot is valued, and where the number stops being evidence.
In short.
- xG values every shot by the outcome of similar shots in the past, then adds them together.
- Distance, angle, body part, the pass before it and the defenders in the way do most of that valuation.
- Over 8 to 10 matches it describes a side better than results do; over one match it is a noisy sample.
- Its six failure modes — penalties, set pieces, game state, red cards, provider differences, finishing skill — are fixable by hand.
- The worked example turns an 8-match gap of +5.6 goals into a ₦500 bet at 5.00 with +10% expected value.
What the number is counting
Every shot in a professional match is logged with its position, the situation it came from and the pass before it. A model looks at hundreds of thousands of past shots with similar attributes and asks one question: what share were scored? That share becomes the shot's value.
You can build the crudest version yourself. Take a season's penalties — say 200 awarded, 158 converted. That is 158 ÷ 200 = 0.79, so every penalty in your ledger is worth 0.79 of a goal. A full model does the same thing with more attributes and far more shots.
Two consequences follow. A shot's value never reaches 1.00, because no shot is certain. And the figure has no idea whether the ball went in — which is the entire point of it.
What one shot is worth
The inputs that move a shot's value most, in rough order of weight:
- Distance from goal. The dominant factor. Value falls away sharply beyond about 16 yards.
- Angle to the posts. How much goal the striker can actually see from where the ball is.
- Body part. Headers are converted far less often than shots with the foot from the same spot.
- The pass before it. A cut-back from the byline leaves the goalkeeper wrong-footed; a cross into a crowd does not.
- Situation. Open play, corner, direct free kick, penalty, or a counter against a broken defensive line.
- Pressure. Defenders between ball and goal, and the goalkeeper's position, in models with tracking data.
What is missing matters as much. The identity of the player is not there: a model built on every shot in the league treats a striker and a full-back identically if the two attempts are geometrically the same.
Adding shots into a match figure
One home side's card from a single match, shot by shot.
| # | Shot | xG |
|---|---|---|
| 1 | Penalty | 0.79 |
| 2 | Foot, 12 yards, central, from a cut-back | 0.31 |
| 3 | Header, six yards, from a corner | 0.24 |
| 4 | Header, 14 yards, from a deep cross | 0.06 |
| 5 | Foot, 18 yards, angled, defender blocking | 0.05 |
| 6 | Foot, 25 yards, open play | 0.03 |
| 7 | Foot, 25 yards, open play | 0.03 |
| 8 | Foot, 30 yards, first time | 0.02 |
| Total | 1.53 |
Eight shots, 1.53 xG — and 0.74 of it came from the penalty and the corner, leaving 0.74 across six open-play attempts. That is thinner than "eight shots" suggests. The visitors had four shots worth 0.61 and won 2-1, which is not an injustice: over one match, that happens constantly.
The gap between goals and chances
The metric earns its place over a run of matches, where the difference between what a side created and what it scored becomes readable. Eight matches for one club:
| Measure | Goals | Expected goals | Gap |
|---|---|---|---|
| Scored | 16 | 10.4 | +5.6 |
| Conceded | 6 | 11.1 | −5.1 |
| Per match, scored | 2.00 | 1.30 | +0.70 |
| Per match, conceded | 0.75 | 1.39 | −0.64 |
This side has been scoring 0.70 goals a match more than its chances justify and conceding 0.64 fewer. Twenty points from eight matches sits on a profile that is marginally negative: 1.30 created against 1.39 allowed. League tables are built from the left-hand column, and so, largely, are prices — and that gap is where the metric earns its keep.
Six ways expected goals will mislead you
| Problem | What goes wrong | What to do about it |
|---|---|---|
| One match is a tiny sample | A single penalty is a third of a typical match total; one refereeing decision rewrites it | Never read one match alone; eight to ten at the same venue is the minimum |
| Penalties are inside the total | Three penalties in five matches inflate an attacking profile by roughly 0.5 a match | Judge attacks on the non-penalty figure; penalties are not a repeatable team skill |
| Set pieces are not open play | A side living on corners looks creative in the totals | Split set-piece from open-play xG, and check the tall centre-back is playing |
| Game state distorts volume | A team 2-0 down after 30 minutes shoots at a defence that has stopped guarding the box edge | Discount xG accumulated while a side was two behind or a man up |
| Providers disagree | Two models can value the same shot at 0.24 and 0.31 on different training data | Pick one source and stay with it; never mix providers in one comparison |
| Players are not average | The model prices the chance, not the finisher or the goalkeeper facing it | Treat a gap that persists over 30-plus matches as possible skill, not automatic regression |
Important. "Regression to the mean" is not a schedule. A side finishing above its chances for eight matches may keep doing so for another eight. The metric tells you the results were flattered; it does not tell you when the flattery stops.
Reading a single match without over-reading it
If you do look at one match, read it in this order. Open the shot map rather than the total: 1.53 from one clear chance and 1.53 from twelve half-chances are different performances. Strip the penalties out. Check when the shots happened — a flurry after the 80th minute at 3-0 down is not evidence of a good attack. Note who was on the pitch, since a red card on the hour makes the last third of the data a different match. Even after all four steps, one match is still one match.
xA, xGOT and xPTS
Three relatives appear on the same stats pages, each answering a different question.
xA — expected assists. The xG of the shot a pass created, credited to the passer. A player with 4 assists on 1.9 xA has had good finishers on the end of his passes.
xGOT — expected goals on target. Recalculated after the shot, from where it ended up in the goal frame; only shots on target have it. The gap between xG and xGOT measures placement: 1.53 xG against 2.10 xGOT means the corners were found and the goalkeeper had a hard evening.
xPTS — expected points. Run the chances through a model of every scoreline, get three probabilities, multiply by the points on offer. At 42% home, 27% draw, 31% away, the home side's expected points are 3 × 0.42 + 1 × 0.27 = 1.53, which is worth comparing with the point they actually took.
Turning the gap into a price
Return to the club above: 20 points from 8 matches, 1.30 xG created and 1.39 allowed. Its next fixture is away at a mid-table side, and the market has priced the results rather than the chances.
Example. The visitors are 1.70 to win, the home side 5.00. Chance profiles that close together do not support a 1.70 favourite on the road; your own figure for the home side, built from the checklist, comes out at 22%. Expected value on the home price is 0.22 × 5.00 − 1 = +10%, which on a ₦500 stake is ₦50 in your favour and a ₦2,500 return if it lands.
Two cautions. Removing the bookmaker's mark-up before comparing is not optional — the method is in turning odds into probability. And an edge built on eight matches carries a wide error band, which is why the stake stays flat at ₦500.
Where to find it. Chance figures sit beside form and line-ups on our match pages. Which of the surrounding numbers are worth reading is covered in the guide to predictive football stats.
FAQ
Is a high xG the same as deserving to win?
No. It means a side created the better chances, which is a different claim. Defending a lead by conceding low-value shots is a tactic, not a failure, and a team that wins 1-0 on 0.6 xG against 1.8 has often done something deliberate rather than something lucky.
Should I use xG for goal lines and GG?
Yes, and it is the natural input for both, since a match total built from each side's chances prices the goal ladder directly. The lines themselves are explained in the guide to over 1.5, 2.5 and 3.5, and the two-sided version of the same question in what GG and NG mean.
Why do two websites show different xG for the same match?
Because they are different models trained on different shot databases, and some include tracking data on defender positions while others do not. Neither is the official figure. Choose one source and use it consistently; comparing across providers produces gaps that are about the models, not the football.
This article is for information only and is not an inducement to gamble. Betting involves the risk of losing money — never stake more than you can afford to lose. 18+. If gambling stops being entertainment, read our responsible gambling guide and seek help.
