What Expected Goals (xG) Actually Tells You About a Match
Decoding Expected Goals: What xG Really Reveals About Football Matches
Beyond the Final Score: Understanding Expected Goals
A 1-0 win and a 4-0 win both count as three points, which is exactly why football tables lie to us for weeks at a time. I've watched sides get battered for ninety minutes, nick a goal from a deflected cross, and then listen to pundits call it a "statement result." The scoreline is a fact. It is not a summary of what happened.
Expected Goals (xG) is the number that fills that gap. Every shot in a match gets a value between 0 and 1 based on where it was struck from, the angle, the body part, the type of pass that set it up, and whether a defender was closing in. A penalty lands around 0.76. A header from six yards out with nobody near you sits high. A hopeful effort from 30 yards is worth about 0.03, which is why your mate shouting "shoot!" from the sofa is usually wrong.
Add up every shot and you get a team's total for the match. That total is a probability, not a promise. A side can rack up 2.4 xG and score once, and that doesn't mean the metric failed; it means they created enough chances to expect two or three goals on an average day.
The gossip worth knowing: recruitment departments at most top clubs were running these numbers long before broadcasters put them on screen.
Strip it back and xG is just history turned into a number. Take thousands of shots from similar positions, under similar pressure, and count how often they went in. A shot from the penalty spot converts roughly three times in four, so it gets a value near 0.76. A low-percentage effort from distance converts around three times in a hundred, so it gets 0.03. Do that for every shot in a match and you have a map of what an average team would have scored from those exact chances.
Think of it like a coin flip where the odds keep changing. A fair coin is 50/50, but imagine a weighted one that comes up heads 76 times out of 100. Flip it once and you might still get tails. Flip it ten times and the results start to look like the weighting. xG works the same way: it describes the probability attached to each chance, not the outcome of any single one.
That distinction matters. A striker who misses a 0.76 chance hasn't beaten the model; he's hit the 24% branch. The number tells you how often that situation produces a goal across a large sample, nothing more.
- Shot location and angle: tighter angles and greater distance drop the value fast.
Type of assist: a through ball or cutback raises it; a hopeful cross or long ball lowers it.
Body part and pressure: headers convert less often, and a closing defender cuts the number.
Real Match Breakdown: Reading the xG Story
Take a match I keep coming back to: Riverside United 1-0 Kingsway Athletic. On the scoresheet it's a narrow win, the kind that gets filed away as "grinding out a result." The shot map says something else entirely. Riverside finished with 2.5 xG, Kingsway with 0.5.
Here's how that gap builds. Riverside's first half alone produced a cutback from the byline to a striker eight yards out (0.42), a free header from a corner (0.31), and a low drive from the edge of the box that the keeper pushed wide (0.09). Add a one-on-one saved at the near post and you're already past 1.2 before half-time. Their goal, when it came in the 67th minute, was a scrappy finish from a scramble — worth maybe 0.18. The biggest chance of the afternoon wasn't the one that went in.
Kingsway's 0.5 came from volume rather than quality: eleven shots, nine of them from outside the box, none above 0.08. Their best moment was a header from twelve yards that sailed over.
Now flip it. Same 1-0 scoreline, but reverse the xG and you have a smash-and-grab — the winner was a 0.04 deflected effort and the losing side missed a penalty. Identical result, opposite story.
That's the whole trick of reading xG: it separates the team that earned the win from the team that survived the afternoon. One number, two very different conclusions.
The Difference Between Expected Goals and Expected Goals Against
One number looks at what a team creates. The other audits what it gives up. Expected Goals Against, or xGA, takes the same shot-quality model and applies it to the chances a defence concedes — the cumulative value of every attempt the opposition got off, weighted by location, angle, assist type and pressure.
Low xGA is the fingerprint of a genuinely mean defence, not one riding a hot goalkeeper. A side can post four clean sheets and still leak 1.6 xGA per game; that run ends, usually painfully. Conversely, a team shipping goals while keeping xGA under 0.9 is getting torched by variance, and the back line is probably better than the table suggests. That gap between Actual Goals conceded and xGA is where the truth hides.
Open, high-risk football is where Expected Goals earns its keep. When both sides push numbers forward, commit full-backs, and leave two centre-backs defending half the pitch, the shot volume climbs and the model's error margin shrinks. Dozens of attempts from a wide spread of positions average out the noise that a single deflected winner would otherwise create. The 2.5 versus 0.5 game above is one example; a better one is a 3-3 draw I still think about, where the two teams combined for 41 shots and the xG read 2.9 to 3.1. Six goals, six deserved ones, and not a single pundit calling it lucky.
The reason is sample size. Ten shots from a mix of positions tells you almost nothing. Thirty tells you plenty. Cautious, low-block matches generate eight or nine attempts, and at that volume a single penalty (0.78) can swing the entire reading. But when a game breaks open, the Actual Goals column tends to land within a goal of the Expected Goals figure, and the teams that keep generating the higher number keep winning the return fixtures. That correlation is what makes the metric worth trusting in exactly the matches where the eye test struggles to keep up.
import math
def shot_xg(distance, angle, is_header=False, is_counter=False):
"""Conceptual model: closer, wider angles, and better situations raise xG."""
# Distance decay: shots from 6 yards are far more dangerous than from 30
distance_weight = math.exp(-0.12 * distance)
# Angle in degrees: 0 = straight at goal, 90 = tight byline
angle_weight = math.cos(math.radians(angle)) ** 2
# Situational adjustments, roughly matching observed conversion gaps
header_penalty = 0.6 if is_header else 1.0
counter_bonus = 1.25 if is_counter else 1.0
raw = distance_weight * angle_weight * header_penalty * counter_bonus
# Clamp to a realistic 0.01-0.95 range, as real models do
return max(0.01, min(0.95, raw))
The Fine Print: When to Be Skeptical of the Metrics
Here's where I stop selling and start warning. Expected Goals is a model built from averages, and averages do not know who is standing in the goal. A defence can allow 1.4 xGA and still keep a clean sheet every week if the man behind them is elite. The model prices the shot, not the save. Post-shot xG exists to correct some of that, but the base number will happily tell you a team was lucky when the truth is simpler: they employ a goalkeeper who saves shots nobody else reaches.
Refereeing is another blind spot. A penalty carries a fixed value around 0.78 in most models, whether it's a stonewall foul or a soft one a striker bought with a stumble. Red cards don't appear in the calculation at all, yet a team down to ten men for half an hour will concede chances that have nothing to do with their defensive structure. The xG column records the shots, not the circumstances that produced them.
Then there's the geometry the model can't see. A deflected strike from 25 yards, a scuffed finish that dribbles over the line, a header from a corner where the marking collapsed: all get bucketed by location and assist type, and none of that captures the mess of a real defensive breakdown. Free kicks outside the box, own-goal pressure, the second ball after a save — the standard parameters handle these crudely or not at all.
Treat xG the way a scout treats a highlight reel. It narrows the conversation. It does not end it. Anyone quoting a single match's figure as proof of anything is using the number to win an argument rather than to learn something. The tool is good. The crystal ball is imaginary.
Off-ball runs before the shot: the model sees the finish, not the decoy run that dragged a centre-back out of position and turned a blocked angle into a clear one.
Set-piece routines: a well-drilled near-post corner produces shots the location model rates highly, yet the training-ground repetition behind it never shows up in the number.
Fatigue and game state: a team chasing a goal at 2-0 down takes low-value shots from range, inflating volume while the model treats each attempt as equally informative.
Substitution timing: fresh legs against a tiring back line generate chances in the final twenty minutes that reflect squad depth, not the tactical structure the xG column implies.
And here's the part that keeps the model honest and the pundits employed: football is played by people with nerves. Expected Goals has no column for a stadium that turns on its own team after twenty minutes, no way to price the moment a full-back's legs go and he starts hoofing clearances into row Z. Panic spreads. I've watched Riverside United dominate 2.1 xG worth of chances and still lose because one centre-back lost his head after a bad touch and the whole back four started playing scared. The model logged the shots. It couldn't log the fear.
Clutch is the other side of it. Some players simply convert when the noise is loudest, and their numbers quietly break the model's assumptions year after year. You know the type: the striker who does nothing for eighty minutes, then buries the only half-chance he gets. His xG per shot looks ordinary. His goal return doesn't. Momentum works the same way, and it's contagious in a dressing room in ways no dataset captures. A keeper makes one ridiculous save and suddenly the entire side believes. No shot-quality model will ever price belief. That's not a flaw to fix, it's the reason we still watch.
Squad Building and Transfer Strategy with xG
Now the part of the job I actually get paid for. Every agent I've had dinner with this season carries a folder of xG numbers, and the smart ones lead with it before they mention goals. Here's why: a winger with 9 goals from 4.2 xG is running hot, and the buying club knows it. A striker with 4 goals from 9.8 xG is the opposite, and that's where the bargains hide. Recruiters call it buying the dip. The finishing hasn't turned bad overnight; the sample has. Regression is coming, and whoever signs him gets the goals at a discount.
xG per 90 minutes is the number sporting directors circle. It measures the quality of chances a player gets himself into, whether or not the ball went in. A 22-year-old attacking midfielder posting 0.45 xG per 90 in a mid-table side is generating elite shot volume from good positions. He may have six goals to show for it. The club paying for those six goals is overpaying. The club paying for the process gets a bargain, and I've watched two Premier League sides circle exactly that profile this window. One looked at the goal tally and walked away. The other looked at the shot map and made the call.
Then there's the sell side, which nobody talks about enough. Clubs now use xG to spot the player whose output is about to fall off a cliff: the 29-year-old with 14 goals from 8.1 xG. Cash in now, because the market is paying for luck. Squads get built on that discipline, moving players a year early rather than a year late.
- Chance creation without the shot: track players whose key passes and expected assists keep producing high-value chances even when their own shot count is low. The playmaker with modest goal numbers can still be the engine of a side's Expected Goals.
- Shot selection per attempt: xG per shot separates the striker who works good positions from the one who blasts hopefully from 25 yards. A low average with high volume is a warning; a high average with modest volume is a finisher waiting for a bigger role.
- Persistent underperformance: when a forward's goals trail his xG across two full seasons, that's a technique question, not bad luck. Watch the tape for placement and body shape before deciding whether he's a buy-low target or a genuine conversion problem.
- Rolling trends, not single seasons: a rising xG per 90 across three campaigns signals a player whose role is expanding. Flat or falling numbers in a weaker league tell you the output was the system, not the man.
The scouting list above only pays off if the players around those targets make sense together. A manager who stacks high-volume shooters ends up with a side that racks up Expected Goals and cannibalises itself: three forwards all wanting the same cutback, nobody making the decoy run that opens it. The opposite mistake is just as common. Fill the team with precise creators and you get lovely build-up ending in a 0.06 xG effort from the edge of the box.
The balance I'd argue for is one volume striker, one high-xG-per-shot finisher, and creators whose expected assists cover the gap between them. Volume forwards drag defenders around and generate rebounds; precise finishers convert the one or two chances a tight game offers. A side whose xG is spread across six players is harder to defend than one whose 2.1 comes from a single source, because shutting down one man kills the entire attack.
At team level, read xG per shot next to total xG. A high total with a low per-shot average means plenty of hopeful efforts. A modest total with a high average means the chances are worth waiting for. Squad builders who track both numbers together tend to buy the right profile rather than the loudest one.
The Verdict: Integrating Analytics into Football Viewing
Watch the match first. Then open the shot map. That order matters, because xG is a check on what your eyes told you, not a replacement for them. If the model says a side deserved more than it got, go back to the chances and ask why: a save, a blocked angle, a striker taking an extra touch. You'll learn more from that five-minute review than from a week of hot takes.
Three habits are worth building. Check xG against xGA together, since a team creating 1.8 and conceding 1.7 is in a different situation from one creating 1.8 and conceding 0.6. Compare xG per shot with total xG to see whether the chances were good or merely numerous. And keep the traditional numbers alongside: possession, shots on target, and the eye test on who was actually running the game.
The supporters who do this will spot a false position in the table weeks before the pundits do. As tracking data improves and models get better at pricing pressure, goalkeeper positioning, and the pass before the pass, the gap between what the scoreboard says and what happened on the pitch will keep shrinking. Football analysis is heading toward a point where the number and the naked eye agree more often than they argue, and the fans who learned to read both will be the ones explaining it to everyone else.