# xG & xA: Reading Advanced Stats
Expected goals and expected assists are the two most quoted numbers in modern football analysis, and also among the most misunderstood. This guide explains what each metric measures, how models are built, and — most importantly — how to read them without being misled.
What xG actually measures
Expected goals (xG) assigns a probability between 0 and 1 to each shot, estimating the likelihood that it becomes a goal. A penalty is typically scored around 76–79% of the time in most public models, so it carries an xG close to 0.78. A shot from the edge of the box under pressure might carry 0.05. Add up every shot in a match and you get a team's total xG for that game.
The key word is *estimate*. xG is not a prediction of the final score; it is a summary of the quality of chances created. A team can generate 2.5 xG and lose 1–0, and that is not a failure of the model — it is the nature of low-scoring, high-variance events. Models describe probability, not destiny.
How the models are built
Most public providers — StatsBomb, Opta, Understat, and others — train models on large databases of historical shots using logistic regression or gradient boosting. The inputs vary, but common features include:
- Distance and angle to the goal from the shot location
- Body part used (foot, head, other)
- Assist type (through ball, cross, cut-back, set piece)
- Defensive pressure — how many opponents are between shooter and goal, and whether a defender is closing down
- Goalkeeper position in more advanced models
- Game state and pattern of play (open play, counter, set piece)
This is why numbers differ between providers. A model that includes shot pressure will treat a blocked effort differently from one that only knows distance and angle. When comparing xG figures across sites, check the source first — mixing providers in one table is a common error.
Reading team xG
Team-level xG is best used over a run of matches, not a single game. Over 10–20 fixtures, it gives a reasonable picture of whether a side creates and concedes good chances. The practical use is spotting divergence from actual results:
- A team with high xG but low goals scored may be creating well and finishing poorly — often a sign that results will improve, though finishing skill is real and persistent for some players.
- A team with low xG but many points is often riding luck or exceptional goalkeeping; regression is likely, but not guaranteed on any fixed timeline.
Treat these as tendencies, not certainties. Football's sample sizes are small, and a season is only 38 games.
Reading player xG
For individuals, xG per 90 minutes helps compare chance-getting across different playing time. A striker with 0.55 xG/90 is consistently arriving in dangerous positions; one with 0.20 is not, regardless of how many goals they have scored in a hot streak.
Two cautions apply. First, xG measures the *quality of the position*, not the quality of the finish — a model does not know that a particular striker is an elite finisher, only that the shot was taken from a certain spot. Second, players who take penalties inflate their xG; separate penalty xG from open-play xG when judging build-up or movement.
xA: the assist equivalent
Expected assists (xA) assigns a probability to each pass based on the likelihood that the recipient scores. It credits the passer for creating the chance, not for the finish. A perfectly weighted through ball that a striker blazes over still earns the passer xA; a five-yard square pass that a teammate curls into the top corner earns very little.
The metric captures creativity better than raw assist counts, which depend heavily on teammates' finishing. A playmaker with high xA but low assists is often creating chances that are being wasted — a signal about the team around them as much as about the player.
Note that xA and "key passes" are different. Key passes count passes leading to a shot; xA weights those passes by the quality of the resulting chance. A player with many key passes from low-value positions may have modest xA.
The xG + xA combination
Adding a player's xG and xA per 90 gives a rough measure of direct goal involvement. It is useful for comparing forwards and attacking midfielders, but it flattens different roles. A deep-lying playmaker who progresses the ball through the lines may contribute little xG or xA yet be central to the team's attack. For those players, look at progressive passes, carries into the final third, and shot-creating actions alongside the expected metrics.
Common misreadings
"They outperformed xG, so they're clinical." Sometimes true over a career; often noise over a handful of games. Finishing skill exists but is small relative to chance quality.
"Low xG means a bad performance." Not necessarily. A team may have controlled the game without fashioning clear chances, or deliberately managed a lead. xG describes chances, not dominance.
"xG predicts the result." It does not. It describes what already happened in terms of chance quality. Post-match xG is retrospective.
"One game's xG proves anything." Single-match xG swings wildly. Use rolling averages over at least five to ten matches.
Practical workflow
1. Identify the provider before quoting any number, and stay consistent. 2. Use per-90 rates for players and rolling averages for teams. 3. Separate penalties from open play when assessing chance creation. 4. Pair with context — game state, opponent strength, and role — before drawing conclusions. 5. Treat divergence from results as a question to investigate, not an answer.
xG and xA are tools for asking better questions about chance quality and creativity. They reward patience and punish overreach. Read them alongside what you see, and they will sharpen your analysis rather than replace it.