Beside every modern scoreline sits a panel of numbers. They are presented with equal confidence and they are not equally meaningful. Some are counts of well-defined events, some are counts of poorly defined events, and some are model outputs presented in the visual language of counts. This page sorts them.
Possession
The most prominent statistic and the most misunderstood. There are two common methods: measuring time in control of the ball, and counting each team's passes as a share of all passes. They produce different numbers for the same match, sometimes by several percentage points, and providers do not always say which they used.
More importantly, possession has no fixed relationship to outcome. A team defending a lead may concede possession deliberately; a team chasing a game may accumulate possession in areas where it can do no harm. Possession describes a style of play. It does not describe control, and treating it as a proxy for dominance is the most common statistical error in football commentary.
Shots
Shot counts are counts of a category with a genuinely fuzzy boundary. Whether a deflected cross, a scuffed attempted clearance, or a blocked effort from thirty-five metres counts as a shot is a judgement, and providers differ. "Shots on target" narrows the definition and improves comparability, but introduces its own edge cases — a shot that would have missed but for a deflection, an effort saved by a defender on the line.
Raw shot counts also weight a tap-in and a speculative long-range effort identically, which is why they correlate with goals only loosely and why the field moved to weighting shots by quality.
Corners
Corners are an unusually clean count — the event is unambiguous and the referee adjudicates it — and an unusually weak indicator. They are as consistent with sustained pressure as with repeated failure to convert pressure into anything better. A high corner count is a fact about territory, not about threat.
Cards and discipline
Card counts mix two distinct things: how a team played, and how a particular referee interpreted what they did. Refereeing thresholds vary between individuals and between competitions, so a disciplinary record aggregated across competitions is comparing measurements taken with different instruments. Within a single competition and season the comparison is far more defensible.
Expected goals and model-based measures
Expected goals assigns each attempt a probability of becoming a goal, estimated from historically similar attempts, and sums them. Its real virtue is stabilising a noisy signal: goals are so rare that a handful of matches tells you very little, while shot quality accumulates faster and predicts future goals better than past goals do.
Two honest caveats. First, models differ — different training data and different input features produce different values for the same match, so figures from two providers are not interchangeable. Second, the model does not know context it was not given: game state, defensive pressure and the identity of the finisher may or may not be inputs.
The wider research programme extends this idea beyond shots to every action on the pitch, valuing passes and carries by how they change the probability of scoring — the approach set out in Actions Speak Louder Than Goals: Valuing Player Actions in Soccer is the standard reference, and the applied end of the same field is on display each year at the MIT Sloan Sports Analytics Conference.
Sample size, which ruins everything
Football's scoring rate is so low that single-match statistics are almost pure noise, and even a full season is a modest sample by the standards of most sports. Any statement of the form "this team is better at X" needs a stated number of matches behind it. The practical rule of thumb is that shot-based measures become interpretable across roughly ten matches, goal-based measures need far more, and anything derived from a single match is an anecdote with a decimal point.
How to read a statistics panel
- Read the scoreline first and let it anchor everything else. It is the only number that is definitionally correct.
- Check whether each figure is a count or a model output. They deserve different amounts of trust.
- Ask who defined the event. If the definition is not published, the number is not comparable across providers.
- Prefer rates over totals when comparing teams that have played different numbers of matches.
See also how to read a result and how the underlying events are recorded.