Stats glossary

24 terms across 4 sections

Every stat this site shows and every term it assumes you know. Each entry says what the number is and — the part that matters — what it does not tell you. Definitions describe what our data feed records, which is not always what a bookmaker settles on.

The markets we model

Seven player counting stats. Each has a fitted model behind it and a trends page showing what has actually happened.

Fouls committed#Fouls committed trends →: Free kicks conceded by a player. Handballs and offsides are not fouls; a foul that also brings a card is still one foul.; Reading it The referee matters more than the player. A strict official lifts every player on the pitch at once, which is why the same player can average 1.2 under one referee and 2.1 under another without changing anything about how he plays.
Fouls won#Fouls won trends →: Free kicks awarded to a player — he was the one fouled. Sometimes called fouls drawn or fouls suffered.; Reading it This belongs to dribblers, not to defenders. It tracks how often a player runs at an opponent with the ball, so it moves with tactical role far more than with form.
Tackles#Tackles trends →: Attempted tackles, whether or not possession was won. Our feed does not separate attempted from won — see Tackles won.; Reading it A tackle count is mostly a statement about where the ball was. A defender in a side that sees little of it accumulates tackles; the same player in a dominant team does not, and neither fact says much about him.
Tackles won#: Tackles where the player's team came away with the ball. **We do not have this stat.** API-Football reports attempted tackles only, and Sportmonks does not separate the two.; Reading it It is listed because the gap matters rather than because we can show it. Bookmakers settle tackle markets on different definitions, and a model fitted on attempted tackles cannot be priced against a book settling on tackles won without a mapping nobody has verified.
Shots#Shots trends →: All attempts at goal: on target, off target and blocked. Volume, with no judgement about quality.; Reading it Two players on the same number can be doing opposite things — one shooting often from bad positions, one rarely from good ones. Shot volume is the more stable of the two to predict, which is why the model prices it before shots on target.
Shots on target#Shots on target trends →: Attempts that would enter the goal without a deflection, including those saved. A shot cleared off the line counts; a shot blocked by an outfield defender does not.; Reading it The widest spread of any market here — roughly thirty-six times between the highest and lowest rates — which is what makes it worth modelling and also what makes small samples dangerous on it.
Cards#Cards trends →: Yellow cards shown to a player. A second yellow that becomes a red counts as both; a straight red is recorded separately and is not in this figure.; Reading it Nearly always priced as “to be carded”, which is a yes/no question rather than a count. Fouls drive it and the referee's card-per-foul rate decides how many of them are punished.
Goalkeeper saves#Saves trends →: Shots on target prevented from becoming goals by the goalkeeper.; Reading it Almost entirely a statement about the opponent and about the keeper's own defence. A keeper behind a bad back four faces more shots and makes more saves without being any better at goalkeeping.

Stats we track but do not price

Recorded, shown on the trends pages, and deliberately not modelled — no bookmaker in our feed quotes them.

Interceptions#Interceptions trends →: Passes cut out by reading them, without a tackle being made.; Reading it Often mistaken for a defensive-quality metric. Like tackles it rises when a team has less of the ball, so it describes the match as much as the defender.
Clearances#Clearances trends →: Deliberate kicks or headers away from the defensive third, under pressure.; Reading it A siege metric. A high clearance count is usually evidence of a defence spending the afternoon defending, not of a defender doing something well.
Duels won#Duels won trends →: Ground and aerial contests won against a direct opponent.; Reading it Broad enough to be steadier match to match than the narrower markets, which is exactly why it is less useful for finding a mispriced line: a stat that barely moves is a stat a bookmaker prices easily.

How the numbers are built

The terms that decide whether a figure on this site means what it looks like it means.

Per 90#: A player's total for a stat, scaled to ninety minutes of playing time.; Reading it The trap on every stats site. A substitute with seventeen fouls in 438 minutes reads 3.49 per 90 and has never committed three fouls in a match. A bet settles on one match, not on a projection to minutes the player will not see — so read the minutes-per-appearance column beside it, always.
Hit rate#: The share of a player's matches in which he cleared a given line. “62% over 1.5 fouls” means he committed two or more in 62% of the matches counted.; Reading it Meaningless without its denominator. 5/5 and 31/50 are both 100% and 62%, and only one of them is evidence. Every hit rate on this site carries the count it came from for that reason.
Match window#: The last N matches for each player — last 5, 10 or 20 — rather than a calendar date range.; Reading it A date range gives a regular eleven matches and a rotated squad player three, then prints both hit rates in the same column as if they were comparable. A per-player window gives every row the same denominator, which is the condition under which two rows can be read against each other.
Shrinkage#: Pulling a player's own rate toward the average for his position and league, by an amount that depends on how many matches he has behind him.; Reading it The single most important rule in the model, and the reason it can disagree with the market. Bookmakers over-weight recent form; five matches of a counting stat is mostly noise. A player with twelve appearances is described more by his position group than by himself, and the projection says so through the sample-size column.
Expected minutes#: How long the model expects a player to be on the pitch, in three confidence states: confirmed lineup, expected lineup, or a guess from his historical start rate.; Reading it The most important input to any counting projection, because a rate only becomes a count once you say how long he plays. Every projection currently on this site is in the weakest state — no lineup data has been captured yet — and is labelled provisional.
Projected distribution#: The full spread of outcomes the model expects — the probability of 0, 1, 2, 3 and so on — rather than a single expected number.; Reading it An expected 1.8 fouls made of 2 every week and an expected 1.8 made of a 0 and a 5 are completely different bets at the same average. The ladder on each match page is where you see which one you have.
Negative binomial#: The distribution used for every count market here, rather than the more common Poisson.; Reading it Poisson assumes the variance equals the mean. Fouls and tackles are more spread out than that, and a Poisson would systematically put too little probability on the high scores — which is precisely where over bets live.
Calibration#: The test of whether a model's stated probabilities happen at that rate. If it says 60% a hundred times, roughly sixty should land.; Reading it A model can be accurate on average and badly over-confident where it is most confident, which produces its biggest fake edges exactly where it would recommend the biggest stakes. Every market here had to pass a calibration gate before it was allowed to produce a projection.

Betting terms

The vocabulary the rest of the site assumes you have.

Line#: The threshold a bet is settled against. “Over 1.5 fouls” wins on two or more.; Reading it Half-lines cannot draw. A whole-number line can: “over 2.0 fouls” is a void at exactly two, and treating that as an under overstates the under's chance. Every line on this site is a half-line for that reason.
Decimal odds#: Total return per unit staked, stake included. 2.50 returns £2.50 on £1 — £1.50 profit.; Reading it The implied probability is 1 divided by the price, but those never sum to 100% across a market. The excess is the bookmaker's margin.
Overround#: How far a market's implied probabilities sum above 100%. A 1X2 market at 105% carries a five-point margin.; Reading it The number bookmakers rely on nobody computing. It is modest on match result and can reach 20–30% on a five-leg same-game builder, which is why a builder price looking generous usually is not.
Closing line value#: Whether the price you took was better than the price the market settled at just before kick-off.; Reading it The only honest short-run measure of whether a model is finding something. Profit over a few weeks is mostly variance; beating the close repeatedly is not. It is the metric this project is judged on, scored as prop prices reach their closing lines.
Quarter Kelly#: A stake of one quarter of the theoretically optimal fraction, capped here at 2.5% of a bankroll.; Reading it Full Kelly is optimal only if the model's probabilities are exactly right, and no model's are. Staking a quarter of it costs a little expected growth and removes most of the risk of ruin when the model is wrong — which it will sometimes be.

Where a bookmaker settles a market on a different definition from the one above, that book and market pair is excluded rather than priced against the wrong rulebook. Tackles are the clearest case: our feed records attempts, and a book settling on tackles won is scoring a different event with the same name.