Category · 22 articles
What the numbers actually measure, in plain language.
Every few years a new metric jumps from analytics Twitter to the broadcast graphics, usually with no explanation of what it means or how it was built. This is where those metrics get unpacked — not just the definition, but the data and the maths underneath, and the situations where each one quietly misleads you.
If you've ever wondered how an expected-goals model is actually trained, why raw tackle counts flatter passive teams, or what PPDA is really counting, start here.
Every scoreline prediction in football rests on the Poisson. I fit it to all 64 games of 2022 — it gets the shape right, but real goals are overdispersed.
Aerial-duel win percentage, why raw counts mislead, and where aerial dominance actually matters — set pieces, target men, and defending crosses.
Expected points (xPts) simulates a match from its shot-level xG to get win, draw and loss odds — points that reward chances created, not goals that went in.
Short versus long goalkeeper distribution, launch %, pass completion under pressure, and how a keeper's distribution choices shape a team's buildup.
Field tilt measures territorial dominance — each team's share of final-third passes or touches. How it is computed and why it differs from possession.
A long throw-in into the box functions like a corner you can take from anywhere down the touchline. Why it works, and the xG case for and against it.
Every team defends corners with some mix of zonal marking, man-marking, and a hybrid of the two. The trade-offs of each, and what the data says.
The current scoreline quietly distorts almost every match statistic. How game state bends possession, shot volume and xG — and why analysts adjust for it.
PPDA puts a single number on pressing intensity. How it is computed from opponent passes and defensive actions, and what high and low values mean.
xG only values shots. Possession-value models score every action - a pass, a carry, a tackle - by how much it changes the odds of scoring. How VAEP works.
A defender who makes lots of tackles might be badly positioned. Why raw tackle and interception counts mislead, and how to read defensive actions in context.
You can dominate the ball and still play it nowhere dangerous. Field tilt measures territory — whose final third the game is actually being played in.
A goal is the end of a chain, not a solo act. Shot- and goal-creating actions credit the two moves before a shot — and surface the players who make chances.
A team dominating possession might just be losing. Why you can't read match stats without the scoreline — and how good analysis adjusts for game state.
Corners feel dangerous and mostly aren't. What the expected-goals value of set pieces actually is, why they're so hard to model, and how smart teams attack them.
Most of the players who create a goal never touch it last. xGChain and xGBuildup credit the whole possession - here's how they work and who they reveal.
A plain-language guide to expected goals — what xG measures, how the model is trained on shot features like distance and angle, and how to read it.
How expected assists (xA) and expected threat (xT) measure creativity that raw assist counts miss - what each one captures, how they are built, and how to read them.
PPDA - passes allowed per defensive action - is how analysts put a number on pressing. What it measures, exactly how it is computed, and where it misleads.
Progressive passes and carries measure who actually moves the ball toward goal. The operational definitions, and why centre-backs and deep midfielders dominate.
Raw tackle and interception counts mostly measure how little a team has the ball. Possession-adjustment fixes that: the problem, the formula, a worked example.
Post-shot xG (PSxG) grades goalkeepers by how hard their shots-faced were to save. How PSxG differs from xG, and what goals prevented means.
