Measuring what matters

Final score: 0–1. The away side nicked it with a penalty. Read only the scoreline and the story writes itself: the winners were better, the losers couldn’t finish. Now watch the same ninety minutes a different way. The team that lost had six clear chances — a tap-in fluffed, two efforts off the post, a one-on-one saved. The team that won had a single shot: the penalty. Who was actually the better side that day? Almost certainly the team that lost. The scoreline is one of the noisiest numbers in sport, and this lesson is about the numbers that aren’t.

The score is an outcome, not a measure

Goals, points, wins — these are outcomes. They tell you what happened, not how likely it was. And as we’ve seen, what happened is skill plus a big dollop of luck. A team can do everything right and lose; a team can ride one moment and win.

The fix is to measure the process instead of the outcome — the quality of what a team created and allowed, rather than whether the ball happened to go in. Process is far less noisy than outcome, so it reveals a team’s true level much faster. It’s the same insight as the long season from the last lesson, packed into a single match: stop scoring the luck, start scoring the play.

Expected goals, worked through

The cleanest example is football’s expected goals, or xG. The idea: not every shot is equal. A tap-in from two yards scores most of the time; a hopeful belt from thirty yards almost never does. So give every shot a probability of scoring, based on chances just like it — distance, angle, header or foot, one-on-one or crowded. Add those probabilities up across a match, and you get the number of goals a team “should” have scored from the chances it made.

Take our losing team. Their six chances, by quality:

• a close header — 0.40 (scores about 40% of the time)
• a good box chance — 0.25
• a half-volley in the area — 0.20
• a shot from the edge — 0.15
• a tight-angle effort — 0.10
• a long-range try — 0.10

Add them: 1.20 expected goals. They created enough quality to score, on average, more than one goal. They got zero — which, from chances like these, happens about one match in four. Unlucky, not incapable.

Now the winners: one penalty, worth about 0.76 xG, and nothing else. Their whole attacking output was that single spot-kick.

So the real contest was 1.20 to 0.76 — the losing team created clearly more and better. Play that same match a hundred times and the “losers” win about 44 of them; the “winners” take only about 19. The 0–1 was a true result for that night and a badly misleading guide to who’s better. And here’s the link back to last lesson: because xG measures the chances, not the bounces, it predicts that the unlucky team bounces back. Their goals will regress up toward the chances they keep making.

Why raw totals lie, in every sport

xG is football’s version of a rule that runs through all of sport: count the quality and the rate, not the raw total.

Basketball learned it the hard way. For years, a team that scored more total points looked better. But a team that plays fast takes more possessions, so it racks up more points without being any more skilful — and a slow, careful team can score fewer total points while being better per possession. The honest measure is points per 100 possessions — efficiency, not volume. Once teams started scoring the rate instead of the total, the whole sport’s strategy shifted: take the most efficient shots (close in, or the three-pointer), skip the pretty but low-value mid-range one.

The pattern repeats everywhere. Total tackles can mean a defender is always out of position and forever chasing. Total saves can mean a keeper’s defence is leaking shots. The raw count rewards volume; the smart metric rewards quality — and quality is the thing that carries from this game to the next.

On the whole

This is the quiet revolution behind modern sport: the shift from scoring the outcome to scoring the process, from the box-score to the model. It’s why clubs hire analysts, why a manager talks about “the performance” after an undeserved defeat, and why a smart fan is calmer than a phone-in caller — they’re watching the chances, not just the conversions.

But hold the humility from the last two lessons. A process metric is better, not true. xG is a model — a tidy estimate of a messy thing, and it can miss what a player does between the shots, or be gamed by a team that piles up cheap half-chances. Every measure simplifies, and someone will eventually optimise for the measure instead of the game. The aim isn’t to swap one magic number for another. It’s to remember that the scoreline is the start of the question, not the answer — and to keep asking what it’s really made of, in sport and anywhere else a single headline number gets mistaken for the truth.