The Numbers Game
One football season, nine statistical traps. A field guide for a rookie analyst.
Nine interactive, animated explorers teaching Bayes' theorem, expected value, survivorship bias, the Abraham Wald problem, Berkson's paradox, the St. Petersburg paradox, regression to the mean, the Hawthorne effect and Simpson's paradox through a single football season.
You are the new analyst at Bandra United — a mid-table club with a small budget and a chairman who loves a dashboard. Over one season, the numbers will try to fool you nine times. Every box below is a working model: drag the sliders, press the buttons, break things. The formulas matter less than the question each trap teaches you to ask.
Bayes' theorem, or what a great trial is worth
Pre-season trials. A seventeen-year-old scores twice in the trial match and the coaches' WhatsApp group is on fire. But brilliant trial performances are common, and brilliant players are rare. Bayes' theorem is just the honest way to combine those two facts: what you knew before the match (the prior) with what you saw in it (the evidence).
The trap: judging the evidence without the base rate. The escape: ask how many ordinary players could have produced the same performance.
Survivorship bias, or interviewing only the winners
The chairman has read that every recent title winner gambled the budget on one marquee striker, and he wants one. Nobody writes long features about the clubs that made the same gamble and went down. Before copying the winners, find the graveyard.
The trap: the sample you can see selected itself by succeeding. The escape: count everyone who tried, not everyone who's still standing.
The Abraham Wald problem, or armour where the holes aren't
Survivorship bias has a famous sting in the tail. In 1943 the statistician Abraham Wald was shown returning bombers riddled with bullet holes and asked where to add armour. The military wanted to reinforce the riddled parts. Wald said the opposite — and his logic applies every time Bandra United studies only the counter-attacks that reached its penalty box. The ones your midfield killed at birth never made the clip reel.
The trap: the missing data isn't just missing, it's missing for a reason. The escape: ask what the failures would have looked like — then armour the engines.
Berkson's paradox, or the trade-off you invented
Intake day: three hundred kids, two numbers each — technique and pace. Across all three hundred, the two are unrelated. But the academy admits only kids whose combined score clears the bar, and inside the building a legend is soon born: "the quick ones can't pass." You manufactured that correlation with a cutoff.
The trap: selecting on a sum makes its parts trade off against each other. The escape: whenever two good things look negatively related, ask who got filtered out.
Expected value, or why the boring pass is right
Eighty-ninth minute, tight angle. Shoot, or square it for the striker? xG is nothing more exotic than expected value: what each choice would be worth if you could live the moment a thousand times. You can't. The simulator can.
The trap: judging a decision by the outcome you happened to get. The escape: judge it by the average over all the outcomes you could have got.
The St. Petersburg paradox, or the game worth infinity
The cup sponsor invents a halftime game. The pot starts at ₹2 and doubles on every heads; the first tails ends the game and you take the pot. The maths says the fair ticket price is infinite — every doubling round contributes ₹1 to the average, forever. Your gut says maybe forty rupees. Your gut has noticed something real about variance, and about how much money you actually have.
The trap: treating expected value as the whole answer. The escape: a bet you can't afford to repeat isn't worth its average — the average lives in the repeats.
Regression to the mean, or the curse of the Golden Boot
Your striker won October's player-of-the-month award, then "lost form." The physio blames fatigue; the pundits blame the new contract. Mostly, it's luck running out. An extreme month is part ability and part fortune, and the fortune doesn't repeat. The same arithmetic explains why sacking the manager "works": you sack at the bottom of the luck cycle, and the bounce was coming anyway.
The trap: reading the swing back to normal as the effect of whatever you did in between. The escape: expect the extremes to drift toward the middle before you credit any intervention.
The Hawthorne effect, or being watched
The club buys GPS vests. High-intensity sprint numbers jump thirteen percent in the first week and the fitness coach wants a bonus. But players run differently when they know the vest is live. Measuring people is itself an intervention — the act of watching changes the thing you're watching.
The trap: attributing to the treatment what belongs to the attention. The escape: design the measurement so the watched and the unwatched can be compared.
Simpson's paradox, or two truths and a flip
The board wants to sell Ravi: Marco's overall conversion rate is nearly double his. Split the shots by type and Ravi is better from open play and better from the penalty spot. Both facts are true at once. The mix of shots is doing the lying — Marco takes far more penalties, and penalties are easy.
The trap: an aggregate can contradict every one of its parts. The escape: never compare two rates without asking what each one is made of.
Nine traps, one question
Every chapter this season was the same failure wearing a different shirt: a number arrived without its context. The missing context was the base rate (chapter 1), the failures (2 and 3), the cutoff (4), the repetitions (5), the bankroll (6), the luck (7), the observer (8), and the mix (9). So the whole season compresses into one habit. When a number walks into the room, ask it two things: compared to what? — and who didn't make it into the sample?