Is a coin flip really 50/50?

The coin flip is civilisation's default randomiser — it starts Super Bowls, settles bar bets and breaks ties in elections. It is also, measurably, not quite fair. The bias is small, human, and fixable, and the story of how it was measured involves a team of researchers flipping coins 350,757 times. (Our coin flip has no such bias — more on why at the end.)

What the physics says

In 2007, mathematician Persi Diaconis — a former professional magician — and colleagues built a coin-flipping machine and worked through the mechanics of a real human flip. The key finding: a flipped coin precesses. It doesn't just spin end-over-end; it wobbles around its axis like a thrown pizza, and that wobble means the coin spends slightly more of its airborne time with its starting face up. Their prediction: a caught coin lands the same side it started up about 51% of the time. Not heads-biased or tails-biased — starting-side biased. The coin itself is symmetric; the flip isn't.

The 350,757-flip experiment

The prediction sat untested at scale until 2023, when a European team led by František Bartoš recruited 48 people to flip coins from 46 currencies and film every one — 350,757 flips. The result: coins landed same-side-up 50.8% of the time, strikingly close to Diaconis's 51%, with the bias varying by flipper (some people's technique was near-perfectly fair, others reached several percent). Knowing the starting side is worth about as much as a casino's house edge — trivial in one flip, decisive over thousands.

Landing on its edge, and other exotica

The beloved movie outcome has also been estimated: an American nickel landing and staying on its edge on a hard surface comes out around 1 in 6,000 — rare, not impossible, and thicker coins do better. Separately, spinning a coin on a table (rather than flipping) can be wildly unfair: spun pennies can come up tails as much as 80% of the time, because the slightly heavier heads face pulls the spin over. Never settle anything on a spun coin.

Streaks feel wrong and aren't

The other way coin flips deceive is in our reading of them. In 100 fair flips, a run of six heads or six tails in a row is more likely than not — yet people shown genuinely random sequences consistently judge them "too streaky" to be random, because our mental model of fairness alternates far more than chance does. This is the gambler's fallacy's engine room: after four heads, the next flip is still 50/50 (well, 50.8), because the coin stores nothing. Real randomness is clumpier than intuition expects — the streaks aren't evidence of bias; their absence would be.

How to flip actually fairly

• Hide the starting side. The measured bias needs knowledge of the starting face to exploit — shake the coin between covered hands first and the 50.8% edge evaporates into an honest 50/50.
• Call while it's in the air — the tradition exists precisely to prevent starting-side gamesmanship.
• Flip, don't spin; catch it rather than letting it bounce (bounces add their own physics).
• Or remove the physics entirely. The digital flip draws from the browser's random number generator — no precession, no starting side, no technique. For choices among more than two options, the random picker generalises the same fairness.

The deeper lesson is a good one to carry around: "fair" processes are engineering achievements, not defaults. Even the humble coin needed a rule — call it in the air — to earn its reputation.