The Unluckiest Number in Every Lottery — the Statistics of the Least-Drawn Ball

The "unluckiest number" is the most-drawn number's twin, and it is produced by exactly the same statistical machinery. One ball must have the smallest count. The only question worth asking is how far below average a fair machine is expected to leave it.

As with its counterpart, this site does not print a fixed "the unluckiest number is 13". Frequency statistics here are computed from imported real draw history, imported game by game, and published on the live pages linked at the end. The answer depends on the window, and it moves.

The minimum of N binomial counts

Same setup as the most-drawn number. A game draws k from N over D draws. Each ball's count is approximately Binomial(D, p) with p = k / N, mean μ = Dp and standard deviation σ = √(Dp(1 − p)).

The expected minimum across N balls mirrors the maximum:

E[min] ≈ μ − σ × z, with z = Φ⁻¹((N − 0.375) / (N + 0.25))

UK Lotto (6 from 59), 1,000 draws

• p = 6/59 = 0.101695, μ = 101.7, σ = 9.56
• z for N = 59 is 2.31
• E[min] ≈ 101.7 − 22.1 = 79.6 appearances

The coldest of 59 balls is expected to land near 80 against an average of 102 — roughly 22% below average — with nothing at all wrong with the machine.

Powerball white balls (5 from 69), 1,000 draws

• p = 5/69, μ = 72.5, σ = 8.20, z for N = 69 is 2.36
• E[min] ≈ 72.5 − 19.4 = 53.1 appearances

About 27% below average, again from a flawless draw.

A small asymmetry, stated honestly

The normal approximation above is symmetric, but a binomial count is very slightly right-skewed. Its skewness is (1 − 2p) / σ, which for the UK Lotto case is (1 − 0.203) / 9.56 = 0.083.

Applying a Cornish–Fisher correction for that skew shifts the estimates apart a little:

• Expected maximum: about 22.7 above the mean (≈ 124.4)
• Expected minimum: about 21.5 below the mean (≈ 80.2)

So the coldest ball sits marginally closer to average than the hottest ball does. It is a real effect and a tiny one — about one appearance in a thousand draws. It matters here only because it is the kind of detail that distinguishes an actual calculation from a plausible-sounding assertion.

The multiple-comparisons problem, quantified

This is the heart of it. When you scan 59 balls for the most extreme one, you are not running one test — you are running 59, and reporting only the winner.

For a single ball under a fair game, the probability of landing beyond ±2σ is about 0.0455. Across 59 independent-ish balls:

P(at least one ball beyond ±2σ) = 1 − (1 − 0.0455)⁵⁹ = 1 − 0.064 = 93.6%

And beyond ±3σ, where a single ball has probability 0.0027:

P(at least one ball beyond ±3σ) = 1 − (1 − 0.0027)⁵⁹ = 1 − 0.853 = 14.7%

For the 69-ball Powerball pool the same arithmetic gives 96.0% and 17.0%.

Read that again. In a perfectly fair 59-ball lottery, there is a 93.6% chance that at least one ball looks two standard deviations out, and roughly a one-in-seven chance that one looks three standard deviations out. A three-sigma result sounds like proof of tampering. Across 59 balls it is a Tuesday.

This is the identical error dissected in clustered pairs and multiple comparisons, where searching thousands of number pairs for the suspicious one guarantees finding several. Searching guarantees finding. The correct response is to correct for the number of tests, or to use a statistic that aggregates all of them at once — see testing hot numbers with chi-square.

Both strategies built on cold numbers are wrong

Cold numbers spawn two opposite superstitions, and both fail for the same reason.

"Avoid the cold ones — they're unlucky." There is no mechanism. The balls have no memory, and a low count is a fact about a finished sequence of past draws.

"Play the cold ones — they're due." This is the gambler's fallacy in its purest form. Under a fair draw the probability that ball 13 appears next is k / N regardless of the previous thousand results. There is no correcting force. The apparent "evening out" over time is the mean growing faster than the spread, not past deficits being repaid — and as the table in the most-drawn number shows, the gap in absolute terms actually keeps widening.

For how long a genuinely cold ball can stay missing before that means anything, see how long a number can go without being drawn.

When a low count would genuinely matter

Coldness becomes evidence only when it survives a properly specified test:

1. A pre-registered hypothesis. "Ball 13 is under-drawn" tested on data collected after you said it is worth something — not chosen because it was already lowest.
2. A chi-square test on the whole pool, whose expected value under fairness is N − 1.
3. A physical mechanism, since statistics alone cannot cause anything. Genuine draw bias comes from ball mass, wear or machine geometry, and it is testable directly — see ball weight and machine bias testing and is the lottery rigged? what auditing involves.

Real draw manipulation, when it has happened, was never detected through frequency charts. It was detected through people: the Eddie Tipton hack was found by tracing a claimant, and the 1980 Pennsylvania fix by betting patterns and informants.

Read the live pages, not a static number

• Powerball frequency statistics
• UK Lotto frequency statistics
• Frequency analyser

Bring the numbers above with you. Compute μ and σ for the draw count shown, and check whether the coldest ball is inside μ − 2.3σ. It almost always is — which is the finding, not a failure to find one.