Do Hot and Cold Numbers Actually Work? A Statistical Analysis

Why everyone talks about hot and cold numbers

Every lottery site, every app and half the conversations at the counter mention hot and cold numbers. The idea is seductive: some numbers are supposedly running hot and will keep coming, while others are cold and therefore due.

We publish hot and cold lists ourselves — they appear on every Powerball statistics and Mega Millions statistics page — but the label describes the past, not the future. This article puts the claim under a microscope with our own data: 1,406 Powerball draws and 929 Mega Millions draws under the current rules, plus a chi-square test that asks the blunt question: is any number behaving in a way fairness cannot explain?

The short answer: no. But the way the data fail to show a bias is more interesting than the answer, and it explains why frequency analysis is still worth doing.

What hot and cold actually measure

A hot number is one that appeared more often than average in the window you are looking at. A cold number appeared less often. That is the whole definition: a count divided by a window — no model, no trend detection, no physics.

The arithmetic is worth doing once. Every draw takes 5 numbers from a pool of 69, so each number has a 5/69 = 7.25% chance of appearing in any given draw. Over n draws the expected count is n × 5/69 and the standard deviation is the square root of n × p × (1 − p). For Powerball:

• Last 300 draws: expected ≈ 21.7 appearances per number, σ ≈ 4.5.
• Last 500 draws: expected ≈ 36.2, σ ≈ 5.8.
• Since October 2015 (1,406 draws): expected ≈ 101.9, σ ≈ 9.7.

Expected value and standard deviation are what turn a raw count into a judgment. In the last 500 draws, 64 appeared 51 times (+2.5σ) and 10 appeared 26 times (−1.8σ). Neither is extraordinary: with 69 numbers, a couple landing two sigmas from the mean is normal behavior in a fair game.

The label also depends on the window: numbers that look hot over 300 draws can look average over 1,400, because a short window has a much larger standard deviation relative to the mean (σ/expected is 21% over 300 draws, 9.5% over 1,406). That is why our statistics pages offer 100, 300 and 500 draws, and why a 20-draw list is noise dressed as insight.

The gambler's fallacy: 'this number is due'

The gambler's fallacy is the belief that a random process owes you something. If a coin lands heads ten times in a row, the next flip is still 50/50: the coin has no memory. Likewise, a Powerball number that has not appeared for 60 draws still has exactly a 7.25% chance in the next draw.

Our own data contains long gaps that feel meaningful and are not. The coldest number in the current Powerball matrix is 13, with 76 appearances against an expected 101.9 — 2.7 standard deviations below average. That is an unusual run of absence with no predictive content: the balls drawn on Saturday do not know that 13 has been quiet.

The hot-hand fallacy: the same mistake in reverse

The symmetric error is believing a number is on a streak. In our 1,406-draw Powerball sample the three most frequent numbers — 21, 61 and 64 — are tied at 125 appearances, +2.4σ. That is the entire claim behind "hot": a number was seen slightly more often than average, so it supposedly will be again.

For a human athlete the hot-hand debate is arguable, because a person can be in form, tired or injured. A ball machine has no form, and nothing in the drawing equipment carries information between draws: the mixing and the selection reset every time. A number that appeared in the last three draws is exactly as likely tonight as one that has not appeared since spring.

If you want the intuition in one line: streaks are something your eye adds to random data. A sequence of 500 fair draws is supposed to contain clusters, gaps and coincidences. That is what randomness looks like.

What probability actually says

Two ideas carry the argument. Independence: each draw is a fresh experiment, so previous results are irrelevant. Uniformity: every ball is identical, so every number — and every combination — has the same chance as any other.

Powerball (5 from 69 plus 1 from 26) has 292,201,338 possible tickets; Mega Millions (5 from 70 plus 1 from 25) has 302,575,350. Your ticket, my ticket and the sequence 1-2-3-4-5 all have the same probability — and no history of past draws changes it.

There is a real edge hiding in frequency data, but not the one people expect. Combinations are equally likely to be drawn, but not equally likely to be played: a line like 1-2-3-4-5 or a birthday-heavy set is chosen by thousands of people, so if it wins the jackpot is split many ways. The honest use of past draws is avoiding popular patterns, not chasing hot numbers — see how to choose lottery numbers.

Real data: Powerball since October 2015

Powerball changed its matrix on 7 October 2015, moving the main pool from 1–59 to 1–69. Our database holds 1,406 draws in that format; the chart below counts how often each number appeared, against the expected 101.9.

The observed counts run from 76 to 125. The gap between the hottest (125) and the coldest (76) looks dramatic until you compare it with the spread a fair game produces: 49 appearances is 5.0 standard deviations wide, but each individual deviation is only about 2.5σ from the mean.

Five of the 69 numbers sit outside ±2σ; a fair game this size would produce about 3.1. Five versus three is not evidence of anything — it is the kind of difference that appears and disappears as the window slides.

Real data: Mega Millions since October 2017

Mega Millions moved to its current 5/70 matrix on 31 October 2017, and our database has 929 draws in that format. Expected appearances are 66.4 per number with σ = 7.85; the observed range is 52 to 86 — narrower than Powerball's because the sample is smaller, and about the same width in standard deviations.

The hottest numbers are 10 (86, +2.5σ), 17 (84, +2.3σ), 31 (80), 3 (79) and 42 (78); the coldest are 67 and 51 (52, −1.8σ), 65 and 45 (53) and 23 (55). Only two of the 70 numbers fall outside ±2σ, against roughly 3.2 expected by chance.

Two games, two matrices, two sample sizes — and the same picture. Both scatter around 1.0 without drift, without clusters, and without any number that behaves consistently across windows. That is what a fair draw looks like when you plot it.

The chi-square test, in plain English

Eyeballing charts settles nothing, so here is the formal test. A chi-square goodness-of-fit compares every observed count with its expectation, adds up the squared differences divided by the expectation, and asks how often a fair process would produce a statistic that large. The p-value is that probability.

One technical detail is routinely ignored. Because each draw takes 5 numbers without replacement, the probabilities sum to 5 rather than 1, so the usual Pearson statistic has an expected value of N − k = 64 instead of N − 1 = 68, and the naive test reports p-values slightly too large. The fix is a single multiplier, (N − 1)/(N − k) = 68/64 for Powerball.

Every p-value is far above the 0.05 threshold: there is no evidence of a biased ball, a favored number or a pattern worth exploiting in either game. The data are exactly as lumpy as fair draws are supposed to be.

The matrix change trap: how a naive test invents a bias

Here is the part that should make you suspicious of every viral "lottery is rigged" analysis. Run the same chi-square over the whole Powerball history — 1,997 draws back to 2010 — and you get χ² = 173.3 with 68 degrees of freedom and p < 0.001. The five coldest numbers are 65, 60, 68, 66 and 67, all between −3.5σ and −4.5σ, and 13 of the 69 numbers fall outside ±2σ. It looks like a smoking gun.

It is an artifact. Those five numbers are 60 or above, and numbers 60 to 69 did not exist before October 2015. The test compared 1,406 draws from a 69-ball game with 591 draws from a 59-ball game, so the extra ten numbers were always going to look under-drawn. Split the data at the change and the effect vanishes: the 5/59 era gives p = 0.99, the current 5/69 era p = 0.10.

Mega Millions is worse: its full history contains 75 distinct main numbers (1–52, 1–56, 1–75 and today's 1–70), so a chi-square over 70 numbers is not even defined there. Powerball's main pool changed on 7 October 2015, and Mega Millions moved to 1–70 on 31 October 2017. Our own database confirms both switches: no Powerball draw before 7 October 2015 contains a number above 59, and no Mega Millions draw since 31 October 2017 contains a number above 70 — while the 5/75 era produced 119 draws that did.

What the charts really show

Go back to the bar chart and look at how ragged the line is. Runs of tall bars alternate with runs of short ones, and the eye wants to name the groups: these are hot, those are cold. But each bar is a single binomial draw, and binomial draws are lumpy: the widest deviation, number 13 at 76 appearances, is a 2.7σ event, and with 69 numbers one that large turns up in roughly four samples out of ten.

The scatter chart makes the same point about both games at once. If frequency data carried information, some numbers would sit consistently above or below 1.0 across windows. They do not: the two clouds look like handfuls of sand thrown at a line, the signature of independent random draws.

Statistics cannot prove a drawing machine is fair. What it can do is flag a result unlikely enough to deserve a physical explanation: a number at +5σ, a p-value below 0.001 that survives era-matching and repeated windows, or a bias that reappears in independent datasets. Nothing in our data comes close to that bar.

If frequencies do not predict, why look at them?

Because prediction is not the only reason to count things. Frequency data has four honest uses, and none of them involve forecasting the next draw:

• It teaches probability. Watching a fair process look lumpy is the fastest cure for the feeling that patterns mean something.
• It audits the data. Published draw histories are the lottery's own record, and counting them is how the matrix change in this article surfaced.
• It helps you avoid popular combinations. Not hot numbers, but popular ones: birthdays under 31, sequences, the line everybody plays. Your odds of matching the draw do not change, but your odds of sharing a jackpot with 300 other people do — the only measurable edge a player has.
• It is genuinely interesting. There is nothing wrong with enjoying the analysis of a game you play, as long as you are honest about what it cannot do.

That is why our statistics pages show windows of 100, 300 and 500 draws and print the expected value next to every count. A frequency without a baseline is decoration, not information.

Using the HOT/COLD bias — and the honest conclusion

Our generator includes an optional number bias with two settings — Favor hot numbers and Favor cold numbers — next to the default Pure random. It weights the picker towards the most (or least) frequent numbers of the last 100 draws, so the lines it produces lean that way. What it cannot do — because nothing can — is improve the probability that any single line wins.

Used honestly, the bias turns a preference into a ticket: if you enjoy playing the numbers you have been watching on our Mega Millions statistics page, it saves you typing them out. But do not confuse a preference with an edge — the generator's HOT and COLD bias is entertainment, like the game itself.

So, do hot and cold numbers work? As descriptions, yes: they say which numbers appeared more or less often in a window. As predictors, no — and the statistics behind that "no" are more useful than the myth they replace. Every draw is independent, every combination equally likely, and the only things you control are how much you spend and how you feel about it.