Home › Guides › Hot and cold numbers

Why Hot and Cold Numbers Don’t Work

This site publishes frequency tables, overdue lists and hot/cold rankings. Here is the calculation showing that every pattern in them is what a perfectly fair machine produces anyway — including the one thing about number choice that is real.

By David Choi · September 4, 2026 · about 9 minutes

Open any lottery statistics site, this one included, and you will find a table of numbers ranked by how often they have been drawn. Number 61 has come up 122 times. Number 13 has come up 78 times. The gap looks enormous, and it invites an obvious conclusion.

The conclusion is wrong, and it is possible to show exactly how wrong with the same data the tables are built from.

1. The claim, and why it cannot be true

Two contradictory folk theories circulate, which should itself be a warning sign:

• Hot numbers — some numbers come up more often, so play them.
• Cold or overdue numbers — some numbers have not come up in a while, so they are due.

They cannot both be right, and in fact neither is, for the same reason: a drawing has no memory. Balls are physical objects in a machine. They do not know their own history and cannot act on it. Each drawing is an independent event with an identical probability distribution, which is precisely what "random" means in this context.

For a number to be genuinely "hot", the physical process would have to be biased — a ball meaningfully heavier, a machine with a defect. Modern lottery draws are held to tolerances specifically designed to prevent that, with balls weighed and sets rotated. And even if a bias existed, it would be far too small to detect within the sample sizes available, for reasons the next section makes concrete.

2. What randomness actually looks like

Here is the thing people get wrong. Random does not mean even. A fair process does not distribute results smoothly — it produces clumps, streaks and gaps, and it produces them reliably.

Flip a fair coin 100 times and you will not get 50 heads. You will typically get somewhere between 40 and 60, and a run of six heads in a row somewhere in the sequence is more likely than not. None of that is evidence the coin is bent.

The same applies to lottery balls, and the size of the expected variation is calculable. If N drawings each select 5 white balls from a pool of p numbers, each number is expected to appear 5N/p times, and the standard deviation of that count is approximately the square root of it.

The interesting question is not "what is the expected count" — it is "how far apart will the highest and lowest counts be, purely by chance?" That is what a hot/cold table actually shows you, and it has a definite answer.

3. Powerball: the expected spread is 46

Powerball has drawn 5 white balls from 69 since October 7, 2015. As of the September 2, 2026 drawing there have been 1,402 drawings under that matrix — 614 at two per week, then 788 at three per week after Monday drawings were added in August 2021.

That gives 7,010 white-ball observations spread over 69 numbers: 101.6 expected appearances per number, with a standard deviation of about 10.1.

We simulated 3,000 alternate universes of 1,402 perfectly fair Powerball drawings and recorded, in each one, the hottest and coldest number. The result:

In 90% of those fair simulations the hottest number landed between 119 and 134 appearances, and the coldest between 72 and 85. A spread of 40 or more occurred in 87% of runs.

Read that again, because it is the whole argument. A gap of forty-plus appearances between the most-drawn and least-drawn Powerball number is not unusual, not suspicious, and not evidence of anything. It is the single most likely thing to happen in a completely fair game. If a lottery's frequency table were flat — every number within a few hits of 101.6 — that would be genuinely strange, and would suggest the draw was not random at all.

Method: 3,000 Monte Carlo simulations of 1,402 draws, each selecting 5 distinct numbers uniformly from 69, using a standard pseudorandom generator. Our own calculation; the code logic is described here in full so it can be reproduced.

4. Mega Millions: the data is far thinner than it looks

Mega Millions is a much starker case, and almost nobody says so.

The game changed matrix on April 8, 2025, moving to 5 of 70 white balls plus 1 of 24 Mega Balls, with the price rising to $5. Drawings before that date were played under different rules and are not comparable — more on that below.

Which means the current-matrix sample is roughly 148 drawings. That is 740 white-ball observations spread across 70 numbers: about 10.6 expected appearances per number.

Running the same simulation on a fair 148-drawing Mega Millions history:

So in a perfectly fair Mega Millions, a number appearing 19 times while another appears 4 times is the ordinary, expected result. One number looks like it appears nearly five times as often as another. Nothing is happening. There is simply not enough data yet for the counts to converge, and there will not be for years — at 104 drawings a year, reaching Powerball's current sample size would take until roughly 2039.

5. The matrix problem most sites ignore

There is a second, more serious error running through this whole category, and it is one of the reasons this site exists.

Both games have changed their number pools repeatedly. Mega Millions has used seven different matrices since 1996 — pools of 50, 52, 56, 75 and 70 white balls, with Mega Ball pools from 15 to 52. Powerball has changed matrix six times since 1992.

A Mega Ball frequency table that pools draws from a 25-ball era and a 24-ball era is not measuring one thing. Ball 25 was drawable before April 2025 and is not now; every other ball's probability changed at the same moment. Counting them together produces a number that describes no real process.

The same applies to the white balls when Mega Millions moved from a 75-ball pool to a 70-ball pool in 2017: numbers 71 to 75 stopped existing, and the odds for 1 to 70 shifted.

Every statistic on this site is computed from current-matrix drawings only, and each page states the matrix and the drawing count behind it. That is unusual, and it is why our Mega Millions counts look so much lower than other sites'. Ours are lower because theirs are pooling incompatible data. The methodology page has the detail.

One thing that is not a matrix change, and is often wrongly treated as one: Powerball added Monday drawings on August 23, 2021. The ball pools did not change, so Monday draws pool perfectly well with Wednesday and Saturday ones. Only the cadence changed.

6. "Overdue" is the same error with a different chart

The overdue view ranks numbers by how many drawings have passed since each last appeared. It is a legitimate description of the past and a completely invalid basis for a prediction.

This is the gambler's fallacy in its purest form: the belief that an independent random process owes a correction. It does not. A number that has not appeared in 60 drawings has exactly the same 5-in-69 chance of appearing in the next drawing as one that appeared last night. The machine has no ledger.

Long gaps are also entirely ordinary. With 69 numbers and 5 drawn per drawing, the chance a specific number is missed in any single drawing is about 92.8%. The chance it is missed 30 times in a row is 0.928 to the 30th power, which is about 11%. Across 69 numbers, gaps of 30-plus drawings are happening somewhere in the table essentially all the time.

7. The one thing about number choice that is real

Here is the part worth keeping, because it is true, it is quantifiable, and it gets buried under all the folklore.

Your number choice cannot change whether you win. It can change how much you win if you do.

Jackpots are split among all tickets matching all six numbers. If you win with a combination that thousands of other people also played, you get a fraction of the prize. If you win alone, you get all of it. The odds of winning are identical either way — the payout is not.

And human number selection is famously non-random:

• 1 through 31 are massively over-played, because people use birthdays and anniversaries. Numbers 32 through 69 are systematically under-chosen.
• Sequences and patterns — 1-2-3-4-5, straight lines and diagonals on the play slip, "lucky" numbers like 7 and 11 — are picked far more often than chance would suggest.
• Numbers from the news, including recent winning combinations, spike immediately after they appear.

So the only defensible strategy in the entire category is a modest one: if you play, include some numbers above 31, and avoid obvious patterns. It will not improve your odds by one part in 292 million. It slightly improves your expected payout conditional on winning, by reducing the chance of a split.

A quick-pick does this for you automatically, which is one reason the majority of jackpot winners are quick-picks — not because the machine is luckier, but because most tickets sold are quick-picks.

8. So why publish the tables at all?

A fair question for a site whose main feature is a hot/cold weighting control.

Three reasons. First, the data is genuinely interesting as a description of what happened, in the same way a season's batting statistics are interesting without predicting the next at-bat. Second, people are going to look up frequency data regardless; better that they find it next to an honest explanation than next to something selling them a system. Third, the simulator's weighting options are a game mechanic — they change how a simulated draw feels, and they are labelled as such on the screen where they are used.

What this site will not do is sell number picks, systems or "winning numbers", or imply that any setting affects a real drawing. Nothing here does. If a site offers to sell you numbers, the simulation above is the reason to close the tab: they are charging you for the shape of ordinary randomness.

If lottery play has stopped feeling like entertainment, the responsible play page has free, confidential help available 24 hours a day.

Sources

• Current matrices and odds: powerball.com prize chart; megamillions.com how to play
• Powerball matrix effective October 7, 2015: NC Education Lottery announcement. Monday drawings from August 23, 2021: powerball.com
• Mega Millions April 2025 changes: megamillions.com, "New Mega Millions arrives in April"; official game rules effective April 5, 2025
• Mega Millions 2013 and 2017 matrix changes: Louisiana Lottery, 2013 and Louisiana Lottery, 2017. Powerball 2012 change: Nebraska Lottery
• Drawing counts and simulation: our own calculation from the draw calendar and 3,000 Monte Carlo runs, described in section 3. Underlying draw data from New York State Open Data.

Keep reading