Is the lottery random?

We ran three decades of US lottery draws through the same statistical tests used to check random number generators. This page shows what we found, including the tests that were designed to catch us out.

Primary dataset: Texas Pick 3 — 26,004 draws since 1993-10-25. Computed 2026-08-24.

Do the same numbers come up on the same date?

This is the most common claim we get asked about. The honest answer needs a denominator: with 537,705 pairs of draws sharing a calendar date, and a 1-in-1,000 chance each pair matches, coincidences are guaranteed.

Match windowPairs comparedRepeats foundExpected by chanceRatio
Exact same date 537,705 535 537.7 0.99×
Within 1 day 4,878,124 4,786 4,878.1 0.98×
Within 3 days 26,515,836 25,981 26,515.8 0.98×
Within 7 days 120,813,200 118,847 120,813.2 0.98×

Widen the window and the number of "amazing coincidences" climbs — but so does the number chance predicts, by the same factor. That is the entire trick.

Are numbers that haven't appeared "due"?

For each digit we measured how often it appeared next, grouped by how many draws it had been absent. If long-gap numbers were due, this line would slope upward.

20.0%25.0%30.0%35.0% expected 27.1% absent 1 draws: 27.4% over 39,599 opportunitiesabsent 2 draws: 26.9% over 28,758 opportunitiesabsent 3 draws: 26.9% over 21,030 opportunitiesabsent 4 draws: 27.4% over 15,368 opportunitiesabsent 5 draws: 27.1% over 11,154 opportunitiesabsent 6 draws: 27.0% over 8,134 opportunitiesabsent 7 draws: 26.6% over 5,940 opportunitiesabsent 8 draws: 27.1% over 4,361 opportunitiesabsent 9 draws: 27.0% over 3,180 opportunitiesabsent 10 draws: 27.1% over 2,322 opportunitiesabsent 11 draws: 26.5% over 1,693 opportunitiesabsent 12 draws: 27.4% over 1,245 opportunitiesabsent 13 draws: 29.2% over 904 opportunitiesabsent 14 draws: 34.7% over 640 opportunitiesabsent 15 draws: 26.8% over 418 opportunitiesabsent 16 draws: 32.4% over 306 opportunitiesabsent 17 draws: 28.5% over 207 opportunitiesabsent 18 draws: 27.7% over 148 opportunitiesabsent 19 draws: 29.0% over 107 opportunitiesabsent 20 draws: 28.9% over 76 opportunitiesabsent 21 draws: 16.7% over 54 opportunitiesabsent 22 draws: 37.8% over 45 opportunities 1 draw 22 draws absent

Flat at 27.1%, which is exactly 1 − 0.9³ — the chance a given digit appears in any draw. A digit absent for 20 draws is no more likely to appear than one drawn yesterday. Across all 22 gap lengths: χ² = 35.8, p = 0.032. The widest bars are the longest gaps, where there is least data — see the limitations note below.

Can you tell real draws from a computer's?

One of these is real. The other came from a random number generator.

Sequence A

1-1-5
4-6-2
5-4-4
6-1-9
9-1-3
0-7-5
3-7-2
7-5-0
1-1-5
4-9-2
2-0-6
5-0-8
5-9-3
3-8-4
0-3-9
5-1-1
5-6-5
6-1-7
5-0-4
9-8-9

Sequence B

8-7-7
4-0-7
8-1-2
6-9-4
8-9-2
1-9-3
0-7-3
5-2-9
1-5-2
5-7-0
6-7-3
8-9-9
6-7-3
7-9-1
5-9-4
4-2-2
7-5-6
9-3-7
4-4-5
1-5-0

Do these tests actually detect anything?

A null result is worthless if the tests are asleep. So we ran the identical code over datasets we deliberately broke, at the same sample size.

DatasetFrequency testSerial test
Real lottery draws
The state data itself		Consistent
p 0.422		Consistent
p 0.225
Seeded PRNG
Known-good pseudo-random source		Consistent
p 0.440		Consistent
p 0.615
Weighted digit (6 at 11.5%)
A biased ball set		Deviation
p < 0.000000000001		Consistent
p 0.483
Seeded recurrence (+3 mod 10)
A deterministic formula		Consistent
p 1.000		Deviation
p < 0.000000000001

Each fault is invisible to one test and obvious to the other. A weighted ball set does not disturb the order of draws; a deterministic formula visits every digit equally often, so it passes a frequency test while the serial test rejects it outright. Both fakes were caught. The real data was not.

Everything, in one chart

We ran 194 separate uniformity tests — every state, game and digit position with enough history. If nothing is going on anywhere, the p-values from those tests should themselves be spread evenly between 0 and 1.

p between 0.0 and 0.1: 19 testsp between 0.1 and 0.2: 16 testsp between 0.2 and 0.3: 23 testsp between 0.3 and 0.4: 26 testsp between 0.4 and 0.5: 22 testsp between 0.5 and 0.6: 14 testsp between 0.6 and 0.7: 21 testsp between 0.7 and 0.8: 12 testsp between 0.8 and 0.9: 18 testsp between 0.9 and 1.0: 23 tests flat = 19.4 per bin p = 0 p = 1

10 tests came in under p = 0.05, against 9.7 expected by chance alone. Testing the p-values for uniformity in turn: p = 0.429. Consistent with randomness

What this does and does not show

These tests show the published sequences look random. They cannot rule out narrowly-targeted manipulation of a handful of draws — no statistical test on the numbers can, and historically real lottery fraud has been caught at the money, not in the sequence. This is an MVP covering five tests; the full battery is larger. Individual states are underpowered on their own, which is why the last chart pools them. Nothing on this page feeds the number-picking tools, and none of it improves anyone's chance of winning. Lottery draws are independent and random — that is the finding, not a disclaimer.

Play for entertainment. If gambling is causing harm, call 1-800-GAMBLER.

Spotted an error in the data? ⚠ Report incorrect data

DrawAnalytics is an informational and entertainment service. We provide historical lottery data analysis and pattern exploration tools. We do not sell predictions, we do not guarantee any outcome, and we make no representation that any tool on this site improves a user's probability of winning any lottery game. Lottery drawings are random. Past results do not predict future drawings. You must be 18 or older (21+ in some states) to play state lottery games. If you or someone you know has a gambling problem, call 1-800-GAMBLER or visit ncpgambling.org.