The core statement
The law of large numbers says that as you repeat a random trial more and more times, the average of the results tends to get closer to the true average of a single trial. Roll a fair die once and you might get a 6. Roll it thousands of times and the average of all the rolls will usually sit very near 3.5.
Two words in that sentence do a lot of work: “average” and “tends”. The law is about averages, not about totals or streaks, and it describes a tendency rather than a schedule. It says nothing about what the next roll will do.
Share versus gap
A coin shows the difference clearly. Flip a fair coin n times. The proportion of heads gets closer to one half as n grows. But the raw difference between heads and tails does not shrink; it typically grows.
The typical size of the departure from the exact middle is described by the standard deviation of the number of heads, which is half the square root of n for a fair coin:
| Flips | Typical gap from n/2 in head count | Typical gap in share of heads |
|---|---|---|
| 100 | about 5 | about 5 percentage points |
| 10,000 | about 50 | about 0.5 percentage points |
| 1,000,000 | about 500 | about 0.05 percentage points |
The “typical gap” is one standard deviation, so results beyond it are quite normal. After a million flips you would not be surprised to be a few hundred heads away from exactly half, yet the share is within a tiny fraction of a percent of 50. The proportion has settled even though the head count has wandered further from the middle.
Settling is not compensating
A frequent misunderstanding is that a long run of one outcome must be balanced by the opposite later. The law of large numbers does not say this.
Suppose a fair coin lands heads 10 times in a row. The next 1,000 flips are still expected to give about 500 heads and 500 tails. The overall total of 1,010 flips is then expected to hold about 510 heads, a share of about 50.5 percent. The early run was not corrected. It was diluted by the later flips until it hardly matters to the share.
Believing that a correction is owed is the gambler’s fallacy. Because independent trials have no memory, each new flip, roll or spin starts fresh with the same chances. A related belief, hot hand belief, expects streaks to continue; for independent chance events that has no support either.
What this does tell you
- Estimating chances from data. If you track many repeats of the same event, the observed rate becomes a better guide to the true rate. This is the reason a larger sample size makes an estimate more trustworthy, though how much better follows a square-root pattern rather than a straight line.
- Why games with a built-in edge are reliable for the operator. A single bet is unpredictable, but across a huge number of bets the average result per bet approaches its expected value. That is why a house edge is dependable for a casino handling vast numbers of bets, while any one player’s session can go either way. For a player, the same law means that repeated play with a negative expectation moves, on average, toward a loss.
- Why simulation works. In a Monte Carlo simulation, running many trials and averaging the outcomes gives a good estimate of the true value.
What it does not tell you
- It does not say that short runs look like the long-run average. Small samples are bumpy and often far from the mean.
- It does not guarantee that luck evens out for you personally. The spread of the total outcome often grows with more plays, even though the average per play settles.
- It does not say a game is beatable, or that continued play converts losses into wins. In a game with the odds set against the player, more plays make the average loss per play more predictable, not smaller.
- It does not apply in the same way to every kind of random process. Some quantities can be so heavy-tailed that averages settle extremely slowly, which matters for games with rare, very large payouts.
- It applies to independent trials from a stable situation. If the odds change between plays, or your skill changes, the average being approached is itself moving.
A dice illustration
Roll one die repeatedly and keep a running average. After a few rolls it may be anywhere from about 2 to 5. After 1,000 rolls the standard deviation of the running average is about 1.71 divided by the square root of 1,000, which is roughly 0.054. So the average will typically be within a tenth of a point of 3.5. After 100,000 rolls the typical wobble is around 0.005. The average hones in on 3.5, but no individual roll ever becomes more likely to be a 3 or a 4 because of earlier results.
Practical takeaways
Use the law to judge how much data you need before trusting an average, and to remember that rare events can still cluster in the short run. Do not use it to predict the next outcome, to justify raising stakes after a losing run, or to believe you are due a win. Chasing losses on that reasoning has no mathematical basis.
Outcomes in chance games remain uncertain in the short run, and the long run favours the house when the game is built that way. If gambling is causing harm, a local support service can offer help.