The statement
The law of large numbers says that as you repeat an independent random trial more and more times, the average of the results tends to move closer to the expected value of a single trial.
For a fair die, a single roll has an expected value of 3.5. After a handful of rolls the average can be anywhere. After thousands, it is very likely to be near 3.5. For a fair coin scored as heads = 1, the average tends towards 0.5.
That is the whole promise, and it is narrower than it sounds.
How quickly the average settles
The spread of an average shrinks with the square root of the number of trials. A fair die has a standard deviation of about 1.71 per roll. The typical error in the average of n rolls is roughly 1.71 divided by the square root of n.
- 100 rolls: about 1.71 / 10 = 0.17.
- 10,000 rolls: about 1.71 / 100 = 0.017.
One hundred times more rolls cut the typical error by a factor of ten, not a hundred. Convergence is real but gradual, and it is statistical, not exact. Averages do not lock onto the expected value at any particular point.
What the law does not say
It does not say early deviations are corrected
Suppose a fair coin gives 60 heads in the first 100 flips, 10 above the even split. The coin has no memory. The next 900 flips are expected to give about 450 heads, so after 1,000 flips the expected total is 510 heads, or 51%.
The proportion moved from 60% to 51% towards the average, yet the excess of 10 heads was never cancelled. It was diluted by the volume of later flips. The law works through dilution, not compensation.
It does not say a streak must reverse
The gambler’s fallacy is the belief that after a run of one outcome the opposite becomes more likely, so things “balance out”. The law of large numbers makes no such claim. For independent trials, the chance of the next outcome is the same whatever came before. This mistaken belief is also called the law of averages, and it is a misreading of the actual law.
It does not say the raw difference stays small
The standard deviation of the count of heads grows as flips increase (it is 5 for 100 flips and 50 for 10,000). The percentage tightens while the absolute gap can widen. A short run versus long run comparison shows this clearly.
It does not say a particular session will match the average
A finite session can finish well above or below the long-run average. The law describes behaviour as the number of trials grows very large. It is not a forecast for the next hour.
Why it matters for games with a house edge
A casino game with a house edge has a negative expected result per unit staked. The law of large numbers says the average result per bet will tend towards that negative figure as the number of bets grows. This is the mathematical basis of the long-run advantage of the house. An even-money bet on a single-zero roulette wheel has an expected result of -1/37 per unit, about -2.7%, and across a very large number of bets the average result per bet tends to sit near that.
The same law also protects fairness testing. If a die or generator is claimed to be fair, observing a large number of rolls and comparing frequencies is a sensible check, because small samples are too noisy to settle the question. A Monte Carlo simulation relies on the same principle: more trials produce steadier estimates.
Weak and strong forms
Mathematicians distinguish a weak law and a strong law, which differ in the exact way convergence is stated. For everyday purposes the common message is the same: given enough independent trials, the sample average is very likely to be close to the expected value, and almost surely converges in the strong form. Both versions assume the trials are independent and share the same distribution, with a finite expected value.
A short summary of the contrast
- Law of large numbers: the average of many independent trials tends towards the expected value.
- Gambler’s fallacy: believing a past streak changes the next independent outcome.
The first is a statement about averages over many trials. The second is a mistaken claim about a single trial.
Nothing here makes it possible to profit from a negative-expectation game. Outcomes are uncertain, losses can be large, and gambling can be harmful. If it is becoming a problem, a local support service can help.