The Law of Large Numbers, Intuitively

Short runs wobble; long runs settle near the true probability

Current run

20 flips: 11 heads, 9 tails

55.0%

target for a fair coin: 50%

flips 20

The law does not promise every short run looks fair. It says the running proportion settles down as evidence accumulates.

Flip a coin ten times and you might get 7 heads. Flip it a thousand times and you'll get very close to 500. Flip it a million times and the proportion of heads will be almost exactly 0.5.

This is the law of large numbers — one of the most fundamental results in probability, and one of the most frequently over-applied.

What it says

The law of large numbers says that as a sample grows, the sample average converges to the true population mean.

More precisely: if $X_1, X_2, \ldots$ are independent and identically distributed with mean $\mu$ , then:

\bar{X}_n = \frac{1}{n}(X_1 + \cdots + X_n) \to \mu \quad \text{as } n \to \infty

In the coin example: each flip is 0 (tails) or 1 (heads) with equal probability, so $\mu = 0.5$ . The average over $n$ flips converges to 0.5 as $n$ grows.

Why it works

The key is the standard error: $\text{SD}(\bar{X}_n) = \sigma / \sqrt{n}$ .

As $n$ grows, the spread of the sample mean shrinks. With $n = 100$ , the typical deviation from the true mean is $\sigma / 10$ . With $n = 10{,}000$ , it's $\sigma / 100$ . The average is getting pinned down.

The rate is $\sqrt{n}$ , not $n$ . Quadruple your sample size to halve the error. This is why large studies are better than small ones — and why the improvement becomes slower as you already have more data.

Casino games

A casino game pays out with probability 0.49 and the house wins with probability 0.51. The house edge is tiny — just 2%.

For a single player on a single bet, the outcome is almost a coin flip. They might win. They might win several times in a row.

But the casino runs this game thousands of times per day across hundreds of tables. By the law of large numbers, the casino's average return converges to its 2% edge with high reliability. The casino's revenue is nearly certain. Each individual player's outcome is not.

The same edge that is invisible in one bet becomes an iron guarantee over a million bets. This is why casinos are profitable businesses and why "I've been losing, I'm due for a win" is wrong.

The gambler's fallacy

The law of large numbers is often misread as: "after a run of bad luck, good luck is due to balance things out."

This is false. Coins have no memory. After 10 tails in a row, the next flip is still 50/50. The law of large numbers does not work by compensating for past results — it works by diluting them.

After 10 tails, you need many more heads to bring the average back to 0.5. If you flip 1,000 more times and get 500 heads, the proportion becomes $\frac{500}{1010} \approx 0.495$ — close to 0.5, with the 10-tails streak barely affecting the total. The average converges not by correction but by swamping.

It requires independence

The law of large numbers assumes each observation is independent. If they're correlated — each observation influenced by the previous ones — the sample average may not converge to the population mean at the $\sqrt{n}$ rate, or at all.

Financial returns are not independent. Votes in a family are not independent. Measurements from the same person on consecutive days are not independent. In all these cases, naive application of "more data = more reliable" can give a false sense of precision.

The effective sample size — accounting for correlations — may be far smaller than the nominal $n$ .

What it doesn't guarantee

The law of large numbers guarantees convergence in the long run. It says nothing about the short run.

In 10 flips, getting 8 heads is perfectly possible and not unusual. In 100 flips, getting 70 heads would be surprising. In 10,000 flips, getting 6,000 heads would be extraordinary. The short run is where randomness lives; the long run is where it averages out.

This has a practical implication for data collection: a small study can give a result far from the truth, and this is entirely consistent with the law of large numbers. The law doesn't tell you how large is large enough — that depends on the variance of what you're measuring. High variance (e.g. income, which has enormous outliers) requires much larger samples than low variance (e.g. coin flips).

The law of large numbers is a guarantee about the infinite limit. In practice, you always have a finite sample. The Central Limit Theorem and confidence intervals are the tools that tell you how close you actually are.

See Sampling for the connection to sampling theory, and Normal Distribution for the Central Limit Theorem.