Plain-language answer
The law of large numbers is a precise mathematical result: as you repeat an independent random process more and more times, the average outcome converges toward the process’s true underlying probability. Flip a fair coin 10 times and 7 heads is unremarkable; flip it 10,000 times and getting 70% heads overall would be extraordinary. The law describes what happens to the long-run average — it makes no claim about any particular short sequence along the way.(Grinstead & Snell, 1997)
Why it matters
The law of large numbers is often paraphrased, incorrectly, as “things even out” — implying that a run of bad luck must soon be balanced by a run of good luck, or that an unusual coincidence early in a process makes similar coincidences less likely afterward. Neither is true. The coin does not “remember” that it came up heads seven times in a row; the mathematics only says that as many more flips are added, the influence of any early run shrinks relative to the growing total. That is convergence through dilution, not correction.
Worked example: convergence without correction
Suppose a fair coin lands on heads 8 times out of the first 10 flips (80% heads). The law of large numbers guarantees that, over millions of further flips, the overall proportion of heads will settle very close to 50%. It does this not by generating extra tails to “cancel out” those early heads, but because 8 extra heads become a vanishingly small share of a much larger total. After 10 flips, 8 heads is a large deviation from 50%; after 10,000,010 flips, the same absolute excess of 8 heads is statistically invisible. The past imbalance is never undone — it is simply outweighed.
This distinction matters directly for coincidences: an early cluster of matches, hits, or notable events in a small sample is not evidence that something has gone wrong with “randomness,” and it does not predict a compensating drought afterward. The formal law of large numbers concerns what a very long run looks like in total; it says nothing about the short-run streaks and clusters that are the raw material of most coincidence stories, which is exactly the domain covered by clustering and runs.
Common misconception
“It’s due” — the belief that a streak makes the opposite outcome more likely soon, so the process can “balance out” — is a well-known reasoning error, not a consequence of the law of large numbers. It sometimes travels under the name the gambler’s fallacy. The actual law of large numbers requires no such compensation and would still hold true even if every early flip happened to land on heads, because the effect of any finite run shrinks toward zero as the number of trials grows, not because later trials tilt to compensate for earlier ones.
Limits and open questions
The law of large numbers assumes the underlying process is genuinely stationary and identically distributed — the same fair coin, the same biased die, unchanging across trials. Real-world processes people apply “it evens out” reasoning to (luck, sports form, market prices) frequently are not stationary in this sense, which means the law’s guarantees may not transfer even when its slogan gets invoked.
Related
- The law of truly large numbers covers the informal (and differently named) idea that with enough opportunities, even very rare events become likely to occur to someone.
- Clustering and runs covers what genuinely random short sequences look like, including why they contain more streaks than intuition expects.
Key takeaways
- The law of large numbers describes convergence of a long-run average toward a true probability — not correction of short-run streaks or imbalances.
- A run of “unusual” outcomes does not make the opposite outcome more likely afterward; each trial in an independent process has no memory of what came before.
- Confusing “long-run averages converge” with “short-run imbalances get corrected” is the mathematical root of the gambler’s fallacy.