The Law of Large Numbers states that as the number of independent, identically distributed trials increases, the sample average converges to the theoretical expected value (population mean).
A: Event A
B: Event B
A two-event view of law of large numbers.
Quick Answer
The Law of Large Numbers states that as the number of independent, identically distributed trials increases, the sample average converges to the theoretical expected value (population mean). In plain terms, a small sample can be far off, but the proportion of heads in 10,000 fair flips will almost surely be very close to 0.5, and a casino's average take per bet settles at its house edge. Use it to explain why large samples give reliable estimates and why a short run of luck proves little. It describes averages over many trials and says nothing about the next single trial.
Why This Matters
The law is why insurance, casinos and opinion polls work: no one can predict one car crash, one roulette spin or one voter, but the average over thousands is steady enough to plan around. It is also why a survey of 1,000 beats a survey of 20, and why a rookie batting .500 after 10 at-bats proves little. Misread it and you fall for the gambler's fallacy, waiting for a run of losses to 'balance out'.
Intuitive Explanation
Flip a fair coin and track the running proportion of heads. After 10 flips you might have 7 heads, 0.70. After 100, perhaps 54, 0.54. After 10,000, something like 5,040, 0.504. The proportion closes in on 0.5. Its typical distance from 0.5 is about n0.5: roughly 0.16 after 10 flips, 0.05 after 100 and 0.005 after 10,000.
The coin closes in without remembering anything, by dilution rather than correction. Suppose the first 10 flips gave 7 heads, 2 more than half. Every later flip is still 50-50, so after 1,000 more flips you expect about 500 more heads, for 507 out of 1,010, or 0.502. The 2 extra heads are still there, but they are swamped. In fact the gap between the counts of heads and tails tends to grow as you keep flipping; only the proportion settles.
The law needs many independent repeats of the same process, and it promises nothing about a small number of tries. A 1-in-100 chance does not have to happen within 100 tries; it fails to happen at all in about 37% of such runs, since 0.99100≈0.37. Nor does the law rescue a biased sample: polling 10,000 people from one neighborhood gives a very precise estimate for the wrong population.
Think of it like this.An early run of luck is like a spoonful of salt in soup: in one cup it is overpowering, but stir it into a big pot and you can barely taste it. Nothing removes the salt; more soup dilutes it. In the same way, extra heads stay in the count but fade in the proportion.
Core idea.Averages of many independent trials settle near the true expected value because each new trial dilutes earlier luck, not because the process corrects itself; the law describes the long run and says nothing about any single trial.
When to Use
Use the law of large numbers when a question is about many trials or a large sample: why a casino profits over a year, why a long season shows a player's true skill, why estimates get steadier as n grows. It also answers 'is heads due?' (no). For the chance of a single outcome, use Basic Probability; for how much an average of n values varies and what shape its distribution has, use the Central Limit Theorem; for a margin of error around an estimate, use a Confidence Interval.
Tip.Ask: Is the question about how an average or a proportion behaves as the number of independent trials grows very large?
How to Recognize It
Before citing the law, check that the claim is about an average over many repeated trials, not about any single outcome.
1.
Is the claim about the proportion or average across many trials rather than about the next trial?
Yes is where the law applies. A claim about the next flip, such as tails being due after five heads, is the Gambler's fallacy; each flip stays 50-50.
2.
Are the trials repeats of the same process, each independent of the others?
Yes means the average will settle at the true mean. If conditions keep changing, as with a player whose skill improves all season, there is no single value for the average to settle on.
3.
Is the question about where the average settles rather than the shape or spread of many sample averages?
Where it settles is this law. The bell shape of sample means, and how wide that bell is, belong to the Central Limit Theorem and the standard error.
4.
Is the sample large, hundreds or thousands of trials, rather than a handful?
Large samples justify trusting the average. With 10 flips of a fair coin, the share of heads is off by 0.2 or more about a third of the time, which is Sampling Variability at work, not a settled long-run value.
Law of Large Numbers vs Central Limit Theorem vs Experimental Probability vs Gambler's fallacy
Each of these comes up when chance plays out over many tries, and the popular 'law of averages' blends them into one wrong idea. Ask whether the claim is about where averages head, their bell-shaped spread, one measured fraction, or the next single trial.
Law of Large Numbers vs Central Limit Theorem vs Experimental Probability vs Gambler's fallacy
Type
Meaning
Key test
Formula
Example
Law of Large Numbers
The Law of Large Numbers says the average of many independent repeats of a chance process settles ever closer to its true mean.
About where the average ends up as trials grow; silent about the next trial.
Xˉn→μ as n→∞
A fair coin's share of heads is often 0.3 or 0.7 after 10 flips but very close to 0.5 after 100,000.
Central Limit Theorem
The Central Limit Theorem says that means of large random samples follow an approximately normal, bell-shaped distribution around the true mean.
About the shape and spread of many sample means, used to find probabilities.
z=σ/nxˉ−μ
Means of 48 rolls of a fair die cluster in a bell shape around 3.5, about 95% of them between 3.0 and 4.0.
Experimental Probability
Experimental probability is the fraction of successes observed in one particular set of trials.
One measured number; the law explains why it becomes reliable as trials pile up.
p^=nx
A spinner lands on green 260 times in 1,000 spins, an experimental probability of 0.26.
Gambler's fallacy
The gambler's fallacy is the mistaken belief that a run of one result makes the opposite result more likely next time.
A claim about the next single trial; independent trials have no memory.
—
Betting on black because the roulette wheel has landed on red eight times in a row.
Law of Large Numbers
Meaning
The Law of Large Numbers says the average of many independent repeats of a chance process settles ever closer to its true mean.
Key test
About where the average ends up as trials grow; silent about the next trial.
Formula
Xˉn→μ as n→∞
Example
A fair coin's share of heads is often 0.3 or 0.7 after 10 flips but very close to 0.5 after 100,000.
Central Limit Theorem
Meaning
The Central Limit Theorem says that means of large random samples follow an approximately normal, bell-shaped distribution around the true mean.
Key test
About the shape and spread of many sample means, used to find probabilities.
Formula
z=σ/nxˉ−μ
Example
Means of 48 rolls of a fair die cluster in a bell shape around 3.5, about 95% of them between 3.0 and 4.0.
Experimental Probability
Meaning
Experimental probability is the fraction of successes observed in one particular set of trials.
Key test
One measured number; the law explains why it becomes reliable as trials pile up.
Formula
p^=nx
Example
A spinner lands on green 260 times in 1,000 spins, an experimental probability of 0.26.
Gambler's fallacy
Meaning
The gambler's fallacy is the mistaken belief that a run of one result makes the opposite result more likely next time.
Key test
A claim about the next single trial; independent trials have no memory.
Example
Betting on black because the roulette wheel has landed on red eight times in a row.
Formula & Notation
Worked Examples
Example 1 — Dilution of an early run of heads
Easy
A fair coin lands heads 8 times in its first 10 flips. (a) What proportion of heads do you expect after 990 more flips? (b) Does the coin now 'owe' tails?
Show the steps and the answer
Expected heads in the next 990 flips: 990×0.5=495.
Each future flip is still 50-50; the coin has no record of the first 10.
Expected total: 8+495=503 heads out of 1,000 flips.
The 8 early heads stay in the count.
Proportion: 1000503=0.503.
The 3 extra heads from the start, out of 1,000 flips, move the proportion only 0.003 above one half.
Compare: after 10 flips the proportion was 0.8, which is 0.3 above one half; after 1,000 it is expected to be 0.503.
The proportion closes in through dilution; tails never has to catch up.
Answer:(a) About 0.503. (b) No. The next flip is still 50-50, and the early run of heads fades in the proportion without any make-up tails.
Takeaway:The law works by swamping early luck with many new trials, not by changing the odds of future trials.
Example 2 — A long run is not a guarantee
Standard
A raffle sells 100 tickets each week, and you buy 1 ticket every week for 100 weeks. Does the law of large numbers guarantee that you win at least once?
Show the steps and the answer
Each week P(win)=1001=0.01, and the weeks are independent.
Every weekly draw is a fresh 1-in-100 chance.
Find P(no win in 100 weeks): 0.99100≈0.366.
You must lose every one of 100 independent draws, so multiply 0.99 by itself 100 times.
So P(at least one win)≈1−0.366=0.634.
'At least one win' is the complement of 'no wins'.
State what the law does say: over many thousands of weeks, the fraction of weeks you win settles near 0.01.
It is a statement about proportions over a very long run, not a promise that a 1-in-100 event happens within 100 tries.
Answer:No. P(at least one win in 100 weeks) ≈ 0.63, so about 37% of the time you win nothing at all.
Takeaway:The law describes long-run proportions; it does not schedule rare events to happen within a fixed number of tries.
Example 3 — Find the mistake: expecting the counts to even out
Application
Max flips a coin 10,000 times and gets 4,940 heads. He says: "The law of large numbers says heads and tails should come out equal by now. I'm 60 short of 5,000, so the coin must be unfair." Find the mistake and correct it.
Show the steps and the answer
Compute the proportion: 100004940=0.494, only 0.006 from 0.5.
The law promises that the proportion gets close to 0.5, and 0.494 is close.
Find the typical gap in the count for a fair coin: 10000×0.5×0.5=50 heads.
In n fair flips the number of heads typically lands within about n/2 of n/2.
Compare: 4,940 is 60 below 5,000, which is 1.2 typical gaps. A fair coin misses 5,000 by 60 or more, in either direction, in about 23% of runs of 10,000 flips.
Something that happens almost a quarter of the time is not evidence of an unfair coin.
Correct the belief: the counts are not pulled together. The typical gap grows with n, about 50 at 10,000 flips and about 500 at 1,000,000 flips, while the gap in the proportion shrinks.
500 out of 1,000,000 is a proportion gap of only 0.0005, so both statements hold at once.
Answer:There is no evidence the coin is unfair: 0.494 is a normal result for 10,000 fair flips. The law says the proportion approaches 0.5, not that the counts become equal.
Takeaway:Convergence is about proportions. The gap in raw counts can grow even as the proportion settles.
Where You'll See It
Free-throw shooting percentage settling down
A player who is truly a 75% shooter might hit 6/10 (60%) on a bad day, but over a whole season of hundreds of attempts the percentage settles near 75% — more trials make the average reliable.
Casino's edge over many bets
On a single spin anything can happen, but because a casino takes millions of bets, its actual payout converges to the expected (house-edge) value, which is why casinos profit in the long run.
Spinner fairness check
Spinning a 4-color spinner 12 times might give lopsided counts, but spinning it 1,000 times makes each color land close to 25% — a lopsided start gets diluted as trials pile up.
Common Mistakes
The slip:After a streak of heads, tails is 'due' to balance things out.
Why it's tempting:People believe the average must self-correct quickly, so they expect a make-up result soon (the gambler's fallacy).
Instead:Each fair flip is still 50/50 regardless of the past. The proportion approaches 50% because more new trials dilute the streak, not because future flips compensate for it.
The slip:The law of large numbers tells me what will happen on the next single trial.
Why it's tempting:It's stated about probability, so it sounds like it predicts each event.
Instead:It only describes the long-run average over many trials. It says nothing about any one outcome; a single flip is still unpredictable.
The slip:With enough trials the count of heads and tails becomes exactly equal.
Why it's tempting:'Converges to 50%' is misread as the counts matching.
Instead:The proportion approaches 50%, but the absolute difference in counts can actually grow. Convergence is about the ratio, not exact equality of totals.
Mini Practice
Try these on your own. Tap Reveal when you want to check.
1.
A fair coin is flipped 10 times and lands heads 7 times. Later it is flipped 1,000 times and lands heads 520 times. Find the proportion of heads each time. Which is closer to 0.5?
Hint:Divide heads by flips, then compare each distance from 0.5.
2.
Each ticket in a game wins $5 with probability 0.1, and nothing otherwise. About how much prize money should a player expect to win over 2,000 tickets, and what is the average per ticket?
Hint:Over many tickets the average win per ticket settles near the expected value.
3.
A fair number cube is rolled 600 times. About how many 3s should you expect? If you got 112, what is the relative frequency, and how far is it from 61?
Hint:61 is about 0.167.
4.
After 100 flips, a coin has landed heads 60 times. If the next 900 flips are exactly half heads, what is the proportion of heads after all 1,000 flips?
Hint:Add up the heads from both parts before dividing.
5.
An insurance company has 50,000 car policies. Separately, one person is deciding whether to buy a single lottery ticket. In which situation does the law of large numbers help predict the result?
Hint:The law is about what happens over very many trials.
6.
A roulette wheel has landed on red 8 times in a row. Pat says, "By the law of large numbers, black is due now." Correct Pat.
Hint:Does the wheel's chance of black change from one spin to the next?
Want the full set?
50 practice questions for this concept — free to try, every one with a complete worked
solution showing the why, not just the answer.
Does the Law of Large Numbers mean a casino can never lose money?
Over a single night a casino can lose, since a lucky player can win a large bet. What the law gives the casino is near-certainty over millions of bets, where its total take settles close to its edge times the money wagered. That is one reason casinos cap bet sizes: a few huge bets would be a small number of trials, where luck still matters.
How large does 'large' have to be?
It depends on how much each trial can vary. A fair coin's share of heads is usually within about 0.03 of 0.5 after 1,000 flips. A lottery with a huge, rare prize needs vastly more plays before the average payout settles, because a single jackpot swings the average enormously. The more extreme the rare outcomes, the more trials you need before the average can be trusted.
Does the law work for averages that are not probabilities, like heights?
Yes. It covers the average of any independent random measurements from the same population, not only proportions of successes. The average height of 5 randomly chosen adults can be far from the population mean, but the average of 5,000 randomly chosen adults will be very close to it. That is why large random samples give trustworthy estimates of a country's average height or income.
What happens if the trials are not independent?
The law can fail. Suppose a factory machine is set up either well or badly at the start of each day, and every part made that day shares the same setting. Measuring 10,000 parts from one day only tells you about that day's setting, not the machine's long-run rate. More trials help only when each one brings new, independent information.
You need Basic Probability (a probability as a long-run share) and Mean as Fair Share (what an average is) first. The law of large numbers adds the guarantee that averages of many independent trials settle near the expected value, and explains why short runs can be far off. Next, the Central Limit Theorem describes the shape of that settling, a normal curve around the mean, and Confidence Intervals use it to put a margin of error on an estimate.