KEY TAKEAWAYS
CONTENTS
The law of large numbers says that the average result of many independent bets moves toward the expected value as the number of bets grows. It does not say that totals even out, that losses are “made up,” or that a win is due. On a European roulette wheel, an even-money bet has an expected result of −1/37, or about −2.70% per bet, and the average of many bets settles near that figure.
This guide explains what the law promises and what it does not. It works through the formulas for how fast an average settles, uses coin flips and roulette as worked examples, shows a table of how the odds of finishing ahead shrink as bets accumulate, and covers the most common misreadings. Assumptions are stated for every number, and no betting pattern changes the expected value.
What the Law of Large Numbers Says
The average converges, the total does not. If a bet has an expected result of −2.70 cents per dollar, then after many bets the average result per bet gets closer to −2.70 cents. The gap between average and expectation shrinks as the number of bets n grows. The total, however, is the average multiplied by n, so the gap in dollars can still grow, only more slowly than n itself.
Think of two different questions. “What is your average result per bet?” becomes more predictable with more bets. “What is your total profit or loss?” becomes more spread out in dollars, although it drifts steadily toward n times the expected value. Both statements are true at once, which is why a long session can feel random while the casino’s results look steady.
Each bet still has a fixed probability. The law describes what happens to averages across many independent trials, not what the next trial will do. This is the key difference from the gambler’s fallacy, which wrongly claims that past results change the next probability.
Independence matters. The law applies cleanly to repeated bets whose probabilities do not change, such as a roulette spin or a coin flip. Bets that depend on one another, or games whose odds shift as cards are removed, need more careful treatment, so every example here assumes identical, independent bets.
The Formulas: How Fast an Average Settles
Two formulas describe the speed of convergence. The standard deviation (SD) of one bet measures how widely a single result varies, as explained in variance in gambling. For n independent bets, the SD of the total grows with the square root of n, while the SD of the average shrinks with it.
where:
SD of one bet = √(E[X²] − (E[X])²), about 0.9996 units for an even-money roulette bet
n = number of independent bets
√n = square root of n
The edge becomes visible when √n times the edge reaches the SD. Setting the expected loss n × edge equal to one SD of the total, SD × √n, gives the following rule of thumb.
where:
edge = house edge per bet, 1/37 ≈ 0.02703 on a European single-zero wheel
SD of one bet ≈ 1 unit for an even-money bet
For this roulette bet, n ≈ (1 ÷ 0.02703)² ≈ 1,369, so after roughly 1,370 bets the expected loss is about one SD of the total. It is a guide to scale, not a threshold. The formulas rely on a normal approximation for ranges, which is less reliable for few bets or lopsided payouts, and they assume every bet is independent.
Worked Examples: Coin Flips and Roulette
Example 1: a fair coin. Suppose a fair coin lands heads with probability 0.5, and you flip it 100 times and see 55 heads.
Step 1: Share of heads = 55 ÷ 100 = 55%, which is 5 points above 50%.
Step 2: SD of the heads count = √(100 × 0.5 × 0.5) = 5, so the surplus is 1 SD.
Step 3: With 10,000 flips, SD = √(10,000 × 0.25) = 50 heads.
Step 4: A surplus of 50 heads is 5,050 ÷ 10,000 = 50.5%, only 0.5 points above 50%.
The surplus in heads grew from 5 to 50 while the share moved from 55% to 50.5%. More flips did not cancel the early surplus; they made it a smaller fraction of the total.
Example 2: even-money roulette. Suppose you bet $10 on red on a European wheel, with a win probability of 18/37 and an expected result of −1/37 per $1.
Step 1: Expected result per bet = $10 × (−1/37) = −$0.27.
Step 2: Over 100 bets, expected total = −$27.03 and SD = $10 × 10 = $100.
Step 3: Over 1,000,000 bets, expected total = −$270,270 and SD ≈ $10 × 1,000 = $10,000.
Step 4: The expected total is 270,270 ÷ 9,996 ≈ 27 SDs from zero.
For a single player the swing dwarfs the average cost over 100 bets, since the SD is about 3.7 times the expected loss. Across a million bets, the expected loss is 27 SDs, so a net gain is effectively out of reach.
How the Odds of Finishing Ahead Change With More Bets
The table applies the same method to $1 bets on red on a European single-zero wheel. The chance of finishing ahead, meaning a net gain greater than zero, comes from the exact binomial distribution rather than the normal approximation. Results are rounded.
| Bets (n) | Expected total (units) | SD of total (units) | SD of average per bet | Chance of net gain |
|---|---|---|---|---|
| 10 | −0.27 | 3.16 | 31.61% | 34.4% |
| 100 | −2.70 | 10.00 | 10.00% | 35.5% |
| 1,000 | −27.03 | 31.61 | 3.16% | 18.8% |
| 10,000 | −270.27 | 99.96 | 1.00% | 0.33% |
| 100,000 | −2,702.70 | 316.11 | 0.32% | effectively 0 |
| 1,000,000 | −27,027.03 | 999.63 | 0.10% | effectively 0 |
The average per bet tightens around −2.70% while the dollar swing widens. The chance of a net gain does not fall steadily at first, because small samples are dominated by chance, and exact break-even results (24.5% of the time at 10 bets, 7.7% at 100) take a large share. After thousands of bets, the edge dominates.
This is why the law of large numbers helps casinos and not individual players. The house takes a huge number of bets, so its results sit close to the edge, as in the roulette house edge calculation. A player’s experience over a few hundred bets can look very different, which is also why the expected loss per hour figure is an average, not a promise.
Common Mistakes With the Law of Large Numbers
Believing results must even out is the first mistake. After a run of reds, a black is no more likely on the next spin. The law describes dilution of the early run over many spins, not a correction.
Confusing totals with averages is the second. The share of heads approaches 50%, but the gap in heads usually grows like √n. Treating a few hundred bets as the “long run” is the third: the table shows that at 100 bets a net gain still occurs about 35.5% of the time.
Thinking the law helps the player is the fourth. It guarantees only that the average result approaches the expected value, which is negative for every standard casino bet. Assuming a betting system changes the outcome is the fifth, because staking plans reshape the pattern of results without changing the expected value. Gambling should be treated as paid entertainment.
Quick Reference and Check Your Answer
Use the table below to estimate how settled results should look after n independent bets of one unit.
| Quantity | Formula | Example (n = 10,000, even money, European roulette) |
|---|---|---|
| Expected total | −n × edge | −270.27 units |
| SD of total | SD × √n | 99.96 units |
| SD of average | SD ÷ √n | 0.9996 ÷ 100 ≈ 0.01, or 1% |
| Expected loss in SDs | edge × √n ÷ SD | 0.02703 × 100 ÷ 0.9996 ≈ 2.70 |
To check your answer, confirm that the expected total equals n times the expected result per bet and that the SD of the total equals n times the SD of the average. If the two do not match, a √n was misplaced.
Related Concepts and Next Steps
To build the foundation for these formulas, read variance in gambling, which defines variance and standard deviation. The gambler’s fallacy guide explains why independent spins have no memory. For the average-result side of the idea, the expected value formula guide shows how to calculate the number the average converges to, and expected loss per hour turns it into a cost per hour. The roulette house edge guide derives the edge used in every table above.
Frequently Asked Questions
