Why long-run results follow the maths
In the short run anything can happen. In the long run, the average result per bet closes in on the expected value. This page shows how fast that happens, with exact numbers.
The law of large numbers
The law of large numbers says that the average of many independent trials gets closer and closer to the expected value as the number of trials grows. For betting, it means your average result per unit staked tends towards the edge built into the bets you make, however lucky or unlucky the early results were.
It does not say the total result gets close to the expected total. It says the average does. That difference is the source of a lot of confusion, so it is worth seeing in numbers.
The numbers for an even-money bet
Take a bet on red on a single-zero wheel: it wins 18 times in 37 on average and pays even money, so the expected result is −1/37 of the stake, about 2.70%. The table shows what happens as the number of 1-unit bets grows. The chance of being ahead is calculated exactly from the binomial distribution.
| Bets of 1 unit | Expected result | Standard deviation | SD as % of total staked | Chance of being ahead |
|---|---|---|---|---|
| 10 | -0.3 | 3.2 | 31.61% | 34.42% |
| 100 | -2.7 | 10.0 | 10.00% | 35.53% |
| 1,000 | -27.0 | 31.6 | 3.16% | 18.76% |
| 10,000 | -270.3 | 100.0 | 1.00% | 0.33% |
| 100,000 | -2,702.7 | 316.1 | 0.32% | < 0.01% |
Reading the table
- The expected loss grows in direct proportion to the number of bets. Ten times as many bets, ten times the expected loss.
- The spread grows more slowly, with the square root of the number of bets. Ten times as many bets gives only about 3.16 times the standard deviation.
- So the spread shrinks relative to the amount staked. The fourth column falls steadily, while the edge stays at 2.70% of every unit.
- The chance of being ahead falls towards zero. After 10 bets it is 34.4%, and a further 24.5% of sessions are exactly level, because 5 wins and 5 losses is a common result. After 1,000 bets it is 18.8%; after 10,000 it is 0.33%; after 100,000 it is too small to show.
This is the whole business model behind any game with a built-in edge. A single bet is uncertain for both sides. A very large number of bets is nearly certain for the side that holds the edge.
"Evening out" is dilution, not correction
Suppose red comes up 10 more times than expected in the first 100 spins. The law of large numbers does not make black more likely afterwards to cancel that surplus (that idea is the gambler's fallacy). The surplus of 10 simply stays there while the total number of spins grows. After 100 spins it is 10% of the total; after 10,000 spins it is 0.10%; after a million it is negligible. The proportion settles, while the absolute gap is free to wander.
What this means in practice
- Short-run results say very little about the long-run average. A good week proves nothing, and nor does a bad one.
- The more you bet in total, the closer your result per unit gets to the edge. With a negative edge, playing longer makes a net loss more certain, not less.
- Staking plans cannot get around this, because the expected loss depends on the total staked (see betting systems).
Understanding this is not about winning more. It is about seeing gambling clearly: as a paid form of entertainment where the price is the edge times the amount staked. The responsible gambling page covers how to keep that cost within limits.