The long run is not a place a player reaches after 100 spins, 1,000 hands, or one bad month. It has no universal finish line.
It is a convergence idea: as the number of comparable trials increases, the average result tends to become more stable around the mathematical expectation. NIST summarizes the simple law-of-large-numbers idea by saying that, under regularity conditions, a sample average should converge toward the true mean as sample size becomes very large. See its Combining Information presentation.
“Tends to” is doing important work. It does not mean every long sample lands exactly on expectation.
The long run has three clocks
Clock 1: Number of decisions
Casino math usually counts wagers, hands, spins, rolls, or games—not calendar days.
A slot player making 600 spins in one hour has more trials than a baccarat player seeing 60 hands in the same hour. Ten years of occasional play may contain fewer decisions than one month of high-speed automated wagering.
Clock 2: Amount wagered
House edge applies to action. Two players can make the same number of bets but generate very different expected losses because their stakes differ.
Expected Loss = Total Amount Wagered × House Edge
A player making 10,000 bets of $1 at a 2% house edge has $10,000 of action and $200 expected loss. A player making 10,000 bets of $100 has $1,000,000 of action and $20,000 expected loss.
Clock 3: Variability of the game
High-variance outcomes can remain far from expectation for longer than low-variance outcomes. A game with rare large prizes needs more trials before its observed average usually looks stable.
That is why “How many plays is the long run?” has no single answer. The answer depends on the distribution, the precision required, and what probability of error is acceptable.
A roulette example: average versus total result
On a standard double-zero wheel, an even-money red bet wins on 18 numbers and loses on 20.
For a $10 wager:
EV per bet = (18/38 × $10) + (20/38 × -$10)
= -$0.5263
The house edge is about 5.26%.
After 1,000 identical $10 bets:
Total Action = 1,000 × $10 = $10,000
Expected Loss = $10,000 × 0.05263 ≈ $526.30
The actual result might be a $300 loss, a $1,100 loss, or even a profit. The long-run statement is not that the player must lose exactly $526.30 after bet 1,000. It is that repeated averages are pulled toward the underlying expected loss as comparable trials accumulate.
Percentage noise shrinks while dollar swings can grow
This is one of the most misunderstood parts of the long run.
As the number of independent identical trials grows:
- the average result per bet usually becomes more stable;
- the percentage difference from expectation usually narrows;
- the possible dollar distance from the expected total can still become larger because more money is being wagered.
For many repeated-trial models, standard deviation of the total grows approximately with the square root of the number of trials:
Total Standard Deviation = Standard Deviation per Trial × √n
If one trial has a $10 standard deviation:
- at 100 trials, total standard deviation is about $100;
- at 10,000 trials, it is about $1,000.
The second sample has a larger typical dollar swing, but the swing is smaller relative to 100 times as much action.
Read Standard Deviation and Variance for the spread around the average.
The long run does not repair the past
Suppose roulette produces red eight times in a row. The next spin is not required to be black to move the wheel toward balance.
The law of large numbers works through future accumulation, not active correction. The first eight reds remain in the record. Their share can become less important as thousands of later spins are added.
Example:
8 reds out of 8 spins = 100% red
458 reds out of 1,000 spins = 45.8% red
4,745 reds out of 10,000 spins = 47.45% red
The early streak was not reversed. It was diluted within a much larger sample.
That distinction defeats the claim that a losing result is “due” to be repaid. See Gambler's Fallacy.
There is no casino-wide long run
Results should be grouped only when the underlying conditions are meaningfully comparable.
Mixing these can create a false average:
- single-zero and double-zero roulette;
- blackjack with different rules or player strategies;
- standard and no-commission baccarat;
- slot games with different paytables or denominations;
- base bets and high-edge side bets;
- promotional periods and normal play;
- manual and automated game speeds.
A large mixed sample can be less informative than a smaller well-defined sample.
“Close enough” must be defined before the test
Analysts often care about a tolerance rather than exact equality.
Suppose a casino wants to know when a machine's observed return is within one percentage point of its 96% theoretical RTP. The question is not merely sample size. It also needs:
- the game's variance;
- independence or dependence structure;
- confidence level;
- stable configuration;
- accurate coin-in and payout data;
- a pre-defined tolerance.
The long run is therefore a statistical planning problem, not a slogan.
A practical convergence table
Assume an observed average loss per $1 wagered is being compared with a true expectation of 2 cents.
| Sample | Observed average loss | Difference from expectation |
|---|---|---|
| 100 bets | 9.0 cents | 7.0 cents high |
| 1,000 bets | 0.5 cents | 1.5 cents low |
| 10,000 bets | 2.4 cents | 0.4 cents high |
| 100,000 bets | 2.1 cents | 0.1 cents high |
This illustration shows increasing stability. It is not a guaranteed path. The 10,000-bet average could be farther away than the 1,000-bet average in a particular sequence.
Convergence describes overall behavior across repeated samples, not monotonic improvement after every new observation.
Player and casino views are mirror images
For the casino, the long run supports forecasting, staffing, capital planning, hold analysis, and product evaluation. The casino can spread risk across many players, games, tables, machines, and days.
For one player, the sample is smaller and concentrated. A person can stop while ahead, lose far more than expectation, or never play enough for the observed average to resemble the theoretical figure closely.
The casino's advantage is not only the House Edge. It is also scale, diversification, limits, time, and the ability to survive variance.
Five claims the long run does not support
- “The game must balance soon.” No fixed short-term balancing schedule exists.
- “A large loss guarantees a recovery.” Expected value does not reimburse individuals.
- “A winning sample proves an edge.” Favorable variance can persist.
- “RTP is a promise for my session.” RTP is a theoretical long-run average under defined conditions.
- “More play makes a negative-edge bet safer.” More play usually makes cumulative expected loss larger.
The useful definition
The long run is best understood as a direction: more comparable evidence, more stable averages, and greater visibility of the underlying expectation.
It is not a deadline. It is not a repayment mechanism. It does not remove variance. It makes the price of repeated play harder to hide.
Continue with Expected Value, Sample Size, Short-Term Variance, RTP, and House Edge Explained.