A probability distribution describes all possible values of a random variable and the probability attached to each value. In casino math, the random variable might be the net result of one bet, the number of wins in 100 rounds, a slot payout, a table’s daily hold, or the waiting time until a particular event.
The distribution is the full mathematical picture. Averages such as RTP or expected value are summaries of that picture.
A simple discrete distribution
Consider a $1 even-money roulette-style bet with two simplified outcomes:
| Net result (x) | Probability (P(X=x)) |
|---|---|
| +$1 | 18/37 |
| -$1 | 19/37 |
The probabilities add to one:
[ \frac{18}{37}+\frac{19}{37}=1 ]
The expected value is:
[ E(X)=\sum xP(X=x) ]
So:
[ E(X)=1\left(\frac{18}{37}\right)+(-1)\left(\frac{19}{37}\right)=-\frac{1}{37}\approx-$0.0270 ]
The average loss is about 2.70 cents per $1 wager over the long run. A single result can only be +$1 or -$1; the expected value is not a possible one-round outcome. It is the probability-weighted average.
Distribution is more informative than the average
Two games can have the same expected return and very different distributions.
Imagine two $1 games, each with 95% RTP:
- Game A returns small amounts frequently and rarely pays a large prize.
- Game B returns nothing most of the time but allocates more return to rare large awards.
Both can average 95 cents returned per dollar over their design cycle, yet Game B will usually have greater variance and a heavier right tail. Short sessions will feel very different.
The NIST/SEMATECH statistics handbook describes probability distributions as a fundamental theoretical and practical concept in statistics.
Discrete and continuous distributions
Casino outcomes can be modeled with different types of distribution:
- Discrete distribution: countable outcomes, such as dice totals, roulette pockets, number of wins, or payout categories.
- Continuous distribution: values across a range, often used as an approximation for revenue, measurement error, or aggregated results.
- Mixed distribution: a point mass at a specific value plus a continuous range, useful in some risk models.
A slot paytable produces a discrete payout distribution. A casino may approximate aggregated daily results with a continuous model for planning, but the approximation should be tested rather than assumed.
Common distributions in casino analysis
| Distribution | Possible casino use | Important limitation |
|---|---|---|
| Bernoulli | one trial with success/failure | only two outcomes |
| Binomial | number of successes in fixed independent trials | requires stable probability and independence |
| Geometric | trials until first success | assumes independent trials with constant probability |
| Multinomial | counts across several outcome categories | fixed category probabilities |
| Normal | approximation for aggregated results | may fit poorly with skewed or heavy-tailed data |
| Poisson | event counts in a period | rate stability and independence may fail |
The name of a distribution should not be selected merely because it is familiar. The model assumptions must match the process.
Variance comes from the same distribution
For a discrete random variable:
[ \operatorname{Var}(X)=\sum (x-E(X))^2P(X=x) ]
Variance measures how far outcomes spread around the expected value. Standard deviation is the square root of variance:
[ \sigma=\sqrt{\operatorname{Var}(X)} ]
A game with rare large payouts can have a much higher standard deviation than a game with many small payouts even when expected value is similar.
This is why house edge alone does not describe bankroll risk, session experience, or the amount of data needed before actual results begin to resemble expectation.
Probability mass is not prediction
A distribution says what outcomes are possible and how likely they are under the model. It does not reveal the next outcome.
If a fair die has distribution:
[ P(X=x)=\frac{1}{6},\quad x\in{1,2,3,4,5,6} ]
rolling five sixes in a row does not remove six from the next-roll distribution. The next roll remains governed by the same probabilities if the die and trials are independent.
Historical data can be used to test whether the assumed distribution remains plausible. It should not be used to invent a short-term “due” outcome.
Tails explain rare but important events
The tails of a distribution contain unusually low or high outcomes. In casino operations, tail events include:
- a large progressive jackpot;
- an extreme table win or loss day;
- an unusually long losing sequence;
- a rare cluster of disputes or failures;
- a high-value credit loss.
Rare does not mean impossible. Planning should consider both probability and consequence.
A manager who budgets only around the mean can be surprised by normal tail variation. A player who treats a rare jackpot as a likely personal outcome can overestimate opportunity.
Conditional distributions
The distribution can change when information changes. For example, the distribution of blackjack outcomes after the first two cards is different from the distribution before the deal. The relevant expression is a conditional probability:
[ P(X=x\mid I) ]
where (I) is known information.
Using valid information is not the same as claiming that recent unrelated results alter the next independent outcome.
Empirical and theoretical distributions
A theoretical distribution comes from game rules and mathematics. An empirical distribution comes from observed data.
Observed frequencies can differ from theoretical probabilities because of sample size, randomness, changing game mix, data errors, or an incorrect model. The correct response is to assess uncertainty and data quality.
For observed counts (n_i) in categories, the empirical probability is:
[ \hat p_i=\frac{n_i}{n} ]
If 2,000 recorded rounds include 990 outcomes in category A, the empirical share is 49.5%. That does not prove the true probability is exactly 49.5%.
Simulation uses a distribution; it does not replace one
A simulation draws repeated outcomes from specified rules or probabilities. It can estimate distributions that are difficult to calculate directly, explore bankroll paths, or show the range of short-term results.
A simulation is only as trustworthy as its game logic, random-number process, assumptions, number of trials, and validation. Repeating a wrong model a million times produces a precise answer to the wrong question.
Player-facing misunderstandings
Common errors include:
- treating RTP as the most likely session return;
- assuming the average payout interval is a countdown;
- believing a tail event is due after a long absence;
- comparing two games only by house edge;
- ignoring net loss when a game produces frequent small returns;
- using a tiny personal sample to reject published mathematics.
The distribution explains why a mathematically stable game can feel chaotic over a short session.
Operational uses
Casinos and suppliers use distributions for:
- game and jackpot risk;
- bankroll and liquidity planning;
- hold and revenue ranges;
- staffing and service-event forecasts;
- alert thresholds;
- simulation and stress testing;
- investigation of unusual results;
- comparison of game versions;
- evaluating whether observed data fit expectation.
The model should be documented with its population, period, game rules, assumptions, and treatment of outliers.
The clean definition
A probability distribution is the complete assignment of probabilities to possible outcomes. Expected value identifies the center. Variance and standard deviation describe spread. Tails describe rare extremes. Conditional distributions incorporate valid information.
For casino readers, the central lesson is that the average is not the next result and may not even be a common result. To understand how a game or operation behaves, you need the shape of outcomes, not only one headline percentage.