Standard Deviation

Standard deviation is a measure of spread. It tells you how far results tend to sit from their average, expressed in the same units as the results themselves.

That makes it especially useful in casino math. Expected value tells you where the long-run center is. Standard deviation tells you how violent the short-run movement around that center can be. A game can have a small negative expectation per wager and still produce swings many times larger than that expectation over a session.

Standard deviation does not predict the next outcome, prove that a game is fair, or tell you how much bankroll is “safe.” It is one piece of the risk picture.

Standard deviation is the square root of variance

Variance and standard deviation describe the same underlying dispersion in different units.

For a population of values, the population variance is:

Variance = average squared distance from the population mean

The population standard deviation is:

σ = √Variance

Written more formally:

σ = √[Σ(xᵢ − μ)² / N]

where:

• σ is population standard deviation;
• xᵢ is each value;
• μ is the population mean;
• N is the number of values;
• Σ means add the values that follow.

The squaring step prevents positive and negative deviations from cancelling each other. The final square root brings the answer back to the original unit.

If outcomes are measured in dollars, variance is measured in dollars squared, which is awkward to interpret directly. Standard deviation is measured in dollars. If outcomes are measured in betting units, standard deviation is also in betting units.

That unit consistency is why standard deviation is usually the more intuitive number to discuss, even though variance is often easier to manipulate mathematically.

Sample standard deviation uses a different denominator

Casino analysts also work with samples: a week of table results, a sample of slot sessions, a set of simulated hands, or a subset of player behavior.

When a sample is used to estimate the standard deviation of a larger population, the common estimator is:

s = √[Σ(xᵢ − x̄)² / (n − 1)]

Here:

• s is the sample standard deviation;
• x̄ is the sample mean;
• n is the sample size.

The denominator uses n − 1 rather than n. This is the usual Bessel correction for estimating population variance from a sample whose mean was also estimated from the same observations.

A report should state which version it used. Mixing the population and sample formulas without saying so creates small but avoidable inconsistencies, especially with limited data.

A double-zero roulette wager shows why the number matters

Take a $10 even-money wager on red on a conventional double-zero roulette wheel.

There are 38 pockets:

• 18 red outcomes win $10 net;
• 20 outcomes lose $10, including 0 and 00.

The expected result per wager is:

E[X] = (18/38 × $10) + (20/38 × −$10)

which equals approximately:

−$0.5263 per $10 wager

That is the familiar 5.26% house edge expressed in dollars on a $10 bet.

Now look at dispersion. Every outcome has a squared value of $100, so:

E[X²] = $100

Variance can be calculated as:

Var(X) = E[X²] − E[X]²

So:

Var(X) ≈ 100 − 0.5263² ≈ 99.723

The standard deviation per wager is therefore:

σ ≈ √99.723 ≈ $9.986

Compare the two numbers:

• expected loss per wager: about $0.53;
• standard deviation per wager: about $9.99.

The short-run swing is almost twenty times the expected loss on one bet. That is why a player can easily be ahead or far behind after a modest number of spins even though the house edge has not changed.

Expected loss and standard deviation grow at different rates

Suppose the same $10 red wager is repeated 100 times under the same rules and stake.

Expected loss grows directly with the number of wagers:

100 × −$0.5263 = −$52.63 expected result

For independent wagers with the same distribution, variances add. Therefore standard deviation of the total grows with the square root of the number of trials:

σ(total) = √n × σ(single wager)

For 100 wagers:

σ(total) = √100 × $9.986 ≈ $99.86

So after 100 wagers:

• expected result: −$52.63;
• standard deviation of the total: about $99.86.

The player’s actual result can therefore be positive even though the expected result is negative. That does not contradict the house edge. It is exactly what a high standard deviation tells you to expect from short samples.

The important long-run relationship is that expected loss grows in proportion to n, while standard deviation grows in proportion to √n. Relative to total action, the noise gradually becomes smaller even though the absolute dollar swing can continue to grow.

The square-root rule needs assumptions

The convenient formula σ(total) = √n × σ is not a magic rule for every gambling sequence.

It works cleanly when the wagers are independent and share the same distribution. Several things can break or complicate that assumption:

• bet size changes from one trial to the next;
• the paytable changes;
• a progressive jackpot moves;
• wagers are correlated;
• one outcome determines whether another bet exists;
• cards are dealt without replacement from a finite shoe or deck;
• bonuses or multipliers change the outcome distribution;
• the player uses a progression that changes stake after wins or losses.

In those cases, total variance must reflect the actual structure. Covariance terms can matter when outcomes are dependent.

This is one reason serious simulations should model the game as played instead of applying a generic square-root shortcut to data that do not meet its assumptions.

One, two, and three standard deviations are not universal probability guarantees

A common shorthand says that roughly:

• 68% of observations lie within one standard deviation of the mean;
• 95% lie within two;
• 99.7% lie within three.

That is the empirical rule for a normal distribution. It is not a definition of standard deviation and should not be pasted onto every casino result.

A single roulette wager has only two possible net outcomes. A slot jackpot distribution may be extremely skewed. Video poker can have a long right tail because rare premium hands contribute a large portion of return. A side bet can have many losing outcomes and a tiny number of very large wins.

For sums of many suitably independent trials, a normal approximation can become useful because of central-limit behavior. But the quality of that approximation depends on the underlying distribution, sample size, and whether rare outcomes dominate the variance.

For small samples or heavily skewed games, exact enumeration or simulation is often more informative than simply saying “two standard deviations.” The variance simulator and long run vs short run simulator are designed for that kind of comparison.

Standard deviation and volatility are related but not identical words

In casino conversation, volatility is often used more loosely than standard deviation.

A player may call a slot “high volatility” because it produces long dry spells and occasional large hits. A game designer may use a more formal volatility measure. A financial analyst may use standard deviation of returns as a volatility measure.

Standard deviation is a defined statistical quantity once the random variable and distribution are specified. “Volatility” can be a broader descriptive label.

That is why two games can both be marketed or described as high volatility while still having different standard deviations under the same stake and measurement period.

The glossary entries for volatility and probability distribution help separate those ideas.

Standard deviation does not say whether the expectation is good or bad

A very important limitation is that standard deviation does not tell you the direction of value.

Two games can have the same standard deviation and very different expected values. One could be slightly favorable to the player under a special promotion, while another has a substantial house edge. Standard deviation only describes dispersion around each game’s own mean.

This creates four broad combinations:

A player choosing between games needs both dimensions. “Lower variance” is not automatically “better value.” “High volatility” is not automatically “bad value.”

Z-scores put a result on a common scale

A z-score expresses how many standard deviations a result is from the mean:

z = (observed result − expected result) / standard deviation

Suppose a casino expected a table to win $10,000 over a period, and the modeled standard deviation of the result was $4,000. If the table actually won $18,000:

z = ($18,000 − $10,000) / $4,000 = 2.0

The actual result is two modeled standard deviations above expectation.

That does not prove advantage play, cheating, bad dealing, or a reporting error. It tells management how unusual the result is under the model being used. The next questions are whether the model matches the actual stakes and rules, whether enough observations were included, and whether the assumed distribution is appropriate.

Standardized measures are useful because a $10,000 variance can be enormous for one table and insignificant for another depending on volume and risk.

Casino managers use standard deviation to judge whether a swing deserves attention

Actual casino win is noisy. A table can run badly while procedures are perfect. A slot bank can outperform theoretical hold for a period without any configuration change. A high-limit customer can win several trips in a row while still generating negative expected value from the player’s perspective.

Standard deviation gives management a way to ask whether the observed swing is ordinary for the exposure involved.

Useful applications include:

• comparing actual table win with theoretical expectations;
• estimating the range of plausible short-term results;
• evaluating high-limit exposure;
• stress-testing bankroll or credit assumptions;
• comparing game volatility at a common stake;
• interpreting simulation output;
• identifying results that deserve operational review.

The last point is important. Statistical unusualness can be a review trigger, not a verdict. An extreme result may justify checking fills, credits, ratings, procedures, game protection, or configuration records. Statistics should guide investigation, not replace it.

Standard deviation is not the same as a confidence interval

The terms are often mixed because both involve uncertainty.

Standard deviation describes spread in a random variable or dataset. A confidence interval is an inferential statement about an unknown parameter estimated from sample data.

For example:

• standard deviation can describe the spread of session results;
• a confidence interval can describe uncertainty around an estimated mean return based on sampled sessions.

The standard error of a sample mean is commonly related to standard deviation by:

standard error = s / √n

That is a different quantity from the standard deviation of individual observations.

The confidence interval and sample size entries cover that distinction in more detail.

Standard deviation does not produce a safe bankroll by itself

A player might ask, “If my bankroll is three standard deviations, am I safe?” The answer is no.

Bankroll risk depends on more than one static standard deviation. It can depend on starting bankroll, bet size, stopping conditions, edge, distribution shape, number of future trials, changing stakes, table rules, and whether the player can replenish funds.

A three-standard-deviation move may be unlikely over one defined period and still become quite plausible across repeated opportunities. A bankroll can also fail before the final observation period even if the ending result would have recovered.

That is why bankroll risk and the bankroll risk calculator should not be reduced to a single “number of sigmas.”

Four questions to ask whenever a casino report quotes standard deviation

A standard-deviation number is only as good as the model behind it. Before using it, ask:

1. What random variable is being measured? Per bet, per hour, per session, per table, or total property result?
2. What is the mean? Standard deviation has no useful context without the center it surrounds.
3. What assumptions were made? Fixed stakes, independence, unchanged rules, and stable paytables matter.
4. What sample or population was used? A precise-looking number from poor data can still be misleading.

If a report cannot answer those questions, “two standard deviations from expectation” may sound more rigorous than it really is.

The practical meaning in casino math

Expected value tells you the direction and average price of repeated play. Standard deviation tells you how much the path can wander around that expectation.

The two belong together.

A negative-expectation game can produce a strong winning session. A positive promotion can still produce a losing session. A casino can have a bad month without the game math changing. A player can run far above expectation without becoming more skilled.

Standard deviation is the scale that makes those swings intelligible.

Use it with expected value, variance, sample size, and the correct probability distribution. It does not tell you what happens next. It tells you how surprising a result is relative to a defined model of what could have happened.