Standard deviation measures the typical size of deviations from an average. In gambling, it answers a question that expected value cannot answer by itself: how widely can actual results swing while the underlying game remains unchanged?
A game may have an expected loss of only a few cents per wager and still produce wins or losses many times larger over a short session. The expected value locates the center. Standard deviation describes the spread around that center.
A number expressed in useful units
Variance and standard deviation describe the same underlying dispersion, but standard deviation is easier to interpret because it returns to the original unit.
If session results are measured in dollars, variance is measured in dollars squared. Standard deviation is measured in dollars. If results are recorded in betting units, standard deviation is also in betting units.
That distinction is why analysts often calculate variance first but discuss standard deviation when explaining the result.
For a complete population of outcomes, the formula is:
[ \sigma = \sqrt{\frac{\sum_{i=1}^{N}(x_i-\mu)^2}{N}} ]
Where:
- (\sigma) is the population standard deviation;
- (x_i) is each observed value;
- (\mu) is the population mean;
- (N) is the number of values;
- (\sum) means add the squared deviations for every value.
The calculation subtracts the mean from each result, squares those differences so positive and negative deviations do not cancel, averages them, and takes the square root.
When the data are a sample used to estimate a larger population, the common sample formula is:
[ s = \sqrt{\frac{\sum_{i=1}^{n}(x_i-\bar{x})^2}{n-1}} ]
Here, (s) is the sample standard deviation, (\bar{x}) is the sample mean, and (n) is the sample size. The denominator (n-1) corrects for the fact that the population mean is being estimated from the same sample. A report should say which version it used.
A roulette calculation
Consider a $10 wager on red in double-zero roulette. There are 18 red pockets and 20 losing pockets, including 0 and 00. Ignoring any special rule, the net result per wager is:
- +$10 with probability (18/38);
- −$10 with probability (20/38).
The expected result per wager is:
[ \mu = \left(\frac{18}{38}\times 10\right)+\left(\frac{20}{38}\times -10\right) = -$0.5263 ]
Because every result is either +$10 or −$10, (E[X^2]=100). The variance is:
[ \operatorname{Var}(X)=E[X^2]-\mu^2=100-(-0.5263)^2\approx 99.723 ]
The standard deviation per wager is therefore:
[ \sigma=\sqrt{99.723}\approx $9.986 ]
That number is almost twenty times the expected loss per wager. This is why a short roulette session is dominated by swing rather than by the smooth average implied by the house edge.
For 100 independent $10 wagers under the same conditions:
[ E[T]=100\times(-$0.5263)=-$52.63 ]
[ \sigma_T=\sqrt{100}\times $9.986=$99.86 ]
The expected total is a $52.63 loss, but the standard deviation of the total is about $99.86. Under a rough normal approximation, a one-standard-deviation range is approximately −$152.49 to +$47.23. That range is not a promise and roulette results are discrete, but it shows why finishing ahead after 100 bets does not overturn the mathematics.
Why the square root of the number of bets appears
For independent wagers with the same distribution, variances add. Standard deviations do not add directly.
If one wager has standard deviation (\sigma), then (n) independent wagers have:
[ \sigma_{\text{total}}=\sqrt{n},\sigma ]
Expected loss grows in direct proportion to (n), while standard deviation grows in proportion to (\sqrt{n}). The average result per wager therefore becomes more stable as the sample grows, even though the absolute dollar swing can still become larger.
This relationship depends on important assumptions. Bet size must be fixed, the rules must stay the same, and outcomes must be independent or modeled with their actual dependence. Changing stakes after wins or losses creates a different distribution. So do bonuses, side bets, progressive jackpots, and correlated outcomes.
The NIST measures-of-scale reference defines standard deviation as the square root of variance and notes that it is expressed in the same units as the data. That unit consistency is what makes the measure practical in casino reports.
What one, two, or three standard deviations mean
People often hear the shorthand that about 68% of observations fall within one standard deviation of the mean, 95% within two, and 99.7% within three. That rule applies approximately to a normal distribution. It should not be copied automatically onto every gambling result.
A single wager with two possible outcomes is not normally distributed. A jackpot game can be heavily skewed. A small sample may be lumpy. The normal approximation generally becomes more useful for totals of many suitably independent trials, but even then analysts should inspect the actual probability distribution.
For a non-normal or highly skewed game, simulation or exact outcome enumeration may give a more honest range than a simple “mean plus or minus two standard deviations.” The Long Run vs Short Run Simulator is useful for seeing how those distributions evolve over repeated play.
Casino uses and common errors
A casino can use standard deviation to estimate table-game exposure, evaluate whether actual win is unusually far from theoretical win, set bankroll or credit limits, compare machine volatility, and design risk reports. A player can use it to understand why two games with similar RTP can feel completely different.
The number is frequently misused in four ways:
- Treating a large deviation as proof of manipulation. Unusual does not mean impossible, and the probability must be calculated under the correct model.
- Ignoring sample size. A deviation of $20 may be large for one low-stakes wager and trivial across a million wagers.
- Mixing games or bet sizes. One standard deviation cannot represent a dataset whose stakes, rules, and exposures were combined carelessly.
- Assuming standard deviation protects a bankroll. It measures risk; it does not remove it.
Standard deviation also differs from a confidence interval. Standard deviation describes the spread of individual or total outcomes. A confidence interval estimates an unknown parameter, such as a mean, using sample data and a stated confidence level.
The practical reading
When someone says a result was “two standard deviations above expectation,” ask four questions before accepting the conclusion:
- What was the expected value?
- How was the standard deviation calculated?
- How many trials were included?
- Did the model match the actual rules, bets, and dependencies?
Without those answers, the phrase sounds precise but may not be informative.
Standard deviation does not predict the next hand or spin. It gives scale to uncertainty. Used with expected value, sample size, and the correct distribution, it helps separate an ordinary swing from a result that genuinely deserves investigation.