The number that explains why a winning session feels like a lie
A player sits down, plays a hundred hands, and walks away up. Another plays the same table, the same rules, the same number of hands, and loses more than they expected. Neither result tells you much about the game. Both are mostly noise. Standard deviation is the tool that measures that noise, and in blackjack it is large enough to bury the house edge for hundreds of hands at a time.
The calculation itself is simple arithmetic once you have the right inputs. The harder part is reading the result honestly: a small expected loss surrounded by a wide band of possible outcomes. This article works through the calculation with hypothetical numbers, labels every input as an example, and shows what the final figure does and does not tell you.
What standard deviation measures in a card game
Standard deviation describes how spread out individual results are around an average. In blackjack, the average is the expected value of a single hand under a given rule set and playing decision. The spread is what happens hand to hand: you win one unit, lose two, push, win one and a half on a blackjack, lose four after a split and a double.
Expected value and standard deviation answer different questions. Expected value tells you the average result per hand over a very large number of hands. Standard deviation tells you how far any short stretch of hands is likely to sit from that average. A game can have a small expected loss per hand and still produce wild swings, because the standard deviation per hand is many times larger than the expected value.
Blackjack is a comparing card game where the objective is a hand total closer to 21 than the dealer's without exceeding it. Published basic-strategy charts derive from the specific rule set in use, including deck count, dealer stand rules, and doubling and splitting rules. Those rule choices change the expected value of a hand. They also change the standard deviation, because hands with more doubles and splits produce larger individual outcomes.
The inputs you need before any calculation
You cannot calculate a meaningful standard deviation without first fixing the game you are describing. The calculation needs three things: the average result per hand, the average squared result per hand, and the number of hands you plan to consider.
The first two inputs come from the rule set and your playing decisions. A single-deck game with liberal doubling rules produces a different distribution of hand outcomes than an eight-deck game with restricted splits. A player who always stands on 16 against a 10 produces different results than one who hits. The table below separates what changes the inputs from what does not.
| Input | What it depends on | What it does not depend on |
|---|---|---|
| Average result per hand | Rule set, basic strategy decisions, bet size | Session length, previous hands, bet progression |
| Average squared result per hand | Rule set, frequency of doubles and splits, bet size | Session length, previous hands, bet progression |
| Number of hands | How long you play, decisions per hour | Rule set, bet size |
Bet size matters in a specific way. If you bet one unit on every hand, the standard deviation scales directly with that unit. If you vary your bets, the calculation becomes a weighted average across the different bet levels, which is messier but still mechanical.
Step-by-step calculation with labelled example inputs
Suppose, as an example, a player bets one unit on every hand at a table where the expected value per hand is minus 0.005 units. That figure is hypothetical and chosen only to make the arithmetic visible; the actual expected value depends on the rule set and the player's decisions.
Suppose also, as an example, that the average squared result per hand is 1.3 units squared. That number comes from the fact that most hands resolve at one unit won or lost, while a smaller number resolve at larger amounts because of doubles, splits, and blackjack payouts. The exact value depends on the rule set.
The variance of a single hand is the average squared result minus the square of the average result. With the example inputs, the square of the average result is 0.000025 units squared, a number so small it barely moves the calculation. The variance is therefore approximately 1.3 units squared per hand.
The standard deviation of one hand is the square root of that variance. The square root of 1.3 is approximately 1.14. So a single hand, under these example inputs, has a standard deviation of roughly 1.14 units.
Now suppose the player plays 100 hands. The variance of the total result scales with the number of hands, so the variance over 100 hands is 100 times 1.3, or 130 units squared. The standard deviation of the total session is the square root of 130, which is approximately 11.4 units.
That is the number that matters for reading a session. The expected result over 100 hands is 100 times minus 0.005, or minus 0.5 units. The standard deviation is 11.4 units. The expected loss is a twentieth of a unit; the typical swing is more than eleven units in either direction.
Why the spread grows slower than the number of hands
Notice what happened to the two numbers as the hand count grew. The expected value grew linearly: 100 hands produced 100 times the per-hand expectation. The standard deviation grew only as the square root of the hand count: 100 hands produced 10 times the per-hand standard deviation, not 100 times.
That relationship is the core of the calculation. Over a single hand, the standard deviation is enormous compared with the expected value. Over 100 hands, it is still large. Over 10,000 hands, the expected value has grown by a factor of 10,000 while the standard deviation has grown by a factor of 100. The gap between the two keeps widening, which is why the long-run average eventually dominates the noise.
The table below shows how the two quantities diverge under the same hypothetical inputs, with all figures labelled as examples.
| Hands played | Expected result (units) | Standard deviation (units) |
|---|---|---|
| 1 | -0.005 | 1.14 |
| 100 | -0.5 | 11.4 |
| 10,000 | -50 | 114 |
| 1,000,000 | -5,000 | 1,140 |
The expected result grows ten times faster than the standard deviation as the hand count rises. At one hand, the standard deviation is more than two hundred times the expected value. At a million hands, it is less than a quarter of the expected value. That is what "the long run" means in arithmetic: the point where the average stops being drowned by the swing.
What decisions per hour do to the calculation
Hand count is the variable you control most directly, and it enters the standard deviation as a square root. Playing twice as many hands multiplies the expected loss by two but multiplies the standard deviation by roughly 1.41. Playing four times as many hands doubles the standard deviation while quadrupling the expected loss.
Decisions per hour matter because they set the hand count. A full table deals fewer hands per hour than a heads-up game. A player who deliberates over every decision sees fewer hands than one who plays basic strategy automatically. The calculation does not care why the hand count is what it is; it only uses the number.
This is also where bet variation enters. If a player raises and lowers their bet, the per-hand standard deviation is no longer a single number. The calculation becomes a weighted average of the squared bet sizes, with larger bets contributing disproportionately to the variance. A player who occasionally bets five units instead of one is not multiplying their standard deviation by five across the board; they are adding a set of hands whose individual variance is twenty-five times larger, which pulls the session number upward.
Progression systems and the redistribution of outcomes
A progression system changes the bet size after wins or losses. It does not change the expected value per unit wagered, because each hand still carries the same edge under the same rule set. What it changes is the shape of the session's distribution.
Increasing the bet after a loss, for example, means the largest bets occur on hands that are no more likely to win than any other hand. The expected value of the session is the sum of the expected values of the individual hands, each scaled by its bet. The standard deviation is the square root of the sum of the squared per-hand standard deviations, each scaled by its bet squared. A progression concentrates more of the total wager on a smaller number of hands, which increases the session standard deviation relative to a flat-betting session with the same total amount wagered.
The result is a redistribution of outcomes, not a removal of the edge. The average result stays negative in proportion to the total amount wagered. The spread around that average widens, so the session is more likely to end far from the average in either direction. A player on a progression is not buying a better average; they are buying a wider range of possible endings for the same expected loss.
Reading your own session against the number
Once you have a standard deviation for your session length and bet pattern, you can use it to read a result. A session result that sits within one standard deviation of the expected value is unremarkable. A result two standard deviations away is less common. A result three or more away is rare under the model's assumptions, though it happens often enough across many players and many sessions.
The calculation is only as good as its inputs. If the expected value per hand is wrong because the rule set is different from the one you assumed, the whole comparison shifts. If the bet pattern is not flat, the single-number standard deviation is an approximation at best. The honest use of the calculation is not to predict a session but to set expectations: to know, before you sit down, how wide the band of ordinary outcomes is, and how little a hundred hands can tell you about the game underneath them.
FAQ
Does the standard deviation formula change if I vary my bet size?
Yes. The flat-bet calculation uses one bet size for every hand. With varying bets, each hand's variance is scaled by the square of its bet, so the session variance becomes a weighted sum across bet levels. The arithmetic is the same in structure, but you cannot use a single per-hand standard deviation for the whole session.
Can I use standard deviation to predict whether I will win my next session?
No. Standard deviation describes the spread of possible outcomes around an average; it does not tell you which side of the average the next session will land on. It can tell you how ordinary or unusual a given result would be under your inputs, but it cannot forecast the result itself.
Where do I find the expected value and variance for the blackjack game I play?
The expected value per hand depends on the specific rule set: number of decks, dealer stand rules, doubling and splitting permissions, and whether blackjack pays at a particular ratio. Published basic-strategy references list these figures for common rule combinations, and the game's information panel or the operator's rules page states the rules you need to match against them.
If you or someone you know has a gambling problem, help is available. In the US, call or text the National Problem Gambling Helpline at the helpline printed in the footer below. In Great Britain, contact the National Gambling Helpline at 0808 8020 133. Please gamble responsibly and only with money you can afford to lose.
Play responsibly
Gambling should be treated as paid entertainment, never as a way to earn income or recover losses. 1-800-MY-RESET (1-800-697-3738)
18+ or 21+ depending on where you are; follow the minimum age that applies to you.
This article is general information about how these mechanics work. It is not legal advice and not a recommendation to gamble or to use any particular operator. Availability and legality differ by jurisdiction — check the rules that apply where you are.