Variance and Standard Deviation Explained

A blackjack game can have an expected loss of only a few dollars per hour and still produce a session that finishes hundreds of dollars up or down. Variance is the mathematics behind that gap.

Expected value tells you where repeated play is centered on average. Variance tells you how widely individual results spread around that center. Standard deviation, the square root of variance, converts that spread into betting units that are easier to interpret.

The practical consequence is simple: a low house edge does not imply a smooth bankroll path.

Expected value and variance answer different questions

Suppose a player flat-bets $25 at a game with a 0.50% house edge and plays 100 hands.

The expected loss is:

[ 25 \times 100 \times 0.005 = 12.50 ]

That $12.50 is the long-run mean cost implied by those assumptions. It says almost nothing about the most likely visible swing during those 100 hands.

Blackjack results come in uneven chunks:

• a normal win earns one unit;
• a normal loss costs one unit;
• a push changes nothing;
• a 3:2 blackjack wins 1.5 units;
• a double can win or lose two units;
• a split can create multiple live wagers from one starting hand.

Because those outcomes are dispersed, the session can wander far from its expected value.

For the cost calculation itself, use Blackjack Expected Loss Per Hour. This page focuses on the spread around that expectation.

Variance is measured in squared units

Let (X) represent the result of one hand in betting units and let (\mu) be the expected result of that hand. The variance is:

[ \operatorname{Var}(X) = E[(X-\mu)^2] ]

The formula says: measure how far each possible result is from the mean, square that distance, and average those squared distances using their probabilities.

The squaring is mathematically useful, but it makes the unit awkward. If your outcome is measured in $25 betting units, variance is effectively expressed in units squared.

That is why standard deviation is usually more intuitive.

Standard deviation puts the swing back into betting units

Standard deviation is:

[ \sigma = \sqrt{\operatorname{Var}(X)} ]

If one-hand variance is 1.303 betting units squared, then:

[ \sigma = \sqrt{1.303} \approx 1.141 ]

The one-hand standard deviation is therefore about 1.14 betting units.

At a $25 base wager:

[ 1.141 \times 25 \approx 28.53 ]

That does not mean a hand normally wins or loses $28.53. It is a statistical measure of spread across the full mix of pushes, ordinary wins/losses, blackjacks, doubles, and splits.

A real rule-set benchmark

The Wizard of Odds blackjack variance analysis reports simulation results for several rule sets using basic strategy. One six-deck game with dealer standing on soft 17, double after split allowed, surrender allowed, and resplitting aces allowed produced approximately:

• expected value: -0.00281 units per initial wager;
• variance: 1.303;
• standard deviation: about 1.141 units per hand.

These numbers are not universal blackjack constants. Change the rules or strategy and the expected value and variance change. They are useful because they let us build a concrete session example from a documented rule set.

Why 100 hands can dwarf the expected loss

For a player playing one hand at a time, a useful approximation for session variance is:

[ \operatorname{Var}(S_n) \approx n v ]

where:

• (n) = number of hands;
• (v) = variance per hand;
• (S_n) = total result across the session.

Session standard deviation is then:

[ \sigma_n \approx \sqrt{nv} ]

Using (v=1.303) over 100 hands:

[ \sigma_{100} = \sqrt{100 \times 1.303} \approx 11.41\text{ units} ]

At $25 per unit:

[ 11.41 \times 25 \approx 285.25 ]

Now compare that with expected loss. The cited rule set has expected value of -0.00281 units per initial bet, so over 100 hands:

[ 100 \times 0.00281 = 0.281\text{ units expected loss} ]

At $25 per unit:

[ 0.281 \times 25 \approx 7.03 ]

So the same 100-hand sample can have a theoretical loss around $7 while its one-standard-deviation scale is roughly $285.

That is why looking at one evening and saying “the house edge cannot be 0.28%; I lost $400” misunderstands the statistic. Short-run noise is much larger than the mean drift.

More hands increase expected loss faster than they increase standard deviation

Expected loss grows roughly in direct proportion to the number of hands. Standard deviation grows with the square root of the number of hands.

Using the same rule set:

Four times as many hands roughly doubles standard deviation, because (\sqrt{4}=2). But expected loss becomes four times as large.

This is one reason the long-run average becomes easier to detect relative to noise as sample size grows, even though the dollar range of possible results continues to widen.

Do not read standard deviation as a guarantee band

People often hear that observations fall within one or two standard deviations of the mean. That rule comes from the normal distribution. Blackjack session results can approach a bell-shaped distribution over sufficiently large samples, but short sessions are discrete, skew can remain, and unusual doubles/splits create non-smooth outcomes.

So “one standard deviation” is best used as a scale of ordinary variability, not a promise that a particular session must stay inside a fixed dollar corridor.

A variance simulator is more useful when you want to see the shape rather than just one summary number. Use the Variance Simulator for that purpose.

Why doubles and splits widen results

A blackjack round does not always stay at one unit of exposure.

Doubling

A correct double adds another wager equal to the original and resolves the hand after one additional card under the applicable rules. The decision can be positive compared with hitting, but the final result is now commonly +2 or -2 initial units rather than +1 or -1.

Splitting

A pair split turns one starting hand into at least two hands and requires another initial-sized wager. Depending on the rules and cards, additional splits or doubles may follow.

The strategy can be correct and the round can still produce a large loss. Correct expected value and high short-run swing are not opposites.

This is why a player who refuses mathematically correct doubles or splits solely because “too much money is out there” may reduce volatility but also worsen expected value.

Blackjack payout rules affect both value and result size

A natural blackjack is another non-even outcome. On a 3:2 table, a $25 blackjack earns $37.50 rather than $25. On a 6:5 table it earns only $30.

The payout change is primarily an expected-value problem — see 3:2 vs 6:5 Blackjack — but it also changes the distribution of hand results.

Rules should therefore not be summarized by variance alone. A lower-volatility game is not automatically a better-value game, and a higher-volatility decision can still be correct if its expected value is better.

Multiple hands at once introduce covariance

The simple (nv) session formula is easiest to interpret for one hand at a time. If you play several blackjack spots simultaneously in the same round, the hands are not fully independent because they share the same dealer outcome and cards from the same shoe.

For (m) simultaneous hands, a more complete variance expression is:

[ \operatorname{Var}(\text{round}) = m v + m(m-1)c ]

where:

• (v) = variance of one hand;
• (c) = covariance between two hands in the same round;
• (m) = number of simultaneous player hands.

The Wizard of Odds simulations cited above report covariance as well as variance for this reason.

This matters when someone assumes that playing two spots simply creates two independent copies of one-hand volatility. Shared dealer results make the relationship more complicated.

Side bets can dominate the feel of the session

Blackjack side bets commonly have payout distributions very different from the main game: frequent losses and occasional large wins. Even a small side wager can therefore add noticeable variance.

If you want to combine main-game and side-bet risk, you need the side bet’s own distribution, not just its house edge. Two wagers can have the same expected loss and radically different variance.

That is why House Edge When Side Bets Are Added should be read as a cost page, while variance is a separate risk dimension.

Three players with the same rules can experience different risk

Suppose all three players face the same blackjack rules and make correct decisions.

Player A: $10 flat bettor

The per-unit variance is the same as anyone else’s, but each unit is $10. A session standard deviation of 11.4 units is about $114.

Player B: $50 flat bettor

The same 11.4-unit standard deviation is about $570. The mathematics in units did not change; the dollar scale did.

Player C: $25 player with frequent side bets and multiple spots

This player may have more total action and a different combined variance structure even though the table placard still says $25 minimum.

“Same blackjack table” therefore does not mean “same session risk.” Bet size, number of spots, extra wagers, and bankroll all matter.

Variance is not risk of ruin

Variance describes dispersion. Risk of ruin asks a different question: given an edge or disadvantage, bankroll, bet size, and betting process, what is the probability of eventually hitting a ruin threshold?

A game can have moderate variance but still grind down a small bankroll because the edge is negative. A high-variance game can produce large temporary wins while also exposing the bankroll to rapid losses.

For bankroll survival questions, use Blackjack Risk of Ruin or the Bankroll Risk Calculator. Do not substitute “I can tolerate a 10-unit swing” for a ruin calculation.

Variance explains why good decisions can look bad

Consider four consecutive decisions:

1. Double 11 against dealer 6 — lose two units.
2. Split 8s against dealer 10 — lose both hands.
3. Stand on hard 17 — dealer makes 18.
4. Receive a natural blackjack — dealer also has blackjack, push.

A player can follow the correct strategy and feel punished on every meaningful decision. Nothing about that sequence proves the strategy is wrong. Strategy is evaluated by the expected value of each decision across the relevant card distributions, not by whether a short run cooperates.

The reverse is equally important. A player can make several poor decisions and win all of them. Winning does not turn a negative-expectation decision into a good one.

A practical unit-based way to think about session swing

Before converting everything to dollars, measure bankroll and results in units:

[ \text{Bankroll Units} = \frac{\text{Bankroll}}{\text{Base Bet}} ]

A $500 bankroll at $25 per hand is:

[ \frac{500}{25} = 20\text{ units} ]

A $500 bankroll at $50 is only 10 units.

If a 100-hand standard deviation is around 11 units for a representative rule set, the second bankroll is obviously much more exposed to an ordinary swing. This does not give an exact probability of ruin, but it makes the scale visible immediately.

More play does not force your bankroll back to the expected line

Another common mistake is to interpret the law of large numbers as a recovery mechanism. If a player is 20 units below expectation after 500 hands, the next 500 hands are not required to “make up” that gap. Future hands still have their own expected value and variance.

As the sample grows, the average result per hand tends to become more stable relative to the true expectation. That is different from saying the cumulative dollar result must return to a particular line on a schedule. A player can remain above or below expected value for a very long time while the per-hand average gradually becomes less noisy.

This distinction is especially important in a negative-expectation game. “I am far below expectation, so I should keep playing until variance corrects” turns a statistical idea into a chasing argument. Additional play adds additional expected loss as well as additional variance.

Suppose the representative rule set above has expected value of -0.00281 units per hand. After another 1,000 hands, the expected change is not a compensating win. It is:

[ 1{,}000 \times (-0.00281) = -2.81 ext{ units} ]

The player may in fact win those 1,000 hands because variance is large, but the mathematical expectation of the new block is still negative. The past deficit does not create a positive expectation in the future block.

That is the cleanest way to separate variance from “being due.” Variance explains why the path wanders. It does not promise that the next segment of the path will repair the previous one.

What variance should change about a player’s decisions

Understanding variance does not create an edge. It should improve interpretation and sizing.

• Do not treat a short winning session as proof of skill.
• Do not treat a short losing session as proof the game is unfair.
• Do not abandon correct strategy because a double or split lost.
• Do not use expected loss as though it were the maximum likely loss.
• Do not increase stakes simply because the bankroll has recently risen.
• Compare bet size with bankroll in units, not only in dollars.
• Treat side bets as separate volatility as well as separate cost.

Most importantly, decide how much loss is acceptable before a normal swing creates pressure to recover it. Variance becomes most damaging when a statistical fluctuation triggers chasing.

The relationship in one view

Blackjack can combine a small house edge with a large standard deviation over the time horizon of an ordinary casino visit. That is not a contradiction. The edge controls the slow average drift; variance controls how noisy the path can be around that drift.