Variance: why the same average can feel so different

Expected value tells you where results are centred. Variance tells you how far they wander from that centre, and it is the reason short-run results look nothing like the maths.

Two bets, one average

On a single-zero wheel, a bet on red and a bet on a single number both have an expected value of −1/37, about 2.70% of the stake (see expected value). Yet a red bet wins almost half the time and pays little, while a single number wins rarely and pays a lot. Variance is the tool that captures this difference.

The formulas for one bet

For a bet that wins with probability p at decimal odds d, with a stake of 1:

mean = p × d − 1   |   variance = p × (1 − p) × d²   |   standard deviation = d × √(p(1 − p))

Standard deviation (SD) is the square root of variance. It is in the same units as the stake, which makes it easier to read.

Worked example

Red: p = 18/37, d = 2. SD = 2 × √(18/37 × 19/37) = 1.000 units.

Single number: p = 1/37, d = 36. SD = 36 × √(1/37 × 36/37) = 5.838 units.

Same mean, but the single-number bet is about 5.8 times as volatile.

Many bets: the square-root rule

For independent bets the means add up and so do the variances. That means the SD grows with the square root of the number of bets, while the expected result grows in direct proportion.

after n bets: mean = n × single mean   |   SD = √n × single SD
100 bets of 1 unit on a single-zero wheel
BetWins onExpected result after 100 betsStandard deviationChance of being ahead after 100
Red / black (pays 1 to 1)18 of 37-2.7010.035.5%
Single number (pays 35 to 1)1 of 37-2.7058.450.9%

The chance of being ahead is calculated exactly from the binomial distribution, not estimated. For red you need at least 51 wins out of 100. For a single number you need at least 3 wins, because each win returns 36 units and 2 wins return only 72, which is less than the 100 units staked.

Notice what this shows. Both bets lose the same 2.70 units on average. Yet the single-number player is ahead more often, because a few lucky hits produce a big win, while a 6.5% chance of not hitting at all over 100 bets is the price of that volatility.

Why high variance misleads people

  • Memorable wins. High-variance bets produce occasional large wins that are easy to remember and talk about. The many small losses that paid for them are forgettable.
  • Short samples prove nothing. With an SD of 58.4 units over 100 bets, a result of +50 or −50 is ordinary noise. You cannot judge a strategy, or anyone's skill, from a short run of high-variance bets.
  • Bankroll swings. The higher the SD, the more money you need to absorb losing runs without running out. That link is the subject of the bankroll maths guide.

A rule of thumb

For a large number of bets, results tend to fall within about two standard deviations of the mean roughly 95% of the time, because the total is close to a normal distribution. That approximation is good for red after a few hundred bets but poor for a 35-to-1 bet, whose results stay lumpy for much longer. When in doubt, compute the exact binomial probability as the table above does.