A range instead of a prediction
Saying that blackjack has a small house edge tells you the average result per hand over a long time. It does not tell you what will happen tonight. A confidence interval is a way to describe that uncertainty honestly: instead of one predicted number, it gives a range in which the result would fall a stated share of the time, under stated assumptions.
The usual choice is a 95 percent interval. That means that if the same session were played many times under the same conditions, about 95 percent of the outcomes would land inside the range and about 5 percent outside it.
The ingredients
Two quantities build the range for a session of identical-sized bets:
- the expected result, which is the house edge multiplied by the number of hands and the bet size
- the spread, which comes from the standard deviation per hand and grows with the square root of the number of hands
The standard deviation per hand in blackjack is usually quoted as a little over one unit of the bet. For illustration, this article uses 1.1 units. Real values depend on the rules and on how often the player doubles and splits.
A worked illustration
Suppose a player plays 10,000 hands at one unit per hand. Assume, purely for illustration, a house edge of 0.5 percent against that player.
- Expected result: 10,000 x 0.005 = 50 units lost, so about -50.
- Spread of the total: 1.1 x the square root of 10,000, which is 1.1 x 100 = 110 units.
- A 95 percent interval is roughly the expected result plus or minus 1.96 times the spread: 1.96 x 110 = 215.6.
That gives a range from about -266 to about +166 units. The most striking feature is that the range is far wider than the expected loss. Even with a house edge working against the player, finishing ahead after 10,000 hands is well within the plausible range. The edge is real, but it is small compared with the noise in any single session.
How the range changes with the number of hands
Because the spread grows with the square root of the hands while the expected loss grows in direct proportion, the two behave differently as sessions lengthen.
- At 100 hands the spread is 11 units and the expected loss is 0.5 unit. The range is about -22 to +21, so the edge is almost invisible.
- At 10,000 hands, as above, the edge is about a quarter of the half-width of the interval.
- At 1,000,000 hands the spread is 1.1 x 1,000 = 1,100 units and the expected loss is 5,000 units. The half-width of the 95 percent range is about 2,156, so the whole range sits below zero, from about -7,156 to about -2,844.
This is the law of large numbers in numeric form: the edge only dominates over very long runs, and most players never get near that many hands.
What a confidence interval does not say
Intervals are easy to over-read. A few cautions:
- It is not a guarantee. About one session in twenty lands outside a 95 percent range.
- It rests on the assumed edge and standard deviation. If those are off, the range is off.
- It assumes bets of equal size and independent hands. Changing bet sizes, or tying them to earlier results, changes the maths.
- A normal-curve approximation is used. It works well for many hands but is rougher for short sessions, where results are lumpy.
- Real results are not symmetrical in detail: a natural pays 3 to 2, doubles and splits create larger swings, and these features are all folded into the single figure for the standard deviation.
Using this sensibly
The useful lesson is not a number to aim for. It is that a single result, good or bad, carries little information about the underlying edge. A winning night does not show that a method works, and a losing one does not show that something is rigged. If you want to know the cost of playing, the expected loss is the figure to keep in mind, and the interval reminds you how far reality can wander from it.
Blackjack carries a house edge under standard rules, and outcomes are uncertain. Set a session budget before you play, remember that chasing losses cannot improve the odds, and seek local support services if gambling stops being a form of entertainment.