Craps probability guide · Reviewed August 18, 2026
Variance explains why a short craps session can finish far above or below its mathematical expectation. It creates uncertainty—not a signal that the next roll has become predictable.
Variance changes results, not the next-roll probabilities
A player can benefit from favorable variance after it happens, but cannot identify a “hot distribution” that changes the next fair-dice probability. Shorter play limits cumulative exposure; it does not turn a negative-expectation game positive.
Variance in plain language
An expected value is a long-run average. Variance measures how widely individual results spread around that average. Standard deviation expresses that spread in the same units as the result being measured.
- Expectation answers: where is the long-run average?
- Variance answers: how dispersed are the possible results?
- Standard deviation makes that dispersion easier to interpret in the original units.
- Independence means the previous fair roll does not alter the probabilities of the next fair roll.
A fair-coin example
For 30 fair coin flips, the expected number of heads is 30 × 1/2 = 15. The standard deviation is √(30 × 1/2 × 1/2), or about 2.74 heads. Results such as 12, 15, 18, or even 20 heads are compatible with a fair coin; they do not require the coin to “correct” itself on flip 31.
| After 30 flips | What it means | Next-flip chance of heads |
|---|---|---|
| 15 heads | Exactly the expected count | 50% |
| 20 heads | Above expectation, still possible by chance | 50% |
| 10 heads | Below expectation, still possible by chance | 50% |
Even 25 heads in 30 flips is possible, but it is far less typical than the examples above. A surprising sample does not change the next fair flip’s 50% probability or create an obligation to balance the count.
A run does not create a debt
The probability of ten consecutive non-heads outcomes from a fair coin is (1/2)10, or about 0.098%. Once those ten outcomes have already occurred, however, the chance of heads on flip 11 remains 50%. The rarity belongs to the complete sequence viewed before it starts; it is not a force acting on the next trial.
Translate the idea to craps
Seven has a one-roll probability of 1/6. In 60 rolls, the expected count is 10 sevens, with a standard deviation of about 2.89. A set of 60 rolls might contain fewer or more than 10. The expectation is not a quota.
Three consecutive sevens have probability (1/6)3 = 1/216 when named before the sequence starts. After two sevens have appeared, the chance that the next fair roll completes the run is still 1/6. Conversely, ten consecutive non-sevens have probability (5/6)10, about 16.15%; after that run, the chance the next roll is also a non-seven is still 5/6.
| Observation | Valid conclusion | Invalid conclusion |
|---|---|---|
| No seven in the last 10 rolls | The observed stretch favored non-seven outcomes | Seven is now due |
| Three sevens in five rolls | The stretch contained unusually many sevens | Seven is now “hot” |
| A shooter made several points | That shooter produced a profitable past sequence for some bets | The next point is more likely to repeat |
Why casinos can tolerate winning sessions
A casino does not need every player or every shift to lose. It accepts volatile individual results while collecting positive expected value over enormous total action. A player experiences far fewer decisions, so session outcomes are much noisier than the long-run average.
This is why both statements can be true:
- A fair craps bet with a house edge has negative expectation for the player.
- The player can still finish a particular session ahead.
The winning session does not disprove the edge, and the edge does not guarantee a loss tonight.
Expected value and volatility answer different questions
Two wagers can have similar house edges but very different session behavior. A bet that resolves rarely for a large payoff may swing more sharply than a bet with frequent smaller results. Conversely, a high hit rate can hide a severe occasional loss.
| Question | Relevant measure |
|---|---|
| What does the wager cost on average? | Expected value / house edge |
| How widely might results move around that average? | Variance / standard deviation |
| How much can I lose in this construction? | Total exposure and outcome rules |
| How likely am I to hit my loss limit? | Distribution of session outcomes, not edge alone |
This distinction matters when adding Odds. The Odds portion carries zero edge, so it does not add expected loss under ideal payout rules, but it does add variance and dollars that can be lost on a point decision.
Variance is not a strategy signal
After results occur, you can describe a run as favorable or unfavorable. You cannot use that label to change the probability of the next independent roll. Switching to the Don’t side because a table “turned cold,” pressing because numbers look hot, or increasing a wager because a result is due adds a prediction that the dice do not support.
A fixed progression can redistribute outcomes—for example, producing many small wins and occasional large losses—but it cannot change the expected value of the underlying bets.
What shorter play actually does
Reducing hours or decisions usually reduces total expected loss because less money cycles through negative-edge wagers. In a short sample, observed averages are less stable; that does not mean its absolute dollar swings must exceed those of a longer comparable session. It does not create a mathematical player edge.
For example, if average action is $10 per decision at a 1.5% edge:
| Decisions | Total action | Expected loss |
|---|---|---|
| 20 | $200 | $3 |
| 100 | $1,000 | $15 |
| 500 | $5,000 | $75 |
Actual results around those averages can be wide, especially over 20 decisions. The expected cost still scales with action.
Practical controls that do help
- Set a maximum affordable loss before play.
- Limit time and total action rather than waiting for a streak to reverse.
- Prefer lower-edge bets when the entertainment is otherwise equal.
- Do not chase losses or raise limits after an unlikely sequence.
- Record total buy-in, cash-out, and hours if you want an honest picture of results.
- Treat a stop-win or stop-loss as an exposure boundary, not a system that changes expectation.
How to keep an honest session record
If you want to compare experience with expectation, record every session rather than only memorable wins:
- Date, game type, table rules, and approximate decisions or time played.
- Total cash or chips brought to the session.
- Total money added after the initial buy-in.
- Final cash-out, including chips carried away.
- Major wager changes, especially side bets or larger Odds multiples.
A small personal sample will remain noisy, so it cannot prove that a system beats the game. It can reveal whether actual spending, hours, and loss chasing match the story you tell yourself.
Frequently asked questions
Do results have to balance after 36 rolls?
No. Thirty-six is the number of ordered dice combinations, not a schedule the dice must complete.
Can a hot shooter stay hot?
The shooter can continue producing favorable results, but the past streak itself does not raise the next-roll probability.
Does walking away while ahead beat the house edge?
It preserves the session result by ending further exposure. It does not prove a repeatable positive-expectation strategy.
Does the law of large numbers mean the casino always wins?
No individual outcome is guaranteed. As independent trials accumulate, average results tend to become more stable around the mathematical expectation.
Turn a seven count into an actual wager result
Repeat a $1 Any Seven at 4 to 1 for exactly 60 independent fair rolls, with no presses or other bets. If K sevens occur, net profit is $4K minus the $1 lost on each of the other 60 − K rolls: 5K − 60 dollars. Ten sevens gives a $10 loss; twelve breaks even; fourteen gives a $10 gain.
The expected seven count is 10, so the expected wager result is −$10. The count’s standard deviation is about 2.89; multiplying by the $5 change in net result per extra seven gives about $14.43 standard deviation in dollars. That is a spread measure, not a guaranteed range or a chance-based reason to raise stakes. It connects frequency, payout and action: hitting the most common total can still lose money on average because the payoff is below fair odds.
