Dice guide
Dragon Dice: variance and the spread of results
Dragon is just the label on this page; the rules are ordinary number dice. The topic is variance: how far a run of results tends to land from the expected value, and how that distance grows with the number of rolls.
Spread of one roll
Take a roll-under target of 20.00 at a 1% edge. The chance is 20% and the payout is 4.950x. A one-unit bet ends at +3.950 or -1. The mean of those two outcomes, weighted by their chances, is -0.0100 units. The standard deviation, the typical distance from that mean, is 1.980 units.
That spread is about 198 times larger than the expected loss. On a single roll, the edge is invisible. Only over many rolls does it start to show.
Spread over many rolls
For independent rolls, the expected net grows in proportion to n, the number of rolls. The standard deviation grows only with the square root of n. That one fact explains most of what players notice. Short sessions are dominated by luck. Long ones drift toward the expected loss.
| Rounds | Expected net | SD of net | Rough 95% band (normal approx.) |
|---|---|---|---|
| 50 | -0.50 | 14.00 | -28.50 to 27.50 |
| 500 | -5.00 | 44.27 | -93.55 to 83.55 |
| 5,000 | -50.00 | 140.01 | -330.01 to 230.01 |
After 50 rolls the rough 95% band runs from a loss of about 29 units to a gain of about 28. A winning session is common. After 5,000 rolls the band is centered on a loss of 50 units and is about 560 units wide.
The band uses a normal approximation, which is rough for 50 rolls, since the win count there is small and skewed. It is better for 500 rolls and good for 5,000.
How the target changes variance
Lower chances carry more variance because the payout is larger. Compare the per-roll standard deviation at three targets, all with the same 1% edge:
- Roll under 5.00: 4.315 units per roll.
- Roll under 20.00: 1.980 units per roll.
- Roll under 80.00: 0.495 units per roll.
Over 1,000 rolls these become 136.5, 62.6 and 15.7 units. The expected loss is 10 units in all three cases.
Why this matters
Variance is why one good evening proves nothing about a target or a method. A gain of 44 units over 500 rolls at 20% is only a one-standard-deviation event. It happens to roughly one session in six. The expected value is the only number that does not depend on luck, and here it is negative.