Why the average is not enough
The average tells you where a random result is centred. It does not tell you how far results typically land from that centre. A die that always shows 3 or 4 and a die that spreads across 1 to 6 can have the same mean, yet they behave differently in play. Variance and standard deviation are the standard ways to measure that spread.
Variance of one die
Start with a fair six-sided die. The mean is 3.5. Variance is the average squared distance from the mean, so for each face compute (value − 3.5)², then average the six results.
| Face | Distance from 3.5 | Squared |
|---|---|---|
| 1 | −2.5 | 6.25 |
| 2 | −1.5 | 2.25 |
| 3 | −0.5 | 0.25 |
| 4 | +0.5 | 0.25 |
| 5 | +1.5 | 2.25 |
| 6 | +2.5 | 6.25 |
The squares add to 17.5, and 17.5 / 6 is about 2.917. That is the variance. (As a check, the exact value is 35/12.)
Squaring does two jobs. It stops positive and negative distances from cancelling out, and it gives extra weight to results far from the mean.
Standard deviation
The variance is measured in squared units, which are awkward to interpret. Taking the square root returns you to the original units. The standard deviation of one die is the square root of 2.917, which is about 1.71.
A rough reading: a typical roll lands about 1.7 away from 3.5. That fits the table, where the faces sit between 0.5 and 2.5 away.
Adding dice
When you add independent random quantities, their variances add. Two dice have variance 2 × 2.917 = 5.833, so a standard deviation of about 2.42. The mean of the total is 7, so most totals land within a couple of points of it.
Here is how the pattern continues for n dice:
| Dice | Mean of total | Variance | Standard deviation |
|---|---|---|---|
| 1 | 3.5 | 2.92 | 1.71 |
| 2 | 7 | 5.83 | 2.42 |
| 10 | 35 | 29.17 | 5.40 |
Notice that standard deviation does not grow in step with the number of dice. It grows with the square root. Ten times the dice gives a spread about 3.2 times larger, not 10 times (5.40 / 1.71 is about 3.16).
Why more dice feel steadier
Compare the spread to the mean. For one die, 1.71 / 3.5 is about 0.49: the typical roll is about half the average away from it. For ten dice, 5.40 / 35 is about 0.15: the typical total is within about 15% of the average. So the total of many dice is relatively more predictable, even though in absolute terms the spread is bigger.
This is the mechanism behind several game design choices. A system that sums many small rolls clusters near the middle, while a system built on a single roll is swingier. Tabletop designers who compare dice pools with single die systems are, in part, choosing between those two behaviours. The same reasoning is the engine of the law of large numbers.
A rough guide to what the spread means
For totals of many dice, the shape of the distribution approaches a bell curve. In that case, roughly two thirds of results fall within one standard deviation of the mean, and roughly 95% within two. For ten dice, that suggests most totals land between about 30 and 40, and a total outside 24 to 46 would be rare. These bands are approximations. With one or two dice the shape is flat or triangular, so the rule of thumb is looser.
You can verify any of this by building a probability table for dice in a spreadsheet, or by plotting outcome histograms of simulated rolls and watching the distribution narrow and round off as dice are added.
Putting it to use
Variance gives a number to a feeling players already have about how swingy something is. Two attacks with the same average damage but different variance will differ in how often they deliver a very low or very high result. That matters when you need a result that clears a threshold, and it matters over short stretches of play, where the spread dominates.
Keep a few cautions in mind. Standard deviation treats upward and downward deviations alike, so it does not say whether the extreme results are good or bad for you. It is also most informative for roughly symmetric outcomes. For lopsided ones, with rare huge payouts, it can understate how unusual the extremes are, and the average and spread together may still mislead. Use it alongside the full list of outcomes whenever the stakes are meaningful.