Why middle values dominate
Roll a single die and every face from 1 to 6 is equally likely. The picture is flat. Now roll two dice and add them. The totals are no longer equally likely: 7 can be made six ways, while 2 and 12 can each be made only one way. The flat picture has become a peak in the middle. That pattern is the start of the bell curve.
The reason is counting. Middle totals can be built from many different combinations of the individual dice, while extreme totals need nearly everything to line up. The more dice you add, the sharper this effect becomes.
Ten dice
Roll ten dice and add them. Each die averages 3.5, so the total averages 35. The lowest possible total is 10 and the highest is 60. There are 6 to the power 10, or 60,466,176, equally likely outcomes, and exactly one of them is all ones. Another single outcome is all sixes. Each of those extremes has a probability of about one in 60 million.
By contrast, totals near 35 can be reached in an enormous number of ways, so they are by far the most common. If you plot how often each total occurs, you get a smooth, symmetric, bell-shaped curve centred at 35. This is the normal distribution.
The size of the spread
The curve’s width is measured by the standard deviation. One die has a standard deviation of about 1.71. For independent dice, the variances add: ten dice have variance 10 × 2.917, about 29.2, and so a standard deviation of about 5.4. The standard deviation of the total grows with the square root of the number of dice, while the average grows in direct proportion to the number of dice.
A useful rule of thumb for bell-shaped data:
- About 68 percent of results fall within one standard deviation of the average.
- About 95 percent fall within two.
- Nearly all (about 99.7 percent) fall within three.
For ten dice, that means roughly two out of three totals land between about 30 and 40, and about 95 percent between about 24 and 46. A total of 50 or more would be unusual, at nearly three standard deviations above average. These figures are approximations, since a real total of dice takes whole numbers and the curve is a smooth model.
Coins show the same thing
Flip a fair coin 100 times. The number of heads has an average of 50 and a standard deviation of 5. About two-thirds of the time you will see between 45 and 55 heads, and about 95 percent of the time between 40 and 60. Getting 70 heads or more is a long way out, four standard deviations above the average, and very unlikely. This is a case of the binomial distribution, which the normal curve approximates well once there are many trials.
Why it keeps appearing
The reason sums of many small independent random pieces tend to a bell shape is a result in mathematics called the central limit theorem. Loosely, it says that when you add many independent contributions of similar size, the total’s shape approaches a normal curve, whatever the shape of an individual piece. A die is flat, a coin is two spikes, yet their sums are bell-shaped.
This is also why averages become more predictable with more trials. The spread of the average shrinks with the square root of the number of trials, which links to what the law of large numbers promises and what it does not, and to how confidence intervals around an observed rate get narrower with more data.
What it is useful for
- Judging what is ordinary. Knowing the spread lets you tell a normal swing from an unusual one. A session result that is within a standard deviation or two of the average is not remarkable.
- Comparing games. Two games with the same average can have different spreads. Wider spread means bigger swings, which is the distinction behind volatility and return.
- Quick approximations. Calculating exact probabilities for sums of many dice or flips is tedious, but the bell curve gives good estimates quickly.
- Planning simulations. Results from a Monte Carlo simulation usually cluster in the same bell shape, which helps in judging how reliable an average from a given number of simulated trials is.
Where the bell curve fails
It is a model, not a law, and it can mislead.
- Few events. With just one die or a few coins the shape is not bell-like at all.
- Dependence. If the events are linked, as with cards drawn without replacement in some patterns, sums can behave differently from the independent case.
- Rare, large outcomes. In games where most plays return little and a few return a lot, totals can be strongly skewed. The bell curve may need a great many plays before it becomes a good fit, and it can understate how often large swings occur. Slot odds are an example of a setup where a naive bell-curve estimate would be unreliable.
- Tails. The normal curve’s extreme tails are only approximate for real processes. Treat very small tail probabilities with caution.
- Skill and changing conditions. If the chance per trial shifts over time, the pooled results are no longer simple sums of identical events.
One caution
A bell curve describes where results usually land, not what will happen next. Knowing that a total is most likely to fall near the average gives no reason to expect it to be pulled back toward the middle after an extreme. Each new independent event is unaffected by earlier ones.
None of this makes a game with a house edge profitable. Typical ranges tell you what swings are normal; they do not remove the long-run disadvantage. Outcomes are uncertain, and a local support service can help anyone who finds gambling hard to control.