The Central Limit Theorem Explained With Dice: Why Averages Look Bell-Shaped

A Dice Demonstration

A single fair die has a flat (uniform) distribution: each of 1 through 6 is equally likely. The mean is 3.5 and the standard deviation is √(35/12) = 1.71. Now roll the die 30 times and compute the average. Repeat this many times. The averages will cluster tightly around 3.5, with a standard deviation of 1.71 ÷ √30 = 0.31. Plotting the averages gives a nearly perfect bell curve, even though the original rolls were flat.

With two dice, the sum already has a triangular shape, and with five or six dice it looks like a bell. The CLT formalizes that pattern.

The Standard Error

The standard deviation of the sampling distribution of the mean is called the standard error: SE = σ ÷ √n. It shrinks with the square root of the sample size. To cut the standard error in half, you need four times as many observations. This relationship underlies confidence intervals and margins of error: a 95% interval for a mean is about ±1.96 standard errors.

For a poll measuring a proportion, SE = √(p(1 − p) ÷ n). With p = 0.5 and n = 1,000, SE = 0.0158 and the margin of error is about ±3.1 percentage points.

What the Theorem Requires

The observations must be independent and drawn from the same distribution with a finite mean and variance. For most distributions, a sample size of 30 or more is enough for the normal approximation to work well in the middle of the distribution; for strongly skewed data, such as incomes with a long right tail, hundreds or thousands may be needed. If the data are already roughly normal, small samples suffice.

The theorem does not apply when observations are strongly dependent, such as time series with trends, or when the distribution has infinite variance, as with certain heavy-tailed phenomena.

Applications

Pollsters use the CLT to estimate how close a sample proportion is to the population proportion. Manufacturers use control charts, plotting the average of small samples over time, to detect shifts in a process. A/B tests compare means or proportions using normal-based tests. Insurers rely on the fact that the average claim cost over many policies is predictable, even though individual claims are not. The same logic explains why investment returns averaged over many assets or periods are less volatile than any single one.

Try It Yourself

Use a random number generator or spreadsheet to simulate 1,000 averages of 30 dice rolls, then make a histogram. Compare the center and spread with 3.5 and 0.31. Then try with a skewed distribution, such as the exponential, and see how larger sample sizes are needed for symmetry.