About the Central Limit Theorem
The Central Limit Theorem (CLT) is one of the most powerful results in statistics: if you draw a large enough random sample from almost any distribution and compute its mean, those sample means will be approximately normally distributed — regardless of the shape of the underlying distribution. Formally, if X has mean μ and standard deviation σ, then the distribution of the sample mean X̄ (from samples of size n) converges to N(μ, σ²/n) as n grows. This explains why so many natural and social phenomena follow bell-curve distributions: they are sums of many independent influences.
Choose a source distribution (Uniform, Exponential, Bimodal, Poisson, or Right-Skewed), set the sample size n, and press Start. The top strip shows the source distribution's shape; the lower histogram accumulates sample means as they are drawn. A normal curve overlay appears in amber, and the stats panel tracks how closely the empirical standard deviation matches the theoretical σ/√n.
Frequently Asked Questions
What exactly does the CLT say?
For independent and identically distributed (i.i.d.) random variables X₁, X₂, …, Xn with mean μ and finite variance σ², the standardised sample mean (X̄ − μ) / (σ/√n) converges in distribution to the standard normal N(0,1) as n → ∞. In practice, n ≥ 30 is often sufficient for symmetric distributions; strongly skewed distributions may need n ≥ 100 or more.
Why does the histogram narrow as I increase the sample size?
The standard error of the mean is σ/√n — it shrinks as the square root of sample size. Doubling n halves the standard error, making the bell curve twice as narrow. This is why averaging repeated measurements is such a powerful tool for reducing uncertainty: collecting four times as many data points halves the uncertainty in the mean.
Does the CLT work for bimodal distributions?
Yes — the CLT applies to any distribution with a finite mean and variance, including bimodal ones. The simulator's Bimodal option uses a mixture of two Gaussians centred at 0.25 and 0.75. Even though the source shape has two peaks, by the time you have drawn a few hundred samples of size n = 30, the histogram of means is already strikingly bell-shaped.
What is the difference between population variance and the standard error?
Population variance σ² describes how widely individual observations are spread around the population mean. The standard error σ/√n describes how widely sample means are spread — it is much smaller because averaging n values "cancels out" some of the individual variability. Confusing the two is a common statistical error: polling companies quote the standard error of their survey, not the spread of individual opinions.
Are there distributions where the CLT does not apply?
Yes. Distributions with infinite variance (such as the Cauchy distribution, or heavy-tailed power-law distributions with exponent ≤ 2) do not satisfy the CLT's conditions. For Cauchy-distributed variables, no matter how many you average, the sample mean has the same Cauchy distribution — it never converges to normal. Financial returns during market crashes sometimes exhibit such heavy tails, which is why normal-distribution risk models can catastrophically underestimate extreme losses.
Why do most measurements in science follow a normal distribution?
Most physical measurements are the result of many small, independent additive errors — instrument imprecision, environmental fluctuations, rounding, and so on. By the CLT, the sum of many small independent errors converges to a normal distribution regardless of their individual shapes. This is why Gaussian (normal) error models are so ubiquitous in experimental science.
What is the "rule of thumb" n ≥ 30?
The rule of thumb that a sample size of 30 is sufficient for the CLT to apply comes from empirical experience with moderately skewed unimodal distributions. For a perfectly symmetric source (like Uniform), n ≥ 5 already gives a near-normal histogram of means. For very skewed distributions (like Exponential), you may need n ≥ 100. In this simulator you can verify this directly by comparing the normal fit quality at different n values.
How is the CLT used in hypothesis testing?
The CLT justifies using z-tests and t-tests even when the underlying population is not normal. For example, when comparing the mean cholesterol level of two groups, the CLT guarantees that the sampling distribution of the difference in means is approximately normal for large enough samples, so a z-test is valid. This makes the CLT the invisible foundation of most classical statistical inference.
What is the connection between the CLT and the bell curve in nature?
Human heights, test scores, measurement errors, and many biological traits follow approximately normal distributions because they result from the additive effects of many independent genetic and environmental factors. By the CLT, the sum of many small independent contributions — whatever their individual distributions — tends to normality. Francis Galton called this the "supreme law of Unreason" when he discovered it in the 1870s.
What does the "Normal fit match" percentage in the stats panel mean?
The simulator computes the theoretical standard deviation of the sampling distribution as σ_theory = σ/√n, then measures the empirical standard deviation of the accumulated sample means. The "match" percentage is 100% × (1 − |σ_empirical − σ_theory| / σ_theory), capped at 0%. A value close to 100% means the histogram is tightly following the CLT prediction; low values indicate you need more samples or a larger n.