Central Limit Theorem — Why Averages Form Bells
Why averages of almost any population pile up into a bell curve
This concept builds on Normal Distribution — The Bell Curve Explained. Consider exploring it first for the best experience.
Start With Any Distribution
The top panel shows the population every draw comes from. Choose from the 'Population Distribution' menu — Uniform, Skewed (Exponential) or Bimodal — and the red bars change shape. A single draw from any of them is unpredictable, and none of the three looks like a bell: Uniform is flat, Skewed has a long tail to the right, and Bimodal has two separate humps. Press 'Next Stage' when you are ready to start sampling.
Compute Sample Means
The bottom panel starts to fill. Each purple block is one sample mean x̄: the average of n fresh draws from the population above, with n chosen on the 'Sample Size (n)' slider. Watch the first 30 means drop in one at a time — each is a single observation from the sampling distribution of the mean. A mean of n draws rarely lands far out, because a high draw is usually balanced by a lower one.
The Magic: Means Form a Bell
As the count climbs toward 100, the histogram of means piles up in the middle and thins out at both ends. Switch the 'Population Distribution' menu to Skewed or Bimodal: the population on top changes completely, yet the means below still build the same symmetric hump. That is the Central Limit Theorem at work. With n = 1 each mean is just one draw, so the bottom histogram copies the top shape; raise n and the bell appears.
Larger n = Narrower Bell
With 300 means collected, a green normal curve is laid over the histogram. Its width is the standard error SE = σ/√n, shown in the 'Std Error (σ/√n)' card. Drag the 'Sample Size (n)' slider up and the bell narrows. Tick Compare Mode to overlay a second sample size as a cyan outline and compare the two widths. Both outlines centre on the same spot, because every sample size estimates the same population mean.
Worked Example: σ = 10, n = 25 gives SE = 2
Keep 'Population σ' at 10 and move the 'Sample Size (n)' slider to 25. The 'Std Error (σ/√n)' card now reads 2.000, because SE = σ/√n = 10/√25 = 10/5 = 2. The 'Measured SD of Means' card, computed from the 300 means on the canvas, lands close to 2 as well. Quadruple n to 100 and SE halves to 1.
The Formal Statement
If X₁, …, Xₙ are independent draws with mean μ and standard deviation σ, then (X̄ − μ)/(σ/√n) approaches the standard normal N(0, 1) as n grows; often n ≥ 30 is enough. You never need the population's shape to make probability statements about x̄ — the basis of confidence intervals and z-tests. The draws must be independent and σ must be finite; a heavily skewed population needs a larger n.
How It Works — Step by Step
-
1
Start With Any Distribution
The top panel shows the population every draw comes from. Choose from the 'Population Distribution' menu — Uniform, Skewed (Exponential) or Bimodal — and the red bars change shape. A single draw from any of them is unpredictable, and none of the three looks like a bell: Uniform is flat, Skewed has a long tail to the right, and Bimodal has two separate humps. Press 'Next Stage' when you are ready to start sampling.
-
2
Compute Sample Means
The bottom panel starts to fill. Each purple block is one sample mean x̄: the average of n fresh draws from the population above, with n chosen on the 'Sample Size (n)' slider. Watch the first 30 means drop in one at a time — each is a single observation from the sampling distribution of the mean. A mean of n draws rarely lands far out, because a high draw is usually balanced by a lower one.
-
3
The Magic: Means Form a Bell
As the count climbs toward 100, the histogram of means piles up in the middle and thins out at both ends. Switch the 'Population Distribution' menu to Skewed or Bimodal: the population on top changes completely, yet the means below still build the same symmetric hump. That is the Central Limit Theorem at work. With n = 1 each mean is just one draw, so the bottom histogram copies the top shape; raise n and the bell appears.
-
4
Larger n = Narrower Bell
With 300 means collected, a green normal curve is laid over the histogram. Its width is the standard error SE = σ/√n, shown in the 'Std Error (σ/√n)' card. Drag the 'Sample Size (n)' slider up and the bell narrows. Tick Compare Mode to overlay a second sample size as a cyan outline and compare the two widths. Both outlines centre on the same spot, because every sample size estimates the same population mean.
-
5
Worked Example: σ = 10, n = 25 gives SE = 2
Keep 'Population σ' at 10 and move the 'Sample Size (n)' slider to 25. The 'Std Error (σ/√n)' card now reads 2.000, because SE = σ/√n = 10/√25 = 10/5 = 2. The 'Measured SD of Means' card, computed from the 300 means on the canvas, lands close to 2 as well. Quadruple n to 100 and SE halves to 1.
-
6
The Formal Statement
If X₁, …, Xₙ are independent draws with mean μ and standard deviation σ, then (X̄ − μ)/(σ/√n) approaches the standard normal N(0, 1) as n grows; often n ≥ 30 is enough. You never need the population's shape to make probability statements about x̄ — the basis of confidence intervals and z-tests. The draws must be independent and σ must be finite; a heavily skewed population needs a larger n.
Key Takeaways
Shape Doesn't Matter
The original population can be skewed, bimodal, or bizarre — sample means always converge to a bell curve. The CLT overcomes our ignorance of the true distribution.
Standard Error = σ/√n
The spread of sample means decreases with sample size. Quadrupling n halves the standard error — this is why larger studies give more precise estimates.
n ≥ 30 Rule of Thumb
For most distributions, sample means are approximately normal when n ≥ 30. Very skewed distributions may need n > 100 for good approximation.
Foundation of Inference
Z-tests, t-tests, confidence intervals — all rely on the CLT guarantee. Understanding CLT makes statistical inference feel inevitable, not mysterious.
Connected Concepts
Normal Distribution
The CLT explains WHY the bell curve appears so often — sums of independent random variables converge to it.
Z-Score
Standardize sample means using Z = (x̄ − μ) / (σ/√n) to test hypotheses.
Confidence Intervals
CIs are built on CLT: x̄ ± z*(σ/√n) works because x̄ is approximately normal.
Hypothesis Testing
Z-tests and t-tests rely on the CLT guarantee that sample means are normally distributed.
Law of Large Numbers
The law says x̄ settles on μ as n grows; the CLT adds how the leftover error is spread — a bell of width σ/√n.
Galton Board
Each ball adds up many left-or-right bounces, and the sums pile into a bell — the CLT made of pegs.
Standard Deviation
σ measures the spread of single draws; divide it by √n to get the spread of their averages.
Updated
- — Steps explain each population shape and the n = 1 case; two new quiz questions and three new connections.
- — Stages now hold long enough to read, a Next Stage button, a σ slider and a worked σ = 10, n = 25 → SE = 2 step.