The Central Limit Theorem
Why averages turn into a bell curve — even when the raw data is nothing like one.
💡
The big idea: Take a messy, lop-sided population, draw a sample, and record its average. Do that over and over, and the collection of averages piles up into a smooth bell curve — no matter what shape the original data had. That surprising fact, the Central Limit Theorem, is what lets statistics say confident things about a whole population from a single sample.
Three different things called 'a distribution'
It is easy to get tangled here, so let us separate three ideas up front:
- The population — every raw value out there (all the incomes, all the wait times). It can be any shape: skewed, flat, two-humped.
- A sample — a handful of values you actually collect, say n of them.
- The sampling distribution of the mean — what you get if you imagine taking many samples and writing down just the average of each one.
The Central Limit Theorem is a statement about that third thing.
🔑
The Central Limit Theorem, in one sentence
If you take samples of size n from almost any population and record each sample's mean, then as n grows the distribution of those means becomes approximately normal (a bell curve) — centered on the true population mean, and getting narrower as n increases. (This assumes each sample is drawn randomly and independently.)
See it happen
Below, pick a population that looks nothing like a bell — try the right-skewed one, or the two-humped 'bimodal' one. The top strip shows the raw population's true shape. The bottom panel collects the averages of the samples you draw.
Two knobs do two different jobs — keep them apart:
- Drawing more samples (the Draw 1 and Draw 100 buttons) simply fills in the bottom histogram so its shape becomes clear. It does not change the shape you are filling in.
- The sample size n is what controls that shape. Slide n down to 1 and each 'average' is just a single raw value — so the bottom matches the lop-sided top. Now raise n: the pile morphs into a bell and, at the same time, gets narrower.
That second knob is the theorem at work: it is the size of each sample — not how many samples you take — that turns averages into a bell.
🎮 Interactive: the CLT in action LIVE
Top: the population (often not a bell). Bottom: the distribution of sample means, which fills in as a bell as you draw more samples. The smooth curve is the normal shape the CLT predicts. Increase n and watch the spread of means shrink.
✨
The two things that surprise everyone
1. The shape. The bell appears for the means even when the raw data is skewed or bimodal — averaging washes out the original shape, and the approximation gets better as the sample size n grows (small samples from very skewed data are only roughly bell-shaped). 2. The narrowing. Bigger samples give averages that cluster tightly around the true mean, because one extreme value gets diluted by more ordinary ones.
How much does the spread shrink?
The spread of the sample means has its own name: the standard error. It is not the same as the population's spread — it is smaller, and it depends on the sample size:
\[ \text{SE} = \frac{\sigma}{\sqrt{n}} \]
Standard error of the mean: the population standard deviation σ divided by the square root of the sample size n.
⚠️
It is √n, not n
Because of the square root, cutting the spread of the averages in half takes four times the sample size, not twice. Going from n = 25 to n = 100 (×4 the data) halves the standard error. Diminishing returns are built into sampling.
📝 Worked example: A population of delivery times is heavily right-skewed with mean 30 min and standard deviation 12 min. You take samples of n = 36 deliveries and average each. What does the CLT predict about those averages?
- Shape: n = 36 is large enough here for the distribution of the sample means to be approximately normal — a bell — even though delivery times themselves are skewed. (Heavier skew would need a larger n.)
- Center: the means are centered on the population mean, 30 min.
- Spread: standard error = σ/√n = 12 / √36 = 12 / 6.
✓ The sample averages form a bell centered at 30 min with a standard error of 2 min — so most sample averages land within a few minutes of 30, even though individual deliveries vary a lot more.
Check your understanding
1. The Central Limit Theorem says that as the sample size grows, the distribution of the sample MEAN becomes approximately…
That is the whole point: the means become approximately normal regardless of the population's shape.
2. A population is strongly skewed. You take large samples and look at the sample means. Their distribution will be…
The CLT applies to the means, not the raw data. Averaging pulls the distribution of means toward a bell even when the population is skewed.
3. The population standard deviation is σ = 20. With samples of size n = 100, the standard error of the mean is…
SE = σ/√n = 20/√100 = 20/10 = 2.
4. You currently use n = 25. To HALVE the standard error of your sample means, you should use a sample size of about…
SE depends on √n, so halving it needs 4× the sample size: 25 × 4 = 100.
✅ Key takeaways
- Population, sample, and sampling distribution of the mean are three different things — the CLT is about the third.
- The Central Limit Theorem: sample means become approximately normal as n grows, whatever the population's shape.
- The means are centered on the true population mean.
- Their spread is the standard error, SE = σ/√n — it shrinks as n grows, but only with the square root.
- Halving the standard error takes four times the sample size, not twice.