Interactive guide
However lopsided the population, the means turn into a bell
Not "the data becomes normal": the data never changes. It's the distribution of sample means, taken over and over, that piles up into a bell curve, centered on the population's own mean, no matter how strange the population looks.
One statistic, repeated many times
Draw n values from the population and average them: that's one sample mean, one single number. Do it again and you'll get a slightly different number. Do it thousands of times and plot every one of those means as a histogram: that histogram is the sampling distribution of the mean, and it's what this whole page is about.
The theorem makes three specific promises about that histogram: it centers on the population's true mean μ, its spread is the population's standard deviation shrunk by σ/√n, and its shape approaches normal as n grows, even when the population itself isn't normal at all.
Pick a population below, including two that are nowhere close to bell-shaped, then draw samples and watch all three promises hold.
One draw, end to end
-
1
Draw n values from the population
Independently, with replacement: each draw doesn't affect the next.
-
2
Average them
One sample, one mean: a single number that summarizes that draw.
-
3
Plot it and repeat
Add that mean to a running histogram, then draw a fresh sample and do it again.
-
4
Watch the histogram settle
More draws resolve its shape more clearly; the bell was always the destination, not something that grows over time.
The population every sample is drawn from, unchanged by anything below.
population, one bar per value population mean (μ)
—
- Population mean μ
- —
- Population SD σ
- —
Draw samples to build up the histogram of their means.
sample means drawn so far theoretical normal curve: mean μ, spread σ/√n
Pick a sample size, then start drawing.
- Samples drawn
- 0
- Mean of means
- —
- SE (empirical)
- —
- SE (theory σ/√n)
- —
Try this: pick Skewed, set n to 1, and press Draw 1000 samples. At n=1 a "sample mean" is just a single draw, so the histogram is the lopsided population itself, tail and all. Now press Reset, bump n up to 30, and draw 1000 more. Same population, same theoretical curve formula, but the histogram now sits almost exactly on top of it. Nothing about the population changed; only how many values got averaged together before being plotted.
The jargons
The Central Limit Theorem is a claim about the sampling distribution specifically — not the population, and not any one sample — so most of the confusion below comes from mixing up which of those three a term actually describes.
The full distribution every sample is drawn from. Fixed and unchanging: the theorem is a claim about what happens to sample means, never about the population reshaping itself.
The average of one sample of size n. A single number: draw a new sample and you'll generally get a different one.
The distribution you'd get by taking the sample mean over and over and histogramming the results: a distribution of a statistic, not of raw data.
The standard deviation of the sampling distribution itself: σ/√n. Larger samples average out more randomness, so SE shrinks as n grows, just slowly, since it depends on the square root.
How many values get averaged into one sample mean. Not how many samples you've drawn; that's a separate count, tracked here as "samples drawn."
How close the sampling distribution's actual shape is to a true bell curve at a given n. Improves as n grows, and grows faster for populations that were closer to symmetric to begin with.