Research Methods · Simplified
Central Limit Theorem Simulator
Two panels, one axis. Only one of them changes shape, and it is not the data.
The simulator
The population, one sample, and the means so far
Draw samples from a population that is nothing like normal. Each sample gives one mean, and the means pile up in the lower panel.
Upper panel: the population, with the most recent sample drawn as ticks beneath it. Lower panel: the mean of every sample drawn so far. Both panels share one value axis unless the lower one is zoomed.
| Distribution of the sample mean | What the theory predicts | What this simulation measured |
|---|
What this shows
The data never became normal
The upper panel has not changed shape once, at any sample size. It cannot: it is the population, and drawing from it does not alter it. Only the lower panel changes, and the lower panel is not data. It is the distribution of a statistic computed from data.
Key idea: The theorem is about the sample mean, not about the sample. As the sample size grows, the distribution of the mean becomes more normal and narrower, at a rate of one over the square root of the sample size. Three things are being confused whenever someone says data become normal with a big enough sample: the population, one sample from it, and the distribution of a statistic across many samples. This figure keeps all three on screen at once so they can be told apart.
Two cautions. The zoom control changes how wide the means look and nothing about how wide they are: read the standard error, not the picture. And "thirty is enough" is a rule of thumb, not a fact. Set the skewed population to a sample size of thirty and the predicted skewness of the mean is still 0.37, while the bimodal population is already close to symmetric by ten. How large a sample needs to be depends on the population it comes from, which is exactly what a fixed number cannot tell you.
The longer version adds four populations, excess kurtosis alongside skewness, a three-part prediction before the simulation, and a challenge on what the theorem does and does not claim. It is at Central Limit Theorem Simulator in the main collection.