Sampling Distributions

Suppose you want the average height of every adult in a city. You cannot measure everyone, so you take a random sample of 50 people and compute their mean height. You get, say, 170.4 cm.

Now imagine a friend does the exact same thing — 50 different people, chosen just as randomly. Their mean will not be 170.4 cm. Maybe 169.1. A third person gets 171.2. Same city, same method, different answers. This wobble has a name: sampling variability. The sample mean is not a fixed fact — it is a number that comes out a little different every time you draw a fresh sample.

The key move in statistics is to stop treating that wobble as a nuisance and start treating it as something with a shape we can describe.

It is easy to blur three different things together. Keep them separate:

• The population distribution — every raw value in the whole group (all the heights). It can be any shape.
• The distribution within one sample — the handful of values you actually collected this time.
• The sampling distribution of the mean — not raw values at all, but the collection of means from many imagined samples.

The first two are about individual data points. The third is about a statistic. Confusing them is the single most common source of error in this topic, so whenever you read the word distribution, pause and ask a distribution of what?

The simulation below lets you watch a sampling distribution assemble itself. The top strip is the population — pick a lop-sided one like the right-skewed or bimodal shape. The bottom panel starts empty and fills in one dot of the sampling distribution each time you draw a sample and record its mean.

Try this: set the sample size to n = 1 and draw many samples — each mean is just one raw value, so the bottom copies the messy top. Now raise n and keep drawing. Two things happen to the pile of means: it becomes more symmetric and bell-like, and it gets narrower, hugging the true center. Those two effects are the heart of everything that follows.

Draw enough samples and you will notice the pile of means centers right on the population's true mean — it does not drift high or low. In symbols, the average of the sample mean over all possible samples equals the population mean:

The sampling distribution has a spread of its own, and it is smaller than the spread of the raw population — averaging several values smooths out the extremes. This spread of the sample mean gets its own name, the standard error, to keep it distinct from the population's standard deviation \( \sigma \):

You now have the two facts that make inference possible: the sampling distribution of the mean is centered on the parameter and has a known spread, the standard error. The remaining question is its shape — and there is a remarkable result, the Central Limit Theorem, which says that for a large enough sample the sampling distribution of the mean is approximately a normal bell curve no matter what the population looks like. That is the next step, and it is what turns these ideas into confidence intervals and tests.