What a Sampling Distribution Is

A parameter is one fixed number describing a whole population: the true proportion of all voters who support a bill, say. You never see it directly. A statistic is what you calculate from a sample, and samples change every time you draw a new one, so the statistic changes too. Survey 500 people today and you might get 54% support. Survey a different 500 tomorrow and you might get 51%. Same population, same question, different people, different answer. That variation is not an error, it is the natural result of chance choosing different individuals each time.

To build a sampling distribution, imagine taking a sample of size n from the population, computing the statistic, writing it down, then putting everyone back and drawing a brand new independent sample of the same size, computing the statistic again, and repeating this thousands of times. Plot all those statistics and you get the sampling distribution. It is not the data from one sample, which just shows individual responses like yes or no. It is a distribution of summary numbers, one per sample, showing how much the statistic itself bounces around.

Sample size changes the spread of this distribution without changing its center. Samples of size 5 produce statistics scattered widely, because one unusual person can swing a small sample a lot. Samples of size 50 cluster much more tightly around the true parameter, because individual quirks average out. This is why larger samples give more trustworthy estimates, and it sets up the next ideas: the sampling distribution's shape, center, and spread let you measure how far a single real sample's statistic is likely to fall from the truth, which is exactly what confidence intervals and significance tests use.