Reading Shape from the Mean and the Median

The median only cares about position: it's the middle value once you line everything up, so one or two extreme numbers can't drag it far. The mean cares about magnitude: it adds up every value and divides, so a handful of very large or very small numbers can pull it noticeably in their direction. That difference in sensitivity is the whole trick. If a few unusually low values stretch out a long left tail, they pull the mean down while the median stays anchored near the bulk of the data, so mean ends up below median. If a long right tail of unusually high values stretches out, the mean gets pulled up above the median.

So the rule of thumb: mean much less than median suggests skew left (a tail of low values), mean much greater than median suggests skew right (a tail of high values), and mean close to median suggests roughly symmetric data. But this is a clue, not a guarantee. You can build lopsided data sets where the pulls happen to cancel out and the mean lands exactly on the median anyway. Numbers alone never prove shape with certainty.

This is exactly why you don't stop at mean and median: you draw the picture. A histogram, dotplot, or boxplot shows the actual tail, the gaps, and any clusters that numbers can hide. Use the mean-median gap as a quick diagnostic that tells you what to expect before you graph, and which measure of center and spread to trust once you see the shape: medians and IQR for skewed data, means and standard deviation when things look symmetric.