Mean, median or mode: which average do you want?

The average is three different calculations wearing one name, and choosing the wrong one is how honest people produce misleading statistics.

The three, and what each finds

The mean adds everything and divides by the count. It uses every value, which makes it sensitive to all of them — including the extreme ones. The median sorts the values and takes the middle one, so it reports the point half the data sits above. The mode is simply the most frequent value.

For evenly spread data all three land close together and the choice hardly matters. The further the data departs from symmetry, the further they separate — and the separation itself is the most useful signal in the whole subject.

Where the mean misleads

Income is the standard example and it is standard because it is stark. In almost any country the mean income sits well above the median, because a small number of very high earners pull the mean upward while the median is untouched by how extreme they are. Quoting mean income describes a typical person badly.

The same asymmetry appears in house prices, wealth, company sizes, download counts and web page views. Any quantity where the maximum is far from the minimum and the distribution has a long tail will have a mean that describes almost nobody. When a distribution is skewed, the median is the honest summary.

Where the median is the wrong tool

The median discards magnitude, which is sometimes exactly what you need and sometimes disastrous. If you are budgeting a total — payroll, fuel, materials — the mean multiplied by the count gives the right total and the median does not. The mean is the only average that reconstructs a sum.

The median is also insensitive to changes at the extremes, which can hide something important. If the highest earners double their income and nobody else changes, the median does not move at all. Whether that is a feature or a flaw depends entirely on the question being asked.

When the mode earns its place

The mode is the only one of the three that works on data with no numeric order — the most common shoe size sold, the most frequent fault code, the most popular delivery slot. For categorical data it is not merely the best choice, it is the only one available.

It also reveals structure the others hide. A dataset with two distinct peaks — a bimodal distribution — usually means two different populations have been mixed together. Both the mean and median will report a single central value that may correspond to nothing real, sitting in the valley between two genuine clusters.

Reading the gap between them

Because the three respond differently to shape, comparing them diagnoses the data. Mean and median close together suggests a roughly symmetric distribution. Mean noticeably above median indicates a tail of high values. Mean below median indicates a tail of low ones. A mode far from both suggests either strong skew or more than one underlying group.

This is why a good calculator shows all three rather than asking you to pick in advance. The relationship between them tells you which one you should have asked for.

Weighted averages, and the trap in them

When values represent different quantities, a plain mean is wrong. Averaging the price per litre across three fuel purchases of different sizes gives a figure nobody paid; weighting each price by the litres bought gives the real average cost. The same applies to grades across modules of different credit values, and to returns across investments of different sizes.

The test is simple: if the things you are averaging are not equally important, an unweighted mean is quietly assuming they are.