Every dataset has a "typical" value — but which measure you pick to describe it changes the story you tell. This page walks through all three with one real example, an outlier you can toggle on and off, and a question to check your instinct.
Each measure answers a slightly different question about the same list of numbers.
mean = (sum of all values) ÷ (count of values)
Add everything up and split it evenly. The mean accounts for every single value — which also means one extreme value can drag it far from where most of your data actually sits.
median = the value in the middle, once sorted
Sort the data low to high and pick the middle entry (or average the two middle entries, for an even count). The median only cares about position, not magnitude — so extreme values barely move it.
mode = the most frequently occurring value
The only measure that works on non-numeric data too (e.g. "most common browser"). A dataset can have one mode, several, or none at all.
Seven food-delivery orders were logged with their delivery time in minutes. One order took much longer than the rest — a jammed intersection, most likely. Toggle it on and off below and watch what happens to each measure.
18, 22, 20, 45, 15, 24, 18 — order #4 took 45 minutes, well outside the usual range.
(18+22+20+45+15+24+18) ÷ 7 = 23.14 — pulled upward by the 45-minute order.
(18+22+20+15+24+18) ÷ 6 = 19.5 — drops by more than 3.5 minutes from one data point.
Sorted: 15,18,18,20,22,24,(45). Middle value is 20 with the outlier, 19 without — barely moves.
18 appears twice — more than any other value, with or without the 45-minute order.
There's no single "correct" measure — it depends on whether outliers are meaningful noise or part of the story.
Your data is roughly symmetric with no extreme outliers, and you need every value to count — e.g. average revenue per user across a large, stable customer base.
Best for: totals, budgets, evenly-distributed dataOutliers exist and would distort the picture — household income, home prices, delivery times, or anything with a long tail on one side.
Best for: skewed data, "typical case" reportingYou're working with categories rather than numbers, or you specifically want to know the single most common outcome — most-ordered menu item, most common shoe size.
Best for: categorical data, inventory, popularity