Mean, median and mode are three different averages
"The average" is not one number. It is at least three of them, they can be far apart, and the phrase itself never says which one you are being given.
Take this set of nine values. It is invented for this example — it is not measured data of any kind, and nothing below should be read as a fact about the world:
2, 3, 3, 4, 5, 6, 7, 8, 122
The three answers
The mean is the total shared out equally. Add everything up and divide by how many there are.
2 + 3 + 3 + 4 + 5 + 6 + 7 + 8 + 122 = 160, and 160 ÷ 9 = 17.78.
The median is the middle value once the list is sorted. Nine values sorted means the fifth one is the middle: four sit below it, four above.
2, 3, 3, 4, [5], 6, 7, 8, 122 — the median is 5.
The mode is the value that occurs most often. Here that is 3, which appears twice while everything else appears once.
So the average is 17.78. The average is also 5. The average is also 3. All three sentences are true and they are nearly twenty apart end to end.
Why the mean drifted
Eight of the nine values are below the mean of 17.78. That is not a paradox; it is what one large value does. The mean uses every number's exact size, so a single 122 pulls it a long way. The median only cares about position in the sorted order, so replacing that 122 with 1,220 would leave the median at 5 and send the mean past 130.
That is the trade. The mean is sensitive to every value including the strange ones. The median ignores how extreme the extremes are.
Which one to ask for
- Mean when you care about the total — budgets, capacity, anything you will multiply back up by the count. The mean times the count gives the total back exactly; no other average does.
- Median when you want a typical case out of a lopsided set. If the values have a floor but no ceiling — nothing below zero, but the top end trailing off — the mean sits above most of the data and the median does not.
- Mode when the values are categories rather than quantities, or when you want the most common outcome rather than a central one. The mean of a set of shoe sizes is a number nobody wears.
When someone quotes an average from a lopsided set without saying which, ask. If the mean and the median are close together the question does not matter much. If they are far apart, the choice was doing work.
The even-count median rule, which is a real bug
With an odd count there is a single middle value and no argument. With an even count there are two middle values, and the median is the mean of those two — not the upper one, not the lower one.
Drop the 122 from the set above and eight values remain:
2, 3, 3, 4, 5, 6, 7, 8
The two middle values are the fourth and fifth: 4 and 5. The median is (4 + 5) ÷ 2 = 4.5.
Take the upper of the two instead and you report 5. Take the lower and you report 4. Both are off, and both look completely plausible, which is why this off-by-one survives in spreadsheets and in code for years. It only ever misreports by the gap between the two middle values, so it is invisible on tightly packed data and wrong by a mile on sparse data.
A quick check: the median of an even-count list often is not a value from the list at all. 4.5 does not appear in the eight numbers above. If your median is always one of the original values, and your counts are sometimes even, you have the bug.
The mean of that eight, for comparison
2 + 3 + 3 + 4 + 5 + 6 + 7 + 8 = 38, and 38 ÷ 8 = 4.75.
Mean 4.75 against median 4.5 — close, because with the 122 removed the set is no longer lopsided. That closeness is itself the signal. Comparing the mean and the median is the cheapest test of skew there is, and it costs one subtraction.
Doing it without the paper
The free Average / Mean tool takes a list separated by commas or spaces and returns the mean, the sum and the count, in your browser, with nothing sent anywhere. It reports the mean only — it does not compute the median or the mode, so for those, sort the list yourself and count in from both ends. Having the sum and the count next to the mean is useful in its own right: it is the fastest way to confirm the tool read every number you meant to give it.