When the average lies: mean, median and what spread tells you
A single average can hide almost everything about a data set. Here is when the mean misleads, why the median resists it, and what standard deviation adds.
Take five numbers: 1, 2, 3, 4 and 100. The mean is 22. Not one of the five values is anywhere near 22 — four sit below 5, and one sits at 100. The average is technically correct and practically useless, which is the whole problem with reporting a single number and calling it typical.
Mean and median answer different questions
The mean shares the total out equally: add everything, divide by the count. Every value influences it, so a single extreme number drags it a long way. The median is simply the middle value once sorted — it only cares about position, so that same extreme number moves it barely at all.
For our five numbers the median is 3. That describes the data far better. Swap the 100 for a million and the mean leaps past 200,000 while the median stays at 3, which is exactly the behaviour you want from a summary of what is typical.
When to use which
- Mean — when data is roughly symmetric and every value genuinely should count. Test scores, repeated measurements, heights.
- Median — when the data is skewed or contains outliers. Incomes, house prices, response times, anything with a long tail.
- Mode — when you want the most common value rather than a central one. Shoe sizes to stock, most-chosen options, categories where averaging is meaningless.
Two data sets, same average, nothing alike
Now consider two groups that both average exactly 50:
| Data set | Mean | Standard deviation | What it looks like |
|---|---|---|---|
| 48, 49, 50, 51, 52 | 50 | 1.58 | Tightly clustered |
| 10, 30, 50, 70, 90 | 50 | 31.62 | Widely scattered |
Identical averages, completely different realities. The first might be a well-controlled manufacturing process; the second, something barely under control at all. Standard deviation is what separates them — it measures how far values typically sit from the mean, in the same units as the data.
Reading a standard deviation
For roughly bell-shaped data, a useful rule of thumb: about 68% of values fall within one standard deviation of the mean, about 95% within two, and about 99.7% within three. So a mean of 50 with a standard deviation of 1.58 means nearly everything sits between roughly 47 and 53 — while the same mean with a deviation of 31.62 stretches from below zero to above 110.
That is also what makes a z-score useful: it re-expresses any value as the number of standard deviations it sits from the mean, so a test mark and a height measurement become directly comparable despite having nothing else in common.
Percentiles: the summary that survives outliers
Quartiles split sorted data into four equal parts. The gap between the first and third — the interquartile range — describes spread while ignoring the extreme ends entirely. Where standard deviation is inflated by a single wild value, the IQR simply does not notice it, which is why box plots and service-level targets tend to be built on percentiles rather than averages.
It is also why "average response time" is a poor service metric. A system answering most requests in 50 ms and a few in 30 seconds can post a respectable mean while a meaningful share of users have a terrible experience. The 95th or 99th percentile tells you what your unluckiest users actually see.
The practical takeaway
Never accept one number as a description of many. Ask for the median alongside the mean — if they differ noticeably, the data is skewed and the mean is misleading. Then ask for the spread, because two data sets with the same centre can behave in completely different ways.
Mean, Median & Mode Compare mean, median and mode on your own data Standard Deviation Measure spread, sample or population Percentile Calculator Find percentiles and the interquartile range