“The average is 40” sounds informative until you discover that most observations are close to 10. A single summary can be correct while describing very few individual experiences.
The useful habit: ask which average is being reported and what the values look like around it.
Mean and median answer different questions
The arithmetic mean adds the values and divides by their count. The median identifies the middle of the ordered values. Both are measures of centre; neither describes the whole distribution.
Consider these fictional waiting times, in minutes: 10, 10, 15, 15, 100.
| Measure | Calculation | Result |
|---|---|---|
| Mean | (10 + 10 + 15 + 15 + 100) ÷ 5 | 30 minutes |
| Median | The third value after ordering | 15 minutes |
| Range | 100 − 10 | 90 minutes |
No one in this example waited exactly 30 minutes. That does not make the mean wrong. It shows that the mean is a calculated centre, not necessarily a value experienced by a typical individual.
See what an extreme value changes
Replace the 100-minute wait with 200 minutes. The mean becomes 50 minutes, while the median remains 15. The larger delay matters greatly for total waiting time, even though the middle observation does not move.
This is why neither measure should automatically win. If you are planning the total staff time needed to handle cases, the long waits matter. If you are describing the middle customer’s wait, the median answers a different useful question.
What happens with an even number of values?
For an even-sized numerical dataset, the usual median is the mean of the two middle values. In 10, 20, 30, 80, the median is (20 + 30) ÷ 2 = 25.
That median is not an observed value either. Also, when values repeat, it is too simplistic to say exactly half of observations are strictly above the median and half strictly below it.
Spread changes the meaning
Two fictional teams could both have a mean turnaround time of 20 minutes. One handles every case in about 20 minutes. The other alternates between very quick and very slow cases. The same centre hides different levels of predictability.
Ask for a range, quartiles or relevant percentiles alongside the average. For a service, a high percentile can help describe longer waits that the median alone leaves out. Read the definition and sample size rather than assuming a percentile guarantees a maximum.
Check who is included
An “average salary” may concern full-time employees, all employees, a particular occupation or a selected sample of respondents. Hourly pay and annual pay answer different questions. Gross and take-home pay are not interchangeable.
A change in the group can change its average even if no individual’s circumstances change. For example, adding more senior employees to a fictional workforce can raise its mean pay without giving existing employees a rise.
Do not casually average averages
Suppose one group of two people has a mean of 10 and another group of eight has a mean of 20. The combined mean is not 15. Weight by group size: (2 × 10 + 8 × 20) ÷ 10 = 18.
Without the group sizes, combining means can be misleading. Combining medians is even less straightforward; group medians and counts alone generally do not determine the overall median.
A checklist for the next average you read
- Is it the mean, median or another measure?
- What units, period and population does it cover?
- How many observations are included?
- Could extreme values or a changing group affect it?
- How widely are the values spread?
- Does the claim concern the whole group or an individual?
Note to keep: an average is a useful compression of information. Ask what that compression leaves out before using it to describe everyone.
Related: how charts mislead and percentages versus percentage points.
