What Kind of Average? Mean, Median, and the Shape of the Data

“The average is 20” is incomplete unless the reader knows how that average was calculated and what observations it summarizes. An arithmetic mean, a median, and a mode can all be described casually as an average, yet they can give different answers for the same dataset.

The difference is especially visible when a few observations are much larger than the rest. It is also important when combining group averages: the size of each group affects the combined mean, and a combined median generally cannot be recovered by averaging the medians.

Five waiting times, two useful summaries

Consider these invented waiting times, in minutes: 2, 3, 4, 5, and 26. The arithmetic mean is the sum divided by the count: 40 ÷ 5 = 8 minutes. The median is the middle observation after sorting: 4 minutes.

Neither calculation is a mistake. The mean incorporates the full numerical size of the long wait. The median identifies the central position in the ordered list. In this example, four of the five waits are shorter than the mean.

If the 26-minute wait becomes 56 minutes, the mean rises to 14 minutes while the median remains 4. The median is less affected by that extreme value, but the extreme wait still occurred. Ignoring it may be inappropriate if the question concerns service reliability or the experience of people who wait longest.

A median is not the midpoint of the endpoints

For 2, 3, 4, 5, and 26, the midpoint between the smallest and largest values is 14. That is the midrange, not the median. The median depends on the ordered observations, not just the endpoints.

With an even number of observations, a common numerical median convention averages the two middle values. For 2, 3, 5, and 26, the median is (3 + 5) ÷ 2 = 4. The result need not be an observed waiting time. A mean need not be an observed value either.

The mode describes the most frequent value or, for grouped continuous data, a prominent region of the distribution. A dataset can have more than one mode or no repeated exact values. Calling it the “typical” value still requires explaining what typical means for the question.

Group sizes change a combined mean

Suppose one class has ten students with a mean score of 80, while another has thirty students with a mean of 60. Averaging 80 and 60 gives 70, but that gives each class equal weight despite their different sizes.

The combined total is 10 × 80 + 30 × 60 = 2,600 points across forty students. The student-weighted mean is therefore 65. The 70 figure answers a different question: the equal-weighted average of the two class means.

Both can be calculated, but their labels must identify the unit being weighted. Are we averaging classes or students? Cities or residents? Stores or transactions? The choice changes the result and its interpretation.

Medians cannot usually be combined the same way

Knowing each group's median and size does not reveal all the underlying observations. Two groups can have identical medians yet very different distributions around those medians. Their combined median depends on where all the observations fall when ordered together.

For that reason, averaging neighborhood median incomes does not generally produce the citywide median household income. The raw or suitably detailed distribution, or an officially calculated combined estimate, is needed. A spreadsheet formula can return a neat number without making it the statistic named in the heading.

The same average can hide different experiences

Two invented sets, 9, 10, 11 and 0, 10, 20, both have mean 10 and median 10. Their spread differs considerably. A summary of reliability, variability, or inequality needs more than the central value.

NIST describes distributions in terms of location, spread, and shape. A histogram, range, interquartile range, or other appropriate measure can reveal information that a single average omits. The useful choice depends on the data and purpose; no one summary is universally best.

When an overall average moves in the opposite direction

A changing mix can reverse the impression from subgroup results. Consider a fictional service with a quick task and a complex task. In Period One, it completes ninety quick tasks averaging 2 minutes and ten complex tasks averaging 20 minutes. The overall mean is (90 × 2 + 10 × 20) ÷ 100 = 3.8 minutes.

In Period Two, quick tasks improve to a mean of 1 minute and complex tasks improve to 18 minutes. But the service now completes fifty of each. The overall mean is (50 × 1 + 50 × 18) ÷ 100 = 9.5 minutes.

Both task types became faster while the combined average became slower. The mix shifted toward complex work. A statement that “performance worsened because average time increased” would miss that change in composition.

This example does not prove that a real service's slower average is harmless. It shows what must be investigated before attributing the change to slower work. The categories, their counts, and the within-category results provide information that the overall mean cannot.

The same principle applies to prices. A store's average transaction can rise because prices increased, because customers bought more items, or because they bought a different mix. A payroll mean can change because employees' pay changed or because the composition of the workforce changed. A responsible explanation identifies the unit of observation and checks the mix.

Rates require an appropriate weight too

Suppose a vehicle travels equal distances at 30 and 60 kilometers per hour. Its average speed over the whole journey is not 45 km/h, because it spends more time on the slower segment. Over two 60-kilometer segments, it takes two hours plus one hour, covering 120 kilometers in three hours: 40 km/h.

If instead it spends equal time at each speed, the average is 45 km/h. The same two speeds produce different averages because the averaging condition differs. State whether time, distance, observations, or groups receive equal weight before selecting a formula.

Read the definition before drawing the conclusion

When a report says an average increased, check whether the calculation, population, weighting, and time period stayed comparable. A changing mix of people or transactions can move the average even if no individual subgroup changed in the same way.

Then ask whether the summary answers the reader's actual question. A mean can help estimate a total when multiplied by the corresponding count. A median can describe the center of an ordered distribution. Neither guarantees that a particular person's experience will resemble it. Naming the statistic and showing enough of the distribution turns “average” from an impression into a reproducible claim.