Five people walk into a café with annual incomes of $30,000, $40,000, $50,000, $60,000, and $820,000.
Their average income is $200,000. Four of them are about to learn that the café is full of surprisingly affluent people.
The arithmetic is correct. The impression may be terrible.
An average compresses a collection into a number. That is useful because we cannot carry every observation through every conversation. The trouble begins when the number is treated as a portrait of a typical person, a prediction for an individual, or a complete explanation of what is happening.
The average person is often a useful imaginary acquaintance. We should be careful about asking them to represent the whole neighborhood.
There is more than one middle
The arithmetic mean is the sum divided by the number of observations. In the café, the total is $1 million and there are five people, producing a mean of $200,000.
The median is the middle observation after ordering the values. Here it is $50,000. Neither number is fraudulent. They answer different questions about the same group.
The mean includes the size of every observation and is affected by an unusually large or small value. The median identifies the center of the ordered list without showing how far the extremes extend.
For these incomes, the median gives a more recognizable sense of the middle person. For another question, such as dividing the group’s total equally, the mean is exactly the quantity you need.
Do not replace one automatic habit with another. The median is not always the heroic truth hiding behind the suspicious mean. It can conceal important change at the edges. A group can have an unchanged median while its lowest and highest values move dramatically apart.
Ask what the statistic is intended to show, then choose the summary that fits that purpose.
Averages can hide the experience of waiting
Imagine a service desk with ten customers. Nine wait one minute; one waits thirty-one minutes. The mean wait is four minutes.
“Average wait: four minutes” sounds fairly comfortable. For the customer who waited thirty-one, it may sound like an organization describing another building.
A distribution helps. How many people waited under five minutes? What happened to the longest waits? Were particular requests delayed because they required a different specialist?
Percentiles can describe parts of that experience. A percentile indicates a point in the ordered observations according to a chosen calculation convention. It does not guarantee that a future individual will receive that outcome. Small samples and different methods can also produce slightly different percentile values.
For practical reporting, a few well-chosen details may be sufficient: the middle wait, the long-wait group, and the reason some cases differ. The objective is understanding, not a dashboard so elaborate it requires its own waiting room.
If a measure improves, check whether the improvement reached the people who had the hardest experience. Lowering the already short waits may move the average while leaving the main problem intact.
Your reference group may not be your group
An average is always an average of something. Which observations were included? Which were excluded? Over what period? Under what conditions?
A product’s average performance under controlled conditions may not describe its performance in your environment. An average cost across a country may not resemble the cost in your city. An average result among people who completed a program may omit those who left it.
These are not reasons to abandon statistics. They are reasons to read the denominator and the selection process.
Suppose a company reports an average satisfaction score from customers who responded to a survey. The number describes those responses. Whether it represents all customers depends partly on who answered and who did not.
Also distinguish an average of individuals from an average of group averages. If one group contains ten people and another contains a thousand, giving both group averages equal weight answers a different question from combining all 1,010 people.
The same numbers can be assembled into different summaries. The assembly instructions belong with the result.
Ask for enough texture to decide
When someone presents an average, try four questions. What kind of average is it? What is the group? How spread out are the values? What decision is this number helping us make?
The answers need not be complicated. A range, a simple chart, or two contrasting examples can reveal what one number conceals.
Avoid demanding every possible statistic before accepting any conclusion. That would make evidence unusable in a different way. Ask for the detail that could materially change your interpretation.
An average is a tool for reducing complexity. It becomes trustworthy when the important complexity remains available nearby.
Back in the café, nobody needs to accuse the $200,000 mean of dishonesty. It has performed its assigned calculation perfectly.
It simply should not order lunch for the other four people.




