The tyranny of the average

“Eighty per cent positive.” It looks good at the top of a report. The number fits in a box, is easy to compare with last year and gives the meeting something to start with.

Now imagine two teams of fifty people each. In one, every respondent is positive. In the other, sixty per cent are. Together, the figure is eighty per cent. It is mathematically correct and a very poor description of what the second team is experiencing.

That is the trouble with a headline average. It can make very different stories look like one story. The number has done its job as a summary. The danger is asking it to do the job of an explanation.

The average is often where the analysis begins, not where the decision should end.

First, which “average” do we mean?

People use the word loosely. A mean is calculated by adding values and dividing by the number of values. A median is the middle value when they are ordered. An employee survey headline may instead be a proportion, such as the percentage of respondents who agree with a statement.

These are different calculations. They share a problem when used alone: they compress many answers into one result. A median can be more useful than a mean when a few very high values pull the mean upwards. The ONS has used the difference between mean and median to explain why growth in average productivity may not describe a typical firm.

Choosing the right calculation matters. Looking at the spread matters too. Even a well-chosen centre cannot tell you whether nearly everyone is close to it or whether two groups sit far apart.

The headline can improve while somebody’s experience worsens

Suppose an organisation’s engagement score rises. That is worth noticing. But which parts of the organisation improved? Did a struggling team recover, or did an already strong team get stronger while another slipped?

The overall figure can also change because the people who answered changed. If response is much lower in one area this year, a year-on-year comparison needs care. You might be seeing a shift in sentiment, a shift in who took part or a little of both.

None of this makes the headline useless. It makes the next question important. Ask where the change occurred, whether it is broad enough to trust and what other evidence might help explain it.

There is a limit to how far you should slice

Once somebody sees a gap between teams, the temptation is to keep splitting the data. Department, location, age, length of service, manager, working pattern. Eventually you can find a striking percentage for almost any tiny group.

A result based on a handful of answers can swing sharply when one more person responds. It may also reveal more about individuals than a confidential survey should. The answer is not to publish every cut with a footnote asking readers to be careful.

Set and respect a minimum reporting threshold. Consider whether a combination of filters could let someone infer a person’s answer even when each table appears large enough on its own. For small groups, look for a safer level of reporting or a different way to listen.

There is a balance here. Hiding every group difference protects nobody if it also conceals a real problem. Publishing highly specific numbers can do harm of its own. Sound analysis needs both curiosity and restraint.

Comments add texture, not a second census

Open comments can explain why a number matters. They can show what people mean by “workload”, describe a decision that felt unfair or point to an issue the questionnaire did not ask about.

They are not a representative vote by themselves. Some people write a great deal; others say nothing. One memorable quote can dominate a room long after the quieter pattern in the data has been forgotten.

Use comments to test possible explanations, find examples and sharpen the next question. Do not take the most vivid sentence and present it as what everyone thinks. The purpose is to understand the range of experience without pretending every form of evidence is interchangeable.

Averages conceal variation. Anecdotes can exaggerate it. The job is to keep both in proportion.

What should a leader ask when the slide appears?

Start with the number of people behind the percentage. Ask how they answered, not just how many fell on the positive side of a line. Look for meaningful differences across groups where reporting is reliable and confidential.

Then ask what decision the result is meant to inform. If the whole organisation has a problem with clarity, an organisation-wide response may make sense. If the problem is concentrated in a particular process or location, a broad campaign may miss the people who need the change.

Finally, ask what you do not know. A survey can identify a pattern without explaining every cause. A careful report can say “this deserves investigation” without feeling that it has failed to produce an answer.

Keep the number, lose the illusion

We need summaries. Nobody can make a decision from thousands of individual answers spread across a table. An overall measure is useful because it gives us a place to begin, a way to track movement and a common language for a conversation.

It becomes tyrannical when it insists that everybody in the group must be having something like the average experience.

The more useful habit is simple: when a number looks unusually neat, ask whose story it has made harder to see. That is often where the real work of understanding begins.