How to read a headline number

A number in a headline has usually passed through four hands: the study, the paper, the press office, and the outlet. Each hand is doing a legitimate job, and each one compresses. By the time it reaches a reader, the figure is accurate and nearly context-free.

Who was in the study

The first and most useful question is who the result happened to. Trial populations are selected. They usually exclude people with certain other conditions, sometimes exclude broad age bands, and always exclude anyone who did not want to join a trial.

That is not a flaw — it is how a controlled study achieves control. But it means the honest reading of any result is narrower than the headline: this happened, on average, to people like the ones who were enrolled.

Average is doing a lot of work

An average result is compatible with an enormous spread underneath it. A reported mean tells you where the middle sat; it tells you nothing about how wide the distribution was, and the distribution is where an individual reader actually lives.

The average is a statement about a group. It is never a prediction about a person.

Compared with what

• Compared with nothing at all, compared with a placebo, or compared with an existing treatment? These produce very different-looking numbers from the same underlying effect.
• Over what period? A result at twelve weeks and a result at two years are not interchangeable, and the shorter one is far cheaper to produce.
• Relative or absolute? A change from two cases per thousand to one per thousand is a fifty per cent reduction and a one-in-a-thousand difference. Both descriptions are true.

Who paid, and does it matter

Funding disclosure is worth reading and worth not over-reading. Industry funding does not make a study wrong; a great deal of good research is funded by people with an interest in the answer. It is a reason to check the design more carefully, not a reason to discard the result.

The general posture that survives all of this is unglamorous: treat a single figure as an invitation to ask what produced it, rather than as a fact you now know. Most of what looks like contradictory health news is two accurate numbers from two different questions.