The Data Detective Summary and key ideas

The Data Detective, by Tim Harford.

Numbers can be abused, and that fact invites an appealing shortcut: assume a statistic is a trick and stop listening. But cynicism leaves us as vulnerable as gullibility. If every number is dismissed, we lose a way to see patterns too large or subtle for ordinary observation. The task is to ask how a statistic was made and what it can actually show.

A striking example is the relationship between storks and babies. Comparisons among countries found that more storks went with more births. The association was strong enough to pass a conventional publication threshold and was described as statistically significant. Yet larger countries have more people, so they have more births and more storks. Country size offers an alternative explanation for the pattern. The numbers show that the two variables move together; they do not show that storks cause births. A correlation can be accurate and impressive while the causal story built from it is wrong.

The rise in lung cancer presented a more consequential puzzle. Motorcars and road tar also became more common, offering a plausible explanation for the increase. To move beyond coincidence and speculation, Doll and Hill compared people with lung cancer to patients of similar sex and age at the same hospital. They asked about the patients’ histories rather than relying on anecdotes or a single proposed cause. Their initial study found that heavy cigarette smoking made lung cancer sixteen times more likely. Doll stopped smoking after seeing the result.

Doll and Hill then followed a much larger group of doctors. They contacted all 59,600 doctors in the United Kingdom, and more than 40,000 responded. Doctors made useful subjects for a long-term study because their smoking could be tracked and their causes of death diagnosed. Over time, the evidence supported the conclusion that smoking causes lung cancer, that greater consumption raises the risk, and that smoking also causes heart attacks. The case shows statistics used as patient investigation: define relevant groups, compare them, and follow outcomes long enough to learn more.

That evidence did not stop tobacco companies from trying to weaken it. Their executives challenged studies and funded distracting research. Rather than demonstrate that cigarettes were safe, they exploited unresolved questions to make the evidence seem less settled than it was. The book reports that, at a 1965 Senate hearing, Darrell Huff used the storks example to argue that the evidence linking smoking and disease was no more convincing than the idea that storks deliver babies. It also reports that the tobacco lobby had paid him. The example shows how a genuine warning about correlation can be turned into a tool for manufactured doubt.

Early COVID-19 presented a different problem: genuine uncertainty from incomplete data. During the crisis’s early stages, politicians faced urgent decisions while epidemiologists, medical statisticians, and economists worked with information that was patchy and inconsistent. Testing was scarce and concentrated among medical staff, critically ill patients, and the wealthy or famous. The available numbers could not reliably show how many mild or asymptomatic cases there were, or the virus’s true fatality. Decision-makers also had to weigh health risks against severe economic consequences. This early account describes open questions and plausible scenarios, not settled conclusions. Missing or uneven data made uncertainty real; it did not make statistics useless.

A first method is to ask where a number came from, who was counted, what it compares, and what alternative explanation could fit. Country size helps explain the storks-and-babies association. Doll and Hill’s matched comparisons and follow-up made their evidence about smoking more informative. When information is incomplete, say what cannot yet be known. Curiosity keeps the inquiry open, while skepticism asks that each claim stay within what the evidence can support.