Here’s a puzzle you may not have thought about.
On the same afternoon, millions of people do two things that seem completely contradictory. They buy insurance against rare disasters — house fires, cancer, flight accidents, theft. And they buy lottery tickets, or throw money into a scratchy, or enter competitions with odds so slim the number is practically zero. The first behaviour seems cautious. The second seems reckless. But they’re being performed by the same people, with the same bank accounts, often in the same shopping trip.
This pattern looks contradictory only if you assume humans price risk the way economists assumed they did for most of the twentieth century — as neutral calculators, multiplying probability by outcome and choosing whichever option has the best expected value. We are not that. We are something stranger and more interesting.
The finding that changed how economists think about risk
In 1979, the psychologists Daniel Kahneman and Amos Tversky published a paper called Prospect Theory: An Analysis of Decision Under Risk. It would go on to become one of the most cited papers in the history of economics. One of its central findings was that humans systematically overweight small probabilities — at both ends.
To understand why this matters, consider how a purely rational calculator would treat risks. If an event has a one-in-a-hundred-thousand chance of happening, a rational calculator would weight it one-hundred-thousandth of the corresponding gain or loss. A one-in-ten-thousand event would get ten times that weight. A one-in-a-thousand event would get a hundred times. And so on, in clean proportion.
What Kahneman and Tversky found, in experiments with real people, was that the clean proportion breaks down at the extremes. Once probabilities become very small — less than about five per cent — people stop weighting them proportionally and start responding to them as categories. A one-in-ten-thousand risk and a one-in-a-million risk are both, functionally, treated as roughly “possible but very unlikely”. A one-in-a-hundred chance of winning and a one-in-ten-thousand chance often feel, subjectively, within shouting distance of each other.
This is why the insurance buyer and the lottery player are the same person. Both behaviours are driven by overweighting small probabilities. The insurance buyer overweights the small chance of catastrophe. The lottery player overweights the small chance of triumph. Neither is being irrational exactly — they’re just running on a brain that was never calibrated to price one-in-a-million outcomes.
The doctor who misread the test
A more worrying version of this plays out in medicine. The German psychologist Gerd Gigerenzer has spent decades showing that even highly trained professionals routinely misread probabilistic information.
His favourite demonstration works like this. Imagine a disease affects one per cent of the population. A test for the disease is ninety per cent accurate. Your test comes back positive. What’s the chance you actually have the disease?
Most people guess around ninety per cent. Most doctors, in Gigerenzer’s studies, also guess around ninety per cent. The correct answer is about ten per cent.
The reason is that the one per cent base rate — the prior probability of having the disease — swamps the test’s accuracy when you work through the maths. In a population of a thousand people, ten would have the disease (one per cent). Of those ten, nine would test positive (ninety per cent accuracy). But of the nine hundred and ninety healthy people, ten per cent of the tests come back falsely positive — which is ninety-nine false positives. So out of a hundred and eight positive tests, only nine are real. That’s about ten per cent.
Gigerenzer’s point isn’t that doctors are bad. It’s that probabilistic information, presented in the way we usually present it — as percentages — is almost uniquely difficult for humans to process. His research found that when the same information is presented as natural frequencies (9 out of 10 with the disease, 99 out of 990 without, and so on) the error rate drops dramatically. Doctors suddenly answer correctly. So do ordinary people. The difficulty wasn’t in the minds. It was in the presentation.
This has implications for almost every decision involving probability. Medical screening. Legal evidence. Financial risk. Insurance premiums. Weather reporting. Public health advice. When the information is given as percentages or rates, most people — including trained professionals — will misread it in predictable ways.
The policy angle
The legal scholar Cass Sunstein has applied this research to public policy, coining the phrase probability neglect. His observation is that when the public responds to risks — terrorism, shark attacks, rare diseases, nuclear accidents — the intensity of response is often driven by how vivid the risk is rather than how likely it is. Politicians and media magnify the effect.
After the September 11 attacks, Sunstein notes, many Americans avoided flying for a year or more, driving instead — which is, statistically, considerably more dangerous per kilometre. Researchers including Gerd Gigerenzer have estimated that the additional road deaths in the United States in the year following the attacks exceeded the death toll of the attacks themselves, by a large margin. The public had not become more rational about risk; they had just shifted the same fear from an emotionally salient category (flying) to a less emotionally salient one (driving).
The same pattern turns up everywhere. Parents worry more about stranger abduction than about car accidents, though the latter is vastly more common. Homeowners insure against house fires more eagerly than against the slow damage of poor maintenance, though maintenance problems will cost them more on average. Investors fear market crashes more than the quiet erosion of inflation on cash savings, though inflation has taken more wealth from more Australians than any crash.
The evolutionary angle
One reason we’re so bad at pricing small probabilities may be that we were never built to be good at it. The psychologist Martie Haselton has developed what she calls error-management theory, which argues that evolution designs minds not to be accurate in the abstract, but to minimise the more costly kind of error in a given environment.
For our distant ancestors, the cost of missing a genuine threat — a predator, a hostile outsider, an approaching storm — was often death. The cost of falsely perceiving a threat that wasn’t there was merely a few seconds of wasted energy. A mind calibrated to run from rustles in the grass, even when the rustle turned out to be the wind, outlived a mind that carefully computed the probabilities. You inherit the genes of the jumpy.
This is why our risk responses feel so disproportionate to the actual data. Our ancestors didn’t need actuarial tables. They needed fast heuristics that erred strongly on the side of catastrophe-avoidance. Those heuristics still run in us, poorly adapted to a world where most of our risks are slow, cumulative and statistical rather than sudden, vivid and acute.
The counter-view worth hearing
Not every researcher is comfortable framing all of this as error. Gigerenzer himself — the same researcher whose work on medical statistics featured above — has spent much of his career arguing that what behavioural economists call “biases” are often adaptive heuristics that work remarkably well in the environments they evolved for.
His argument is roughly this: in a world of genuine uncertainty, where probabilities aren’t clean, where evidence is incomplete, where you have limited time and cognitive resources, the heuristics humans use often outperform formal probabilistic reasoning. A firefighter who senses, without consciously calculating, that a building is about to collapse is doing something a probability calculator could not have done in time. A doctor whose intuition correctly flags a patient as more ill than the tests show is often reading signals that no formal risk calculation could capture.
So the honest picture is probably this: humans are badly calibrated for certain kinds of precisely statistical risks, especially at very low probabilities and in unfamiliar domains. But the same mental shortcuts that fail us in those contexts serve us well in others — often the ones that matter most urgently. The skill worth developing is not becoming a human probability calculator, which you can’t. It’s learning to recognise when your intuitions about risk are in a domain they were built for, and when they aren’t.
Sanity-checking your own sense of risk
A few small practices, drawn from this research, can help when a risk is making a large decision for you.
Ask yourself: What’s the base rate? Before reacting to a specific risk, find out how common it actually is in the general population. This is the single most effective debiasing move, and one of the least practised.
Convert percentages to frequencies. Ninety per cent accurate becomes nine in ten. A one-in-a-million risk becomes if we lined up a million people. Your brain handles frequencies much better than it handles ratios.
Ask whether the risk you’re responding to is vivid or common. Often they’re not the same thing, and the vividness is doing much of the emotional work.
And finally, notice which catastrophes you insure against and which you ignore. The ones you insure against are probably the ones your mind finds it easy to imagine. The ones you ignore may be the ones that are more likely to actually occur.
The question worth carrying is one that sounds simple but rewards sustained thought:
Of the things you fear, how many could you actually estimate the probability of, and of the things you don’t fear, how many have you just never thought to picture?
Key research referenced: Kahneman and Tversky’s prospect theory (1979) and their work on probability weighting; Gerd Gigerenzer’s research on risk literacy and natural frequencies; Cass Sunstein on probability neglect (Risk and Reason, 2002); Martie Haselton’s error-management theory.