Why some risks frighten more than their numbers warrant, and when that is defensible

The ranking does not follow the rates

Ask people to order hazards by how dangerous they are and the result correlates only loosely with any measure of expected harm. Some hazards that cause few deaths generate sustained alarm; others that cause many are accepted with little comment. The gap is large, stable and well documented, and it is usually the opening move in an argument about whether the public is irrational.

That argument goes wrong at the first step, because it assumes the ranking was meant to be a ranking by expected harm. When people are asked how risky something is, they answer a broader question, and a good deal of research has gone into working out what that broader question contains.

The dimensions that predict the response

The research tradition here is descriptive: rate many hazards on many attributes, then look at which attributes cluster and how well the clusters predict concern. The recurring picture is two main dimensions. One is roughly dread — whether the harm is uncontrollable, catastrophic, involuntary and inequitably distributed. The other is roughly unfamiliarity — whether the hazard is new, unobservable, and its effects delayed.

These are factor structures derived from ratings rather than causal accounts, and they should be read as a compact description of what people are responding to rather than as a mechanism. What they establish reasonably well is that the response is systematic. It is not noise, and it is not ignorance of the numbers, because the pattern holds even among people who have the numbers.

Some of the response is a value judgement

Several of those attributes are legitimate grounds for treating two hazards differently even when the expected harm is identical. Whether the exposure was chosen. Whether the person exposed also receives the benefit. Whether the harm falls on people who had any say in the decision. Whether it can be undone. None of these is a probability, and none of them is a mistake.

A person who accepts a larger risk they chose over a smaller one imposed on them is not miscalculating. They are applying a view about consent and fairness that most ethical frameworks endorse, and calling that an error simply because it does not maximise a single quantity is a substantive philosophical claim being smuggled in as arithmetic.

Some of it is a straightforward mistake

The other component is real error, and it deserves to be named as clearly. People misestimate frequencies, in a pattern that follows what is reported and memorable rather than what is common. Rare dramatic causes are overestimated and common undramatic ones underestimated, and this has been shown often enough to be treated as established.

The two components are separable in principle and get mixed together in practice, which is what makes these disputes so unproductive. An argument that begins by correcting somebody’s frequency estimate and then implies their values are also wrong will fail at both, because the second claim contaminates the first.

The catastrophic case deserves more respect than it gets

One dimension is worth defending specifically. People treat a single event harming many at once as worse than the same total spread across many separate events, and this is often cited as a plain irrationality.

There are reasons why it need not be. Concentrated losses fall on connected people, so a community can lose its capacity to function rather than merely losing individuals. Large simultaneous events overwhelm response capacity in a way distributed ones do not. And they carry information about the system that produced them, since something capable of failing at that scale can plausibly do it again. Whether these considerations justify the size of the difference in concern is arguable. That they justify some difference is hard to deny.

Separating the strands in an actual argument

The productive move in any dispute of this kind is to ask which claim is being made. Is the disagreement about how often the harm occurs, which is a factual question with an answer, or about how bad a given quantity of harm is in this form, which is a question about values that evidence cannot settle?

Most public arguments about risk contain both and label neither. Splitting them does not guarantee agreement. It does mean the factual half can be resolved with evidence, and the remaining disagreement can be conducted as what it actually is, which is a disagreement about consent, fairness and what people are willing to accept.