# At Risk?

*By Bea Groves-McDaniel and SAL-9000*
*Research Assistant and Amanuensis: SAL-9000*
*11 November 2026*

## Two Words for One Thing

There is a section in a set of board papers I read recently headed 'Key risks'. It runs to sixteen paragraphs, and not one contains a probability.

That is worth saying at the outset, because the word 'risk' has two lives. The strict sense of risk is a quantified probability of an unwanted outcome, stated against an exposure and a baseline. The loose sense is an uneasy feeling, a category of thing that might go wrong, or a device for warming a room. Actuaries and safety engineers work in the strict sense; everybody else, most of the time, works in the loose one. Both uses are legitimate. What happens constantly, and almost never on purpose, is the substitution of one for the other while the vocabulary stays exactly the same.

The strict sense is not even straightforward. Sven Ove Hansson, in the Stanford Encyclopedia of Philosophy, notes that what psychologists call *risk perception* is a misnomer, since the underlying assessments are *"a matter of attitudes and expectations"* rather than perception. Daniel Pritchard, in the same discussion, offers a modal alternative to the probabilistic account: *"to say that a target event is risky is to say that (keeping relevant initial conditions for that event fixed) it obtains in close possible worlds"*. Even at its most technical, risk comes in two flavours, and the choice between them is a choice about what we are trying to find out.

I want to make a narrow claim. Most public argument about risk is not a disagreement about numbers. It is the substitution of the loose sense for the strict one, and the substitution does real work in the world, because the loose sense is cheaper and better suited to persuasion than the strict one ever will be.

## The Arithmetic of the Imagined

Daniel Kahneman's great contribution was to show that we do not assess risk with a calculator, even where one is available. System 1, the fast thinking process, does the assessing, and it does so rather badly: it judges likelihood by how easily examples come to mind, and it commits what Amos Tversky and he called the 'law of small numbers' (1974). Where the number of cases is small, the ratio we compute is unstable, and we have no way of knowing that it is unstable. A percentage struck from three cases looks exactly like a percentage struck from three thousand, though it means something quite different.

The example everybody reaches for is flying against driving. The official figures are instructive. The Department for Transport records 1,624 people killed on the roads of Great Britain in 2023, over 334 billion vehicle miles travelled, a rate of five fatalities per billion miles (Department for Transport, 2024). The International Civil Aviation Organization records that scheduled commercial air transport in 2023 produced one fatal accident and 72 fatalities, an accident rate of 1.87 per million departures and a fatality rate of 17 per billion passengers carried (International Civil Aviation Organization, 2024). Set those side by side and the plane looks better. Swap the denominators and the comparison tilts, because one counts vehicle miles and the other counts people carried.

That is not a quibble. It is the main point. The same body of facts supports opposite stories depending on the denominator. The denominator is precisely what gets dropped when a risk is asserted rather than measured. The vivid case also gets a vote that the mundane one never receives. A road death is a million unremarkable events; an air crash is one very remarkable issue.

The same arithmetic governs our sense of who is dangerous. Of those who had experienced rape or assault by penetration, including attempts, since the age of sixteen, over half (54.7 per cent) were assaulted by a partner or ex-partner (Office for National Statistics, 2025). The overwhelming majority of sexual violence is done by somebody the victim knows. The stranger is vivid, and the stranger is rare, and the two facts are traded against one another as though rarity were the same thing as probability.

The case that most deserves care is where a small number does enormous work. Consider the proposition that a transgender person is a grave risk to women in single-sex facilities. The legal position, asserted in both directions, is more interesting than either side makes it. Section 22 of the Gender Recognition Act 2004 makes it an offence for a person who has acquired protected information in an official capacity to disclose it, and it defines the protected information to include, where an application is still pending, the applicant's gender before it becomes their acquired gender. The Equality Act 2010 permits service providers to run single-sex services and to exclude those with the gender reassignment protected characteristic where that is justified. And in For Women Scotland Ltd v The Scottish Ministers [2025] UKSC 16 the Supreme Court held that the appointment of a trans woman who holds a certificate does not count towards a statutory target of fifty per cent women on a public board. The law is precise here, and it is not what either side says it is.

The statistics are where the law of small numbers earns its keep. On 31 March 2025 there were 339 transgender prisoners in England and Wales, with nine more prisoners known to hold a gender recognition certificate: 3.9 transgender prisoners per 1,000 prisoners (HM Prison and Probation Service, 2025). These are very small numbers, and smallness is precisely the condition under which people over-read them. The rate is stable enough to be quoted and unstable enough to be abused This is a very dangerous combination. The honest response is neither dismissal nor amplification. It is to ask what the number is a rate of, and then to ask what comparison would make it meaningful. Three point nine per thousand is not, on its own, a finding about anything.

There is a general logical error here, and it does not belong to one side of any political divide. It runs: one person in a category has committed a serious crime; therefore the category is a grave risk. The first step is often legitimate, since it may reveal that the category contains a feature that raises the probability. The second is a category error, because it converts a proposition about individuals into a property of a group and then treats the group as a thing with a rate. Rates belong to populations defined by exposure, not to labels.

This is where risk becomes a weapon. Unrealistic risks, blown up out of all proportion, are the raw material of aggressive politics, because they work so well. The UK Government published 404 pieces of guidance in response to a call for input on single-sex spaces, and found that around ten per cent appeared to misread the Equality Act, in a variety of ways, including by failing to acknowledge that the Act allows providers to exclude those with the gender reassignment protected characteristic where justified (Office for Equality and Opportunity, 2024). When professional guidance is confused about a rule that is settled in the statute, the confident demagogue has an easy job.

## Asking What the Number Is a Rate Of

The remedies are procedural, and I would offer four.

- Ask what the number is a rate of. A figure without a denominator, baseline, comparison and time period is an impression with decimal points.

- Keep the two senses apart in your own speech. If you are frightened, say that you are frightened. If you are claiming a probability, produce the probability.

- Do the denominator work deliberately. The same facts support opposite stories under different denominators, and the choice of denominator is a choice about the argument rather than about the world.

- Distrust precision that comes from small numbers. That is the law of small numbers turned on oneself, and it cuts in every direction.

- And learn the law, where law exists. Nobody holds an informed view of a risk created by a statute they have not read. The most instructive risk argument I read this year was a judicial one, and it turned on the difference between a person's sex and their acquired gender, and on who holds the statutory authority to decide which counts for what.

There is a reason the loose sense persists. Stories are cheap and travel further than dull statistics. Our public life runs largely on the former kind, and the colloquial vocabulary lets us pretend the story and the stat's are the same thing.

## Coda

SAL-9000 researched and verified every source in this essay, and carried out the statistical and legal checking. The argument, its weaknesses and its consequences are mine, and the reader should hold me to them.

## Bibliography

Department for Transport (2024) *Reported road casualties in Great Britain, annual report: 2023*. London: Department for Transport. Available at: https://www.gov.uk/government/statistics/reported-road-casualties-great-britain-annual-report-2023/reported-road-casualties-great-britain-annual-report-2023 (Accessed: 11 November 2026).

Equality Act 2010, c. 15. London: The Stationery Office. Available at: https://www.legislation.gov.uk/ukpga/2010/15/contents (Accessed: 11 November 2026).

For Women Scotland Ltd (Appellant) v The Scottish Ministers (Respondent) [2025] UKSC 16. London: The Supreme Court. Available at: https://supremecourt.uk/cases/judgments/uksc-2024-0042 (Accessed: 11 November 2026).

Gender Recognition Act 2004, c. 7. London: The Stationery Office. Available at: https://www.legislation.gov.uk/ukpga/2004/7/contents (Accessed: 11 November 2026).

Hansson, S.O. (2022) 'Risk'. *The Stanford Encyclopedia of Philosophy*. First published 2007; substantive revision 8 December 2022. Available at: https://plato.stanford.edu/entries/risk/ (Accessed: 11 November 2026).

HM Prison and Probation Service (2025) *Offender Equalities Annual Report 2024-25*. London: Ministry of Justice. Available at: https://www.gov.uk/government/statistics/hmpps-offender-equalities-annual-report-2024-to-2025/hmpps-offender-equalities-annual-report-2024-25--2 (Accessed: 11 November 2026).

International Civil Aviation Organization (2024) *Safety Report 2024 edition*. Montreal: International Civil Aviation Organization. Available at: https://www.icao.int/sites/default/files/sp-files/safety/Documents/ICAO_SR_2024.pdf (Accessed: 11 November 2026).

Kahneman, D. (2011) *Thinking, Fast and Slow*. New York: Farrar, Straus and Giroux.

Office for Equality and Opportunity (2024) *Response to call for input on single-sex spaces guidance*. London: GOV.UK. Available at: https://www.gov.uk/government/publications/response-to-call-for-input-on-single-sex-spaces-guidance/response-to-call-for-input-on-single-sex-spaces-guidance (Accessed: 11 November 2026).

Office for National Statistics (2025) *Sexual offences in England and Wales overview: year ending March 2025*. Newport: Office for National Statistics. Available at: https://www.ons.gov.uk/peoplepopulationandcommunity/crimeandjustice/bulletins/sexualoffencesinenglandandwalesoverview/yearendingmarch2025 (Accessed: 11 November 2026).

Pritchard, D. (2016) 'Epistemic risk'. *The Journal of Philosophy*, 113(11), pp. 550–571.

Tversky, A. and Kahneman, D. (1974) 'Belief in the law of small numbers'. *Psychological Bulletin*, 76(2), pp. 105–110.
