Why they diverge

Three instruments measure disability in the UK. They do not agree, and they are not supposed to.

Three questions, three answers

The legal test, the survey question and the benefits assessment each draw a boundary in a different place. The Equality Act 2010 asks whether a physical or mental impairment has a substantial and long-term adverse effect on the ability to carry out normal day-to-day activities — a threshold filtered through legal interpretation. The major continuous surveys ask respondents whether they have a long-standing illness, disability or impairment that reduces their ability to carry out day-to-day activities: self-identification, with no external verification. The Personal Independence Payment assessment applies a points-based functional test administered by a healthcare professional against a prescribed descriptor set.

Three instruments, three populations — and the populations they define do not map onto each other. The survey count tends to be larger than the benefit-receipt count, because not everyone who self-identifies as disabled claims a benefit, and not everyone who claims a benefit considers themselves disabled in the everyday sense of the word. The legal population is, in principle, broader still: entitlement under the Act does not require a claim or a survey response.

The divergence is structural, not a measurement error. Each instrument was designed for a purpose: the statute to set a threshold for protection from discrimination; the survey to track prevalence and the employment and pay gaps that depend on it; the assessment to determine financial entitlement. Alignment was never the goal. A policy debate that treats the three as interchangeable — or that uses whichever count is most convenient — misrepresents what each one measures.

The practical consequence is visible in who publishes what: different bodies report from different instruments on different schedules, which is why headline figures quoted in the same week can differ by millions. The gap between them is not noise. It is information about what the question was.