The wording decides the number

A single word substituted in a survey question can shift the disability count by millions. This is not measurement error — it is measurement.

The question is the instrument

Every published disability prevalence figure traces back to a question, and every question contains choices: which conditions count, how long they must have lasted, whether the respondent or a clinician decides, and whether the focus is the condition or its effect on daily life. Change any of those choices and you change the count. The mechanism is not subtle and it is not a flaw — it is how survey measurement works. The number the headline carries is inseparable from the words that produced it.

The clearest demonstration comes from duration thresholds. The Annual Population Survey and the Family Resources Survey have both used variants of a "long-standing illness, disability or impairment" filter at different points in their histories. "Long-standing" is not defined for the respondent; it invites a judgement call. Replace that phrase with "a condition expected to last, or which has already lasted, 12 months or more" and you impose a concrete threshold. People with conditions they regard as permanent but which may not yet have reached a year, and people uncertain whether their condition qualifies as "long-standing", make different routing decisions under the two wordings. The resulting counts differ — and the difference is not trivial.

The Equality Act 2010 definition requires that an impairment has a substantial and long-term adverse effect on day-to-day activities, with "long-term" meaning has lasted, or is likely to last, at least 12 months. Survey designers working after 2010 had reason to align their duration language with that statutory threshold. But alignment is not identical to equivalence: a survey respondent self-assessing against a question is not the same instrument as a tribunal applying a legal test. The words may converge while the populations they capture continue to diverge.

What moves when the wording moves

When the Health Survey for England revised its question sequencing in ways that changed how the disability filter was applied, prevalence estimates shifted. The Office for National Statistics has documented cases in which changes to question order, routing logic or the specific terminology of a filter question produced step changes in the series — breaks that are real in the data but reflect the instrument rather than any change in the population. The same individuals, surveyed under two different wordings in the same week, would route differently through the questionnaire.

The effect is largest at the margin: people with moderate, fluctuating or recently acquired conditions. Someone with a condition of ten months' duration reads "long-standing" differently from "expected to last 12 months". Someone with well-controlled diabetes may regard their condition as managed rather than limiting, and answer accordingly — or not, depending on whether the question asks about the condition or its effect. The self-report mechanism means the question is doing active cognitive work on the respondent, not merely recording a fact that exists independently of it.

This is why comparability over time is so difficult to establish. A wording improvement — one that better captures the intended concept — necessarily breaks the series. Analysts cannot have both precision and continuity; they must choose one or document the discontinuity carefully and publish both old and revised series in parallel where possible. The Office for National Statistics and the Department for Work and Pensions have handled such breaks differently at different times, and the degree of transparency in how a break is signalled varies across bulletins.

Why it matters for the published figures

Readers who cite a disability prevalence figure without naming the question that produced it are citing something incomplete. "Around a fifth of the UK population is disabled" and "around a quarter" are both defensible depending on the instrument, the year and the precise wording deployed. Neither is wrong in isolation; both are meaningless without the source instrument attached.

The practical implication for anyone reading a statistical bulletin is straightforward: find the filter question, note the duration criterion, check whether the question asks about conditions or limitations, and look for any methodological note flagging a break in series. The number comes last. The wording comes first.

Threshold is an independent publication about disability statistics and governance. It does not provide advice, guidance or information to individuals about benefits, entitlements, legal rights or services of any kind.