Administrative data
Photo: Kampus Production / Pexels
When a record is not a person
Administrative data — the records kept by government systems as a by-product of delivering services — offer a count that surveys cannot: near-complete coverage of everyone who has passed through a particular process. For disability statistics, the most prominent examples are the claimant counts produced from Personal Independence Payment and its predecessor Disability Living Allowance, the records from the Employment and Support Allowance caseload, and — since 2013 — the Universal Credit statistics published by the Department for Work and Pensions.
These figures are published regularly, carry large denominators, and do not suffer from the sampling uncertainty that qualifies survey-based estimates. That precision is real. What it measures, however, is something narrower than "disabled people."
A claimant count is a count of people who applied, were assessed, and were awarded a particular benefit under the criteria in force at the time of assessment. Change the assessment criteria — as happened repeatedly during the transition from DLA to PIP, which began in 2013 — and the count changes, not because the underlying population changed but because the instrument changed. The resulting series is not comparable year-on-year without careful qualification, and DWP's own statistical releases carry explicit warnings to that effect.
Photo: Mikhail Nilov / Pexels
The coverage problem runs in both directions. Administrative data includes people not captured by household surveys — those in residential care, for instance, are outside most survey frames but may appear in benefit records. At the same time, many disabled people, as defined by the Equality Act 2010 or identified by survey self-report, are not benefit claimants: they are employed, do not meet the specific benefit criteria, or have not applied. The claimant count is therefore neither a floor nor a ceiling for the disabled population — it is a different shape entirely.
A further complication is that administrative records capture the moment of assessment, not a stable characteristic. Someone awarded PIP at a given award level may deteriorate or improve; the record reflects the award, and periodic reassessment produces its own discontinuities in the series.
None of this makes administrative data less valuable. Claimant counts are, for some questions — geographic distribution of claimants, award rates by impairment type, transition between benefit categories — the only source available at the required granularity. The discipline required is to read them as what they are: counts of people in a system, which is not the same thing as a count of people.
