Annual Deprivation Index (ADI) — Neighbourhoods (LSOA)
Autonomy Data Unit, Autonomy Institute
https://adi.apps.autonomy.work
Dataset release: 2026-09-04

CONTENTS
  adi-lsoa-employment.csv       Claimant Count, Nomis NM_162_1
  adi-lsoa-crime.csv            Recorded street crime plus separate ASB incidents
  adi-lsoa-health.csv           Modelled QOF prevalence and two coverage indicators
  adi-lsoa-data-dictionary.csv  Metric/indicator definitions, units, sources and availability
  adi-lsoa-geography.csv        Current LAD and region codes/names for each area

SHAPE AND GEOGRAPHY
  Each domain CSV is UTF-8, long by year, and contains exactly 404,988 data
  rows: 33,749 unique areas x 12 years (2014-2025). `code`, `name` and
  `year` identify a row. Missing values are empty fields; numeric zero is a
  published value, not the missing-value marker (and may reflect source rounding).

  LSOAs use 2021 boundaries and are rolled up to 2025 local authority district
  (LAD) and region boundaries. This release contains 33,749 of England's 33,755
  2021 LSOAs; six complex 2011-to-2021 boundary-change LSOAs are excluded. The
  geography CSV supplies the current parent codes and names needed to select a
  council or region. Higher-level `pop` values sum the 33,749 included LSOAs.

POPULATION, COUNTS AND RATES
  `pop` is the ONS mid-year population estimate for all ages (Nomis NM_2014_1,
  2021 LSOA vintage) for that area and year. It is not an adult or working-age
  denominator. The 2025 value repeats 2024 because the source series ends in 2024.

  Each deprivation metric is a three-column group:
    `<count>`       the metric count or modelled count
    `<count>_pop`   the population covered by that count
    `<count>_rate`  round(`<count>` / `<count>_pop`, 8)

  Use the metric-specific `<count>_pop`, not `pop`, to reproduce a nonblank rate.
  Counts may be fractional and retain the decimal precision needed to reproduce
  the published eight-decimal rate. At LSOA level, a measured metric's coverage
  population normally equals `pop`; all three metric fields are blank when it is
  unavailable. At higher levels, coverage population can be below `pop` because
  it sums only LSOAs with that metric available.

COUNT SEMANTICS
  Employment `claimant_count` is the mean of 12 monthly Claimant Count values in
  the calendar year. Claimant Count combines Jobseeker's Allowance with the
  relevant Universal Credit component: UC claimants not in employment early in
  the series, then those in the `Searching for Work` conditionality regime from
  April 2015. The same person can appear in both components. Nomis independently
  rounds each monthly observation to the nearest five before download, so the
  annual mean can be slightly above or below the unrounded value, with the
  largest relative effects in low-count areas. It is neither a unique-person
  count nor a sum of monthly values.

  Crime counts are annual street incidents assigned or apportioned to LSOAs.
  `recorded_count` sums the 13 police-recorded crime categories and excludes
  `Anti-social behaviour`. ASB remains available as its own incident series: it is
  governed by the National Standard for Incident Recording rather than the main
  police-recorded crime collection and must not be added to `recorded_count` when
  describing recorded crime. British Transport Police incidents are excluded.
  These measures are not survey estimates of all crime.

  Health `_afflicted` values are modelled estimates, not observed resident counts.
  They combine published GP-practice QOF prevalence with LSOA GP-registration
  patterns, then multiply the estimated prevalence rate by the ONS resident
  population; fractional people are therefore expected. They will not reconcile
  to NHS England's raw QOF register totals because GP practice lists and resident
  populations are different measures. Use `_rate` for area comparisons and
  `_afflicted` for roll-ups within these ADI files. Source gaps of at most two
  consecutive years are filled only when they are interior gaps bounded by an
  observation on each side. Leading and trailing gaps remain blank. The CSV has
  no per-cell flag distinguishing those interpolated estimates.

HEALTH COVERAGE INDICATORS
  `registration_coverage` is the area's GP registrations divided by its ONS
  resident population. It can exceed 1 because registrations and residents are
  different administrative measures. It is reported for interpretation; low
  values are not used to suppress estimates.

  `qof_coverage` is the share of those GP registrations at practices included in
  that year's QOF publication with a usable list size. It lies between 0 and 1.
  It is an overall practice-publication measure; coverage for a specific disease
  can be lower, and a disease estimate is withheld below 80% disease-specific
  coverage. The internal registration counts used to calculate these indicators
  are not additional deprivation metrics and are not included in the CSV.

  Missing LSOA indicator rows (years not listed have zero):
    registration_coverage: 2018: 42; 2019: 9
    qof_coverage: 2018: 42; 2019: 9

YEARS
  Employment and crime use calendar years. Health uses QOF financial years and is
  labelled by the ending year: health `2021` means QOF 2020-21 (April 2020 to
  March 2021). A shared year label therefore does not mean identical periods.

AVAILABILITY AND ADJUSTMENTS
  A territorial police force-year is accepted only when all 12 monthly files are
  present and non-empty and at least 90% of records that could belong to England
  carry an English LSOA code. A rejected force-year leaves its resident LADs and
  LSOAs blank. Exact rows with every crime metric blank are:
    year   missing LAD rows   missing LSOA rows
    2014                  0                   0
    2015                  0                   0
    2016                  5               1,038
    2017                  5               1,038
    2018                 14               1,735
    2019                 47               5,619
    2020                 16               2,095
    2021                 16               2,095
    2022                 16               2,095
    2023                 10               1,702
    2024                 10               1,702
    2025                 15               2,740
  Region and England rows remain available on their reduced metric-specific
  coverage populations.

  Source values that cannot be bracketed by two observations remain blank. LSOAs
  with every published health metric blank are:
    year   LSOAs with all 22 metric rates blank
    2014                                     64
    2015                                      0
    2016                                      0
    2017                                      0
    2018                                      0
    2019                                      0
    2020                                      0
    2021                                      0
    2022                                      0
    2023                                      0
    2024                                      1
    2025                                      2
  Non-diabetic hyperglycaemia (NDH) was not a published QOF register before
  QOF 2020-21, so every output row is intentionally blank through 2020 rather
  than missing through data loss. It begins in output year 2021, when
  16 LSOA rows remain blank. Eight LSOA epilepsy (EP) values
  in 2016 and seven LSOA heart-failure (HF) values in 2021 were rejected as
  implausible one-year spikes; aggregate coverage excludes them.

  Depression (DEP) in 2024 is interpolated from adjacent LSOA rates where both
  exist; 2 LSOAs without both anchors remain blank.
  Osteoporosis (OST) in 2015 is likewise interpolated where both anchors exist;
  64 LSOAs without both anchors remain blank. Both corrected
  metrics are reaggregated from LSOAs.

  NHS Digital warns that QOF 2020-21 implementation changes may make indicator
  values inaccurate and comparisons with earlier years unreliable. Obesity was
  particularly affected, and asthma and COPD register definitions changed. This
  is a comparability warning, not evidence that every condition moved downward.
  See https://digital.nhs.uk/data-and-information/publications/statistical/quality-and-outcomes-framework-achievement-prevalence-and-exceptions-data/2020-21

  CVD primary prevention (CVDPP) is published for output years 2014-2020 (QOF
  2013-14 through 2019-20) and is blank from 2021 after the register was
  withdrawn. There is a sharp England-level break between output years 2014 and
  2015, so comparisons across that boundary should be treated cautiously.
  Dartford's output-year 2019 rate (0.00624) also reverses sharply against 2018
  (0.00303) and 2020 (0.00412); it is retained but has not been source-validated.
  Smoking (SMOK) and hypothyroidism (THY) are excluded because each is available
  in only the first source year. Exact availability is in the data dictionary.

LICENCE
  Open Government Licence v3.0. Boundaries (c) ONS / Crown copyright.

Generated 2026-09-03. Methodology: https://adi.apps.autonomy.work/about
