# Historical population surface, Great Britain 1801 to 2021

Every census year's population as population-weighted points, the
surface behind the station-distance series on this site. Built by the
`Atlas/histpop` pipeline (source published beside the datasets as
`atlas_pipelines_src.zip`); `VALIDATION.md` beside this file gives each
year's total against the published census by nation, its coverage and
its measured placement error, and `SHA256SUMS` the checksums of this
build.

## population_surface_<era>.csv.gz

One row per point, about 6.3 million rows across 22 census years, in
four files by era (the same eras as the GeoPackages, and the same
columns in each): `early` 1801 to 1841, `victorian` 1851 to 1911,
`interwar` 1921 to 1961, `modern` 1971 to 2021. Concatenate them for the
whole surface.

| column | meaning |
|---|---|
| `census_year` | the census the count comes from (1801 to 1911 and 1921 to 2021 at the decennial census; there was no 1941 census) |
| `easting`, `northing` | British National Grid (EPSG:27700), metres |
| `population` | people at this point; the points of one year and nation sum to that census within the tolerance shown in VALIDATION.md |
| `coverage` | which nations the year holds; every year is `GB` |
| `placement` | how the points of that year were placed, see below |

The `placement` column holds one of `early`, `modelled`, `official`.
`early` is the name-placed parish tier (1801 to 1841); `modelled` is a
count spread inside a census unit over a density basis (1851 to 1961);
`official` is a publisher's own population-weighted centroid, or a
density basis where none is published (1971 to 2021). The unit each
year rests on, by nation:

| census year | unit, England and Wales | unit, Scotland | placement |
|---|---|---|---|
| 1801 to 1831 | parish, 1831 Comparative Account, placed by name on GB1900 labels inside the registration district the 1852 index gives the name where it gives one (spread over that district's labels where no label inside carries the name: `district-density`), else on a label inside the county, else by county label density | civil parish polygons | `early` |
| 1841 | parish, 1841 Enumeration Abstract, placed as 1831 | civil parish polygons | `early` |
| 1851 to 1901 | registration sub-district, seeded by name inside its registration district, spread over GB1900 label density | civil parish polygons | `modelled` |
| 1911 | local government district, spread over GB1900 label density | civil parish polygons | `modelled` |
| 1921, 1931 | civil parish inside rural districts where the County Report's Table 3 proved (356 of 589 rural districts in 1921, 484 of 602 in 1931); boroughs, urban districts and unproven rural districts on district rows; spread over GB1900 label density | civil parish, NRS Table 8, spread over GB1900 label density | `modelled` |
| 1951, 1961 | local government district, spread over the 1981 enumeration district surface | civil parish, NRS Table 8, spread over the 1981 surface | `modelled` |
| 1971 | enumeration district centroids | the 1981 surface scaled (see VALIDATION.md) | `official` |
| 1981 | enumeration district centroids | postcode grid references, calibrated to the 1991 Monitor | `official` |
| 1991 | enumeration district counts spread over 2001 centroids (no 1991 centroid product exists) | postcode grid references | `official` |
| 2001, 2011, 2021 | output area population-weighted centroids | postcode grid references | `official` |

A point is not a household: it is a share of a census unit's count,
spread inside that unit over a density basis (map labels of the 1900s,
or modern centroids), so distances computed from these points carry the
placement error VALIDATION.md measures for each tier.

`interpolation_years.csv` says, for every year 1801 to 2026, which census
shape the year borrows (`census_shape`), how many years it is from that
census (`years_from_shape`) and the national total used, for anyone
interpolating between censuses.

## distance_percentiles.csv

The distance-to-nearest-station series by census year, computed over
this surface and the published station table (`station_history/`),
population weighted, straight line; the series the site charts. One row
per census year:

| column | meaning |
|---|---|
| `census_year` | the census year |
| `stations_open` | public passenger stations open on the census night, judged at the precision of each station's own dates (METHODS.md, definitions); rows the station table marks `duplicate_of` are counted once |
| `stations_sourced_only` | the same count over stations whose dates come from a record rather than a model |
| `stations_year_only_in_census_year` | stations whose year-only opening or closure falls in the census year, so the convention (opening counts as open, closure as closed) decided them |
| `stations_within_a_month_of_census_night` | stations with an opening, closing or reopening date within 31 days of the census night, the ones a month's error in the records would move across it |
| `population_basis` | `modelled` or the official basis of the year (`ED centroids`, `OA centroids`, with the Scottish source) |
| `coverage` | `GB` in every year |
| `population` | the surface's total for the year |
| `p50_mi`, `p50_km` | the median person's distance to the nearest open station; likewise `p75_mi`, `p75_km`, `p90_mi`, `p90_km`, `p95_mi`, `p95_km`, `p99_mi`, `p99_km` for the 75th, 90th, 95th and 99th percentile person |
| `p50_err_mi` (and `p75_err_mi`, `p90_err_mi`, `p95_err_mi`, `p99_err_mi`) | the modelled error band for that percentile, miles: placement, extrapolation, coverage, station-date, record-date, Scotland-model, name-placement and district-spread terms in quadrature, each measured except the name-placement term, which is estimated (VALIDATION.md). Quadrature assumes the terms are independent; the three placement terms will correlate, so the band understates by that much. This is a ceiling, see below |
| `record_date_err_p50_mi`, `record_date_err_p90_mi` | the record-date term on its own: the shift when every station within a month of the census night whose date the sources give only to the month or the year is flipped to the other side of it; day-precise dates are not flipped |
| `census_night_sensitivity_p50_mi`, `census_night_sensitivity_p90_mi` | not a term: the shift when every station within a month of the census night is flipped regardless of precision, which is how much the answer depends on the exact night (large in 1841, when the Great Western opened lines on 31 May and 14 June) |
| `check_bias_p50_mi`, `check_bias_p90_mi` | where a tier has a direct check (1851, 1861, 1921, 1931): the tier's own spreading method minus the truth held by named 1841 parish points, signed, miles; a floor, see below |
| `missing_stations_p50_mi`, `missing_stations_p90_mi`, `missing_stations_p99_mi` | sensitivity: the shift when the share of closed stations the station table knows it lacks is dropped at random from that year's stations (reduced minus full, miles); a lower bound on the missing-station effect, since the lack is counted only for the counties Wikipedia lists |
| `placement_basis` | how the year's points were placed and what its band rests on |

**What the band is.** The ± is a combined heuristic band: measured
shifts, one estimated term and one bound (the record-date flip) added
in quadrature. It is not a confidence interval and carries no coverage
probability; read it as the scale of the known error terms, not as odds.

**Ceiling and floor.** The error band was calibrated on 1971 enumeration
districts and applied to every modelled year with the year's own unit
widths, so it is defined the same way throughout; where a tier can be
checked directly with 1841 parish points the check comes out well inside
it (1851 p90 0.58 modelled against 0.31 checked; 1921 0.50 against 0.08).
The band is therefore a ceiling and the check a floor: the check holds
the 1841 name-placed points as truth, so a parish on the wrong village is
invisible to it, and it says nothing about the counts themselves. The
chart draws the ceiling. The years with no check (1871 to 1911, 1951,
1961) have only the ceiling, and the direction of its error there is not
known. The early tier's name placement has a term of its own, an
estimate from re-placing the index-constrained rows county-wide and
scaling to the share the index does not know (VALIDATION.md), and its
district-density rows a measured spread term.

The file is computed from OpenStreetMap-positioned stations, so under
the ODbL it is a Produced Work of the station table rather than a
derivative database; it carries the station table's attribution
("Map data (c) OpenStreetMap contributors, ODbL") and is otherwise
published under this dataset's CC BY-SA 4.0.

## The transcriptions

The figures behind the parish tiers were read from page images of the
printed census volumes. The rows are published as read, with the proof
each row passed, so a reader can check any figure against the page:

- `transcription_ew_parish_1801_1831.csv`: the 1831 Comparative Account
  (Internet Archive b30450019), England and Wales parishes with 1801,
  1811, 1821 and 1831 counts; `image_page` and `printed_page` locate the
  row, `match` says how the name was placed: `district` and
  `district-fuzzy` on a label inside the district the 1852 index gives
  the name, `district-density` spread over that district's labels
  (no coordinates in this file), `exact`, `variant`, `despaced` and
  `fuzzy` on a label inside the county, `heading` and `parent` at the
  borough or parish the row is listed under, with `|scattered` and
  `|by-parent` qualifiers; blank means the row sits on county density.
  `rd_units` lists the registration districts the index allows the name
  (SN 9032 unit ids), `easting`/`northing` is the placed point,
  `is_total` and `is_heading` mark rows that carry no population.
- `transcription_ew_parish_1841.csv`: the 1841 Enumeration Abstract
  (Internet Archive india.history.resource.73451); a row stands only when
  its printed males and females sum to its total; `leaf` locates it,
  `placement` says how it was placed: `1831-placed` at the 1831 row of
  the same name (rejected where it lies outside the district the index
  gives the name), `gb1900-district` on a label inside that district,
  `district-density` spread over that district's labels, `gb1900` on a
  label inside the county, `1831-sequence` recovered from the alphabetical
  order of the page, `county-density` spread over the county's labels.
- `transcription_ew_parish_1921.csv` and `transcription_ew_parish_1931.csv`:
  Table 3 of every England and Wales County Report (histpop.org), civil
  parishes with acreage, the earlier census, persons, males and females;
  `mno`, `page` and `row` locate the cell on histpop's page images;
  `kind` classifies the row (`parish`, `sub`, `district`, `total`,
  `county`, `entire`, `header`, `other`); `basis` is the proof the
  persons figure passed (`sum`: persons equals males plus females as
  read; `agree`: two readers agree; `split`, `single`, `derived`: weaker,
  see BUILD_EW_1921_1931.md in the source bundle); `raw` is the reader's
  text; `name_tess` a second reading of the name.
- `transcription_scot_parish_1921_1971.csv`: Scottish civil parishes 1921
  to 1971 from the 1971 County Report's Table 8, supplied by National
  Records of Scotland; `basis` is `ocr` (read from the scan and proved on
  its already-known columns), `t14` (the 1921 Vol. I Table 14), `hand`
  (transcribed by hand), `closed` (a parish absorbed by then) or
  `model_1911_1981` (the share-model backbone, not a reading). Data from
  National Records of Scotland.

## The same points as GeoPackages

Four era files (`population_surface_early.gpkg`, `_victorian`, `_interwar`,
`_modern`; a layer per census year, about 740 MB together) are kept in
the project archive rather than served here, being too large for the
repository. Ask through the contact page if you want them; their
checksums are in `SHA256SUMS`.

## Licence and credit

CC BY-SA 4.0 for the surface as a whole, because its census inputs are
CC BY-SA deposits (UK Data Archive SN 4554, 4570, 9035, 9180). Please
credit "Mind the Gates railway atlas; data from National Records of
Scotland; contains OS data © Crown copyright and database right; GB1900
gazetteer (CC0); ONS and Nomis data under the Open Government Licence
v3". `ATTRIBUTION.md` at the root of the published atlas lists every
source with its terms. Corrections are welcome with a source.