Which parts of London never came back?
London has more people in it than ever before. Fifteen of its thirty-three boroughs are still smaller than they used to be, and together they are two million people short of their own record.
Published · Data as of
London has never had more people in it. The 2021 census counted 8,799,727 usual residents, which is 184,727 more than the city held at its old peak in 1939. That is the story the Greater London Authority published in 2015, and a national news cycle ran it. It is true, and it is the least interesting thing in this spreadsheet.
Here is the more interesting thing. Of London’s thirty-three boroughs, fifteen are still smaller than they used to be — and together they are 2,080,658 people short of their own records. The city set an all-time high with a hole in it 11 times the size of the surplus.
Greater London's population rises from 1,097,000 in 1801 to 8,615,000 in 1939, falls 23.3% to 6,608,598 by 1981, then climbs 33.2% to a record 8,799,727 in 2021.
Show the numbers
| Year | Population | Counted as | Census? |
|---|---|---|---|
| 1801 | 1,097,000 | persons present | yes |
| 1811 | 1,303,000 | persons present | yes |
| 1821 | 1,573,000 | persons present | yes |
| 1831 | 1,878,000 | persons present | yes |
| 1841 | 2,207,000 | persons present | yes |
| 1851 | 2,651,000 | persons present | yes |
| 1861 | 3,188,000 | persons present | yes |
| 1871 | 3,841,000 | persons present | yes |
| 1881 | 4,713,000 | persons present | yes |
| 1891 | 5,572,000 | persons present | yes |
| 1901 | 6,510,000 | persons present | yes |
| 1911 | 7,162,000 | persons present | yes |
| 1921 | 7,387,000 | persons present | yes |
| 1931 | 8,110,856 | persons present | yes |
| 1939 | 8,615,000 | persons present | mid-year estimate |
| 1951 | 8,196,807 | persons present | yes |
| 1961 | 7,997,094 | persons present | yes |
| 1971 | 7,452,343 | persons present | yes |
| 1981 | 6,608,598 | residents | yes |
| 1991 | 6,679,699 | all usual residents | yes |
| 2001 | 7,172,057 | all usual residents | yes |
| 2011 | 8,173,941 | all usual residents | yes |
| 2021 | 8,799,727 | all usual residents | yes |
Source: Historical Census Population (OGL-2.0) — full attribution
The fall is not the Blitz. It starts before the war and it keeps going for four decades after it: from 8,615,000 in 1939 down 23.3% to 6,608,598 in 1981, which is the bottom. Everything since has been the climb back.
Fifteen of London’s thirty-three boroughs held fewer people in 2021 than at their own historical maximum, and every one of those maxima falls before the Second World War.
FactThe boroughs that did not come back
City of London holds 93.3% fewer people than at its 1801 maximum, Westminster 61.0% fewer than in 1871, down to Waltham Forest at 2.6% below its 1939 maximum.
Show the numbers
| Borough | Peak year | At its peak | In 2021 | Below peak |
|---|---|---|---|---|
| City of London | 1801 | 129,000 | 8,583 | 93.3% |
| Westminster | 1871 | 524,000 | 204,236 | 61.0% |
| Islington | 1901 | 436,000 | 216,589 | 50.3% |
| Southwark | 1901 | 596,000 | 307,637 | 48.4% |
| Tower Hamlets | 1901 | 597,000 | 310,306 | 48.0% |
| Camden | 1901 | 377,000 | 210,136 | 44.3% |
| Kensington and Chelsea | 1901 | 250,000 | 143,375 | 42.6% |
| Hammersmith and Fulham | 1921 | 288,000 | 183,157 | 36.4% |
| Hackney | 1901 | 389,000 | 259,146 | 33.4% |
| Lambeth | 1931 | 421,000 | 317,654 | 24.5% |
| Newham | 1921 | 448,000 | 351,036 | 21.6% |
| Wandsworth | 1931 | 388,000 | 327,506 | 15.6% |
| Haringey | 1931 | 307,391 | 264,238 | 14.0% |
| Lewisham | 1931 | 326,844 | 300,553 | 8.0% |
| Waltham Forest | 1939 | 286,000 | 278,425 | 2.6% |
Source: Historical Census Population (OGL-2.0) — full attribution
The list has a shape. City of London is the extreme and the joke: it peaked in 1801, the very first column of this spreadsheet, at 129,000 people, and now holds 8,583 — 93.3% fewer. Nobody moved out of the Square Mile so much as the Square Mile stopped being a place people lived.
After that come the inner boroughs, and they are the ones the Victorians filled first. Eight of the fifteen reached their maximum in 1901 or earlier: City of London, Westminster, Islington, Southwark, Tower Hamlets, Camden, Kensington and Chelsea, Hackney. Waltham Forest is the near miss, 2.6% short of its 1939 figure — close enough that the measurement seams described below could account for it on their own.
Every borough peaked either before the war or right now
18 boroughs are at their all-time maximum in 2021. The other 15 peaked between 1801 and 1939. No borough reached its maximum in the intervening decades.
Show the numbers
| Peak year | Boroughs | Which |
|---|---|---|
| 1801 | 1 | City of London |
| 1871 | 1 | Westminster |
| 1901 | 6 | Camden, Hackney, Islington, Kensington and Chelsea, Southwark, Tower Hamlets |
| 1921 | 2 | Hammersmith and Fulham, Newham |
| 1931 | 4 | Haringey, Lambeth, Lewisham, Wandsworth |
| 1939 | 1 | Waltham Forest |
| 2021 | 18 | Barking and Dagenham, Barnet, Bexley, Brent, Bromley, Croydon, Ealing, Enfield, Greenwich, Harrow, Havering, Hillingdon, Hounslow, Kingston upon Thames, Merton, Redbridge, Richmond upon Thames, Sutton |
Source: Historical Census Population (OGL-2.0) — full attribution
The split is total. Every borough at its maximum reached it in 2021 itself. Every borough below its maximum reached it before the Second World War. There is no borough whose best year was 1961, or 1991, or 2011 — no borough that filled up, emptied, and half-filled again. Whatever happened to London happened to two sets of places in opposite directions, and it has not reversed.
The arithmetic underneath the record is simple enough to state in one line. Since 1939, the fifteen boroughs still below their peak have lost 1,044,203 people between them. The other eighteen have gained 1,227,376. London’s record is what is left over.
The city did not recover. It redistributed, and the total crossing its old line is a coincidence of the two movements roughly cancelling.
InterpretationThree things this chart is not telling you
1939 is not a census. There was no census in 1941 — the war cancelled it — so the peak everyone quotes should not be in a census series at all. It is here because the publisher put it here, and its own footnote says what it is: a mid-year estimate built on the September 1939 National Register. It is drawn hollow above for that reason, and every comparison with it is a comparison with an estimate.
The ruler changes twice. This series measures persons present from 1801 to 1971, residents for 1981, and all usual residents from 1991 to 2021. “Persons present” counts who was physically in a place on the night. “Usual residents” counts who lives there. They are different quantities, and the change is invisible in the numbers.
The record itself crosses both seams: 8,799,727 usual residents in 2021 is being compared with 8,615,000 persons present in 1939, a difference of 2.1%.
FactThat last point deserves its weight. The margin by which London broke its old record — 2.1% — is smaller than the definitional change sitting underneath it. The publisher says so itself: when both measures were published for 1971, the resident count came out 23,000 higher than persons present for London overall. I am confident London is at a record. I am not confident it is at a record by 184,727.
Everything before 1931, and 1939, is a reconstruction. Those figures are not counts but estimates made by the Greater London Council, working backwards onto today’s borough boundaries from a city that was not organised that way. The boroughs this piece leans on hardest — the ones whose maxima fall in 1901 or earlier — are exactly the ones most exposed to that.
Waltham Forest, at 2.6% below its 1939 figure, would cross back over its own record under a consistent definition.
Hypothesis Test: A reconstruction of the 1801–1931 series on a single population definition, or borough-level resident estimates for 1971 to sit beside the persons-present ones.Behind the data
Method. One workbook from the London Datastore, borough level, every year in the series from 1801 to 2021. For each of the thirty-three boroughs the maximum is found across all twenty-three columns and compared with its 2021 value; ties go to the earlier year. Nothing is joined, nothing is modelled, nothing is interpolated, and no borough boundary is redrawn — the publisher has already reconstructed the whole series onto constant modern boundaries, which is what makes it comparable and is also somebody else’s modelling choice. The Greater London line is the publisher’s own total rather than the sum of the boroughs, because the publisher’s footnote warns that the two can differ.
Caveats. The population base changes twice, at 1981 and at 1991, and the headline record crosses both. 1939 is a mid-year estimate rather than a census and is the only such year here. Everything to 1921, and 1939, comes from a Greater London Council reconstruction rather than a count. The census years are unevenly spaced and there is a twenty-year gap across the war, so a line between points implies a smoothness the data does not have. For 2021 the publisher notes that no table gave definitive totals, so the figures are the sum of females and males. And “borough” here is an administrative shape, not a neighbourhood: a borough below its maximum can contain streets that are fuller than they have ever been.
This file is also not stable. It was fetched on 30 August 2026 and again on 12 September 2026 from the same address, and the second version had a different layout, an extra check row and fuller footnotes. Every figure used here is identical across the two, and the pipeline refuses to write if the shape moves in a way it does not recognise.
What would change my mind. A reconstruction of this series on one population definition throughout would move boroughs in and out of the fifteen: Waltham Forest is 2.6% short of its own record, which is inside the margin the definitional change could account for on its own. So would evidence that the Council’s pre-1931 reconstruction is materially wrong for the inner boroughs, which is where this piece puts most of its weight. Neither would touch the shape of the finding — that the boroughs below their maximum all peaked before the war and the rest are peaking now — but either could change the count.
Sources
- Historical Census Population — Office for National Statistics, via the Greater London Authority (London Datastore). OGL-2.0. Retrieved . One workbook, borough level, 1801 to 2021, twenty-three years of which twenty-two are censuses. Five properties of this file are the piece rather than footnotes, and none of them is visible in the numbers. First, the ruler changes twice, not once. The dataset page says "persons present 1801 to 1971 and usual residents for 1981 to 2021" — two regimes. The workbook's own footnote says three: "Population is Persons present from 1801 to 1971, Residents for 1981, and Usual Residents from 1991 onwards." The file is the better authority because it travels with the numbers. The publisher also quantifies the 1971 seam: resident estimates published for that year were "slightly higher than persons present for most boroughs and 23,000 higher for London overall". Second, 1939 is in this series and is not a census. The workbook's footnote: "1939 figures are mid-year estimates based on the September 1939 National Register of population." So the peak this subject is usually argued over is present after all, and transform.py flags it so that no chart and no sentence can treat it as a count. Third, the early figures are reconstructions. The sources block names Greater London Council Research Memorandum 413 for 1801 to 1921 and for 1939; the 1961 Greater London Report for 1931, 1951 and 1961; and the individual censuses from 1971. For 2021 the publisher adds that "there was no table providing definitive population totals", so the figures given are the sum of females and males. Fourth, the file's own date metadata is wrong and must not be trusted: the resource page prints "Last Updated: October 2014" beside a file whose contents run to 2021. That is ISSUES #039, and it is why transform.py prints the last year it actually found and refuses to run if that year is not 2021. Fifth, and learned the hard way: THIS FILE IS NOT STABLE. It was fetched on 2026-08-30 as a 24,458-byte workbook with `data` and `metadata` sheets, and again on 2026-09-12 as a 39,914-byte workbook with one `Sheet1`, a header on row 4, two rival definitions of Inner and Outer London printed one after the other, and a publisher's own "Sum of boroughs" check row. Same URL, same dataset page. Every borough and London figure this piece uses is identical across the two versions — the re-issue changed the layout and added documentation, not the numbers — and the re-issue is better documented. But the cross-check in transform.py refused to write on both encounters, once at 136% and once at exactly 200%, and it was right both times. A future run should expect the shape to move again. The download URL carries a UUID under the dataset slug expjm, and the resource page and the URL disagree about the filename — the page calls it population-borough-1801-2021.xlsx and the link serves "population 1801 to 2021.xlsx". fetch.py therefore resolves the link from the dataset page by title rather than hard-coding either name, the same discipline the phone piece needed for a different reason.
- Research Memorandum 413, The Changing Population of the London Boroughs — Greater London Council. OGL-2.0. Retrieved . Published on the same Datastore page as the workbook, 3.03 MB, and cited by the workbook itself as the source of every figure from 1801 to 1921 and of the 1939 estimate. It is therefore the primary source for the reconstructed half of this series, and it is named here rather than described as "estimates" because a reader who wants to check the Victorian numbers needs somewhere to go. Not downloaded by fetch.py: nothing in the pipeline reads it, and it is cited rather than processed.
Attribution
- OGL-2.0 Historical Census Population, Research Memorandum 413, The Changing Population of the London Boroughs
Contains public sector information licensed under the Open Government Licence v2.0. The GLA cannot warrant the quality or accuracy of this data.
Data as of . Last fetched . Pipeline code is available on request.