Skip to main content

UK Census Data for Local Stories

The census is the only free dataset that describes every street in the country on the same day, using the same definitions. It is also the one most often reported wrongly, because its geography, its disclosure control and its pandemic timing all behave in ways that are invisible in a spreadsheet. This guide covers choosing the right geography, getting the data out, and the traps that ruin small-area census stories.

Last reviewed: Next review due:

Method, not numbers: this guide deliberately avoids quoting census population figures. Every claim here is about how the data is built, protected and released. Take the figures themselves from the live table you are citing, and link to it.

Why the census is the local reporter’s best free dataset

Almost every other dataset a local journalist uses is a sample, an administrative by-product, or a return that some bodies file well and others file badly. The census is none of those. It is a mandatory, whole-population count, taken on one day, with one set of definitions, published down to areas of a few hundred people, and free to reuse. That combination is why it ends up as the denominator underneath most other local statistics you will ever quote.

The quality is genuinely high. For Census 2021 in England and Wales the person response rate reached 97% of the usual resident population, and over 88% in every local authority, against targets of 94% and 80% respectively. Local authorities were invited to review provisional estimates for their own areas before publication.

It has exactly one structural weakness, and it is worth keeping in mind throughout: a census is a snapshot, and it starts ageing the moment it is taken. Everything below is about extracting the parts that stay true, and flagging the parts that were already unusual on the day.

Three censuses, two dates, three agencies

“The 2021 census” is not one thing. ONS ran England and Wales, NISRA ran Northern Ireland, and National Records of Scotland ran Scotland — and Scotland ran a year late.

England and Wales — 21 March 2021
Run by ONS. This is the census behind Nomis, Create a Custom Dataset, the topic summaries and the multivariate tables. Its outputs are accredited official statistics, with one exception covered below.
Northern Ireland — 21 March 2021
Run by NISRA, on the same census day as England and Wales, which makes it the more straightforward devolved comparison. NISRA publishes its own results and its own Flexible Table Builder for building custom Census 2021 tables.
Scotland — 20 March 2022
Run by National Records of Scotland. Scottish Ministers moved the date on NRS advice after, as OSR puts it, "COVID-19 restrictions delayed or prevented NRS’s logistical planning and testing of census systems and processes". Collection was then extended to the end of May 2022 after a slow start: the national return rate stood at 79% at the end of April, rose to 87.9% by the end of the extension, and reached 89.2% after final receipting in September 2022. NRS therefore leaned harder on coverage adjustment than ONS did, and that is worth a line in any Scotland-versus-England comparison.

ONS says the timing difference “makes the production of UK results more challenging than usual”, and produces UK totals by interpolating Scottish values for March 2021 between the 2011 and 2022 censuses. It will not do this below country level, because interpolation for smaller areas is “additionally challenging, with the quality of the resulting values liable to be lower”. There is no UK census figure for your patch, and the interpolated Scottish values do not replace NRS’s own results as the definitive statistics for Scotland. For how the three agencies relate to one another, see our guide to the UK statistical bodies.

The geography ladder: pick the right rung

Most census stories succeed or fail at this step. Go too small and the data is suppressed, noisy, or describes an area no reader recognises. Go too large and the story disappears into an average. The statistical hierarchy runs output area to LSOA to MSOA; wards, parishes and local authorities are administrative geographies assembled from those building blocks.

Output area (OA)
40–250 households; 100–625 usual residents
Lowest level of census geography

Street-scale patterns, and the building block from which every larger area is assembled. Rarely publishable on its own once disclosure control bites — and never for a rare characteristic.

Lower layer super output area (LSOA)
400–1,200 households; 1,000–3,000 residents
33,755 in England, 1,917 in Wales

The workhorse for neighbourhood stories, and the geography most other datasets — deprivation, health, housing — are published on, so it joins well.

Middle layer super output area (MSOA)
2,000–6,000 households; 5,000–15,000 residents
6,856 in England, 408 in Wales; fit within local authorities

Big enough to survive disclosure control for most characteristics, small enough to mean something locally. Start here when an LSOA breakdown comes back blocked.

Electoral ward
Varies enormously — no population threshold
6,904 in England, 762 in Wales, as at 5 May 2022

The geography readers and councillors actually recognise. Census outputs are fixed to the 2022 boundaries, so check against your council’s current wards before naming one.

Parish or community
Varies; mainly rural in England, whole-country in Wales
10,480 parishes in England, 878 communities in Wales as of 2022

Village-level stories where a ward is too coarse. Expect heavy suppression on anything beyond headline counts.

Local authority
The whole council area
Upper and lower tier both published

Comparisons between councils, and the level at which almost every other official statistic is available for a fair join.

Two practical notes. Because wards and parishes are assembled from output areas rather than counted independently, ONS selectively realigned around 200 OA boundaries to ward boundaries for 2021 to improve the fit — a reminder that ward figures are constructed, not native. And if you are not sure which ward, council or constituency covers the address in your story, our Find Your Patch tool will resolve a postcode to all of them.

Where to actually get the data

Nomis
University of Durham, on behalf of ONS
The query tool most reporters should learn first. Launched in 1981, it holds Census 2021 alongside 2011, 2001 and earlier, plus labour market and benefits data. Use the dataset finder to identify a table by topic, the ready made tables for common cross-tabulations, and area profiles for a one-area summary. It downloads clean CSV with ONS geography codes attached.
ONS Create a Custom Dataset
ONS
ONS’s own flexible table builder for Census 2021 in England and Wales. You pick a population type — all usual residents, all households, residents in communal establishments and so on — then variables, then an area type. It is the right tool when no published table crosses the two variables your story needs.
ONS Open Geography Portal
ONS
Boundary files, names and codes, and lookup tables. This is where the map layer comes from, and where you get the OA-to-ward or LSOA-to-local-authority lookups that let you aggregate census data into an area your readers recognise.
Scotland’s Census
National Records of Scotland
Scotland’s Census 2022 results and reports live on the NRS census site, on its own timetable and with its own classifications. Never assume an England and Wales table has a Scottish twin with the same categories.
NISRA Flexible Table Builder
Northern Ireland Statistics and Research Agency
Northern Ireland’s Census 2021 outputs, with both a custom table builder and a library of ready-made tables. NI ran on the same census day as England and Wales, which makes it the easier of the two devolved comparisons.
Census 2021 Dictionary
ONS
Not a data source but the page you should open before every census story. Each variable has its definition, its classifications, the exact question asked, and a comparability rating against the 2011 Census. Skip it and you will eventually compare two things that are not the same thing.

ONS releases census results as a family of products rather than one file: topic summaries carrying mostly single-variable data with commentary, multivariate datasets crossing two or more variables, origin-destination or flow data showing movement between places, and anonymised microdata samples. If the cross-tabulation you need is not in a published table, the custom dataset builder is usually the answer before an FOI is.

Statistical disclosure control: the small-area trap

This is the most important thing on this page, and the reason small-area census journalism goes wrong. ONS is obliged to protect the confidentiality of individual responses, and it does that by changing the data before you ever see it. Not by hiding some of it — by altering it.

The trap: small census counts are not exact counts

ONS protected Census 2021 three ways: it swapped records between areas, applied a cell key method to every table, and applied disclosure rules deciding which tables could be published at all. Targeted record swapping moved unusual or identifying households to similar areas: the geographies were changed for between 7% and 10% of households, and between 2% and 5% of individuals were swapped between communal establishments.

Cell key perturbationthen adds small amounts of noise to the published counts. ONS says “a typical dataset would have around 14% of cell counts perturbed by a small amount”, and that “small counts were more likely to have been perturbed so datasets with large counts receive less noise than those with many small counts”. The perturbation is consistent — the same cell appearing in two different datasets receives the same treatment — but ONS warns directly that “the randomness may mean small changes to totals”, and that across two differently constructed datasets “the totals of all cells may in turn be different”.

So: a count of 7 in an output area is not a statement that seven people are there. A difference of 2 between two small areas is not a finding. A total that refuses to reconcile with its own components is the method working, not a mistake to hunt down. And the smaller and rarer the thing you are counting, the more of what you are looking at is protection rather than population.

The third mechanism shows up as suppression. Create a Custom Dataset has rules that stop a dataset being produced at all if it would be too disclosive, and the guidance on the trade-off is genuinely useful: where the system suppresses data, you may be able to get less detailed data for the same area, or the same detail for a larger area. That is a choice to make consciously and then explain in the piece.

None of this makes small-area census data unusable. It makes it usable for patterns rather than values. A run of neighbouring LSOAs all pointing the same way is a story. One output area with an eye-catching number is not.

2021 was a pandemic census

England and Wales were counted on 21 March 2021, during a national lockdown, with guidance to work from home wherever possible and millions of people on furlough. ONS has published a standing note on which topics that distorted, and it names five.

Students
Lockdown began in January 2021, so many students never returned to their term-time address before census day. ONS reworked the guidance and then corrected the data, saying it "identified over 90,000 students who would otherwise have been incorrectly enumerated at their non-term-time address". University-town population and age structure still need handling with care.
Labour market
Furloughed workers were advised to report themselves as employed and economically active, but ONS says "guidance on answering work-related questions may not have been consistently followed". Census 2021 estimated around a million fewer employed 16 to 64 year olds than the Labour Force Survey. ONS advises that "care should be taken when using Census 2021 data to draw conclusions about the labour market in other years".
Travel to work
Home workers were told to "answer for the longest part, by distance, of your current journey to work", while furloughed and isolating people were told to answer how they used to travel. ONS says the results are "unlikely to represent pre-pandemic or post-pandemic travel patterns or workplaces" and to take extra care below region level.
Geographic location of usual residents
Where people were counted is itself a pandemic artefact in places. Second homes, seasonal workforces and anyone temporarily relocated were all captured mid-lockdown, and that flows through into every per-area figure built on the resident base.
International migration
Another topic ONS lists as most likely affected. Treat country-of-birth and recent-arrival change since 2011 as indicative of structure rather than as a clean measure of flows over the decade.
Everything else
Tenure, household composition, ethnicity, religion, national identity, qualifications, disability, unpaid care and health are far less exposed to lockdown distortion. These are where the strongest and safest local stories are.

ONS’s own advice on the commuting data is the model for how to handle all of this: take care not to overinterpret, take extra care below region level, and accept that “it is difficult to know whether patterns detected are concentrations of furloughed populations, pandemic-related changes in behaviour, or longer-term trends”. If you cannot separate those three explanations, do not pick one in the headline.

Comparing with 2011 without inventing a trend

“Up 30% since the last census” is the most common census headline and the easiest to get wrong, because two separate things can move underneath it: the definition and the boundary.

  • The variable definition moved: The Census 2021 Dictionary rates each variable as comparable, broadly comparable, or not comparable with 2011. "Broadly comparable" means "changes may have been made to the question or options that people could choose from or how write-in answers are classified" — enough to move a percentage without anything changing on the ground.
  • The question guidance moved: ONS reviewed and redeveloped all question guidance for 2021. In some cases that slightly changed the data collected even where the question wording looks similar. The dictionary page for the variable is where this is disclosed.
  • The output areas moved: 2011 OAs that still sat inside the 40 to 250 household and 100 to 625 resident thresholds were carried forward unchanged. Others were merged or split into new 2021 OAs, and LSOAs and MSOAs were reshaped the same way. Around 200 OA boundaries were realigned to ward boundaries.
  • The ward boundaries moved: Census outputs are published for wards as they stood on 5 May 2022. Councils redraw wards at boundary reviews, so the ward in your census table may not be the ward on your council’s current map, or the one the councillor you are quoting represents.
  • One census ran a year late: Scotland was enumerated in March 2022, England, Wales and Northern Ireland in March 2021. Any three-nation comparison is comparing different moments, and ONS explicitly declines to interpolate Scottish values below country level.

The practical test before publishing any change-since-2011 figure: open the variable’s page in the Census 2021 Dictionary and read its comparability rating and its notes, then confirm the area code you are using means the same ground in both censuses. Both checks take a couple of minutes and both belong in your notes.

The new 2021 questions — and one accreditation change

Census 2021 in England and Wales asked about sexual orientation and gender identityfor the first time, both voluntary and both asked only of people aged 16 and over. It also added a question on previous service in the UK armed forces, which gave local reporters a veteran population estimate for their area for the first time. Because the two identity questions were voluntary, some over-16s left them blank — and ONS treats those blanks as legitimate responses rather than imputing values to fill them, so be precise about whether a percentage is of all over-16s or only of those who answered.

Gender identity: no longer accredited official statistics

On 5 September 2024 ONS asked that the Census 2021 gender identity estimates for England and Wales cease to be accredited official statistics and be reclassified as official statistics in development. On 12 September 2024the Office for Statistics Regulation published its final review and agreed. Ed Humpherson’s letter to ONS states: “OSR supports your proposal to publish the Census 2021 gender identity statistics for England and Wales as official statistics in development rather than accredited official statistics.”

OSR’s finding was specific: “We consider that there is sufficient evidence that the gender identity question in the census has underperformed, with some groups of people being more likely than others to misunderstand the question.” ONS says the evidence “does show there is potential for bias in how the gender identity question was answered by those in England and Wales who responded that they did not speak English well”. Its quality guidance goes further: smaller group breakdowns “should not be used as precise estimates of the trans population and comparisons between areas or groups should be considered with caution, particularly where there are different levels of English language proficiency”.

For a local reporter that warning lands squarely on the story you were probably going to write. A local-authority or ward-level trans population figure, or a table ranking areas by it, is exactly the breakdown ONS is telling you not to treat as precise — and areas differ in English language proficiency, which is the specific mechanism suspected of biasing it. Report the estimates as official statistics in development, say what the review found, and do not build a league table on them.

Two things this change is not. It is not a judgement on the rest of the census: Humpherson’s letter is explicit that “all other outputs from the Census 2021 in England and Wales are unaffected by this decision and remain designated as accredited official statistics”. And it is not a downgrade of the sexual orientation estimates, which came from the same voluntary question block and remain accredited official statistics. Getting that distinction right matters, because conflating the two misrepresents both.

Rates, bases and change over time

1
Use the census for both numerator and denominator where you can
The census’s quiet advantage is that it gives you the population of exactly the same area at exactly the same moment. If you take counts from one source and population from another, you have introduced a mismatch the census would have solved for you.
2
Say which population base you used
All usual residents, residents in households, households, and household reference persons are different bases, and they produce different percentages from the same underlying data. ONS makes you choose one in Create a Custom Dataset for a reason. Name it in the copy.
3
Percentage of what, exactly
The voluntary questions were answered by a subset of the population, and ONS did not impute the missing answers — it treats a blank as a legitimate response. A percentage calculated against everyone aged 16 and over is a different number from one calculated against those who answered.
4
Check the boundary vintage on both sides
Match Census 2021 data to 2021 output area, LSOA and MSOA boundaries, and ward outputs to the 5 May 2022 ward boundaries. Joining 2021 data to 2011 geography, or to this year’s redrawn wards, produces a map that is wrong in a way nobody will spot in the newsroom.
5
Treat change since 2011 as two questions
Did the thing change, or did the measurement change? Check the variable’s comparability rating in the Census 2021 Dictionary, then check whether the area itself was merged, split or realigned between the two censuses.
6
Do not build a rank on perturbed small counts
League tables of output areas or wards on a rare characteristic are the classic misuse. The noise ONS added lands hardest on exactly the small counts your ranking is made of.

Mapping census data

Census data maps well because ONS publishes the boundaries to go with it. The Open Geography Portal — “the UK’s official Office for National Statistics Open Data Geography Site” — carries boundary files, names and codes, and the lookup tables that connect one geography to another.

  • Join on ONS codes, never on names. Area names repeat across the country; codes do not.
  • Match the vintage. Census 2021 data belongs on 2021 output area, LSOA and MSOA boundaries, and ward outputs on the 5 May 2022 wards.
  • Map a rate or a percentage, not a count. A choropleth of raw counts is a map of where people live, which your readers already know.
  • Beware the large rural area. Output areas are equalised on population, not on land, so a sparsely populated area dominates the map while representing few people.
  • Do not map down to a level the disclosure control has made unreliable. If the pattern only appears at output area level, it is probably noise.
  • Say in the caption which geography you mapped and which census it came from.

For the mechanics of building the map itself, see our mapping UK data guide.

A snapshot that ages

The census is a photograph of one day, and that day is receding. Some of what it measures is structural and ages slowly — housing stock, tenure mix, the age profile of a settlement, entrenched patterns of deprivation and unpaid care. Some of it ages fast, particularly anything about work, commuting or recent migration, all three of which were already unusual in March 2021.

The rule of thumb: use the census for structure and composition, and a more current source for levels. If you need today’s population of a ward as a denominator, ONS mid-year population estimates are the better choice; the census is what tells you what kind of ward it is.

There will be a next one. In June 2025 the UK Statistics Authority recommended that the government commission ONS to deliver a questionnaire-based census of England and Wales in 2031, alongside continued development of an admin-based population and migration statistics system; in July 2025 Cabinet Office Minister Georgia Gould confirmed the government was commissioning it. NRS and NISRA also recommended a census in 2031. A consultation on which topics Census 2031 should cover ran from 28 October 2025 to 4 February 2026, with a full report due later in 2026 — a live local story in itself for anyone whose patch depends on a question that might not survive.

Red flags in your own copy

  • A "UK census" figure for a town, region or local authority — the three censuses ran on two different dates and ONS does not interpolate below country level.
  • A ranking of wards or output areas built on small counts, presented without mentioning that ONS perturbed them.
  • A percentage change since 2011 for a variable the Census 2021 Dictionary does not rate as comparable.
  • A commuting, workplace or employment story leaning on Census 2021 as though March 2021 were a normal month.
  • A trans or gender identity figure for a local area quoted as an accredited official statistic.
  • A map that joins 2021 census data to 2011 boundary files, or to this year’s redrawn wards.
  • A student-population claim about a university town that does not acknowledge the lockdown enumeration problem.
  • The phrase "the census shows" attached to a number that is now several years old, where a mid-year population estimate exists.

Before you publish checklist

  • I have named which census the figure comes from — England and Wales 2021, Northern Ireland 2021, or Scotland 2022 — and linked the table.
  • I have named the geography and the population base, and confirmed both are the same on every side of my comparison.
  • I have checked the variable in the Census 2021 Dictionary for its definition and its comparability rating against 2011.
  • If the story rests on a small area or a rare characteristic, I have said in the copy that ONS perturbs small counts, and I have not built a ranking on them.
  • If the topic is students, work, commuting, migration or where people were counted, I have flagged that Census 2021 was taken during lockdown.
  • If I am quoting gender identity data, I have described it as official statistics in development and explained the review finding.
  • If mapping, I have matched the boundary vintage to the census and joined on ONS codes.
  • I have put the interpretation to the council, the university or the relevant body and included their response.

Start from the geography, not the spreadsheet

Most census stories begin with a place rather than a table. Resolve the postcode to its ward, council and constituency first, then pull the census figures for exactly those areas — and where the census stops short, the council usually holds the rest.

Find Your PatchCouncil FOI Requests

Frequently asked questions

Was the census taken on the same day across the UK?
No, and this is the first thing to get right. Census day in England, Wales and Northern Ireland was 21 March 2021. Scotland’s census was taken a year later, on 20 March 2022, after Scottish Ministers moved the date on the advice of National Records of Scotland — the Office for Statistics Regulation records that COVID-19 restrictions "delayed or prevented NRS’s logistical planning and testing of census systems and processes". ONS says the difference in timing "makes the production of UK results more challenging than usual", and builds UK-wide figures by interpolating Scottish values for March 2021 between the 2011 and 2022 censuses. Crucially, ONS does not do this below country level: it says it will not produce those statistics for smaller areas because interpolation there is "additionally challenging, with the quality of the resulting values liable to be lower". So there is no such thing as a UK census figure for your town.
What is the smallest area I can get census data for?
The output area (OA), which is the lowest level of geography for census statistics. An OA contains between 40 and 250 households and a usually resident population of between 100 and 625 people, with exceptions such as areas containing a large prison. Above it sit lower layer super output areas (LSOAs), usually four or five OAs, with 400 to 1,200 households and 1,000 to 3,000 residents; and middle layer super output areas (MSOAs), usually four or five LSOAs, with 2,000 to 6,000 households and 5,000 to 15,000 residents. MSOAs fit within local authorities. Being able to get a number at OA level is not the same as being able to publish it — see the disclosure control section, because that is where small-area reporting goes wrong.
Why do my census numbers not add up?
Because ONS deliberately altered them. It protected confidentiality in Census 2021 three ways: swapping records between areas, applying a cell key method to every table, and applying disclosure rules about which tables could be published at all. Targeted record swapping changed the geography of between 7% and 10% of households, and swapped between 2% and 5% of individuals between communal establishments. Cell key perturbation then adds small amounts of noise to counts: ONS says "a typical dataset would have around 14% of cell counts perturbed by a small amount", and that "small counts were more likely to have been perturbed". ONS states plainly that "the randomness may mean small changes to totals" and that where you build two different datasets, "the totals of all cells may in turn be different". A row that will not reconcile is usually the disclosure control working as designed, not an error you can chase down.
Can I report the Census 2021 gender identity figures?
You can, but not as accredited official statistics, and not without the caveat. On 5 September 2024 ONS asked that the Census 2021 gender identity estimates for England and Wales stop being accredited official statistics and be reclassified as official statistics in development. OSR published its final review on 12 September 2024 and agreed: Ed Humpherson wrote that "OSR supports your proposal to publish the Census 2021 gender identity statistics for England and Wales as official statistics in development rather than accredited official statistics". OSR’s report concluded there is "sufficient evidence that the gender identity question in the census has underperformed, with some groups of people being more likely than others to misunderstand the question". ONS says the evidence "does show there is potential for bias in how the gender identity question was answered by those in England and Wales who responded that they did not speak English well", and warns that smaller group breakdowns "should not be used as precise estimates of the trans population", with comparisons between areas treated cautiously where English language proficiency differs. Two things follow. First, a local-authority or ward-level trans population figure from Census 2021 is exactly the kind of breakdown ONS is warning you about. Second, this change applies only to gender identity: Humpherson’s letter says "all other outputs from the Census 2021 in England and Wales are unaffected by this decision and remain designated as accredited official statistics", and the sexual orientation estimates from the same question block remain accredited.
Can I compare Census 2021 with Census 2011?
Variable by variable, and only after checking. The Census 2021 Dictionary gives each variable a comparability rating against 2011. ONS defines the middle rating like this: "A variable that is broadly comparable means that it can be generally compared with the same variable used in the 2011 Census. However, changes may have been made to the question or options that people could choose from or how write-in answers are classified." Some variables are simply not comparable. The geography moved too: 2011 output areas that still met the household and population thresholds were kept unchanged, but others were merged or split to create new 2021 OAs, and the same was done to LSOAs and MSOAs — so an area code that exists in both censuses does not always describe the same ground. Around 200 OA boundaries were also selectively realigned to ward boundaries.
Does the pandemic make Census 2021 unusable?
No, but it makes five topics unsafe to treat as normal. ONS identifies the topics most likely to have been affected as the geographic location of usual residents, students, international migration, labour market, and travel to work. On students, ONS says it "identified over 90,000 students who would otherwise have been incorrectly enumerated at their non-term-time address". On work, furloughed people were advised to report themselves as employed and economically active, but ONS notes that "guidance on answering work-related questions may not have been consistently followed", and Census 2021 estimated around a million fewer employed people aged 16 to 64 than the Labour Force Survey. On commuting, ONS is blunt: "Census 2021 results are unlikely to represent pre-pandemic or post-pandemic travel patterns or workplaces." Everything else — tenure, household composition, age structure, ethnicity, religion, qualifications, disability, unpaid care — is far less exposed.
Where do I get boundary files to map census data?
The ONS Open Geography Portal, which describes itself as "the UK’s official Office for National Statistics Open Data Geography Site". It publishes the boundary files, names and codes, and the lookup tables that let you join one geography to another. Two rules save most of the pain. Match the vintage: use 2021 output area, LSOA and MSOA boundaries with Census 2021 data, not the 2011 versions. And take the lookups from the portal rather than joining on names — area names repeat across the country, ONS codes do not.
When is the next census, and what do I use until then?
In June 2025 the UK Statistics Authority recommended that the government commission ONS to deliver a questionnaire-based census of England and Wales in 2031, alongside continued development of an admin-based population and migration statistics system. In July 2025 Cabinet Office Minister Georgia Gould confirmed the government was commissioning ONS to run that census; National Records of Scotland and NISRA also recommended a census in 2031. A topic consultation on what Census 2031 should ask ran from 28 October 2025 to 4 February 2026. In the meantime, a census figure is a snapshot that ages — use ONS mid-year population estimates for current denominators and treat census characteristics as structural context rather than a current headcount.

Related guides

Data JournalismData Journalism HubRead →Data JournalismUK Statistical Bodies: ONS, NRS, NISRA and UKSARead →Data JournalismMapping UK DataRead →Data JournalismUK Crime Data GuideRead →BeatsLocal Government Reporting BeatRead →

Primary sources

Check the live release. Census outputs are added to and revised after first publication, and quality guidance is updated inside the ONS, NRS and NISRA methodology pages rather than announced separately. Verify every figure, and every accreditation status, against the current publication before you file.