Open, machine-readable datasets on police complaints, stop and search, use of force and deaths following police contact, for the United Kingdom, the United States and Australia.
Correction, 2026-08-30 (v1.1). The UK disparity figures in v1.0.0 were computed on an incomplete denominator. Census 2021 uses 2021 local-authority codes, and the local-authority to police-force lookup we joined against uses 2024 codes, so the 17 districts abolished in the April 2023 local government reorganisation matched nothing and 1,686,893 residents were silently dropped. Cumbria and North Yorkshire lost most or all of their populations; Avon and Somerset lost the whole of Somerset. A partial denominator produces a wrong ratio rather than a blank one, so the affected figures were understated. Corrected values: Avon and Somerset Black 4.3x → 5.9x, North Yorkshire Black 4.0x → 5.6x, North Yorkshire Asian 1.3x → 2.6x, national Black 4.4x → 4.5x, and Cumbria is computable for the first time. The build now fails loudly on any unmatched local authority instead of continuing.
Correction, 2026-08-31 (v1.1.1). The UK deaths series was understated by 163 deaths. The IOPC footnotes two of its year column headers, and our ingest matched year headers on a pattern that did not allow for a trailing footnote marker, so 2010/11 and 2015/16 were dropped silently and published as zero rather than failing. All of the difference falls in one category: deaths following other police contact, 1,299 to 1,462. The corrected all-category total is 3,666, not 3,503. The parser now asserts that exactly 21 year columns are found and aborts otherwise. The correction table is in
METHOD.md. Versions of this dataset published before 2026-08-31, including Zenodo v1.1.0, carry the understated figure and should not be used.
Everything here is derived data: computed from official open sources by code, then published with the source, the period and the denominator attached. No figure in this repository was written, estimated or rounded by a language model.
Published by PoliceComplaint.com, which presents the same figures as readable pages. This repository is the underlying data.
Most of the inputs are already public. Two of these outputs are not published in this form anywhere we could find, and one is a straightforward tidying of a source that is otherwise only available as a spreadsheet.
| Dataset | What it adds |
|---|---|
uk/stop-search-ethnic-disparity-by-force.csv and -national.csv |
Each force's stop-and-search shares by ethnicity divided by that ethnic group's share of that force area's own resident population, built by joining data.police.uk to Census 2021 via the ONS local-authority-to-police-force-area lookup |
uk/complaint-review-outcomes-by-force.csv |
Every force ranked on how often its own complaint decisions were overturned on review |
us/police-killings-by-state.csv |
Per-state aggregates including whether any officer was criminally charged, and how the prosecution ended |
australia/complaints-and-deaths-in-custody-by-state.csv |
The Report on Government Services police tables as tidy per-state rows rather than a 3,328-row wide spreadsheet |
The UK government already publishes per-force stop-and-search rates by ethnicity. Ethnicity Facts and Figures publishes "stop and search rate for every 1,000 people, by ethnicity and area", by police force area, using the same 2021 Census.
This dataset differs in two ways, and neither is "they don't publish it":
- Recency. Their most recent published year ends 31 March 2023. This covers rolling 36-month windows running to mid-2026.
- Measure and granularity. They publish a rate per 1,000 people. This publishes a ratio of shares: a group's share of recorded searches divided by its share of residents, alongside outcome data on the same rows.
Both are legitimate measures and they answer slightly different questions. If you need the official statistic, use theirs. If you need something more current, or you need outcome and ethnicity on the same row, use this.
All verified against the files in this repository at the commit you are reading.
United Kingdom
- 1,397,874 stop and search records across 41 of 44 forces. Three forces published no stop-and-search data at all and are named in the file rather than dropped.
- Nationally, Black people accounted for 18.2% of searches with a recorded ethnicity
against 4.0% of residents, a ratio of 4.5x. That national figure is computed over
1,151,285 searches across the 36 forces with a publishable denominator, and is in
uk/stop-search-ethnic-disparity-national.csvwith the shares it is derived from. - 24.5% of completed complaint reviews found the force's own handling was not reasonable and proportionate. The highest was Devon & Cornwall at 40.1%; the Metropolitan Police was 36.0%.
- Disparity ratios are published for 36 forces. Five forces have search data but no
publishable denominator and are marked
publishable=falsewith the reason.
United States
- 15,550 people killed by police, 2013 to 2026, across 51 states and 208 departments.
- 267 of those cases (1.7%) led to a criminal charge against an officer.
- 91 ended in a conviction or a guilty plea. 40 ended in acquittal.
Australia
- Complaints per 100,000 people in 2024-25 ranged from 6.3 in Victoria to 97.6 in the Northern Territory.
- 445 deaths in police custody, 2007-08 to 2024-25. 93 involved a person recorded as Aboriginal or Torres Strait Islander.
Read these before drawing a conclusion from any file here.
- A disparity ratio is not proof of unlawful conduct. It is a measured difference in outcomes. It does not establish a cause and it does not identify anyone's behaviour.
- The numerator and denominator use different bases. Searches carry officer-defined
ethnicity; the Census records self-defined ethnicity. Nationally, people recorded as
Mixed are 0.5% of searches against 2.9% of residents, a gap too large to be real:
officers rarely record "Mixed", so those people are recorded as something else at the
point of search, which inflates whichever category absorbs them. The direction of the
finding is solid; the magnitude is uncertain. See
METHOD.md, which sets out the test we ran against the government's own published rates. - City of London is an artefact, not an outlier. Its ratio is the highest in the UK file at 10.6x, because the City has roughly 8,600 residents and several hundred thousand daily workers and visitors. A residence-based denominator is the wrong denominator there. The figure is published because dropping inconvenient rows is worse, but do not rank it against territorial forces.
- Complaint counts reflect recording practice. A force that records complaints readily logs more of them. That is not the same as a force whose officers behave worse. This is most visible in the Australian file, where the 15x spread between Victoria and the Northern Territory is far too large to be conduct and is very largely definitional.
- Denominators are smaller than totals wherever a field is often blank. Body-camera and mental-health shares are computed only over records where the field was completed, and the denominator is published in its own column so you can see it.
- Periods differ by force. data.police.uk coverage is not uniform; each row carries its
own
period. - A recorded complaint is an allegation, not a finding. A death recorded as following police contact is categorised that way by the publisher and does not imply the police caused it or acted wrongly.
- Scotland and Northern Ireland are absent from the UK files. Police Scotland and the PSNI do not report into these datasets and the IOPC's statistics cover England and Wales only. There is no Scottish or Northern Irish data here because there is none to compute from, not because it was overlooked.
- Ireland has no dataset at all. We checked all 12,985 tables in the Central Statistics Office catalogue. No Irish police complaints, use-of-force or death-following-contact data exists in machine-readable form.
| Source | Covers | Licence |
|---|---|---|
| data.police.uk | Stop and search with outcomes, England, Wales, Northern Ireland | Open Government Licence v3.0 |
| Independent Office for Police Conduct | Complaints and review outcomes, England and Wales | Open Government Licence v3.0 |
| Home Office use of force statistics | Use of force reports | Open Government Licence v3.0 |
| ONS Census 2021 (TS021) and the LAD-to-PFA lookup | Population denominators by ethnic group | Open Government Licence v3.0 |
| Mapping Police Violence | People killed by US police, 2013 onwards | Per the publisher's terms; aggregates only here |
| US Census Bureau NST-EST2024 | State population denominators | Public domain |
| Report on Government Services 2026, Part C Section 6 | Australian complaints and deaths in custody | CC BY 4.0 |
Only aggregates are redistributed here. No source's record-level data is republished.
See METHOD.md for how each figure is computed, including the exact joins,
the publishability thresholds and the rounding.
The pipeline is deliberately boring: Python reads the source files, counts, divides, and writes. Where a rate cannot be computed honestly, the row is marked unpublishable and the reason is recorded rather than the row being dropped.
Every file here is also rendered as a browsable page, so a reader can check a figure without downloading a CSV. Each page names the source file it was computed from.
| Page | Renders |
|---|---|
| Open data downloads and methodology | All nine files, with per-column notes and the publishability thresholds |
| Who gets stopped and searched in England and Wales | uk/stop-search-ethnic-disparity-by-force.csv and the national file |
| All 44 UK police forces ranked by complaints | uk/complaint-review-outcomes-by-force.csv and uk/stop-and-search-outcomes-by-force.csv |
| Deaths during or following police contact, 21 years of data | Both deaths files |
| US police departments ranked by people killed by police | us/police-killings-by-agency.csv |
| The US state index | us/police-killings-by-state.csv |
| Police complaints and deaths in custody in Australia, by state | australia/complaints-and-deaths-in-custody-by-state.csv |
| Compare two UK police forces side by side | A tool built on the UK files |
| Press and media | Who publishes this, what may be reused, and the commercial interest disclosed in full |
Cite the concept DOI, which always resolves to the most recent version:
PoliceComplaint.com. (2026). Police accountability data: United Kingdom, United States and Australia [Data set]. Zenodo. https://doi.org/10.5281/zenodo.22165476
To cite this specific release, use 10.5281/zenodo.22165477.
If you rely on an underlying figure rather than our calculation, please cite the original publisher as well. Each dataset names its sources.
Derived data in this repository: CC BY 4.0. Attribute to PoliceComplaint.com and, where you rely on an underlying figure, to the original publisher named in the table above.
If anything here is wrong, tell us and we will check it against the source and correct it: open an issue, or email info@policecomplaint.com. That applies to police forces and oversight bodies writing about their own numbers as much as to anyone else.