How to Read Police Statistics Without Fooling Yourself
Police statistics are easy to quote and easy to misread. Start with the numerator, denominator, definition and comparator — then ask what the data can actually establish. Stop and search shows why the method matters.
Audio edition
≈ 6 min · narrated
Audio edition
≈ 6 min · on-device voice
Police statistics are not self-explanatory. A rate that looks precise can still answer the wrong question if the numerator, denominator or comparator does not match the claim being made.
Stop and search is a good example because the Home Office itself now publishes unusually clear warnings about the method.
In the year ending March 2025, official force-area rates were calculated using resident population estimates. The Home Office says those figures do not account for transient populations such as commuters and tourists, can hide substantial variation within a police force area, and also need to be read alongside the fact that stop and search is concentrated among particular age and sex groups. Home Office, year ending March 2025.
That does not make stop-and-search disparity statistics meaningless. It means the headline ratio is the beginning of the analysis rather than the end.
Start with the numerator
Before interpreting any rate, ask what has actually been counted.
In stop and search, the Home Office counts search incidents, not unique people. A person searched several times can therefore appear several times in the numerator. Some records also have missing ethnicity, age or sex information and are excluded from particular analyses. Home Office, 2025.
The same problem appears throughout policing data.
A missing-person report is not necessarily a unique missing person. A use-of-force record can contain several tactics. Recorded crime depends on reporting and recording rules. A complaint is not a misconduct finding.
The label on a chart should never substitute for understanding the unit being counted.
Then inspect the denominator
Rates are useful because they put a count into context. They also create a new analytical choice: what should the count be divided by?
For stop and search, resident population is transparent and nationally available. It allows standardised comparisons across ethnic groups and areas. The limitation is that police encounters happen where people are, not only where they live.
The Home Office therefore warns that force-area rates do not capture commuters, tourists and other transient populations. It also notes that large force areas can contain places with very different population compositions and very different levels of police activity. Home Office, 2025.
A separate government analysis of geographical differences makes the same broader point: the spatial distribution of ethnic groups, deprivation and stop-and-search activity can affect the size of disparity estimates. Race Disparity Unit, 2021.
The correct conclusion is not “the disparity is explained by geography”. It is that the resident-population rate alone does not identify the mechanism producing the disparity.
Bias, deployment, exposure, local crime patterns, officer decision-making and geography are empirical questions. They need evidence rather than being selected according to which explanation is politically convenient.
A rate should be written as a sentence
One way to improve police performance reporting is to force every important rate into plain English.
Instead of:
Black people are X times more likely to be stopped.
write:
In the specified period, the number of recorded stop-and-search incidents involving Black people per 1,000 Black residents was X times the equivalent rate for White people, using the published resident-population denominator.
The longer version is less elegant. It is also much harder to overinterpret.
It exposes the period, the fact that incidents rather than unique people are counted, and the denominator used. A subsequent sentence can then explain relevant methodological limits and additional evidence.
This habit works for almost any policing statistic.
Comparisons need a reason
A number usually becomes interesting only when placed beside another number.
The comparison could be the previous year, a similar force, a control group, a local target or an expected value. Each answers a different question.
Year-on-year comparison can be distorted by seasonality or a recording change. Force comparison can be distorted by different population or demand structures. Target comparison can tell you whether an organisation hit its ambition without telling you whether the ambition was sensible.
A comparator should therefore be selected because it helps answer the policy question, not because it produces the most dramatic percentage.
Recorded data can change without underlying harm changing equally
Police data is especially vulnerable to measurement changes because the police are part of the mechanism that creates the dataset.
Improve reporting access and recorded crime can rise even if underlying offending does not. Tighten recording compliance and the same thing can happen. Change a classification rule and a time series can break without any behavioural change in the public.
The reverse is also possible. A fall in reporting can make recorded demand look better while victims are simply becoming less visible.
This is why the site’s Worthing shoplifting analysis treats reporting behaviour as part of the evidence rather than a technical footnote. The same principle should apply elsewhere: ask whether the measurement process changed before interpreting the trend as a change in the underlying problem.
Statistical caution should not become statistical paralysis
The purpose of these questions is not to make every number unusable.
A rate can be imperfect and still informative. A police-recorded series can have limitations and still show a meaningful change. An observational comparison can support a sensible operational hypothesis without proving causation.
The discipline is to make the conclusion no stronger than the measure.
For a senior leader, four questions catch a large proportion of avoidable errors:
- What exactly is the numerator?
- What exactly is the denominator?
- Compared with what, and why that comparator?
- What else could have moved the number besides the explanation being proposed?
If those answers are visible, the statistic becomes easier to challenge and more useful to act on.
That is the objective of the More or Less Policing strand: not fewer numbers, but better claims built from them.
Explore the site’s data work in the Crime Dashboard →.
Sources and further reading
Discussion questions
- 01
Which dashboard rate in your force would change most if a different but defensible denominator were used?
- 02
Does your performance process distinguish a measurement problem from a genuine operational change?
