More or Less: Five Habits for Reading Police Statistics

The most useful thing about BBC Radio 4’s More or Less is not that it catches bad numbers. It teaches a repeatable way of interrogating numbers that look perfectly plausible.

Tim Harford has presented the programme since October 2007. He describes it as a programme about the way numbers are used in the news and public debate; the Open University, which co-produces the series, describes the same mission as asking what statistics really measure and what kind of truth they support. Tim Harford — More or Less; Open University — More or Less.

That makes it unusually relevant to policing. Forces produce large volumes of statistics, dashboards and performance measures. Most errors are not arithmetic errors. They are interpretation errors.

Five habits from the programme’s way of thinking are especially useful.

1. Ask for the denominator

A count is not automatically a rate and a rate is not automatically meaningful.

“1,000 offences” can describe a serious local problem or a very low national rate depending on the population, time period and opportunity for the offence. “A 50% increase” can mean an increase from two incidents to three.

The first question should therefore be: compared with how many?

In policing, denominator choice often changes the interpretation. Shoplifting per resident population answers a different question from shoplifting per daytime population or retail footfall. Stop and search per resident population does not account for commuters and tourists. Use of force per officer answers a different question from use of force per arrest or per encounter.

There is rarely one universally correct denominator. There is a denominator that matches the question being asked. Worthing’s run as the “shoplifting capital of the UK” shows what the choice does in practice: a daytime-population denominator reorders the whole league table.

2. Ask what was actually counted

Many statistical arguments are really arguments about definitions.

Police-recorded crime is not the same measure as victimisation in the Crime Survey for England and Wales. A complaint is not the same thing as a substantiated misconduct finding. An arrest is not a conviction. A missing-person report is not a unique missing person if the same individual is reported repeatedly.

The label on the chart can conceal these distinctions.

A useful discipline is to rewrite the statistic in full before discussing it. Instead of “crime is up 20%”, say: police-recorded shoplifting offences in this force area were 20% higher in the year ending March 2026 than in the previous 12 months. The longer sentence exposes the period, measure and geography that the shorter slogan hides.

3. Ask “compared with what?”

A single number is often just a dot.

Performance starts to become interpretable when there is a meaningful comparison: the same place last year, a comparable force, a control group, a counterfactual estimate or a long-term trend.

The comparator has to fit the claim. Comparing a tourist city with a rural county because both are police-force areas may tell you less than comparing similar local authorities or micro-locations. Comparing this month with last month can be misleading when the offence is seasonal.

The point is not to collect more comparisons. It is to choose one that helps answer the question.

4. Do not turn association into causation

Police dashboards are full of before-and-after stories.

A force introduces an operation and crime falls. A training programme begins and complaints fall. More neighbourhood officers arrive and confidence rises.

Any of those changes may be real. None of the before-and-after patterns alone proves the intervention caused the outcome.

Other things may have changed at the same time. The underlying trend may already have been moving. Reporting practices may have altered. The intervention may have been targeted precisely where the problem was expected to improve or worsen.

The stronger the causal claim, the stronger the design needed to support it. The familiar “domestic abuse rises 38% when England lose” slogan is a case study in a number outrunning its design — the underlying evidence is both narrower and stronger.

This does not make ordinary operational data useless. It means leaders should distinguish monitoring from evaluation.

5. Be most suspicious when you like the answer

Harford’s book How to Make the World Add Up places emotional reactions near the centre of statistical judgement. The danger is not only being fooled by somebody else’s number. It is wanting the number to be true. Tim Harford.

That is especially relevant in organisational performance.

A statistic showing your intervention succeeded receives less scrutiny than one showing it failed. A figure supporting a policy you already dislike attracts methodological questions that were absent when the same method supported your position.

The defence is procedural rather than moral: apply the same checklist regardless of whether the conclusion is welcome.

What is the numerator? What is the denominator? What is the definition? What is the comparator? What does the design establish? What important information is missing?

Why this belongs on a policing site

Evidence-based policing is not simply about finding research papers. It is about learning how much confidence to place in a claim.

More or Less models that habit in an accessible form. It asks questions senior police leaders, analysts and practitioners should ask routinely of their own numbers, including numbers produced internally with good intentions.

The objective is not cynicism. A reader who assumes every statistic is manipulated is no better informed than one who believes every chart.

The better position is curiosity with method.

A number is evidence only after you understand what produced it and what conclusion it can carry. That is a useful half-hour lesson for policing every week.

---

Sources and further reading

• Tim Harford — More or Less
• Open University — More or Less
• Tim Harford — How to Make the World Add Up