More or Less Policing: how to read the numbers without being fooled

For more than twenty years, Radio 4’s More or Less has done something quietly radical: it takes a number that everyone is repeating — in a headline, a manifesto, a select committee — and asks the boring, devastating question. Where did that come from, and does it mean what they say it means? Tim Harford and the team have made a national pastime of the polite teardown, and along the way they have taught a generation of listeners — me included — that a statistic is not a fact. It’s a claim, with a method behind it, and the method is where the truth lives.

This is the first piece in a strand that borrows their habit of mind and points it at policing. Call it More or Less Policing. The premise is simple. Police data has never been more available — data.police.uk publishes millions of records a month under an open licence — and never more easy to misread. So each piece here will take one police number, or one kind of police number, and show you how to read it. And because reading is easier when you can see the thing, each one is paired with the live data explorer, where the same figures sit waiting, stamped with the month they describe and the day they were pulled.

A grateful nod to More or Less before we start: this strand is an homage, not an affiliation. If you don’t already subscribe, fix that. Then come back.

The number everyone gets wrong

Here is a figure you have seen in some form: Black people are several times more likely to be stopped and searched than white people. It is true. It is also, on its own, almost meaningless — and the reason why is the single most useful idea in all of statistics.

Look at the national stop-and-search snapshot in the explorer. You can see, this month, the share of searches carried out on each ethnic group. It is a stark picture. But notice what that chart is not telling you. It shows the numerator — how many searches. It says nothing about the denominator — how many people there were to search in the first place.

A rate is a numerator over a denominator. Change the denominator and the same numerator tells a completely different story. “Several times more likely” usually divides searches by residential population — who lives there, counted at the last census. But searches don’t happen where people sleep. They happen where people are: town centres at midnight, transport hubs, the night-time economy, the places policing is deployed. The population available to be searched at the time and place of searching is not the population of the borough. Sometimes it’s wildly different.

This is not an argument that disproportionality is fine. It plainly is not, and the evidence that some of it reflects bias rather than deployment is real and serious. It is an argument that the raw ratio cannot tell you how much — and anyone who quotes the ratio as if it settles the question, in either direction, has skipped the step that matters. The honest answer requires the right denominator, and the right denominator is hard.

Three questions to ask any rate

When a police statistic is a rate — and most of the ones that start arguments are — ask:

1. What’s on top, exactly? A “stop” can mean a search under PACE, a traffic stop, a stop-and-account. Count different things, get different numbers.
2. What’s on the bottom, exactly? Residents? Daytime population? People actually present? The choice often matters more than the data.
3. Compared to what? A force’s number means nothing until you set it beside last year, beside a similar force, beside what you’d expect. A single figure is a dot; insight starts at the second dot.

That last one is why the explorer always shows you every force at once, and why the standing health warning at the top of every view says the same three things: the data isn’t everywhere, isn’t now, and isn’t exact.

What this strand will do

Future pieces will take the same scalpel to other favourites: why “recorded crime went up” can be good news, what an “outcome rate” actually counts, why the crime map lies a little on purpose, and why the latest figure is never the latest month. Each will be short, each will name the trap, and each will hand you straight to the data so you can check my working — which is, after all, the whole point.

A statistic is a claim. Let’s read the claims carefully.

Explore the figures behind this piece in the Crime Dashboard →