Comparing two postcodes: a worked example.

A comparison sounds simple. Two numbers, smaller is better, the smaller place wins. It isn't. The headline total carries hidden assumptions about what the radius contains, who is reporting, and what categories make up the count, and the moment two areas differ on any of those axes the arithmetic stops behaving as advertised. This piece walks through a hypothetical comparison between two anonymised postcodes — call them Postcode A, an inner-city suburb of a large English city, and Postcode B, a market town in the same county — and shows where it goes wrong if the totals are read naively. The two places are stand-ins, not real postcodes; every figure below is illustrative.

The setup

Postcode A sits inside the urban footprint of a large city. Its half-mile radius covers tightly built terraced streets, two transport hubs, a high street with evening trade, a further-education college, and a small retail park. The resident population of the broader ward is around 12,000, but the weekday daytime population is materially higher because of the college, the shops, and the people moving through the stations. Postcode B is a market town in the same county. Its half-mile radius takes in a market square, three supermarkets, a primary school, residential terraces, semi-detached streets on quieter roads, and the edges of surrounding farmland. The resident population of the broader ward is also around 12,000.

The two are matched on the variable most readers reach for first — population. In every other respect they are different places. Choosing them this way makes visible something the postcode-level data cannot see: the built environment behind the count.

The headline numbers

The hypothetical 12-month totals are 612 incidents in A and 198 in B. At a glance, A looks roughly three times "worse" than B. A reader who stops there has a clean answer and a wrong one — or, more charitably, one that is true in a narrow arithmetic sense and misleading in every sense that matters for a decision. Three things distort the headline ratio, and each needs unpicking.

Adjusting for what the half-mile radius actually contains

The half-mile circle is the same shape in both places, but it is not measuring the same thing. A circle around A contains far more residences, businesses, vehicles, and pedestrians than the same circle around B. Even if the per-resident victimisation rate were identical, A's count would be higher because there is more of everything inside the radius — more parked cars to break into, more shops to lose stock from, more people on the street late at night. The headline number divides by an implicit denominator (people, opportunities, exposure) that is inconsistent between the two. The comparison is in fixed units of area when the underlying activity scales with what the area contains.

The fix is not a corrected ratio — the data gives no clean exposure denominator — but to hold the difference in mind as a discount on the raw gap. Some unknown share of 612-to-198 is simply that A's circle is denser.

Adjusting for category mix

The breakdown is more revealing than the total. In the hypothetical, A's 612 splits roughly as 220 anti-social behaviour, 140 violence and sexual offences, 80 vehicle crime, 70 shoplifting, 50 public order, 30 burglary, and the rest thin. B's 198 splits as 110 ASB, 25 vehicle crime, 22 shoplifting, 15 violence and sexual offences, 10 burglary, 8 public order.

Read side by side, the picture changes. Residential burglary — the category most readers actually care about when they ask "is it safe to live here" — is 30 in A and 10 in B. Per resident the rates are similar; neither postcode is a burglary outlier. Most of A's headline excess sits in categories that track commercial and transit activity rather than residential life. Shoplifting and ASB cluster around shops and high streets. Public order rises near licensed venues. Violence and sexual offences are weighted toward the transport hubs. The 3:1 ratio collapses to something much smaller once the comparison is restricted to categories that scale with where people live rather than where they pass through.

Adjusting for reporting density

A separate distortion sits on top of category mix and is not visible in the breakdown alone. An urban suburb has more CCTV, more witnesses, more proactive policing, and shorter response times than a market town. Incidents that would never enter the recording system in B enter it routinely in A. A late-night altercation on a station concourse is recorded; the same altercation on a quiet road outside B might be resolved informally and never reach the police. Part of the 612-to-198 gap is real difference in what happened, and part is difference in what made it into the system. Without separating the two, the comparison overstates the underlying difference.

We have no clean way to subtract the reporting effect from a postcode-level total. What we can do is treat the gap with humility, and ask whether the categories driving most of the difference are the ones most likely to be affected by reporting density. ASB and public order are heavily reporting-sensitive; burglary is much less so.

Reading category-by-category

This is where the comparison becomes useful rather than misleading. Residential burglary, normalised mentally for property density, is broadly similar — neither postcode is an outlier on the category most directly tied to living somewhere. ASB is higher in A absolute, but the spatial pattern, where the data exposes it, sits near the transport hubs and the high street, which is where ASB tends to land in a suburb of that shape. Vehicle crime is higher in A absolute but probably similar per parked car once kerbside density is taken into account. Shoplifting tracks the retail footprint. Violence and sexual offences are higher in A, and that gap is the most genuinely concerning piece, though even there the concentration around the transport hubs suggests the residential streets are not where most of it is happening.

The comparison that initially read "A is three times worse than B" reads, after disaggregation, as "A and B are similar at the residential street level, but A has substantially more activity around its commercial and transport zones, and that activity produces the headline gap." Those are different statements, and they support different decisions.

The trend dimension

One more lens is worth applying. Are the two areas trending the same way? In the hypothetical, both are drifting upward at roughly 5 to 8 per cent year on year. Whatever is happening at the macro level — recording standard changes, methodology shifts, post-pandemic normalisation — is affecting both similarly, so the comparison-of-trends carries less information than the comparison-of-levels. The trend line is mostly common shock, not divergent local behaviour. When trends diverge — one area climbing while the other stays flat, or one falling while the other holds — comparing trends becomes meaningful, because the common component has cancelled out and what remains is local.

What we recommend

A short approach is enough to keep the comparison honest. Don't lead with the headline total — lead with the category breakdown, because the shape of the count carries more information than the size of it. Mentally normalise for what is in the radius — residential, commercial, transport, mixed — before treating two figures as comparable. Compare like with like where possible: two postcodes of similar density and land-use mix make a defensible comparison; an inner-city suburb and a market town make a category error dressed up as one. Use trends only when comparing at scale or across years; for a single year between two specific places, the levels and breakdowns do most of the work. The arithmetic is the easy part. Choosing the right denominator, time window, and disaggregation is what decides whether the answer is any good.