National trends and local departures

About this explorer

This is an independent replica built on the structure of Andrew Wheeler's CrimeDecomp, which separates US city-level crime patterns into trend, season and residual components. This version uses England & Wales Community Safety Partnership (CSP) and Police Force Area (PFA) data from the Home Office's Police Recorded Crime open data tables instead, at quarterly rather than monthly resolution.

Methodology and how this differs from the original

Each offence group is modeled separately. A logit-scale additive decomposition is fit per crime type: a national natural-spline trend plus a saturated financial-year-quarter effect (the quarterly analogue of the original's monthly Fourier harmonics), population-weighted across CSPs. Each CSP and PFA then gets its own trend and seasonal deviation from that national curve, fit as a simple per-area regression rather than a single joint mixed-effects model. This is a practical simplification, not a re-implementation of the original's glmmTMB random-effects model: smaller CSPs' curves are not statistically shrunk toward the national pattern the way a true mixed model would shrink them, so noisy small-area estimates should be read with that in mind.

Quarterly counts occasionally go negative in the Home Office data ahead of 2019, reflecting crimes later transferred to another force or cancelled and netted against the quarter in which that was discovered, rather than corrected retrospectively. These are genuine values, floored at zero only for the purposes of the rate/logit calculation used here, not zeroed out or hidden.

"Most serious violence" is not a category in the underlying Home Office tables. This is a legacy indicator (originally National Indicator 15, pre-2012) that some Community Safety Partnerships and Police and Crime Commissioners still construct locally, broadly as homicide plus the most serious wounding/endangerment offences. Treat it as a locally-constructed aggregate rather than an official published series.

Data transformations and area exclusions

A small number of areas were excluded outright. The City of London (both as a CSP and as its own Police Force Area) is dropped throughout: its resident population is around 8,600 but its daytime/working population runs into the hundreds of thousands, making any population-normalised rate for it an extreme, non-comparable outlier. Similar could be contended regarding Westminster in London, where the annual visitor numbers exceed tens of millions and activity in the West End creates similar outlying dynamics. The Isles of Scilly CSP is dropped for the same reason in miniature, a resident population of roughly 2,000 means its crime rate can swing wildly on a handful of offences. Manchester Airport is combined into the Manchester CSP, since the airport falls within the Manchester CSP boundary rather than constituting a separate comparable area.

Some CSPs were renamed or merged over the data's time range and are treated as a single continuous series under the current name. Historic Merthyr Tydfil and Rhondda Cynon Taf counts are combined into the CSP now named Cwm Taf.

In several police force areas the CSP boundaries published in the crime tables didn't line up with the local-authority boundaries used in the population estimates. Where CSPs were renamed/merged to align with the population data, the corresponding older CSPs were combined to match:

• Cumbria: Barrow-in-Furness, Eden and South Lakeland → Westmorland and Furness; Carlisle, Allerdale and Copeland → Cumberland
• Dorset: Bournemouth and Poole → Bournemouth, Christchurch and Poole
• Northamptonshire: Corby, East Northamptonshire, Kettering and Wellingborough → North Northamptonshire; Daventry and South Northamptonshire, and Northampton → West Northamptonshire
• Thames Valley: Aylesbury Vale, Chiltern, South Bucks and Wycombe → Buckinghamshire

Suffolk's LADs and CSPs don't align at all, so all of Suffolk except Ipswich was merged into a single Suffolk area in both the crime and population datasets: Suffolk Coastal, Waveney and Western Suffolk → Suffolk.

The reverse adjustment was also needed in other force areas, merging local-authority population figures together so they matched a CSP that is itself made up of multiple LADs:

• Devon and Cornwall: East Devon and Mid Devon → East and Mid Devon; North Devon and Torridge → Northern Devon; South Hams, Teignbridge and West Devon → South Devon and Dartmoor
• Hampshire: Basingstoke and Deane, Hart and Rushmoor → North Hampshire
• Kent: Dartford and Gravesham → Dartford and Gravesham
• Nottinghamshire: Broxtowe, Gedling and Rushcliffe → South Nottinghamshire
• South Wales: Rhondda Cynon Taf and Merthyr Tydfil → Cwm Taf
• Suffolk: Babergh and Mid Suffolk → Suffolk
• Warwickshire: Stratford-on-Avon and Warwick → South Warwickshire
• West Mercia: Bromsgrove, Redditch and Wyre Forest → North Worcestershire; Malvern Hills, Worcester and Wychavon → South Worcester/South Worcestershire

Finally, four Kent CSPs: Maidstone, Medway, Tonbridge and Malling, and Tunbridge Wells are excluded from the final crime dataset entirely. Since the 2024/25 financial year, the Home Office data has not assigned their crime totals accurately at CSP level; the counts instead appear under an "Unassigned Kent" category. Kent's Police Force Area total is unaffected by this exclusion, since it's built independently from force-level totals rather than by summing these CSPs.