Methodology

1. Two types of data

The two data types are always clearly labelled and never blended into a single score.

2. Underreporting caveat

Recorded crime statistics capture only thefts reported to police. Vehicle theft is widely underreported for several reasons: insurance may not require a police reference, owners may consider it futile, or recovery happened before a report was filed. In the UK, the Crime Survey for England and Wales consistently estimates actual vehicle crime at significantly higher levels than recorded crime.

We show the recorded rate prominently with a caveat on every page. We never present these figures as absolute counts of all vehicle thefts. Community reports provide complementary (though also incomplete) recency data.

3. All motor vehicles — not motorcycles specifically

Both source datasets aggregate all motor vehicles:

• France (SSMSI) — indicateur Vols de véhicule: cars, vans, motorcycles, mopeds, and other motor vehicles combined.
• UK (Home Office) — offence code 48 (theft of a motor vehicle): similarly all vehicle types.

Motorcycle/scooter-specific splits are not available at commune or local-authority level in official data. Community reports on this site do capture vehicle type, making them the only bike-specific data available. We state the "all vehicles" limitation plainly on every official data display.

4. Denominators — why per 1,000 residents?

Raw theft counts are meaningless for comparison — Paris will always have more thefts than a village. We normalise by resident population per 1,000 because:

• Population figures are freely available at commune and local-authority level from INSEE (France) and ONS (UK).
• Registered-vehicle counts at this granularity are not publicly available for both countries in a consistent form.

We rank areas within each country only. French communes and UK local authorities have very different catchment sizes and population structures — a cross-country league table would be misleading.

5. French data suppression (~53% of communes)

SSMSI suppresses counts below a statistical threshold to prevent individual identification. Approximately 53% of French communes have suppressed vehicle theft figures. These appear in our data as confidence = suppressed.

Suppressed does not mean low-risk. We show suppressed communes separately, never in the ranked table, and never as evidence of safety. The honest message is: insufficient data.

6. Community report handling

• Anonymous. No account required. IP is hashed (not stored in clear) for rate-limiting only.
• Recency-weighted. We compute a score approximating a 90-day half-life — a report from today scores ~1; one from 62 days ago ~0.5. This surfaces active areas without overweighting historical clusters.
• Jittered pins. Display locations are shifted by ±~150 m deterministically from the UUID. Exact submission coordinates are never shown.
• Post-moderation. Reports are published immediately then reviewed. Spam, abuse, or implausible entries are removed; removed reports are excluded from all public aggregates.
• No data ≠ safe. Absence of community reports never implies low risk. We show an honest empty state.

7. Data licences