CrimeRate

Methodology

How CrimeRate risk scores are computed: data sources, modeling approach, and limitations.

CrimeRate publishes crime risk scores at neighborhood resolution across US metros and surrounding counties. Every score on this site is computed by PerilScore using the same data layer used by insurance and risk management professionals.

Data sources

We start from public records. Every input is auditable, and we don’t use proprietary or paywalled data.

  • FBI Uniform Crime Reporting (UCR): standardized offense counts published by participating law-enforcement agencies.
  • FBI National Incident-Based Reporting System (NIBRS): detailed incident-level records where the agency reports under NIBRS.
  • Local law-enforcement open data: city and county open-data portals where published.
  • US Census population estimates: used to compute rates per 100,000 residents so areas are comparable across population sizes.

Modeling approach

We aggregate incident records to neighborhood-scale sample points, then compute frequency and severity indicators for both violent crime (homicide, rape, robbery, aggravated assault) and property crime (burglary, larceny, motor-vehicle theft). The result is a single 0 to 10 probability score at each neighborhood-scale sample point (about 5 km²).

Output metrics include offense totals, rates per 100,000 residents, violent and property category scores, individual offense counts, a 3-year change indicator, and a 5-year trend slope.

Resolution

Scores are computed at neighborhood resolution: approximately 5 km² sample points across the contiguous US. This is much finer than the county-level averages most public crime data provides.

Update cadence

The model is refreshed annually as UCR, NIBRS, and local data publishers release the prior year’s records. Major model updates are versioned and disclosed.

Validation

Models are evaluated against held-out historical years and benchmarked across multiple reporting sources where available. The exact validation protocol is documented in the PerilScore technical papers.

Limitations

  • Forecast boundary. CrimeRate reflects the historical reporting record. Short-term incident patterns can change month to month.
  • Property-specific detail. Scores reflect a neighborhood-scale sample point. For a property-specific score that incorporates property type, security features, and lighting, use the free PerilScore app.
  • Source coverage varies. Some jurisdictions report under NIBRS, others under summary UCR, and some publish local open data. We document source coverage and credibility weighting for the local model area.
  • Reporting changes can affect the score. Increases in a category can reflect changes in reporting practice, including the NIBRS transition or new offense codes, alongside changes in underlying incidents. We surface this where it materially affects scoring.

Attribution

Risk scores powered by PerilScore. Visit perilscore.com for the full platform, API access, and commercial-use licensing.

Methodology

Public data. Real science. No black boxes.

Every score is computed from decades of public weather records using physics-based probability modeling. It's the same approach used by insurance and risk management professionals.

Decades of public weather data

Hurricane tracks, storm intensities, fire perimeters, hail reports, all drawn from public scientific archives. We don't use proprietary data. You can audit every input.

Physics-based probability modeling

Scores reflect how the actual peril behaves: wind fields, fire spread, ground shaking, and storm tracks. The model keeps the physics visible instead of flattening every place into a broad average.

Used by professionals

The same PerilScore data layer is used by insurance and risk management professionals. We publish it here so anyone can find authoritative risk numbers for their location.

Frequently asked questions

Where does the crime risk score come from?
Every score is computed from years of public incident records (FBI UCR and local law-enforcement reporting) aggregated to neighborhood-scale sample points (about 5 km² each), much finer than the county-level data most public sources publish. It’s the same data layer insurance and risk management professionals use for property risk assessment.
Is this score a forecast?
No. It’s a long-run probability score derived from years of historical incident reporting. It reflects the underlying risk environment, not a prediction for any single month or year.
Does this number apply to my exact address?
The score reflects the neighborhood-scale sample point at the location you searched. For a property-specific score that incorporates building, occupancy, and protection factors, use the free PerilScore app.
How is this different from a published crime map?
Public crime maps typically show recent incidents on a map. CrimeRate translates years of incident reporting into a single comparable 0-10 probability score per location, designed for risk modeling rather than situational awareness.

Want the full picture for a specific property?

The scores on this site show the representative crime layer for a local area. Enter a street address to add building age, construction type, roof details, occupancy, surroundings, and property-level context.

Check an address Browse local scores

Free results for any US street address.