U.S. city crime data, ready to query

What is in the sample?

Four modeled indexes: major crime, persons crimes, property crimes and social disorder. An index of 100 is the U.S. rate per resident; 200 is twice it. These estimates are not a count of police reports or a safety rating.

The files also carry each place's high, medium and low confidence shares. A share of 0.395 means 39.5% of residents live in block groups with that confidence level; it is not a probability that the city estimate is correct.

Columns and types

Use place_geoid as the join key, stored as text. A city name can occur more than once. Parquet preserves types automatically; when importing CSV, keep the leading zeros in IDs such as Davis's 0618100.

place_geoidstring

Seven-character Census place GEOID; retain leading zeros.

place_namestring

Place name; names are not unique geographic keys.

statestring

Two-letter state abbreviation, including DC.

place_typestring

Census place type; blank where the source omits it.

residentsinteger

Resident population used by the model; not people present.

major_crime_indexnumber

Modeled persons plus property crimes per resident, U.S. = 100.

persons_crime_indexnumber

Modeled murder, rape, robbery and aggravated assault per resident, U.S. = 100.

property_crime_indexnumber

Modeled property crimes per resident, U.S. = 100.

disorder_indexnumber

Modeled social disorder per resident, U.S. = 100; separate from major crime.

confidence_high_sharenumber

Share of residents in high-confidence block groups, 0 to 1.

confidence_medium_sharenumber

Share of residents in medium-confidence block groups, 0 to 1.

confidence_low_sharenumber

Share of residents in low-confidence block groups, 0 to 1.

model_releasestring

Source model release identifier; not a police-record coverage date.

source_urlstring

Canonical CrimeMetric place page.

Download the machine-readable schema. Empty numeric fields indicate an unavailable estimate.

Coverage, method and limitations

This sample contains the places that pass CrimeMetric's publication quality gate in this release. It includes cities, towns, villages and census-designated places, but does not cover every U.S. place. It contains no ZIP-code crosswalk, street addresses, individual incidents or neighborhood boundaries.

The model combines FBI agency totals and local incident records at census block-group level, then aggregates the estimates to places. Per-resident indexes can be high in places with many visitors and few residents. Compare places of similar size and read the confidence shares alongside the index.

The model release date is distinct from the police-record dates shown on individual city pages. This download stays fixed when the website receives newer incident records.

Method and validation·Data sources·Reading the index

Use and attribution

The sample license permits your own research and commercial analysis, with attributed charts, findings and selected figures. Standalone resale or redistribution of the dataset requires a separate agreement. Keep the release and confidence information available, and do not use the data for decisions about individual people.

CrimeMetric modeled estimates, release 2026-2026-11-01. Vorp Labs. https://crimemetric.com/data

For address-level queries, block-group detail and current API responses, see the CrimeMetric API. Its terms are separate from this sample license.