How the estimates are made
Starting point: what police agencies report
Most police agencies report their crimes to the FBI. Those agency totals, by crime type, are the base of the numbers on this site; where an agency doesn't report, its totals are estimated from its own past reports or from similar agencies. Disorderly conduct and trespassing are modeled, because the FBI records them only through arrests.
Allocating crime to neighborhoods
A model spreads each agency's totals across the 2020 census block groups it serves, using residents, jobs, daytime population and points of interest, and the local incident records an agency publishes where they exist. A city's figures add up the block groups that lie mostly inside it, so they can differ from the totals of the agency that polices it. The index compares crimes per resident with the U.S. rate: 100 is the U.S. rate and 200 is twice it.
How accurate it is
The model is tested on places it was not trained on. For each place that publishes its own incident records, the model is fit without that place's records, and where it puts crime is compared with what the records show. City totals come from FBI agency reports, so this tests the part the model adds: where crime falls within a place.
Across 1,123 places and 16 crime types (6,120 tests), compared with spreading each place's crime by where people live:
| CrimeMetric | By residents | |
|---|---|---|
| Crime placed in the wrong neighborhoods (lower is better) | 30% | 43% |
| Crime found in the tenth of neighborhoods ranked highest | 33% | 16% |
| Rank agreement with actual neighborhood crime (1 is perfect) | 0.59 | 0.15 |
"Crime placed in the wrong neighborhoods" is the share of a place's recorded crime that would have to move between block groups (those mostly inside the place) for the estimate to match the records. Each test is one place and one crime type with at least 30 recorded incidents, and the figures are averages with every test counted equally.
By size of place
| Crime placed in the wrong neighborhoods | CrimeMetric | By residents |
|---|---|---|
| Cities of 250,000+ | 36% | 48% |
| 100,000 to 250,000 | 34% | 47% |
| 25,000 to 100,000 | 30% | 44% |
| Under 25,000 | 20% | 32% |
| Sheriff and unincorporated areas | 39% | 48% |
Larger places spread crime over more neighborhoods, so there is more room to misplace it. These rows use places whose records are crime reports rather than calls for service.
Uncertainty
Each neighborhood estimate comes with an 80% interval and a confidence level. Indexes are withheld for block groups with fewer than 100 residents, and expected counts are still published.
Figures for the model updated Oct 20, 2026 (release 2026-2026-10-20-2).