Methodology
CrimeScore is a small-area statistical model. It is designed to help product teams present neighborhood context, not to make decisions about individual people.
CrimeScore produces a 0 to 100 Safety Score for Census block groups. Higher scores represent safer relative standing, and A is the best grade.
The deployed score combines violent crime, property crime, disorder, and frequency components. The API returns those component values with every successful score response.
Crime data comes from city and county open-data portals and is normalized into shared offense categories. Census predictors come from ACS 5-year estimates and related geographic data.
Race and ethnicity are blocked from model features and are not shown as score contributors. Contributor labels are descriptive model signals, not causal claims.
The current production path uses precomputed model estimates. Those estimates are converted into deployable scores with a blend of county-relative and national percentile ranks.
That blend helps the score stay locally useful while still preserving national context.
Police incident data reflects reporting and enforcement patterns, not total true crime prevalence. ACS data is lagged and may miss rapid neighborhood change. Some rural areas have weaker open-data coverage.
CrimeScore should not be used for predictive policing, individual risk scoring, lending or housing discrimination, or any use prohibited by fair-housing or civil-rights law.
FAQ
No. CrimeScore is a statistical location-risk model. It does not claim that a feature causes crime.
No. Race and ethnicity are blocked from model features and hidden from score contributors.
The score API resolves locations to Census block groups. Map layers also support county and state level browsing.