Crime Rates per Capita

Extruded height = reported offenses per 100,000 residents. Surface colour = population density. FBI Uniform Crime Reports.

Averaging window

Sums offences and population across neighbouring years before dividing, which is not the same as averaging the rates. Rare crimes cannot be steadied by geography alone — a region of 100,000 people expects about 6.5 murders a year, so its murder rate carries roughly 39% error however the boundary is drawn. Three years is three times the events. The window narrows at each end of the series.

Geographic area

A region is a group of neighbouring counties in one state, merged until it holds about 100,000 people. Small counties swing wildly year to year — under 10,000 residents the median violent rate moves 45% a year and the murder rate 100% — and grouping them makes a trend visible. Large counties are left alone. Region and county data are fetched on first click; if a fetch fails, Status says so and clicking again retries. ZIP and census tract have no nationwide source at all; see Data notes.

Crime type

Year

2019

1960 SRS ▏NIBRS 2019

Height exaggeration

1.0×

Fade below

off

Dims everything shorter than this so the tall units stand out. Measured against the fixed scale, so it does not drift with the year or with height exaggeration.

Height shows

Crime rate scale

Both are fixed across all years, so the year slider shows real change.

Density colours

Data notes & caveats

Sources. 1960–2019: FBI Summary Reporting System state totals via the CORGIS State Crime dataset. 2020 onward: pulled live from the FBI Crime Data Explorer API. Rates are reported offences per 100,000 residents.

The 2019/2020 seam is real. The two eras are different collection systems (SRS vs NIBRS). Monthly CDE counts are summed to annual totals and divided by that year's population; years without all 12 months reported are omitted entirely rather than shown as a partial-year collapse. Treat cross-boundary comparisons with care.

Reporting coverage. NIBRS adoption caused a sharp drop in participating agencies — nationally about 75–79% of the population was covered in 2021, against ~96% by 2023. States below 80% coverage are drawn de-saturated and semi-transparent, with the exact figure on hover. A short pillar in those years may mean under-reporting rather than less crime.

County view. Built by summing the counts individual police agencies reported, from UCR Offenses Known and Clearances by Arrest (Return A), 1960–2024. That construction carries real hazards, which is why each county-year shows the share of its population that actually reported: city police and county sheriffs operate in the same county; an agency that reported nothing looks identical to one with no crime; and agencies that straddle county lines or serve no single county — state police, multi-county districts — are excluded entirely. See Kaplan on county-level data before drawing conclusions from a single county.

What the county layer excludes. An agency-year without all twelve months reported is dropped rather than summed, because a part-year total renders as a crash that never happened. Across 1960–2024 that removes about 337,000 agency-years. Aggravated assault is the aggravated count only — the Return A "total" column bundles simple assault, which is roughly ten times larger.

Why not ZIP or census tract? ZIP and census-tract selectors are disabled because no public nationwide dataset carries this offence breakdown at those levels — they are published city-by-city, never nationally.

Rape definition change. The UCR revised its rape definition in 2013. Figures before and after that year are not directly comparable.

Voluntary reporting. Agencies report to the UCR voluntarily. Coverage varies by state and year, and a low number can mean low crime or low reporting. Earlier years are considerably less complete.

Population density is computed from each state's true spherical land+water polygon area, so it will differ slightly from Census land-only density figures.

Status

What is actually loaded right now. A layer that fails says so here rather than just not appearing.