FBI Crime Data Explorer (CDE)

One country, tens of thousands of police agencies, one shared vocabulary for what got taken and when. FBI Crime Data Explorer is the Federal Bureau of Investigation's official presentation of the Uniform Crime Reporting program: offense counts collected from law enforcement nationwide and organized so an agency, a county, a state or the whole country reads side by side. Summary Reporting System tables carry series that reach back to the 1960s depending on the table; incident-based NIBRS collections widen from 1991 onward; hate crime, arrests, LEOKA and annual estimates round out the shelf.

Two structural facts matter most. First, geography runs deep: national roll-ups, all fifty states and territories, counties, tribal and university jurisdictions, and individual agencies identified by their Originating Agency Identifier - so the same offense question answers at whatever zoom the analysis needs. Second, these are aggregates, not incidents: rows carry counts by agency, year and offense category rather than one row per burgled house. For anyone modeling demand for alarm installation and monitoring, that aggregation is the point - a census-shaped baseline, not a case file.

Because burglary, robbery, larceny-theft and motor vehicle theft sit among the Part I offenses reported per agency per year, the archive reads as a map of where property crime concentrates. Get a sample of this dataset to see the full field dictionary and the rows behind it before you commit.

One row per agency, year and offense - the shape a delivery lands in:

# one row per agency x year x offense - row shape shown; every value pinned to live records in your sample
ori         : <reporting agency identifier>    state_abbr : <two-letter state code>
data_year   : <calendar year>                  population : <covered residents>
offense     : burglary-breaking-or-entering
actual      : <offenses known to police>

ori         : <reporting agency identifier>    state_abbr : <two-letter state code>
data_year   : <calendar year>                  population : <covered residents>
offense     : motor-vehicle-theft
actual      : <offenses known to police>

Three things to notice. First, the grain is deliberate: counts, not incidents, which keeps the table small enough to sweep nationally yet sharp enough to separate one suburb from the next. Second, population rides on the same row as the count, so per-resident rates need no second lookup. Third, ori is the join key - the stable agency identifier that lets demographics, housing stock or competitor locations attach without fuzzy matching.

The rows above show record shape; a requested sample pins every value to live records.

Six fields define every row, dividing cleanly into three jobs. Identity: ori names the reporting agency, state_abbr locates it. Time and subject: data_year fixes the calendar year, offense carries the UCR category. Measure: actual holds the count of offenses known to police, and population supplies the covered residents that turn counts into rates.

Additional fields on request. The collections behind the aggregates run deeper than the six-column core: incident-based segments with victim-offender pairings and property-loss values, hate-crime bias motivations, LEOKA circumstances, arrest counts by demographic slice, and the participation rosters that say which program each agency reports under in which years. Each folds into your delivery once named, so the dictionary you build on is the dictionary you tested.

• Geography: the entire United States, nested. National roll-ups at the top; all states and territories; counties; tribal and university jurisdictions; individual agencies at the bottom, each with its own identifier. One frame covers a corner-store catchment and the whole country.
• Temporal: depth varies by table. Summary series reach back to the 1960s depending on the table, incident-based participation widens from 1991 onward, and everything runs current through the latest published year. Decade-scale trend work and recent-vintage comparisons draw from the same well.
• Granularity: offense counts aggregated by year, jurisdiction and offense category - agency-year-offense rows. Incident-level texture sits in the NIBRS segments and folds in on request.

That combination - full nesting, long memory, one grain - is why this set anchors the security-alarm pack: it is the denominator everyone else argues against.

API, files, or your warehouse. Daily, weekly, or hourly.

You pick the channel, the cadence and the scope; the six-field rows above travel unchanged through all three. Every delivery ships the complete field dictionary, the sample rows and the coverage profile already mapped, so a national sweep lands as one predictable table rather than a stitching project. Agency identifiers arrive normalized, counts typed as integers, populations attached - rate math becomes a WHERE clause instead of an afternoon. Cadence changes are a settings conversation, not a re-integration project, and a scoped sample comes first either way.

• Market researchers and consultants size the alarm installation and monitoring market from the offense base, cutting burglary and theft counts by state, county and agency catchment. See the market researchers use cases page for the workflow patterns.
• Competitive intelligence and product teams explain monitoring-demand shifts with the crime trend underneath them, benchmarking rival footprints against where property crime actually concentrates. The competitive intel product teams use cases page has the playbook.
• Sales and growth teams rank territories by property-crime intensity before committing dealer-recruiting budget. The sales growth teams use cases page covers the motion.
• Investors and quant researchers feed decades-long offense series into models linking incidence to security spending and subscriber economics. The investors quants use cases page shows the patterns.
• Data scientists and ML engineers treat the agency identifier as the entity spine joining counts to permits, registries and firmographics. The data scientists use cases page details the joins.

Market researchers get the cleanest denominator in the industry pack: offense counts that nest from nation to single agency without a stitch in sight. Competitive intelligence teams get context - the trend line explaining why monitoring contracts appear and lapse where they do. Sales and growth teams get a ranking machine: property-crime intensity per resident sorts territories without another vendor spreadsheet. Quant researchers get sixty years of memory in one schema. Data scientists get a stable join key that survives scope changes from one metro to the whole country. Across all five, the constant is the same: a census-shaped aggregate beats a thin proprietary panel whenever the question is where property crime concentrates.

Three things, all knowable upfront. First, the grain is aggregates: if your model needs incident-level texture, say so and the NIBRS segments fold in - but the headline counts stay counts, which is exactly what makes national sweeps cheap. Second, temporal depth differs by table and by program; summary series reach deepest and incident-based participation widens from 1991, so name your years and the sample tells you exactly how far back they go. Third, agency participation shifts over decades - the participation rosters fold in on request, so your trend lines rest on agencies that actually reported in the years you are reading.