CrimeContext Methodology

The primary data source is the FBI Uniform Crime Reporting Program, accessed through the FBI Crime Data Explorer. The UCR program collects voluntary submissions from more than 18,000 federal, state, local, and tribal law enforcement agencies covering both violent (murder, robbery, aggravated assault, rape) and property (burglary, larceny-theft, motor vehicle theft) offenses. The FBI also publishes annual estimates that statistically adjust for non-reporting agencies.

Population denominators come from the U.S. Census Bureau's annual Population Estimates Program, used to compute per-capita rates per 100,000 residents. Where available, we cross-reference figures against the Bureau of Justice Statistics publications, which often provide the deeper context — including the National Crime Victimization Survey, the canonical source on unreported crime.

Underlying FBI and Census data are public domain. CrimeContext analysis, scoring, and editorial framing are our own.

We never display raw crime counts. A city of two million residents reporting 1,000 burglaries is fundamentally safer than a city of 50,000 residents reporting 500, but raw counts suggest the opposite. The standard criminological unit — incidents per 100,000 residents — is the only fair way to compare cities of different sizes, and it is the unit the FBI itself uses in published reports.

Every page on this site enforces this rule. Even when the underlying FBI release exposes raw counts, our derived metrics are population-adjusted before display. The only exception is the offense-share charts on city pages, which show the relative composition of a city's crime mix (for example, what share of property crime is motor vehicle theft) and are inherently rate-neutral.

Every city in our dataset receives a Safety Context Score on a 0–100 scale, which maps to an A–F letter grade. The composite weighs three signals:

• Per-capita violent crime versus the national average — 40% weight. Violent crime includes murder and nonnegligent manslaughter, robbery, and aggravated assault. The score uses a log-scaled curve so that a city at 2.5x the national rate does not look almost identical to one at 3x: doubling from 1x to 2x is a much bigger jump for residents than from 3x to 4x.
• Per-capita property crime versus the national average — 30% weight. Property crime covers burglary, larceny-theft, and motor vehicle theft. Same log-scaled normalization against the FBI national rate.
• 5-year trend direction — 30% weight. A linear regression across five years of total per-capita crime, capped between -20% (maximum boost) and +20% (maximum penalty). Cities with a clearly improving trend can score well even with above-average current rates; cities with rising trends are penalized even if their current snapshot looks fine.

The letter grades are: A (80–100, significantly safer than average with neutral or improving trends), B (65–79, below-average rates), C (50–64, near the national average), D (35–49, above the national average), and F (below 35, well above average — usually with worsening trends).

CrimeContext currently includes U.S. cities with at least 25,000 residents that report consistently to the UCR program. Smaller communities are sometimes covered in state and county aggregates, but the year-to-year noise at small populations makes per-capita rates unreliable as decision inputs — a single additional incident in a town of 5,000 swings the rate by 20 per 100,000.

Cities flagged by the FBI as having incomplete agency reporting in a given year retain their grade only if estimated totals are available from the FBI's statistical adjustment. Where coverage is too thin to estimate reliably, the city is excluded from the cohort for that year and noted in the source documentation.