What makes CrimeGrade™ scores accurate
Three things make CrimeGrade™ scores hold up at a business address. Every model is tested on places and months it never saw. Every estimate is made block by block. Every crime is measured against what it targets.
Tested on what it never saw
Each crime's model is judged on whole metro areas and later months left out of its training, the same test a new market puts it to.
How the models are testedEstimated block by block
CrimeGrade™ estimates crime for each hexagon about 66 meters on a side. A heatmap smoothed from neighborhood totals gives every block the same average.
See the differenceThe right yardstick
Shoplifting is measured against retail floor space, vehicle theft against parked cars and street robbery against the time people spend in public.
See every yardstickTested on places and months it never saw
Each crime's model learns from police records across the country. To show how well it predicts crime somewhere new, CrimeGrade™ scores each model on whole metro areas and later months that were left out of its training. Each model keeps only the factors that improve it.
The police counts behind the models are checked first. Every report names the release it was built on, and CrimeGrade™ updates its data once a year.
The scores a team acts on have been tested on ground the models never trained on.
The same thefts, measured two ways
Take a park with 10 crimes a year and 2 residents. Per resident, it reads as one of the most dangerous places in the country. Measured against the thousands of hours visitors spend there, the same 10 crimes describe a safe park. Shopping districts and downtowns work the same way: few residents, many shoppers and workers.
The map on the left ranks Washington, DC the way most crime indexes do, by neighborhood and per resident. The map on the right shows the CrimeGrade™ theft score for each hexagon, measured against the time people spend in public and the cars parked nearby.
Theft per resident, by neighborhood
CrimeGrade™ theft score, by hexagon
A store reads on the risk to the people and cars that are actually there, block by block.
Each crime's yardstick, mapped
CrimeGrade™ divides each crime by what that crime targets, which it calls the crime's ideal denominator. Each map shows one yardstick across Washington, DC on the hexagon grid, from fewer (light blue) to more (dark blue). CrimeGrade™ builds the same layers for every hexagon in the country.
Retail floor area
Shoplifting follows the size of the sales floor, so it is measured per 1,000 square feet of retail floor area. Theft from businesses is measured the same way against all commercial floor area.
Businesses
Each business is a target, so robbery of a business is measured per 100 businesses. Burglary and arson of businesses are counted per 100 businesses too, and their scores use the household and housing-unit measures, like all burglary and arson.
Person-hours
The hours people spend in a place, at home, at work or out in public. Violent crime follows people, so its rate follows where people are, around the clock.
Estimated hour by hour for every hexagon, from where people live, work, shop, study and visit.
Public person-hours
The hours people spend outside their homes: on the street, at work, in stores and in parks. Street robbery happens in public, so time at home is left out.
Open parking spaces
The risk to a car follows the cars parked around it. CrimeGrade™ counts about 1.5 billion parking spaces nationwide in six classes: street, lot, driveway, home garage, public garage and airport. Vehicle theft is measured against the open ones, everything except home garages.
Theft from vehicles is measured per 1,000 parking spaces.
Households
Occupied homes. Most burglaries target homes, so burglary is measured per household.
Housing units
Homes, occupied or empty. Arson strikes buildings whether or not anyone lives in them, so the yardstick counts every home.
Housing units and businesses
Vandalism strikes homes and businesses alike, so its yardstick counts both.
Public person-hours and parking spaces
Theft takes from people and from cars, so its yardstick blends the two, about 64% people and 36% vehicles nationally.
Which number answers which question
Yearly counts in the rings
How much crime the 1-, 3- and 5-mile rings hold in a year. Counts size patrols and guard coverage.
CrimeGrade™ scores for each crime
Each crime on its ideal denominator, on one national 0 to 1000 scale. A score means the same level of risk in a city center and a rural county.
CrimeGrade™ store scores
A business compared with the typical store of its type, where 100 is typical and 200 about twice the incidents.
Every crime and its yardstick
The full list, with each measure's unit, is on the Data definitions page.
| Crime | Measured against | Why it fits |
|---|---|---|
| Shoplifting | Retail floor area | Shoplifting follows the size of the sales floor |
| Theft from businesses | Commercial floor area | Theft follows the size of the business |
| Robbery of businesses | Businesses | Each business is a target |
| Robbery of people | Public person-hours | Street robbery happens away from home |
| Murder, rape and sexual assault, aggravated and simple assault | Person-hours | Violence follows where people are, at any hour |
| Burglary | Households | Most burglaries target homes |
| Theft | Public person-hours and parking spaces | Theft takes from people and from cars |
| Vehicle theft | Open parking spaces | The risk to a car follows the cars parked nearby |
| Arson | Housing units | Counts the homes at risk, occupied or empty |
| Vandalism | Housing units and businesses | Vandalism hits homes and businesses alike |