How the Safety Index works

Data sources

1. City open-data sites. Each of CommunitySafe's 57 cities posts its own police reports on an official public website (for example: San Diego Police, Los Angeles Police, San Francisco Police, Chicago Police, New York Police, Seattle Police, Boston Police, DC Police, and Philadelphia Police. See the Cities directory for all 57.) We pull the reports straight from the same public source the city itself uses.
2. FBI Crime Data Explorer 2023. The national average we compare each city against. It is shown as the number of reports per 100,000 people, so big and small cities can be compared fairly. Violent (Persons): 379.5, Property: 1,934.1. Source: cde.ucr.cjis.gov. The FBI puts out new numbers every October, and we update ours to match.
3. US Census 2023-2024 population estimates. The population count we use to work out reports per 100,000 people. Source: census.gov.

How the Safety Index is computed

1. Count the recent reports. We pull every police report the city has posted recently (usually the last 30 to 180 days). We sort them into the two groups the FBI publishes a national average for: Violent (Persons) and Property. Other offenses (public-order ones like drug or weapon offenses) are still tracked, but they don't count toward the score because the FBI has no national average for them.
2. Estimate how many people live there. For a whole city, we use its real US Census 2023-2024 population. For a single neighborhood, we estimate the population in one of two ways:
   • By size (preferred): we take the city's population and give each neighborhood a share based on how much of the city's land it covers, using the city's official neighborhood map. This assumes people are spread out evenly. It is close enough to rank neighborhoods against each other, but it is an estimate, not a census count.
   • By share of reports (backup): when we don't have a neighborhood map, we start from the city's rate and adjust it by how busy the neighborhood is compared to the average. A neighborhood with an average number of reports gets the city's rate. One with twice as many reports gets twice the rate.
3. Turn it into a yearly rate per 100,000 people. We scale the recent count up to a full year, then divide by population and multiply by 100,000: (reports × 365 ÷ days counted ÷ population) × 100,000. This is the same unit the FBI uses to compare cities to the national average.
4. Compare it to the fairest baseline. A neighborhood's grade is compared to its OWN city's rate, not the national average. Here's why. The national average mixes together rural, suburban, and city areas. Cities tend to have more reports than rural areas, and busy neighborhoods have even more. So comparing a single city neighborhood straight to the national average would make almost every neighborhood look worse than it really is. Comparing it to its own city is fairer. We still show the national average too, so you can see how each city stacks up against the country. For a whole city's score, the national average IS the fair comparison, so that is what we use.

An adjustment for three cities

Three cities — Cleveland, New Orleans, and Las Vegas — only publish 911 calls rather than finished crime reports. Counting calls tends to roughly double or triple the numbers, because:

• One real crime often sparks several calls (the first 911 call, follow-ups, and so on).
• Many calls turn out to be nothing after police look into them (false alarms, mistakes).
• Some calls aren't crimes at all (welfare checks, traffic complaints). We filter those out, but the count still runs higher than a finished-report city.

So that those three cities don't look unfairly worse, we shrink their rate by a set amount before grading:

• Cleveland: × 0.35
• New Orleans: × 0.40
• Las Vegas: × 0.50

These adjustments come from studies that compared 911-call counts to finished crime reports. Each of these cities shows a clear “adjusted” badge on its score card so you can see the change. All other cities are left exactly as-is (no adjustment).

What the score IS

• A simple summary of how many police reports there have been per resident.
• Built to line up with the FBI's national rate per 100,000 people.
• Checkable: you can recreate every number from the sources above using the steps above.

What the score IS NOT

• Not a prediction of what will happen next.
• Not a count of all crime. It only counts crimes that were reported and published.
• Not a replacement for professional safety advice, 911, or local know-how.
• Not something to base housing, lending, insurance, or hiring decisions on. Reports pile up in areas that have long been treated unfairly. Using the score this way can break the Fair Housing Act.
• Not a label of any neighborhood as “dangerous” or “safe.” We count reports, not the character of a place or the people who live there.

Known limitations

• Reports take time to appear. Cities can take 7 to 30 days to post a report. The score reflects what has been posted, which may not be the very latest.
• Some areas report more than others. Where people trust the police, more crimes get reported. Where trust is lower, fewer do. So the score partly reflects how much people report, not just what actually happens.
• Neighborhood population is an estimate. Splitting people up by land area assumes everyone is spread out evenly, which isn't true in cities with a packed downtown and spread-out suburbs. We show the estimate next to the score so you can judge it.
• Cities sort crimes a little differently. Most cities use the FBI's standard Persons / Property / Other groupings. A few (notably Chicago) sort some crimes differently (for example, counting robbery as Persons instead of Property). We use each city's own sorting, so keep that in mind when comparing cities.
• Other offenses don't count toward the score. The FBI has no national average for public-order offenses, so they can't be part of the score. You can still see them in other tabs (Crime Map, Trend Feed).

Demographic data exclusion

CommunitySafe does NOT collect, show, or analyze race, ethnicity, religion, age, gender, or sexual orientation from any source. Before we pull data from a city, our code lists exactly which fields it is allowed to read, and demographic columns are dropped no matter what the city publishes. This is built into the code. See src/server/services/crime-data/adapters/ for the list of allowed fields per city.

Individual identification

CommunitySafe only groups data down to the neighborhood level. We never name, identify, or track individual people. On the Crime Map, each dot shows the type of offense and the block where a report was filed. It never shows victim or suspect names. CommunitySafe posts are anonymous and pass through a check that blocks names, exact addresses, and license plates.