Look up the same ZIP on two crime-rating sites and you will often get numbers that seem to contradict each other. One says 42, another says 98. One calls it moderate, the other extreme. Neither is necessarily wrong. They are answering different questions, and the differences come from six choices every rating has to make.
1. Which direction the scale runs
This is the first thing to check and the most common source of confusion. Some ratings score safety, so 100 is best. CrimeScore scores crime intensity, so 100 is the most reported crime. A 42 on a safety scale and a 98 on a crime scale might be describing the same place. Read the axis before you read the number.
2. Which incidents are counted
A rating that counts only violent crime, or weights it heavily, will rank a shopping district with frequent theft below a residential area with rare assaults. A rating that counts every report does the opposite. CrimeScore weights incident types by severity, so a robbery counts for more than vandalism, but every type counts. A vendor that will not tell you its weighting is asking you to trust an opinion.
3. Which data feeds it
Some ratings are modeled from agency-level totals, the annual numbers a department reports for its whole jurisdiction, spread across neighborhoods with statistical assumptions. Others are built from individual reported incidents with locations. The second is harder and more honest about local variation. CrimeScore is built entirely from incident reports collected by SpotCrime since 2007, more than 60 million of them, and the model uses no demographic, economic, or housing variables at all.
4. Which boundary
There is no official ZIP boundary. Ratings use Census ZCTAs, some use their own approximations, and some score smaller units like block groups and then average up. Two boundaries for the same code will contain different incidents. CrimeScore uses 2020 ZCTAs, consistently, every month, and the ZIP versus ZCTA post explains the difference.
5. Which comparison
A score is only meaningful against a baseline. Against the whole country, a mid-size city's downtown looks busy. Against other downtowns, it might be average. CrimeScore uses a national percentile for ZIPs against all scored ZIPs, and for counties against all scored counties, and shows a state rank alongside so you have both comparisons.
6. Which month, which model
Crime data lags, and models change. A rating with no date and no version cannot be reconciled with anything. Every CrimeScore carries a snapshot month and a model hash, and ZIP pages keep the monthly history. If two CrimeScores for one ZIP differ, the month or the hash will tell you why.
How to choose
Ask what question you are trying to answer.
- "What happened near here recently?" is an incident question. Use a map of reports, such as SpotCrime.
- "How does this area compare with others?" is a score question. Use a rating with a stated baseline, a date, and a version.
- "Is this address safe?" is a question no ZIP-level rating can answer, from anyone. The area is too large and the word too loaded. Be suspicious of any source that pretends otherwise.
When two ratings disagree, list their answers to the six choices above. In our experience the disagreement dissolves into "different scale, different counting, different boundary" almost every time. What remains is the choice of which question you actually needed answered.
The full description of CrimeScore's choices is on the methodology page.