Crime Data Without the Panic: Using the FBI's Explorer

Two towns went on my shortlist in the same week last year. Similar size, similar housing cost, six hours apart by car. One of them showed a violent crime rate a little under half the other's on the first aggregator page I landed on, and for about a day I quietly moved it to the top.

Then I went to look at the underlying data and found that the "safer" town's police department had not submitted a full twelve months. The rate was not low. The count was short.

That is the whole problem with crime numbers in a relocation search. Almost every other figure you pull for a candidate town — the assessor's tax bill, the utility's rate schedule, the flood zone — comes from an authority that is obliged to produce it. Crime data does not work that way. Reporting to the FBI is voluntary, and the number you end up staring at is the sum of whatever the agencies chose to send in.

So this piece is not about which towns are safe. I have no standing to tell you that, and the FBI itself explicitly warns against ranking places this way. It is about the three things to check before you let a crime figure move a town up or down your list.

Where the number actually lives

The FBI publishes through the Crime Data Explorer, at cde.ucr.cjis.gov. The old cde.fbi.gov address still gets passed around; the working host is the CJIS one.

It is a JavaScript application and the FBI reorganizes it between releases, so I am not going to promise you a button label that may be gone by the time you read this. What stays stable is the downloads page — cde.ucr.cjis.gov/LATEST/webapp/#/pages/downloads — which holds the flat files, including the participation data. When the interactive views confuse me, I go straight there.

Scale, so you know what you are working with. For data year 2024, according to the FBI's own summary published on the Explorer, 14,601 agencies covering 296,527,998 people submitted through NIBRS and another 2,074 agencies covering 28,572,514 people submitted through the older Summary Reporting System — 86.3% of agencies, 95.6% of the population. The 2025 figures went up on 14 August 2026, three days before I wrote this, and coverage was the widest since before the NIBRS switch: more than 96% of the population inside a reporting agency, with violent crime down about 9% year over year.

Ninety-six percent sounds like near-total coverage. It is — nationally. It tells you nothing about whether your town's department is inside it.

Check one: did this agency report, and for how many months

This is the check that would have saved me a day of wrong ranking, and it takes about four minutes.

Find the specific agency. Not the county, not the metro — the department that polices the address you are considering. Then look for its participation record: which years it submitted, and how many months in each of those years.

The threshold to care about is twelve. The FBI's own rate tables are built from agencies with 12 months of complete data, but plenty of downstream sites scrape whatever count exists and divide by a full-year population anyway. An agency that reported eight months and had its total divided by twelve months' worth of people produces a rate roughly a third too low, and nothing on the page will say so.

Three patterns worth flagging in your notes:

• Zero months. The agency is absent that year. Any site showing you a number for it is estimating, and you should find out from what.
• A partial first year. Very common in the year an agency switched systems. An analyst who follows this closely, Jeff Asher, documented the Los Angeles Police Department turning NIBRS on in May 2024 with theft and motor vehicle theft not fully captured at first. If a big-city department can produce that, a 40-officer town department can.
• A jump that lines up exactly with a system change. Not a crime wave. A counting change. More on that below.

A partial year is visible before you open a chart at all. The Explorer's agency lookup — cde.ucr.cjis.gov/LATEST/agency/byStateAbbr/IL, plain JSON, no key — returns a nibrs_start_date for every agency in the state, and it is frequently not 1 January. Read on 18 August 2026, it puts the Cook County Sheriff's Office (ORI IL0160000) at 1 December 2022 and the Los Angeles Police Department (CA0194200) at 1 May 2024 — one month and eight months, at the outside, in those first years. The same response carries an is_nibrs flag that comes back false for 1,194 of the 1,439 Pennsylvania agencies listed, so swap the two-letter code for your candidate state before you assume there is anything there to read.

Check two: 2021 is a wall in your chart, not a trend

On 1 January 2021 the FBI switched off the Summary Reporting System and required everyone to report through NIBRS. Many departments were not ready.

The result, per the Congressional Research Service's write-up on NIBRS participation rates and data quality: about 7,000 of roughly 18,000 agencies did not get data through for 2021, and the agencies that did covered around 65% of the population. The New York and Los Angeles police departments submitted nothing at all. The FBI then reversed course and began accepting SRS submissions again to patch the gap, which is the only reason those 2,074 SRS agencies are in the 2024 count at all.

Two consequences for anyone drawing a ten-year line for a candidate town.

The dip may be a reporting dip. If the town's department is one of the ones that went dark in 2021 and came back in 2022 or 2023, your chart shows a valley and a recovery. Neither happened on the street.

The rise may be a counting change. SRS used the Hierarchy Rule: in an incident involving several offenses, only the most serious one was counted. NIBRS records the offenses in an incident rather than collapsing them to one. Same incidents, more recorded offenses. A department that switched systems can post a higher total for a year in which fewer things happened.

Recovery has been real, and it has been uneven in a way that is geographic rather than national. The Bureau of Justice Statistics NIBRS page sets the two numbers next to each other for May 2024: all 50 states and the District of Columbia are certified to report to NIBRS, but only 82% of the population is actually covered by a NIBRS-reporting agency. Even among the big departments the gap is visible — 125 of the 154 agencies serving places of 250,000 or more were reporting, so 29 were not.

BJS publishes that coverage as a state map. Find your candidate state on it before you decide how much of the surrounding context you can see at all.

Check three: the denominator

The third failure is arithmetic, and it is the one that produces the "most dangerous city in America" headlines.

A city police department polices the incorporated city. Its numbers cover the city limits. So the population you divide by has to be the city's — the Census Bureau's place estimate, which lives in the SUB-EST2025 file on the City and Town Population Totals: 2020–2025 page, Vintage 2025, released May 2026, with estimates running through 1 July 2025. Pulling that estimate lands you in the same trap from the other direction: one town name can return an incorporated place, a county subdivision and a metro area, which is the geography problem I worked through in four Census tables worth pulling.

What people do instead is compare a city-limits crime count against their mental image of "the area," which is the metro. Places with tight historic boundaries get punished for it. St. Louis is the standard teaching case: the city is a legally independent jurisdiction holding a small share of its metro population, so the city police department's numerator sits over a denominator that excludes most of the region. Pull the two figures yourself — the place estimate from SUB-EST2025 and the metro estimate from the Census metro tables — before you accept any comparison that says "St. Louis" without saying which one it means.

Also check who actually polices your address. Unincorporated land is usually the county sheriff, not the nearest city PD, and that is a different agency with a different participation record and a different population base.

The pull sheet

One row per candidate town. Everything here is free, and it takes under half an hour once you have done it twice.

That last row is not filler. The FBI revises prior-year figures as late submissions arrive, so a number you copied in March may not match the same cell in September. Write the date next to it or you will not be able to explain your own sheet in three months.

CDE will give you its own half of that date. cde.ucr.cjis.gov/LATEST/refresh-date returns one short object of load dates; on 18 August 2026 it reported the UCR and police employee files last refreshed 15 August, LEOKA on the 17th, use-of-force on the 14th. With both dates in the sheet — when the FBI last loaded the file, when you copied the cell — a number that moved stops being a mystery.

What this tool cannot answer

The Explorer's smallest unit is the reporting agency. A mid-size city is one number. Whether the street you are looking at resembles that number is a question the federal data cannot touch.

For that you need the department's own open data portal or public crime map, and those follow local rules rather than federal ones. Some states also run their own UCR portals that are more current than the federal release — Texas at txucr.nibrs.com and Illinois at ilucr.nibrs.com are two I have used. Check whether your candidate state has one; it often carries monthly data well ahead of CDE.

How far ahead is worth checking, and it takes one request. Chicago's Crimes – 2001 to Present sits on a Socrata portal that answers ?$select=max(date); its own metadata promises daily updates covering everything but the most recent seven days, and on 18 August 2026 the newest incident it held was 9 August. Austin's Crime Reports is labelled weekly and ran through 15 August on the same day. Days behind, against a federal file whose headline release lands once a year — the catch being that every city names its fields its own way, so nothing you build for one town transfers to the next.

Why I stopped using crime as a screening filter

I would stop treating crime as a screening filter. I used it as one early on, cutting towns off the list on a single figure, and at least one of those cuts was made on a partial-year count I never looked into.

Now it sits at the end of the process. Cost, tax and insurance decide the shortlist, because those are numbers an agency is obliged to publish and I can verify to the dollar. Crime data comes in afterward, as context on a town that has already cleared everything else — and only after I have written down how many months that agency actually reported.

Sources, read 17 August 2026 and rechecked 18 August: FBI Crime Data Explorer (2024 agency and coverage figures from the Explorer's UCR Summary of Reported Crimes in the Nation, 2024); FBI caution against ranking; Congressional Research Service, NIBRS participation and NIBRS: Benefits and Issues; BJS NIBRS page; Census SUB-EST2025.

Next action: open the Explorer, find the exact agency for one candidate address, and write down a single number — how many months it reported in the most recent published year. If it is not 12, you are done for today. The rate on that page is not usable yet.