The Number the FBI Isn't Allowed to Print

Page three of the FBI's June 2025 newsletter for crime-reporting agencies sits between a list of data deadlines and a phone number in West Virginia. It says the bureau's use-of-force collection now covers 77.6 percent of the country's estimated 937,000 sworn officers, from 11,955 agencies. It does not say how many times those officers used force. The FBI isn't allowed to.

The reason is a condition written into the collection before the first report came in. It has shaped every national figure on police use of force since. If you build anything that touches police data, it's worth knowing exactly where the line sits, because the line is a number and the number keeps moving.

A table with four rows

The FBI's National Use-of-Force Data Collection is voluntary. No law required the bureau to build it and no law requires a department to send anything. When the FBI asked the Office of Management and Budget to approve the collection in September 2018 (the Paperwork Reduction Act generally requires that sign-off for federal data collections), OMB approved it with strings attached.

The strings are a four-row table. The Government Accountability Office reprinted it in a December 2021 report, and it's the whole story in miniature:

OMB added one more condition. If the collection didn't reach 60 percent by the end of 2022, the FBI was to shut it down and try something else. OMB officials told GAO they set the bar high because use-of-force data are “highly influential” and so warrant a high standard of quality.

Notice what the unit is. Participation isn't measured in agencies, or in incidents, or in people. It's measured in officers. A department counts toward the threshold in proportion to how many sworn officers it employs, whether or not any of them ever fires a weapon.

What gets counted, and what doesn't

The collection asks about a narrow slice of police force. An incident qualifies if an officer discharges a firearm at or in the direction of a person, or if the force results in death or serious bodily injury. For each one, the agency reports the date, time and place, the type of force, the reason for the contact, and the age, sex, race and ethnicity of the officers and people involved.

Everything below that bar is invisible here. A takedown with no serious injury is out. So is pepper spray, a Taser deployment that leaves a bruise, or a pointed gun that never goes off.

That matters more than it sounds. Andrew Wheeler, reviewing the research on police oversight this year at CrimeDe-Coder, found some of the clearest evidence in exactly that lower tier: departments that required officers to document pointing a firearm, or that tightened rules on pepper spray and conducted energy devices, saw measurable changes. In his view, simple standardized forms for lower-level force are what make patterns visible over time. The federal collection was never designed to hold those forms.

Percentages without a pie

At 60 percent the FBI could publish shares, and it does. USAFacts walked through the 2025 release in January. Of the incidents reported that year, 54.3 percent began with a response to unlawful or suspicious activity. Traffic stops accounted for 12.9 percent and routine patrol for 7 percent. Across the five years of published data, about 30 percent of reported incidents ended in at least one death. Firearms were the most commonly reported type of force in 2025.

All of that is real information. None of it tells you how many. Thirty percent of what? You know the shape of the pie and nothing about its size, which means you can't compute a rate, compare a year to the year before, or say whether a region has more incidents than another. You can only say how its incidents are divided up.

USAFacts put the gap in one line: what the data can't show is “how often these incidents happen.”

One more figure from the same article is worth sitting with. In each month of 2025, around 1 percent of participating agencies reported at least one incident. The other ninety-nine percent, if they reported at all, reported zero. The FBI's newsletter warns about what happens when agencies skip even that: “cold spots” appear on the map because only some agencies are sending anything. In a dataset where nearly every honest submission is a zero, a missing zero and a real zero look identical on a map. Small numbers misbehave in ordinary crime feeds too, which is why we wrote a whole post about how little a three-incident block can tell you.

A finish line measured in officers

The participation history is short and lumpy. GAO reported 44 percent for 2019 and 55 percent for 2020. (USAFacts gives 46 percent for 2019. Participation figures get revised as late submissions arrive, so the two snapshots probably reflect different dates.) The Justice Department announced in June 2022 that the collection had crossed 60 percent, a few months before the shutdown deadline, and the FBI began publishing percentages. USAFacts reports coverage reached 80 percent in 2024 and slipped to 78 percent in 2025. In January 2024 the FBI switched to measuring participation over a rolling 12-month window, and as of USAFacts' January 2026 article the total counts still hadn't been released.

Here is the part that surprised us. The denominator moved. In 2021, FBI documentation cited by GAO used a total of 860,000 sworn officers eligible to participate. The June 2025 newsletter uses 937,000. The arithmetic, using only published figures:

officers covered, May 2025    0.776 × 937,000 ≈ 727,000
80% bar at 937,000            0.800 × 937,000 = 749,600
gap to the bar                                ≈  22,500 officers

same 727,000 against 860,000  727,000 / 860,000 ≈ 84.5%

agencies, 2025 (USAFacts)     12,035 / 19,277 ≈ 62.4%

Read the middle line with care. We don't know why the estimate grew by seventy-seven thousand officers. It may be real hiring, a broader definition of who counts, or a better census of small agencies. The two totals may not be built the same way, so 84.5 percent is not a claim about where the collection “really” stands. The point is narrower. When the denominator is an estimate, the threshold is an estimate too, and a collection can lose ground without a single agency dropping out.

The last line says something about who is missing. Sixty-two percent of agencies cover 78 percent of officers, which can only happen if participating agencies are larger, on average, than the ones that don't participate. The holdouts skew small. For a national total that's a modest problem. For anyone who wants to look at a rural county, it's the whole problem.

Who's counting instead

The public does have a count of one thing: deaths. Campaign Zero's Mapping Police Violence project, which assembles its database from news reports, public records and other sources, says in its 2025 year-end report that police killed at least 1,314 people in the United States last year, a 5 percent decrease from 2024.

That number and the FBI's percentages are not measuring the same thing, and you shouldn't splice them together. The Mapping Police Violence figure counts people killed. The FBI collection counts incidents, includes serious injuries and shots that hit nobody, and comes from the agencies themselves. A developer who divides one by the other has built a number that means nothing.

If you're building on this data

Most of our customers will never touch the use-of-force collection directly. Some will, especially teams building civic dashboards or police-accountability tools. A few notes from the way we handle partial-coverage feeds:

1. Store participation next to every figure.A percentage from the 2025 release should carry its coverage (78 percent of officers) and its reporting window in the same record. If the coverage field is empty, the figure shouldn't render.
2. Keep missing and zero apart. An agency that reported zero incidents and an agency that reported nothing need different values in your schema. Collapsing them to 0 is how a cold spot becomes a safe spot.
3. Use the shares as shares.“Traffic stops were 12.9 percent of reported incidents” is a statement about the collection. It isn't a rate per stop, per resident or per officer, and it can't be turned into one without a count the FBI hasn't published.
4. Write the scope into the label.“Deaths, serious injuries and firearm discharges” is the definition. Users will read “use of force” as everything, including the much larger volume of lower-level force that this collection excludes.
5. Go local for anything below national.Many departments now publish their own use-of-force dashboards or annual reports, often with a lower bar for what counts. They're inconsistent with each other, but each one is complete for its own agency, which the national file isn't. We made the same argument about body-camera records and about dispatch files: the agency's own export usually knows more than the federal roll-up.

If you want the background on how federal crime reporting works in general, our NIBRS explainer covers the system the use-of-force collection is bolted onto. And the zero-population ZIP post is about a different denominator with the same habit of quietly deciding what a number means.

Two and a half points

The FBI has the reports. They sit inside the same UCR program that produces the national crime figures, filed by twelve thousand agencies, most of them reporting zero, month after month. What it can publish is a breakdown: this share from traffic stops, this share fatal.

The newsletter that announced 77.6 percent also promised the bureau was “moving toward 80 percent.” At 937,000 officers, that's about twenty-two thousand more officers' worth of departments. Somewhere in the country there's a sheriff's office whose monthly zero would help get them there.