Why the FBI's Switch to NIBRS Made National Crime Numbers Harder to Read

Imagine a federal agency telling the country that it has retired a ninety-year-old measuring stick, replaced it with a better one, and that several thousand police departments have not yet found where they put the new one. That is not a joke about government. That is, in outline, what happened to American crime statistics on January 1, 2021, and every pundit who has ever waved a national crime chart at you owes you an explanation of it.

I want to give you the explanation, because it matters to anyone who argues about crime for a living or at the dinner table. The switch from the old Summary Reporting System to the National Incident-Based Reporting System, known as NIBRS, was a good idea executed unevenly. The result is that comparing 2019 to 2022 on a national chart is harder than it looks. A conservative who understands why is harder to fool, and harder to embarrass.

The definitions you need first

Three terms carry this whole story.

Uniform Crime Reporting (UCR) is the FBI's program for collecting crime statistics from local, county, state, tribal and federal law enforcement agencies. Participation by the agencies is voluntary. Congress has never ordered a city police department to report to the FBI, and that fact explains most of what follows.

The Summary Reporting System (SRS) was the original UCR method. An agency counted the number of offenses it knew about each month, in a handful of categories, and sent in the totals.

The National Incident-Based Reporting System (NIBRS) collects a record of each incident: the offenses, the victims, the offenders, the arrestees, the location, the weapon, the property involved. It is a database of events, not a tally of totals.

The first is the program, the second is the old way, the third is the new way. Keep them straight and the rest is arithmetic.

A short history of counting crime

Before 1930 there was no national crime count at all. Police chiefs, through the International Association of Chiefs of Police, saw the problem: every city defined burglary and robbery a little differently, and no one could say whether the country was getting safer. A committee of the association worked out uniform definitions and a reporting scheme. In 1930 the FBI took over the job of collecting the reports, and it has run the program since.

The original design counted seven "Part I" offenses, chosen because they were serious, frequent and likely to be reported to the police: criminal homicide, rape, robbery, aggravated assault, burglary, larceny-theft and motor vehicle theft. Congress added arson to the list later, in 1979. Everything else, from fraud to drug offenses to simple assault, fell into "Part II," and for those the program collected only arrest counts.

For the better part of a century, this was the national crime record. The famous "crime index," the number reporters cited in the 1970s and 1990s, was built from Part I offenses reported by agencies that chose to participate. It had flaws that were obvious to statisticians from the start, and the FBI said so in its own publications. It was still the only game in town, and the trend lines it produced told a recognizable story: a long rise from the 1960s, a peak in the early 1990s, a long decline after that.

The hierarchy rule and why it had to go

Here is the part of the old system that does the most damage to anyone who wants a detailed picture. The SRS used what the FBI called the hierarchy rule. When a single incident involved several offenses, the agency counted only the most serious one.

A worked example, labelled as a hypothetical: suppose a man breaks into a house at night, steals a television, and assaults the homeowner who comes down the stairs. That is a burglary, a theft and an aggravated assault in one incident. Under the hierarchy rule, the agency reports one offense, the most serious, which is the aggravated assault. The burglary and the theft vanish from the national count. Three crimes happened. The statistics recorded one.

There were exceptions: arson was counted alongside whatever else happened, and motor vehicle theft was counted separately. But the rule was the rule. It was a sensible shortcut when the data moved by paper and mail. It was a poor fit for an age of databases, because it threw away information that a database could easily hold.

NIBRS drops the hierarchy rule. Each incident can carry multiple offenses, and every one is recorded. It also records who the victim was, what the relationship was between the victim and the offender, where the crime happened, and what weapons were involved. If you want to know how many robberies in a given year involved a firearm and a stranger, NIBRS is built to answer that. The SRS could not.

Premise one of the case for NIBRS, then, is simple: it counts what happened, and the old system counted what was most serious about what happened.

Group A and Group B

NIBRS reorganized the offense list too. The old Part I and Part II distinction became Group A and Group B.

Group A offenses are the ones for which NIBRS collects full incident detail. The list is far longer than the old Part I: as the Congressional Research Service describes it, it covers more than seventy offenses across more than two dozen categories, including the old index crimes but also fraud, embezzlement, counterfeiting, extortion, drug and narcotic violations, weapon law violations, human trafficking, and several offenses against children and families.

Group B offenses are a short list of lesser offenses, such as disorderly conduct, driving under the influence, liquor law violations and trespass, for which NIBRS collects only arrest data, not full incident reports.

For a reader who wants a broader picture of crime, this is real progress. The FBI also revised the national definition of rape in 2013, replacing the narrow old "forcible rape" language with a broader definition. Add that to the move to NIBRS and you can see why you cannot casually line up a rape figure from 2010 against one from 2022 and call it a trend. The definition of the thing being counted moved.

The deadline, and who missed it

The FBI announced in 2016 that it intended to make NIBRS the national standard, and it set January 1, 2021 as the date when the SRS would be retired. That gave local agencies a few years. The deadline proved too tight for many: for its 2022 and 2023 data the FBI again accepted submissions in both formats.

Now remember that reporting is voluntary and that switching costs money. To report in NIBRS, an agency needs a records management system that can capture incident-level data in the NIBRS format, staff trained to enter it, and a state program, in most states, that collects the data from local agencies and forwards it to the FBI. A small sheriff's office with a vendor that updates its software on schedule might make the change without noticing. A large metropolitan department with a decades-old system, a union contract, a budget fight and a city council with other priorities might not be ready on a fixed date.

That is what happened. When the FBI published its 2021 crime data, the share of agencies that had made the transition was well short of full coverage, and the agencies that had not were not a random sample. The Congressional Research Service reported that the FBI received 2021 data from 9,881 of 18,818 agencies, about 53 percent, and that major departments such as the New York and Los Angeles police departments did not participate. Because big cities carry a disproportionate share of violent crime, a gap there is not a small gap in the national picture.

Participation figures were described differently as the FBI published later releases, so check the current numbers yourself in the FBI's Crime Data Explorer and the program's own documentation. What matters is the mechanism: voluntary reporting plus an unfunded technical mandate plus a fixed deadline equals a patchy first year.

How the FBI filled in the gaps

The FBI has always estimated national totals. Even in the best years of the old program, not every agency reported, so the published national figures were estimates built from reporting agencies and adjusted for the missing ones, using the population and type of the agencies that did report.

After the 2021 transition, the FBI faced the same problem with a heavier lift. It could not produce its traditional national estimates for that year and instead published a trend analysis built on the agencies that did report. In plain language, the bureau used the reporting agencies to infer the picture for those that sent nothing, and attached confidence intervals to show how imprecise the result was.

An estimate like this is only as good as the match. If the missing agencies are unlike the reporting ones, say because the largest cities have different crime patterns than the typical reporting town, then the estimate will be off in a direction the method cannot see. The FBI was candid about the limitation. The Congressional Research Service noted that such estimates are subject to sampling and reporting error. Figures can also be revised as more agencies come in, which is the right thing to do and also the reason a number quoted in a news story from one year may not match the same year's number in a later table.

Premise two: a national total during the transition is an estimate with a margin the headline does not show.

The steelman: it is only a technical hiccup

Let me state the opposing view at its best. The defenders of how this was handled say that the switch was overdue, that a country cannot demand better data without tolerating a messy transition, that the FBI was transparent about gaps, and that by now the coverage keeps improving each year. Criticizing the numbers, they add, risks turning into a general distrust of statistics, which helps no one.

Every part of that is partly right. The switch was overdue. The FBI did publish its caveats. And a skeptic who responds to the transition by deciding that no crime statistic can be trusted has swapped analysis for a shrug.

But here is where the defense stops working. The people who need the numbers are not statisticians. They are voters, city council members and sheriffs deciding whether crime is rising. When national figures arrive with gaps and then get revised, and when two cable networks cite two different numbers for the same year, the public learns to doubt all of it. That is a cost, and it was entirely predictable, because voluntary reporting meant the federal government could not compel the big cities to finish on time. A better-funded, better-phased transition would have cost more up front and bought a lot of credibility.

The conservative principle here is not exotic. It is competence in the basic functions of government. Counting crime accurately is a core public job, closer to the Constitution's census clause than to any program that gets a press conference. If a government cannot tell its citizens reliably whether they are safer than they were, it has not earned a great many of the other things it asks them to trust it with.

What the Founders would recognize

The Constitution orders a count of the population every ten years, in Article I, Section 2, the "actual Enumeration." The Founders understood that self-government requires knowing the facts about the governed. They did not write a crime clause, because policing belonged to the states under the Tenth Amendment, and it still does. That is exactly why the UCR is voluntary. The federal government collects and publishes; it does not command the local police chief to hand over his records.

That is the right design, in my view, and it also explains the weakness. A federal system gets national statistics only by cooperation. The way to better data is therefore not a federal mandate on every town's police department; it is clear standards, reasonable grants for software and training, and a state program that does the heavy lifting for small agencies. The 1930 arrangement was built on cooperation and trust, and the 2021 gaps are what cooperation looks like when the cooperation is uneven.

How a careful reader compares years across the switch

You do not have to give up. You have to read more carefully. Here is a method, in numbered steps.

1. Start with homicide. Murder and nonnegligent manslaughter are the offense least affected by the change, because a death is hard to hide and is recorded by medical examiners and in health statistics as well. The Centers for Disease Control and Prevention's death records provide an independent check on the FBI's homicide totals. When the two sources point in the same direction, trust the trend.
2. Prefer same-agency comparisons. If a city reported in the same system before and after the change, compare the city to itself. Do not compare an agency's old summary count to its new incident count without checking how the agency's own documentation treats the break.
3. Know which offenses the hierarchy rule shrank. Under the old rule, a burglary that came with an assault was counted only as the assault. Under NIBRS, both show up. So counts of lesser offenses can rise under NIBRS with no change in the real world. The FBI has published research on how NIBRS and SRS compare when the same agency reports in both, and you should look it up before you claim a trend in burglary or larceny.
4. Check the coverage before the headline. Ask what share of the population lives in agencies that reported for each year. A change in coverage can mimic a change in crime.
5. Use the other instrument. The National Crime Victimization Survey asks households about crimes whether or not they were reported. It is a separate source with a separate design, and I discuss it in The Crime Survey That Asks Victims Instead of Asking Police Departments. When police data and survey data both point the same way, you have found something real.
6. Treat the first release as provisional. The FBI has revised its tables. Check the date on the table you are quoting.

The longer lesson on what crime numbers can and cannot support is in Seven Things Crime Statistics Cannot Tell You, and Two Things They Can. It applies with extra force across a break in method.

Why this matters for the arguments you will hear

Let me draw out the practical consequences, because they cut in every direction.

If someone tells you crime fell sharply in a given year, ask whether that figure comes from an estimate, and whether the agencies in the estimate are the ones that matter. If someone says crime is out of control, ask the same question. Both claims may be true or false; the point is that the transition made it easier than ever to quote the number that suits you, and harder than ever to be sure it is the number that is right.

Conservatives have a particular stake in doing this honestly. We argue that policy, from policing levels to bail rules to sentencing, has real effects on public safety, and we are right to demand evidence. The case against policies like those described in The Case Against Cashless Bail Starts With the Victim, Not the Defendant and What California's Proposition 47 Changed and What Proposition 36 Undid rests on specific claims about offenses and outcomes. If we get the baseline wrong, we hand the other side an easy rebuttal and we deserve it.

There is also a political lesson in who loses when the data is messy. A neighborhood that feels unsafe is not helped by an argument about whether the national index moved two percent. Residents know what they see. The statistics are meant to serve the public, not to overrule it, and a government that publishes patchy numbers has to expect the public to trust its own eyes over the table.

What good reform looks like

Here is what I would want, and what the record supports. First, funding tied to specific outcomes: grants that pay for the software and the training that let a small agency report, with no strings beyond the reporting itself. Second, state programs that act as the aggregator for their counties, so that a rural sheriff's deputy is not expected to run a data operation. Third, a transparent method for estimating missing data, published in plain language and revised openly. Fourth, a long overlap period in which agencies report in both systems, so analysts can build a bridge across the break instead of guessing at it.

None of that is exotic. It is the ordinary discipline of changing a system without losing the record. The country moved its crime count to a better system, and the better system will, in time, give researchers a richer picture than the old one ever could. But the transition was a reminder that the data we argue over is made by human institutions, with budgets and deadlines and gaps, and that the first duty of anyone who cites it is to say where it comes from.

The case in brief

1. The FBI's Uniform Crime Reporting program began in 1930 and relied on voluntary participation by local agencies, which is why it depends on cooperation rather than command.
2. The old Summary Reporting System counted only the most serious offense in each incident, which discarded information that NIBRS now keeps.
3. When the FBI retired the old system at the start of 2021, many agencies, including large ones, were not ready, so national figures for the transition years rest partly on estimates.
4. A careful reader leans on homicide, compares agencies to themselves, checks coverage, uses the victimization survey as a second source, and treats first releases as provisional.
5. Accurate crime statistics are a basic government function, and conservatives make the best case for law and order when they cite the numbers honestly.

What you can do: open the FBI's Crime Data Explorer, find your own county's agencies, and check whether each one reported under NIBRS for the years you care about before you quote a trend to a neighbor.