Stitches or No Stitches

The Los Angeles County Sheriff's Department has a statistical code guide, and in it is a rule for fistfights. If hands, fists, or feet cause broken bones, internal injuries, or a wound that needs stitches, the deputy codes an aggravated assault. A bruise or a minor cut makes it a simple assault. One counts as violent crime and one doesn't, and the difference can be a needle and some thread.

The county's Office of Inspector General quoted that rule in a September 2014 report, and then added the line that makes this whole subject hard. “It is not uncommon for a victim to refuse medical treatment even when it is obvious that their injuries require stitches,” the inspectors wrote. When that happens, the deputy makes a judgment call.

This post is about that judgment call. Aggravated assault is the largest piece of the violent-crime count in the United States, bigger than robbery, rape, and murder combined. It is also the one category whose boundary gets drawn by hand, at the scene, by whoever writes the report. If you build anything on a violent-crime rate, you are building on that line.

A category that isn't in the penal code

Start with an odd fact. California has no crime called “aggravated assault” and no crime called “simple assault.” The Sheriff's Department said so itself, in a November 2014 internal audit: “There is no crime within the California (State) Penal Code that defines ‘aggravated’ or ‘simple’ assaults.” The terms exist only in the FBI's reporting manuals, as a bridge between fifty different sets of state statutes.

So every assault gets classified twice. Once under state law, for charging. Once under the federal scheme, for counting. The second classification is the one that ends up in your dataset, and nobody gets prosecuted for it.

The FBI's NIBRSuser manual (version 2021.1) draws the line as code 13A versus code 13B. Aggravated assault involves a dangerous weapon, or an “obvious severe or aggravated bodily injury,” or a risk of serious injury. The manual lists what counts: apparent broken bones, loss of teeth, possible internal injury, severe laceration, loss of consciousness. Then it defines a severe laceration as “one that should receive medical attention.”

Should. Not did. (Pepper spray, for the record, is a weapon. So is a shoe, if somebody swings it.)

The manual also says local prosecution policy “should not influence classification.” We think that sentence is there because it so often does. Officers in most places are trained to write up a wobbler, a crime chargeable either way, as the felony. The felony version of a fistfight looks a lot like a 13A.

Los Angeles, 2014: the line moves down

In August 2014, the Los Angeles Times reported that the LAPD had recorded nearly 1,200 violent crimes as minor offenses in the year ending September 2013. Most were aggravated assaults filed as simple assaults. Corrected, the paper said, violent crime for that year would have been nearly 7% higher and aggravated assaults almost 14% higher. The LAPD's answer was that 1,200 was about 1.3% of the roughly 94,000 crimes it recorded that year.

Both numbers are true. That is the problem with this category. A tiny share of all crime can be a large share of one violent category, because the category is small and the misfiled cases all point the same way.

The Times went back further. In October 2015 it reported that the department had misclassified roughly 14,000 serious assaults between 2005 and 2012, enough to raise violent crime by 7% and aggravated assaults by 16% over that stretch. The reporters, Ben Poston and Joel Rubin, trained two machine-learning classifiers on the text of more than 400,000 crime records obtained through a public records request. The paper published its code and sample data on GitHub, at datadesk/lapd-crime-classification-analysis. (It is archived now, and still worth an afternoon.)

The LAPD's own inspector general then ran an audit. As later reported by the Times and NBC Los Angeles, that 2015 audit estimated more than 25,000 aggravated assaults misclassified as minor incidents from 2008 to 2014. In 2017, Capt. Lillian Carranza, then commanding the Van Nuys station, filed a claim alleging the problem had continued, and put the underreporting she found at roughly 10% of aggravated assaults across several divisions. The department said errors are corrected when found and that accusations about report accuracy would be investigated.

Same county, same year: the line moves up

Here is the part that almost nobody quotes. After the Times story, the county Board of Supervisors told its Office of Inspector General to check whether the Sheriff's Department was doing the same thing. The inspectors pulled 240 assault reports, forty from each of six stations, half coded aggravated and half coded simple. At least two staff members reviewed each one.

They found the error running the other way. Of the 120 simple assaults, four should have been aggravated. Of the 120 aggravated assaults, thirty-eight should have been simple. At Century and Compton stations, roughly half the sampled aggravated assaults looked like simple assaults to the inspectors. “Initial classification of assaults significantly overreport simple assaults as aggravated assaults,” the report concluded.

Two months later the Sheriff's Department released its own, larger audit: 907 reports from July 2013 through June 2014. Its auditors found aggravated assaults coded correctly 80% of the time and simple assaults 91% of the time. Of the ninety aggravated assaults that didn't meet the standard, fifty-three (59%) were domestic violence reports. The most-cited reason was the lack of a serious injury.

One more wrinkle, because it is instructive. The inspector general found no classification errors at the Marina del Rey station. The department's audit, a few months later, scored Marina del Rey's aggravated assaults at 55% correct, the worst in the county. Twenty-ish reports per station, two teams of reviewers, two answers. That is what small samples do.

Chicago: the label was fine, the count wasn't

Chicago's Office of Inspector General audited 2012 assault data and published in April 2014. Inspector General Joseph Ferguson's staff drew a random sample of 383 assault-related incidents and compared each narrative to its code. Twelve were wrong, or 3.1%. The FBI's training material treats 10% as acceptable, so on the question everyone was asking, Chicago passed.

The audit then found a different problem. Chicago counted aggravated assaults and batteries by incident report, while the Illinois program requires one offense per victim. A bar fight with three victims went in as one. In the sample, that rule alone undercounted victim offenses by 24.2%. A separate misreading of the rule for crimes against protected persons, police officers among them, cut reported aggravated assaults by another 5.7% and aggravated batteries by 3.2%.

Nothing was mislabeled. The arithmetic was wrong. We bring it up because a lot of people treat “downgrading” and “bad crime data” as the same subject, and Chicago shows the counting rules can move a number further than the labels do. It is the same family of problem as the Target store that doubled San Francisco's shoplifting: a procedure, not a crime wave.

What the line looks like in a public file

You can't audit narratives from the outside. You can look at the ratio. In any assault feed, divide aggravated assaults by all assaults and watch where that share sits and whether it moves. Here is the query we ran against the LAPD's legacy open-data file for calendar 2023, the last full year before the department's NIBRS handoff. Codes 230 and 236 are assault with a deadly weapon and intimate-partner aggravated assault. Codes 624 and 626 are their simple-assault counterparts.

curl -s -G "https://data.lacity.org/resource/2nrs-mtv8.json" \
  --data-urlencode "\$select=area_name, crm_cd, count(*) as n" \
  --data-urlencode "\$where=date_occ between '2023-01-01T00:00:00' and '2023-12-31T23:59:59' and crm_cd in ('230','236','624','626')" \
  --data-urlencode "\$group=area_name, crm_cd" \
  --data-urlencode "\$limit=500"

Citywide, that comes to 15,687 aggravated and 29,710 simple, so aggravated assaults were 34.6% of the total. By division, the share runs from 22.2% in West LA to 43.4% in 77th Street.

We want to be careful here. That spread doesn't show misclassification. Places with more guns produce more aggravated assaults, and South LA and the Westside are not the same streets. We left out codes 231, 235, 625 and the other assault variants to keep the query readable, and adding them shifts the shares a few points without changing the order.

What the ratio is good for is change. A division's aggravated share should drift slowly. If it drops five points in a quarter while its simple-assault count rises by about the same number of cases, something happened to the line, not the streets. That is the fingerprint the Times was chasing, and it is visible in public data that anyone can pull. The same signal shows up when an agency migrates records systems and a dropdown changes.

Why this matters right now

According to the Real-Time Crime Index, which tracks 565 agencies covering about 113 million people, murder fell 16.3% in January through July 2026 compared with the same months of 2025. Robbery fell 13.3%. Aggravated assault fell 2.9%.

We don't know why aggravated assault is lagging, and we'd be skeptical of anyone who says they do from these numbers alone. It could be real. Domestic violence, which made up most of the Sheriff's over-coded cases in 2014, doesn't track the street-crime cycle the way robbery does. It could also be partly paper. Murder has a body and a death certificate. Aggravated assault has a deputy, a cut, and a victim who may or may not go to the hospital. In our view, the categories with the softest boundaries deserve the widest error bars, and aggravated assault has the softest boundary of the violent four.

If you're building on assault data

Five things we do, or would tell you to do, with any assault feed:

1. Track the aggravated share per agency over time. Store it as its own series. Flag any quarter where it moves more than a few points and the total assault count doesn't.
2. Don't rank agencies on aggravated assault alone. The LA County audits are enough reason. Show all assaults next to it, so a user can see when one place is simply coding up.
3. Read the NIBRS fields, not just the code. A 13A with weapon code 40 (personal weapons) and an injury of N (none) or M (apparent minor injury) is sitting right on the line. A rising share of those inside an agency's 13As is worth a look.
4. Check the counting unit. Incidents, offenses, and victims are different numbers, as Chicago found. Know which one your feed reports before you compute a rate.
5. Widen the interval at small scale. A block with four aggravated assaults this year and two last year tells you about two judgment calls. See the enforcement confound for the other way a human decision ends up in the count.

We skipped the long history of downgrading allegations in New York, Milwaukee, and elsewhere. They're well documented, but the Los Angeles and Chicago audits are the ones that publish their samples, and a sample is something you can check.

One last detail from the 2014 inspector general's report. When the Sheriff's Department changed a code after the fact, it was supposed to file a report documenting why. The inspectors looked at seventy-one reclassified assaults. Two had the report.