Domestic: True

Chicago's public crime file has a column called domestic. It holds one of two values. In 2025 it said true on 22,647 of the city's 42,670 batteries, which is 53 percent. It also said true on 2,832 thefts and 14 narcotics cases.

That column is one of the few places an outsider can test a claim now going around police circles. On October 3 the Police Executive Research Forum published a note under the headline “Violent crime is falling. Intimate Partner Violence is not.” It may well be true. The evidence behind it is harder to find than the sentence suggests, and most crime feeds couldn't confirm or refute it if you asked them to.

Four definitions, one word

PERF's support for the headline is a survey of its members, most of whom said intimate partner homicides and assaults weren't declining at the same pace as other violence. Some said theirs were rising. That is informed testimony from people who answer these calls. It isn't a count, and PERF concedes the difficulty: at its September meeting, Lynn Addington of American University pointed out that “domestic violence” is defined differently across data sets.

Here is how differently, using the four sources a developer is most likely to touch.

• Chicago's flag follows the Illinois Domestic Violence Act, which covers spouses, relatives, dating partners, caregivers and “persons who share or formerly shared a common dwelling.” A fight between two former roommates can qualify. So can one between sisters.
• NIBRS has no domestic field at all. The 2025 technical specification gives Data Element 35, a two-letter relationship code (BG, XS, SBand two dozen more), and it must be left blank when the offender is unknown. Anyone who wants “domestic” has to choose which codes count.
• The FBI made that choice in its February 2026 special report. Children and parents are in. Siblings, grandparents and in-laws are out. Simple assault is out too, since the report covers only murder, rape, robbery and aggravated assault.
• The National Crime Victimization Surveyseparates intimate partner violence (spouses, boyfriends, girlfriends, current or former) from domestic violence (those plus other family). It also includes simple assault, and it can't include homicide, because it works by interviewing victims.

So “intimate partner” survives in just one of those four as a published category. In the Chicago file you can't pull it out at all. Every Chicago number below is domestic in the Illinois sense, and we would not relabel it.

Run the query yourself

With that caveat attached, the file does let you do something rare: split one city's assaults and batteries by the flag and watch both sides for a quarter century.

curl -sG "https://data.cityofchicago.org/resource/ijzp-q8t2.json" \
  --data-urlencode "\$select=date_extract_y(date) as y, domestic, count(*) as n" \
  --data-urlencode "\$where=primary_type in('ASSAULT','BATTERY')" \
  --data-urlencode "\$group=y, domestic" \
  --data-urlencode "\$order=y, domestic"

Between 2001 and 2025, flagged incidents fell 47.1 percent and unflagged ones fell 49.7 percent. Over twenty-four years the two lines move almost together. Since 2019, the flagged side has actually done better. Here is the recent stretch, with “aggravated” meaning any description containing AGG.

Read the top two rows and PERF's headline looks wrong for Chicago. Read the third and it looks right. The share of domestic incidents coded as aggravated went from 11.4 percent to 18 percent in ten years, while the same share on the other side of the flag stayed near 27 percent the whole time.

We can't tell you why from this file. Severe violence may have grown while lesser incidents shrank. Victims with minor injuries may have stopped calling. Officers and supervisors may have changed how they code the same injury, which is the problem we covered in Stitches or No Stitches. We didn't fit a model, because one city and a handful of annual counts wouldn't earn one. In our view the honest reading is that the aggregate hides a real divergence in severity, and the cause is open.

One note if you rerun this. Chicago renamed its descriptions along the way, so AGGRAVATED: HANDGUN became AGGRAVATED - HANDGUN. An exact match on the old spelling returns 714 flagged assaults in 2019 and none in 2025. Match on the pattern, or you will report a collapse that happened in a text field.

The percentage that rises when nothing does

The national figure most often attached to this argument is a share. The FBI's report says violent crimes involving domestic relationships went from 25.6 percent in 2020 to 27.5 percent in 2024. That's accurate, and it tells you less than it seems to.

Chicago shows why. In 2020 the domestic share of assaults and batteries jumped from 48.4 percent to 50.5 percent, its highest point in the file. Domestic incidents that year fell 11 percent. Other incidents fell 18.3 percent. A share goes up whenever the other side falls faster, and in a decade when street violence is the thing falling, a rising domestic share is close to what you would expect by arithmetic alone.

The FBI's window has a second wrinkle. It uses only NIBRS data, and those five years are when the national program went NIBRS-only, with some of the largest agencies arriving late. We couldn't find a fixed panel of agencies described anywhere in the report. Their mix of cities changed underneath that percentage, by how much we don't know.

Fifty-four people

Homicide is the count nobody can argue away, so it is where the claim sounds strongest. Chicago's homicide total fell from 597 in 2024 to 441 in 2025. Rows flagged domestic went the other way, from 44 to 54. The unflagged count dropped 30 percent, the flagged count rose 23 percent, and the domestic share went from 7.4 percent to 12.2 percent. You could write that story in January and it would be correct in every digit.

Then look at January through September, comparing like with like. Flagged homicides went 33, 43, 24 for 2024, 2025 and 2026. The share went 7.2, 12.6, 7.3. Every one of those is a person, and none of them is a trend. A move from 44 to 54 is about one standard deviation under a Poisson assumption, the same issue we walked through in Three Burglaries Is Not a Trend. And the 2026 figure may still move, since a relationship usually can't be recorded until someone knows who the offender was. (NIBRS makes that explicit: unknown offender, blank field.)

If you are comparing years, use a multi-year average and print the interval. Your city's number is probably smaller than Chicago's and noisier.

A count that moves when the violence doesn't

Every police number here passes through two decisions before it becomes a row. Somebody has to call, and somebody has to write it down as domestic. Both decisions move.

The Bureau of Justice Statistics' survey, in Criminal Victimization, 2024, has the share of domestic violence reported to police going from 47.7 percent in 2023 to 64.5 percent in 2024, a statistically significant jump in one year. Then BJS tested a redesigned questionnaire on the same year. The two instruments' estimates nearly match on how much intimate partner violence happened: 782,900 victimizations on the old form, 782,470 on the new. They disagree on how much reached the police: 60.8 percent versus 40.3 percent.

That last figure comes from a 2024 Council on Criminal Justice paper by Alex Piquero and Andrew Wheeler, which pairs NIBRS with survey reporting rates. Note the middle figure as well. The survey doesn't show intimate partner violence diverging from stranger violence since 2020. It shows both climbing off a pandemic low, on estimates with wide margins, which is awkward for anyone citing it on either side. The gap between survey and police counts is the subject of our dark figure post.

The second decision is the officer's, and it can push the count in either direction. PERF's own note says that when the police chief in Rathdrum, Idaho, began requiring a report for every allegation of physical violence, the department's domestic numbers rose. Denver saw a similar rise after opening its family justice center. “What we have is an increase in reported domestic violence,” Commander Matt Clark said. Better police work, higher line on the chart.

It runs the other way too. One PERF participant described investigators reconstructing seven or eight years of calls because many had never been coded as intimate partner violence. And New York Focus reported last month on disciplinary files covering 18 State Police employees with substantiated domestic abuse allegations. In one 2019 case a responding sheriff's deputy declined to write the domestic incident report state law requires, later saying under oath that he considered it a “civil matter.” (The husband, a trooper, has said he acted in self-defense.) No feed will ever carry that report. We ingest police data for a living and we have no way to see a row that was never written.

If you build on this data

Two notes before the list. The flag matters even if you never chart domestic violence, because it changes what your other numbers mean. And the location field makes it a privacy question. Of Chicago's flagged assaults and batteries in 2025, 75.9 percent were recorded at an apartment or residence. For unflagged ones it was about a fifth.

1. Carry the definition with the field.Store which statute or code list produced a domestic flag. Don't merge Chicago's flag with a NIBRS-derived one under a single name, and read our taxonomy post before you try.
2. Publish counts next to shares.A share with no counts beside it can't separate more of one thing from less of another.
3. Split by severity. In Chicago, 82 percent of flagged assaults and batteries in 2025 were simple. A trend in that total is mostly a trend in calls for lesser incidents.
4. Hold agencies constant. For anything built on NIBRS, compare only agencies that reported across the whole window. The NIBRS transition alone can manufacture a trend.
5. Treat blank as unknown.A missing relationship means no offender was identified or nobody recorded one. It doesn't mean a stranger did it.
6. Think hard before scoring an address on it. In our view a safety grade should report flagged incidents separately rather than dropping them or folding them in. Separating them says nothing about seriousness: the FBI found no injury for 37.4 percent of domestic victims against 60.9 percent of others. It says the risk to a person on that block differs from the risk to a person in that home.
7. Don't map the dots. We would not ship a public layer of domestic=true points, even at block level. On a quiet block, it points at a household. See our notes on suppression.

Which brings us to Dallas. Its Police Incidents file has a column named family, documented as “Yes or no if the offense is family violence.” We queried it this morning. It says false on 1,400,785 rows and is blank on 139,517. The count where it says true is zero.

Those rows may be withheld before publication, which would be a defensible choice. The column doesn't say. Ask it for family violence in Dallas and it returns an empty set.