2022-01-01 to 2024-12-31 · 47,588 victims · Martin Ngoh
Black women's reported assault rate in Baltimore is about 2.5 to 3 times White women's, and age, police district and neighborhood do not explain it. How it compares with Hispanic women's cannot be settled: ethnicity is unknown for 42% of women victims.
Per 100,000 residents a year, the reported assault rate for Black women is 3,605: 1.6 times Hispanic women's, 2.6 times White women's, 5.6 times Asian women's.
Adjusted for age, year, police district, neighborhood socioeconomics, the gap against all other women is 2.84x (95% CI 2.57 to 3.15).
Women only (20,233 Black women, 4,876 other women). Each cut asks whether a plain explanation accounts for the gap.
Police data holds only what was reported. If Black women report more or less often than other women, every rate here moves, and this data cannot say which way.
How much of the gap survives adjustment, one control at a time. Poisson rate models at the census-tract level, women only, 5,010 tract by group by age by year cells, covering 96.7% of located Black women victims and 88.8% of other women.
A 2.84-fold gap remains after every measured control. None of the crude excess is explained: adjusted, the gap is wider than the crude 2.62x.
Residents are the denominator, so exposure away from home is unmeasured. Reporting behavior is invisible to police data. Nothing here measures offenders or circumstances.
Legacy SRS 2022 to 2024 against NIBRS January 2025 to August 2026 (September 2026 left out for reporting lag). 13B simple assault matched to 4E common assault; 13A aggravated assault to 4A to 4D plus shootings. Women victims, Hispanic of any race first by the ethnicity column, otherwise the recorded race, in both. A woman listed under both 13A and 13B in one case is counted once in the total, as aggravated (1 in this period). Neither file records victim type, so assaults on officers are in both.
The data says the reported assault rate for Black women is higher. It does not say why. Nothing here measures causes, offenders or circumstances.
Every number is a report that reached the police. Willingness to report, and recording practice, differ by group, area and time.
Rates count reports. Someone assaulted twice counts twice, so a rate is not the share of people assaulted.
Rates divide by where people live, not where they spend time.
Race is recorded by officers; the population counts people who are Black alone. There are 5% more people who are Black alone or in combination. If multiracial victims are recorded as Black, the worst case is 1.54x Hispanic, 2.46x White, 5.35x Asian instead of 1.62x, 2.58x, 5.62x.
1.2% of Black residents are also Hispanic, so they sit in both denominators.
2,093 victims are left out of the rates: their sex is unknown, or their race is unknown or outside the compared groups. Race is unknown or outside the groups for 2.3% of women and 5.4% of men.
Police reports, not how often women are hurt. In the national victimization survey, which counts assaults whether or not police learned of them, Black and White women describe being assaulted at about the same rate nationally and about 1.5 to 2 times in large cities. A follow-up (Police-Records-vs-Survey-Assault-Victims-by-Race-and-Sex-2015-2025, on GitHub) tested why police records differ more: not reporting rates, not (or only a little) how police write up a call, not the same women counted repeatedly, but largely where assaults happen and who calls. Hospital emergency departments, which do not depend on a call to police, see a gap like the police one (about 4.6 times for women in 2021 to 2022), which points to the survey undercounting assaults on Black women.
Police data reflects where officers patrol and who calls them. This data cannot separate more policing or more reporting from more assaults.
Baltimore records ethnicity apart from race, and began recording it in June 2021. It is unknown or blank for 41.6% of women victims. 41.9% of women recorded as White have unknown ethnicity, and where it is known, 18.1% of them are Hispanic. They are counted as White here. If they were Hispanic, Black women's rate would be 0.82x Hispanic women's instead of 1.62x, and 4.85x White women's instead of 2.58x. 934 women with unknown race but Hispanic ethnicity are counted as Hispanic.
523 women victims (2.0%) have unknown race. It is most common in the Southern district (3.0%) and least in the Western district (1.2%), and it is more common where fewer Black women live (correlation -0.76 across districts). That understates other groups' rates slightly, which widens the gap. Spread over the four groups like known victims of the same assault type in the same district, the ratios move to 1.61x Hispanic, 2.57x White, 5.61x Asian from 1.62x, 2.58x, 5.62x.
Baltimore redrew its police districts in 2023, and the data names the old district before mid-2023 and the new one after. Here every assault is placed in the 2023 district that holds its coordinates, for all three years, so one map is used throughout. 99.9% of victims are placed; 65 have no usable coordinates.
Police district and tract poverty, income, unemployment and housing are shaped by segregation and disinvestment. Here the adjusted gap (2.84x) is wider than the crude one (2.62x), so these controls do not account for it. Where a control narrows a gap, the gap has been located, not explained away.
No code or field marks assaults by a partner, so the partner test from the Los Angeles and DC analyses cannot run. In both, partner assault was close to half of assaults on women.
The data does not record the victim type, so assaults on police officers are counted with everyone else's. Los Angeles and DC left them out.
Common assault (code 4E) and aggravated assault (codes 4A, 4B, 4C, 4D, 9S, 3), which includes 1,739 non-fatal shootings Baltimore codes apart from aggravated assault. Homicides are left out. 148 American Indian or Alaska Native victims and 76 Native Hawaiian or Pacific Islander victims are counted in the totals but not compared, because counts this small give unstable rates. 396 women victims with unknown age fall out of the age test only. When an assault has several victims, each is counted.
Victims cover January 2022 to December 2024: full years with ethnicity recorded, before Baltimore police moved to NIBRS in January 2025. The population is the ACS 2024 5-year estimate, an average over 2020 to 2024.
The Asian comparison moved +39% between the legacy system (2022 to 2024) and NIBRS (January 2025 to August 2026) (+54% for aggravated assault), against 11% or less for the others. It rests on 151 Asian women victims in 2022 to 2024 and 59 Asian women victims in January 2025 to August 2026, so it is left out of the headline.
Rates are victims per 100,000 residents per year over 3.0 years; cells under 2,000 residents are not rated. Women only from the tests onward. Sources: Baltimore Police Department, Part 1 Crime Data (Legacy SRS), Open Baltimore; Baltimore Police Department, NIBRS Group A Crime Data, Open Baltimore (replication); Baltimore City police districts, 2023 boundaries, Open Baltimore; US Census Bureau ACS 2020 to 2024 five-year estimates via Census Reporter.