In San Francisco jail data, about 9 in 10 Hispanic bookings are recorded as “White”
Anyone who counts San Francisco jail bookings by race alone sees almost no Hispanic people and a White count close to twice its real size.
Drafted by Claude (Anthropic’s AI) at Eric Niehaus’s direction, from his research question. Not yet reviewed by Eric Niehaus. All figures come from official aggregate data and published reports, not from person-level records. The 2025 and 2013 booking figures were re-checked by a second automated query against the DataSF API on October 31, 2026, and matched; other years and the SFPD report figures were not re-checked.
of Hispanic people booked into San Francisco’s jail in 2025 sit inside “White” in the race field (an estimate from totals). The share has been 87–90% every year since 2019.
The claim
The Sheriff’s jail booking data have a race field with no Hispanic option and a separate ethnicity field that does have one. Any count built from the race field alone moves about 89% of Hispanic bookings into “White”. That over-counts White bookings by about 79% and makes Hispanic bookings disappear.
The data
2025 San Francisco Sheriff bookings (14,897), from DataSF dataset pnsi-p43w, official aggregate counts.
| Group | Race field only | Best estimate of true share | Gap |
|---|---|---|---|
| White (best estimate: non-Hispanic White) | 53.9% | 30.1% (range 28–31%) | +23.8 points (+79%) |
| Hispanic / Latino | 0% (no such value) | 26.7% (range 25–32%) | −26.7 points (−100%) |
| Black | 34.1% | 33.7% | +0.4 points |
| Asian / Pacific Islander | 5.9% | 6.1% | −0.2 points |
How the 89% is computed: (White in the race field − non-Hispanic White in the ethnicity field) ÷ Hispanic total = (8,035 − 4,486) ÷ 3,978 = 89.2%.
The pattern holds every year
| Year | Bookings | White, race field | White, ethnicity field | Hispanic, ethnicity field | Hispanics inside “White” |
|---|---|---|---|---|---|
| 2013 | 23,917 | 42.9% | 32.7% | 14.9% | 68% |
| 2016 | 18,262 | 45.7% | 31.4% | 19.5% | 73% |
| 2019 | 17,540 | 51.0% | 29.7% | 23.5% | 90% |
| 2022 | 10,498 | 53.2% | 26.3% | 30.2% | 89% |
| 2024 | 12,320 | 54.4% | 28.7% | 28.9% | 89% |
| 2025 | 14,897 | 53.9% | 30.1% | 26.7% | 89% |
| 2026 (Jan–Sep) | 12,410 | 53.2% | 29.9% | 26.9% | 87% |
Three sources agree on the Hispanic share
| Source | Hispanic | White | Black | Asian |
|---|---|---|---|---|
| Sheriff bookings, ethnicity field (2025) | 26.7% | 30.1% | 33.7% | 6.1% |
| SFPD arrests, 96A report (Q2 2025) | 26.6% | 34.1% | 30.1% | 5.4% |
| SFPD arrests, 96A report (Q4 2024) | 28% | 31% | 30% | 6% |
| Police stops ending in arrest, officer-perceived (Jan 2024–Jun 2026, 16,577 people) | 27.7% | 30.1% | 32.2% | 4.7% |
It used to be worse
SFPD’s arrest system dated from 1972 and allowed only black, white or other. In 2010 SFPD reported Hispanic people as about 1.5% of its arrests. After a 2012 news investigation (KQED / Bay Citizen), SFPD said it would record race and ethnicity separately. The Sheriff’s ethnicity field put Hispanic people at 14.9% of bookings in 2013, about ten times the 2010 SFPD figure (a different dataset, so the comparison is rough). Any long-term San Francisco trend that includes years before 2013 is unreliable for Hispanic and White.
What this shows, and what it does not
- Shown: counts that use the race field alone misstate the racial makeup of San Francisco bookings by a large and stable margin.
- Shown: the ethnicity field already records the Hispanic share, and two other sources agree with it. The distortion appears wherever someone reports race alone, as many national datasets do.
- Not shown: that anyone does this on purpose. The most likely cause is form design and the federal rule that treats Hispanic as an ethnicity, not a race. The fix is cheap: always report race and ethnicity together.
- Not shown: whether the ethnicity field itself misses some Hispanic people. National research finds prisons record only about two-thirds of Hispanic people correctly. Testing this in San Francisco needs name-level booking records, which are not public.
- Estimate, not a match: the 89% comes from totals, not from matching each person’s two fields. The public data have no person-level crosstab. Other categories also differ between the two fields (Unknown is 770 in race and 294 in ethnicity), so the true share may be somewhat lower or higher.
The strongest objection
“This isn’t mislabeling. Federal rules say Hispanic is an ethnicity, and Hispanic people can be White.” Agreed on the rule. The problem is the use: a race-only count is read as “White people”, and it mixes two groups with very different booking rates. The finding is that race-only counts mislead, not that the jail lies.
Other objections, in brief: bookings are not arrests or crimes (true; they are reported here as bookings). Officer perception at stops is not fully independent evidence (agreed; it is a check, not proof). All three sources come from police and jail staff (true; a name-level surname check is the next step).
Next
The same check, run the same way, for every California county, the 50 largest US cities and every state, with one scorecard per agency: recorded versus corrected shares, how complete the ethnicity field is, and the sources. Rates per agency only; no individual is ever labeled.
Sources
- DataSF: Sheriff Jail Bookings (pnsi-p43w); metadata for bookings by race (5zbm-xnsp) and by ethnicity (36n6-w97s)
- DataSF: Police Department Stop Data, RIPA (ubqf-aqzw)
- SFPD 96A quarterly reports: Q2 2025; Q4 2024; report index
- SF District Attorney Justice Dashboard guide (2020)
- KQED / Bay Citizen: SF police underreport arrest rates for Latinos, Asians (2012); CBS SF (2012); SF Public Press (2012)
- RIPA regulations, 11 CCR 999.226
- Census QuickFacts: San Francisco County
- Finlay, Luh and Mueller-Smith, Race and Ethnicity (Mis)measurement in the U.S. Criminal Justice System (NBER, 2024)
- ProPublica: Hispanic drivers disappear from Louisiana traffic records (2021)