San Francisco crime rates

San Francisco crime rates in South of Market

Incidents per 1,000 residents, counted the S.F. Chronicle's way · source updated Oct 9, 2026.

South of Market had 40.3 reported violent crimes per 1,000 residents in the twelve months to Sep 2026, rank 1 of 38 neighborhoods, 4.3× the citywide rate, and no clear change from the prior twelve months. 1,039 incidents on 25,795 residents. Reported crime here depends on an officer arriving and writing a report: in this area, 39% of Priority B police calls in the last twelve months closed as gone on arrival or unable to locate (citywide 33%), and such calls almost never produce an incident report.

Method

Source: SFPD Incident Reports: 2018 to Present (wg3w-h783 on DataSF). The counts follow the S.F. Chronicle's method, so the figures can be compared directly. Only initial reports count, because supplemental reports would count an incident twice. Rows are counted as published. Violent crime is assault plus robbery. Property crime is burglary plus larceny theft plus motor vehicle theft. Homicide is left out of these counts because it is often filed confidentially. It is counted separately, one per victim, from the list of every homicide since 2018 that SFPD's Homicide Unit released in response to a public records request. Each row has the case number, the date SFPD counts it on, the block or intersection, and the police district. Each block is placed at the middle of its street segments on the city's street map (3psu-pn9h on DataSF), and each intersection at the city's own point for it. The homicide then counts in the neighborhood and supervisor district the city assigns to that place. Police districts are the ones SFPD recorded. The list matches SFPD's quarterly victim data (cd9v-umhr on DataSF), which has the same counts as its CompStat reports, in every police district and quarter it covers. On the police district map in effect on the day, 378 of the 382 victims fall in the district SFPD recorded or on a street at its edge. The list runs through Oct 4, 2026, so homicide figures are shown only for periods that end by then. Rates divide by the 2020–2024 American Community Survey five-year population estimate, the same source and years the Chronicle uses. The Chronicle's published neighborhood populations match the survey's tract totals to within 0.2%, and every rate on this site uses this estimate. The survey publishes estimates down to the block group, and census blocks (on DataSF) are used to fit them to each area. A neighborhood is the sum of its tracts. An association area drawn along streets, or a police district, takes whole blocks, because streets form the edges of blocks. A supervisor district or benefit district takes each block group's estimate in proportion to the share of it inside the boundary. Where a boundary cuts through blocks, as benefit district boundaries do, the resident count could be off, and a rate is shown only when the largest possible error is under 10%. Areas under 1,000 residents (the parks) show counts but no rate. Twelve months means complete months, compared with the twelve before, as the Chronicle counts them. The 12-week and 4-week periods end two weeks before the newest report, because reports keep arriving after an incident. For property crime, about one in eight reports is dated more than a week after the incident. The year-earlier figure for these periods counts only reports filed within the same time after that period, so both sides are equally complete. Rates for these shorter periods count incidents per 1,000 residents during the period itself. Every comparison is tested before the page states it. An area is compared with the city by testing its count against what the rest of the city's rate would give its residents (an exact Poisson test). A period is compared with a year earlier by testing the two counts against each other (an exact binomial test). A difference is called higher, lower, up or down only when chance would produce one that large less than 5% of the time. Otherwise it is shown in gray as within normal variation. When neither count reaches 10, the figure is not compared. With many areas and categories on one page, about one in twenty differences that are only chance will still pass the test. Reported crime is not all crime. Confidential reports (about 6% of all reports, over half of them domestic violence) and incidents involving juveniles are withheld from the public data, and a single store can swing a neighborhood's larceny count.

Every figure here is also in https://sfresponse.org/data/crime.json.