Home › Methodology
Methodology
Every number on this site is computed from public Census Bureau tables and a public boundary file from the City of San Francisco. This page lists the sources, the exact tables, and every step between the raw data and a neighborhood page, so you can check or reproduce any figure.
Sources
- Neighborhood and district figures
- American Community Survey (ACS) 5-year estimates, 2020-2024, at census-tract level. Earlier points use the 2015-2019 and 2010-2014 vintages.
- Citywide trends
- ACS 1-year estimates for San Francisco County, 2005 to 2024. There is no standard 2020 release.
- Full counts
- 2010 Census Summary File 1 and 2020 Census Demographic and Housing Characteristics file, at tract level.
- Boundaries
- San Francisco Planning Department Analysis Neighborhoods, 2020 census tracts assigned to neighborhoods, published on DataSF (dataset sevw-6tgi). The same file carries each tract's 2022 supervisor district.
- Tract changes
- Census Bureau 2010 to 2020 tract relationship file for California, used to bring older vintages onto current boundaries.
- Current market values and rents
- Zillow Research public data files: the Zillow Home Value Index (ZHVI) by Zillow neighborhood, ZIP code and city, and the Zillow Observed Rent Index (ZORI) by ZIP code and city. Currently August 2026.
- Inflation
- Consumer Price Index for All Urban Consumers (CPI-U), annual averages, from the Bureau of Labor Statistics via FRED.
Tables used
B01003 population; B25001 housing units; B25002 occupancy; B25003 tenure; B25004 vacancy status; B25008 population in occupied units by tenure; B25014 occupants per room; B25024 units in structure; B25034 year built; B25038 year householder moved in; B25063 gross rent distribution; B25070 gross rent as a share of income; B25075 home value distribution; B19001 household income distribution. Decennial: P1, H1, H3, H4 (2020) and P001, H001, H003, H004 (2010).
From tracts to neighborhoods
The Census Bureau has no neighborhood geography. Its smallest unit with usable housing data is the tract, and San Francisco has 242 of them. The Planning Department assigns every tract to exactly one of 41 Analysis Neighborhoods so that city departments can report census data consistently. This site uses that assignment as published, with no adjustments. Boundaries do not always match how residents describe their neighborhood: Dogpatch is inside Potrero Hill, Cole Valley inside Haight Ashbury, and so on. Those names redirect to the neighborhood that contains them.
4 of the 41 areas are excluded because they have fewer than 500 homes: Golden Gate Park, Lincoln Park, McLaren Park, The Farallones. That leaves 38 neighborhood pages.
Counts and rates
Counts such as housing units, renter households or vacant units are summed across a neighborhood's tracts. Rates are computed from the summed counts, never averaged from tract rates. Vacancy rate is vacant units divided by all units. Rental vacancy rate follows the Census definition: units for rent, divided by renter-occupied units plus units for rent plus units rented but not yet occupied. Rent burden is renter households paying 30 percent or more of income on gross rent, out of households for which the ratio could be computed. Overcrowding is occupied homes with more than 1.0 occupants per room. Average household size is population in occupied units divided by occupied units.
Medians
Tract medians cannot be averaged. Instead, each neighborhood's median rent, home value and household income is re-estimated from the ACS distribution tables: the households in each dollar bin are summed across the neighborhood's tracts, and the median is interpolated linearly inside the bin that contains the midpoint. When the midpoint falls in the survey's open top bin ($3,500 or more for rent, $2,000,000 or more for value, $200,000 or more for income), the page shows the bin floor with a plus sign, because the true median is above it. Medians for the citywide comparison row are computed the same way from all tracts, so the comparison is like for like. The citywide figures on the housing overview and home page are the Census Bureau's own published 1-year medians.
Margins of error
ACS estimates are survey estimates with a 90 percent margin of error. When counts are summed, margins combine as the square root of the sum of squared margins, the Census Bureau's documented approximation. Each count carries a reliability flag based on its coefficient of variation: high under 15 percent, medium to 30 percent, low above that. Pooled medians are flagged by the number of households in the distribution. Combining several tracts into a neighborhood roughly halves the relative margin compared with a single tract, which is the main reason this site reports neighborhoods rather than tracts.
Earlier years
Consecutive five-year ACS releases overlap by four years, so the site shows non-overlapping vintages only: 2010-2014, 2015-2019 and 2020-2024. The two older vintages are published on 2010 census tracts. Their values are allocated to 2020 tracts in proportion to shared land area using the Census Bureau's relationship file, then rolled up to neighborhoods. Where a 2010 tract was simply split in 2020, which covers most changes in San Francisco, this is exact. Where boundaries moved, it is an approximation. The 2010 Census counts are treated the same way. Dollar figures for earlier years are shown in nominal dollars and in 2024 dollars using the CPI-U annual average ratio.
Zillow market figures
Census rent and value figures describe everyone who lives in a neighborhood, averaged over five survey years, so they run well below the current market: a median rent includes tenants on decades-old rent-controlled leases, and a median home value is what owners guess their home is worth, capped at $2,000,000. To show what a home costs today, each page also carries two Zillow Research indexes for August 2026. ZHVI is Zillow's estimate of the typical value of homes in the middle third of the market, single-family and condo, smoothed and seasonally adjusted. ZORI is the typical asking rent on new listings, across all home types.
Zillow publishes ZHVI for its own neighborhood areas, which do not match the Planning Department boundaries. Each Zillow area is assigned by hand to the Analysis Neighborhood that contains most of it (Cole Valley to Haight Ashbury, Dogpatch to Potrero Hill, and so on), and a neighborhood's value is the plain average of the Zillow areas assigned to it. Each page lists which areas went in. Where no Zillow area fits, the value comes from ZIP code indexes instead. ZORI is only published by ZIP code, so every neighborhood's asking rent is the housing-unit-weighted average of the ZIP codes its census tracts fall in, with each tract assigned to the ZIP code that covers most of its land. Supervisor districts use the ZIP method for both. A neighborhood needs at least half its homes in a ZIP code with data to get a figure.
The asking rent premium is Zillow's asking rent divided by the Census median gross rent for the same area, minus one. Gross rent includes utilities and asking rent does not, so the true premium on a like-for-like basis is slightly larger than shown. One-year and five-year changes compare the same month in earlier years. Yearly charts use the August value of each year. Zillow data is used under its public research data terms, with attribution, and is not affiliated with this site.
Maps and charts
Maps are drawn from the tract polygons in the DataSF file, simplified for the web. Charts are generated from the same data as the text. Nothing on the site is hand-entered.
Updates
The Census Bureau releases a new five-year ACS each December. The site is rebuilt from the new release when it arrives, and the vintage shown in each page's header is the one in use. Zillow's indexes are refreshed monthly, a few days after Zillow publishes the prior month; the month in use is shown wherever a Zillow figure appears. Census data on this site was last built on 2026-09-17.