San Francisco / Data methodology

Know what each number measures.

Version: September 18, 2026. Downloads are reviewed snapshots. There is no scheduled refresh or automatic publication. Source data may be revised after this snapshot.

Reading across datasets

San Francisco city and county figures describe the local edition. The Muni stop inventory follows its network and may include stops outside city limits.
ZORI represents market rent; ACS represents a five-year survey of occupied housing. Do not treat them as the same rent measure.
The four Census additions use San Francisco city geography and 2020 to 2024 estimates. The existing renter cost burden extract uses San Francisco County geography. Check each file's universe and geographic identifier before comparisons.
Notice counts are administrative filings. No population denominator, completed eviction count or neighborhood risk score is inferred.
Missing values remain blank in CSV and null in JSON. Zero is used only when a measure is truly zero, including no observed notice IDs for a fully covered month.
Each JSON download contains metadata and records. CSV column names use explicit units where possible. No private renter survey or subscription records are included.

Market rent trends

City ZORI, all homes plus multifamily, smoothed monthly series. Retain missing months. Round displayed dollars to cents. Year-over-year change compares the same calendar month in the previous year.

Typical observed market rent index; not a median lease, bedroom-specific quote, neighborhood rent or rent paid by all incumbent tenants. Historical values can be revised. Missing values are not zero.

Fields: month, zori_usd, yoy_percent, geography, zillow_region_id, attribution.

Coverage: 2015-01 to 2026-08.

Source and original documentation ↗

Zillow terms apply: https://www.zillow.com/z/corp/terms/ . No Homzora relicensing.

Renter cost burden

Select GEO_ID 0500000US06075 from 2024 ACS five-year table B25070; preserve estimates and 90% margins of error. Gross rent includes contract rent plus applicable utilities and fuels.

Five-year period estimates, not a 2024-only survey or a current market snapshot. Universe: renter-occupied housing units paying cash rent. Not-computed income shares are separate. Do not sum margins of error. Total overlaps every component row. No derived burden percentage or statistical significance claim is published here.

Fields: variable, gross_rent_share_of_household_income, estimate_units, moe_90_units, geoid, period.

Coverage: 2020–2024 ACS five-year estimates.

Source and original documentation ↗

U.S. Census Bureau public data.

Eviction notice filings

Count distinct eviction_id grouped by month of file_date. Query starts 2015-01-01 and ends before 2026-09-01. No names, addresses, notice identifiers or point locations are republished.

Notices filed with the Rent Board are not court judgments, completed evictions, unique households, or an eviction rate. Late filings and revisions can change prior months. No neighborhood risk ranking is inferred.

Fields: month, distinct_notice_ids, geography, count_basis, retrieved_utc.

Coverage: 2015-01 to 2026-08; excludes the partial current month.

Source and original documentation ↗

DataSF source license: Open Data Commons Public Domain Dedication and License

Transit stops and scheduled service

Apply GTFS calendar plus calendar_dates additions/removals for 2026-08-28. Select active trips, then departure-time stop calls from 06:00 inclusive to 22:00 exclusive in local service-day time. Count trips and distinct stop IDs per route. Separate stop inventory includes coordinates and active-route count for this window.

The feed downloaded September 18 is expired; these are historical schedules, not current service, actual arrivals, reliability, commute times or walking accessibility. Stop IDs can represent opposite sides/platforms at one location. Trip counts are not frequency estimates. No causal rent/transit relationship is claimed.

Fields: route_id, route, route_name, service_date, window_local, scheduled_trips_with_stop_calls, distinct_stop_ids, scheduled_stop_calls.

Coverage: HISTORICAL: Friday 2026-08-28, 06:00–22:00; downloaded feed calendar spans July 23–August 28, 2026.

Source and original documentation ↗

SFMTA Transit Data License Agreement; see bundled license and source page. No Homzora relicensing.

Median gross rent by bedroom count

Select GEO_ID 1600000US0667000 from the 2024 ACS five-year summary file table B25031. Pair every estimate with its published 90% margin of error; retain the complete category hierarchy. No percentages or combined margins of error are calculated. For the rent table, Census code -333333333 means an open-ended interval median with no computable margin of error; the normalized estimate and margin are blank/null, while raw tokens and a status field are preserved.

Universe: Renter-occupied housing units paying cash rent. Units: USD per month.

Period estimates covering 2020 to 2024, not a current market survey or 2024 alone. Totals and subtotals overlap their components. Do not add all rows together or add margins of error. Differences need a statistical test before being interpreted as significant. These are medians for occupied homes, not asking rents or a quote for an available apartment. Dollar estimates are expressed in 2024 inflation-adjusted dollars.

Fields: variable (Census estimate ID), label (full category hierarchy), estimate, margin_of_error_90, estimate_raw and moe_raw (unchanged source tokens), status, unit, geoid, geography, period. Blank CSV or null JSON values mean a nonnumeric estimate, not zero. The rent table open-ended interval row is not a precise rent amount. Margins of error use the Census 90% confidence level.

Original summary file · Variable definitions · Public-use statement

Housing types by renter or owner occupancy

Select GEO_ID 1600000US0667000 from the 2024 ACS five-year summary file table B25032. Pair every estimate with its published 90% margin of error; retain the complete category hierarchy. No percentages or combined margins of error are calculated. For the rent table, Census code -333333333 means an open-ended interval median with no computable margin of error; the normalized estimate and margin are blank/null, while raw tokens and a status field are preserved.

Universe: Occupied housing units. Units: housing units.

Period estimates covering 2020 to 2024, not a current market survey or 2024 alone. Totals and subtotals overlap their components. Do not add all rows together or add margins of error. Differences need a statistical test before being interpreted as significant. Building size does not establish property condition, rent regulation coverage, amenities or whether a home is available.

Original summary file · Variable definitions · Public-use statement

Vehicle access by renter or owner occupancy

Select GEO_ID 1600000US0667000 from the 2024 ACS five-year summary file table B25044. Pair every estimate with its published 90% margin of error; retain the complete category hierarchy. No percentages or combined margins of error are calculated. For the rent table, Census code -333333333 means an open-ended interval median with no computable margin of error; the normalized estimate and margin are blank/null, while raw tokens and a status field are preserved.

Universe: Occupied housing units. Units: housing units.

Period estimates covering 2020 to 2024, not a current market survey or 2024 alone. Totals and subtotals overlap their components. Do not add all rows together or add margins of error. Differences need a statistical test before being interpreted as significant. Vehicle availability does not identify parking spaces, commuting choices, travel times or the quality of transit service.

Original summary file · Variable definitions · Public-use statement

Household internet subscriptions

Select GEO_ID 1600000US0667000 from the 2024 ACS five-year summary file table B28002. Pair every estimate with its published 90% margin of error; retain the complete category hierarchy. No percentages or combined margins of error are calculated. For the rent table, Census code -333333333 means an open-ended interval median with no computable margin of error; the normalized estimate and margin are blank/null, while raw tokens and a status field are preserved.

Universe: Households. Units: households.

Period estimates covering 2020 to 2024, not a current market survey or 2024 alone. Totals and subtotals overlap their components. Do not add all rows together or add margins of error. Differences need a statistical test before being interpreted as significant. Subscription counts do not identify available providers, prices, speeds, reliability or service at an individual address. Some category branches overlap; read the full label.

Original summary file · Variable definitions · Public-use statement

Homzora renter experience

Use the existing voluntary renter survey. No raw responses are exposed by this data expansion. Reference the existing survey field dictionary and release criteria.

Unpublished. Minimum 100 eligible answers per measure, closed quarter and editorial review; subgroup and complement at least 15. A voluntary sample is not representative. No automated results publication.

Fields: quarter_utc, measure, eligible_n, estimate, method_version, editorial_review_date.

Coverage: No results published.

Source and original documentation ↗

Homzora schema: CC BY 4.0; this does not license private responses.

Rental fees and concessions

One observation per public listing, observation date and item type. Record quoted base rent, lease length, mandatory fee or concession, payment cadence and source URL. Unknown amounts remain blank; refundable deposits stay distinct from fees. Deduplicate listing/date/item before analysis. Editorial verification is required before publication.

Empty staff collection template, not a findings dataset or an operating public intake form. No claims about prevalence, legality or market averages. Do not include tenant names, emails, private documents or exact unit addresses. Concessions are conditional advertised offers, not guaranteed savings.

Fields: observation_id, observed_date, listing_reference, public_source_url, area_label, bedrooms, base_rent_usd_month, lease_months, item_type, item_label, amount_usd, cadence, mandatory_status, refundable_status, conditions, verified_date.

Coverage: No verified observations published.

Source and original documentation ↗

Homzora schema: CC BY 4.0. Linked listing materials retain their own terms.

Fee observation protocol

The empty template is for documented editorial observations. Use item_type = fee, deposit or concession; cadence = one_time, monthly, annually or other; mandatory_status = yes, no or unknown; refundable_status = yes, no, unknown or not_applicable. Keep monetary amounts in USD and dates as YYYY-MM-DD. Record each separate quoted item on its own row. Blank means unknown, not zero.

Retain lease length and written concession conditions. Do not automatically subtract one-time concessions from monthly rent or combine refundable deposits with costs. No prevalence estimate should be published without a defined sampling frame, collection window, deduplication and review. The template does not add a new public submission form.

Updates and corrections

For a future release, retrieve the same series and definitions, check source vintages, rerun transformations, compare totals and review changes before replacing the versioned snapshot. Keep prior versions and record corrections. Current source URLs may return a later vintage than the files in this release.

Report a suspected issue through Contact Homzora, including dataset name, period and the disputed field. Research signup remains separate from survey answers.