ATM in San Francisco, CA

How to read the atm in San Francisco numbers

Start with one distinction: what we publish for San Francisco, CA are medians, not averages.

A median is the midpoint: half of the local market sits above it and half below. Averages get dragged around by a handful of extreme values — one very expensive property, one unusually large loan, one graduate earning far more than their classmates — so a median is the safer number to plan against. Locations on record of 198 and institutions represented of 48.

Treat these as a baseline for picking a branch or an institution to bank with, not a quote. Your own position depends on details no dataset holds: credit history, the exact street, the specific product on offer this week.

Turning the San Francisco, CA data into a decision

Context does most of the work here. A number is only high or low next to something else, so we publish San Francisco, CA and how it sits inside California against a wider benchmark on the same measure. State of California.

Small gaps tell you nothing — sampling error alone can produce them. Large gaps almost always have a physical cause: the age of the housing stock, commuting distance, the balance between renters and owners, or a single large employer setting local pay.

Read the measures as a set. Income on its own says little; income next to housing cost and next to what similar households pay elsewhere says a great deal.

How San Francisco, CA compares

Comparison is the fastest way to tell a genuinely unusual local market from an ordinary one. We benchmark San Francisco, CA against California and against the national picture on the same definitions, in the same period, so the difference you see is a real difference rather than an artefact of two datasets disagreeing.

Definitions matter as much as the numbers. Two sources can both publish a "median rent" and mean different things — one including utilities, one not; one covering every unit, one only those recently let. Mixing them produces gaps that look meaningful and are not.

When the underlying sample is too small for a reliable estimate, we publish nothing for that measure rather than modelling a plausible-looking figure. Fewer numbers you can trust beats more you cannot.

Five checks worth running first

None of this needs a spreadsheet. It needs the discipline of checking the same handful of things every time, whether you are picking a branch or an institution to bank with for the first time or the fifth.

• Verify the branch is still open and staffed before you travel; registers lag closures by a quarter or more.
• Check whether deposit insurance covers your full balance, and how joint accounts change that limit.
• Check the fee schedule and the fine print, not the headline rate — that is where the real cost usually hides.
• Ask what the number looks like in a bad year, not an average one, and make sure the plan still holds.
• Write down the monthly number you can genuinely cover, not the maximum a lender or admissions office says you qualify for.
• Compare at least three offers. Pricing on the same product varies more between providers than most people expect, and the gap is yours to keep.

Questions people ask about San Francisco, CA

Method and limitations

The page is assembled from federal branch and office registers, specifically the FDIC office file and the NCUA credit union branch file. Nothing here is modelled from a proprietary black box: every figure traces to a published record, and where we derive a ratio we show both inputs so the arithmetic can be checked.

The limits are the usual three: sampling error grows as geography shrinks, published data always trails reality by months or years, and a midpoint says nothing about the extremes at either end of the local market.

Where a source revises a figure, the revised value replaces the old one rather than sitting alongside it, so the page always reflects the current official position. If you spot a number that looks wrong, it is worth telling us — corrections get made against the source record, not by guesswork.