San Antonio Crime Map

It shows reports, not convictions, and not all crime

Every number here is an offence reported to SAPD. Some reports are later unfounded. Nobody on this map has been convicted of anything, and no individual is named, because the City’s data contains no names, no addresses, and no coordinates. That is a feature, not a gap.

It also is not all crime. The published dataset covers NIBRS Group A offences only, which excludes DUI, disorderly conduct, and trespass, among others. Crimes that are never reported appear nowhere in it, and for some categories that is most of them.

ZIP code is as precise as the data goes

You may have seen crime maps that drop a pin on a block. They are working from a private police records feed. The City’s public dataset has no street address and no latitude or longitude: the finest geography it publishes is the ZIP code. So this map stops at the ZIP code, rather than inventing a precision it does not have.

A ZIP code is also not a neighbourhood. It is a mail-delivery boundary that can span a wealthy subdivision, an industrial park, and a highway interchange at once. Treat a ZIP-level figure as a coarse signal, never as a verdict on a street.

A percentage change is the number every crime map leads with, and it is the number most likely to mislead you, because it is really a ranking of small denominators. A ZIP code with a modest baseline can swing by half on nothing more than an ordinary month.

So the What actually changed panel above does not rank by percentage. It asks a harder and more useful question: given what this ZIP code normally does at this time of year, is this month actually surprising? Each ZIP is compared against its own baseline, adjusted for the season and the length of the month, and we report how many standard deviations away it landed.

The percentage change is shown right next to it, deliberately, so that you can watch the two disagree. They frequently do. That gap is the entire argument for building this thing.

Two details that most published analysis skips, and both change the answer:

Offence counts are overdispersed. Crime clusters, so real month-to-month variance is well above what a textbook Poisson model assumes — in some ZIP codes several times above it. Test against plain Poisson and you will discover “statistically significant” spikes every single month that are nothing at all. We estimate each ZIP code’s own dispersion from its own history and widen the interval to match.

We are testing dozens of ZIP codes at once. At the reflexive 95% confidence threshold, roughly three of San Antonio’s ZIP codes would be flagged as unusual by chance alone, every month, forever. A map that reported them would manufacture a scare story on a monthly schedule. We only call a ZIP code unusual beyond three standard deviations.

The honest consequence: on a typical month, this panel tells you that nothing unusual happened. That is not a broken feature. That is the answer, and almost nobody publishes it.

The pipeline

Offence records come from the City of San Antonio’s open data portal (SAPD Offenses, CC BY licence) through its CKAN SQL API. They are aggregated to one cell per ZIP code, per month, per offence type, and joined to the City’s own ZIP boundary layer, which is projected to flat SVG geometry at build time. No map library, no tiles, and no third-party tracker loads on this page.

The build asserts its own totals. It re-counts the source table independently and refuses to publish if the aggregate does not reconcile to the row count, because the City’s datastore truncates large responses silently and a partial extract would look perfectly normal.

What we throw away, and why we tell you

Of 516,255 published records, 5,196 (1.01%) carry a blank or unmappable ZIP code. They are excluded, because there is nowhere honest to put them. The alternative, quietly assigning them somewhere, is how a map starts lying.

We count offences, not incidents. Those 516,255 offences come from 472,301 distinct police reports, because a single report can carry more than one offence. A dashboard that counts reports will show you a smaller number than this one, and neither of us is wrong. Ask which is being counted before you compare two crime figures.

The City republishes monthly, so the most recent few weeks are never present. If you need what happened last night, this is the wrong tool, and we would rather say so than imply a freshness we do not have.

Every figure in this section is read live from the same data file the map is drawn from, so it cannot drift out of step with the map above it. Data built 2026-07-12. Population: U.S. Census ACS 5-year (2023), table B01003.