Methods & sources

Data sources

Definitions

Crime categories

Violent crime = murder and manslaughter, rape and sexual assault, robbery, aggravated assault. Property crime = burglary of a building, motor-vehicle theft, theft (including theft from vehicles), arson. These follow the FBI’s standard “Part I” definitions so Austin can be compared with other places. Everything else is grouped into plain-English buckets (other crimes against persons; drugs, DWI and alcohol; fraud; vandalism and public order; weapons; warrants, traffic and other). APD assigns each report an internal offense code; we map the codes to groups with the table below, which you can search.

Vehicle burglary (break-ins to cars) is classed by APD, and by the FBI, as theft rather than burglary. We show it separately because it is common and behaves differently from other theft. Shoplifting and other theft are combined in trends because shoplifting was recorded inconsistently before about 2011.

Rates and populations

“Per 100,000 residents” divides by Austin’s population in that year. We use Census counts for 2000 (656,562), 2010 (790,390) and 2020 (961,855) with smooth growth in between, and the FBI-reported population for 2022 onward (already rebased to the 2020 Census). FBI-reported populations for 2011–2021 are pre-rebase estimates that overshoot by about 2–4% and drop abruptly in 2022, so we do not use them; the sensitivity table shows the effect of using them instead. Neighborhood rates use 2020 Census population for every year.

Neighborhoods

Each incident carries a census block group. We assign each block group to the neighborhood reporting area it overlaps most (at least half its area), so boundaries are approximate; about 6% of incidents (block groups outside any reporting area, or with no location) could not be placed. Council districts are assigned the same way using the current district map.

Clearance

“Cleared” means APD marked the incident cleared by arrest or by exceptional means. It measures police outcomes, not charges or convictions.

How the district-attorney analysis works

1. Terms. Rosemary Lehmberg (2009–2016), Margaret Moore (2017–2020), José Garza (2021–present). Analyses compare calendar years: 2017–2020 (“Moore years”) with 2021–2025 (“Garza years”).
2. Austin versus the nation. For each crime type we compare the change in Austin’s rate between those periods with the change in the U.S. (and Texas) rate: the percent change in the ratio Austin ÷ comparison. Austin is measured with APD’s own incident series because it uses one method for all 23 years; the FBI-reported Austin series has a 2019 methodology break (the move to NIBRS, which counts more offenses per incident). The U.S. and Texas come from the FBI, which has its own 2021 reporting dip (large agencies missing during the NIBRS transition).
3. Sensitivity. We repeat the estimate with a longer baseline, without the pandemic years, adjusting for Austin’s pre-2021 trend, with different population figures, and with the FBI-reported Austin series. We report the range, not one number.
4. Synthetic control. We build a weighted blend of 64 large-city police departments (FBI data) whose log annual rates, each demeaned by its own 2010–2020 average, best match Austin’s over 2010–2020, and compare 2021–2025. The same procedure is run on every other city as a placebo; we report the share of cities with a comparably good pre-2021 fit whose relative rise was larger than Austin’s. We run it twice, with APD’s series and with Austin’s FBI-reported series, because each has a different reporting break.
5. Same-city comparison. About 3% of Austin’s incidents fall in Williamson County (different prosecutor) and a handful in Hays County. We compare change in incidents in the Travis part with the Williamson part, with a 95% interval computed by resampling the yearly counts as Poisson noise (4,000 draws). The interval reflects chance variation only; it does not capture differences in the two areas’ growth or character.
6. Courts and jail. Texas OCA felony case outcomes and sentences, and Texas Commission on Jail Standards jail counts, for Travis and the other largest Texas counties, shown side by side rather than combined into an estimate.

District attorney terms used

We do not label any prosecutor “progressive” or “traditional” ourselves; the charts show each county’s numbers and leave that judgment to the reader.

Limitations

• Reported crime is not all crime. It reflects what people report and what police record. Changes in reporting or enforcement look like changes in crime.
• Highest offense per incident. A robbery with an assault counts once, as a robbery. Counts are incidents, not victims, so they run below victim counts and below FBI offense counts.
• No causal claim about the DA. Anything that differs between Austin and the nation after 2020 could be due to policing levels, city or state policy, population change, or chance. The same-city comparison narrows this but is imperfect and its interval is wide.
• Prosecutor effects are mostly invisible in these data. The public court totals are aggregates. They show how many felony cases end in dismissal but not why, they do not include cases declined before filing, and they cannot follow a particular defendant. Travis changed how it coded rejections versus dismissals in August 2020 (per KXAN), which affects journalists’ comparisons across that date.
• Recent data are incomplete. 2026 is partial, the most recent few weeks of APD data are still being filed, and the FBI’s most recent months are incomplete.
• Other cities’ FBI data have their own breaks. Many large departments switched to NIBRS reporting around 2021, and some under-reported for months. That adds noise to the national and multi-city comparisons (the synthetic-control placebos show it as large 2021 swings).
• Neighborhood rates use 2020 population for all years, and neighborhoods with many visitors but few residents (downtown, the airport) have inflated per-resident rates.

What would improve this

• Person-linked case data: prior arrests, release decisions and their decision-makers, and outcomes for each felony arrest. Texas courts hold this; it is not published in bulk.
• Other jurisdictions’ own open data to build a better counterfactual than national averages.

Reproduce it

python3 -m venv .venv && .venv/bin/pip install pandas numpy geopandas shapely pyarrow scipy statsmodels requests
cd pipeline
python build_incidents.py     # raw APD CSV + Census geography -> incidents.parquet
python build_austin_data.py   # incidents -> site JSON (monthly, neighborhoods, council, clearance)
python resolve_cities.py      # find FBI agency IDs for 64 large cities (needs FBI_API_KEY in .env)
python fetch_fbi.py           # FBI CDE monthly series for the U.S., Texas and those cities
python build_national.py && python build_pop.py
python analyze_da.py          # eras, DiD, sensitivity, county split
python synth_control.py       # synthetic control + placebo tests
python build_da_process.py    # Texas OCA annual court data (legacy)
python build_dispositions.py  # OCA CARD felony outcomes + TCJS jail counts
python build_claims.py        # @AustinJustice sample + claim checks
cd ../site && python3 -m http.server

Offense crosswalk

How each APD offense type is grouped (types with at least 20 reports). Counts are all-time reports.