About the data

What’s in here

Missouri requires police agencies to report data on every traffic stop to the state Attorney General: who got stopped, why, what happened, and whether a search occurred. We extract that data from yearly PDF reports published by the AG’s office.

This dataset includes counts and rates, broken down by the race of the driver. The metrics tracked include:

• Stops: total stops, resident stops, non‑resident stops
• Outcomes: arrests, citations, searches, and contraband found
• Rates: arrest rate, citation rate, search rate, contraband hit rate, stop rate, resident stop rate
• Arrest reason (e.g., drug violation)
• Citation/warning reason (e.g., moving violation)
• Driver age (17 and under, 18–29, 30–39, 40–64, 65+)
• Driver gender (male, female)
• Location of stop (city street, county road, interstate, etc.)
• Officer assignment (dedicated traffic, general parol, special assignment)
• Reason for stop (investigative, moving, equipment, etc.)
• Stop outcome (arrest, citation, warning, no action, etc.)
• Type of contraband found (drugs, weapons, etc.)
• Type of search / probable cause (consent, incident to arrest, etc.)
• Search duration (0–15 min, 16–30 min, 31+ min)
• What was searched (vehicle property, driver, driver property)
• Population estimates from the Decennial Census and the American Community Survey (ACS)

Where it came from

The Missouri Attorney General is required by statute to compile and publish the Vehicle Stops Report.

Each report typically includes an executive summary, statewide aggregates, agency‑specific reports, and a separate document of agency comments (if submitted). We extract structured data from the agency‑specific reports and from the agency responses, which are later joined together.

Source reports and comments are available from the AG’s office:

• Vehicle Stops Report landing page
• 2020+ format — agency‑specific report PDFs:
  • 2024: https://ago.mo.gov/wp-content/uploads/2024-VSR-Agency-Specific-Reports.pdf
  • 2023: https://ago.mo.gov/wp-content/uploads/VSRreport2023.pdf
  • 2022: https://ago.mo.gov/wp-content/uploads/vsrreport2022.pdf
  • 2021: https://ago.mo.gov/wp-content/uploads/2021-VSR-Agency-Specific-Report.pdf
  • 2020: https://ago.mo.gov/wp-content/uploads/2020-VSR-Agency-Specific-Report.pdf
• 2020+ format — agency response PDFs:
  • 2024: https://ago.mo.gov/wp-content/uploads/2024-Agency-Responses-1.pdf
  • 2023: https://ago.mo.gov/wp-content/uploads/VSRagencynotes2023.pdf
  • 2022: https://ago.mo.gov/wp-content/uploads/2022-agency-comments-ago.pdf
  • 2021: https://ago.mo.gov/wp-content/uploads/2021-VSR-Agency-Comments.pdf
  • 2020: https://ago.mo.gov/wp-content/uploads/2020-VSR-Agency-Comments.pdf
• Pre‑2020 format (2000–2019): a single combined statewide PDF per year listing every agency in sequence, with substantially fewer metrics per agency. No separate agency‑comments PDFs exist for these years.
  • 2019: https://ago.mo.gov/wp-content/uploads/2019agencyreports.pdf
  • 2018: https://ago.mo.gov/wp-content/uploads/2018agencyreports.pdf
  • 2017: https://ago.mo.gov/wp-content/uploads/2017agencyreports.pdf
  • 2016: https://ago.mo.gov/wp-content/uploads/2016agencyreports.pdf
  • 2015: https://ago.mo.gov/wp-content/uploads/2015agencyreports.pdf
  • 2014: https://ago.mo.gov/wp-content/uploads/2014agencyreports.pdf
  • 2013: https://ago.mo.gov/wp-content/uploads/2013-agency-reports.pdf
  • 2012: https://ago.mo.gov/wp-content/uploads/2012agencyreport.pdf
  • 2011: https://ago.mo.gov/wp-content/uploads/2011agencyreport.pdf
  • 2010: https://ago.mo.gov/wp-content/uploads/2010agencyreports.pdf
  • 2009: https://ago.mo.gov/wp-content/uploads/2009agencyreports.pdf
  • 2008: https://ago.mo.gov/wp-content/uploads/2008agencyreports.pdf
  • 2007: https://ago.mo.gov/wp-content/uploads/2007agencyreports.pdf
  • 2006: https://ago.mo.gov/wp-content/uploads/2006agencyreports.pdf
  • 2005: https://ago.mo.gov/wp-content/uploads/2005agencyreports.pdf
  • 2004: https://ago.mo.gov/wp-content/uploads/2004agencyreports.pdf
  • 2003: https://ago.mo.gov/wp-content/uploads/2003agencyreports.pdf
  • 2002: https://ago.mo.gov/wp-content/uploads/2002agencyreports.pdf
  • 2001: https://ago.mo.gov/wp-content/uploads/2001agencyreports.pdf
  • 2000: https://ago.mo.gov/wp-content/uploads/2000agencyreports.pdf

Currently, we extract data for reports 2000–2024 (published 2001–2025). The 2000 report is included in the underlying parquet/CSV downloads but suppressed in the web UX because coverage is too sparse to chart meaningfully. Pre‑2020 reports use a different format and report a smaller set of metrics; see Key normalization below for how the two eras are reconciled. Coverage in the earliest years (2000–2003) is partial: roughly half of agencies are missing in those years because of blank race columns in the source PDFs.

Agency metadata (names, addresses, contact info) comes from a 2025 copy of the Missouri law enforcement agencies database provided by Jesse Bogan at The Marshall Project. The latest version of this data is available via data.mo.gov and will be integrated after the 2025 VSR is released (spring 2026).

Because agency names vary between the agencies database and the VSR, we built a crosswalk to join their information.

The address from the agency data is run through Geocod.io to attach geographic identifiers for each jurisdiction and to geocode the agency address. These identifiers are joined with Census cartographic boundary files to display jurisdiction maps and spatial relationships.

The processing pipeline is an open source Python/Dagster project originally developed at The Marshall Project.

How the data is structured

Row model (the core table)

The central dataset is a row‑based table where each row represents one metric for one agency and one year. Key fields:

• agency — the canonical agency name (see Name normalization below)
• year — report year
• table, section, metric — human‑readable labels
• table_id, section_id, metric_id — slugified identifiers
• row_key — era‑specific table_id--section_id--metric_id (the original, format‑specific identifier)
• canonical_key — era‑independent metric identifier shared between the pre‑2020 and 2020+ formats (see Key normalization below)
• row_id — year-agency-row_key (globally unique)
• Race columns: Total, White, Black, Hispanic, Native American, Asian, Other. Pre‑2020 reports use the header Am. Indian; the combine step collapses it into Native American so the column is populated across all years.

All values are numeric or null (. in the PDF becomes null).

Name normalization

Agency names vary between the AG’s reports and the agency reference database, and they vary year over year within the AG’s own reports (capitalization, abbreviations, “Department” vs “Dept”, whitespace, parenthetical notes, OCR artifacts). The pipeline ships a CSV crosswalk (data/src/agency_crosswalk.csv) that maps every observed name variant to a single canonical name.

• The combined dataset replaces raw names with their canonical form before downstream processing.
• Agency agency_slug is derived from the canonical name and is stable across years.
• The agency_index keeps every observed variant in a names array so callers can still match on a raw name from a source PDF.

Key normalization

The pre‑2020 (2000–2019) and 2020+ reports use different table layouts and different row_key strings for equivalent concepts. To support cross‑era analysis, every row carries a canonical_key populated from a crosswalk (data/src/canonical_crosswalk.csv).

Examples:

In the released v2 dist outputs, row_key is replaced by canonical_key so frontend callers see a single era‑independent identifier.

Rate rows and normalization

Rates in the reports (e.g., “search rate”) are captured as provided. The pipeline also computes its own per‑agency rates from raw counts at the normalization layer:

• search-rate, arrest-rate (all years), and contraband-hit-rate (2020+) are recomputed per agency per race from the raw numerator/denominator counts. The computed rows use canonical keys (e.g. search-rate); pre‑computed rates from the PDF are kept for validation.
• Statewide rates are recomputed from totals for consistency; we do not sum rate rows.
• YYYY ACS pop fields are normalized to acs-pop so population variables are comparable across years.
• In early years (2020–2021), population rows that look like YYYY-pop are also mapped to acs-pop.

Versioning

Data releases follow semantic versioning (vMAJOR.MINOR).

• Major — breaking schema change: columns renamed/removed, row_key/canonical_key structure changed, canonical metric definitions revised.
• Minor — backward‑compatible addition: new years added, new canonical metrics added, new derived columns added.

The S3 bucket layout is:

s3://{bucket}/
  releases/
    v1/    ← frozen v1 (2020–2024, era‑specific row_keys)
    v2/    ← current release (2000–2024, canonical_key, name normalization)
  manifest.json   ← top‑level pointer to the current frontend‑facing release

Each release directory contains a manifest.json describing the version, release date, years covered, schema version, the list of canonical_metrics, and a changelog. The current release is v2.0 — it adds 2000–2019 pre‑2020 format data, normalizes agency names through the crosswalk, replaces row_key with canonical_key in dist outputs, and partitions per‑agency JSON by year.

Agency comments

Agency comments are parsed from the separate response PDFs and attached downstream. Each comment entry has:

• agency
• year
• comment (string, with paragraph breaks preserved as \n\n)
• has_comment
• source_url

Line breaks inside paragraphs are collapsed to a single space. Paragraph breaks are preserved.

Downloads and file formats

All public downloads live under:

https://data.vsr.recoveredfactory.net/

Combined JSON (all datasets)

This file contains all datasets in one JSON object with keys:

• vsr_statistics
• agency_index
• agency_comments

Download:

• https://data.vsr.recoveredfactory.net/releases/v2/downloads/missouri_vsr_2000_2024_downloads.json

Per‑dataset CSV + Parquet

For analysis in pandas/R/SQL, each dataset is also provided as CSV and Parquet:

VSR statistics (with rank/percentile/percentage rows)

• https://data.vsr.recoveredfactory.net/releases/v2/downloads/missouri_vsr_2000_2024_vsr_statistics.csv
• https://data.vsr.recoveredfactory.net/releases/v2/downloads/missouri_vsr_2000_2024_vsr_statistics.parquet

Agency index (names + metadata)

• https://data.vsr.recoveredfactory.net/releases/v2/downloads/missouri_vsr_2000_2024_agency_index.csv
• https://data.vsr.recoveredfactory.net/releases/v2/downloads/missouri_vsr_2000_2024_agency_index.parquet

Agency comments

• https://data.vsr.recoveredfactory.net/releases/v2/downloads/missouri_vsr_2000_2024_agency_comments.csv
• https://data.vsr.recoveredfactory.net/releases/v2/downloads/missouri_vsr_2000_2024_agency_comments.parquet

Download manifest

The manifest includes file sizes for dynamic download UIs:

• https://data.vsr.recoveredfactory.net/releases/v2/downloads/missouri_vsr_2000_2024_downloads_manifest.json

Statewide sums (subset)

A slim statewide summary used for homepage charts:

• https://data.vsr.recoveredfactory.net/releases/v2/dist/statewide_year_sums_subset.json

Included canonical keys:

• stops
• probable-cause--consent
• stop-outcome--arrests
• stop-outcome--citation
• stop-outcome--warning
• stop-outcome--no-action

Per‑agency downloads (convenience)

These files are convenient for single‑agency use, but the primary way to get the data is still the combined downloads above.

Agency year files (index)

In v2, per‑agency data is split into one file per year. To get a full agency history, request each year individually using the pattern releases/v2/dist/agency_year/{slug}/{year}.json. Years available: 2000–2024 (with sparse coverage in the earliest years).

Format quirks and implementation notes

Agency index (agency_index)

The agency index is an aggregation of agency metadata and the VSR names. Notable fields:

• agency_slug — slugified canonical name (apostrophes removed, punctuation collapsed)
• canonical_name — the single canonical agency name (see Name normalization)
• names — array of known names for the agency (canonical, crosswalked, and VSR variants)
• city, zip, phone, county — from the agency reference database
• census_geoid — US Census GEOID for the agency’s jurisdiction
• stops and all_stops_total — the most recent total stops, for weighting/search
• years_with_data — array of years in which the agency reported at least one non‑null value
• latest_year_with_data — the most recent year in years_with_data (or null)

CSV quirk: names is serialized as a JSON array string in CSV (e.g. "[\"Agency A\", \"Agency A PD\"]"). In Parquet and JSON, it is a native list.

Agency comments (agency_comments)

• comment preserves paragraph breaks as \n\n.
• Line breaks within a paragraph are collapsed to spaces.
• Text is minimally cleaned; odd characters present in the source PDFs are preserved.

Statewide sums (statewide_year_sums)

• Rows ending in -rate are excluded from summation; statewide rates are recomputed from totals.
• no-mshp--* excludes the Missouri State Highway Patrol.
• avg-no-mshp--* is an average across agencies (after excluding MSHP), not a sum.

VSR statistics (reports_with_rank_percentile)

Derived rows are added per metric:

• -rank (dense rank, 1 = highest)
• -percentile (0–1 scale)
• -percentage for non‑rate metrics (race ÷ total)

Metric definitions (rates)

Rates are defined according to the VSR documentation:

• Stop rate: (stops / previous year ACS population) × 100
• Resident stop rate: (resident stops / previous year ACS population) × 100
• Search rate: (searches / stops) × 100
• Contraband hit rate: (searches with contraband found / total searches) × 100
• Arrest rate: (arrests / stops) × 100
• Citation rate: (citations / stops) × 100

Metrics tracked (canonical key list)

In v2, the dist outputs index metrics by canonical_key rather than the era‑specific row_key. Some metrics appear only in certain years (e.g., the disparity index was discontinued after 2022; per‑race rates and contraband totals are 2020+ only). Misspellings preserved from the original reports include parol and equpiment.

For convenience, each canonical key has a per‑metric JSON file at: https://data.vsr.recoveredfactory.net/releases/v2/dist/metric_year/{canonical_key}.json These are handy for single‑metric use, but the combined downloads are the primary interface.

All canonical keys currently present: