Latino voters shifted towards Prop. 50. Here’s how we analyzed it.

An analysis of 2025 election results for Prop. 50 using precinct-level data. This repository contains the code and processing pipeline used to build precinct-level datasets. For the full narrative methodology and findings framing, see Show Your Work.

Published story: Latinos in California are mad at Trump. Their votes for Democrats’ gerrymandering show it

Data sources

• Precinct results, geographic, and voter demographics files from the Statewide Database.
• Census Voting Age Population (CVAP) disaggregated by the Redistricting Voting Hub.

Filepaths

• ./inputs/census/CVAP_2020-2024_ACS_csv_files.zip
  • Census CVAP special tabulation zip consumed by 00_census.py (CVAP_ZIPPED_DATA_FP)
  • Data used for validation and experimentation
• ./inputs/rdh/ca_cvap_2024_2020_b_csv/ca_cvap_2024_2020_b.csv
  • Redistricting Data Hub 2024 CVAP block-level extract consumed by 00_census.py (RDH_CVAP_DATA_FP)
  • Data used for interpolating precinct demographics
• ./inputs/statewide_db/ General Election 2024 files
  • Expected core files used by 02b_results_2024.py:
    • Election results: state_g24_sov_data_by_g24_srprec.zip
    • Voter demographics: state_g24_voters_by_g24_srprec.zip
    • Precinct geographic boundaries: srprec_state_g24_v01_shp.zip
• ./inputs/statewide_db/S25/ Special Election 2025 files
  • Expected county-level 2025 election results, voter demographics, and precinct boundaries inputs used by the 2025 results workflow

Network-dependent data

These files were excluded from version control due to file size limits. If missing, notebooks will try and fetch them over the network:

• 00_census.py
  • Local expected file: ./inputs/census/tl_2020_06_tabblock20.zip
  • Fallback URL: https://www2.census.gov/geo/tiger/TIGER2020/TABBLOCK20/tl_2020_06_tabblock20.zip
• 02b_results_2024.py
  • Local expected file: ./inputs/statewide_db/srprec_state_g24_v01_shp.zip
  • Fallback URL: https://statewidedatabase.org/pub/data/G24/state/srprec_state_g24_v01_shp.zip

Quick start

This project uses Marimo (an interactive Python notebook) and uv (a fast Python package manager). This project uses just to quickly run project scripts.

Setup (first time only)

1. Install uv and just (if not already installed):

 brew install uv
 brew install just

2. Install dependencies:

 uv sync

Run the pipeline

After required inputs are in place under ./inputs/ (see above), run stages in order or use the combined recipe:

Step	Command	Purpose
CVAP / county bounds	just generate-cvap-file	00_census.py
Precinct geometries	just generate-precincts-file	01_geography.py
Standardized results	just generate-results-file	02a_results_2025.py, 02b_results_2024.py
Interpolation	just interpolate-cvap	03_interpolation.py
Analysis exports	just generate-analysis-exports	04_analysis.py
Full sequence	just generate-all-data	All of the above (also updates vis/static/county_precincts.json)

Individual notebooks can be opened interactively with uv run marimo edit <notebook>.py.

Limitations and practical notes

• Large inputs: Census zips and statewide database files are large; allow plenty of disk space under ./inputs/ and ./outputs/.
• Network fallbacks: If automatic downloads fail, place the files at the paths listed under Network-dependent data.
• Runtime: Full runs are heavy geospatial work; expect long runtimes on a laptop for the whole pipeline.
• Redactions: Some precincts have suppressed counts in source data; the pipeline preserves blanks where administrative redactions apply.
• Interpolation: Cross-year precinct boundary and areal-interpolation steps introduce assumptions; see our Show Your Work.

Data processing workflow

The notebooks follow a sequential pipeline:

1. 00_census.py — Census ETL: county bounds, CVAP by tract, CVAP by block

• Output: outputs/county_bounds.geojson, outputs/cvap_tracts.gpkg, outputs/cvap_blocks.gpkg
• Run with: just generate-cvap-file

2. 01_geography.py — Processes precinct geographic files from all counties

• Output: outputs/precincts.gpkg
• Run with: just generate-precincts-file

3. 02a_results_2025.py and 02b_results_2024.py — Clean and standardize precinct-level election results

• Output: 02a → outputs/precinct_results.gpkg; 02b → outputs/precinct_results_2024.gpkg
• Run with: uv run marimo edit 02a_results_2025.py or uv run marimo edit 02b_results_2024.py (interactive), or just generate-results-file to run both

4. 03_interpolation.py — Interpolates Census CVAP demographics to precincts and 2024 vote data to 2025 precincts

• Input: outputs/precinct_results.gpkg, outputs/precinct_results_2024.gpkg, outputs/cvap_tracts.gpkg, outputs/cvap_blocks.gpkg
• Output: outputs/precincts_results_cvap_tracts.gpkg, outputs/precincts_results_cvap_blocks.gpkg, outputs/precincts_2024_results_cvap_blocks.gpkg

5. 04_analysis.py — Analysis and exports from interpolated datasets

• Input: merged/interpolated GeoPackages from 03_interpolation.py
• Output: outputs/precinct_results_plus_demographics_*.gpkg, outputs/partner_export.csv, GeoJSON used for optional map tiles (see notebook)

6. 05_fact_check.py — Fact-check workbook for story and methodology claims

• Input: outputs/precincts_results_cvap_blocks.gpkg, outputs/precincts_2024_results_cvap_blocks.gpkg, inputs/statewide_db/state_g24_sr_blk_map.csv
• Output: in-notebook verification text/metrics used to validate published statements (no data export)

Notebooks

If you want to use a particular notebook (for example 01_geography.py):

uv run marimo edit 01_geography.py

This opens the notebook in your browser, runs code interactively, and auto-saves changes to the .py file.

Note: Marimo notebooks are plain Python files—you can edit them in any editor, but the browser UI helps run and visualize results.

Development workflow

1. Open the notebook: uv run marimo edit FILENAME
2. Make changes in the browser interface
3. Changes are automatically saved to FILENAME.py
4. Share your changes via git (the .py file is the notebook)

01_geography.py — Precinct geographic data cleaning

Reproject the voting precincts from each county into NAD83/California Albers and normalize the properties for each feature (precinct) so that it has the following attributes:

• county — The county containing the precinct
• precinct_id — The precinct ID from the county
• precinct_name — The human-readable name included by the county, otherwise None

00_census.py — Census ETL

Produces three GIS outputs in NAD83/California Albers (EPSG:3310): California county boundaries (outputs/county_bounds.geojson), CVAP by census tract (outputs/cvap_tracts.gpkg), and CVAP by block (outputs/cvap_blocks.gpkg). Run with just generate-cvap-file.

02a_results_2025.py and 02b_results_2024.py — Precinct-level results standardization

Standardizes statewide election results into a consistent schema for downstream geospatial analysis.

03_interpolation.py — Merge and interpolate CVAP to precinct geography

Builds precinct-level analysis layers by combining election results with CVAP inputs and running tract/block interpolation workflows. The notebook includes:

• Geometry/data validation and reprojection checks
• Merge audits and county-level mismatch diagnostics
• Export of precinct-level CVAP-enriched outputs (including outputs/precincts_2024_results_cvap_blocks.gpkg)

04_analysis.py — Demographics and precinct-level results analysis

Runs exploratory and comparative analysis on the merged precinct datasets, including statewide and county-level majority-group summaries, vote-shift calculations, and dataset-specific breakouts.

05_fact_check.py — Fact-check checks for editorial claims

Recomputes core published facts values from the merged precinct datasets. This notebook is for validation of published facts and presents dynamic claim markdown text reproducing the published. If you are looking to trace the data and analysis for our published report, you should start in this notebook.

Output files

Generated files land in outputs/ (gitignored except as noted). Typical artifacts:

File	Produced by	Description
county_bounds.geojson	00_census.py	California counties (EPSG:3310)
cvap_tracts.gpkg, cvap_blocks.gpkg	00_census.py	CVAP layers by tract / block
precincts.gpkg	01_geography.py	Combined precinct polygons (county, precinct_id, precinct_name, geometry)
precinct_results.gpkg	02a_results_2025.py	2025 Prop. 50 results joined to geometry
precinct_results_2024.gpkg	02b_results_2024.py	2024 presidential-style fields on 2024 precincts
precincts_results_cvap_*.gpkg	03_interpolation.py	Interpolated CVAP + votes
precinct_results_plus_demographics_blocks.gpkg	04_analysis.py	Analysis-layer export (blocks / Prop 50)
precinct_results_plus_demographics_blocks_2024.gpkg	04_analysis.py	Analysis-layer export (blocks / 2024 presidential comparison)
partner_export.csv	04_analysis.py	Tabular partner export (see data dictionary below)
precinct_results_plus_demographics_blocks.geojson	04_analysis.py	Optional map/interactive intermediate

Note: Most output files are gitignored (see .gitignore).

Data dictionary (primary exports)

partner_export.csv (from 04_analysis.py)

Percent columns in the CSV are stored as fractions between 0 and 1 (for example, 0.5 for 50%), suitable for spreadsheets that format as percent.

Column	Description
county	County name
precinct_id	Precinct identifier (joined key)
total_votes_25	Total votes counted for Prop. 50 in the precinct
yes_votes, no_votes	Vote counts for Yes / No on Prop. 50
yes_pct, no_pct	Yes / No shares of two-party vote (fractions 0–1)
registered_voters	Registered voters where available from source data
turnout	Turnout measure used in analysis (fraction 0–1)
total_votes_24	Total votes in 2024 presidential contest (interpolated to 2025 precinct where applicable)
dem_votes_24, rep_votes_24	Democratic / Republican presidential votes (interpolated baseline)
dem_pct_24, rep_pct_24	Democratic / Republican presidential vote shares (fractions 0–1)
vote_shift_net	Net swing metric between elections
majority_racial_group	CVAP-based majority racial/ethnic group label for the precinct
plurality_racial_group	Plurality group where coded
largest_racial_group_pct	Share for the largest single CVAP group (fraction 0–1)

Key GeoPackage layers

• **precincts.gpkg** — Attributes: county, precinct_id, precinct_name; geometry in EPSG:3310.
• **precinct_results.gpkg / precinct_results_2024.gpkg** — Standardized vote and registration fields for 2025 and 2024 respectively; see notebook export lists in 02a_results_2025.py and 02b_results_2024.py for exact column names per vintage.
• **precinct_results_plus_demographics_*.gpkg** — Enriched analysis layers from 04_analysis.py combining votes with interpolated CVAP demographics; column sets follow the analysis dataset configuration inside 04_analysis.py.

For interpretation of demographic groupings, net shift, and limitations, rely on our Show Your Work

Data use

While the contents of this repo are shared under an Apache 2.0 license, CalMatters/The Markup would appreciate any credit or attribution you're willing to give. We're also interested to learn how you used it, so feel free to send us a message or open an issue if you do. If you have any questions, feel free to contact us as well.

CalMatters is a nonpartisan, nonprofit journalism venture committed to explaining how California’s state Capitol works and why it matters.

Contact and issues

For bugs, unclear documentation, or questions about the pipeline, use GitHub Issues or send us an email.

Note on AI-assisted work

We used AI coding tools to assist with the analysis. All results were manually reviewed by a member of our team.

License

Code in this repository is licensed under the Apache License, Version 2.0; see LICENSE. Dependency licenses are managed by PyPI packages; see NOTICE.