U.S. county-level election results: President, U.S. House and U.S. Senate, 1990-2024

Heideman, Paul

doi:10.5281/zenodo.22976553

Open dataset · v1.0.0 · CC BY 4.0

County-level U.S. election results: House, Senate and President, 1990–2024

Free, documented county-level election data for the United States: every regular U.S. House general election by county since 1990, plus Senate and presidential results. Each row gives the candidate, party and votes in one county, and says where the numbers came from.

Download all files (zip, 9.5 MB) Explore the map Zenodo · DOI

Map of the 48 contiguous states shaded by county: the Democratic share of the two-party vote for U.S. House in 2024, from deep red to deep blue. — U.S. House, 2024: Democratic share of the two-party vote in each county. Republican Democratic Unopposed seat, not on the ballot No data Solid blocks are uncontested races. The interactive map covers every year and office, with candidate names on hover.

At a glance

54,939House county-years, 1990–2024

37,283Senate county-years, 1990–2024

27,969Presidential county-years, 1992–2024

1,523documented sources

County-level House results back to 1990. Open datasets usually start in 2016 (MEDSL) or cover only some states and years (OpenElections). The earlier years here come from official state reports, canvasses and blue books, many transcribed from scans or microfiche.
Candidate-level, not just shares. Names, party labels (including fusion lines such as Working Families), a DEM / REP / OTHER party group, and votes.
Split counties handled. A county divided between congressional districts has one row per district, and the summary file adds them up to one row per county.
Every number traceable. Each row names its source. SOURCES.csv describes each of the 1,523 sources: publisher, document, where to find it, how it was transcribed, and its terms of use.
Errors found and fixed in the open. 135 entries in data_corrections_log.csv record source errors (doubled counts, missing counties, wrong party labels) and how each was handled. Checked against FEC and state certified totals.
Gaps listed, not hidden. 41 state-years are not fully covered. Most are House seats won unopposed and never printed on the ballot, or Senate years with only a special election. Each one is listed with its reason.

Download

Files are attached to the GitHub Release and archived on Zenodo. Column definitions are in DATA_DICTIONARY.md.

File	Contents	Size
us_county_results_long.csv .parquet	One row per year × office × county × district × candidate × party line: candidate, party, party group, votes, source, quality flag	46 MB 3 MB
us_county_results_summary.csv	One row per year × office × county: Democratic, Republican, other and total votes; two-party and total shares	12 MB
us_county_results_gaps.csv	Every state-year-office not fully covered, with the reason	5 KB
us_county_results_no_ballot.csv	House seats whose unopposed winner was not on the ballot (FL, LA, OK) or not tabulated (AR), by county	133 KB
SOURCES.csv	Every source: publisher, document, locator, transcription method, license terms	1.7 MB
data_corrections_log.csv	Every error found in the sources and how it was handled	109 KB

Quick start

Read FIPS codes and districts as text so leading zeros survive.

# R
library(readr)
url  <- "https://github.com/pmheideman/us-county-election-results/releases/download/v1.0.0/us_county_results_summary.csv"
cty  <- read_csv(url, col_types = cols(.default = col_character()))
house_2024 <- subset(cty, office == "house" & year == "2024")

# Python
import pandas as pd
url = "https://github.com/pmheideman/us-county-election-results/releases/download/v1.0.0/us_county_results_long.parquet"
long = pd.read_parquet(url)   # year and votes are integers; FIPS codes are strings
senate = long.query("office == 'senate' and year >= 2016")

Coverage

Office	Years	County-years	Candidate detail
U.S. House	1990–2024	54,939	Every candidate, per district
President	1992–2024	27,969	Every candidate from 2000; the two major-party nominees before that
U.S. Senate	1990–2024	37,283	Every candidate from 2016; the two major-party nominees before that

48 contiguous states; Alaska, Hawaii and DC are not included. Regular general elections only (Louisiana and Texas runoffs are included where they decided the seat); special elections are excluded. Counties use 5-digit FIPS codes.

Frequently asked questions

Where can I download county-level U.S. House election results?

Here. This dataset has county-level results for every regular U.S. House general election from 1990 to 2024 in the 48 contiguous states, with candidate names, parties and votes. It's free under CC BY 4.0. Use the download links above or the Zenodo record.

How are counties split between congressional districts handled?

The long file has one row per county, district and candidate, so a county split between three districts appears three times, once per district. The summary file adds these up to one row per county and year, with the number of districts in n_districts.

How is this different from MEDSL, OpenElections or Dave Leip's Atlas?

MEDSL's county presidential file starts in 2000, and its House and Senate precinct files start in 2016. OpenElections is volunteer-built, so coverage varies by state and year. Leip's Atlas sells its data. This dataset starts from those open sources and fills the rest from official state reports and canvasses to reach back to 1990. Errors found along the way are corrected and logged, and every row says which source it came from.

Can I get vote shares instead of raw votes?

Yes. us_county_results_summary.csv has Democratic, Republican, other and total votes, the Democratic two-party share (D ÷ (D + R)) and the Republican share of all votes, for every county, office and year.

Why are some counties blank for the House?

Mostly because the seat was won unopposed. Florida, Louisiana and Oklahoma leave unopposed candidates off the ballot, and Arkansas doesn't count their votes, so no county returns exist. These are listed in us_county_results_no_ballot.csv. Six state-years have no county-level source found yet; they're listed in us_county_results_gaps.csv.

Is the data free to use, including commercially?

Yes. The compiled data is licensed CC BY 4.0: use it for anything, with credit. Vote counts are public records, but a few state websites publish results under their own terms. The license columns of SOURCES.csv list them.

How accurate is it?

State totals are checked against the FEC's certified results and the states' own canvasses. For President (2000–2024), 648 of 658 state-level nominee totals match the certified figures within 0.5%. House and Senate results (2004–2022) are compared with the FEC district by district and state by state. Known remaining problems are listed in the corrections log and flagged row by row in quality_flag.

Cite

Heideman, Paul (2026). U.S. county-level election results: President, U.S. House and U.S. Senate, 1990–2024 (v1.0.0) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.22976553

Please also cite the upstream sources you rely on (listed in SOURCES.csv). The code that builds the dataset is on GitHub under the MIT license. Design and data decisions are documented in DECISIONS.md, and changes between releases in the changelog.