What this dataset measures
Built to test the common claim that the night before Thanksgiving ("Blackout Wednesday") is the deadliest night of the year for drunk driving. Every night is included so any holiday can be ranked against all others, not only against other holidays.
- Geography
- United States
- Unit of observation
- One night (6 p.m. to 5:59 a.m.), United States
- Update cadence
- Annually, when NHTSA releases a new FARS data year
- Curator
- FactHub Research Team
Supporting files
Source snapshots, manifests, quality reports, documentation,
and other materials used to produce or review this release.
-
analyze.py
Transformation manifest · text/x-python · 7.8 KB (7,951 bytes)
Reads the FARS accident and MIDRVACC files, assigns each crash to an hour, and sums alcohol-impaired deaths into nights; prints the named-night table.
-
export.py
Transformation manifest · text/x-python · 2.1 KB (2,146 bytes)
Writes the three CSV tables from the tables built in analyze.py.
-
build_dataset.py
Transformation manifest · text/x-python · 6.9 KB (7,054 bytes)
Computes every saved-analysis result in this dataset (window sensitivity, holiday periods, head-to-head counts).
-
results-for-dataset.json
Derived table · application/json · 13.6 KB (13,885 bytes)
Every saved-analysis result as JSON, written by build_dataset.py.
Saved analyses
Annual alcohol-impaired and total traffic deaths
Question: Do our totals match NHTSA's published national counts?
Result: Alcohol-impaired deaths ranged from 10,196 (2019) to 13,599 (2021).
Operation: Sum over all crashes each year, including crashes with unknown hour.
Scope: United States, 2015-2024
View this analysis
- Counts only deaths in crashes on public roads that FARS records; injuries and non-fatal crashes are not included.
- Alcohol-impaired deaths are estimates: when a driver's blood-alcohol test result is unknown, NHTSA fills it in with 10 imputed values.
- 2024 data are from NHTSA's first (annual report) release of the 2024 file, posted April 1, 2026; the final file, due about a year later, can change counts slightly.
Replication
FactHub's estimates beside published external estimates of the
same quantity. The difference is FactHub's value minus the
external value.
Sources
-
NHTSA Fatality Analysis Reporting System (FARS), national CSV files 2015-2024
Crash-level records: date, hour, deaths (accident.csv) and imputed highest driver BAC (MIDRVACC file, 10 imputations).
Retrieved November 16, 2026. Public domain (U.S. government work). Unrestricted; FactHub links the files rather than hosting them.
Reliability:
Government.
FARS is a census of fatal crashes on U.S. public roads, coded by state analysts under contract with NHTSA. Blood-alcohol values missing from police and medical records are filled in by NHTSA's multiple imputation.
-
NHTSA, Fatality Analysis Reporting System Analytical User's Manual, 1975-2024 (DOT HS 813 794), Appendix E
Defines the MIDRVACC file (10 imputed values of the highest driver BAC in each crash) and the BAC value range used to apply the .08 threshold.
Retrieved November 16, 2026. Public domain (U.S. government work). Linked, not hosted.
Reliability:
Government.
Official documentation of the FARS files by NHTSA's National Center for Statistics and Analysis. Appendix E (p. E-2) describes the imputed alcohol files.
-
NHTSA Traffic Safety Facts, Alcohol-Impaired Driving: 2024 Data (DOT HS 813 815)
Published national alcohol-impaired death counts used to check this dataset's totals (2023 final file: 12,382; 2024: 11,904).
Retrieved November 16, 2026. Public domain (U.S. government work). Linked, not hosted.
Reliability:
Government.
NHTSA's annual fact sheet on alcohol-impaired driving, computed from the same FARS files with the same imputation method.
-
NHTSA Traffic Safety Marketing, Thanksgiving: Buzzed Driving Is Drunk Driving (2026 campaign page)
NHTSA's Thanksgiving holiday window (6 p.m. Wednesday to 5:59 a.m. Monday) and its 2024 count of drunk-driving deaths in that window, used as a check.
Free copy
Retrieved November 16, 2026. Public domain (U.S. government work). Linked, not hosted.
Reliability:
Government.
NHTSA's public-campaign page for the Thanksgiving impaired-driving campaign. NHTSA runs the campaign, so it has an interest in highlighting the risk, but its counts come from FARS.
License: FactHub's selection, cleaning and combined tables are
licensed under CC BY 4.0.
Source data keeps the terms listed above (terms).
Limitations and reproducibility
Reproducible: Download the FARS national CSV files for 2015-2024 from NHTSA and run analyze.py, export.py and build_dataset.py (Python 3 standard library only) on the folder holding them. The scripts are attached.
- Counts only deaths in crashes on public roads that FARS records; injuries and non-fatal crashes are not included.
- Alcohol-impaired deaths are estimates: when a driver's blood-alcohol test result is unknown, NHTSA fills it in with 10 imputed values.
- 2024 data are from NHTSA's first (annual report) release of the 2024 file, posted April 1, 2026; the final file, due about a year later, can change counts slightly.
- Crashes are placed in time by their recorded local hour; 1,248 alcohol-impaired deaths over the ten years had an unknown hour and are not assigned to any night.
- A death counts on the night of the crash, even if the person died later (FARS counts deaths within 30 days).
- Counts are deaths, not drunk drivers on the road. More travel on a night can raise deaths without more drinking per driver.
- The Dec. 31, 2024 night is left out because its early-morning hours fall in 2025 data, not yet released.
Publications using this dataset
Version details
The exact citing versions, and the version of this dataset each cites.
-
Article
thanksgiving-eve-drunk-driving-deaths —
version 2
(published) cites version 1
-
Chart
thanksgiving-eve-drunk-driving-rank —
version 1
(published) cites version 1
-
Briefing
thanksgiving-facts —
version 3
(published) cites version 1
-
Briefing
thanksgiving-facts —
version 2
(published) cites version 1
Node: dataset:us-drunk-driving-deaths-by-night:1:analysis:annual-totals