How the figures are made

Last checked against the NHTSA pages: 2 Oct 2026

Each count comes from the national FARS files of 1975 to 2024. The rates use the population estimates of the Census Bureau and the vehicle-mile tables of the Federal Highway Administration (FHWA). This edition rests on the FARS 2024 annual file. Data to 31 Dec 2024.

What FARS counts

FARS holds each crash on a public road in which a person died in 30 days or less after the crash. A crash on private property is not in FARS. A crash with no death is not in FARS.

"Fatal crashes" is the number of these crashes. "Deaths" is the number of persons who died in them. The glossary has the words of NHTSA. The guide How FARS counts a fatal crash gives more detail.

The annual file and the final file

NHTSA releases the first file of a crash year as the annual file. It replaces this file with a final file about one year later. The final file has the cases and the updates that came after the annual file.

The newest year on this site is an annual file until NHTSA releases its final file. Counts of that year can change. The page FARS release dates shows which file each year has.

Years and periods

The accident table is read for each year from 1975 to 2024.
The vehicle and person tables are read from 2015. Their codes for the factors are the same in these years.
"The ten years" are 2015 to 2024. Ten-year rates, factor counts and the tables by weekday, time and road class use these years.
Route pages use the records of routes from 2000.
The years before 2015 are read once from a fixed copy of the files. The site does not track later revisions of these years.

Deaths for each 100,000 residents

The rate of one year divides the deaths of the year by the residents of the year, times 100,000. The rate of the ten years divides the deaths of the ten years by the sum of the residents of each year. A rate has one decimal.

The residents are estimates of the Census Bureau. Each year comes from one file:

2000 to 2009: the intercensal estimates of 2000 to 2010.
2010 to 2019: the estimates of vintage 2020.
2020 to 2024: the estimates of vintage 2025.

The glossary explains a vintage. A new vintage replaces the estimates of all earlier years, and the rates change with it. A change of vintage is not a change in the data of NHTSA.

The vintage 2025 file of the Census Bureau has the planning regions of Connecticut and not its eight counties. FARS still codes Connecticut by county. This site uses the estimate of 2020 for each of these counties from 2021. The rates of these counties are less exact for that reason.

This site has no rate before 2000. A rate is for the residents of a place. It is not a rate for the persons who use its roads. The deaths include persons who do not live in the place.

Deaths for each 100 million vehicle-miles

This rate divides the deaths in a state by the vehicle-miles that vehicles travel in the state. The miles are those of table VM-2 of the FHWA, from 2015. The United States line uses the total of the 50 states and the District of Columbia. Only states and the United States have this rate.

The FHWA gives the miles of some road classes as estimates of the states. The glossary has the note of the FHWA.

The classes of the county map

A county map shades each county by its rate of the ten years. The limits of the classes are the same for every state, so one shade has one meaning on every map.

Fewer than 10 deaths for each 100,000 residents in a year.
10 to 14.9 deaths for each 100,000 residents in a year.
15 to 19.9 deaths for each 100,000 residents in a year.
20 to 29.9 deaths for each 100,000 residents in a year.
30 or more deaths for each 100,000 residents in a year.

A county with few residents can show a high rate from few deaths. A class shows the size of a rate. It is not a judgement of the county. A county with a count of 1 or 2 has no class. The map shows it with a hatch, as the table has no rate for it.

Counties

A crash is in the county that FARS records. Each county of the Census Bureau's list with a FARS record has a page. FARS codes Connecticut by its eight old counties, and this site does the same.

The Census Bureau changed a few county codes after 1975. This site joins each of these old codes to the code of today, so that the series by year has no break. For some Alaska areas that were divided, the join is an approximation.

Factors

A factor is something that FARS records about a crash or a person. A recorded factor is not the cause of a crash. The list below names the FARS field of each factor. The glossary has the words of NHTSA for each field.

A driver recorded as drinking (DR_DRINK): fatal crashes in which FARS records one or more drivers as drinking. FARS does so when police reported alcohol involvement or when a test of the driver was positive for alcohol. This is not NHTSA's alcohol-impaired measure, which uses a blood alcohol concentration of .08 g/dL or higher and estimated values.
A speeding-related driver (SPEEDREL): fatal crashes with a driver coded as racing, above the speed limit, too fast for conditions, or speeding with no detail.
Pedestrians, bicyclists and motorcyclists (PER_TYP and BODY_TYP): persons killed of each type. A bicyclist count includes other pedalcyclists. A motorcyclist is a driver or passenger of a vehicle with a body type of 80 to 89.
No restraint used (REST_USE): occupants killed with no restraint in use, among the occupants killed whose restraint use FARS records. Motorcycle riders are not in this count.
In the dark (LGT_COND): fatal crashes with a light condition of dark, among those with a reported light condition.
On a rural road (RUR_URB): fatal crashes on a road coded rural, among those coded rural or urban.

NHTSA says that alcohol data is often missing, so the count of drivers recorded as drinking may be too low. For the years before 2015 the count is from the element DRUNK_DR, which counts a blood alcohol concentration of .01 g/dL or more. From 2015 the count counts .001 g/dL or more. NHTSA says that DRUNK_DR is unreliable for 1977, 1981 and 1999 to 2007, and this site gives no count for these years.

Routes

A route is the trafficway that FARS records for the crash, where FARS classes it as an Interstate, a US highway or a state highway. A route has a page when FARS records 20 or more fatal crashes on it in one state from 2015 to 2024. The same route number in a different state is a different page. A local street or a county road has no page.

Arkansas writes the section of a route after its number, for example "SR-18-04". This site reads the number before the section as the route. For an Interstate or a US highway, it checks the number against the routes of Arkansas.

FARS has a second trafficway field for a crash at a junction. This site reads the first trafficway only. A trafficway with two route numbers counts under the first number.

A route page that was built stays on the site. If the route goes below 20 fatal crashes, its page says so. The tables of mile ranges show the crashes that have a milepoint in FARS.

Small counts

A count of 1 or 2 for one county, one route or one county in a year is written "1 or 2". No rate is given for it, and the map gives it no class.
A county with fewer than 20 deaths in the ten years shows totals of five years, not single years.
A share is given only where the base is 10 or more fatal crashes.

Pages and the search index

Each county of the Census Bureau's list with a FARS record has a page. Search engines list a county page when the county has 100 deaths or more in the ten years 2015 to 2024. A count of this size gives a rate and shares that are stable enough to quote.
A route has a page when it has 20 fatal crashes or more in the ten years. Search engines list a route page when the route has 200 fatal crashes or more.
Each state has a page. Search engines list the pages of the 2023 and 2024 years of a state. The pages of the earlier years stay on the site, and the page of the state links each one.
The pages of the states, the lists of routes, the factors, the findings, the guides and the other hub pages are always listed.
A page that exists keeps its address. A page under a limit of the index stays, and you can open it from the lists of the site. The limits keep the search index to pages with enough figures of their own.

Findings

A finding answers one question for all states or for the largest counties. Its table orders places by a stated figure. The figure is a count, a change, or a rate for each 100,000 residents or each 100 million vehicle-miles. No finding orders routes. A county is ranked by rate only with 200 deaths or more in the ten years. This stops a small count from making a large rate. A change in percent needs 30 deaths or more in the earlier year. The order is not a judgement of a place.

Editorial policy

Every figure comes from a source file that the page names. The program writes each sentence from the figures. The same fact has the same words on every page.
Tables of named places are in name order. The pages of findings are the one exception: they order states or counties by a figure that the page names, and they never order routes.
No page says that a figure is good or bad. No page names a cause.
A page uses the terms of NHTSA. Where NHTSA has a definition, the site quotes it, with its source and the date on which the site copied it.
The site has no page about one crash or one person.
An error that the site made is corrected at the next build and recorded on the page Corrections.

Known gaps

Puerto Rico is in FARS and is not on this site. Some tables of NHTSA include it.
NHTSA's alcohol-impaired figures use estimated values. This site gives the count of drivers recorded as drinking, so its figures are different.
The newest year is an annual file until NHTSA releases its final file. The counts can change.
The factors and the road classes are from 2015. Earlier years have counts of crashes and deaths only.
FARS writes some route numbers in a form that this site does not read. A crash on such a trafficway is in the totals of its county and not on a route page. Counts of routes can be lower than the FARS records.
FARS wrote most US routes of some states with the number of a state route, for example "US-SR 1". These are Alabama in 2000 to 2018, Georgia in 2019 and Tennessee in 2000 to 2011. This site does not read this form, so a US route page of these states shows too few crashes in these years. The page says so above its table of years.
This site does not guess a route from a road name or from a state route number. A crash on a trafficway with a name and no route number has no route.
This site has no figures for one city or town. A city in the search opens the page of its county.
The county map uses a Census Bureau boundary file of 2021, with generalised outlines.