How We Count

How We Count

You never have to take our word for a figure. Every number on this site starts as a raw federal record and ends as a figure you can check. Not marketing copy. Not talking points. This page shows every step in between, with every caveat named.

We count every death, not a sample

Our death counts come from the federal fatal-crash census. NHTSA calls it the Fatality Analysis Reporting System, or FARS. We use FARS data from 2000 through 2024. FARS is a full count of every fatal motor vehicle crash on a U.S. public road, not a survey or a sample. It draws on police reports, coroner and medical examiner records, and state crash files. A crash counts if it killed someone within 30 days, so every death on this site is counted, not estimated. NHTSA usually finalizes a year's data about two years after it ends, and our numbers update on that same schedule.

We estimate injuries, and we say so

FARS only counts deaths, so we turn to NHTSA's Crash Report Sampling System (CRSS) to estimate injuries. CRSS isn't a full count. It's a sample of police-reported crashes of every severity. NHTSA scales it up to stand in for the whole country. That's why every injury number on this site is labeled an estimate, with a link back to this page. CRSS puts the ratio at about 54 people injured for every person killed. That's a national estimate, and it only holds for recent years. CRSS can't support a state or county injury count, so we never present one as a state or local statistic.

We adjust for population

Raw death counts favor big states, so we adjust for population before comparing places. Every rate on this site is deaths per 100,000 residents. We use U.S. Census Bureau estimates for that same year. We always use the latest population number available for that place.

We keep comparisons honest

NHTSA changed how it labels hit-and-run crashes over time. From 2009 through 2019 it used a separate "unknown" label. NHTSA retired that label later. Every hit-and-run trend or comparison uses 2009 through 2024 data only. The labels don't match further back than that.

What we won't claim

A federal dataset can't answer every question. Some numbers would require guessing at a cause NHTSA doesn't record. Some would rank small counties on a single year's rate, or combine two datasets that measure different things. When a number would need us to guess, we say so instead of publishing it.

When we get one wrong, we fix it

If you find an error in a number, a broken source link, or a method we've described wrong, reach us through our about page. We correct published numbers as soon as we confirm the fix, and we note meaningful changes on the page they affect. See our editorial policy for the full correction process. Our data sources page shows where every dataset comes from.