How we compute these numbers

The sources

Crash and fatality counts come from the Fatality Analysis Reporting System (FARS), the federal government's crash-level record of every traffic fatality in the country, maintained by the National Highway Traffic Safety Administration.

Population figures for per-capita rates come from the U.S. Census Bureau's county and place population estimates, matched to the same years as the crash data.

We pull both datasets ourselves, run our own queries against them, and publish the result. We don't report a number we can't reproduce.

Fatal crashes only

Every count and rate on our Road Safety pages counts fatal crashes, crashes where at least one person died. We don't include injury-only crashes or property-damage crashes. This makes our numbers smaller than "all crashes" totals you might see elsewhere, and it's deliberate: fatal crashes are the most completely and consistently reported category in FARS, so it's the honest baseline for comparison across cities, counties, and years.

The per-capita rate

A raw crash count favors big cities and counties just because more people live there. To compare places fairly, we also compute a rate per 100,000 residents, annualized: fatalities divided by total person-years in the window, times 100,000.

This is why a smaller city can rank more dangerous than a bigger one even with fewer total deaths. Both the raw count and the rate are real. They answer different questions, so we show both.

Trend figures

A trend figure compares the first and last year of a stated window (for example, "2015 vs. 2024"). We show the full year-by-year table alongside every trend claim, so you can check the math yourself instead of taking our word for the percentage.

Corridors, not exact locations

Where we rank a road, we rank it at the named-corridor-by-county level (for example, "I-10, Maricopa County"), not an exact mile marker or intersection. FARS records a road name and a county, not GPS-precise crash locations we'd trust for that granularity, so we don't claim more precision than the data supports.

Why we don't publish an intersection-by-intersection ranking

Arizona's fatal-crash records leave the intersection-type field blank far more often than the rest of the country does. We could estimate around that gap, but a ranking built on an estimate would read as more precise than it is. Instead, we publish the gap itself as a finding: it's a real, useful fact about Arizona's crash data, and it's more honest than a ranking we can't stand behind.

Pedestrian figures

A pedestrian fatality is any person coded as a pedestrian in FARS who died in the crash. We track this as its own category because pedestrian deaths have moved differently than the overall crash trend in Arizona over the past decade.

Motorcycle figures

A motorcycle fatal crash is any fatal crash where FARS codes at least one involved vehicle as a motorcycle or moped body type. We count the crash, not the number of people on the bike.

Truck and commercial vehicle figures

A truck/commercial fatal crash is any fatal crash where FARS codes at least one involved vehicle as a cab-chassis, single-unit straight truck, truck-tractor, or medium/heavy truck body type.

Speeding figures

A speeding-related fatal crash is one where FARS codes a vehicle's speed as a contributing factor, including racing, exceeding the posted limit, or driving too fast for conditions.

Teen-driver figures

A teen-driver fatal crash is one where FARS records a driver aged 15 to 19 among the people involved. This counts the crash where a teen was driving, not necessarily who was at fault.

Senior-driver figures

A senior-driver fatal crash is one where FARS records a driver aged 65 or older among the people involved. As with teen-driver figures, this counts involvement, not fault.

The small-sample floor

We don't publish a rate or ranking built on fewer than 10 records for a given geography and category. Below that floor, a single crash can swing the number too far for the result to mean anything. Where a county or category falls below the floor, we say so instead of publishing a number that looks precise but isn't.

Where a real finding sits in a small-population place (La Paz County's fatality rate, for example), we don't suppress it, we label it. The sample is healthy, but a handful of crashes moves the rate a lot in a place with few residents. We say that plainly on the page instead of hiding the caveat.

What we don't publish yet: impaired-driving figures

We're still finalizing our impaired-driving numbers. Two different ways of counting alcohol-involvement in our own records disagree with each other, and we won't publish a DUI figure until we've resolved which definition is right. Check back, or ask us for this cut directly.

What we deliberately don't claim

• We don't claim exact-intersection or exact-location precision. Our corridor rankings go to road-name-by-county, no finer.
• We don't rank a geography or category below our 10-record floor.
• We don't count injury-only or property-damage crashes in any figure on these pages, only fatal crashes.
• We don't publish a DUI/impaired-driving figure until our internal definition discrepancy is resolved.
• We don't treat a composite score (where we publish one) as a measured fact. We label it as a model and show the formula.

Questions about a number

If a figure on our site looks wrong to you, or you want a cut of the data we haven't published, reach out. We'll check our work.