FMCSA Crash and Carrier Data: A Method Guide

There is no single federal number for commercial vehicle crashes. There is a report that assembles several collections which were built to different specifications, and understanding why they disagree is more useful than picking one.

The assembling report

Large Truck and Bus Crash Facts is a recurring annual report from the Federal Motor Carrier Safety Administration containing descriptive statistics on fatal, injury, and property-damage-only crashes involving large trucks and buses. Its own landing page states that the content is compiled by the agency's Crash Data Analytics Division, and then lists the collections it draws on.

A detail worth recording, because it is the kind of thing that quietly propagates: the page introduces the list as four major sources and then sets out five bullets. Anyone citing a count of contributing sources should read the list rather than the sentence above it. The list is what follows.

The five collections named

The Fatality Analysis Reporting System, published by the National Highway Traffic Safety Administration, described there as a census of fatal crashes involving motor vehicles travelling on public trafficways. Census is the operative word. It aims at completeness rather than at a sample, which is why it supports fatality counts and cannot support anything about crashes without a fatality.

The General Estimates System, a probability-based nationally representative sample of police-reported crashes across all severities. It was retired in 2017 and replaced.

The Crash Report Sampling System, which builds on its predecessor and begins with data for 2016. The agency attaches a warning to it that is the single most consequential sentence on the page for anyone building a time series:

Although CRSS and GES are both samples of police-reported crashes involving all types of motor vehicles, CRSS includes a more efficient and flexible sample using updated traffic and demographic information. As a result, comparisons of 2016 (and later) CRSS estimates with older GES estimates should be performed with caution.

Federal Motor Carrier Safety Administration, Large Truck and Bus Crash Facts

A trend line drawn straight across that boundary is not measuring a change in crashes. It is comparing two sampling designs and reporting the difference between them as news.

The Motor Carrier Management Information System crash file, which is FMCSA's own collection and the one that connects crashes to identified carriers.

Highway Statistics, from the Federal Highway Administration, which supplies exposure. Crash counts without vehicle miles travelled cannot produce a rate, and rates are usually the thing people actually want.

Which source answers which question

The practical value of the list is that it partitions the questions.

• How many people died: the fatality census, because it is a census.
• How many crashes of all severities occurred: the sampling system, as an estimate with sampling error attached, never as a count.
• Whether a rate is rising: crash figures over exposure from Highway Statistics, with both series checked for definitional breaks.
• Anything about a specific carrier: the carrier-linked records, not the national series.

Mixing these is the standard error. A sampled estimate quoted as though it were a count, or a fatality census used to say something about injury crashes, will both produce a number that survives casual checking and is answering a question nobody asked.

Carrier-level lookups

For an individual operator, the agency runs SAFER, which returns a company snapshot from the carrier information system. It is a live administrative record rather than a research dataset, and it should be read as an answer to what the agency currently holds about this operator rather than as a complete safety history.

Two limits are worth stating. The record reflects reported and processed events, so recent matters may not appear. And carrier identity is administrative: operating authority can be reorganised, which means the entity in the system and the business a reader has in mind are not always the same thing across a long period.

Broader aggregates and the agency's own analytical products sit in its Analysis and Information portal, and a number of the underlying files are published through the Department of Transportation's data portal, which is generally the more convenient route for bulk work.

Access notes, because they cost time

Both the FMCSA and NHTSA web hosts reject automated requests with a forbidden response while serving the same pages normally to an ordinary browser. Adding browser headers to a command line client does not change this. The consequence for anyone maintaining citations is specific and counterintuitive: an automated link checker will report these pages as broken when they are fine, so a failure against these hosts has to be confirmed by hand before acting on it.

The opposite failure is also present. The agency's own landing page for the carrier information system could not be located at any of three documented addresses, all of which return not-found. That is a genuinely dead reference, not a blocked one, and the working route is through the analysis portal instead. Both conditions look identical to a script and are opposite in meaning.

Published report years currently run through 2022. Anyone needing something more recent is working with the underlying collections directly rather than with the assembled report, which is a materially different exercise.

The companion guide covers the Texas state collection, which records a great deal the federal sources never see and excludes a category they include. For the analytical material that this kind of source feeds, our other work on court data is the way back.