FMCSA, FARS, and NTSB: Three Crash Sources That Answer Different Questions

FMCSA: reported commercial-motor-vehicle crash information

FMCSA describes its public Crash File as records drawn from federal, state, and local police crash reports involving commercial motor carriers. It publishes daily data from a database that is already 24 hours old and warns that the file is not real time. Reporting and processing delays remain relevant even when a downstream download is recent.

The agency also explains that one crash may have multiple records because a separate report is entered for each involved commercial motor vehicle. Its documented identifying fields include report state, number, date, time, and sequence. Counting rows without understanding those relationships can misstate the number of incidents or vehicles involved.

Sources: FMCSA Open Data Program

FARS: annual information about fatal traffic crashes

NHTSA describes FARS as a nationwide census of fatal motor-vehicle traffic crashes, with yearly data available for research. That population differs from a broader commercial-vehicle crash file. A nonfatal event can appear in another source without belonging in FARS, so the absence of a FARS result cannot establish that a carrier has never had a crash.

For historical analysis, preserve the year and release version used. Consult the year-specific documentation before interpreting carrier identifiers or coding fields. A field that resembles a USDOT number needs its documented identifier type; missing or unknown codes must not become actual carrier numbers. Record the matching method alongside the result so it can be reviewed later.

Sources: NHTSA FARS overview and documentation

NTSB CAROL: investigations and recommendations

CAROL provides searchable NTSB investigations and recommendations across transportation modes. Its help page describes data for highway and other non-aviation modes from 2010 onward. This is an investigation resource, not a complete census of every carrier’s crashes. A search returning nothing should be described as no matching investigation found.

An investigation may offer detailed factual material, findings, and recommendations that a summary crash count cannot supply. Identify the case number, status, report date, and the document being cited. A preliminary account and a completed investigation are not equivalent evidence. If a later report changes the account, keep a record of the revision rather than silently overwriting the earlier review.

Sources: NTSB CAROL help

Link observations without erasing their differences

Imagine one fictional fatal highway crash involving two commercial vehicles. There may be two vehicle-level records in one dataset, a fatal-crash record in another, and an investigation file in a third. Combining those rows into a single total of four crashes would be wrong. Collapsing them without preserving which vehicle and carrier each record describes would also lose information.

Use available event dates, locations, report or case identifiers, and vehicle relationships to investigate overlap. Preserve each source observation and explain the connection. If evidence is insufficient to establish that two records describe the same event, mark the relationship unresolved rather than choosing whichever total looks cleaner.

A fair comparison needs a denominator and a window

Raw counts do not describe exposure. A large fleet and a small fleet can have the same count while operating very different numbers of miles and vehicles. A rolling period and a calendar-year file can also produce different totals without a contradiction. Before comparing carriers, document the population, period, missing data, and any exposure measure used.

Do not manufacture a rate by dividing recent crashes by an old or unverified mileage figure. If compatible exposure data are unavailable, show the observations and the limitation. A precise-looking rate built from mismatched dates is less informative than a clearly labeled count. Likewise, missing inspection or crash data should remain unknown rather than automatically become favorable evidence.

How this fits into a broker review

When information is disputed, preserve the disputed record and the carrier’s supporting material, then follow the relevant official correction route. FMCSA points users to DataQs for challenges to safety information. Do not erase a record merely because it is disputed, or present an unresolved challenge as a completed correction.

Sources: FMCSA: SAFER frequently asked questions

Keep source education separate from product availability

These databases are useful research sources, but explaining them does not mean every record is already integrated into VerifyCarrier. FARS and CAROL are research candidates in the current enrichment design. Check product coverage before expecting them in a carrier panel. A clear source guide should help brokers investigate an issue even when the relevant evidence remains outside the product.