The federal safety rating barely predicts claim risk — it covers under 3% of carriers, and even there, its link to FMCSA's own crash data is not statistically significant

Summary

No. The federal safety rating — the Satisfactory/Conditional/Unsatisfactory label FMCSA assigns under 49 CFR Part 385 — is a weak and largely unusable predictor of claim frequency, for two separable reasons. First, it barely exists: only 2.4% of active motor carriers (53,222 of 2,237,648) carry any rating at all; the other 97.6% are simply "Not Rated" because FMCSA has never conducted a compliance review on them. Second, even among the carriers that DO have a rating, a live regression against FMCSA's own crash data shows essentially no explanatory power (R² = 0.0003) and the difference in mean crash rate across Satisfactory/Conditional/Unsatisfactory is not statistically significant (ANOVA F=1.73, p=0.18, n=13,334). No government table anywhere publishes cargo or liability insurance claims at the carrier level — that data is proprietary to private insurers — so this analysis uses FMCSA's own recordable crash rate as the closest available public proxy, and even on that weaker test the rating does not clear significance.

Step 1 — What "claims" data actually exists at carrier grain (and what doesn't)

I searched this corpus's full catalog for cargo/liability claims, loss ratio, and insurance-claim data joinable to individual motor carriers. Nothing exists at that grain. transport.fmcsa_carriers (the FMCSA Company Census File, ~4.47M rows) carries registration, fleet size, and — only for carriers that received a compliance review — a safety_rating and a computed recordable_crash_rate (crashes per million miles). It carries no cargo-damage, liability-loss, or insurance-claim field of any kind, and the catalog search explicitly flagged "trucking" as an unmatched term with no table covering claims. Cargo and liability insurance claims are transactions inside private carriers (Progressive, Sentry, Reliance Partners, Great West, etc.) and are not reported to any federal agency in a form that becomes public microdata — unlike NFIP flood claims or crop-insurance indemnities, which ARE public because a federal program pays them. This is a genuine, structural data gap, not a search failure: private commercial-liability claims are simply outside what any government source publishes at the carrier level.

Given that gap, the analysis below does two things: (1) reports what the published research literature says about the predictive validity of FMCSA's safety-scoring systems for crash risk (the outcome government researchers actually study), and (2) runs the closest available live test using this corpus's own carrier-level crash data as a proxy for claim-triggering events, since a reportable crash very often generates both a liability claim and, if freight is damaged, a cargo claim — while noting explicitly that a proxy is not the same thing as a claims count (crashes are one but not the only driver of cargo/liability claims; theft, refused loads, and handling damage generate claims with no crash involved at all).

Step 2 — The federal safety rating covers almost no carriers

Querying live: of 2,237,648 active USDOT-registered carriers, only 53,222 (2.4%) carry an official safety rating — 41,219 Satisfactory, 11,234 Conditional, and just 769 Unsatisfactory. FMCSA only assigns this rating after a compliance review, and reviews are triggered selectively (complaints, crash involvement, size, prior violations), so the rated population is not a random sample of carriers — it already skews toward carriers FMCSA had a reason to look at. For 97.6% of active carriers, the field is simply blank. A predictor that is undefined for the overwhelming majority of the population it's being asked to predict over cannot be doing much predicting for the fleet as a whole, regardless of how it performs on the sliver where it exists.

Step 3 — Live test: does the rating predict FMCSA's own crash rate, among carriers that have one?

Exclusion disclosure, checked live against the same table. The predicate recordable_crash_rate IS NOT NULL excluded 39,888 of the 53,222 rated carriers (74.9%), because FMCSA only computes this field for a carrier with a compliance review that also produced qualifying inspection/mileage data. The predicate recordable_crash_rate NOT IN (999.0,999.99,999.999) excluded zero additional rated carriers — none of FMCSA's three "not computed" sentinel values are present among rated carriers, so this predicate excluded nothing beyond what the NULL filter already excluded. The predicate power_units IS NOT NULL likewise excluded zero rated carriers, since that field is never null in this table. The predicate mcs150_mileage IS NOT NULL excluded a further 9 rated carriers from the regression sample only. Net: 13,334 of 53,222 rated carriers (25.1%) remain, and this surviving quarter is not a random subset — carriers with a valid crash rate have larger average fleets (avg 26 power units for Satisfactory-rated carriers with a valid rate vs. 7–9 for Conditional/Unsatisfactory) than the rated population as a whole. There is no valid "headline stat without this exclusion" to show alongside it: the 39,888 excluded carriers have no crash-rate value to average, and substituting zero for a missing measurement would manufacture a number FMCSA itself does not report.

Restricting to the 13,334 rated carriers with a valid rate: mean crash rate rises in the expected direction — Satisfactory 1.48, Conditional 2.02, Unsatisfactory 2.31 crashes per million miles — compared with 1.20 for unrated carriers with a valid rate. That ordering looks like confirmation on its face. But a one-way ANOVA across the three rated groups, run live via the hypothesis-test tool, returns F=1.73, p=0.18 — not significant at conventional thresholds, meaning the observed gaps are compatible with chance given the sample sizes involved (Unsatisfactory has only n=89 carriers with a valid crash rate, producing a very wide standard error). A companion OLS regression (crash rate on Conditional/Unsatisfactory indicators, controlling for fleet size and annual mileage), also run live, confirms the same result more strongly: R² = 0.00026, overall F-test p=0.48, Conditional vs. Satisfactory borderline at p=0.070, Unsatisfactory statistically indistinguishable from Satisfactory (p=0.61). Fleet size and mileage themselves show no measurable effect either. The median crash rate is 0 in every single rating bucket — most carriers, rated or not, show zero recorded crashes in the window FMCSA tracks, so these means are driven by a small right-tailed subset, which is exactly the kind of distribution where a 3-category label struggles to separate anything.

Step 4 — What the published research says (a different, related question: crash risk, not claims)

The federal government's own methodological reviews of FMCSA's carrier-safety scoring back up this null result, though they were studying a related but distinct system. GAO's 2014 review and a 2017 National Academies of Sciences panel (Improving Motor Carrier Safety Measurement) evaluated the newer, more granular Compliance-Safety-Accountability (CSA) Safety Measurement System — seven percentile-ranked BASIC scores (Unsafe Driving, Hours-of-Service, Vehicle Maintenance, Controlled Substances/Alcohol, Hazmat, Driver Fitness, Crash Indicator) that superseded the older three-tier rating as FMCSA's primary internal risk tool. That review found predictive validity for future crash risk varies enormously by BASIC: Crash Indicator and Unsafe Driving correlate strongly with future crashes, but Vehicle Maintenance correlates only around 65%, and Controlled Substances/Alcohol and Hazmat Compliance correlate at just 31–34% — GAO explicitly flagged "weak or negative correlations" for some categories as a methodological concern. Crucially, none of this literature is about cargo/liability claims either — it is about future crash involvement, one input into claims but not the whole of it. I found no peer-reviewed or government study that directly measures the correlation between any FMCSA safety score and private cargo/liability claims frequency. What exists instead is industry/insurance-broker commentary (cited below) asserting informally that underwriters cross-reference CSA BASIC scores — particularly Vehicle Maintenance — against claims history when pricing cargo coverage; this is stated as underwriting practice in marketing content, not as a verified statistical finding, and none of it reports an effect size, a sample, or a methodology.

Bottom line

Answering the question as asked — the official federal safety rating, and cargo/liability claims specifically — the honest answer is: this cannot be tested directly with any public data source, because carrier-level cargo/liability claims are never published by any government agency. Using the best available public proxy (FMCSA's own crash rate), the rating shows no statistically significant relationship even among the small share of carriers it covers, consistent with longstanding GAO/National Academies criticism that FMCSA's safety-scoring systems are weak predictors for several of their component categories. Insurers who price cargo/liability coverage appear, per industry sources, to rely more on the finer-grained CSA BASIC percentiles and each carrier's own private claims history than on the blunt three-tier rating — and even those percentile scores have documented weak spots. A shipper or broker using the S/C/U label alone as a claims-risk signal is relying on a marker that is absent for 98% of carriers and, where present, does not clear a basic significance test against FMCSA's own data.

What This Report Does Not Answer

• Identify/use a measure of cargo or liability claim frequency at carrier grain: No government table (in this corpus or identified via external search) publishes private cargo/liability insurance claims at carrier grain — that data is proprietary to insurers. Used FMCSA's own recordable_crash_rate as the closest public proxy and labeled it explicitly as a proxy, not a claims measure, throughout.

Every query behind this report

5 warehouse calls ran in this session, in order. Each is reproducible against the same snapshot.

SELECT safety_rating, COUNT(*) AS n_carriers,
       COUNT(recordable_crash_rate) FILTER (WHERE recordable_crash_rate NOT IN (999.0,999.99,999.999)) AS n_with_rate,
       AVG(recordable_crash_rate) FILTER (WHERE recordable_crash_rate NOT IN (999.0,999.99,999.999)) AS avg_crash_rate,
       median(recordable_crash_rate) FILTER (WHERE recordable_crash_rate NOT IN (999.0,999.99,999.999)) AS median_crash_rate
FROM transport.fmcsa_carriers
WHERE status_code = 'A'
GROUP BY safety_rating
ORDER BY avg_crash_rate DESC

SELECT safety_rating, recordable_crash_rate
FROM transport.fmcsa_carriers
WHERE status_code = 'A' AND safety_rating IN ('S','C','U')
  AND recordable_crash_rate IS NOT NULL
  AND recordable_crash_rate NOT IN (999.0,999.99,999.999)

SELECT safety_rating,
       COUNT(*) AS n,
       AVG(CASE WHEN recordable_crash_rate > 0 AND recordable_crash_rate NOT IN (999.0,999.99,999.999) THEN 1.0 ELSE 0.0 END) AS share_any_crash,
       AVG(power_units) AS avg_power_units,
       AVG(mcs150_mileage) AS avg_mileage
FROM transport.fmcsa_carriers
WHERE status_code='A' AND recordable_crash_rate IS NOT NULL AND recordable_crash_rate NOT IN (999.0,999.99,999.999)
GROUP BY safety_rating
ORDER BY share_any_crash DESC

SELECT recordable_crash_rate,
       CASE WHEN safety_rating='C' THEN 1 ELSE 0 END AS is_conditional,
       CASE WHEN safety_rating='U' THEN 1 ELSE 0 END AS is_unsatisfactory,
       power_units,
       mcs150_mileage
FROM transport.fmcsa_carriers
WHERE status_code='A' AND safety_rating IN ('S','C','U')
  AND recordable_crash_rate IS NOT NULL AND recordable_crash_rate NOT IN (999.0,999.99,999.999)
  AND power_units IS NOT NULL AND mcs150_mileage IS NOT NULL

SELECT
  COUNT(*) AS total_rated,
  COUNT(*) FILTER (WHERE recordable_crash_rate IS NULL) AS null_crash_rate,
  COUNT(*) FILTER (WHERE recordable_crash_rate IN (999.0,999.99,999.999)) AS sentinel_crash_rate,
  COUNT(*) FILTER (WHERE power_units IS NULL) AS null_power_units,
  COUNT(*) FILTER (WHERE mcs150_mileage IS NULL) AS null_mileage
FROM transport.fmcsa_carriers
WHERE status_code='A' AND safety_rating IN ('S','C','U')