Does inspection history predict casualties?

A barge with a history of recorded deficiencies is commonly treated as more likely to be involved in a casualty later. It is a reasonable assumption and it is cheap to act on, because the deficiency record is public and the casualty record is public.

Maury holds both, for the same hulls, over the same years. So the assumption can be checked rather than assumed. Below is that check, in full, including the parts that weaken it.

Lift is the band's rate divided by the rate across all 101,374 vessel-years (27.8%). A lift of 1.00 means the band is indistinguishable from the population. The spread across every band, from no recorded deficiencies to ten or more, is 0.04.

There is no trend to read here. The band with ten or more recorded deficiencies has a casualty rate marginally below the band with none.

A null result is worth nothing on its own — a blunt enough instrument finds nothing anywhere. So the same cohort, the same years and the same calculation, applied to a different prior:

A factor of 2.16, on the same data that showed nothing for deficiencies. The method can separate groups. It does not separate them by deficiency history.

It does not say a vessel with deficiencies is sound, or that inspection is not worth doing. Inspection finds and forces the correction of real faults; a fault corrected is one that cannot be counted here. That alone could account for a flat result, and this archive cannot distinguish it from the alternative.

It does not say anything about any individual vessel. This is a statement about a population of 17,663 hulls. Maury does not score, rate, or grade vessels, and a reader who takes a number from this page to a conclusion about one barge has reproduced the exact error that removed the compliance-status field from this archive before it was ever published.

The cohort is not the fleet. 17,663 vessels, not the 532,794 in the archive: a hull enters only where the record supports a prior year and an outcome year.

The window is 2004–2021, not the archive's full span.

The cohort is casualty-enriched. 8,608 of 17,663 hulls (49%) have at least one recorded casualty, so the 27.8% base rate is a rate within this cohort and is not a casualty rate for the US barge fleet. Do not quote it as one. The comparison between bands is unaffected — every band is drawn from the same population.

Recorded is not the same as occurred. Both sides are what the Coast Guard wrote down. Roughly 41.1% of the archive comes from one district, and a deficiency not written down is not in the left column.

This is a correlation over a window, not a causal claim.

Every hull counted here has a page on this site, with the inspections and deficiencies the count is built from. The archive is rebuilt from Coast Guard exports by a documented runbook, and the queries behind this page run against that archive at build time — the numbers above are computed when the site is generated, not typed in.

Cohort definition. One row per vessel-year. Prior deficiencies counts deficiencies recorded against the hull before that year began; casualty means the hull appears in a Coast Guard incident investigation for that year. A hull enters only where the record supports both a prior year and an outcome year.

Coverage, known gaps and method · Search the 532,794 vessels · Correct a record