Queensland publishes a safety rate for every mine. For most of them it measures size

The dataset is called Mine site safety performance, it is published by Resources Safety and Health Queensland on the Queensland open data portal under a Creative Commons licence, and it covers the seven years from 2018-19 to 2024-25 across five workbooks: underground coal, surface coal, underground mineral mines, surface mineral mines, and quarries.

Each row is a named site in a named year, with the number of serious accidents, the number of high potential incidents, the hours worked, and two frequency rates. Publishing that at site level is unusual and it is the right instinct. The difficulty is what happens when the rate column is read as a ranking.

What the rates actually are

The workbooks do not define their own terms, and the regulator's own website was not reachable to us, so we took the definitions from the legislation instead, which is the better authority in any case. Under the Mining and Quarrying Safety and Health Act 1999, a serious accident is an accident at a mine that causes the death of a person or a person to be admitted to a hospital as an inpatient for treatment for the injury. A high potential incident is an event, or a series of events, that causes or has the potential to cause a significant adverse effect on the safety or health of a person. The Coal Mining Safety and Health Act 1999 carries the same two definitions for coal mines.

The frequency rates are not defined in either Act, because they are a calculation rather than a legal term. So we derived the formula from the data and checked it: the serious accident frequency rate is simply serious accidents divided by million hours worked, and the high potential rate is the same form. Tested against every row of the 2024-25 underground mineral sheet, the formula reproduced the published figure in 14 cases out of 14.

Why that denominator swallows the story

A rate per million hours is a sensible measure for a large operation. Most of the sites in this dataset are nowhere near a million hours.

What a single serious accident becomes at different site sizes, using the published formula. The hours figures are the actual distribution of the 1,637 site-years in the mineral and quarry workbooks.

Eighty-nine per cent of the site-years in these workbooks fall under 200,000 hours. The median is 15,500, which is roughly eight full-time people. At that scale the rate column is not really measuring incident frequency at all. It is measuring how few hours the site worked, with a single event as the trigger.

This is not a flaw anyone introduced deliberately, and it is not unique to Queensland. It is what happens to any per-unit rate when the denominator gets small, and it is the reason hospitals, schools and police divisions all publish rates with population floors attached. The point is that this dataset does not carry one, and it is published in a form that invites exactly the comparison it cannot support.

What we are deliberately not printing

We could list the sites with the highest published frequency rates in Queensland. We are not going to, and the reason is the arithmetic above.

Every site at the top of that list is a small one where a single incident met the denominator. A table of those names would read as a league table of the state's most dangerous operations, and it would not be one. It would be a list of small sites that had one bad day, which is a different thing and a materially unfair thing to publish about identifiable businesses. The one site we have named, Weipa, sits at the favourable end and is named only to anchor the scale.

That is also why the finding here is about the dataset rather than about any operator. The numbers are real, the regulator published them in good faith, and the transparency is a credit. The rates simply cannot be compared between sites of different sizes, and nothing in the workbooks says so.

Our view, labelled as such

Publishing per-site data is the right call and we would rather have this dataset than not. Two small changes would make it far harder to misread, and neither requires collecting anything new.

The first is a note in the workbook stating what the rates are and how they are calculated, which currently exists nowhere in the file. The second is suppressing or flagging the rate where hours worked fall below a threshold, in the way statistical agencies routinely suppress small-cell figures. Where the underlying counts are one or zero, the counts themselves are the honest presentation and the derived rate adds nothing but the illusion of comparability.

Until then, the practical advice for anyone using this data, whether a journalist, a procurement team or a worker checking a prospective employer, is to read the serious accident and high potential incident counts and the hours worked, and to treat the two rate columns as unusable for comparison between sites.

Work in mining or resources regulation, or know of published guidance on these rates that we could not reach? Tell us and we will read it and log the outcome on this page.