Practical limits on reporting parameter values

I’d be interested to hear views on the use of a limit on reporting parameter values, particularly where there are a large % of left censored results. My example is the reporting of RCS exposures in a SEG where we have 6 samples, 2 being results <LOD (0.003), the remaining results are between LOD and 0.006. The calculated AM comes out at being 0.0043, the EF is 1.1e-06% [0-0.37]. I report these in a table with all the other SEGs and whilst I agree that the scientific notation and CI is the correct expression of data, for the reader of the report who may be the HSE manager or mine manager it is lost on them. In these situations I have simply reported the EF as <0.001%.

Hello Peter,

Thanks for this first ever post ! I took the liberty of moving the topic to the “strats and stats” category.

Just a clarification, what you call EF is the ratio of the AM to the OEL right ?, expressed in %.

Then your result is that the AM is estimated at 0.000001% of the OEL, but could be as high as 0.37% of the OEL. Do I interpret your results correctly ?

I should have been clearer, EF is exceedance fraction. so it is very small in this example. Typically we may report the results of up to 30 SEGS in an annual report for large sites. The results are listed in a table and Exceedance fraction is one of the parameter values.

Oh sorry, I see now. My bad
You report EF<0.001% instead of point estimate 1.1e-06% with CI : [0-0.37%]

I agree in principle, but would use the UCL instead of the point estimate, here EF < 0.37% (so we are sure).

One possibility (e.g. for expostats) would be to report estimate + CI only if the UCL is > 1% (arbitrary choise). If UCL is <1%, we can say with confidence EF is very small compared to the 5% threshold. If UCL is >1% it might be worthwhile to provide the full info.

something like that ?

I did consider using the UCL estimate of the EF, but would then need to report it for all of the SEGS to maintain consistency. Clients have a hard time understanding simple statistical measures, EF included, the UCL of EF could be a stretch for some.

I do like your suggestion of an arbitrary threshold for UCL…

1 Like