LOQ

[mckrause]

Tom's answer is correct except at the LOQ. Since the measured value is less than the LOQ the reported value should be <0.05% or <LOQ, and then if a numerical value is included (as in J flagged data) the reported value would be 0.048%, since it is less than the LOQ.

In any case the value is less than the LOQ. If the reporting limit is the LOQ then you MUST report the value as less than the LOQ, not a rounded value equivalent to the LOQ. Some labs (almost all environmental labs) report a numeric value that is associated with the reported <LOQ. These values are flagged (thus my reference to a "J" flag) as being less than the LOQ and therefore being less reliable than a value reported above the LOQ.

My original post showed <0.050%. I did not intend to imply additional precision.

[Peter Apps]

Unless it is standard procedure to apply the rounding at different stages for validations/ calibrations and samples, then Tom's answer is correct. Rounding is part of the measurement (and contributes uncertainty to the result). The measurement result is the signal times calibration then rounding, so 0,05%. This is equal to the LOQ, so is reported as such.

Peter

[crimelist]

You know tell me the references to what was said?

[Peter Apps]

The Guide to Uncertainty in Measurement is the complete and final answer to all questions about measurement results and how to express them - Google "BIPM Uncertainty" etc.

Peter

[crimelist]

Thanks Peter.
I found this:
http://www.bipm.org/utils/common/docume ... 2008_E.pdf

but I have not found the part that interests me.

[Peter Apps]

IUPAC probably has some guidance also.

I am not going to look up chapter and verse for you - you do your own reading.

Peter

[lmh]

(Sorry about this response... I'm not trying to be nasty!) I still maintain that you're solving the wrong problem.

The whole point of a LOQ is that it guarantees a particular precision. If the LOQ is quoted in such a way that the precision it should guarantee is actually impossible (i.e. 0.05 instead of 0.050) then it's self-contradicting and meaningless.

Put simply: You cannot quantify, at the LOQ, with the precision guaranteed by the LOQ, and that's plain daft.

It is utterly essential to quote a LOQ with enough decimals such that the question you're trying to answer ("What should I do, if the answer I get is a tiny, tiny bit below the LOQ but rounds to the same as the LOQ?") becomes irrelevant. If the difference between the rounded value and the un-rounded, measured value is large enough to bother you, then the quoted precision at the LOQ (and by extension on all measurements) is inadequate for the task in hand.

I have to agree with Tom's original answer, and everything that Peter has added. The GUM files are frightening though!

Being asked to use a LOQ of 0.05 (no decimals) is as useless as being told you're allowed to eat bananas that are orange and round.

[Peter Apps]

I agree with lmh, you are worrying about the wrong thing - or rather worrying when there is no need to. Your measurement result is 0.05 (whatever units), not 0.048. O.048 is not the measurement result, it is just a step in the calculation chain.

What strikes me as really strange is that you are stressing over a 0.002 rounding when your action limit is 1% - your LOQ of 0.05 (presumably %) is ten times better than it needs to be.

Peter

[crimelist]

I agree with lmh, you are worrying about the wrong thing - or rather worrying when there is no need to. Your measurement result is 0.05 (whatever units), not 0.048. O.048 is not the measurement result, it is just a step in the calculation chain.

What strikes me as really strange is that you are stressing over a 0.002 rounding when your action limit is 1% - your LOQ of 0.05 (presumably %) is ten times better than it needs to be.

Peter

Definitely for you it is a wrong problem. But for me it is very important because it creates confusion for operators (analysts)
My analytical result is 0.0482384592385235 etc.etc.

But my LOQ is 0.05%.

So back to the question:

my ultimate value will be:

0.04 less than LOQ or 0.05 equal LOQ?


This is important because if the single impurity value (this is the case) is more LOQ I have to sum it in total impurities otherwise not be counted.

[Peter Apps]

For the third or fourth (and final) time;

the numbers that you are citing are NOT the analytical result. The analytical result is what you get AFTER you have rounded the numbers that the software spits out. The numbers that the software spits out are simply part of the calculation of the result, with exactly the same status as the area of the analyte peak.

You cannot solve the problem of your samples being out of spec by fiddling with rounding.There is no way that you can round 0.048... down to 0.4, no matter how much you want to pretend that the sample contains less impurity than it really does.

If your operators are confused then they need better training, but the easiest way to solve the problem is to require the number produced by the software to be rounded to 2 decimals so that nobody in your lab gets confused by lots of decimals.

Peter

[crimelist]

For the third or fourth (and final) time;

the numbers that you are citing are NOT the analytical result. The analytical result is what you get AFTER you have rounded the numbers that the software spits out. The numbers that the software spits out are simply part of the calculation of the result, with exactly the same status as the area of the analyte peak.

You cannot solve the problem of your samples being out of spec by fiddling with rounding.There is no way that you can round 0.048... down to 0.4, no matter how much you want to pretend that the sample contains less impurity than it really does.

If your operators are confused then they need better training, but the easiest way to solve the problem is to require the number produced by the software to be rounded to 2 decimals so that nobody in your lab gets confused by lots of decimals.

Peter

Peter you're free to stop responding, but what you say makes no sense.

the analytical result is "also" the one generated by the system

I'm not trying to take shortcuts to indent the value of my impurities. Otherwise arrotonderei 0048 to zero

[Peter Apps]

You like playing with numbers, so sum the post count of the posters who go for 0.05 % as the result you should report, and divide it by the sum of the post count of those who go for <LOQ (you and mckrause) - then take the trouble to read the literature that you have been referred to and come back to tell us whether the authoritative and definitive guides to measurement agree with the majority or the minority.

Peter

[crimelist]

You like playing with numbers, so sum the post count of the posters who go for 0.05 % as the result you should report, and divide it by the sum of the post count of those who go for <LOQ (you and mckrause) - then take the trouble to read the literature that you have been referred to and come back to tell us whether the authoritative and definitive guides to measurement agree with the majority or the minority.

Peter

Ok I will study literature.

But now for me (and Excel) 0.048 is below LOQ (0.05)

[Peter Apps]

Good luck when the auditors came around.

Peter

[lmh]

One last attempt as I've been thinking about this.

The LOQ is the concentration corresponding to an RSD on your measurements of 10%.

For your method, the LOQ is 0.05, implying a standard deviation of 0.005. You measured a value of 0.048. This differs from the LOQ by 0.002, much smaller than the standard deviation of the measurements. Therefore Tom's original post is not only correct from a regulatory perspective, it is correct from a statistical/measurement perspective: There is no significant difference between your measurement and the LOQ, and therefore, to the best estimate you can make, measurement = LOQ.

If you state that your measurement is less than LOQ, you're doing exactly the same as anyone who claims an increase in X based on triplicate measurements where actually there is no statistically significant change in X were they to carry out a t-test.

In the end it's your call, but you should be aware of what has been done, if anyone asks you to explain the result.