LOQ

[crimelist]

Hi.. i have a question.

My software automatically calculates the value of impurity. The result is 0.048% The LOQ Value is 0.05%!

My result is > or < or = that LOQ?

Thanks

[tom jupille]

Measured values should be reported to the same number of significant figures as the limits. Therefore, 0.048% would be rounded to 0.05% and thus equal to an LOQ of 0.05% (note that 0.054% would also round to 0.05% and be reported the same way).

If the LOQ had been stated as 0.050%, then 0.048% would not be rounded and would be reported as less than LOQ.

[lmh]

... oooh, this brings out the pedant in me. A LOQ is the concentration necessary to guarantee a previously specified accuracy and precision in the result. It's amazing how often it's defined that way, but then used without any indication of what specified accuracy is being guaranteed (and I'm as guilty as everyone else of this)!

But here's the pedantic bit: You'd usually define the precision of an answer by stating its RSD, i.e. saying what standard deviation you expect on the answer. A standard deviation implies that the frequency of getting any particular answer follows a bell-curve around a mean value. Now if you define the LOQ as 0.05 (not 0.050), and interpret this to mean that the results can also only be reported to a precision of two decimal places (and I am sure Tom's right that this is how things are done), then the measurements are no longer anything like a continuous distribution around the mean. In fact if the true standard deviation of the true results is a good bit less than 0.01, you could get a whole string of results that are actually quoted as totally identical: 0.052, 0.048, 0.049, 0.051, 0.049, 0.050 would become 0.05, 0.05, 0.05, 0.05, 0.05, 0.05.... RSD = 0!

I'm not a statistician, but I do feel that this is a bit unsafe. Say our LOQ has been picked so that we know our precision should be RSD = 30%. This means that we have quantified all the little errors that drag a true measurement a little from the mean, and worked out that the random walk of errors from the mean, at the LOQ, gives us a bell-curve with s.d. = 30%. We probably did this using measurements made to quite a lot of decimal places. We now make real measurements that have wandered to the same extent, s.d. = 30%, and we add an extra, final, random jump one way or the other, to the nearest second decimal. This is another added error, and therefore increases the actual standard deviation of the measurements. Therefore we no longer have the guaranteed precision at the LOQ. On top of that, we're now throwing away the numbers we need, if we're to assess what the actual precision is.

What I really mean is that in stating the LOQ, you need to consider the underlying precision that you think you are guaranteeing. If your underlying precision is 50%, then you can have a LOQ of 0.05. If it is 10%, then a LOQ of 0.05 would be entirely wrong because given a sample that truly contained 0.055, every measurement would have an error of at least 10% before considering any analytical variability. Otherwise it's like saying you are measuring humans' heights to the nearest foot 5 feet or 6 feet, with a precision of 3 inches... so what does that mean about someone who's 5'6"?

Right, that's a completely irrelevant pedantic rant.

[Peter Apps]

lmh, if you think that you are a pedant, take a look at the Guide to the Expression of Uncertainty in Measurement; http://www.bipm.org/en/publications/guides/gum.html - those guys are real professional pedants.

Somewhere in there you will find that you are right about rounding and readability errors - and uncertainty budgets need to include an estimate for how much uncertainty they contribute.

Peter

[lmh]

Strewth! I'm outclassed completely! If I want to finish that document before retirement age, I'd better develop a faster style of reading. It is great to know that these issues are being dealt with by people with brains the size of a smallish planet. Thanks for the link, though, it is actually really interesting.

[Peter Apps]

Strewth! I'm outclassed completely! If I want to finish that document before retirement age, I'd better develop a faster style of reading. It is great to know that these issues are being dealt with by people with brains the size of a smallish planet. Thanks for the link, though, it is actually really interesting.

I worked in metrology for a while. I knew I was entering an alternative reality when the marks for my entrance exam were given to three significant figures !

The GUM does make it look complicated and cumbersome, but it's not so bad once you get used to it. I would really like to see chemists using proper units all the time instead of %, ppm etc, which can mean different things in different contexts. Also the ridiculous estimation of LOD from a peak that is 3 x baseline noise that appears in nearly every paper I get to referee just has to go.

Peter

[crimelist]

Ok bro.. so my result will be 0.05% (and stop).


because on my analysis request I must indicate the value obtained or if less than LOQ

[lmh]

Yup, it's an alternative reality, especially for those of us who thought a guide for metrology was a girl who likes the weather.

It's a pity, though, that other walks of life take so little care about propagation of errors, precision, and confidence intervals, especially in contexts where the result matters intensely. For example, I can't see any point in the forensic analyst worrying about detection limits and statistical quality of his results when the court that finally makes the decision about guilt can't be bothered to define what it means by "beyond reasonable doubt", as decided by 12 good men and true?

[Peter Apps]

Yup, it's an alternative reality, especially for those of us who thought a guide for metrology was a girl who likes the weather.

It's a pity, though, that other walks of life take so little care about propagation of errors, precision, and confidence intervals, especially in contexts where the result matters intensely. For example, I can't see any point in the forensic analyst worrying about detection limits and statistical quality of his results when the court that finally makes the decision about guilt can't be bothered to define what it means by "beyond reasonable doubt", as decided by 12 good men and true?

It's not all bad - speed cameras for traffic offences have to be calibrated against standards that are ultimately traceable to the speed of light in a vacuum. Similarly with blood alcohol, reference materials trace to the kilogram and mole. These don't go to jury trial though.

Peter

[Consumer Products Guy]

O.J. did it.

[mckrause]

No, your value will be 0.048%, reported as less than the LOQ (so, <LOQ (0.048%))

[crimelist]

No, your value will be 0.048%, reported as less than the LOQ (so, <LOQ (0.048%))

I don't understand: my LOD is 0.05& not 0.050%

Mi value is 0.048 (aprox 0.05)

Then... 0.05% or 0.05(<LOD) ??

Right?

[Peter Apps]

No, your value will be 0.048%, reported as less than the LOQ (so, <LOQ (0.048%))

I don't understand: my LOD is 0.05& not 0.050%

Mi value is 0.048 (aprox 0.05)

Then... 0.05% or 0.05(<LOD) ??

Right?

Wrong. The very first answer from Tom is correct. Ignore everything and everybody that disagrees with it.

Peter

[crimelist]

No, your value will be 0.048%, reported as less than the LOQ (so, <LOQ (0.048%))

I don't understand: my LOD is 0.05& not 0.050%

Mi value is 0.048 (aprox 0.05)

Then... 0.05% or 0.05(<LOD) ??

Right?

Wrong. The very first answer from Tom is correct. Ignore everything and everybody that disagrees with it.

Peter

OK!!!

So... my final result is 0.048%

And I report?

1) 0.05%
2) 0.05% <LOD

LOD=0.05%
Limit <1.00%

[Peter Apps]

I don't understand: my LOD is 0.05& not 0.050%

Mi value is 0.048 (aprox 0.05)

Then... 0.05% or 0.05(<LOD) ??

Right?

Wrong. The very first answer from Tom is correct. Ignore everything and everybody that disagrees with it.

Peter

OK!!!

So... my final result is 0.048%

And I report?

1) 0.05%
2) 0.05% <LOD

LOD=0.05%
Limit <1.00%

Refer to my previous answer.

Peter