by
lmh » Thu Nov 17, 2016 1:01 pm
I think the confusion is happening because of the way the problem has been posed. In effect, you're doing it the wrong way round at the moment!
You would like to demonstrate that the amount present is zero, but you're worried that you need to measure it +/- a certain percentage, so you're trying to achieve this percentage. The difficulty is that because your sample contains nothing, you are trying to generate the result "0" with an RSD of 2%, which means no error whatsoever (clearly impossible, as you recognised). Therefore quite reasonably you're trying to guess a concentration a bit higher than zero, where you can achieve the RSD of 2%, so you can check that the precision of your measurements is acceptable.
The difficulty is (1) that you're obliged to guess a concentration and hope that you can achieve a target %RSD, and (2) that the target %RSD isn't helpful to the underlying problem. If 1ng of X is going to cause trouble, being knowing you can measure 10ng with 2% RSD doesn't help. What you want to know is whether you can detect 1ng.
This is why it's better to measure the Limit of detection, and simultaneously estimate how much of the problem peak you can actually tolerate in your application before it causes trouble. If the LOD is appropriately less than the estimated amount of X that causes trouble, then you can detect problematic levels of X with adequate reliability. As a 'bonus', if the LOQ is less than or equal to your measured value, then you can promise a certain level of precision to your quantitative answer about how bad the problem is.
The point about the %RSD approach is that you use it on reasonably large peaks, in order to be able to claim that you can measure the amount of a chemical that is present (rather than a chemical that probably isn't, but might be!) with the precision someone has required.
