francis - 17-9-2011 at 23:53
I measured Zn and Cu in a water samples by flame AAS.
Two water samples were treated with alum and polyacrylamide flocculants, and two blank samples (MilliQ water) were treated the same way.
After preparing standards and plotting calibrations curves, the concentrations were estimated from the equation of the line.
The treated blank was 0.005ppm Cu and the confidence interval at 95% was calculated to be 2ppm. Zinc gave similar results for the blank and the
sample.
So I would report that as 0 + / 2ppm. But the detection limit of the instrument is 0.001ppm.
How can the concentration be negative? Do I report this simply as 'below the detectable limit'?
bbartlog - 18-9-2011 at 04:58
Uncertainties can be asymmetrical (so rather than +/- 2ppm, +2ppm/-0.005ppm). But I would suggest reviewing your calculation of the confidence
interval. If you measure something at 0.005ppm (i.e. 5ppb) using an instrument that can detect down to 1ppb, it seems surprising that the CI would be
relatively gigantic like this. It's like saying you estimate that approximately fifty people passed by, but you can only be 95% sure it wasn't more
than twenty thousand.
blogfast25 - 18-9-2011 at 05:06
Definitely review your calculation. This kind of question is better posted in another part of the forum though...
DDTea - 18-9-2011 at 09:42
If the concentration is "negative," then it is below the detection limit. The analyte signal is not fully resolved from the background signal, so it
cannot be accurately quantified. These results are typically reported as "less than" values, e.g., "< 2 ppm" if the blank concentration is 2 ppm.
Instrument manufacturers playing up their instruments' detection limits are a lot like digital camera manufacturers playing up their cameras' pixels:
it's never that simple. Detection limits vary from element to element, with operating conditions, and with experimental protocol. So your best bet,
when performing any kind of method development activities, is to do the statistical analysis yourself and to determine your own detection limits. It
takes a bit of time, but it's not difficult. Also, it's a good idea to run samples in duplicate.
[Edited on 9-18-11 by DDTea]