Hi Sinned
Sorry for the late reply, I had to dig in my files for an example.
I would not have assumed that the measurement error of the entire system is applicable to each element. The OES has separate element "windows" which presumably are tailored to the individual elements.
Some elements are likely to be present in larger amounts (eg Cr, Ni) and some in smaller amounts (eg P, S).
The instrument is calibrated by measuring on several standard reference samples, measuring the actual number of photons obtained which can be attributed to each element (which in some cases can overlap), and applying a best fit regression for each element individually.
Based on that, even though I've never tried to quantify the individual or aggregate errors, I don't see why the precision would be the same for Ni as for S. Also, my Spectro even had two different calibrations for Ni, one for below 5 % and one for above 5 %. I presume that this is because the concentration is actually non-linear with counts, but the best fit was done with linear regression. More modern instruments probably can do polynomial or exponential fits as appropriate.
The best I can do is show you an example of what happened when I repeatedly sparked the same sample. I always sparked reference samples just before running the actual sample. A couple of things that are coming back to me now....the C-content would consistently drop with each "burn" and I never got to the bottom of it. Also the Si-calibration was remarkably stable.
I hope this helps...