3.5. Reproducibility
When considering instrumental long-term reproducibility of the analysis and in order to monitor instrumental problems and to check for accuracy during each run, we analyze one of several working standards during each run. The samples selected as working standards cover much of the typical range of oxide concentrations encountered in the analysis of most common silicate rocks, and provide a framework for evaluating long-term repeatability and for consideration of how well individual oxides can be measured. First of all, relative standard deviations vary with concentration. Overall, Si, A1, Fe, and Ca have relative standard deviations better than ± 5%, suggesting that in most cases these four elements can be measured exceedingly well. Judging from the results, for the other oxides, ± 5% precision can be obtained on MgO and TiO2 above ~ 3 wt%, and on K2O above ~ 4 wt%. The precision for Na2O appears to be inherently lower, but ± 5% (relative) at 3 wt% concentration is quite acceptable. Results for MnO and P2O5 are also acceptable at the low concentration levels involved. While evaluating the trace element concentrations the reproducibility could be achieved with ~ 2–3% bias values. However, for Ba and few exceptions with Hf, La and Ta having lower concentrations could not be detected. Further, the repeatability of XRF analysis depends on counting statistics, short term error (caused either by imprecise re-setting of instrument parameters or sample presentation), instrumental drift and sample preparation errors. In our spectrometer, sample presentation imprecision is minimum because an automatic sampler changes discs and the same sample holder was always used.