ESTIMATION OF THE MAXIMUM INFLUENCE OF SAMPLE PREPERATION ON LONG TERM ANALYTICAL PRECISION, PETER HOOPER GEOANALYTICAL LAB

A variety of people, from experienced technicians to first time visitors, prepare samples in Washington State University’s Peter Hooper GeoAnalytical Lab. It is therefore important to test the consistency of our preparation methodology to assess the impact of sample preparation bias.

To evaluate this potential variability, in-house basaltic (BGR) and granitic (GGP) standards were chipped, split into ~40g aliquots with a rotary splitter, and a single technician (ST) prepared 48 of each of the check standards for separate and combined analysis by X-ray Fluorescence (XRF) and Inductively Coupled Mass Spectrometry (ICP-MS). 47 more aliquots of the granite and basalt standards were then inserted into the sample stream as blind unknowns (BU) for a period of just over 11 months, these results were compared to ST prepared standards. We consider the results for GGP to be representative of a worst-case scenario as it is coarse crystalline, and therefore the chosen 40g aliquot size is considerably smaller than ideal, increasing the likelihood of a heterogeneous aliquot due to sampling bias.

For the ST standards, average values were determined, and relative standard deviations (RSD) for all elements in both BGR and GGP were calculated to be <3% for all Major elements, as well as Ba, Ga, Rb, Sc, Sr, Y, and Zr. RSDs were <8% for Cu, Nb, Ni, Pb, V, Zn; and <13% for Ce, La, and Nd. For both standards, Cr and Th approach the limits of reliable quantification via XRF (<10 ppm), which resulted in anomalously high RSDs of 24% (Cr) and 12% (Th).

For the BU samples prepared by multiple users over 11 months, RSDs for the majority of elements were ≤1% higher than that of the ST prepared standards, with the exception of Ce, Nd, and Zr (≤1.5%), Cr (4%), Ni (6%), and Zn (7%).

Based on these results, we conclude that technician variability in sample preparation bias is a very small, though measurable, portion of overall “sample to spreadsheet” error, and should be evaluated when reporting precision for any analytical method.