Evaluación estadística de Materiales de Referencia Geoquímica del Centre de Recherches Pétrographiques et Géochimiques (Francia) aplicando un esquema de detección y eliminación de valores desviados
Keywords: geochemical reference materials, outlier statistical tests, analytical geochemistry
AbstractThe present study shows the results of a statistical evaluation to establish reliable composition values (central tendency and dispersion parameters) for eight Geochemical Reference Materials (GRM) available from Centre de Recherches Pétrographiques et Géochimiques (CRPG; Nancy, France): bauxite BX-N, diorite DR-N, sienite DT-N, potash feldspar FK-N, granite GS-N, microgabbro PM-S, serpentine UB-N, and dolerite WS-E. The methodology consisted in the preparation of updated composition databases for each GRM and their statistical data processing. A statistical filter was applied in each element database with ni >5, consisting of simple and multiple deviation/spread, Grubbs, Dixon, and moments of high order variant tests. The objective of this procedure was to detect and to eliminate discordant outlier values in population samples that according to the theory, tend to be normally distributed. Concentration values obtained in this evaluation were compared with working values proposed by CRPG, which have been generated by applying a central tendency indicator approach (robust method). An ANOVA statistical evaluation (at 99% confidence level) indicated significant variance differences between the literature working values and those generated in this work in 39% (BX-N) to 70% (UB-N) of studied cases. Values of %Rsd obtained in this work were generally lower in comparison with those reported by CRPG. Relative standard deviation percentage difference (∆%Rsd) varied from 25 to 100%, although in some cases was higher, reaching until ~1400% (Sb for BX-N). According to ANOVA evaluation, no significant differences were observed between microgabbro PM-S average values proposed in this work and those suggested in the literature. For the rest of GRM, differences in composition were observed from 1.9% (GS-N) to 18.4% (BX-N) of evaluated cases. In general, considering the eight GRM, the magnitude of deviation of average value (∆x) was <18.6% for 89% of evaluated elements, whereas that differences for rest of cases were larger. Extreme cases with a high contrast (∆x >100) were detected in: (a) Na2O, Cs, and Rb for BX-N; (b) Sn for DR-N; (c) MgO, CaO, and Cl for DT-N; (d) Fe2O3, Th, and F for FK-N; and (e) Ge for UB-N. The information generated in this work could be used for several purposes that include better calibration models for analytical methods, quality control of new geochemical data, and, as a consequence, improved geochemical models for geological processes.