Evaluation of GSJ intrusive rocks JG1, JG2, JG3, JG1a, and JGb1 by an objective outlier rejection statistical procedure

  • Mirna Guevara Centro de Investigación en Energía, UNAM, Priv. Xochicalco S/No., Col. Centro, Apartado Postal 34, Temixco, Mor. 62580, Mexico.
  • Surendra P. Verma Centro de Investigación en Energía, UNAM, Priv. Xochicalco S/No., Col. Centro, Apartado Postal 34, Temixco, Mor. 62580, Mexico.
  • Fernando Velasco-Tapia Centro de Investigación en Energía, UNAM, Priv. Xochicalco S/No., Col. Centro, Apartado Postal 34, Temixco, Mor. 62580, Mexico. Facultad de Ciencias de la Tierra, UANL, Hacienda de Guadalupe, Carretera Linares- Cerro Prieto Km 8, Apartado Postal 104, Linares, N.L. 67700, Mexico.
Keywords: outlier, reference material, statistical tests, univariate sample, analytical geochemistry.

Abstract

An objective procedure involving fourteen statistical tests (a total of thirty-four variants) for detection and rejection of outliers in a univariate sample was applied to five geochemical Reference Material (RM) databases of intrusive rocks from the Geological Survey of Japan (GSJ). All available concentration data for JG1, JG1a, and JG3 granodiorites, JG2 granite, and JGb1 gabbro were first grouped in eight general analytical methods. These groups were tested for systematic differences using analysis of variance (ANOVA). After this process, the databases were evaluated by means of statistical tests for detection of possible outliers at a strict confidence level of 99%, minimizing the danger of rejecting a valid observation as an outlier. New concentration mean values and other statistical parameters were computed from final normal distributions. Although 23% of mean concentration values were practically identical to those reported in earlier literature, 67% differ by about 1% to 20%. The present statistical scheme provides a better perspective to evaluate existing geochemical databases than probably erroneous two-standard deviation method involving prior subjective judgments used by the U.S. and Japanese scientists. The present mean values with generally lower %RSD may contribute to a better estimation of precision, accuracy, and sensitivity of routine analysis for intrusive rock samples.

Published
2018-07-04
Section
Regular Papers