Derivation of recommended values from compilations: an opinion

Abstract

In the July 1989 special issue of Geostandards Newsletter, Govindaraju indicated that a minimum of 40 analytical results, arrived at by at least four independent methods, are needed to derive working values or recommended concentrations, and that the working values are improved when derived from an even larger data set. To make the needed data more readily available, he suggests the desirability of annual updates for geochemical reference sample databases. It is the purpose of this opinion piece to comment on these issues. First, the need for 40 or more analytical results to derive working values is a direct contradiction to the National Institute of Standards and Technology (NIST) approach in producing standard reference materials (SRMs). NIST certifies the elemental concentrations in SRMs, based on agreement between results from a minimum of three independent methods of analysis. Recently, the United States Geological Survey adopted a similar criterion in deriving recommended values for the shale reference sample, SDO-1 (1989). Consensus values published by Gladney et al. (1987) do not differ significantly from original certified values, despite being based on as many as 50 independent results. Admittedly, the Gladney et al. (1987) compilation provides information on many elements not covered in the earlier certification program, and is, in this sense an improvement. However, the fact that consensus values for 19 elements in NBS coal SRM 1632, for example, are based on five or fewer literature values, and an additional 12 values are based on six to ten results, is itself a contradiction to the Govindaraju statement that 40 or more analyses are required for appropriate derivation of working values. It indicates the futility of annual updates to improve the quality of recommended concentrations, since even these few results were obtained only over twelve years. Comparison of Flanagan's (1976 and 1984) original compilations for USGS geochemical reference samples with those published later by Gladney and Roelandts (1988a, b) similarly leads to the conclusions that 1) good working values can be derived from fewer than 40 laboratory results, providing all are obtained under stringent quality control, and 2) annual\ updates of compilations are likely to be redundant for all but a few elements.