STATISTICAL ASSESSMENT OF ONLINE ANALYZERS

The use of on-line analyzers which are capable of producing a quality index for raw material in real time provides an economically viable al ternative to traditional laboratory assays. Such analyzers are, in pri nciple, capable of making quality determinations to a substantially hi gher order of precision. In this paper some statistical methods useful in assessing precision and bias of an analyzer relative to one or two reference instruments are considered. In particular, the roles of pre paration variance and of scale bias between reference instruments are highlighted. A large preparation variance and/or substantial scale bia s between reference instruments effectively precludes proper evaluatio n of an analyzer. Precision estimates proposed by Grubbs [F.E. Grubbs, J. Am. Stat. Assoc; 43 (1948) 243-264] and by Hahn and Nelson [G.J. H ahn, W. Nelson, Technometrics 12 (1970) 95-102] in an analogous contex t are shown to be numerically identical. A test for scale bias between analyzer and reference instruments is also proposed. (C) 1997 Elsevie r Science B.V.