Assessment of the Precision and
Bias of an On-Line Analyzer
Using a Single Reference Instrument
Fred Lombard
I consider estimation of the precision
of an on-line gauge. Typically, this type of estimation involves independent
results from two or more reference instruments (laboratories). The properties
of the estimator are then independent of the variability of the product
(quality variation). The use of more than one reference instrument, however,
entails significant additional costs.
The Grubbs estimator which is based
on results from a gauge and a single reference instrument (laboratory)
has the unpleasant property that its standard error, is heavily dependent
on the quality variation. I propose a new estimator of which the standard
error is much less dependent on the quality variation. In fact, the
standard error can be less than that of the Grubbs estimator based on the
use of two reference instruments.
In order to function optimally,
the new method requires some a priori knowledge regarding the true quality
variance and gauge measurement error variance. A theoretical analysis of
the properties of the new method is quite complicated and is, as yet, incomplete.
In particular, a simple formula for the standard error of the estimator
is lacking. I propose (for the time being) that the bootstrap be used to
estimate the standard error in practical applications. The efficacy of
the new method is illustrated by theoretical calculations and by Monte
Carlo simulation results.
FRED LOMBARD
Department of Statistic
Rand Afrikaans University
P.O. Box 524
Auckland Park 2006, South Africa
fred@rau3.rau.ac.za
A Comparison of Statistical Techniques
for Handling Raw Material Variation
in the Food Industry
Tormod Naes, Ingunn Berget, and
Bjorn-Helge Mevik
In many industries, the raw materials used in production vary according to for instance biological variation. Such variation can have a strong influence on the quality of the manufactured product if not handled properly. In this talk a number of statistical approaches to the problem will be discussed. Among these are continuous updating of processing conditions, sorting of raw material into homogeneous categories and setting of process conditions that are robust to raw material variation. The different possibilities will be illustrated by examples from food research.
TORMOD NAES
Department of Quality Analysis
MATFORSK
Olsovegen 1
1430 Aas, Norway
tormod@sn.no
tormod.nas@matforsk.no
tormod.naes@matforsk.no
Recent Developments in Receptor
Modelling
Clifford H. Spiegelman, Eun Sug
Park, and Byron Gajewski
Receptor models are used to estimate pollution recipes, pollution amounts, and location of polluters from toxics and particulate data, as well as from meteorological data. We will present our findings for several sites in Texas as well as present our current methodology. We will also review work by other contributors.
CLIFFORD H. SPIEGELMAN
Department of Statistics
Texas A&M University
College Station, TX 77843-3143,
USA
cliff@stat.tamu.edu
