(Received 9 October 1992; accepted 28 April 1995)
| ||Format||Pages||Price|| |
|PDF (164K)||6||$25||  ADD TO CART|
Cite this document
The current statistical analysis of interlaboratory test data involves tests of significance for outliers. In this paper we show that the use of tests of significance is faulty and we propose instead an analysis according to a row-linear model. This approach involves an examination of straight lines fitted by least squares for all laboratories against the averages over all laboratories for each level in the two-way table. By this method, a laboratory can be judged both in terms of the position of its response line when compared to other laboratories, as well as in terms of the residuals of the observed measurements from the fitted line.
Guest researcher, Chemical Science and Technology Laboratory, National Institute of Standards and Technology, Gaithersburg, MD
Stock #: JTE11406J