TESTS FOR PRECISION AND ACCURACY OF MULTIPLE MEASURING DEVICES

We propose a multivariate linear model for analyzing measurement-device data that follow Grubbs's model. The multivariate model generates the well-known regression (multiple correlation) test for equality of device variances. It also yields a new simultaneous test for equality of variances and biases. The multivariate model readily permits testing of interesting sub-hypotheses involving the variances and biases. We examine the power of these tests for one-dimensional hypotheses compare them to tests proposed by Grubbs, and propose generalizations of Grubbs's tests. We also consider extensions of Grubbs's model that allow similar regression and simultaneous tests to be performed.