Tests of the Recalibration Period of a Drifting Instrument

The use of a drifting instrument requires that an adequately short recalibration period be chosen. After several periods, the calibration data can be used to test the adequacy of the choice. This paper discusses statistical tests of the recalibration period and applies these tests to continuous analyzers for sulfur dioxide. These tests check the sequence of calibrations for evidence of rapid drift-rate changes, which are obscured by random measurement error and by slow drift-rate changes for which the recalibration period is adequate. This paper presents two tests, a test of the second differences of the calibration sequence for normality and a test of the upper part of the spectrum for flatness. Both of these tests are needed because they have different power against different alternatives to constant drift rate. This paper illustrates these tests with two sequences each consisting of about fifty recalibrations of a sulfur dioxide analyzer. Also, the power of the tests and some approximations made in their formulation are investigated by Monte Carlo experiments.