Year
1981
Abstract
Hakkila, et al.l and Shipley2 have couched the inspector's verification problem of testing the hypothesis of diversion and falsification in the framework of the general linear statistical model.3 Three specific models are investigated: (1) Diversion and k distinct falsifications (k-i-1 degrees of freedom); (2) Diversion and the accumulation of all the falsifications (2 degrees of freedom); (3) Diversion only (1 degree of freedom). A test statistic has been derived for models(1) and (2) by the likelihood ratio procedure under the hypothesis of zero falsification and zero diversion versus a one-sided alternative of positive diversion, positive falsification, or both. An analogous test has been developed for model (3) for diversion only. A detailed discussion of this development is given by Shipley.2 Utilizing his notation, the test variables called Inspector's £ufficient Statistics - ISSg,ISS^, and ISS2 - are used for testing the above three models (1), (2), and (3) respectively. Note that ISS2 corresponds to the wellknown MUF-D statistic,which is currently in vogue. These two tests, for purposes of this discussion, are equivalent. The objectives of this paper are to review the ISS procedures, develop optimal critical regions for ISS tests,and compare detection probability (power) curves for models (1), (2), and(3). Optimal critical regions depend upon values obtained from one—sided chi-squared tests with more than one degree of freedom. Details of such a procedure are not to be found in the literature, and this strange neglect was one of the motivating factors in this study.