Nonsensitiveness regions for threshold ellipsoids. (Q558567)

Threshold region in a mixed linear statistical model is defined as a set in parametric space. On the boundary of this set the power function of the test of a linear hypothesis on the model parameters attains the prescribed value (near to 1). In the case of normality the shape of this set is an ellipsoid. Position and a size of the threshold ellipsoid depends on the values of the covariance matrix parameters. Nonsensitiveness region is such a neighbourhood of a given covariance matrix vector parameter that for points of its interior the value of the power function is larger than the prescribed value minus sufficient small number. The boundary of the nonsensitiveness region is determined in the paper.