Insensitivity Regions and Outliers in Mixed Models with Constraints

Let the mixed linear regression model when mean value parameters satisfy some linear constraints be under consideration. To detect rough errors or mistakes (outliers) in observations needs a knowledge of a covariance matrix of an observation vector. If unknown variance components occur in it, some problems arise how to recognize, whether estimates or approximations of them can be used instead of their true values. Insensitivity approach is presented in the paper. The aim of the paper is to give a procedure how to detect outliers and to determine proper insensitivity regions for testing hypotheses.