A COMPARISON OF TWO TESTS IN MULTIVARIATE MODELS

In multivariate models the test usually used for a linear hypothesis is the Wilks lambda test. There does not seem to be a need for any alternative. However, if the covariance matrix is parameterized by a few parameters only, and the parameters can be estimated, the problem arises whether a "plug-in" test, which approximates the test in a situation where the covariance matrix is totally known, is more suitable. A partial answer to this question is given in the paper. The sensitivity approach is used in order to check how this principle works in a situation where a plug-in procedure can be compared with another, already established, procedure.