Testing the equality of regression matrices when some additional data sets are available

Abstract

We consider the problem of testing the equality of several multivariate multiple regression matrices when some data from an unidentified subset of these models are available in addition to the data sets from the identified models. The standard test (T₀) discards the additional data and tests the equality of the regression matrices using only the data sets from the identified models. We propose two test procedures (T₁ and T₂) utilizing all the available data because the additional data may contain some useful information about the regression matrices which are tested for equality. The test procedure T₁ seems to have larger power than T₀ for all parameter values. Further T₂ has the largest power for certain parameter values

Keywords:

• multivariate multiple regression
• power of the test
• non-centrality parameter matrix