MULTIVARIATE REGRESSION WITH A SINGULAR COVARIANCE MATRIX

                  Dietrich von Rosen

       Matematisk Statistik, Uppsala Universitet

Classical multivariate linear models analysis is extended in 
order to cover high-dimensional data. The basic assumption is
that the dispersion matrix is singular. Consequences of this
assumption are studied. Among others the Growth Curve model
with a singular dispersion matrix is treated. Both tests and
estimators are given. We will show how one can analyse data.
Several new aspects appear. For example, sometimes we have to
preprocess the data by performing certain projections before
using the data in the models.