William E. Strawderman, Rutgers University
Improved Estimation of Location Parameters
We consider
simultaneous estimation of the parameters of a multivariate location
family. About half the talk will be devoted to an intuitive discussion
of certain types of so-called shrinkage estimators. The discussion
will focus on; (1) What type of estimators might be expected to
perform better than the classical least squares estimators, (2) When
can large improvements be expected, and (3) How large an improvement
can be expected. The aim will be to indicate that the phenomenon of
improvement over classical procedures via shrinkage is not
particularly dependent on the assumption of an underlying normal
distribution. We give some theoretical results which give support to
this thesis.