Student Seminar Series – December 6, 2007
University of Minnesota
School of Statistics
College
of Liberal Arts

 

Consistent Model Selection in High Dimensional but Low-Sample Data



Liliana Forzani


Thursday, December 6, 2007
10:00 AM,
300 Ford Hall
Minneapolis, East Bank Campus

Refreshments at 9:30 AM
300 Ford Hall

Abstract



 
 

Using as a starting point the hypothesis of normality of X|(Y=y), we developed a methodology for dimension reduction for the forward regression of Y|X under the sufficient dimension reduction paradigm.  This methodology includes finding the dimension of the central subspace, maximum likelihood estimation for a basis of the central subspace, testing for predictors and prediction.

 

In order to develop this methodology necessary and sufficient conditions for each method are studied.  After that, maximum likelihood estimators for the parameters are found.  This allows us to develop inference procedures that are shown to work in simulation studies.  Finally, the models are applied to get conclusions in real data set examples.