Department of
presents
Dr. Leonard Stefanski
stefanski at stat dot ncsu dot edu
ABSTRACT
I will describe two new approaches to variable selection in regression modeling based on research with Dennis Boos and students Xiaohui Luo and Yujun Wu. The key idea in both approaches is to calibrate an existing tunable selection method in order to achieve desirable properties of selected models. I will restrict attention to forward selection for which the tuning parameter is the familiar alpha-to-enter, aka SLENTRY to SAS users. In the first approach, Noise Added Model Selection (NAMS), parametric bootstrap-like data sets are generated by incrementally adding noise to the response variable, and SLENTRY is tuned by tracking the effect of added noise on selected models' mean squared errors for different SLENTRY values. In the second approach, Variable Added Model Selection (VAMS), random phony predictor variables are added to the data set, and SLENTRY is tuned by tracking the proportion of falsely included phony variables in the models selected for different SLENTRY values.
Friday, September 29, 2006
3:35 - 4:35 pm
206 Cox Hall
Refreshments will be served on the second floor of Dabney Hall (left of Room 222) at 3:00 pm.