Department of
presents
Dr. Howard Bondell
bondell at stat dot ncsu dot edu
ABSTRACT
In many common situations, interest focuses on selecting a subset of variables to predict a response. Often it is the case that the predictor variables under consideration have an underlying grouping structure to them, either known or unknown to the scientist. Known grouping structures include the case of Analysis of Variance (ANOVA) models where multiple `dummy' variables are used to represent a single factor. In this situation, it is desired to either include or exclude the entire set of variables together as a group. Unknown grouping structure occurs when some predictors are interrelated in such a way that they may be highly correlated with one another and have a group effect on the response, such as a group of genes sharing a common pathway, for example. Supervised clustering techniques attempt to find these `predictive groups' by combining information from the response along with the relationships among the predictors. In this talk, a new penalized regression approach to perform variable selection along with supervised clustering will be discussed. A penalized approach to select entire factors in the ANOVA model is also introduced. The form of this penalization has the ability to simultaneously collapse levels within the selected factors to perform what is typically done in a post-hoc analysis involving pairwise comparisons.
For those of you unfamiliar with the title: The `Odd Couple’ refers to a Broadway play (originally from 1965) that was made into a film (1968) and later became a television series in the 1970’s, in which the two main characters were named Oscar and Felix. To see how it relates to the talk, you will just have to wait and see.Friday, November 10, 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.