Student Seminar Series - February 27, 2007
University of Minnesota
School of Statistics
College of Liberal Arts
Graphical
Methods of Determining Predictor Importance and Effect
Aaron Rendahl
Tuesday, February 27, 2007
10:10 AM, 155
Ford Hall
Minneapolis, East Bank Campus
Refreshments
at 9:45 AM
300 Ford Hall
Abstract
One of the goals of a
regression analysis is to understand how a particular predictor
affects
the response. In this research, a criterion for evaluating
and
creating graphical methods for this purpose is proposed. The
method
is used to evaluate and study existing methods, including the
component-plus-residual
plot and the added-variable plot. Additionally, two
apparently new
plots are discussed, one called a marginal-plus-residual plot, and
one
that is a new version of the added variable plot for generalized
linear models. The proposed work is to expand this
framework to
study plots for more general regression models and algorithms.