Statistics Seminar: Bruce Brown -- Cramer-von Mises smoothing of covariate curves

Bruce Brown
School of Mathematics and Statistics
University of New South Wales 

Location: Carslaw 173 

Time: 2pm Friday, June 4, 2010 

Title: Cramer-von Mises smoothing of covariate curves 

Abstract: Smooth curve fitting can be carried out as a stand-alone exercise, or for the
effect of individual covariates within linear models.  Penalty function methods for
doing this lead to a spline-type formulation, and the ’most common’ method for choosing
the penalty coefficient, as in journal citations, is cross-validation.  Because
cross-validation uses a criterion different from those in a penalty function
formulation, and different again from indicators of consequent unwanted bias in linear
models, there are a number of inter-related, potentially confusing questions about how
these models should be set up and fitted.  Following an intuitive non-parametric
approach leads to a Cramer-von Mises method which may settle some of these questions, as
well as leading to a surprising alternative formulation.