Department of
presents
John Staudenhmayer
Density estimation and heteroskedastic
measurement error
Abstract
This talk is about density estimation
when the variable of interest is subject to heteroskedastic
measurement error. We take a non-parametric approach and model both the density
and the variance function with penalized mixtures of B-splines.
The model is fit with Bayesian methods. Contributions of the work include an
explanation of the biases incurred by incorrectly assuming homoskedastic
measurement errors and the derivation of an equivalent kernel for a spline based density estimator. The talk will also discuss
other approaches to density estimation in the presence of measurement error,
including deconvolution kernels, non-parametric
maximum likelihood, and flexibly specified / ordinary mixed models. A
motivating dataset from nutritional epidemiology will be discussed. This is
joint work with John Buonaccorsi (UMass-Amherst)
and David Ruppert (
Friday, December
7, 2007
3:35 - 4:35 pm
301 Riddick Hall
Refreshments will be served in the common area of 301 Riddick at 3:00 pm.