CSI779

David Marchette

The project I am working on for CSI 779 is to replicate and extend the Monte Carlo study reported in Gray (see the abstract). The basic idea of the study is that given the assumption that data comes from a mixture (in this case a 2-component mixture), there is a bias introduced when the component terms are misspecified, either by incorrectly choosing the component densities, or by making unwaranted assumptions about the parameters of the densities (eg: incorrectly assuming equal variances).

My plan is to replicate parts of the study, and then to extend the study in three ways:

• First, add another component density to the list of those studied.
• Second, instead of looking at parameter bias, consider density error as measured by mean integrated square error.
• Third, add another density estimate, a nonparametric method called the Filtered Kernel Estimator, which uses a mixture as a pilot estimate.

A preliminary version of the paper describing this work is available both as a postscript and HTML document.

Created by David Marchette