Department of Statistics Seminar
North Carolina State University

presents

Laine Elliott, Dhruv Sharma, Weihua Cao

A SAMPLER BY NCSU STATISTICS STUDENTS

1. Speaker: Laine Elliott

Title:  Adjusting for Measurement Error by Matching Moments with the Latent Variable of Interest

Abstract: A variety of complications arise when imperfect measurements, W, are observed in place of a true variable of interest, X.  When interest focuses on estimation of the density of X, this presents an obvious problem, because the mis-measured data do not have the same distribution as the latent variable X.  In the context of linear and non-linear regression models where X is a covariate, regression parameter estimators obtained when W is substituted for X may be substantially biased, and statistical power may be compromised.  The effects of measurement error have been thoroughly studied.  Many strategies for correcting for measurement error depend on the specific modeling or regression context and can be intractable in highly non-linear models.  In our work, we take an alternative perspective focused on re-creating the distribution of X from the observed W, either as the primary quantity of interest or as a means to improving parameter estimation. We obtain estimates of X for which the first M sample moments are unbiased for the corresponding moments of X. We investigate the benefit of substituting these estimates for X in density estimation and a variety of non-linear regression models.  This method has the advantage that once the estimates of X are obtained, they can be substituted in any model, including complicated non-linear models.  Regression parameter estimators that depend primarily on M moments will be virtually unbiased.   I will describe some of the results from this work.
*************

2. Speaker: Dhruv Sharma.

Title: Simultaneous Variable Selection and Clustering for SVMs

Abstract: Variable selection for high-dimensional data analysis offers unique challenges, particularly when there are many redundant noise features and high collinearity. Support Vector Machines have been successfully used to study classification problems. In this paper we propose the PACS algorithm, a new procedure to perform automatic variable selection for classification problems under such conditions. The procedure simultaneously selects variables while grouping them into predictive clusters, further aiding in the investigation of groups of features with similar behavior. The procedure is shown to compare favorably with existing methods in terms of both classification accuracy and dimension reduction, while yielding the additional group structure information.
******************

3. Speaker: Weihua Cao

Title: Improving Efficiency and Robustness of the Doubly Robust Estimator for a Population Mean with Incomplete Data

Abstract: Considerable recent interest has focused on so-called doubly robust estimators for a population mean response in the presence of incomplete data, which involve models for both the propensity score and the regression relationship between outcome and covariates. The usual doubly robust estimator may yield severely biased inferences if neither of these models is correctly specified and can exhibit nonnegligible bias if the estimated propensity score is close to zero for some observations. We propose alternative doubly robust estimators that achieve comparable or improved performance relative to existing doubly robust estimators, even with some estimated propensity scores close to zero.

Friday, September 12, 2008,------3:35:4:35 pm,-------321 Riddick

Refreshments will be served in the Riddick Reading Room at 3:00pm.***NOTE:*** No food or drink is allowed in any of the classrooms in Riddick Hall.