Department of Statistics Seminar
North Carolina State University

presents

Pabak Mukhopadhyay, Steven Novick and Jared Lunceford

North Carolina State University

A Sampler of Graduate students

ABSTRACT

Three graduate students of the Department of Statistics at NCSU will present brief overviews of their current research.

• Pabak Mukhopadhyay - "Bayesian Analysis of Quality Adjusted Lifetime Data"
  Quality of Life is an important aspect of evaluating clinical trials of chronic diseases such as Cancer and AIDS. It is assumed that patients progress through a series of health states which differ in quality of life. Bayesian methodologies to model such data using sampling based methods are presented. Simulation illustrations are also provided.

• Steven Novick - "Reducing Measurement Error Bias in Parametric Modeling with Monte-Carlo Corrected Score Functions"
  When estimating parameters in a model relating a response variable to a vector of covariates, a typical assumption is that the covariates are precisely measured. Unfortunately, it is often the case that some or all of the covariate's components are measured with error. Correcting the score is a technique proposed independently by Stefanski (1989) and Nakamura (1990) to reduce measurement error bias in parameter estimation. By varying the procedure of Cook and Stefanski's (1994) simulation extrapolation, it can be shown that a Monte-Carlo corrected score is obtainable for a large class of smooth true-data score function.

• Jared Lunceford - "Benchmark Dose Estimation for Continuous Heteroscedastic Data"
  A class of risk assessment methods, collectively referred to as Benchmark Dose (BMD) methods, are currently under development as quantitative strategies for identifying critical exposure levels for environmental toxicants. We studied the effects of variance modeling and method of estimation on BMD point estimates and lower confidence bounds using a factorial design of estimation methodologies and model parameters. The definition of BMD utilized here is the toxicant dose corresponding to a 5 percent increase in response mean over background, relative to the total response range. Such a definition is appropriate for bounded, monotone, dose-response models such as the Hill equation used in this study. The methods of Maximum Likelihood (ML), Ordinary Least Squares (OLS), and Generalized Least Squares (GLS) are compared under various levels of heteroscedastity. The experiment is also performed using data generated under alternative variance models so that the assumed model is incorrect. Simulation results demonstrate the need to incorporate variance modeling into BMD estimation. The underlying variance model and the method of estimation have a significant influence on the BMD lower confidence bounds and coverage probabilities.

Friday, August 27, 1999

3:35 - 4:35 pm

206 Cox Hall

Refreshments will be served on the second floor of Dabney Hall (left of Room 222) at 3:00 pm.