ST 790 Final Projects
The projects are to result in thorough but concise, professional quality technical reports of not more than ten 1.5-spaced pages (not including listings of raw data, computer output and R/WinBUGS code) followed by a 15 minute seminar style presentation. Projects are to be turned in by April 28, 2010. If handed in before that date, the instructor will give feedback and allow you to make improvements. One paragraph proposals are due by April 7, 2010. These proposals should spell out briefly the data source, the main goal of the study, and your basic statistical methodology. Pick a subject that interests you (preferably on a topic that you learned in this course) and then execute and analyze the results of a study intended to increase
your understanding of the subject.
All presentations will take place in 5270 Sas Hall on Wednesday, April 28, 2010.
- 1:30-1:55: Eric Reyes
Title: Bayesian Average Error Based Approaches to Sample Size Calculations for Hypothesis Testing
Abstract: In the classical framework, sample size calculations are made to adequately maintain pre-specified Type-I and Type-II error rates associated with the hypothesis test of interest. We present a Bayesian approach to hypothesis testing and sample size determination based on average errors. Computational details for several examples
are provided, with an emphasis on one and two sample binomial experiments. We also compare the estimated sample size for several examples with those obtained from classical frequentist methods.
- 1:55-2:20: Phillip Schulte
Title: Bayesian Modeling of Rare Event Data
Abstract: Medical researchers are often interested in computing the probability of observing an event or number of events when such events are infrequent. Here, we examine a theoretical meta-analysis of trials testing efficacy of a drug on a rare event. In such situations, frequentist methods often break down given estimation close to the boundary. Thus, we use a Bayesian approach, comparing several Bayesian modeling possibilities and their corresponding assumptions. A simulation study will be conducted which may shed light on methods that perform well in properly identifying treatment differences.
- 2:20-2:45: SeungJun Shin
Title: Maximum tolerated dose estimation using Bernstein polynomial based nonparametric Bayesian response curve estimators
Abstract: In clinical trials, especially in phase I studies, one of the main goals is to find the maximum tolerated dose (MTD) using a dose response model. Several parametric methods have been developed to estimate MTD. However the accuracy of the estimator critically depends on the assumed form of dose response curve. We consider a Bayesian semi-parametric method based on Bernstein polynomials, proposed by Curtis and Ghosh (2010) to estimate MTD. A numerical algorithm is developed to obtain a posterior estimate of the MTD based on Curtis and Ghosh's Bernstein polynomial based response curve estimator and compare its performance to a standard logistic regression based estimator. The computation of the MTD estimate is illustrated using real data sets.
Last updated on: April 23, 2010