21.06.2014 Views

Annual Report 2008.pdf - SAMSI

Annual Report 2008.pdf - SAMSI

Annual Report 2008.pdf - SAMSI

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

used to conditionally model high category events conditional upon the occurrence of a hurricane.<br />

A Markov transition structure is used to propagate information over time.<br />

The second paper, on the impact of various treatments on the future drinking behavior of<br />

alcoholics, also uses a GPD structure. The goal is to determine which of three approaches- (1)<br />

twelve step program, (2) motivational treatment, (3) behavioral therapy- does the best job of<br />

deterring poor future drinking behavior. The novelty of the approach used here lies in the mode<br />

of prior elicitation. It is performed on parameters in the predictive space. This differs from the<br />

traditional approach in that we can avoid using non-informative priors on poorly understood<br />

parameters associated with the design matrix, focusing instead on quantities that are actually<br />

observed and thus better understood.<br />

Sourish presented his work at a postdoc/grad student seminar while at <strong>SAMSI</strong>, and he<br />

intends to present the second paper mentioned above at the New England Statistical Symposium<br />

in April. He also participated in a one-day program aimed at facilitating a better understanding of<br />

research in the mathematical sciences among undergraduate students. Each of these research<br />

topics will also contribute a chapter to Sourish’s Ph.D. thesis.<br />

Summary of Papers:<br />

(a) Analyzing Extreme Hurricane Activity using Multinomial-Dirichlet Model<br />

(b) Analyzing Fatal Drinking Behavior of Patients suffering Alcohol Dependence Disorder using<br />

Pareto Regression<br />

The second work above is currently being sponsored by the University of Connecticut Health<br />

Center.<br />

From Sourish Das<br />

Relationship of <strong>SAMSI</strong> Research to Ph.D. Dissertation<br />

I am developing Bayesian Method of analyzing extreme category in Multinomial-<br />

Dirichlet model, especially, in the context of the Hurricane data of Indian Ocean (southern<br />

hemisphere region) and Pacific Ocean (West pacific region). Here the storms are categorized into<br />

5 categories; where estimating the probability of rare category (that is category 5 hurricane) is<br />

challenging. This work will be a part of the 3 rd chapter of my Ph.D. dissertation.<br />

I also developed the Bayesian analysis of Pareto Regression model with unknown shape<br />

parameter, for excess over threshold values, designed for longitudinal studies. We developed<br />

Pareto regression model in Generalized Linear Model (GLM) framework, using log-link between<br />

the shape parameter of random component and systematic component of the model. We put into<br />

practice one-step Markov transition model to embrace the longitudinal component in the model.<br />

Then we implemented our methodology to examine the association of GABRA2 gene with fatal<br />

alcohol intake, with the effect of psychological treatment as comparative study of efficacy of<br />

three treatments on fatal alcohol intake.<br />

Research Contributions<br />

1. <strong>SAMSI</strong> Technical <strong>Report</strong> # 2007-08 “On Bayesian Analysis of Generalized Linear<br />

Models: A New Perspective”<br />

2. <strong>SAMSI</strong> Technical <strong>Report</strong> # 2007-09 “On Bayesian Inference for Generalized<br />

Multivariate Gamma Distribution”

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!