Monday, February 28, 2005
| Title |
Semiparametric Analysis of Longitudinal Data with Informative Observation Times |
| Speaker |
Do-Hwan Park
University of Missouri |
| Time |
3:00-4:00 p.m. |
| Place |
ENB 108 |
| Note |
The speaker is a candidate for the Asst. Professor position in Statistics. |
Abstract
Statistical analysis of longitudinal data is an important topic faced in a
number of applied fields including epidemiology, public health and medicine. In
general, the information contained in longitudinal data can be divided into two
parts. One is the set of observation times that can be regarded as realizations of
an observation process and the other is the set of actually observed values of the
response variable of interest that can be seen as realizations of a longitudinal
or response process. For their analysis, a number of methods have been proposed
and most of them assume that the two processes are independent. This greatly
simplifies the analysis since one can rely on conditional inference procedures
given the observation times.
However, the assumption may not be true in some applications. We will consider
situations where the assumption does not hold and propose a semiparametric
regression model that allows the dependence between the observation and response
processes. Inference procedures are proposed based on the estimating equation
approach and the asymptotic properties of the method are established. The results
of simulation studies will be reported and the method is applied to a bladder
cancer study.
Friday, February 25, 2005
| Title |
Robust Estimation of Mixture Complexity |
| Speaker |
Mi-Ja Woo
University of Georgia, Athens |
| Time |
2:00-3:00 p.m. |
| Place |
PHY 109 |
| Note |
The speaker is a candidate for the Asst. Professor position in Statistics. |
Abstract
Developing statistical procedures to determine the number of components, known
as mixture complexity, remains an area of intense research. In many applications,
it is important to find the mixture with fewest components that provides a
satisfactory fit to the data. This talk focuses on consistent estimation of
unknown number of components in finite mixture models, when the exact form of the
component densities are unknown but are postulated to be close to members of some
parametric family. Minimum Hellinger distances are used to develop a robust
estimator of mixture complexity, when all the parameters associated with the model
are unknown. The estimator is shown to be consistent. When there is no model
misspecification, Monte Carlo simulations for a wide variety of target mixtures
illustrate the implementation and performance of the estimator. Robustness of the
estimator examined via model misspecification shows that, in contrast to an
estimator based on Kullback-Leibler distance, the performance is unaffected by
model misspecification. An example concerning hypertension is revisited to further
illustrate the performance of the estimator.
Wednesday, February 23, 2005
| Title |
General Convex Stochastic Orderings and Related Martingale-type Structures |
| Speaker |
Francisco Vera
University of South Carolina |
| Time |
3:00-4:00 p.m. |
| Place |
LIF 267 |
| Note |
The speaker is a candidate for the Asst. Professor position in Statistics. |
Abstract
Over-dispersion of a population relative to a fitted baseline model can be
accounted for in various ways. For example, one way is by using a mixture over the
family of baseline models. Another is via a martingale structure if the Total Time
on Test (TTT) Transform of the population “dominates” that of the
baseline model. Here these latter ideas are extended to stochastic orderings in
terms of Tchebycheff systems and related to a martingale-type of structure, called
a k-mart, between the population and the baseline model. These ideas are
illustrated for a binomial baseline model using the Saxony 1876-85 sibship census
for families with twelve siblings. In addition the construction of a “most
identical” distribution in the case of 1-mart is presented.
Friday, February 4, 2005
| Title |
Analysis of Gene Expression Data and Chemosensitivity Prediction |
| Speaker |
Florence George |
| Time |
3:00-4:00 p.m. |
| Place |
PHY 109 |
Abstract
Microarrays are part of a new class of biotechnologies which allow the
monitoring of expression levels for thousands of genes simultaneously. Microarray
technology can provide important insights about the underlying genetic causes of
many important biological questions. We discuss the computational methods of four
important tasks: (1) The identification of differentially expressed genes, (2) the
discovery of clusters of differentially expressed genes, (3) identification of
features from the clusters and (4) the classification of biological samples.
The study is on gene expression levels of 55 advanced stage ovarian cancer
patients. 33 of these patients showed complete response to chemotherapy, while the
rest had a progressive disease at the completion of therapy. We sought to
determine whether the gene expression levels were sufficient for the prediction of
chemosensitivity.
Friday, January 21, 2005
| Title |
Review of Extreme Value Distributions with Examples |
| Speaker |
Rajaram Lakshminarayan |
| Time |
3:00-4:00 p.m. |
| Place |
PHY 109 |
Abstract
Extreme value theory has turned out to be one of the most important statistical
disciplines in the last few decades. One of the most outstanding features of
extreme value analysis is the objective to quantify the stochastic behavior of a
process at unusually large (or small) levels. The central platform of extreme
value theory is the three types of theorem of Fisher and Tippet, which asserts
that there are only three types of distributions that can arise as limiting
distributions of extreme values in the random samples.
The topic of this seminar mainly focuses on the review of extreme value
distributions, especially Generalized Extreme Value (GEV) and Generalized Pareto
Distributions (GPD) with examples as applied to the existing real data on rainfall
and sea-levels.
Possible applications of extreme value theory to the area of pharmacokinetics
to model the maximum drug concentrations in blood after the infusion of a drug
along with appropriate covariates will be discussed.
Friday, January 14, 2005
| Title |
How to Perform an Analysis of Variance Procedure in S-Plus |
| Speaker |
George Kimber |
| Time |
3:00-4:00 p.m. |
| Place |
PHY 109 |
Abstract
The commands in S-Plus that generate the one-way ANOVA, the factorial ANOVA,
and the nonparametric ANOVA will be demonstrated using datasets from several
disciplines. Illustrations of how to test the underlying assumptions will be
presented. Several of the post-hoc procedures will also be reviewed. A general
discussion of the rationale for and the interpretation of the Analysis of Variance
procedures and their related tests will also be conducted.