Department of Mathematical Sciences

University of Nevada, Las Vegas

Statistics Colloquium/Seminar Series



[2006-2008] [2008-2009] [2009-2010] [2010-2011] [2012-2013] [Current Year]


For more information, contact the Colloquium/Seminar Coordinator, Dr. Hokwon Cho

(To see Math Dept Colloquia/Seminars, click next: Math Dept Seminar)


Fall 2011

Friday, September 9
CBC-C128, 11:00 am

(refreshments at 10:45am)

Dr. Subir Ghosh
Department of Statistics

University of California, Riverside

Title: Non-Iterative Methods and Local Linear Approximations

for Probit Regression Models

[Abstract] Numerous methods are available for iteratively solving the maximum likelihood estimating equations (MLEEs) for probit regression models. This paper introduces new noniterative methods for solving MLEEs using the exact solutions for all possible paired observations and the exact solution obtained by the local linear approximations of two weight functions in MLEEs. The exact solutions in such special circumstances permit us to establish the bias in the estimators for proposed methods. A real data is used to demonstrate the closeness of solutions of the proposed new methods with that of the standard methods available in literature and statistical softwares. The simulation is also performed to make the comparison. Theoretical properties are established to evaluate the proposed methods.

Friday, October 14
CBC-C128, 11:00 am

(refreshments at 10:45 am)

Dr. Mai Zhou
Department of Statistics

University of Kentucky

Title: Empirical Likelihood Method and Survival Analysis

[Abstract] Empirical likelihood test is a general nonparametric statistical test method. We shall use the first half of the talk to give an introduction of this method which is mostly summarized in Owen (2001) book. In the second half of the talk we shall focus on the development of the EL method in the survival analysis, with right censored data. We discuss efforts of use EL in regression models (AFT model and Cox model), two sample problems, in estimation with over determined estimating equations and use with other data types.

Friday, November 4
CBC-C128, 11:00 am

(refreshments at 10:45am)

Dr. Jungwon Mun
Department of Mathematics & Statistics

California State Polytechnic University, Pomona

Title: Notes on Coefficients of Intrinsic Dependence: Improvement and

Relationship with Existing Tests

[Abstract] As high dimensional data arise often in modern studies, correlation measures have recently regained the attention of statisticians and researchers in many fields such as genetics, bioinformatics, psychology and behavioral science s. Dimension reduction becomes a major concern in the analysis of high dimensional data and correlation measures play a key role in it. In 2005, Hsing and his co-authors introduced a new measure, called Coefficient of intrinsic dependence (CID) that overcomes drawbacks and broadens the applicable area. As we have studied and investigated the CID, we found a way to refine the measure that improves its performances and a more logical way to compare the behaviors of CID with other correlation measures. In addition, we derived theoretical relationships between CID and ANOVA. Various simulations and figures are presented illustrating the updated behaviors of CID and our claims of its relevance with ANOVA.


Spring 2012

Friday, January 27
CBC-C128, 11:00 am

(refreshments at 10:45am)

Dr. Chul Ahn
Professor & Director of Biostatistics and Research Design

Department of Clinical Sciences

University of Texas Southwestern Medical Center

Title: Design and Analysis of Correlated outcomes

[Abstract] Correlated outcomes commonly occur in biomedical research. Examples include repeated measurement studies, clustered outcome studies such as dental studies, ophthalmologic studies, and community intervention studies. The correlations among observations must be taken into consideration in the design due to the potential association for observations within a cluster. I present parametric and nonparametric methods to calculate sample sizes and powers for studies with correlated outcomes. Design and analysis of correlated outcomes will be illustrated using examples from studies of diagnostic sensitivity and specificity, community intervention studies and repeated measurement studies.

Thursday, March 8
CBC-C214, 1:00 pm

(refreshments at 12:45 pm)

Dr. Mingan Yang
School of Public Health

Saint Louis University, Saint Louis, MO

Title: Bayesian Semiparametric Latent Variable Model:

Application to study of bleeding in relation to Fibroid Tumor


[Abstract] In parametric hierarchical models, it is standard practice to place mean and variance constraints on the latent variable distributions for the sake of identifiability and interpretability. Because incorporation of such constraints is challenging in semiparametric models that allow latent variable distributions to be unknown, previous methods either constrain the median or avoid constraints. In this article, we propose a centered stickbreaking process (CSBP), which induces mean and variance constraints on an unknown distribution in a hierarchical model. This is accomplished by viewing an unconstrained stick-breaking process as a parameter-expanded version of a CSBP. An efficient blocked Gibbs sampler is developed for approximate posterior computation. The methods are illustrated through a simulated example and an epidemiologic application.

Tuesday, March 13
CBC-C322, 1:00 pm

(refreshments at 12:45 pm)

Dr. Gengxin Li
Department of Epidemiology & Public Health

Yale University

Title: The Improved SNP Calling Algorithms for Illumina Beadarray Data


[Abstract] Genotype calling from high throughput platforms such as Illumina and Affymetrix is a critical step in data processing, so that accurate information on genetic variants can be obtained for phenotype-genotype association studies. A number of algorithms have been developed to infer genotypes from data generated through the Illumina BeadStation platform, including GenCall, GenoSNP, Illuminus, and CRLMM. Most of these algorithms are built on population-based statistical models to genotype every SNP in turn, such as GenCall with the GenTrain clustering algorithm, and require a large reference population to perform well. These approaches may not work well for rare variants where only a small proportion of the individuals carry the variant. A fundamentally different approach, implemented in GenoSNP, adopts a SNP-based model to infer genotypes of all the SNPs in one individual, making it an appealing alternative to call rare variants. However, compared to the population-based strategies, we propose the two-stage SNP calling procedures, to improve call accuracy for both common and rare variants. The effectiveness of our approach is demonstrated through applications to genotype calling on a set of HapMap samples used for quality control purpose in a large case-control study of cocaine dependence. The increase in power with our proposed method is greater for rare variants than for common variants depending on the model.

Monday, March 19
CBC-C214, 1:00 pm

(refreshments at 12:45 pm)

Dr. Hongxia Yang
IBM Watson Research Center

Yorktown, NY

Title: Cascade Model with Dirichlet Process for Analyzing

Multi-Relational Dynamic Matrices


[Abstract] Dyadic matrices are natural data representations in a wide range of domains, including social media, business analytics, and recommendation systems. Dyadic data often involve two types of abstract objects, and the dyadic matrix is based on the observed pairs with each element from either object. Due to the increasing needs from the practical applications, dyadic data analysis attracts more attentions and many techniques have been developed recently. However, most of the existing approaches, such as co-clustering and relational reasoning, only handle single dyadic table and lack flexibility to reveal the patterns from multiple relational matrices. We propose a general nonparametric Bayesian framework with a cascaded structure to model multiple relational dyadic matrices, and then perform an efficient hybrid Gibbs sampling algorithm for posterior inference and analysis. We later extend the methodology via Nonparametric Bayesian collective matrix factorization. Extensive evaluations are conducted over both synthetic data and real data and the experimental results show that the proposed model captures the hidden structure of data and generalizes clustering analysis and predictive inference in a unique way.

Tuesday, March 20
CBC-C214, 1:00 pm

(refreshments at 12:45 pm)

Dr. Tao Lu
National Institutes of Health

Washington D.C.

Title: Statistical Methods for Dynamic Models with Application Examples


[Abstract] A dynamical system in engineering and physics, specified by a set of differential equations, is usually used to describe a dynamic process which follows physical laws or engineering principles. The parameters in the dynamical system are usually assumed known. However, an interesting question to ask is how to estimate these parameters when they are not known before. In this talk, I show you two examples where various statistical methods are applied to dynamic models for estimating unknown parameters based on observed data. Eventually, we are interested in predicting the future behavior of the dynamic system. The first example is on modeling HIV viral load dynamics from a clinical trial study. The second is on modeling a complicated interactive network.



Statistics Colloquium/Seminar Series