Friday,
September 13
CBCC116, 11:00 am
(refreshments at 10:45am)
Prof. Daniel Jeske
Department of Statistics
University of California, Riverside
Title: Neutral Zone
Classifiers and their Applications
[Abstract] Neutral zone classifiers
allow for a region of neutrality when there is inadequate information to
assign a predicted class with suitable confidence. A neutral zone
classifier is defined by classification regions that trade off the cost of an
incorrect classification against the cost of remaining neutral. In this
talk, Bayes neutral zone classifiers will be discussed and it will be
demonstrated that they outperform previous neutral zone classifiers with
respect to the expected cost of misclassifications and also with respect to
computational complexity. A neutral zone classifier is illustrated with
a microbial community profiling application in which no training data is
available. Previous applications of neutral zone classifiers have only
dealt with the scenario where training data exists. Extensions to
sequential neutral zone classification will also be discussed.
Friday,
November 8
CBCC116, 11:00 am
(refreshments at 10:45am)
Prof. Bimal
Sinha
Department of Mathematics & Statistics
University of Maryland, Baltimore County
Title: Data Analysis
under Confidentiality Protection
[Abstract] Statisticians working in most federal
agencies are often faced with two conflicting objectives: (1) collect and
publish useful data sets for designing public policies and building scientific
theories, and (2) protect confidentiality of data respondents which is
essential to uphold public trust, leading to better response rates and data
accuracy. In this talk I will provide a survey of a few statistical methods
currently used at the Census Bureau.
Friday,
November 22
CBCC116, 11:00 am
(refreshments at 10:45am)
Prof. T. N. Sriram
Department of Statistics
University of Georgia
Title: Sequential FixedWidth
Confidence Interval Based on
BhattacharyyaHellinger
Distance Estimator
[Abstract] We consider the construction of a
sequential fixedwidth confidence interval for the Bhattacharyya Hellinger (BH) functional, T(g), when the random sample
comes from an unknown probability mass function g. Using the empirical mass function, ^{^}f_{n},
to estimate g, we consider the
minimum BH distance estimator, T(^{^}f_{n}),
introduced by Simpson (1987, JASA) to construct a sequential fixedwidth
confidence interval for T(g) of the form [T(^{^}f_{Nd}) − d, T(^{^}f_{Nd})
+ d] based on a fullysequential
sampling scheme, N_{d}.
The sequential procedure is shown to be asymptotically consistent and
efficient, as d −> 0. When g
= g_{θ}_{,α,L} = (1−α)f_{θ}
+ αδ_{L},
which models an experiment where observations distributed according to a
parametric model f_{θ}
are mixed with 100α% gross errors located at a large value L, we
investigate the theoretical coverage probability of our sequential
fixedwidth confidence interval based on BH estimator of θ and the
expected sample size, as d > 0. Here, we reparametrize
L = L_{d} such that L_{d} −>
1 as d −> 0. Some interesting contrasts are made with earlier
studies involving the robustness of sequential fixedwidth confidence
intervals. Monte Carlo results are presented when f_{θ}_{ } is
Poisson with mean θ.

Friday,
February 7, 2014
CBCC114, 11:30 am
(refreshments at 11:15am)
Prof. S. Rao Jammalamadaka
Department of Applied Probability and
Statistics
University of California, Santa Barbara
Title: MiddleCensoring
[Abstract] In connection
with survival analysis and related problem, there is considerable literature
which treats data that is censored from the left, the right or both. In this talk, we consider situations where the
data becomes unobservable if it falls inside a random interval in the middle,
which we call middlecensoring. This happens in clinical trials and lifetime
studies where a subject is temporarily absent or withdrawn from the study and
the event of interest occurs during this period, so that the exact time of
occurrence cannot be observed. This is also applicable in cases where some of
the observations are so imprecise that they are stated as intervals, as for
example the situation where peoples’ willingness to pay for a natural
resource. In this talk, we present a ”selfconsistent estimator” and
the Nonparametric MLE for the survival function, study its largesample
properties, and demonstrate how well it works even in the presence of heavy
censoring. Presence of covariates can also be dealt with both in a parametric
and semiparametric setup.
Friday, March 7, 2014
CBCC114, 11:30 am
(refreshments at 11:15 am)
Dr. Chandrasekaran Kandhasamy
National Institute for Research in
Tuberculosis
Indian Council of Medical Research (ICMR)
Ministry of Health, India
Title: Spatial count data modeling with CAR priors
in Bayesian approach
[Abstract] In spatial
analysis, Poisson models are typically used to analyze count data such as
disease incidence. One approach to minimize the unexplained variation in such
models is to assume prior distributions for the linear or nonlinear
predictor term to include an extra random effect in Bayesian framework. The
random effects in the above linear predictor model may have two components:
correlated heterogeneity due to spatial aspect and uncorrelated heterogeneity
due to unobserved covariates. Conditional autoregressive (CAR) models are
assumed for the correlated heterogeneity and zero mean Gaussian distributions
are assumed for the uncorrelated heterogeneity. In this talk, we will discuss
the various currently available CAR models for spatial count data in the
presence of covariate information. After an appropriate model choice using
the Deviance Information Criterion (DIC), a map of the random effects from
the resulting model is then prepared. We illustrate this procedure using
Tuberculosis and HIV data from India for the year 2011.
Thursday,
April 3, 2014
CBCC116, 1:00 pm2:15pm
(refreshments at 12:45 am)
Prof. Nitis
Mukhopadhyay
Department of Statistics
University of Connecticut, Storrs
Title: Can We Make Teaching Probability and
Statistical
Inference
Exciting? You Decide
[Abstract] Innovative
teaching leads to serious research and I find this process very exciting.
Lately, I have published extensively on various topics in probability,
statistical inference, and linear models. Related research topics and
publications originated directly from teaching graduate and undergraduate
level courses in statistics. In this presentation, I will touch upon some
selected topics from: independence, RaoBlackwell
and LehmannScheffé theorems, multivariate
normality, and invariant tests. I will highlight their origin and resolution.
Friday,
April 4, 2014
CBCC114, 11:30 am
(refreshments at 11:15 am)
Prof. Nitis
Mukhopadhyay
Department of Statistics
University of Connecticut, Storrs
Title: On
Sufficiency, Minimal Sufficiency, Information and Ancillarity:
Examples
and Counterexamples
[Abstract] Sufficiency, minimal sufficiency, information, and ancillarity are some of the deepest notions forming
the very core of all statistical science laid down by R. A. Fisher nearly 100
years ago. Yet, intricacies are abound and many
socalled “obvious results” often turn out false due to conceptual mixup. In
this presentation, I will happily explore some of the basic issues with the
help of examples and counterexamples.
