|
Department of
Mathematical Sciences
2009-2010 For more information, contact the Colloquium/Seminar
Organizer, Dr. Hokwon Cho (To
see Math Dept Colloquia/Seminars, click next: Math
Dept Seminar) |
|
Fall 2009 |
|
·
Fri.
12:30 p.m. November 20, SEB-1240:
Dr. Donatello Telesca Department
of Biostatistics, Abstract: Functional data often exhibit a common shape but
also variations in amplitude and phase across curves. The analysis often
proceeds by synchronization of the data through curve registration. We
propose a Bayesian hierarchical model for curve registration. Our methodology
is extended to define a class of probability models, which combine curve
registration with functional mixed effects modeling, discriminating phase and
amplitude variability in a joint fashion. We discuss this class of models
with a focus on penalized smoothing splines and
propose Bayesian inferential procedures based on Markov Chain Monte Carlo
samples from the posterior distribution of the functions of interest. We
illustrate the application of our model using simulated data as well as to
two datasets, namely, the ·
Fri.
1:00 p.m. September 25, SEB-1240:
Dr. Peter Müller Department
of Biostatistics, Abstract: We consider statistical inference for high
throughput genomic data. Most traditional statistical methods implicitly
assume independent sampling (conditional on some hyperparameters).
Recognizing the limitations of independent modeling we develop a model that
includes a simple dependence structure across genes (or proteins). The
important features of the proposed model are the ease of representing typical
prior information on the nature of dependencies, model-based parsimonious
representation of the signal as a ordinal outcome, and the use of a coherent
probability model over both, structure and strength of the conjectured
dependencies. |
|
Spring
2010 |
|
·
Seminar
topics and schedule will be forthcoming. |
|
2008-2009
Statistics Colloquium/Seminar |
|
Spring
2009 |
|
·
Fri.
11:30 a.m. April 24, CBC C-224:
Dr. Charles Davis President,
Environmetrics and Statistics Ltd Abstract: Lognormal (LN) distributions are often assumed for
environmental contaminants, with perhaps some justification. But decisions
are made from measurements, not the unobservable concentrations themselves.
These often do not have LN distributions. Rather, at fixed concentrations
distributions of measurements are often normally distributed, and if
low-level measurements are unbiased one has negative values; standard LN
inference techniques fail in this setting. This reality is universally
ignored; measurement values are censored at a Reporting Limit, the negative
values are never seen, and we continue to develop (and publish) methods for
left-censored LN environmental data. ·
Fri.
11:30 a.m. February 27, CBC C-224:
Dr. Kevin Quinn Department
of Government & Institute for Quantitative Social Science, Abstract: We amass a new,
large-scale dataset of newspaper editorials that allows us to calculate
fine-grained measures of the political positions of newspaper editorial
pages. Collecting and classifying over 1500 editorials adopted by 25 major US
newspapers on 495 Supreme Court cases from 1994 to 2004, we apply an item
response theoretic approach to place newspaper editorial boards on a
substantively meaningful—and long validated—scale of political preferences.
We validate the measures, show how they can be used to shed light on the
permeability of the wall between news and editorial desks, and argue that the
general strategy we employ has great potential for more widespread use. ·
Fri.
11:30 a.m. February 13, CBC C-224:
Dr. Barry Arnold Department of Statistics, Abstract: The Azzalini skew-normal
density of the form 2φ(x)Φ(λx) can be
viewed as having arisen by considering a bivariate
random variable (X,Y) with a classical bivariate
normal density and focussing on the conditional
distribution of X given Y < E(Y). The same family of distributions is
encountered if we consider the conditional distribution of X given Y >
E(Y). A slightly more general family is provided by considering the
conditional distribution of X given Y > y0 where y0
is not necessarily equal to E(Y). The resulting model (which we can call a
hidden truncation model, since we only observe X if the unobserved or hidden
variable Y exceeds a threshold value, is a flexible extension of the
classical univariate normal model with potential to
.t a broad spectrum of data configurations which may not be well fitted by a
classical normal model. In the present paper we consider several other basic bivariate non-Gaussian models and investigate the nature
of their corresponding hidden truncation models. In particular, it is of
interest to identify situations in which hidden truncation fails to augment
the basic model. Additive component representations provide an alternative to
the hidden truncation paradigm in the normal case. It is conjectured that it
is only in the normal case that the two models coincide. ·
Fri.
1:00 p.m. February 6, CBC C-224:
Dr. Grace Chiu Department of Statistics and
Actuarial Science, Abstract: We propose a model-based approach for constructing
ecological health indices through statistical inference. Our latent health
factor index (LHFI) is obtained by estimating an unobservable health factor
term in a mixed-effects ANOCOVA that directly models the relationship among
indicator variables (or metrics) and health. Unlike conventional indices (e.g. IBI and
O/E index) that rely on domain-specific calibrations of metrics against
reference conditions whose non-constancy is largely unaccounted for, our
methodology (a) involves no explicit reference conditions while metrics are
intrinsically "calibrated" in the context of multiple comparisons,
and (b) can naturally incorporate spatio-temporal
influences on calibration schemes. ·
Fri.
2:30 p.m. January 23, CBC C-224:
Dr.
Abel Rodriguez Department of Applied Mathematics and Statistics, Abstract: This talk discusses
clustering procedures for nested samples of curves, where multiple profiles
are collected for each subject in the study. We start by considering
the application of standard functional clustering tools to this
problems, which lead to groupings based on the average profile for
each subject. After discussing some of the shortcoming of this
approach, we present a model based on a generalization of the nested
Dirichlet processes that uses the information on the distribution of curves
to generate the clusters. The method is illustrated using data from the
Early Pregnancy Study on hormone profiles along multiple menstrual periods
for a cohort of women. The resulting model simultaneous cluster both
curves and subjects, allowing us to identify outlier curves for each group of
women, as well as outlying women whose distribution of profiles differs from
the rest. |
|
Fall 2008 |
|
·
Friday
11:30 a.m. November 21, CBC C-224: Dr.
N. Balakrishnan Department
of Statistics, Title: Over/Under-Dispersed
Poisson Distributions and Processes Abstract: In this talk, I will establish several connections of the Poisson weight function to overdispersion and underdispersion. Specifically, I will show that the logconvexity (logconcavity) of the mean weight function is a necessary and sufficient condition for overdispersion (underdispersion) when the Poisson weight function does notdepend on the original Poisson parameter. I will also discuss some properties of the weighted Poisson distributions (WPDs). I will then introduce a notion of pointwise duality between two WPDs and discuss some associatedproperties. Next, after presenting some illustrative examples and providing a discussion on various Poisson weight functions used in practice, I will make some concluding remarks. Finally, I will use these results to introduce and discuss over/under-dispersed Poisson processes.
·
Friday
11:30 a.m. November 7, CBC C-224: Dr. Glen Meeden Department
of Statistics, Title: A Noninformative Bayesian Approach to Finite Population
Sampling Using Auxiliary Variables Abstract: In finite population
sampling prior information is often available in the form of partial
knowledge about an auxiliary variable, for example its mean may be known. In
such cases, the ratio estimator and the regression estimator are often used
for estimating the population mean of the characteristic of interest. The Polya posterior has been developed as a noninformative Bayesian approach to survey sampling. It
is appropriate when little or no prior information about the population is
available. Here we show that it can be extended to incorporate types of
partial prior information about auxiliary variables. We will see that it
typically yields procedures with good frequentist properties
even in some problems where standard frequentist
methods are difficult to apply. Moreover one does not need to select a model
which explictly relates the characteristic of
interest to the auxiliary variables. ·
Friday
11:30 a.m. October 10, CBC C-224: Dr. Lurdes Inoue Department
of Biostatistics, Abstract: In this talk we discuss some modeling approaches to
investigate disease progression. First, we propose a model that links
longitudinal biomarker and disease progression. Specifically, we consider an
underlying latent disease process that describes the onset of the disease and
models the transition to an advanced stage of the disease as dependent on the
biomarker levels. Next, we propose a variation of the above model to
investigate disease progression using data prospectively collected in a
screening study. We illustrate our methods through simulations and a case
study in prostate cancer. ·
Friday
11:30 a.m. October 3, CBC C-224: Dr. Ben Kedem Department
of Statistics, Title: Bayesian
Spatial Prediction Abstract: We discuss Bayesian
spatial/temporal prediction in transformed Gaussian random fields where the
transformation belongs to a parametric family. ·
Friday
3:30 p.m. September 12, CBC C-224: Dr. Ashis SenGupta Department
of Statistics, University of California, Riverside and Applied
Statistics Unit, Indian Statistical Institute, Kolkata, India Abstract: Observations
on angular propagations, directional orientations, and even strictly periodic
phenomena can be cast in the arena of directional data (DD). Such
observations are frequently encountered in almost every sphere of applied
science, ranging from e.g. agriculture to zoology, chronotherapy
to defence, etc. There has been a paucity of
probability distributions to model DD, even on 3-smooth manifolds, e.g.
torus, cylinder, and hence on their higher dimensional generalizations. We
present here unified approaches to derivations of such distributions. Tests
for orthogonality of the directional random
variables are then obtained based on these distributions. Next, models for
regression with linear and circular variables are presented and related
inference procedures are developed. Both classical and Bayesian approaches
are discussed. Finally, the proposed methods are illustrated by several
real-life examples. |
|
2008-2009
Statistics Colloquium/Seminar |
|
Spring
2009 |
|
·
Fri.
11:30 a.m. April 24, CBC C-224: Dr. Charles
Davis President,
Environmetrics and Statistics Ltd Abstract: Lognormal (LN) distributions are often assumed for environmental
contaminants, with perhaps some justification. But decisions are made from
measurements, not the unobservable concentrations themselves. These often do
not have LN distributions. Rather, at fixed concentrations distributions of
measurements are often normally distributed, and if low-level measurements
are unbiased one has negative values; standard LN inference techniques fail
in this setting. This reality is universally ignored; measurement values are
censored at a Reporting Limit, the negative values are never seen, and we
continue to develop (and publish) methods for left-censored LN environmental
data. ·
Fri.
11:30 a.m. February 27, CBC C-224:
Dr. Kevin Quinn Department
of Government & Institute for Quantitative Social Science, Abstract: We amass a new,
large-scale dataset of newspaper editorials that allows us to calculate
fine-grained measures of the political positions of newspaper editorial
pages. Collecting and classifying over 1500 editorials adopted by 25 major US
newspapers on 495 Supreme Court cases from 1994 to 2004, we apply an item
response theoretic approach to place newspaper editorial boards on a
substantively meaningful—and long validated—scale of political preferences.
We validate the measures, show how they can be used to shed light on the
permeability of the wall between news and editorial desks, and argue that the
general strategy we employ has great potential for more widespread use. ·
Fri.
11:30 a.m. February 13, CBC C-224:
Dr. Barry Arnold Department of Statistics, Abstract: The Azzalini skew-normal
density of the form 2φ(x)Φ(λx) can be
viewed as having arisen by considering a bivariate
random variable (X,Y) with a classical bivariate
normal density and focussing on the conditional
distribution of X given Y < E(Y). The same family of distributions is
encountered if we consider the conditional distribution of X given Y >
E(Y). A slightly more general family is provided by considering the
conditional distribution of X given Y > y0 where y0
is not necessarily equal to E(Y). The resulting model (which we can call a
hidden truncation model, since we only observe X if the unobserved or hidden
variable Y exceeds a threshold value, is a flexible extension of the classical
univariate normal model with potential to .t a
broad spectrum of data configurations which may not be well fitted by a
classical normal model. In the present paper we consider several other basic bivariate non-Gaussian models and investigate the nature
of their corresponding hidden truncation models. In particular, it is of
interest to identify situations in which hidden truncation fails to augment
the basic model. Additive component representations provide an alternative to
the hidden truncation paradigm in the normal case. It is conjectured that it
is only in the normal case that the two models coincide. ·
Fri.
1:00 p.m. February 6, CBC C-224:
Dr. Grace Chiu Department of Statistics and Actuarial
Science, Abstract: We propose a model-based approach for constructing
ecological health indices through statistical inference. Our latent health
factor index (LHFI) is obtained by estimating an unobservable health factor
term in a mixed-effects ANOCOVA that directly models the relationship among
indicator variables (or metrics) and health. Unlike conventional indices (e.g. IBI and
O/E index) that rely on domain-specific calibrations of metrics against
reference conditions whose non-constancy is largely unaccounted for, our
methodology (a) involves no explicit reference conditions while metrics are
intrinsically "calibrated" in the context of multiple comparisons,
and (b) can naturally incorporate spatio-temporal
influences on calibration schemes. ·
Fri.
2:30 p.m. January 23, CBC C-224:
Dr.
Abel Rodriguez Department of Applied Mathematics and Statistics, Abstract: This talk discusses
clustering procedures for nested samples of curves, where multiple profiles
are collected for each subject in the study. We start by considering
the application of standard functional clustering tools to this
problems, which lead to groupings based on the average profile for
each subject. After discussing some of the shortcoming of this
approach, we present a model based on a generalization of the nested
Dirichlet processes that uses the information on the distribution of curves
to generate the clusters. The method is illustrated using data from the
Early Pregnancy Study on hormone profiles along multiple menstrual periods
for a cohort of women. The resulting model simultaneous cluster both
curves and subjects, allowing us to identify outlier curves for each group of
women, as well as outlying women whose distribution of profiles differs from
the rest. |
|
Fall 2008 |
|
·
Friday
11:30 a.m. November 21, CBC C-224: Dr.
N. Balakrishnan Department
of Statistics, Title: Over/Under-Dispersed
Poisson Distributions and Processes Abstract: In this talk, I will establish several connections of the Poisson weight function to overdispersion and underdispersion. Specifically, I will show that the logconvexity (logconcavity) of the mean weight function is a necessary and sufficient condition for overdispersion (underdispersion) when the Poisson weight function does notdepend on the original Poisson parameter. I will also discuss some properties of the weighted Poisson distributions (WPDs). I will then introduce a notion of pointwise duality between two WPDs and discuss some associatedproperties. Next, after presenting some illustrative examples and providing a discussion on various Poisson weight functions used in practice, I will make some concluding remarks. Finally, I will use these results to introduce and discuss over/under-dispersed Poisson processes.
·
Friday
11:30 a.m. November 7, CBC C-224: Dr. Glen Meeden Department
of Statistics, Title: A Noninformative Bayesian Approach to Finite Population
Sampling Using Auxiliary Variables Abstract: In finite population
sampling prior information is often available in the form of partial
knowledge about an auxiliary variable, for example its mean may be known. In
such cases, the ratio estimator and the regression estimator are often used
for estimating the population mean of the characteristic of interest. The Polya posterior has been developed as a noninformative Bayesian approach to survey sampling. It
is appropriate when little or no prior information about the population is
available. Here we show that it can be extended to incorporate types of
partial prior information about auxiliary variables. We will see that it
typically yields procedures with good frequentist properties
even in some problems where standard frequentist
methods are difficult to apply. Moreover one does not need to select a model
which explictly relates the characteristic of
interest to the auxiliary variables. ·
Friday
11:30 a.m. October 10, CBC C-224: Dr. Lurdes Inoue Department
of Biostatistics, Abstract: In this talk we discuss some modeling approaches to
investigate disease progression. First, we propose a model that links
longitudinal biomarker and disease progression. Specifically, we consider an
underlying latent disease process that describes the onset of the disease and
models the transition to an advanced stage of the disease as dependent on the
biomarker levels. Next, we propose a variation of the above model to
investigate disease progression using data prospectively collected in a
screening study. We illustrate our methods through simulations and a case
study in prostate cancer. ·
Friday
11:30 a.m. October 3, CBC C-224: Dr. Ben Kedem Department
of Statistics, Title: Bayesian
Spatial Prediction Abstract: We discuss Bayesian
spatial/temporal prediction in transformed Gaussian random fields where the
transformation belongs to a parametric family. ·
Friday
3:30 p.m. September 12, CBC C-224: Dr. Ashis SenGupta Department
of Statistics, University of California, Riverside and Applied
Statistics Unit, Indian Statistical Institute, Kolkata, India Abstract: Observations
on angular propagations, directional orientations, and even strictly periodic
phenomena can be cast in the arena of directional data (DD). Such
observations are frequently encountered in almost every sphere of applied
science, ranging from e.g. agriculture to zoology, chronotherapy
to defence, etc. There has been a paucity of
probability distributions to model DD, even on 3-smooth manifolds, e.g.
torus, cylinder, and hence on their higher dimensional generalizations. We
present here unified approaches to derivations of such distributions. Tests
for orthogonality of the directional random
variables are then obtained based on these distributions. Next, models for
regression with linear and circular variables are presented and related
inference procedures are developed. Both classical and Bayesian approaches
are discussed. Finally, the proposed methods are illustrated by several
real-life examples. |
|
2007-2008
Statistics Colloquium/Seminar |
|
Spring
2008 |
|
·
Friday
11:30 a.m. May 2, CBC C-224: Dr. Kaushik Ghosh Department of Mathematical Sciences, Abstract: In longitudinal studies of
patients with the Human Immunodeficiency Virus (HIV), objectives of interest
often include modeling of individual-level trajectories of HIV Ribonucleic
Acid (RNA) as a function of time. Empirical evidence suggests that individual
trajectories often possess multiple points of rapid change, which may vary
from subject to subject --- both in number and in location. Presence of such changepoints make
the modeling of individual viral RNA levels difficult, since usual methods
become unsuitable. In this
talk, we present a new robust multiple-change point model for longitudinal
trajectories. The proposed method uses a joint model to incorporate
information from the longitudinal data as well as from informative dropouts,
which are common in such studies. A Dirichlet process prior is used to model
the distribution of the changepoints. The Dirichlet
process leads to a natural clustering, and thus, sharing of information among
subjects with similar trajectories. A fully Bayesian approach for model
fitting and prediction is implemented using the Gibbs sampler on the ACTG 398
clinical trial data. ·
Friday
11:30 a.m. March 28, CBC C-224: Dr. Yuedong Wang Department of Statistics and Applied Probability, Abstract: Almost all of the current nonparametric regression
methods such as smoothing splines, generalized additive models and varying coefficients
models assume a linear relationship when nonparametric functions are regarded
as parameters. In this talk we present a general class of nonlinear
nonparametric models that allow nonparametric functions to act nonlinearly.
They arise in many fields as either
theoretical or empirical models. We propose new estimation methods based on
an extension of the Gauss-Newton method to infinite dimensional spaces and
the backfitting procedure. We extend the
generalized cross validation and the generalized maximum likelihood methods
to estimate smoothing parameters. Connections between nonlinear nonparametric
models and nonlinear mixed effects models are established. Approximate
Bayesian confidence intervals are derived for inference. We will also present
a user friendly R function for fitting these models. The methods will be
illustrated using two real data examples. ·
Friday
2:00 p.m. February 1, CBC C-224: Toby
White, Ph.D candidate Department of Statistics, Abstract: Latent class transition models are used to partition
a population into a small number of relatively homogeneous subgroups so that the movement of individuals among these
subgroups can be followed through time. One context for these
models involves the |
|
Fall 2007 |
|
·
Fri.
11:30 a.m. November 30, CBC C-225: Dr.
Anton Westveld Department of Mathematical Sciences, Abstract: An extensive literature in international and
comparative political economy has focused on the how the mobility of capital
affects the ability of governments to tax and regulate firms. The
conventional wisdom holds that governments are in competition with each other
to attract foreign direct investment (FDI). Nation-states observe the
fiscal and regulatory decisions of competitor governments, and are forced to
either respond with policy changes or risk losing foreign direct investment,
along with the politically salient jobs that come with these
investments. The political economy of FDI suggests a network of
investments with complicated dependencies. We
propose an empirical strategy for modeling investment patterns in 24 advanced
industrialized countries from 1985-2000. Using bilateral FDI data we
estimate how increases in flows of FDI affect the flows of FDI in other
countries. Our statistical model is based on the methodology developed
by Westveld & Hoff (2007). The model
allows the temporal examination of each notion's activity level in investing,
attractiveness to investors, and reciprocity between pairs of nations.
We extend the model by treating the reported inflow and outflow data as
independent replicates of the true value and allowing for a mixture model for
the fixed effects portion of the network model. Using a fully Bayesian
approach, we also impute missing data within the MCMC algorithm used to fit
the model. A working paper can be found at: http://faculty.unlv.edu/westveld/Papers/FDI.pdf. ·
Fri.
11:00 a.m. October 12, CBC C-225:
Dr. Junyong Park Department of Mathematics and Statistics, Abstract: We consider the problem of
testing whether a high dimensional observation vector has signal, i.e.,
testing all the mean values are zero versus the alternative that non-zero
means exist. The setup is when the dimension of vector is large, and the mean vector is 'sparse', e.g., the small fraction
of mean values is non-zero. We suggest a test which is not sensitive to the
exact tail behavior under normality assumption. In particular, if the
'moderate deviation' tail of the distribution is represented as the product
of a tail of a standard normal and a `slowly changing' function, our
suggested test is robust. In particular, a need for robust test is expected
when the observations are of the normalized form where normality assumption
is commonly used from C.L.T. |
|
|
|
Fall 2006 |
|
·
Fri.
11:00 a.m. November 3, CBC C-225:
Dr. Nitis Mukhopadhyay Department of Statistics, Abstract: A horticulturist was considering the number of days
each marigold variety took from planting seeds to reach a stage when first bud
appeared. The primary interest was to estimate the maximum waiting time
between “seeding” and “first budding” among three varieties. It was thought
that a 99% confidence interval of width one day would suffice since
the data could be recorded with accuracy of one-half day. We assumed a normal
distribution for the response variable. The horticulturist provided positive
lower bounds for the variances that led to unequal pilot sample sizes.
Accordingly, a new two-stage sampling design had to be developed and
implemented. We will show that the data validated all assumptions made during
the course of this investigation. Some of the important exact as well as large-sample
properties of the proposed methodology will also be summarized. Interpretations
of the properties would be highlighted with real data.
Finally, we will argue that the new methodology is theoretically
superior to an existing methodology in case the pilot sizes could somehow be
“chosen” equal. Using the data on hand, the superiority of the new
methodology will be indicated. |