>
Dear All
<>Re: Biostatistics Seminars in Limerick
Biostatistics Seminar Series Academic Year 2006/7 - please
note that in the second semster Friday seminars will be
held at 14:00 in A2002 and not at 15:00 as previously
announced.
First Semester
1. H-likelihood approach to spatio-temporal modelling
Prof. Youngjo Lee, Seoul National University,.South Korea.
October 3rd, 3:15 Room A2002.
2. Double Hierarchical Generalized Linear Models
Prof. John Nelder, Imperial College, London , UK
October 24th, 3:00 Room A2002.[Postponed due to illness ]
3. Title: Modelling of mean-covariance structures in
generalised estimating equations for longitudinal data
Prof . Jianxin Pan , University of Manchester, UK.
November 15th, 3pm Room A2002
When used for modelling longitudinal data generalised estimating
equations specify a working structure for the within-subject covariance
matrices, aiming to produce efficient parameter estimators. However,
misspecification of the working covariance structure may lead to a large
loss of efficiency of the estimators of the mean parameters. In this
talk I will introduce an approach for joint modelling of the mean and
covariance structures for longitudinal data within the framework of
generalised estimating equations. The resulting estimators for the mean
and covariance parameters are consistent and asymptotically Normally
distributed. Real data analysis and simulation studies show that the
proposed approach produces efficient estimators for both the mean and
covariance parameters.
<>
4. Analysis of Multivariate Survival Data via HGLMs
Prof. Il Do Ha, Daegu Haany University, Daegu City, South Korea.
November 22nd 3pm Room, A2002.
<>Recently, random-effect survival models such as frailty models or
mixed-effect models have been widely used to analyze multivariate (or
correlated) survival data in the form of recurrent or multiple-event
times which often arise in the research fields of medicine or
econometrics. In particular, these data can be unbalanced and/or
correlated including the bivariate form, and also can be censored and/or
truncated due to the study design as in classical univariate survival
data. For inference, marginal likelihood methods (e.g. MCEM, GHQ), which
require integration out the random effects, have been mainly developed,
but become computationally heavier as the number ofrandom components
increases (Gueorguieva, 2001; Huber et al., 2004). This difficulty has
limited the wider application of such models.
Random-effect models have been recently extended to HGLMs (hierarchical
generalized linear models, Lee and Nelder, 1996, 2001,2006), which allow
various random structures such as crossed and/ornested structures,
structured dispersion, or spatial and temporal correlations. The HGLM
method based on h-likelihood (or hierarchica llikelihood) provides a
statistically efficient and simple unified framework for various
random-effect models. Thus, random-effect survival models can be
modelled and fitted via the HGLM (Ha and Lee,2003, 2005; Ha, Lee and
MacKenzie, 2006).
In this talk, we introduce the various forms of multivariate survival
data and then show how to model, fit and analyze such data via the HGLM.
We also discuss about the practical uses and further work..
5. Improvement of Watterson's and related estimates for the
recombination rate based on shrinkage.
<> Prof. Andreas Futschik, University of Vienna, Austria
December 1st, 2pm, Room A2002
We focus on the estimation of the scaled mutation parameter $\theta$, which is one of the parameters of key interest in population genetics.
One of the most popular estimates in this context is Watterson's estimate. We show that Watterson's estimate is inadmissible when taking the mean squared error as the measure of performance.Subsequently we propose an alternative estimator that is easy to calculate and
has a smaller mean squared error than that of Watterson's estimate for all possible parameter values $0<\theta<\infty.$ We show furthermore
that this estimate is admissible in the class of all linear estimates. We also propose improved versions of related estimates and derive a
class of Bayes-estimates.
------------------------------------------------------------------------
Second Semester
1. Why I hate minimisation
Prof. Stephen Senn , University of Glasgow, UK
January 26th (Friday), 14:00 Room A2002.
Minimisation is a technique of sequentially marginally balancing
clinical trials. It is not based on sound design theory, brings marginal
advantages compared to randomisation as regards orthogonality and some
disadvantages as regards blinding. Furthermore its very debatable merits
have been over-exaggerated by its proponents who tend to use the fact
that they have minimised as an excuse to ignore prognostic information.
In this talk I shall argue that conditioning is the way to make
inferences valid in the presence of covariate information and that
minimisation has no useful role in designing clinical trials.
<>
2. Fisher information & design of quantum experiments
Prof. Peter Jupp, University of St.Andrew's, Scotland,UK
February 16th (Friday), 14:00, Room A2002.
Quantum theory is (by its very nature) probabilistic, and so gives rise
to problems in statistical inference. In classical parametric inference
an important question is `What parts of the data are informative about
parameters of interest?'. Key concepts here are those of Fisher
information, sufficient statistic, and cut. This talk will explore some
analogous concepts for quantum statistical inference.
<> 3. Examining Spatial Heterogeneity through Geographically Weighted
Regression
Prof . Stewart Fotheringham, NUI .
March 7th (Wednesday) ,15:00, Room A2002
Abstract to follow.
4. Stochastic Models for Patient Care
Prof. Sally McClean, University of Ulster, UK
March 28th (Wednesday), 15:00, Room A2002.
<> Abstract to follow.
5. Hidden Markov Chain Models in Statistical Genetics
Dr. David Ramsey, University of Limerick, Ireland
April 18th (Wednesday), 15:00, Room A2002
Abstract to follow.
6. TBA *
Prof. Goeran Kaumerman, Bielefeld University, Germany.
May 11th (Friday) 14:00 Room A2002
Abstract to follow.
7. Modelling of molecular biological processes with genomic data
Prof Ernst Wit, Lancaster University, UK
May 25th (Friday) 14:00 Room A2002.
<>The current flood of all types of genomic data raises the challenge to
make our models for the underlying biological processes both relevant
and feasible. We give several approaches of model-based inference of
such biological processes using e.g. microarray data. <>
*Some details to be confirmed.
All welcome ...
<>We take tea 15 mins before the seminars.
<>Best
Gilbert
Seminar organiser.
--
_____________________________
Prof. Gilbert MacKenzie
Dept. of Mathematics & Statistics,
University of Limerick,
Limerick
Ireland
CoB ~ http://www.ul.ie/biostatistics
ISA ~ http://www.istat.ie.
Gilbert ~ http://www.staff.ul.ie/mackenzieg
Email: [log in to unmask]
Tel: +353 (0)61 213499
Fax: +353 (0)61 334927
_________________________
--
|
