School of Mathematical Sciences
                   Queen Mary, University of London


                              Summer 2002

             STATISTICS SEMINAR: DESIGN OF EXPERIMENTS

                             All are welcome


The talks are held in the Mathematics Seminar Room (103)
on Level 1 of the Mathematics Building, Queen Mary, University of London.

The nearest underground station is Stepney Green.
Turn left at the exit and walk 400 yards.

_______________________________________________________________________
   DATE SPEAKER TITLE
-----------------------------------------------------------------------
9 May 2002
        16.30 L.M. Haines Optimal Design for Random
                University of Natal Intercept Models
                Pietermaritzburg, Durban
-----------------------------------------------------------------------
16 May 2002
        15.00 J.D. Godolphin Identifying Designs which
                University of Surrey are Vulnerable to Observation
                                                Loss

        16.00 Coffee Break

        16.30 A.F.M. Smith Bayesian Methods for Nonlinear
                Queen Mary, UL Classification and Regression
-----------------------------------------------------------------------
30 May 2002
        15.00 M. Keogh Brown A Study of Internet Request
                Queen Mary, UL Patterns

        16.00 Coffee Break

        16.30 A. Biswas Variance-Adjustment for Odds
                Indian Statistical Ratio in '2 by 2' Contingency
                Institute, Calcutta Tables
-----------------------------------------------------------------------


                          A B S T R A C T S
                          -----------------

---------------------------------------------------------------------
                           L. M. Haines
          Optimal Design for Random Intercept Models

This talk is concerned with optimal design for a simple linear
regression model with a random intercept term and an explanatory
variable belonging to the set {0,1,...,k}. Individual
designs comprising up to k+1 distinct values of the explanatory
variable are assumed to be available, there being 2^{k+1}-1 such
designs. The problem of constructing a population design which
allocates weights to these individual designs in such a way that
the information associated with the fixed effects is in some sense
maximized is addressed. In particular D- and V-optimal designs
are discussed and a geometric approach to confirming their global
optimality is introduced.

The work is joint with my Ph.D. student Legesse Kassa.
---------------------------------------------------------------------


                          J.D. Godolphin
  Identifying Designs which are Vulnerable to Obervation Loss

It is well known that a design may become disconnected during the
course of an experiment because of the loss of S, a subset of the
original observations, for reasons beyond the control of the
experimenter. If this situation arises then certain treatment
contrasts are inestimable and it is impossible to test the usual
hypothesis that all treatments have the same effect. We say that
a design is vulnerable to observation loss if there are subsets S
which consist of only a few observations. The purpose of this talk
is to describe a procedure for identifying designs which are
vulnerable in this way. Applications of the method to several
incomplete block and row-column designs are discussed.
---------------------------------------------------------------------


                            A.F.M. Smith
  Bayesian Methods for Nonlinear Classification and Regression

An overview will be given of a variety of flexible nonlinear regression
approaches to curve and surface fitting, and to classification
problems using tree and partition models. Implementation relies on
stochastic search procedures using Markov chain Monte Carlo.
---------------------------------------------------------------------


                           M. Keogh Brown
                A Study Of Internet Request Patterns

We propose a method for extracting traffic activity patterns from
web cache logs and present a statistical time series model which
reflects the behaviour of the internet data. We analyze some
statistical properties of the model. We also consider the physical
properties of the caches and cache networks and their influences
on the traffic.
---------------------------------------------------------------------


                                A. Biswas
   Variance-adjustment for odds ratio in '2 by 2' contingency tables

While analyzing '2 by 2' contingency tables,
log-odds ratio is often approximated by a normal distribution with some
variance. In the present paper we show that the expression of that
variance is erroneous in the presence of correlation between two
binomial distributions of the contingency table. We provide some
adjustment to this variance expression. By extensive computations
it is illustrated that this variance-adjusted
normal approximation is a much better approximation for such data. The
method is applied to some real life datasets. The
whole development is dependent on the existence of a bivariate binomial
distribution. A multivariate (and hence bivariate) binomial distribution
is motivated and derived for that purpose.

This is joint work with
Jing-Shiang Hwang
Institute of Statistical Science, Academia Sinica, Taipei, Taiwan

--------------------------------
For more information ask:

Barbara Bogacka
School of Mathematical Sciences
Queen Mary, University of London
Mile End Road
London E1 4NS

Tel: 020 7882 5497
e-mail: [log in to unmask]
----------------------------------

The seminar information is kept on:

http://www.maths.qmw.ac.uk/~rab/seminars.html
---------------------------------------------