UNIVERSITY OF GLASGOW
STATISTICS SEMINAR PROGRAMME
Wednesday, 26 January, 3 pm (Glasgow-Strathclyde partnership seminar)
Service in a Poisson rain
Serguei FOSS (Institute of Mathematics, Novosibirsk)
(visiting University of Strathclyde)
Wednesday, 16th February, 3pm
Bayesian analysis of mixtures with an unknown number
of components --- an alternative to reversible jump methods
Matthew STEPHENS (University of Oxford)
Wednesday, 23rd February, 4 pm (Joint Meeting with Glasgow RSS Local Group)
Veterinary modelling --- a trip to the zoo
George GETTINBY (University of Strathclyde)
Wednesday, 1st March, 3 pm
Testing effects in repeated measures designs
Toni RIETVELD (University of Nijmegen)
Seminars take place in Room 1f(203), Mathematics Building,
University of Glasgow
For further information please contact the seminar organiser:
Ilya Molchanov
University of Glasgow : e-mail: [log in to unmask]
Department of Statistics : Ph.: + 44 141 330 5141
Glasgow G12 8QW : Fax: + 44 141 330 4814
Scotland, U.K. : http://www.stats.gla.ac.uk/~ilya/
ABSTRACTS
SERVICE IN A POISSON RAIN
The talk will deal with a queueing problem that involves a server
moving with a constant speed along a line in a positive direction.
Customers appear on a line in a ``Poisson rain''. A server has to stop
and to serve every customer it meets. A number of asymptotical results
concerning the server's location $X(t)$ at time $t\to \infty$ will be
established. A number of related queueing problems will be discussed
also.
BAYESIAN ANALYSIS OF MIXTURES WITH AN UNKNOWN NUMBER OF COMPONENTS ---
AN ALTERNATIVE TO REVERSIBLE JUMP METHODS
Mixture distributions are typically used to model data in which each
observation is assumed to have arisen from one of a number of
different groups. They also provide a convenient and flexible class of
models for density estimation. The analysis of such models has a long
and distinguished history, and continues to attract interest and
present difficulties. In this talk we will briefly review some of
these difficulties before concentrating on the particularly difficult
problem of deciding how many components to use in a mixture model.
Richardson and Green (1997) consider a Bayesian approach to this
problem using a Markov Chain Monte Carlo (MCMC) approach, which makes
use of the ``reversible jump'' methodology described by Green (1995).
We describe an alternative MCMC method which views the parameters of
the model as a (marked) point process, extending methods suggested by
Ripley (1977) to create a continuous time Markov birth-death process
with an appropriate stationary distribution. We illustrate our method
on both univariate and bivariate data, make some comparisons with the
reversible jump methodology, and describe some general difficulties
with choosing suitable priors for the model parameters.
VETERINARY MODELLING --- A TRIP TO THE ZOO
no abstract provided
TESTING EFFECTS IN REPEATED MEASURES DESIGNS
Repeated measures designs (also called within-subjects designs) are
very frequently used in psycholinguistics. They are potentially very
powerful, and yet require only a relatively small number of subjects,
by presenting them all items at issue. Many aspects of these designs,
however, have been subject of debate, and of continuing research,
among which are the question as to whether items (words) used in
current psycholinguistic experiments should be considered as random or
fixed effects, and the associated problem of finding statistics which
yield a fair balance between power and the chance of making Type I
errors.
In this contribution I will will review current conventions in
reporting the results of psycholinguistic research in which repeated
measures designs are used, and then comment on these conventions in
the light of available statistical knowledge.
In order to estimate probabilities of Type I and Type II errors
associated with different statistics and different characteristics of
data sets, a number of data sets were generated. For each condition
10000 data sets, all with two within-subject factors (Type, with three
levels and Words nested under Type, 10 words per level of Type) and 30
subjects. Thus a complete data set comprised 900 data points
(simulated RT-measurements).
The effects of five characteristics of the data sets on Type I and
Type II-errors were determined: 0effect size of factor Type, with
values 0,0,0 (no effect), -20, 0, +20 (small effect), -40, 0, +40
(large effect); 1presence or absence of interaction subject x types;
2presence or absence of missing data; 3presence or absence of skewness
in distributions from which the items and the error component were
drawn. Presence or absence of dependency between words in different
groups (thus the matching of words in different Types was simulated).
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|