. Note this is the last formal seminar this academic session.
CENTRE FOR MEDICAL STATISTICS
SEMINAR SERIES on BIOSTATISTICS 1999/2000
NO 9: Wednesday, June 14th, 2000, at 2:30 pm
Modelling Salamander Mating Data Using Importance Sampling
by
Dr. Jianxin Pan
(Centre for Medical Statistics, Keele University)
The salamander mating data arise from an experiment in which
each female animal was paired with three males from her own
population and with three males from another population.
In the experiment, twenty female and twenty male animals
were involved. The experiment was conducted three times
using a pre-chosen design protocol and hence 360 dependent
binary observations were generated.
A primary objective of the experiment is to study the
probability of a successful mating for each of the four
combinations of gender and population, and to see
if there exists heterogeneity among individual animals.
Modelling the salamander mating data, however, is
a challenging problem because the likelihood of
the model involves a 360-dimensional integral which
is analytically intractable. Laplace-based approximations
such as penalized quasi-likelihood (PQL) approach
lead to severely biased estimates and are not appropriate
for modelling these data.
In this talk, I present a new estimation technique which
is not restricted by the dimension of the integral. The
likelihood and its 1st- and 2nd-derivatives are calculated
using importance sampling. Moreover, I consider how to
chose an importance distribution for sampling and compare
the findings with PQL and MCMC approaches.
All welcome!!
Venue:
Room CBB0.004
Chancellor's Building - Extension
(Bananna Wing)
Keele University
http://www.keele.ac.uk
http://www.keele.ac.uk/depts/ma/seminars/medstats.html
http://www.keele.ac.uk/university/campus/maps/camus.htm
Best
Gilbert
_____________________________
Prof. Gilbert MacKenzie
Director,
Centre for Medical Statistics
Keele University
Staffordshire ST5 5BG
England
Tel: (01782) 583651
Fax: (01782) 584268
______________________________
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
