************** Please note change of address *********************
***** Our department has moved to new premises. ***************
***** A map with instructions on how to get to our **************
***** department can be found at the following web site: *********
***** http://www.ucl.ac.uk/Stats/map.html ************************
Everyone is welcome. Tea, coffee and biscuits are provided!
Details of the seminar:
================
Monday 7 February 2000 - 4pm, Room 102, 1-19 Torrington Place,
Department of Statistical Science - University College London.
Speaker : Prof. J. Nelder - Dept. of Mathematics - Imperial College
Title: A large class of models derived from GLMs
Abstract:
Originally GLMs allowed errors from a one-parameter exponential family, and
a transformation to linearity of the systematic effects via a link
function. The following extensions comprise a large class of useful models
with algorithms all based on iterative weighted least squares.
(1) Data-driven curves for continuous covariates via splines (GAMs)
(2) Quasi-likelihood models for variance functions not in the GLM family.
(3) Joint modelling of mean & dispersion via two interlinked GLMs.
(4) Hierarchical GLMs (HGLMs) allowing random effects both normal and
non-normal, with a fitting criterion that avoids integration.
(5) Combination of (3) and (4) to give joint modelling of mean &
dispersion with fixed and random effects in models for both components.
(6) Generalization of correlated responses to non-normal GLM
distributions, relevant to analysis of longitudinal data.
(7) Generalization of the Kalman filter via HGLMs for dynamic models.
----------------------------------------------------------------------------
----------
K. Skouras, Lecturer in Statistics, Department of Statistical Science, UCL,
1-19 Torringhton Place, London WC1E 6BT, United Kingdom.
Tel: 020 - 76791862 Fax: 0171-7383 4703
e-mail: [log in to unmask]
----------------------------------------------------------------------------
-----------
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|