GENERAL APPLICATIONS SECTION
Half Day Meeting on Mixed Models in Data Analysis
Tuesday 12th February, 2pm (Tea 3.30pm) at the RSS, Errol Street, London
Sensitivity analysis for incomplete longitudinal data
Professor Geert Molenberghs
(Limburgs Universitair Centrum)
A variety of methods have been proposed for the analysis of
incomplete(longitudinal data). Some are too simplistic (complete case
analysis, last observation carried forward), some rely on simplifying
assumptions such as missing at random or ignorable. Recently, more methods
have become available to allow for non-random missingness. However, also
these pose issues of identifiability, sensitivity to model assumptions, etc.
In this talk, we will sketch some of the issues and outline strategies for
sensitivity analysis.
Small samples
Dr James Roger
(Livedata(UK) Ltd)
Standard errors for fixed-effect parameter estimates and tests based on
fixed-effect contrasts in standard REML procedures assume that the
random-effect parameters are known. For small samples these approximations
may give misleading results. Several practical examples will be described.
Better approximations for use with small samples and ongoing research will
be discussed.
Smoothing splines in mixed models
Sue Welham
(London School of Hygiene and Tropical Medicine)
Recently, methods have been developed so that smoothing spline models
can be fitted within a linear mixed model framework. As well as standard
cubic smoothing splines, other types of smoothing spline can also be fitted
as linear mixed models. Examples from agriculture and epidemiology will be
used to demonstrate the use of smoothing splines in both investigating the
structure of data and in quantifying features of the fitted curves.
Extending multilevel models to complex cross classified and multiple
membership data structures.
Professor Harvey Goldstein, Mr Jon Rasbash and Dr William Browne
(Centre for Multilevel Modelling, Institute of Education,
University of London)
Multilevel models have been fitted successfully to many kinds of
hierarchically structured data. Many data sets, however, exhibit more
complex mixtures of cross classifications and multiple membership
structures. The talk will explore ways of fitting such data using both
likelihood based and Bayesian methods. Examples will be drawn from
Education, Veterinary Epidemiology and Demography.
ALL ARE WELCOME
*****************************************************************
Dr Karen Vines (Secretary, GAS)
Department of Statistics
The Open University
Walton Hall
Milton Keynes, MK7 6AA
e-mail: [log in to unmask]
|