Seminari

[Giovanna De Santi <gdesanti@ds.unifi.it>]

DIPARTIMENTO DI STATISTICA "G. Parenti" - UNIVERSITA' DI FIRENZE


Si comunica che nei giorni 22-26 novembre 1999, il prof. Dankmar Böhning, della Free University  of Berlin, terrà una serie di seminari dal titolo

A Course on Mixture Models: Outline

I seminari verranno tenuti presso l'aula del Dipartimento secondo il seguente progamma:

Monday: Introduction
09:30 - 12:30
- Population heterogeneity: the natural genesis of mixture models
- Some examples
- Parametric or nonparametric mixture models?
14:00 - 16:30
- Connection to empirical Bayes estimation
- Classification using posterior Bayes
- Missing covariates and mixture models

Tuesday: Theory of nonparametric mixture models
09:30 - 12:30
- The likelihood and its properties
- The directional derivative and the gradient function 
- The general mixture maximum likelihood theorem
- Applications of the theorem
14:00 - 16:30
- Existence and number of support points of the NPMLE
- A case study on the Simar's accident data
- Exercises on real data sets: homogeneity or heterogeneity? (Tutored by Tan Böhning)
- Diagnostic tests for heterogeneity
- Testing the number of components with the likelihood ratio
- 
Wednesday: Algorithms
09.30 - 12.30
Finding the number of components
- The vertex direction method (VDM)
- The vertex exchange method (VEM)
- Step-length choices
- Introducing C.A.MAN
14:00 - 16:30
Number of components known
- The EM algorithm for mixture models
- Mixing normls with different variance components
- Exercises with C.A.MAN (Tutored by Tan Böhning)

Thursday: Applications of Mixtures: Meta-Analysis
09:30 - 12:30
- The conventional approach

- Strategies of optimal pooling and a likelihood approach
- Heterogeneity
- Classification of studies using posterior Bayes
14:00 - 16:30
- Introducing META
- Exercises with META (Tutored by Tan Böhning)
- Adjusting for heterogeneity using DerSimonian-Laird
- Adjusting for heterogeneity using Sarol-Böhning

Friday: Applications of Mixtures: Disease Mapping
09:30 - 12:30
- The conventional approach

- Empirical Bayes estimators and their maps
- Estimating map heterogeneity - estimating the prior
- Heterogeneity in perinatal mortality for the North West Thames Health Region - data: a case study
14:00 - 16:30
- Introducing DISMAP
- Exercises with DISMAP (Tutored by Tan Böhning)
- Adjusting for heterogeneity using DerSimonian-Laird

The course will give a state-of-the-art introduction into the area and various field appliactions. The course will use the book by the author on mixture models:
Dankmar Böhning (1999) Computer-Assisted Analysis of Mixtures with Applications: Meta-Analysis, Disease mapping and Others. Chapman & Hall /CRC, Boca Raton.

La S.V. è cordialmente invitata.

Per ulteriori informazioni sul seminario contattare abiggeri@ds.unifi.it