[Forum SIS] Seminar: Roberta Varriale on Thursday 7JULY2009 at 4:30pm (DEC Bocconi)

[Marco Bonetti]

*************************************************************************************
                                                      ANNOUNCEMENT
*************************************************************************************
On THURSDAY, July 9th 2009 at 4:30pm, in Room C of Bocconi University
(Via Sarfatti 25, Milan)


                                                      Roberta Varriale
                                     Department of Statistics  
"G.Parenti"
                                               University of Florence,  
Italy

will hold the seminar:

=================================================================
"Robust Random Effects Models: a diagnostic approach based on the  
forward search"


Abstract:

This work presents a simple robust method for the detection of  
atypical observations and the analysis of their effect on model  
inference in the random effect linear models (McCulloch and Searle,  
2001). In particular, we extend the approach used in the fixed effect  
framework (Bertaccini and Varriale, 2007) using a Forward Search  
procedure that orders the observations by their closeness to the  
hypothesized model (Atkinson and Riani, 2000).
The starting point of the FS is to fit the model to very few  
observations chosen with a robust procedure, order all the  
observations by their closeness to the fitted model, increase the  
subset size and refit the model. The process continues with increasing  
subset sizes until all data are fitted.
In random effect models, also known as multilevel models, the outliers  
may affect the data at each level of observation. Attention is limited  
to two hierarchical levels. During the search, at each stage, we  
monitor some informative quantities, such as parameter estimates,  
residual plots and other relevant statistics in order to identify the  
outliers. In particular, we focus on the effect of outliers on the  
second-level variance using the likelihood ratio test suggested by  
Self and Liang (1987). A cut-off point separating the outliers from  
the other observations is identified through a graphical analysis of  
the information collected at each step of the Forward Search; the  
Robust Forward LRT is the value of the classical LRT statistic at the  
cut-off point. Through some Montecarlo simulation studies we are able  
to claim the clear superiority of our proposal since the probability  
of the type I error computed with the FS method is much lower than the  
one computed with the classical approach when data are contaminated,  
without any loss in terms of power when data are not contaminated.

(Work in collaboration with B. Bertaccini, Department of Statistics  
"G.Parenti", University of Florence, Italy)
====================================================================

You are warmly invited to participate.

Sincerely,

Marco Bonetti


---
Marco Bonetti
Department of Decision Sciences
Bocconi University
Via Guglielmo Roentgen 1
20136 Milan, Italy
Tel +39 02 58365670
Fax +39 02 58365634


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.stat.unipg.it/pipermail/sis/attachments/20090707/94619cff/attachment-0001.html>