[Forum SIS] avviso di seminario - N. Salvati, Università di Pisa
Maria Giovanna Ranalli
giovanna a stat.unipg.it
Mar 5 Ott 2010 11:05:48 CEST
.: Nicola Salvati, Dipartimento di Statistica e Matematica Applicata
all'Economia, Università di Pisa :.
Venerdi 22 Ottobre, ore 12, Aula 201
Università degli Studi di Perugia, Dipartimento di Economia, Finanza e
Statistica
.: M-quantile and Expectile Random Effects Regression for Multilevel
Data :.
Abstract:
The analysis of hierarchically structured data, for example
longitudinal data or geographically clustered data, is usually carried
out by using random effects models. The primary goal of random effects
regression is to model the expected value of the conditional
distribution of an outcome variable given a set of explanatory
variables while simultaneously accounting for the dependence structure
of hierarchical data. The expected value, however, does not always
offer a complete picture of this conditional distribution. In this
paper we show how one can overcome this problem by fitting
hierarchical linear models to the M-quantiles of the conditional
distribution of the outcome variable given the covariates. Our
proposed M-quantile random effects regression model extends M-
quantile regression (Breckling and Chambers, 1988) and can be
considered as an alternative to the quantile random effects model
(Geraci and Bottai, 2007). M-estimation is synonymous with outlier-
robust estimation, and so robust estimation of both fixed and random
effects is straightforward. An advantage of the M-estimation framework
is that it also allows for the use of expectile regression. This can
potentially lead to efficiency gains when the use of outlier-robust
estimation methods is not justified but there is still interest in
modelling the conditional distribution of the variable of interest. We
show how a modified maximum likelihood approach can be used to fit the
M-quantile random effects regression model, and inference for
estimators of the fixed and random effects parameters is discussed.
The performance of this approach is then evaluated in a series of
simulation studies. We conclude the paper by describing a case study
where both M-quantile and expectile random effects regression are used
to analyse repeated measures data collected from a rotary pursuit
tracking experiment.
tutte le informazioni su http://www.ec.unipg.it/DEFS/
cordiali saluti
M. Giovanna Ranalli
~ Dipartimento di Economia, Finanza e Statistica
~ Sezione di Statistica
~ Via Pascoli
~ Universita' degli Studi di Perugia
~ 06123 Perugia - Italy
~ Tel +39 075 5855939
~ Fax +39 075 5855950
~ url: www.stat.unipg.it/~giovanna
Maggiori informazioni sulla lista
Sis