[Forum SIS] avviso di Seminario Prof. Bartolucci

[Julia Mortera]

A tutti gli interessati, si porta a conoscenza del* SEMINARIO* 
organizzato nell'ambito del dottorato di Metodi statistici per 
l'economia e l'impresa (MSEI)

* *

*Francesco Bartolucci, Università degli Studi di Perugia*

Latent variable models for longitudinal data


*MERCOLEDÌ 01 FEBBRAIO 2012, ORE 14,30 – Aula 24- 3° piano *

* *

*Dipartimento di Economia Università degli Studi Roma Tre* - *Via Silvio 
D’Amico 77, 00145 Roma*

* *

* *

*Abstract*

*I briefly review two approaches for the analysis of longitudinal 
data,which are based on a common root: using a sequence of latent 
variables to model the unobserved heterogeneity between sample units. 
The first*

*approach is based on assuming that the latent variables are 
discrete-valued and then the resulting latent process follows a Markov 
chain. The second approach is based on assuming, instead, a latent 
AR(1)process. After a brief discussion of the pros and cons of the two 
approaches, I propose a third approach based on assuming that the latent 
process follows a mixture of AR(1) models. The advantage is an 
improvement in the fit (close to that reachable by a latent Markov 
model), while retaining a parsimonious structure (close to that of the 
latent AR(1) model). For the latent mixture AR(1) model based on the 
proposed approach, I illustrate how to perform maximum likelihood 
estimation by an Expectation-Maximization algorithm implemented by 
well-known recursions in the hidden Markov literature. The use of the 
model is illustrated by an application involving a longitudinal dataset 
coming from the Health and Retirement Study, about the self-reported 
health status by a sample of subjects aged 50 and over.*

-- 
Julia Mortera
Dipartimento di Economia
Università Roma Tre
Via Silvio D'Amico 77
00145 Roma
Tel: +39 06 57335732
Fax: +39 06 57335771
URL: http://host.uniroma3.it/facolta/economia/vedi_docente.asp?id=49