The paper proposes a comparison between dynamic models with continuous and discrete latent variables for panel data. We consider Limited Dependent Variable models (LDV) in the first case, and Latent Markov (LM) models in the second case. In both cases the maximum likelihood estimation method through the EM algorithm is used. Since the likelihood of LDV models is not tractable analytically, we implemented the Gauss Hermite and the Adaptive Gauss Hermite quadrature methods for approximating the integrals involved in it. The comparison between the two classes of models is carried out by means of a simulation study
Continuous versus discrete latent structures in dynamic latent variable models
CAGNONE, SILVIA;
2014
Abstract
The paper proposes a comparison between dynamic models with continuous and discrete latent variables for panel data. We consider Limited Dependent Variable models (LDV) in the first case, and Latent Markov (LM) models in the second case. In both cases the maximum likelihood estimation method through the EM algorithm is used. Since the likelihood of LDV models is not tractable analytically, we implemented the Gauss Hermite and the Adaptive Gauss Hermite quadrature methods for approximating the integrals involved in it. The comparison between the two classes of models is carried out by means of a simulation studyFile in questo prodotto:
Eventuali allegati, non sono esposti
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.