Não foi possível enviar o arquivo. Será algum problema com as permissões?

Essa é uma revisão anterior do documento!


Aplicações de inferência bayesiana aproximada para modelos gaussianos latentes espaço temporais

Aplicações de inferência bayesiana aproximada para modelos gaussianos latentes espaço temporais

The family of latent Gaussian models is flexible and suitable for a wide range of applications requiring complex modelling. In particular, spatial temporal data is among the most challenging structures for statistical models. The present work revises some strategies for modelling such data, including spatio-temporal interactions. Inference for such models usually relies on computational intensive algorithms such as MCMC (Monte Carlo Markov Chain). However, routine implementation of algorithms of this kind for spatial and/or temporal problems typically faces difficulties related to dimensionality and dependence structure. Proposals for new methods and algorithms for this class of models are still investigated by on recent literature. The INLA (integrated nested Laplace approximation) approach recently proposed by RUE, MARTINO and CHOPIN, 2009 has proved efficient and promising for different applications and is investigated here and used in the analysis of three problems with different targets and data structures. The first refers to the assessment of water quality with an assumed Gaussian distribution for the response variable. The second has dengue mosquito (\textit{Aedes aegypti}) egg counts collected at ovitraps in the municipality of Recife/PE as a negative binomial response variable. The final application refers to citrus leprosis disease assuming a binomial distribution for the presence/absence binary data. Close results for the two approaches were obtained in the first case. For the second some relevant differences were found such as significance of coefficients associated to covariates being investigated, although with comparable predictions of spatio-temporal effects. The third problem proved more ch alleging, with no reliable estimates of spatio-temporal interactions and showing greater differences between the two approaches with an apparent over-smoothing of spatial effect obtained by GAM associated with low coverage for prediction intervals. Measures of concordance between predicted and observed values such as mean and absolute squared error, correlation and coverage rates were used when comparing the two approaches within the three applications. Results show that in general INLA provides a more realistic and flexible approach producing more reliable results.

Códigos dos exemplos

Funções adicionais


QR Code
QR Code pessoais:wbonat:nutri (generated for current page)