Non-linear regression models assuming  that the error term has a finite mixture of normal distributions

Vicente Garibay CANCHO[1]

Jorge Alberto ACHCAR[2]

Edwin Moises Marcos ORTEGA[3]

§    ABSTRACT: n this paper, we explore the use of Markov Chains Monte Carlo (MCMC) methods to develop a Bayesian analysis for non-linear regression (NLMR) models assuming that the error term has a finite mixture of normal distributions. We use the Metropolis-Hastings algorithm to simulate samples for the model. We illustrate the the proposed methodology with two examples, considering simulated and real data sets.

§    KEYWORDS: Non-linear regression; selection of models; Markov Chain Monte Carlo; mixture of normal distribution.



[1] Departamento de Matemática Aplicada e Estatística, Instituto de Ciências Matemáticas e de Computação -- ICMC, Universidade de São Paulo – USP, CEP:13560-970, São Carlos, SP,Brasil. E-mail: garibay@icmc.usp.br

[2] Departamento de Estatística, Universidade Federal de São Carlos - UFSCar, Caixa Postal 676, CEP 13565-905, São Carlos, SP, Brasil. E-mail: jachcar@power.ufscar.com.br

[3] Departamento Ciências Exatas, Universidade de São Paulo, campus de Piracicaba – ESALQ/USP, Caixa Postal 9, CEP 13418-900 , Piracicaba, SP, Brasil. E-mail: edwin@ciagri.usp.br