MODELOS DE ÁRBOL DE REGRESIÓN BAYESIANO: UN ESTUDIO DE CASO
Keywords:
Monte Carlo Markov Chain, Regression tree, Metric on Models, Log. Likelihood, Integrated LikelihoodAbstract
The Bayesian method for selecting regression models proposed by Chipman et al. (1998a) as well as some related results of the
same authors are explored. In applications of this method, forming groups of the generated models has resulted a very useful tool
(models are generated by Monte Carlo Markov Chains) so some metrics are required and here we propose a new one for
grouping the models. The possibility of reducing the number of necessaries models to be generated in order to determine (with
Chipman et al. (1998a) approach) a tree which give a satisfactory explanation of the data is explored by means of a simulation
study. We apply the method to the data used by Denison et al. (1998) and some comparisons are made. Lastly we analyze the
results of the application to the data of some socio-economical study
