Bayesian estimation for multinomial logistic regression

Abstract

This paper discusses the applicability of Bayesian multinomial logistic regression model on the prediction of happiness of Somaliland youth by two variables; age and educational level. It presents two approximate methods on multinomial logistic regression estimation; classical method and Bayesian method or Markov Chain Monte Carlo (MCMC), to obtain the marginal posterior density for the parameters. A comparison of these two methods is carried out to determine the usefulness of Bayesian method on multinomial logistic regression estimation. R and WinBUGS (Bayesian Inference using Gibbs Sampling) programs have been used to fit the model. As both of these two methods have suggested; happiness increases with educational level and decreases with age. In addition, it is also shown that Bayesian Multinomial logistic regression is useful in direct computations and it produces very accurate approximations to the posterior density.