253 – SPEED: Government Statistics, Health Policy, and Marketing
Analysis of the Australian Election Study Using Bayesian Quantile Regression Models
Charles Au
University of Sydney
S. T. Boris Choy
University of Sydney
Quantile regression is useful for modeling the conditional quantile of the dependent variable. Recently, quantile regression has also been applied to discrete choice models, where the dependent variable is binary or ordinal. They can be estimated using the Bayesian Markov chain Monte Carlo (MCMC) approach when the error terms are assumed to follow, for example, the asymmetric Laplace distribution. This paper proposes the application of Bayesian quantile regression models to survey data from the Australian Election Study (AES). The binary and ordinal quantile regression models will be used for investigating the factors that influence Australian voters’ choice for certain political parties and their level of interest in politics generally. In addition, to assist with the interpretations of regression coefficients, this paper proposes to calculate the marginal effects of the explanatory variables. The main objectives are to investigate the differences in the coefficients estimates and marginal effects of the regression models at various quantile levels. Comparisons will also be made to binary and ordinal probit models.