Keywords: Discrete Choice, MCMC, Hierarchical Bayes, Unbalanced Data, Statistical Methodology and Artificial Intelligence, Machine Learning, Big Data.
The paper introduces a new functional form for modeling discrete consumer choices which can flexibly fit any dataset based on the amount of asymmetry (conversely symmetry) in the data, yielding results identical to traditional symmetric models when the data warrants, yet giving superior model fits, prediction and inference results to existing methodologies when the data diverge from traditional model assumptions both in and out-of-sample. Furthermore, a new Hierarchical Bayesian estimation methodology is introduced, which is seen to be superior to existing models, with very few MCMC iterations, because of the flexibility of the functional form, in the presence of unbalanced or poorly sampled data. Together they present an integrated framework which extends the current discrete choice modeling literature in the social sciences, as well as, Statistical and Econometric methodology for both traditional as well as Big Data contexts.