Robust discrete choice models with t-distributed kernel errors

摘要: Models that are robust to aberrant choice behaviour have received limited attention in discrete analysis. In this paper, we analyse two alternatives the multinomial probit (MNP) model. Both alternative models belong family of robit models, whose kernel error distributions heavy-tailed t-distributions. The first model is (MNR) which a generic degrees freedom parameter controls heavy-tailedness distribution. second alternative, generalised (Gen-MNR) model, has not been studied literature before and more flexible than MNR, as it allows for alternative-specific marginal For both devise scalable gradient-free Bayes estimators. We compare MNP, MNR Gen-MNR simulation study case on transport mode behaviour. find deliver significantly better in-sample fit out-of-sample predictive accuracy MNP. outperforms due its Also, gives reasonable elasticity estimates MNP particular regarding demand under-represented class-imbalanced dataset.