Empirical identification in the mixed logit model: Analysing the effect of data richness