On the efficiency of Mixed Logit parameters estimates: analysing the effect of data richness