Modeling discretionary activity location choice using detour factors and sampling of alternatives for mixed logit models

Excessively large choice sets have been identified as an important issue influencing the prediction accuracy of individuals' activity location models. In this work, a constrained choice modeling approach with sampling of alternatives is applied to analyze an individual's location choice for discretionary activities. The issue of large choice sets is tackled through two constraining methods: (i) adequacy of destination, in which only locations which are suited to a certain type of activity are selected and (ii) use of a delimitation rule depending on the type of trip chain to which the discretionary activity belongs. A mixed logit model with sampling of alternatives is specified to estimate individuals' location choice for different types of discretionary activities. The estimation results show sampling alternatives using an individual's constrained choice set based on both adequacy destination and detour provides significantly better prediction accuracy compared to that using only adequacy of destination. We conducted several experiments with respect to constraining methods, number of sampled alternatives and bias-correcting methods for sampling of alternatives in a mixed logit model. The results show that the Naïve method for sampling of alternatives in a mixed logit model provides better goodness-of-fit than estimation with correction terms.