Abstract
Activity-based models in transportation science focus on the description of human trips and activities. Modeling the spatial decision for so-called secondary activities is addressed in this paper. Given both home and work locations, where do individuals perform activities such as shopping and leisure? Simulation of these decisions using random utility models requires a full enumeration of possible outcomes. For large data sets, it becomes computationally unfeasible because of the combinatorial complexity. To overcome that limitation, a model is proposed in which agents have limited, accurate information about a small subset of the overall spatial environment. Agents are interconnected by a social network through which they can exchange information. This approach has several advantages compared with the explicit simulation of a standard random utility model: (a) it computes plausible choice sets in reasonable computing times, (b) it can be extended easily to integrate further empirical evidence about travel behavior, and (c) it provides a useful framework to study the propagation of any newly available information. This paper emphasizes the computational efficiency of the approach for real-world examples.