QUASI-RANDOM SIMULATION OF DISCRETE CHOICE MODELS

This paper examines ways to perform simulation of travel behavior with less computer time for a given level of accuracy, or greater accuracy for a given amount of computer time. The properties of two quasi-random draws, (t,m,s)-nets and Halton draws, are described. Four types of (t,m,s)-nets, two types of Halton draws, and independent draws are compared in an application of maximum simulated likelihood estimation for a mixed logit model. All of the quasi-random procedures are found to perform far better than independent draws. The best performance is attained by one of the (t,m,s)-nets. The properties of the nets theoretically imply that two of them should outperform the other two, and the results confirm this expectation. The two more-accurate nets outperformed both types of Halton draws, while the two less-accurate nets performed worse than the Halton draws. Since the coverage properties of the Halton draws and (t,m,s)-nets cannot be directly compared, it is not possible to determine beforehand whether a given (t,m,s)-net will perform better or worse than Halton draws