抄録
In transportation demand modeling, it is a common practice to utilize discrete choice models in estimating people's choice upon several modes of transportation based on observed socio-economic backgrounds and mode-specific features. The nature of the models, however, dictates the requirement of providing a substantial number of observations. An interesting question is whether it would still be possible to estimate a reasonably good model given only a small amount of data set available. One way to make this small number matters is through simulation. This paper reports an exercise of applying bootstrap technique to generate new data sets from a small number of observations and to estimate new models from them. From this simulation, distributions of models' parameters can be provided, and hence the parameters and their standard errors can be calculated from their respective distribution, and their biases can be corrected. The performance of this simulation-based model is then discussed.
