Monte Carlo Simulation of Sydney Household Travel Survey Data with Bayesian Updating using Different Local Sample Sizes

Abstract

There is increasing interest in the potential to simulate household travel survey data as an alternative to collecting large sample household travel surveys, or as a means to augment sample sizes, well beyond what can usually be considered. In prior research on simulating such data, it has been shown that it is possible to reproduce, within reasonable bounds of accuracy, an actual household travel survey, It has also been found that the procedure of updating the distributions of the simulated variables, using Bayesian updating with subjective priors, can provide significant improvement in the accuracy with which an actual household travel survey can be simulated. In work performed to date, it has not been determined what the optimal size would be for the update sample to be used in the Bayesian updating. Rather, prior work has used a sample of approximately 500 households, largely as a matter of convenience and cost. In this paper, we report on further research that compares different sample sizes for the local update data. It was found that a reasonable updating could be obtained from a sample as small as 300 households, chosen through a stratified sampling procedure, and that results improved substantially when the update sample was increased to 500. However, an increase in the sample to 750 did not produce very much additional improvement, suggesting that sample sizes of this size and larger may not be economically justified. At the same time, the research suggests that there may be room for a more targeted sampling procedure which could allow smaller samples to be more cost-effective.