Determining Effective Sample Size to Calibrate a Transit Assignment Model: A Bayesian Perspective
place - australasia, planning - methods, technology - passenger information, ridership - forecasting
transit assignment model, passenger loads
A transit assignment model is used to predict passenger loads in order to evaluate existing and future transit network scenarios. One fundamental issue affecting the calibration of a transit assignment model is the amount of data required. The present study is designed to determine the effect of different sample sizes on the accuracy level of the estimated passenger flow. A Bayesian model is adapted for transit assignment, and the sample size for three types of priors, namely: uninformative, informative, and overly informative, are examined. In order to assess the value of prior information on passenger flow, the root-mean-square error (RMSE) between each posterior estimate and the actual observation is computed. The posterior estimate that minimizes the %RMSE defines the effective sample size (ESS). This paper uses one day’s automatic fare collection data from the South East Queensland (Australia) transit network to evaluate the effect of sample size and prior information on the posterior passenger flow estimates. The results show that it is not possible to determine the ESS for the Bayesian model with an uninformative prior. With an informative prior, the ESS is 50% of the population and, for the model with an overly informative prior, the ESS is 10% of the population. This means that the lack of prior information cannot simply be compensated by increasing the sample size in this Bayesian model. However, good prior information reduces the necessary sample size substantially.
Permission to publish the abstract has been given by SAGE, copyright remains with them.
Rahbar, M., Hickman, M., Mesbah, M., & Tavassoli, A. (2018). Determining Effective Sample Size to Calibrate a Transit Assignment Model: A Bayesian Perspective. Transportation Research Record. https://doi.org/10.1177/0361198118781182