An evidence based analysis of orthogonal and efficient experimental designs for transportation stated preference surveys

Abstract

Discrete choice models have had wide application for mode choice simulation as part of the multi-modal four-step transport demand modelling process in South Africa. For all models since 2000, urban commuter modal trip preference data for the mode choice sub-model has been collected using stated preference (SP) surveys. The SP experimental designs have been fractional factorial types (FFD). The SP data and the subsequent mode choice models were directly dependent on the soundness of the underlying experimental designs. This paper presents a review of the experimental designs of four South Africa metropolitan SP survey case studies to derive empirical evidence of the efficiency of these designs as measured by the D-error. The analysis showed there were several inadequacies in the experimental designs. Furthermore, the multinomial logit (MNL) models estimated from the SP data for each metro highlighted the shortcomings of the experimental designs. The paper illustrates that there is enough insight into prior estimates of the attribute parameters for use in efficient experimental designs and that these designs provide improved D-error measures over the equivalent orthogonal fractional factorial designs. The paper recommends that efficient designs be used in the future for all SP transport applications.