Travel time reliability and the bimodal travel time distribution for an arterial road

In relation to the development of travel time reliability metrics, previous studies had suggested that the travel time variability distribution might follow either normal or lognormal forms. From substantial new data observation undertaken in Adelaide, including assessment of two sets of longitudinal trip time data, these distributions appear inadequate. Generally this is because the upper tails of the observed distributions are more substantial than those of the normal or lognormal distributions. Additionally, there is some evidence of bimodality in some of the actual travel time distributions. Consequently, a number of questions arise: When does bimodality occur? How does it affect the measurement of travel time variability and reliability? Hence, this paper investigates the existence of bimodality in travel time distributions and tries to answer those questions. From the data it was found that the bimodality occurred when travel time observations for a given link actually belong to two different populations. This occurs on links with traffic signals, in circumstances in which the traveller experiences two distinct modes of traffic behaviour: (1) when there is no delay at the traffic signals, and (2) when the traveller encounters a long delay. Bimodality is known to occur in other fields, such as the life test process in reliability engineering. The paper uses current research to measure the probability of the separate populations. Using the mixture normal distribution functions, we can then refine and extend the previous travel time reliability metrics.