Isolating Different Factors Affecting Travel Time Reliability in an Observational Before/After Study

AbstractIt is increasingly recognized that travel time reliability is important to travellers and goods, and hence there is an increasing demand to include reliability in the evaluation and appraisal of transport projects and programs, as a separate factor in its own right. A number of the policy options already in use as congestion mitigation policies have also positive impacts on the travel time reliability. It is therefore important to evaluate impacts of any remedial action separately both for average travel time (congestion) and for variability of that travel time (reliability). The impacts are generally assessed with an observational before/after study. As many things, independent from the system to be assessed, change between the periods before and after the installation of the system, observational before/after studies take also into account a reference site, where the effect of these changes is identified.In a previous paper Bhouri and Aron, (2013) presented the reliability impacts of a managed lane case consisting in a Dynamic Hard Shoulder Running (HSR) during rush hours on a French urban motorway. The study was conducted by comparing reliability indicators based on data collected in 2002 before the implementation of the system and in 2006 after this implementation. Jacques Chirac, former president of France, launched in 2003 an important campaign for road safety and against speeding. This modified the travel time distribution, differently according to the lane. As reliability is based on the percentiles (and not only on averages), it was necessary, in order not to confound the effect of this campaign with the HSR effect, to develop the observational before/after study method for determining that time effect by level of travel time: to do this, the before and after travel time distributions of the reference site are established, then their difference is analyzed as a function of the travel time, taking into account the percentiles. This method and the results are presented in this paper