Dynamic Route Choice Modelling of the Effects of Travel Information using RP Data

Abstract

Traffic congestion is experienced by a great number of travellers during peak hours and is directly influenced by travel-related decisions, such as route choice decisions. In order to minimize problems that arise from congestion, such as delays, uncertainty and environmental effects, there has been an increased interest to investigate the role of travel information on travellers’ route choice behaviour. When travel information is provided, travellers have to decide whether to comply with the travel information. The complexity of such a decision increases, for instance, with travellers’ expectations about the traffic situation, the quality and reliability of travel information and route choice habits. A review of the literature shows that estimation of route choice utility models based on stated preference data is the most traditional way to investigate the relationship between travellers’ route choice behaviour and travel information. Stated preference experiments, however, are subject to the main drawback of external validity, i.e. it is arguable whether the experimental design and its results are valid and can be replicated in a real setting. Although revealed preference experiments are not subject to issues regarding external validity, there are three main modelling issues associated to the estimation of route choice utility models based on revealed preference data collected in real networks: choice set sampling is difficult, utilities may share unobserved attributes and the attributes of the alternatives (e.g. travel times) are in general unknown. This thesis has two main contributions: an empirical and a methodological. The empirical contribution is to investigate the role of travel information on travellers’ route choice behaviour based on revealed preference data collected in a real congested network. The methodological is to address for the first time issues regarding choice set sampling and attributes of the alternatives in a dynamic setting supported by real data about the traffic conditions in the network. A revealed preference experiment making use of GPS devices, travel diaries and interviews was conducted to investigate the route choice behaviour of 32 travellers for a period of 9 weeks in the Netherlands. Travellers were subject to different sources and timing of travel information provision. The target group that was investigated consisted of commuters living within 19 kilometres distance from work, who, according to the Dutch institute of mobility research, correspond to 60% of the 2 million car commuters in the Netherlands (Dutch MON database, 2008). This means that the investigated target group corresponds to the largest group of travellers driving during peak hours in the Netherlands. The sample resulted in 897 valid GPS traces and travel diaries, among which 374 refer to the first 3 weeks of data collection in which only free/public sources of travel information were available and 523 correspond to the subsequent 6 weeks of data collection in which personalized real-time travel information was also available. Moreover, data about the traffic conditions in the network was also collected for the whole study period at 1 minute intervals. The group subject only to free/public sources of travel information can be categorized as regular travellers. Otherwise, they are labelled informed travellers. Model estimation was done based on a new modelling framework denominated Recursive Logit Dynamic (RL Dyn), which is based on the sequential link-choice decisions of the Recursive Logit (RL) (Fosgerau et al., 2013). The RL can be consistently estimated with revealed preference data and has the advantage of not requiring of choice sets to be defined. Although both the RL and RL Dyn share the same properties, the latter is extended to a dynamic setting. In other words, while in the static formulation of the RL the network is defined in term of links and nodes, in the dynamic one it is defined in term of links, nodes and time intervals. This way, the time dimension is incorporated to the route choice problem and not only the route to chosen matters, but also the moment the route is chosen. This thesis shows the pertinence of using the RL Dyn for the proposed investigations. In particular, combination of the interviews conducted at the beginning of the experiment with the GPS traces indicate that travellers are not fully aware of the route choice set. Furthermore, the (predictively) validity of the RL Dyn and a substantially higher predictive power than its static counterpart (RL) were demonstrated. This indicates that adding the time dimension to the route choice problem compensates the higher complexity and computational demand for model estimation. Regarding the role of different sources of travel information, estimation results indicate that the more detailed the travel information is, the more travellers are sensitive to it. Both for pre-trip and en-route information, travellers who consult TomTom are significantly more sensitive to travel time than those who consult radio. Besides this, although estimation results indicate that road links with variable message signs (VMS) panels significantly influence route choice behaviour. It appears that consulting or not consulting the information displayed on them similarly influence travellers’ decisions. This suggests that the VMS panels are well located, but not much can be said regarding the use of travel information. Besides this, equipping travellers with TomTom devices increases the likelihood that they will use travel information to assist their route choices. With respect to the timing of providing travel information (i.e. pre-trip or en-route), estimation results suggest that travellers who only consult pre-trip information are significantly less sensitive to travel time than travellers who do not consult information at all. As pre-trip information is provided at the beginning of the trip, being aware of the traffic conditions may provide some sort of comfort to travellers. Besides this, outcomes of the quantitative and qualitative analyses suggest that finding out the best route choice considering the departure time is “now” appears to be more important than planning an “ideal” departure time to avoid congestion. As reported by the participants, they usually consulted travel information just when entering the car and not beforehand while still at home or in the office. Regarding compliance with travel information, outcomes of the quantitative and qualitative analyses show that (in general) the more detailed the source of the travel information is, the higher the compliance rates are. Besides this, compliance rates appear to differ for informed and regular travellers: informed travellers are more willing to comply with travel information than regular travellers. In addition, travellers are more willing to comply with pre-trip, rather than en-route, information. Despite the potential effects of travel information, the impact on the route choice behaviour of commuters is smaller than expected. Although informed travellers are about 30% more sensitive to travel time than regular travellers, the importance of habit to commuters’ route choice decisions is very strong. Habit turned out to be a highly significant parameter in all model specifications tested. In particular, even when incurring delays longer than expected, travellers tend to stick to their preferred planned routes. This suggests that in order to change the behaviour of commuters either the penalty associated to a late arrival or the benefit for on time/earlier arrival should be or perceived as higher. Based on these findings the (overall) conclusion can be drawn that the role of travel information to alleviate congestion, and the consequent effects on network performance, resultant from commuting trips is limited. This means that it is not expected that travellers will change their route choices due to provision of travel information unless one is dealing with extreme conditions. Similar outcomes are expected for other situations in which commuters living within 19 kilometres from work are investigated in a real setting under regular traffic conditions.