Route choice is one of the most difficult travel choice dimensions to analyze and to interpret. Moreover, when facing a route choice situation, road users are influenced by a lot of unobserved factors such as habit, inertia and personal preferences which lead to a choice that might be different from the expected “shortest” one when looking with an objectivity lens. This is crucial for the effectiveness of using ITS (Intelligent Transport System) or ATMS (Advanced Traffic Management System) technologies, which need high precision data to provide route information to the users. Thus, a lot of researchers have explored in depth route choice behavior in order to understand its key determinants. In this study we focus on route switching behavior, namely the analysis of whether road users, moving between the same OD pair, use the same route or change depending on the level of satisfaction of several elements that are not directly known to the researcher. The aim of this paper is to understand which determinants lead road users to switch between different routes during a period of two weeks. Several past researches faced on the route switching behavior, and can be divided into two main groups: the first group used GPS data to improve the knowledge about the attributes which affect the route choice behavior, comparing the observed routes with the simulated ones (Abdel-Aty, et al., 1994; Li, et al., 2005; Spissu, et al., 2011; Arifin & Axhausen, 2011; Levinson & Zhu, 2013); the second group used data from questionnaires and laboratory experiments to understand the effect of the information provision on the route choice behavior (Khattak, et al., 1995; Mannering & Kim, 1994; Polydoropoulou, et al., 1996; Srinivasan & Mahmassani, 2003; Jou, 2004; Xu, et al., 2010; Ben-Elia & Shiftan, 2010). However, limitations can be recognized in previous literature: (i) the route switching behavior was analyzed only in relation to the provision of external inputs (like route information); (ii) none of the studies used high resolution data in modeling applications.
The objective of this paper is to analyze in depth route switching behavior from a pure behavioral perspective, focusing only on the actual choices of the road users. Indeed, while the previous works were based on understanding a response to a specific input (route information provision), in this research we want to explore route switching in a normal day by day case, considering the inter and intra personal factors that the road users actually consider when making a route choice. So we can merge the two research fields which already approached route switching and close the gap between them in order to study route switching behavior without external stimuli and high resolution data. This is important because this kind of data provide very precise information about the chosen routes (with low respondent’s burden) and also the users are not forced to think about the answer and do not perceive any lack of realism on the choice situation (which could happen in case of providing questionnaires or other techniques), so their answer is not influenced from other factors which can introduce bias in the data. To this end, we used data acquired during a survey, named "Casteddu Mobility Styles” (CMS), conducted by the University of Cagliari (Italy) in the metropolitan area of Cagliari between February 2011 and June 2012. Each participant was asked to carry a smartphone with built-in GPS in which an application called “Activity Locator” – implemented by CRiMM (Centre for Research on Mobility and Modeling) – had been installed (Meloni, et al., 2011). A total of 8831 trips were recorded by 109 individuals, of which 4791 referring to the car driver mode. Each GPS track (consisting of a sequence of referenced position points) was then treated with map-matching techniques that allowed associating each “GPS point” to a link of the network, thus creating the observed route database. Because of the density of the network where the map-matching was performed, as well as the precision of the registered GPS points, it happens that a starting place of a trip is not ever associated to the same node of the network; so, two trips which start from the same location could have different starting nodes, although they are pretty close. This affects the analysis, because becomes harder to identify analytically if two trips have the same origin or destination and, then, to build the database to perform the modeling analysis. In order to avoid this problem, the starting and ending nodes were firstly merged in clusters, so to make it possible to identify trips with the same origin-destination pair. Several values of the cluster’s radius were tested to find the best in terms of data quality, aggregation and precision. The reference value is 100 meters for both the origins and the destinations, which consist of a total of 864 trips related to 77 users and 187 origin/destination couples. However, also other different sizes of the clusters will be tested in order to verify if the dimension affects the results of the analysis.
We tested the effects of several variables, main related to three categories: objective characteristics of the route (like travel time, distance, percent of congested roads, number of traffic lights, number of turns, roads’ category), socio-economic information of the users (like gender, age, driving experience, household’s characteristics, monthly expenditure for transport) and activity-based data (like departure time of the trip, motivation of the trip, day of the week, pre-trip activity, place of departure/arrival of the trip). Binary logit models were tested in order to understand what is the contribution of each determinant in leading the users to switch with respect to the previous choice for the same OD pair. Because of the characteristics of the data, binary mixed logit models were also estimated in order to better understand the behavior of the users, taking also into account of the panel effect. Results suggest that road users will switch to routes which are shorter or after encountering high friction factors along the previous route. Differences between commute and non commute trips can be found, because it’s well known that the former are more influenced by the experience and the knowledge of the network.