Day-to-day route choice modeling incorporating inertial behavior

Accurate route choice modeling is one of the most important aspects when predicting the effects of transport policy and dynamic traffic management. Moreover, the effectiveness of intervention measures to a large extent depends on travelers’ response to the changes these measures cause. As a complement to utility maximization theory there is growing interest in alternative approaches which debate assumptions of perfect information, rationality and homogeneity. Empirical research on suboptimal choice behavior from the perspective of travel time has led to various theories including bounded rationality, habit, satisficing and perception bias. Although these theories may differ in a strict behavioral sense, there are various analytical similarities that make it impossible to empirically distinguish between them based on observed choice alone. However, one of the phenomena that can be observed in route choice is inertial behavior, i.e. the tendency of travelers to repeatedly choosing their current path, even if this path is or becomes the non-shortest path. Using observations from a real-world route choice experiment, a new day-to-day dynamic route choice model, consisting of a Dynamic Expected Shortest Path Module and a Choice Strategy Module was developed, which accounts for four identified choice strategies including inertia. The data contained actual route choice data for 5 OD-pairs that were collected for a panel of 20 subjects over 20 days during peak hours on regular working days. Experienced travel conditions were recorded using a GPS device while a pre- and post-task questionnaire was included, with information about demographics, experience, but also on perceptions about traffic conditions and preference levels. First, individual’s route choice behavior is investigated, contributing to the current understanding of the daily route choice behavior of individuals, and more particular, providing insights in and understanding of factors and mechanisms that contribute to inertial behavior. Therewith, it complements the current knowledge of route choice behavior. The new model was calibrated and validated using an enter-wise regression method and a jack-knife cross validation method, and compared with a number of existing model approaches: based on the shortest path, on prospect theory, regret theory, fixed thresholds theory and on the SILK theory. Additionally, the model was extended using an agent-based approach based on Bayesian simulation in order to see the effect of this approach on the model performance. Lastly, the indifference band is quantified by altering the model attribute related to travel time within the developed model. Besides this, the data analysis and the fixed threshold theory were used to quantify the indifference band for comparison. Results showed that individual characteristics and characteristics of the choice situations were most important in explaining exposed choice strategies, while variables on experience were found to be less important. The 2-step-model provided correct predictions in 75.35% of the cases and had one of the most constant performances when applied on different choice situations, which places it among the state-of-the-art models with the highest performances. When examining model performances of the new and state-of-the-art models for every OD-pair, it is found that some models perform better on certain OD-pairs than others and vice versa (some performed very low < 50%). This implies that in certain circumstances or choice situations a certain route choice model would be the best to use, and suggests that a hybrid model using different choice models for different choice situations could significantly improve current modeling practice. Additionally, it was found that the model performance reduced considerably when parameter correlations were ignored which highlights that explanatory variables of route choice behavior are strongly correlated and are therefore crucial in obtaining accurate model results. Lastly, inertia thresholds between 12.1% and 22.1% of the average trip travel time were found on an individual level, and 12.6% to 16.3% of the average trip travel time on the situational level (i.e. per OD-pair). These findings in conjunction with the observation that about one-third of the choices were inertial choices, offers indication of the extent individuals allow suboptimal choice situations, and as such the potential to realize system improvements. The data available for this research was limited in the number of subjects, choice situations and trips. It requires bigger and richer data to enable a much more thorough analysis of how people move through traffic networks, and when and how they access and react to change in the traffic system, as well as to validate the modelling results. When considering the issue of data for travel behaviour analysis, and in particular the analysis of travellers’ choices for choice alternatives, the behavioural phenomena explored in this research are generally hard or impossible to include in simulated or hypothetical environments. Specifically issues as perception error, habit or bounded rationality must preferably be studied in real-life environments, where actual trips are being made without experimenter instructions. Using new ICT technologies it is possible to overcome some of the deficiencies of revealed preference (RP) data as used in this study or stated preference (SP) data, and obtain bigger and richer data than was up till now the case. Future research will use Experience Sampling (ES) by means of a smartphone app to collect additional data for model refinement and validation. For each user of the app a mobility profile is collected, which enables a longitudinal analysis of travel behaviour. The ES-method combines SP with RP by asking a subject about its preferences in a certain situation (SP) directly after that particular situation has revealed itself (RP), thus the context where the stated preference is recorded is directly linked to a real life situation that has just occurred. ES is a powerful method for understanding a range of psychological phenomena, such as mood, behaviour, thoughts or feelings, as they occur in the daily lives of individuals. However, this method is not often used in travel behaviour research. In a later stage a multimodal route planner facility of the app will be used to analyse travellers’ choices and reasoning in response to specific information messages they receive. The paper elaborates on the experimental design.