The transportation system is undergoing major technological and infrastructural changes, such as the introduction of autonomous vehicles, high speed rail, carsharing, ridesharing, flying cars, drones, and other app-driven on-demand services. While the changes are imminent, the impact on travel behavior is uncertain, as is the role of policy in shaping the future. Literature shows that even under the most optimistic scenarios, society’s environmental goals cannot be met by technology, operations, and energy system improvements only – behavior change is needed. Behavior change does not occur instantaneously, but is rather a gradual process that requires years and even generations to yield the desired outcomes. That is why we need to nudge and guide trends of travel behavior over time in this era of transformative mobility. We should focus on influencing long-range trends of travel behavior to be more sustainable and multimodal via effective policies and investment strategies. Hence, there is a need for developing policy analysis tools that focus on modeling the evolution of trends of travel behavior in response to upcoming transportation services and technologies. Over time, travel choices, attitudes, and social norms will result in changes in lifestyles and travel behavior. That is why understanding dynamic changes of lifestyles and behavior in this era of transformative mobility is central to modeling and influencing trends of travel behavior. Modeling behavioral dynamics and trends is key to assessing how policies and investment strategies can transform cities to provide a higher level of connectivity, attain significant reductions in congestion levels, encourage multimodality, improve economic and environmental health, and ensure equity.
This dissertation focuses on addressing limitations of activity-based travel demand models in capturing and predicting trends of travel behavior. Activity-based travel demand models are the commonly-used approach by metropolitan planning agencies to predict 20-30 year forecasts. These include traffic volumes, transit ridership, biking and walking market shares that are the result of large scale transportation investments and policy decisions. Currently, travel demand models are not equipped with a framework that predicts long-range trends in travel behavior for two main reasons. First, they do not entail a mechanism that projects membership and market share of new modes of transport into the future (Uber, autonomous vehicles, carsharing services, etc). Second, they lack a dynamic framework that could enable them to model and forecast changes in lifestyles and transport modality styles. Modeling the evolution and dynamic changes of behavior, modality styles and lifestyles in response to infrastructural and technological investments is key to understanding and predicting trends of travel behavior, car ownership levels, vehicle miles traveled (VMT), and travel mode choice. Hence, we need to integrate a methodological framework into current travel demand models to better understand and predict the impact of upcoming transportation services and technologies, which will be prevalent in 20-30 years.
The objectives of this dissertation are to model the dynamics of lifestyles and travel behavior through:
• Developing a disaggregate, dynamic discrete choice framework that models and predicts long-range trends of travel behavior, and accounts for upcoming technological and infrastructural changes.
• Testing the proposed framework to assess its methodological flexibility and robustness.
• Empirically highlighting the value of the framework to transportation policy and practice.
The proposed disaggregate, dynamic discrete choice framework in this dissertation addresses two key limitations of existing travel demand models, and in particular: (1) dynamic, disaggregate models of technology and service adoption, and (2) models that capture how lifestyles, preferences and transport modality styles evolve dynamically over time. This dissertation brings together theories and techniques from econometrics (discrete choice analysis), machine learning (hidden Markov models), statistical learning (Expectation Maximization algorithm), and the technology diffusion literature (adoption styles). Throughout this dissertation we develop, estimate, apply and test the building blocks of the proposed disaggregate, dynamic discrete choice framework. The two key developed components of the framework are defined below.
First, a discrete choice framework for modeling and forecasting the adoption and diffusion of new transportation services. A disaggregate technology adoption model was developed since models of this type can: (1) be integrated with current activity-based travel demand models; and (2) account for the spatial/network effect of the new technology to understand and quantify how the size of the network, governed by the new technology, influences the adoption behavior. We build on the formulation of discrete mixture models and specifically dynamic latent class choice models, which were integrated with a network effect model. We employed a confirmatory approach to estimate our latent class choice model based on findings from the technology diffusion literature that focus on defining distinct types of adopters such as innovator/early adopters and imitators. Latent class choice models allow for heterogeneity in the utility of adoption for the various market segments i.e. innovators/early adopters, imitators and non-adopters. We make use of revealed preference (RP) time series data from a one-way carsharing system in a major city in the United States to estimate model parameters. The data entails a complete set of member enrollment for the carsharing service for a time period of 2.5 years after being launched. Consistent with the technology diffusion literature, our model identifies three latent classes whose utility of adoption have a well-defined set of preferences that are statistically significant and behaviorally consistent. The technology adoption model predicts the probability that a certain individual will adopt the service at a certain time period, and is explained by social influences, network effect, socio-demographics and level-of-service attributes. Finally, the model was calibrated and then used to forecast adoption of the carsharing system for potential investment strategy scenarios. A couple of takeaways from the adoption forecasts were: (1) highest expected increase in the monthly number of adopters arises by establishing a relationship with a major technology firm and placing a new station/pod for the carsharing system outside that technology firm; and (2) no significant difference in the expected number of monthly adopters for the downtown region will exist between having a station or on-street parking.
The second component in the proposed framework entails modeling and forecasting the evolution of preferences, lifestyles and transport modality styles over time. Literature suggests that preferences, as denoted by taste parameters and consideration sets in the context of utility-maximizing behavior, may evolve over time in response to changes in demographic and situational variables, psychological, sociological and biological constructs, and available alternatives and their attributes. However, existing representations typically overlook the influence of past experiences on present preferences. This study develops, applies and tests a hidden Markov model with a discrete choice kernel to model and forecast the evolution of individual preferences and behaviors over long-range forecasting horizons. The hidden states denote different preferences, i.e. modes considered in the choice set and sensitivity to level-of-service attributes. The evolutionary path of those hidden states (preference states) is hypothesized to be a first-order Markov process such that an individual’s preferences during a particular time period are dependent on their preferences during the previous time period. The framework is applied to study the evolution of travel mode preferences, or modality styles, over time, in response to a major change in the public transportation system. We use longitudinal travel diary from Santiago, Chile. The dataset consists of four one-week pseudo travel diaries collected before and after the introduction of Transantiago, which was a complete redesign of the public transportation system in the city. Our model identifies four modality styles in the population, labeled as follows: drivers, bus users, bus-metro users, and auto-metro users. The modality styles differ in terms of the travel modes that they consider and their sensitivity to level-of-service attributes (travel time, travel cost, etc.). At the population level, there are significant shifts in the distribution of individuals across modality styles before and after the change in the system, but the distribution is relatively stable in the periods after the change. In general, the proportion of drivers, auto-metro users, and bus-metro users has increased, and the proportion of bus users has decreased. At the individual level, habit formation is found to impact transition probabilities across all modality styles; individuals are more likely to stay in the same modality style over successive time periods than transition to a different modality style. Finally, a comparison between the proposed dynamic framework and comparable static frameworks reveals differences in aggregate forecasts for different policy scenarios, demonstrating the value of the proposed framework for both individual and population-level policy analysis.
The aforementioned methodological frameworks comprise complex model formulation. This however comes at a cost in terms of prolonged computation and estimation times. Due to the non-convex nature of the objective function, direct maximization of the likelihood could become difficult and highly unstable. An alternative approach would be to use the Expectation Maximization (EM) algorithm instead of traditional gradient descent algorithms. This particular statistical learning technique is more stable, and requires fewer iterations to converge by taking advantage of the conditional independence structure of the model framework. This dissertation will provide rigorous derivation, formulation and application of the EM algorithm for mixture models and hidden Markov models with logit kernels, which constitute the building blocks of the generalized dynamic framework. Using such a statistical learning technique, i.e. the EM algorithm, model estimation time will be reduced from the order of many hours to minutes.
The line of work initiated throughout this dissertation is critical in this era of transformative mobility in terms of developing a generalized model that accounts for adoption styles and dynamic modality styles. The proposed dynamic, disaggregate discrete choice framework models the evolution of travel and activity behavior over time in addition to the adoption and diffusion of new transportation services. The proposed framework can be integrated with current travel demand models through the construct of adoption styles and modality styles, which shall provide a deeper understanding of behavioral dynamics and trends of travel behavior in an attempt to better inform long-range policy making. This dissertation provides the building blocks to advance the field of travel demand modeling in order to guide transformative mobility into the envisioned sharing economy future.
