We are in an era of rapid urbanization, technological advances, transportation transformation, and increasingly big data and computation power. We have witnessed how shared transportation (Uber, Lyft, Lime, Bird, etc) intrudes into daily lives in just a few years, how online shopping, including same-day grocery delivery, has changed day-to-day travel trajectories, and how the emerging work-from-home lifestyle would fundamentally change people's location choices. At the same time, large-scale data becomes more accessible than ever; so does the computation power needed to process such data. It is therefore a good time to retrospect existing paradigms of dynamic behavior models and keep exploring the potentials and new opportunities.

While studies on long-term travel behavior, such as residential location choice and working location choice, have been the emphasis of a substantial body of prior work, most empirical studies adopt a static approach to behavior modeling. For the small body of work that allows for dynamic behavior modeling, only backward-looking behavior, i.e., time-dependency, is incorporated, and the role of forward-looking behavior, i.e. by considering future expectations in sequential decision-making, has long been neglected. This is with good reasons: the estimation of a truly dynamic choice model is extraordinarily difficult due to (a) computational tractability associated with big data and large-scale dynamic programming to accommodate forward-looking and (b) scarcity of longitudinal data on long-term travel behaviors, such as residential moving trajectories. Yet long-term travel behavior is inherently dynamic, and this has led to concerns that estimates from static models may be biased.

In economics, dynamic discrete choice models (DDCM) have been used to model many aspects of transportation behaviors; however, this approach has several limitations, including its assumptions of optimal human behavior, conditional independence, extreme value distribution, etc. In the recent decade, advances in artificial intelligence, especially in inverse reinforcement learning (IRL), have inspired new approaches to solving complex dynamic behavioral problems. In particular, IRL can circumvent several assumptions common in DDCM, while still reconstructing problems and estimating models in a tractable way. However, the research worlds of economics and of artificial intelligence rarely reference each other; one objective of this dissertation is to bridge these two disciplines to address the challenging problem of modeling large-scale long-term forward-looking travel behavior.

We do not necessarily need the forward-looking assumption in all situations. In practice, for short-term and medium-term behavior trajectories, such as mode choice, car usage and car ownership, the backward-looking assumption can be sufficient. This is because these choices are associated with much lower costs both financially and psychologically and the impact of future expectations can be trivial. There is a rich amount of literature on backward-looking dynamic models, including studies on identifying policy and environment triggers that shift travel behavior, studies that investigate the role of key life events on travel behavior change, and studies on lifestyle analysis which treat lifestyle transition as a higher-level orientation of behavioral change. However, most frameworks on backward-looking dynamic modeling concentrate on analysis of single-dimensional choice and ignore the interdependence and multi-dimensionality of travel behaviors. Furthermore, few prior work consolidates all these dynamics components in a single framework to analyze the joint effect of different sources of triggers. Therefore, another objective of this dissertation is to develop a unified modeling framework that accounts for time-varying economic and policy context (external dynamics), life events (internal dynamics), lifestyle, and multi-dimensional interrelated choices.

The first component of this dissertation formulates a mathematical framework for representing long-term travel behaviors as sequential actions under the Inverse Reinforcement Learning (IRL) framework, which aims to address the forward-looking limitation. In the proposed framework, the individual observes the environment and takes action (i.e., move to a new location or not) accordingly by evaluating action-dependent future rewards received from the environment. The reward can be a function of built-in environment attributes, which shares a similar concept with the utility function in discrete choice models. Three highlights of the first component of this work are presented below.- In the classic IRL setting within the domain of artificial intelligence, the agents (usually robots) are often assumed to have homogeneous behavior and do not own any internal dynamics associated with the agents. Our work extends the IRL framework to accommodate heterogeneous household behavioral dynamics, and derives its corresponding learning algorithm to estimate the parameters associated with the attributes. - We provide an in-depth theoretical comparison between Dynamic Discrete Choice Model (DDCM) in economics and IRL in artificial intelligence from different aspects, including terminologies, assumptions, and model structures. - To validate the existence of forward-looking behavior and the methodological feasibility of the proposed framework, we use a large-scale infused data set of household relocation trajectories in Texas over a 7-year period (2005-2011). - The empirical results are three-fold. First, all households have a positive preference to locate in areas with higher degree of land-use mix, higher accessibility to jobs, and lower employment density. Our model also shows that low-income households focus more on current needs and are less forward-looking compared with households with higher income level. And low-income households present less willingness to pay for neighborhood amenities such as land-use mix and accessibility to jobs. In terms of goodness of fit, our proposed model outperforms the DDCM model (for high-income and low-income households), backward-looking model and the static model.

The second component of this work addresses the limitation on backward-looking models. The HMM framework has gained increasing attention in the transportation arena (in applications from car ownership to mode choice) due to its latent hierarchical structure, favorable model performance, and intuitive interpretation. Highlights of this component are as follows. - We extend the framework of heterogeneous Hidden Markov Model (HMM) from single-dimensional discrete choice to multi-dimensional discrete and continuous choices, and derive its recursive parameter learning algorithm. Building on this framework, we propose a unified model that conjoins lifestyle, life events, external environment, and multi-dimensional travel behavior dynamics. - We evaluate the feasibility and robustness of the proposed methodology via a case study: a retrospective survey in the San Francisco Bay Area consisting of 830 households. - To fully explore the potentials of the proposed framework, we provide trend analysis of car ownership and mode use based on estimation results, and conduct sensitivity analysis of changes in fuel price and unemployment rate. - We identify four latent lifestyles: auto-oriented-2-car group with rare use of other travel modes, auto-oriented-1-car group with rare use of other travel modes, multi-modals group that own at least one car, and auto-free group that have the lowest car ownership and car usage. The results demonstrate how life events, policies, and the economic environment influence people on lifestyle transitions.

In sum, this dissertation provides building blocks to evolve the field of dynamic behavior modeling by incorporating advances in artificial intelligence. Throughout this dissertation, when providing theoretical improvements building on each mathematical framework, we ground each methodology with case studies. Empirical results have shown our methodologies can effectively help quantify the triggers that prompt individuals and households to change their travel behavior, better predict the trend of future mobility, and help transportation planning and policy-making.

The field of transportation and travel behavior research has long been interested in answering causal questions. Take the recent COVID-19 pandemic as an example: the transportation sector in particular was one of the most heavily impacted sectors, and transportation researchers found themselves with a plethora of questions to answer regarding the current and future impacts of the pandemic on the transportation system. Yet, out of roughly 250 pandemic related transportation research papers we reviewed, only about 10 explicitly reference the causal inference literature and are explicit about their causal designs, despite the fact that a significantly larger portion of those papers are trying to answer causal queries. This disconnect is the motivation behind this dissertation.

It is important to acknowledge that transportation researchers have indeed used and contributed to several areas of the causal inference literature. Notable contributions include addressing self-selection bias between residential choice and travel behavior, addressing omitted variable bias through integrated choice and latent variable models (ICLVs), and addressing endogeneity between multiple travel outcomes using joint discrete choice models.

Despite those contributions, many advances in the causal inference literature have yet to enter the field of travel demand modeling. There are many reasons for this disconnect, some of which stem from long-rooted beliefs and practices within the travel behavior and demand modeling literature on model development, selection, and validation. For starters, there is a tendency for discrete choice modelers in transportation to assume their models allow for causal interpretations because they rely on a behavioral theory of human decision making, like variations of random utility theory. While this criterion typically entails endogeneity checks between the outcome and the regressors, it overlooks important nuances about the data generating process and sources of variation in the exogenous variables. This is further exacerbated by the heavy reliance in the field of transportation demand modeling on goodness-of-fit statistics and statistical tests of significance when finalizing modeling specifications, making them prone to the issue of “bad controls”. For instance, adding post-treatment or mediator variables that are exogenous to the outcome but endogenous to the treatment variable of interest, undermines the causal interpretation of the model coefficients, even if adding those variables results in improved predictive model performance and goodness-of-fit statistics. Finally, causal identification strategies rarely appear in the transportation literature. In the causal inference literature, the analyst states the assumptions before drawing any causal conclusions from the model by explicitly specifying the source of variation in the treatment of interest. Such assumptions are referred to as identification strategies: they are assumptions about the data generating process that, only if true, allow the modeler to interpret the model parameters as causal. Those strategies are rarely explicitly stated in transportation demand models, even though those models are often used to evaluate the impact of policy interventions.

To address this gap, this dissertation comprises two parts: 1) a conceptual part, and 2) an applied part.

The conceptual part consists of Chapter 1, where I elaborate in detail on the disconnect described above and point to specific examples from the transportation literature where such misconceptions about causality are most evident. Next, I provide a review of the key concepts in causal inference, and give an overview of the main causal identification strategies used in the empirical sections of this dissertation. I also give an overview of causal graphical models, an alternative causal inference framework to the more well-known potential outcomes (PO) framework which has been gaining popularity in recent years. I focus the overview on parts where I believe this framework, and Directed Acyclic Graphs (DAGs) more specifically, are most useful to transportation researchers.

The applied part consists of Chapters 2, 3 and 4, where I apply some of the causal identification strategies presented in Chapter 1 to answer three different empirical causal research questions in transportation, two of which rely on observational data, while one involves randomized experiments. Each chapter results in domain-specific empirical contributions in its respective area. The common theme across all three chapters is the explicit focus on estimating causal parameters, the clear statement of the causal identification strategies used to estimate those parameters, and the transparency about the source of variation in the treatment. This is in contrast to the common practice in travel behavior modeling where models are specified and estimated without explicitly stating the assumptions under which the estimated parameters can have causal interpretations, and can often lead to erroneous conclusions and misapplications of those models.

In Chapter 2, I quantify the causal effect of telecommuting on travel frequency and distance traveled. This question is motivated by the unprecedented rise of telecommuting in the past two years and its likely persistence in a post-pandemic world. To answer this question, I collected and used five waves of original U.S.-based survey data combined with passive smartphone Point-of-Interest (POI) data collected over the course of the pandemic. I quantify the effects of changes in the frequency of telecommuting on the total number of daily and weekly trips that a telecommuter makes, as well as their total daily and weekly distance traveled. Crucially, I overcome important limitations of related work in the literature by controlling for unobserved individual confounders, a limitation of existing research that predominantly relies on cross-sectional observational data. I do this by leveraging the longitudinal aspect of the data and implementing causal quasi-experimental designs like first-differences and two-way fixed effects regressions with individual, time, and individual by time controls. I find strong evidence that telecommuting causes the generation of new non-commute trips. Specifically, I find that individuals make an average of 1 additional non-commute trip on telecommuting days relative to commute days. This trip is on average shorter than the two-way commute trips, meaning that the net effect of telecommuting on total distance traveled is negative. Importantly, I find that this relationship persists at the weekly level, where I estimate that 1 additional day of telecommuting per week causes an increase of about 1 additional non-commute trip. This means that the additional travel on telecommuting days is the result of a newly generated trip, not a trip that has been shifted from other days of the week. The results are more robust than those in the existing literature and suggest that the trip reduction effects of telecommuting could be overestimated if telecommuting-induced new trip generation is not properly accounted for.

In Chapter 3, I quantify the causal effect of vaccines on reversing pandemic induced mobility trends. The question is motivated by the extensive literature analyzing and forecasting the stickiness of pandemic-induced behavioral and mobility trends in a post-pandemic world. Using the same data as the one used in Chapter 2, I show how the pandemic behavioral response of people in the U.S. was heterogeneous: individuals with low levels of concern about being infected with COVID-19 engaged in riskier behaviors than those with higher levels of concern, including traveling more, attending large gatherings, and using public transportation. Then, using difference-in-differences designs, I show how getting vaccinated affected those behavioral differences. Specifically, I find that getting vaccinated caused an increase in mobility, with vaccinated individuals increasing their number of weekly trips by 4.8 trips per week after getting vaccinated, compared to 1.8 trips for the unvaccinated during the same time period. The difference-in-differences estimate is 3 trips per week, or 170% of the increase in trips for the unvaccinated. The collective results provide important insights on human and travel behavior during the pandemic impact and recovery periods, and how vaccines affect those behaviors.

The third and final empirical question is answered in Chapter 4, and is on learning and optimizing user behavior at plug-in electric vehicle (PEV) charging stations. Research on consumer behavior in the PEV charging context is limited, data is lacking, and consumer preferences, especially at workplace charging stations, remain poorly understood. I address this gap by designing and implementing randomized pricing experiments, the gold standard of causal inference, and quantify key behavioral quantities like the willingness and required incentives to delay charging, and the relationship between plug-in duration and hourly prices. I then propose a novel optimization framework that incorporates those learned behaviors into the optimization objective and significantly improves the operational efficiency of PEV charging stations. My analysis shows that incorporating behavioral theory in the optimization framework results in significantly lower operational costs (up to 17%) and higher net revenues (up to 50%) for the charging station operator compared to the uncontrolled baseline, without sacrificing the user experience.

Aside from the conceptual contributions (Chapter 1) and domain-specific empirical contributions (Chapters 2, 3, and 4), the dissertation also makes data contributions through the collection of two original datasets that will soon be made publicly available to the transportation research community. I have participated in those data collection efforts playing a primary role in designing the surveys, defining the sample, and implementing randomized pricing experiments. The two datasets are:An extensive dataset on a panel of U.S. participants with a comprehensive set of behavioral questions over the course of the pandemic. The data comprises passively collected POI data, as well as five waves of surveys which included questions regarding respondents’ employment status, travel and telecommuting behavior, vaccination status, demographic characteristics, and ideological beliefs. An anonymized and de-identified version of this dataset, along with aggregated mobility metrics from the POI data will be made publicly available for transportation researchers. An experimental dataset of user behavior at PEV charging stations that includes exogenous variation in the prices and the resulting real life behavioral responses of study participants. This dataset fills a critical gap in the human behavior literature in PEV charging research, where most pricing data are observational and do not allow for price perturbations. The data will be made publicly available for academic researchers along with an accompanying data manuscript that describes it.

Collectively, this dissertation emphasizes the importance of being rigorous about the design and assumptions under which researchers and modelers can draw causal conclusions from models, an important task in transportation research. After all, causal inferences are only valid if their underlying assumptions are correct, yet those assumptions are rarely discussed and formally stated in transportation models. This dissertation fills this gap.

Latent modal preferences, or modality styles, are defined as behavioral predispositions towards a certain travel mode or set of travel modes that an individual habitually uses. They are reflective of higher-level orientations, or lifestyles, that are hypothesized to influence all dimensions of an individual's travel and activity behavior. For example, in the context of travel mode choice different modality styles may be characterized by the set of travel modes that an individual might consider when deciding how to travel, her sensitivity, or lack thereof, to different level-of-service attributes of the transportation (and land use) system when making that decision, and the socioeconomic characteristics that predispose her one way or another. Travel demand models currently in practice assume that individuals are aware of the full range of alternatives at their disposal, and that a conscious choice is made based on a tradeoff between perceived costs and benefits associated with alternative attributes. Heterogeneity in the choice process is typically represented as systematic taste variation or random taste variation to incorporate both observable and unobservable differences in sensitivity to alternative attributes. Though such a representation is convenient from the standpoint of model estimation, it overlooks the effects of inertia, incomplete information and indifference that are reflective of more profound individual variations in lifestyles built around the use of different travel modes and their concurrent influence on all dimensions of individual and household travel and activity behavior.

The objectives of this dissertation are three-fold: (1) to develop a travel demand model framework that captures the influence of modality styles on multiple dimensions of individual and household travel and activity behavior; (2) to test that the framework is both methodologically flexible and empirically robust; and (3) to demonstrate the value of the framework to transportation policy and practice.

In developing an appropriate framework, the dissertation builds on Latent Class Choice Models (LCCMs) used previously in the literature, synthesizing recent advances in the sub-domains of taste heterogeneity and choice set generation, and contributing methodologically to the sub-domains of preference endogeneity and simultaneous choice models. With regards to preference endogeneity, discrete choice models in the past have usually subscribed to the neoclassical assumption that preferences, as denoted by taste parameters and choice sets, are characteristics of the decision-maker that are exogenous to the choice situation and stable over time. Such a representation would be adequate if an individual's modality style were expected to be invariant across time. However, modality styles are subject to external influences, such as changes to the transportation system. For example, one would expect the introduction of transportation policies such as the London Congestion Charge or the implementation of infrastructural initiatives such as Bogota's Transmilenio bus rapid transit system to lead to changes in modality styles, and an apposite framework should be able to model and predict these changes. With regards to simultaneous choice models, discrete choice models in the past have introduced correlation across multiple dimensions through the covariance structure of the utility specification corresponding to each of the dimensions. Though such an approach has been shown to result in a significant improvement in fit, the covariance structure is a black box that it does not offer any insight to the underlying source of correlation. We introduce correlation through the modality styles construct, conditioning multiple dimensions of individual and household travel and activity behavior on a single overarching modality style, and thereby offering a behavioral rationale to the underlying source of correlation.

The proposed framework has the following structure: modality styles are specified as latent classes. Heterogeneity across modality styles is captured by allowing taste parameters and choice sets corresponding to the class-specific choice models to vary across classes. Preferences are endogenized by defining class membership as a function not only of the characteristics of the decision-maker, as is standard practice, but also of the consumer surplus offered by each class, which in turn is a function of alternative attributes, taste parameters and choice sets. Choices across multiple dimensions are correlated by conditioning the class-specific choice models for all dimensions of interest on the class membership model.

We apply the framework to study the relationship between individual modality styles and travel mode choice behavior using two very distinct travel diary datasets from two very culturally and geographically distinct regions. The first dataset was collected in Karlsruhe, Germany and comprises a relatively small sample of 119 individuals surveyed over a fairly long observation period of six weeks. Estimation results indicate the presence of habitual drivers who display a strong bias for using the automobile and multimodal individuals who exhibit variation in their modal preferences. Multimodal behavior is further distinguished by those who appear to be sensitive to travel times and those who appear to be insensitive. The second dataset was collected in the San Francisco Bay Area in the United States and consists of a relatively large sample of 26,350 individuals surveyed over a fairly short observation period of two days. Estimation results uncover six modality styles that are distinguishable from one another by the kinds of individuals that belong to each of the six modality styles, their latent preferences for different travel modes and the relative importance that they attach to different level-of-service attributes of each of the travel modes. For example, two of the six modality styles comprising 30% of the sample population only consider the car when deciding how to travel. These two modality styles, labeled inveterate drivers and car commuters, can further be distinguished from one another by their value of travel time. Inveterate drivers have a very low value of in-vehicle travel time of 0.55 $/hr for mandatory tours and are insensitive to in-vehicle travel times for non-mandatory tours. Car commuters have a value of in-vehicle travel time of 6.95 $/hr for mandatory tours and are insensitive to travel costs for non-mandatory tours, indicating a very high value of in-vehicle travel time for the same. Consistent with findings in the social sciences and multiple streams within economics that have shown preferences to be endogenous, the case study shows that a decision-maker's value of time is sensitive to the level-of-service of the transportation system, and an increase in overall travel times can induce decision-makers to lower their value of time.

The framework is subsequently adapted to study the evolution and persistence of modality styles and travel mode choice behavior in a dynamic context. Individual modality styles are still represented as latent classes, but an individual is allowed to have different modality styles at different time periods. The evolutionary path is hypothesized to be a Markov process such that an individual's modality style in the current time period is dependent only on her modality style in the previous time period. As before, travel mode choices for a particular time period are conditioned on the individual's modality style for that time period. The framework is empirically tested using travel diary data collected in Santiago, Chile. The dataset comprises a sample of 220 individuals surveyed over four one-week periods that span a time period of twenty-two months that includes the introduction of Transantiago, a complete redesign of the city's public transit system. Estimation results identify three modality styles: unimodal auto users who only consider the automobile, unimodal transit users who only consider the public transit system and have a low value of time, and multimodal users who consider all travel modes and have a high value of time. The case study further finds that the distribution of individuals across modality styles is highly sensitive to a shock to the transportation system such as that represented by the introduction of Transantiago. Results from a sample enumeration show that nearly a quarter of the sample population changed its modality style post-Transantiago.

For all three datasets, estimation results find that modality styles are strongly correlated with more long-term travel and activity decisions, such as level of vehicle ownership and residential location. In examining the influence of individual modality styles on travel mode choice behavior, the model framework for both the static and the dynamic context took one or more of these decisions as exogenous inputs. However, such a causal representation risks endogeneity, leading us to reverse the representation and include these dimensions explicitly as dependent variables. In doing so, we recognize that dimensions such as level of vehicle ownership represent decisions that are not made by individuals in isolation from other members of the household. An individual's preferences and choices are strongly shaped by the opinions and behaviors of the people around her, particularly when a choice is made collectively by a group of individuals, as in the case of a household. Therefore, interaction between household members must be understood to influence, among other things, individual modality styles. To reflect this influence, we introduce the household modality styles construct, characterized by the modality styles of the respective individuals that make up the household. We build upon the LCCM framework described previously, replacing the individual modality styles construct with the household modality styles construct and conditioning both individual and household level dimensions on the household's modality style, therefore introducing correlation between preferences of the individuals that constitute the household.

The framework is used to examine the relationship between household modality styles, level of vehicle ownership, transit season pass possession and travel mode choice behavior using travel diary data from Karlsruhe, Germany. The dataset comprises a sample of 96 male and female household heads belonging to 48 households surveyed over a six-week observation period. Estimation results identify four distinct household modality styles. The model uncovers both significant correlation between modal preferences of heads of the same household and notable differences as well. In general, female household heads are found to be less reliant on the automobile for their mobility requirements than their male counterparts. Short-term individual decisions, such as mode choice, are found to be inextricably linked with more long-term individual and household decisions, namely level of vehicle ownership and transit season pass possession, both of which vary considerably across different modality styles.

Modality styles have important implications for transportation policies and infrastructural initiatives seeking to change existing patterns of travel mode choice behavior. Travel demand models constitute an important component of the planning and policy-making process, being widely used to make forecasts, which in turn are driven by the assumptions that these models make about how individuals arrive at decisions. We use estimation results from the BATS 2000 dataset to compare forecasts from different model specifications for scenarios evaluating the impact of increased auto congestion and improvements to the public transit system on travel mode choice behavior. Findings reveal that models of travel mode choice behavior that ignore the influence of modality styles can overestimate expected gains from transport policies and infrastructural initiatives seeking to reduce automobile use by factors of between one-and-a-half and three. The dissertation further demonstrates how incremental improvements in the transportation system, unless accompanied by corresponding shifts in the distribution of individuals across different modality styles, will result in far smaller changes in travel behavior than would be predicted by a traditional model of travel mode choice. This dissertation makes the case that what is needed is a dramatic change to the transportation system that forces individuals to reconsider how they travel.

Though the applications presented in the dissertation restrict their attention to the influence of modality styles on four specific dimensions of individual and household travel and activity behavior - travel mode choice for work/mandatory tours, travel mode choice non-work/non-mandatory tours, transit season pass possession and level of vehicle ownership, our results serve as a good starting point for a more comprehensive framework that recognizes the influence of modality styles on all dimensions of individual and household travel and activity behavior. The model framework developed by this dissertation is shown to be both methodologically flexible and empirically robust. Using a model of individual modality styles and travel mode choices in a static context as the foundation, we were able to expand the framework in multiple directions, extending it to a dynamic context, including additional dimensions of decision-making such as transit season pass possession and level of vehicle ownership, and incorporating the influence of intra-household interactions on individual preferences. Despite differences in observation period, sample size, local topography and cultural context across the three datasets, the framework was consistently found to outperform traditional models of travel behavior in terms of both statistical measures of fit and behavioral interpretation. The dissertation concludes with a discussion on how the framework might be extended further to include dimensions such as destination choice, vehicle miles traveled and residential location. We identify some of the major hurdles to their inclusion and suggest possible solutions, laying out an extensive road map for future research in the area. When complete, the line of work initiated by this dissertation is expected to result in a comprehensive model of individual and household travel and activity behavior that integrates travel demand and land use analysis through the modality styles construct with the objective of offering a deeper understanding of decision-making and greater predictive power than current models in practice.

Every day, decision-makers make choices among finite and discrete sets of alternatives. For example, people decide whether to walk, bike, take transit, or drive to work; shoppers decide which of the available brands of toothpaste to buy; and firms decide which vacant buildings they will rent for office space. Across these disparate domains, discrete choice models mathematically represent the procedures that analysts believe decision-makers are using to make such choices.

Historically, the field of discrete choice modeling grew mainly out of economics, and this lineage has had long-lasting methodological ramifications. In particular, despite the great mathematical similarity between discrete choice models and models in statistics, machine learning, and causal inference, discrete choice research remains mostly siloed, seldom drawing from or contributing to methods in these related disciplines.

In this dissertation, we help demolish the methodological silo around discrete choice re- search. Drawing from recent techniques in statistics, machine learning, and causal inference, we remove substantive limitations on the decision-making processes that could be be represented and predicted with previously available discrete choice methods. At the same time, by addressing concerns of discrete choice modelers, we make methodological contributions to the fields of statistics and machine learning, and we identify future research areas where discrete choice modelers are well suited to advancing the state of the art in causal inference.

Importantly, the methodological advances described above were not divorced from to- day’s societal concerns. Given that more and more government agencies are (unsuccessfully) attempting to raise bicycle commuting rates in their jurisdictions, we guide our interactions with the statistics, machine learning, and causal inference literatures by trying to more accurately model an individual’s choice of commuting by bicycle. In particular, we use parametric link functions from statistics to better model the adoption and abandonment of bicycling. From machine learning, we use decision trees to represent the non-compensatory decision protocols that individuals appear to follow when deciding whether to commute by bicycle, and we use diagrams from the causal inference literature to gain insight into how we can bet- ter model the effects of bike lane investments on bicycle commute mode shares. All together, we not only make methodological contributions to the fields of discrete choice, statistics, machine learning, and causal inference, but we contribute to the efforts of transportation planners and modelers who are trying to make our cities and regions more sustainable and environmentally friendly. The methods developed in this dissertation have applications to strategic bicycle planning, helping analysts understand when certain interventions are not enough to cause people to abandon non-bicycle modes of travel at the desired rates and what alternative interventions might be more effective.

In total, the specific contributions of this dissertation are the following:

1. We create a new spatial unit of analysis (the zone of likely travel) for the incorporation of roadway-level variables such as presence and type of bicycle infrastructure, roadway slopes, and traffic speeds into mode choice models.

2. We propose and demonstrate the novel use decision-tree methods for directly including roadway-level variables in mode choice models.

3. We create a new class of closed-form, finite-parameter, multinomial choice models that avoid an undesirable symmetry property that we describe in Chapter 3.

4. We create a procedure for using this new class of models to extend many existing binary choice models to the multinomial setting for the first time.

5. We create methods for creating new, symmetric and asymmetric, binary choice models.

6. We provide a microeconomic framework for interpreting decision trees by showing that decision trees represent a non-compensatory decision rule known as disjunctions-of-conjunctions and that such rules generalize many of the non-compensatory rules used in the discrete choice literature so far.

7. We propose and estimate the first bayesian model tree, thereby combining decision trees and discrete choice models in the first two-stage, semi-compensatory model that jointly:

a) uses disjunctions-of-conjunctions for the choice-set generation stage,

b) allows for context-dependent preference heterogeneity in the choice stage, and

c) quantifies analyst uncertainty in the estimated disjunctions-of-conjunctions

8. We identify techniques such as the use of causal diagrams that can be borrowed from the causal inference literature to improve the ability of discrete choice modelers to predict outcomes under external changes or policy interventions such as investing in on-street bicycle lanes.

9. We identify areas of the causal inference literature that can be improved through the incorporation of techniques from discrete choice or through the application of causal inference techniques that are very relevant to discrete choice modellers yet only infrequently researched by traditional causal inference researchers.

Through this dissertation, we empirically demonstrate most of our contributions using commute mode choice data from the San Francisco Bay Area. In every case, we found that the new models developed as part of this dissertation fit our data better than traditional discrete choice models. These results were stable across all measures of fit that were used, whether the measures were in-sample or out-of-sample, frequentist or bayesian. Beyond fit, all of our new models also proved to be qualitatively different than traditional discrete choice methods. Our new models provided insights and forecasts that both made more sense and were more accurate than their traditional counterparts. Finally, our contributions related to causal inference are the only items from the list above without empirical demonstrations. Instead, these contributions are bolstered by substantial literature review, discussion, and thought exercises that show the (general and bicycle specific) benefits of merging discrete choice and causal inference techniques.

To address issues of climate change, information about greenhouse gas emissions is more and more often being presented to people. For example, labels and signs displaying pounds and kilograms of CO2 are showing up in trip planners, in car advertisements, and even in restaurant menus. This is being done under the assumption that the information about the environmental impacts of different alternatives will encourage more sustainable behavior. However, little is known about whether or not this strategy affects the choices people make. In order for people to change their behavior for the benefit of reduced emissions, they first need to place positive value on those reductions. This research aims to answer three questions regarding the efficacy of presenting people with this information: first, do people place significant value on reducing their greenhouse gas emissions, second, can people consistently value a pound of CO2, and third, how does this value of green vary across the population? To answer these questions, five experiments were designed and conducted using discrete choice experiments, a framework typically used to investigate how people value reducing their travel time. UC Berkeley students and residents of the San Francisco Bay Area made stated and revealed preference transportation choices from a set of alternatives. With knowledge of the attributes of their chosen alternative as well as those of their available alternates, their choices were analyzed using logit, mixed logit, and hybrid choice models. The findings include that not only can people consistently interpret and place value on the pounds of CO2 associated with their transportation alternatives, but also that there exists a discrete distribution in the value of green. For all except one experiment, the best models indicate that while a majority of the population does not act as though they care about reducing their greenhouse gas emissions, there is a small group with a willingness to pay of $2.68 per pound of CO2.

Metropolitan Planning Organizations (MPOs) regularly perform equity analyses for their long-range transportation plans, as this is required by Environmental Justice regulations. These regional-level plans may propose hundreds of transportation infrastructure and policy changes (e.g. highway and transit extensions, fare changes, pricing schemes, etc.) as well as land-use policy changes. The challenge is to assess the distribution of impacts from all the proposed changes across different population segments. In addition, these agencies are to confirm that disadvantaged groups will share equitably in the benefits and not be overly adversely affected. While there are a number of approaches used for regional transportation equity analyses in practice, approaches using large scale travel models are emerging as a common existing practice. However, the existing methods used generally fail to paint a clear picture of what groups benefit or do not benefit from the transportation improvements. In particular, there are four critical shortcomings of the existing transportation equity analysis practice. First, there is no clear framework outlining the key components of a transportation equity analysis at the regional-level. Second, the existing zonal-level group segmentation used for identifying target and comparison groups are problematic and can lead to significant biases. Third, the use of average equity indicators can be misleading, as averages tend to mask important information about the underlying distributions. Finally, there is no clear guidance on implementing scenario ranking based on the equity objectives.

In addressing the first shortcoming of existing equity analysis practices, we present a guiding framework for transportation equity analysis that organizes the components of equity analysis in terms of transportation priorities, the model, and the equity indicators. The first component emphasizes the need to identify the priority transportation improvement(s) relevant for communities, as this guides the transportation benefits (or costs) to be evaluated. The second component is the model to be used for facilitating scenario analysis and measuring the expected transportation and land-use changes. The third component refers to the selection of equity indicators (ideally selected based on the transportation priorities identified), and the evaluation of these indicators. This three-part framework is also useful for outlining the research needs for transportation equity analysis. Among other key research needs, the literature indicates that the development of meaningful distributional comparison methods for transportation planning and decision-making and the use of more comprehensive measures of transportation benefit (for use as equity indicators) are critical.

The primary contributions of this dissertation relate to the third component; we develop an advanced approach for evaluating transportation equity outcomes (as represented by the equity indicator(s)). Our proposed analytical approach to transportation equity analysis addresses the existing shortcomings with respect to zonal-level group segmentation and average measures of transportation equity indicators. In addition, our approach emphasizes the importance of scenario ranking using explicit equity criteria. Our approach leverages the disaggregate functionality of activity-based travel demand models and applies individual-level data analysis to advance the existing equity analysis practices.

There are four steps in our proposed equity analysis process. The first step is to select the equity indicators to be evaluated and segment the population into a target group and comparison group(s). In this case we advocate for an individual -unit of segmentation and therefore individual-level equity indicators. This minimizes the biases associated with aggregate group segmentation and average equity indicators. The second step is to calculate the indicators from the model data output, which involves determining the exact measures (formulas) for the selected equity indicators. Here we advocate for measures that are comprehensive and sensitive to both transportation system changes and land-use factors, such as the logsum accessibility and consumer surplus measure. The third step in the process is to generate and evaluate distributions of the individual-level equity indicators. In particular, we advocate for the use of what we refer to as the "Individual Difference Density" comparison, which compares distributions of individual-level changes for the population segments across the planning scenarios. This comparison allows for the "winners" and "losers" resulting from the transportation and land-use plans to be identified. The fourth and final step in the process is to identify equity criteria (associated with the chosen equity standard (objective)) and rank the planning scenarios based on the degree to which they meet the equity criteria.

We present two conceptual demonstrations of the advantages of distributional comparisons, relative to average measures. The first case uses a synthetic data set and simple binary mode choice model to show and the second case uses an empirical data set (the 2000 Bay Area Travel Survey) and more sophisticated mode choice model. These demonstrations show that distributional comparisons are capable of revealing a much richer picture of how different population segments are affected by transportation plans, in comparison with average measures. Further, distributional comparison provides a framework for evaluating what population's characteristics and conditions lead to certain distribution transportation outcomes.

Our proposed process for regional transportation equity analysis is subsequently applied in a case study for the San Francisco Bay Area. We evaluate joint transportation and land-use scenarios modeled using the Metropolitan Transportation Commission's state-of-the-art activity-based travel demand model. We demonstrate the power of individual-level data analysis in a real-world setting. We calculate individual-level measures of commute travel time and logsum-based accessibility/consumer surplus using the model output and compare the scenario changes across income segments. We generate empirical distributions of these indicators and compare the changes associated with the planning scenarios for low and high income commuters. Further, we apply criteria for a set of equity standards (which represent alternative equity objectives) and rank the planning scenarios. There are four key takeaways from this case study. First is that our results show a significant difference in equity outcomes when using the individual-level population segmentation approach, compared to using the zonal segmentation approach done in practice. In fact we find opposite results. For average commute travel time, the Metropolitan Transportation Commission's zonal segmentation approach indicates that low income commuters are worse off than all other commuters, while the individual segmentation approach (in our case) indicates that low income commuters are significantly better off than high income commuters. While the underlying causes for these results warrant further investigation, we hypothesize that this difference is due to the fact that the zone-based approach only captures 40% of the target (low income) group. The individual-level segmentation approach is able to capture 100% of the target group. Second is regarding the equity indicators evaluated. The commute travel time indicator results indicate that low income commuters are better off than high income commuters, while the accessibility/consumer surplus results indicate that low income commuters are worse off than high income commuters. The underlying causes for these results warrant further investigation, but we hypothesize that this difference in results to due to the fact that the logsum accessibility/consumer surplus measure by design is able to capture transportation and land-use related factors, while the travel time measure only captures one dimension of transportation user factors. Focusing on travel time may be misleading because it does not fully capture the true benefits of the transportation scenarios. Third is regarding the use of distributional comparisons, relative to average measures. We find that distributional comparisons are much more informative than average measures. The distributional measures are capable of providing a much richer picture of individuals-level transportation impacts, in terms of who gains and who loses due the transportation planning scenarios. Using the accessibility/ consumer surplus measure, the Individual Difference Densities show that as many as 33.3% of low income commuters experience losses, compared to 13.4% for high income commuters. Finally, we make the case that the use of equity standards for scenario ranking plays an important role in the equity analysis process. Our results show that different equity standards result in different rankings for the transportation planning scenarios. This points to the need for agencies (and communities) to make conscious decisions on what equity standard(s) should be used and apply this/these in the scenario ranking process.

This dissertation work includes the first known full-scale application of a regional activity-based travel model for transportation equity analysis that involves distributional comparisons of individual-level equity indicators and scenario ranking based on equity criteria. We find that while the existing practice is to use average measures to represent how difference are affected by transportation plans, distributional comparison are able to provide for a richer evaluation of individual-level transportation impacts. Distributional comparisons provide a framework for quantifying the "winners" and "losers" of transportation plans, while average measures and be misleading and uninformative. We make significant progress with regard to evaluating equity indicators (part three of the guiding framework). However, our proposed process is flexible and can be extended to include a number of additional advances, including more environmental and long-term land-use related equity indicators (e.g. emissions exposure, gentrification and displacement risk, employment participation, etc.) and additional population segments (e.g. age, ethnicity, household type, auto-ownership class, etc.). Among other important research directions, our analytical framework for regional transportation equity analysis can be applied to investigating why certain groups are more likely to be "losers" and what factors of transportation planning scenarios to modify in order to arrive at a more equitable transportation and land-use plan.