Event Type: PhD Defense

This dissertation introduces a novel model for integration into an Agent-Based Model (ABM) framework, aimed at estimating and predicting workers’ commuting behaviors in a post-pandemic context. The model is designed to inform policy-making by analyzing commuting patterns and their responsiveness to policy changes. The methodology involves three stages: first, identifying worker commuting preference classes through latent class analysis; second, predicting class membership based on demographic features; and third, estimating commuter types (commuter, hybrid commuter, or telecommuter) using logistic regression. The study utilizes the ASU Covid Future Panel Survey data to identify diverse worker classes with distinct telecommuting experiences and preferences. Key findings include the discovery of various worker classes, such as regular commuters and telecommuters, and a class inclined towards high-frequency telecommuting under supportive policies. The research also explores intra-class commuter type estimation and factors influencing commuting choices, offering valuable insights for future commuting pattern predictions and policy development.

The purpose of this study is to develop a methodology for processing vehicle trajectory data which are presented as a series of discrete positions of vehicles recorded over consecutive time intervals. The framework combines vehicle trajectory smoothing and imputation, ensuring that speeds and higher-order derivatives of positions are consistently defined as symplectic differences in positions, while adhering to physically meaningful bounds determined by traffic laws, drivers’ behaviors, and vehicle characteristics.

This dissertation explores the evolving landscape of prepared food consumption, particularly restaurant meals, and proposes optimization strategies for managing delivery fleets. In the context of COVID-19, I examined shifts in meal consumption in California using Heterogeneous ordered logit models. I analyzed meal delivery, uncovering unique dynamics across regions and times and emphasizing the role of deliveries in enhancing food access for marginalized communities. Using graph theory, I also explored fleet management optimization, comparing Hopcroft-Karp and Karp algorithms. This research informs policy interventions, aids platform operators, restaurant owners, and urban planners, and bridges academia and practice through an interdisciplinary lens.

Cities around the world vary in terms of their transportation network structure and travel demand patterns, with implications for the viability of shared mobility services. Recently, the urban mobility sector has witnessed a significant transformation with the introduction of several new types of Mobility-on-Demand (MOD) services that vary in terms of their capacity and flexibility of routes, schedules, and user Pickup and Dropoff (PUDO) locations. This dissertation proposes models and algorithms to analyze sharing in transportation networks and Mobility-on-Demand (MOD) services in two comprehensive studies. The first study addresses the lack of metrics that jointly characterize a region’s travel demand patterns and its transportation network in terms of the potential for travelers to share trips. I define sharing potential in the form of person-trip shareability and introduce ‘flow overlap’ as a fundamental shareability metric. Then, I formulate the Maximum Network Flow Overlap Problem (MNFLOP), a math program that assigns person-trips to network paths that maximize network-wide flow overlap. The results reveal that the shareability metrics can (i) meaningfully differentiate between different Origin-Destination trip matrices in terms of flow overlap, and (ii) quantify demand dispersion from a single location considering the underlying road network. Finally, I validate MNFLOP’s ability to quantify shareability by showing that demand patterns with higher flow overlap are strongly associated with lower mileage routes for a last-mile microtransit service. The second study proposes a scalable algorithm for operating shared-ride MOD services with flexible and dynamic PUDO locations (or virtual stops)—called C2C (Corner-to-Corner) services—in a congestible network. I compare four MOD service types: Door-to-Door (D2D) ride-hailing, D2D ride-pooling, C2C ride-hailing, and C2C ride-pooling by evaluating operator and user costs. The results show that ride-pooling reduces operator costs while slightly increasing user costs, whereas C2C reduces operator costs but significantly increases user costs. Combining ride-pooling and C2C further reduces operator costs and decreases vehicle miles traveled in MOD systems.

Traffic safety is a major public health issue worldwide. The Kingdom of Saudi Arabia (KSA) is facing alarming road safety concerns with traffic-related injuries being the third leading cause of fatalities in the country. A better understanding of factors influencing injury severity outcomes of traffic crashes is vital for the proactive and effective implementation of suitable countermeasures. Various non-parametric machine learning (ML) based techniques have been increasingly used lately to address the drawbacks of statistical schemes for modeling the injury severity of traffic crashes. The dissertation examines the application of six different ML algorithms for injury severity prediction of traffic crashes based on three years data on interstate rural highways in KSA. Injury severity modeling performance of proposed algorithms was assessed in terms of various evaluation indices. Experimental results revealed that some models outperformed others based on the selected evaluation metrics. To address the ML model’s non-interpretability issue, feature importance and SHAP analysis were also employed. An analysis of several significant factors which aggravate the chances of fatal and severe injuries was done. The study also proposes the application of the Information Root Node Variation (IRNV) technique for extraction of significant decision rules highlighting the circumstances for the categorization of specific crash injury severity instances. For comparison purposes, multinomial logit (MNL) models were also developed and the consistency of injury severity risk factors between MNL and ML models was investigated. The outcomes and findings of the current study can yield valuable insights to safety practitioners for timely and effective implementation of suitable mitigation measures.

Apart from the COVID-19 pandemic, two chronic problems affecting Californians are high housing costs and road congestion. Although high housing costs and the determinants of commuting have separately received a lot of attention from academic researchers, to my knowledge very few papers have analyzed the linkage between them. In this dissertation, I present three essays that will enhance our understanding on the relationship between commuting, land use, housing costs, and the impact of COVID-19 on telecommuting. In all three essays, I use Structural Equation Model (SEM).

In my first essay, I propose a framework for understanding the impact of housing costs on commuting time and commuting distance in one worker-households in Los Angeles County, which is the most populous county in the US. After analyzing data from the 2012 California Household Travel Survey (CHTS), I find that households who can afford more expensive neighborhoods have on average a commute 3.1% shorter per additional $100k to their residence median home values.

In my second essay, I analyze the commutes of two-worker households to understand some of the trade-offs they need to make, since two-worker households have dual work constraints. My data for this essay come from 2017 National Household Travel Survey (NHTS) respondents who reside in five U.S. MSAs (San Francisco, Los Angeles, Dallas, Houston, and Atlanta). Results show that women do not commute as far as men on average, although their commuting time is not necessarily shorter than men’s, and that the commuting times of men and women are weakly positively correlated. Moreover, households have faster commutes by 14.5% for men and 22.7% for women per additional $1000 to their residence median monthly housing cost.

My third essay investigates the impact of the COVID-19 pandemic on telecommuting by analyzing a unique dataset collected at the end of May 2021 by IPSOS via a random survey of California members of KnowledgePanel®. I find that an additional 4.2% of California workers would engage in some level of telecommuting and more educated workers are expecting to telecommute more (0.383* for bachelor’s degree) post-pandemic.

Teasing out the impact of high housing costs on commuting is important at a time when concerns about the environmental impacts of transportation have turned reducing vehicle-miles traveled (VMT) into a policy priority. More generally, a better understanding of the determinants of commuting is critical to inform housing and transportation policy, improve the health of commuters, reduce air pollution, and achieve climate goals.

Public transportation in the U.S., including in California, is under siege. Over the last two decades, ridership has been steadily declining, possibly because traditional transit users gained more accessibility to private cars and because of the emergence of transportation network companies (TNCs, i.e., Uber and Lyft). The COVID-19 pandemic worsened a bad situation. Public transportation officials are now confronted with the challenge of restoring the health of public transit so it can contribute to a more equitable and sustainable transportation system. Therefore, in this dissertation, I first investigate how Transportation Network Companies (TNCs, e.g., Uber and Lyft) impacted transit ridership pre-pandemic, before analyzing how COVID-19 affected transit and other modes. I rely on both discrete choice and quasi-experimental models to analyze data from the 2009 and 2017 National Household Travel Surveys and from a California survey administered in May 2021 by Ipsos.

In Chapter 2, my results for the U.S. show that individuals/households who use either public transit or TNCs share socio-economic characteristics, reside in similar areas, and differ from individuals/households who use neither public transit nor TNCs. In addition, individuals/households who use both public transit and TNCs tend to be Millennials or belong to Generation Z, with a higher income, more education, no children, and fewer vehicles than drivers. To the best of my knowledge, this is the first nationwide study to contrast public transit and TNC users that relies on cross-nested logit structures. My second contribution here is a comparison between individual and household-level models to account for intra-household dependencies of mode choice.

In Chapter 3, I quantify the impact of TNCs on household transit use by comparing travel for households from the 2017 NHTS (who had access to both transit and TNCs) matched with households from the 2009 NHTS (who only had access to transit) using propensity score matching. Overall, I find an 18.2% daily reduction in transit use for the entire U.S., with a 21.8% drop for weekdays and a 15.2% decrease for weekends. My main contribution here is to tease out the causal link between the emergence of TNCs and the decline of transit at the household level using propensity score matching, as previous studies relied either on descriptive statistics, correlation analyses, or considered aggregate ridership changes.

In Chapter 4, I analyze how Californians changed transportation modes due to COVID-19 and explore their intentions to use different modes after the pandemic. I find that driving but especially transit and TNCs, could see substantial drops in popularity after the pandemic. Many Hispanics, African Americans, Asians, lower-income people, and people who would like to telecommute more intend to use transit less. Key obstacles to a resurgence of transit after COVID-19 are insufficient reach and frequency, shortcomings that are especially important to younger adults, people with more education, and affluent households (“choice riders”). In addition to addressing these concerns, effective transit policies must be integrated into a comprehensive framework designed to achieve California’s social and environmental goals.

Overall, my findings highlight the danger for public transit to enter into outsourcing agreements with TNCs, neglect captive riders (people with no alternatives to transit), and to expose choice riders to TNCs. Key priorities for transit agencies should therefore be to increase the frequency of their service (as appropriate), and extend their reach to solve the “first and last mile problem”, possibly by creating partnerships with micromobility providers.

Keywords: Public Transit; Transportation Network Companies; Active Modes; COVID-19; Cross-Nested Logit; Generalized Ordered Logit; Propensity Score matching

Trucking electrification combined with connected and automated technologies promises to cut the cost of freight transportation, reduce its environmental footprint, and make roads safer. If electric trucks are powerful enough to cease behaving as moving bottlenecks, they could also increase road capacity and reduce the demand for new infrastructure, a consequence that has so far been overlooked by the literature. In this dissertation, I study the traffic and infrastructure demand impacts of electrifying and connecting (via cooperative adaptive cruise control, CACC) heavy-duty drayage trucks (HDDTs) that serve the San Pedro Bay Ports (SPBP; the ports of Los Angeles and Long Beach, which is the largest port complex in the U.S), quantify the resulting health, environmental, and Environmental Justice impacts, and explore how to maximize the benefits of connected vehicles with active demand management.

In Chapter 2, I explore the potential traffic and infrastructure implications of replacing conventional HDDTs that serve the SPBP with electric and/or connected HDDTs. I rely on microscopic simulation on a freeway and arterial network centered on I-710, the country’s most important economic artery. My results show that 1,000-hp electric/hydrogen trucks can be a substitute for additional road capacity. Accounting for the traffic impacts of new vehicle technologies is critical in infrastructure planning, and my results suggest shifting funding from building new capacity to financing zero-emission (ZE) 1,000 hp HDDTs until the market for these vehicles has matured.

In Chapter 3, I quantify the health and GHG reduction benefits of replacing the HDDTs serving the SPBP with ZE-HDDTs. I simulate ZE-HDDTs on a regional freeway network to analyze their PM2.5 and CO2 emissions in 2012 and 2035 using MOVES3 emission factors. I then estimate their contribution to PM2.5 concentrations with InMAP and health impacts with BenMAP. I find that despite technology improvements and air quality regulations, SPBP HDDTs would still cause 106 premature deaths (valued at $1.3 billion in $2022) and 2,142 asthma attacks (>2/3 accruing to disadvantaged communities) in 2035 due to population and drayage traffic growth, not to mention at least $220 million in climate costs. With ZE-HDDTs becoming attractive from a total cost of ownership point-of-view, the main cost of achieving ZE road drayage is a scrappage program for non-ZE-HDDTs. My results justify implementing this program by 2035.

In Chapter 4, I study the performance impacts of lane management strategies implemented on I-710 to support the deployment of CACC-enabled vehicles and their potential to absorb the 2035 projected growth in cargo demand at the SPBP. I find that a designated lane for CACC-enabled vehicles can decrease congestion by creating more platooning opportunities, thus maximizing CACC benefits.

Mobility systems are undergoing a major transformation due to emerging autonomous and shared mobility technologies. A primary aspect of such technologies consists of improving the mobility system inefficiencies via a reduction in the number of vehicles needed to fulfill the transportation needs. This would impact the use of vehicles and their expected lifetime. This dissertation is focused on the importance of the increased usage of vehicles and how the system can benefit from an optimization with a vehicle point of view. The improvements come from mainly two aspects of shared mobility – carsharing and ridesharing – which are both implemented in the modeling and optimization framework. An analysis of the current vehicle ownership and trip distributions is presented. A vehicle usage cost function is designed to incorporate the changed relative importance of fixed and usage-based variable costs. It presents a framework that analyzes the interactions between all the elements, including a pricing scheme for benefit-cost analysis and optimizations from a service provider perspective. With shared mobility, ownership paradigms can also change to subscription-based use
of vehicles from fleet service providers, as included implicitly in the interaction framework. Modeling is carried out for idealized networks, as well as a real-world network of a reasonable size from the city of Irvine, CA. The results capture the increased use of shared and/or autonomous vehicles (SAV) and the benefits of optimizing the system with properly updated costs. Results and conclusions are provided on the viability of service provider plans as well as on system benefits in terms of the replacement ratio indicating how many personal vehicles can be removed using autonomous fleets.

Traffic congestion is known to have many negative impacts both for travelers and society as a whole (e.g., emissions, noise). This dissertation is centered on the modeling and analysis of vehicle and trip flow dynamics. Several transportation systems are considered, from a single bottleneck to a network region, in order to propose effective management strategies to reduce traffic congestion. Since different transportation systems are considered, different models and space paradigms are needed to model each system’s dynamics. At the local level, it is reasonable to study the traffic flow in the physical road and how it evolves spatially and temporally with traditional traffic flow theory models. In contrast, when the network is highly complex (e.g., corridors with many on- and off-ramps or urban cores), it is reasonable to disregard the detailed vehicle-network interaction at the distinct physical locations and model traffic dynamics with more aggregated models in the relative space.