Abstract

The three chapters in this dissertation study and apply econometric models to answer questions in transportation economics. Chapter 1 and 2 analyze the Berry, Levinsohn and Pakes (BLP) discrete choice model for combined micro- and macro-level data. Chapter 1 considers the concerns of choice set aggregation and estimating consistent standard errors within the BLP Model. These concerns are studied within the context of a vehicle choice application with interest in estimating household valuation of fuel efficiency. Chapter 2 studies the numerical properties of the maximum likelihood approach to estimating this BLP model. Chapter 3 applies a Poisson-Log Normal panel data model to study the effect of red light cameras on collision counts in Los Angeles. The camera program suffered from weaknesses in enforcement that dampened the effectiveness of the program over time. The model considered here controls for this dampening effect. Chapter 1 finds that choice set aggregation affects the point estimates obtained from the BLP model and causes standard errors to be too small. The use of inconsistent sequential standard errors also underestimates the magnitude of standard errors. These findings may partly explain the disparity across existing estimates from choice models on the value households place on vehicle fuel efficiency. Chapter 2 finds that the maximum likelihood estimation approach is able to find the global minimum regardless of choice of starting values, optimization routine used and tightness of convergence criteria. These findings highlight the benefits of estimating the BLP model on combined micro- and macro-level datasets using the maximum likelihood approach compared to using the nested fixed point approach and only macro level data where numerical stability is difficult to obtain. Chapter 3 finds that controlling for the dampening effect from poor enforcement, the Los Angeles Automated Red Light Camera program decreased red light running related collisions but increased right-angle and injury collisions, as well as collisions overall.

Abstract

This thesis is concerned with specifying and estimating multivariate models in discrete data settings. The models are applied to several empirical applications with an emphasis in banking and monetary history. The approaches presented here are of central importance in model evaluation, policy analysis, and prediction. The first chapter develops a framework for estimating multivariate treatment effect models in the presence of sample selection. The methodology deals with several important issues prevalent in program evaluation, including non-random treatment assignment, endogeneity, and discrete outcomes. The framework is applied to evaluate the effectiveness of bank recapitalization programs and their ability to resuscitate the financial system. This paper presents a novel bank-level data set and employs the new methodology to jointly model a bank’s decision to apply for assistance, the central bank’s decision to approve or decline the assistance, and the bank’s performance. The article offers practical estimation tools to unveil new answers to important regulatory and government intervention questions. The second chapter examines an important but often overlooked obstacle in multivariate discrete data models which is the proper specification of endogenous covariates. Endogeneity can be modeled as latent or observed, representing competing hypotheses about the outcomes of interest. This paper highlights the use of existing Bayesian model comparison techniques to understand the nature of endogeneity. Consideration of both observed and latent modeling approaches is emphasized in two empirical applications. The first application examines linkages for banking contagion and the second application evaluates the impact of education on socioeconomic outcomes. The third chapter, which is joint work with Professor Ivan Jeliazkov, studies the formulation of the likelihood function for simultaneous equation models for discrete data. The approach rests on casting the required distribution as the invariant distribution of a suitably defined Markov chain. The derivation resolves puzzling paradoxes highlighted in earlier work, shows that such models are theoretically coherent, and offers simple and intuitive linkages to the better understood analysis of continuous outcomes. The new methodology is employed in two applications involving simultaneous equation models of (i) female labor supply and family financial stability, and (ii) the interactions between health and wealth.

Abstract

In this research, we propose a series of models designed to take advantage of the availability of data–both structured and unstructured–from a variety of sources ranging including passive data, questionnaires, and social media data to analyze underlying patterns and trends in travel and activity behavior, and to provide results that support enhancements both in transportation planning and the application of programming to support such efforts. First, we introduce a framework for automatically inferring the travel modes and trip purposes of human movement when tracked by a GPS device. We utilize a multiple changepoints algorithm to divide trajectories into segments using only speed data. Then, Random Forest is used to classify segments into moving and not-moving types. For moving segments, travel mode (car, bus, train, walk, and bike) is predicted. Next, multiple machine learning algorithms are employed, validated, and tested to identify the most suitable model for inferring trip purposes. The overall accuracy for prediction is over 80% on the testing set, both with and without data on socio-demographic variables. The model also predicts “shop” trips with an accuracy of 92.1%, while its accuracy for “go home” and “studying” trips reaches 100%. Additionally, we utilize the classification results in the first stage of research to compare households’ travel patterns from before a new light rail transit line began service to two periods of time after service began. Our results indicate that, although the average of activity duration varies significantly over days of week and waves, the random effect of these two factors on activity duration was minor; time of day contributed over one third of the total variance in the duration. Finally, the dissertation demonstrates uses of Twitter data as a potentially important data source to understand comments, criticisms, and responses about light rail in Los Angeles. The results of information flow analysis, sentiment estimation, topic modeling, and its application can be useful for exploring trends among commuters and how their sentiment changed according to the light rail line they used, the time of day, and the day of the week

Abstract

A macroscopic relation between the network-level average flow-rate and density, which is known as the macroscopic fundamental diagram (MFD), has been shown to exist in urban networks in stationary states. In the literature, however, most existing studies have considered the MFD as a phenomenon of urban networks, and few have tried to derive it analytically from signal settings, route choice behaviors, or demand patterns. Furthermore, it is still not clear about the definition or existence of stationary traffic states in urban networks and their stability properties. This dissertation research aims to fill this gap.

I start to study the stationary traffic states in a signalized double-ring network. A kinematic wave approach is used to formulate the traffic dynamics, and periodic traffic patterns are found using simulations and defined as stationary states. Furthermore, traffic dynamics are aggregated at the link level using the link queue model, and a Poincare map approach is introduced to analytically define and solve possible stationary states. Further results show that a stationary state can be Lyapunov stable, asymptotically stable, or unstable. Moreover, MFDs are explicitly derived such that the network flow-rate is a function of the network density, signal settings, and route choice behaviors. Also the time for the network to be gridlocked is analytically derived.

Even with the link queue model, traffic dynamics are still difficult to solve due to the discrete control at signalized junctions. Therefore, efforts are also devoted to deriving invariant continuous approximate models for a signalized road link and analyzing their properties under different capacity constraints, traffic conditions, traffic flow fundamental diagrams, signal settings, and traffic flow models. Analytical and simulation results show that the derived invariant continuous approximate model can fully capture the capacity constraints at the signalized junction and is a good approximation to the discrete signal control under different traffic conditions and traffic flow fundamental diagrams. Further analysis shows that non-invariant continuous approximate models cannot be used in the link transmission model since they can yield no solution to the traffic statics problem under certain traffic conditions.

For a signalized grid network, simulations with the link queue model confirm that important insights obtained for double-ring networks indeed apply to more general networks.

Abstract

This dissertation provides modifications and extensions to the Household Activity Pattern Problem (HAPP) to help move existing formulations from a laboratory prototype toward a more useable activity-based demand modeling product. Previous research on HAPP has been based on a pickup and delivery problem with time window constraints (PDPTW), which does not lend itself easily to application that is compatible with an activity-based forecasting model. Meanwhile, other research on activity-based modeling lacks of the integration of household decisions regarding time-of-day arrival, activity duration and traffic congestion effects on travel. We borrow concepts from economic research and consider that each household member tries to obtain maximum utility by choosing arrival time of activities, choosing activity duration while minimizing travel times and travel costs throughout the course of the day.