Dissertations, Theses, and Capstone Projects

Date of Degree

9-2015

Document Type

Dissertation

Degree Name

Ph.D.

Program

Economics

Advisor

Jonathan R. Peters

Subject Categories

Economics

Keywords

Big data; Large scale statistical computing; Mechanism design; Taxicab; Transportation; Travel demand

Abstract

This dissertation consists of four essays on the economic analysis of transportation systems. In the first chapter, the conventional disaggregate travel demand model, a probability model for the modeling of multiple modes, generally called random utility maximization (RUM), is expanded to a model of count of mode choice. The extended travel demand model is derived from general economic theory -- maximizing instantaneous utility on the time horizon, subject to a budget constraint -- and can capture the dynamic behavior of countable travel demand. Because the model is for countable dependent variables, it has a more realistic set of assumptions to explain travel demand then the RUM model. An empirical test of the theoretical model using a toll facility user survey in the New York City area was performed. The results show that the theoretical model explain more than 50 percent of the trip frequency behavior observed in the New York City toll facility users. Travel demand for facility users increase with respect to household employment, household vehicle count, and employer payment for tolls and decrease with travel time, road pricing, travel distance and mass transit access.

In the second chapter, we perform a statistical comparison of driving travel demand on toll facilities between Electronic Toll Collection (ETC) users, as a treatment group, and non users, as a control group, in order to examine the effect of ETC on travel demand that uses toll facilities. The data that is used for the comparison is a user survey of the ten toll bridges and tunnels in New York City, and the data contains individual user's travel attributes and demographic characteristics, as well as the frequency of usage of the toll facilities so that the data thus allows us to examine the difference in travel demand of E-ZPass, the Electronic Toll Collection System for Northeastern United States' highway ETC system and compare tag holders and non tag holders. We find that the estimated difference of travel demand between E-ZPass users and non-users is biased due to model misspecification and sampling selection, and E-ZPass has no statistically significant effect on travel demand after controlling for possible sources of biases.

In the third chapter, we develop a parallel sparse matrix-transpose-matrix multiplication algorithm using the outer product of row vectors. The outer product algorithm works with the compressed sparse row (CSR) form matrix, and as such it does not require a transposition operation prior to perform multiplication. In addition, since the outer product algorithm in the parallel implementation decomposes a matrix by rows, it thus imposes no additional restrictions with respect to matrix size and shape. We particularly focus on implementation of this technique on rectangular matrices, which have a larger number of rows and smaller number or columns for per- forming statistical analysis on large scale data. We test the outer product algorithm for randomly generated matrices. We then apply it to compute descriptive statistics of the New York City taxicab data, which is originally given by a 140.56 Gbytes file. The performance measures of the test and application shows that the outer product algorithm is effective and performed well on large-scale matrix multiplication in a parallel computing environment.

In the last chapter, I develop a taxi market mechanism design model that demonstrates the role of a regulated taxi fare system on taxi drivers' route choice behavior. In this model, a fare system is imposed by a taxi market authority with the recognition of asymmetric information, which in this case is about road network and traffic conditions, between passengers and drivers, and taxi trip demand is different and uncertain at its origin and destination. I derive a prediction from the model that shows the drivers have an incentive to make trip longer than optimal if they have passengers

Download

Included in

Economics Commons

Share

COinS