Data-driven modeling of taxi trip demand and supply in New York City

The taxicab is an important transportation service in New York City (NYC). This dissertation addresses three primary questions related to taxi demand and supply in NYC using innovative data including: a large amount of taxi Global Positioning System (GPS) data that the NYC Taxi & Limousine Commission (TLC) has collected over a two-year period, detailed transit schedule information, and neighborhood characteristics. The three questions are: Is there variability and imbalance between taxi demand and supply across different times of day and different locations in NYC? What factors determine the distribution of demand for taxicab services in NYC by time and by location? How can the imbalance between taxicab demand and supply be identified and quantified in order to guide policies to improve transportation access in NYC? iii The hypothesis is that the variability and imbalance between taxi demand and supply at different times of day and locations exists. Neighborhood characteristics are factors that are expected to be related to the distribution of taxi trips. In the supply analysis, the taxi supply is quantified, an analysis of taxi use every day identifies factors related to the variability of taxi supply over time, and a customer search model was developed to identify areas where drivers of vacant taxis tend to go and to quantify how efficiently drivers of vacant taxis search for their next customer. The model indicates that the NYC taxicabs could be redistributed more efficiently as a system if the taxi drivers are assigned to specific passenger pickups. In trip generation modeling, six important explanatory variables are identified that influence taxi trips: population, education, supply, income, Transit Access Time (TAT), and employment. More complex models involving count regression and accounting for spatial autocorrelation are then developed to understand the imbalance between taxi demand and supply by controlling for the above-mentioned factors. The errors of the trip generation model provide insights of when and where there is insufficient taxi supply or surplus taxi supply relative to taxi demand. A case study is introduced to compare the total cost for two modes of transportation (transit and taxi) between NYC Penn Station and three New York area airports (JFK, LGA and EWR). Transit is found to be more cost-effective than taxi for most times of the day.Ph.D.Includes bibliographical referencesby Ci Yan