Broadening Understanding of the Interplay Among Public Transit, Shared Mobility, and Personal Automobiles (2018)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

61 To begin to establish a better understanding of how transportation network company (TNC) trips interact with public transit, the research team performed an exploratory analysis of the relationships between underlying demographic variables and the level of TNC use and the level of transit availability at different times within the study areas. While this exploratory analysis cannot establish the magnitude or direction of the relationships it identifies, it does point to areas for further research to investigate. The core question is whether the area demographics that are associated with TNC supply match, or contrast with, those associated with transit supply. The purpose of this analytical approach is to understand the ways in which TNCs complement or substitute for public transit. • Complementary relationship: If different underlying characteristics more reliably explain the different services’ supply, then we would assume a complementary relationship, in which the two services together represent a broadening of the transportation market (e.g., by increasingly the supply of reliable transportation effectively available at non-peak times, in areas away from concentrations of employment or without convenient connections to high-frequency transit corridors). In a complementary scenario, an individual might use a TNC to access a public transit stop that would have otherwise been inaccessible, or take transit rather than a personal auto at a time when transit is running, knowing they can reliably return via TNC once transit stops running for the night. • Substitution relationship: The more the spatial availability of the two services is tied to common characteristics of the physical environment and customer base implied by the area’s demo- graphics, the more they are likely to substitute for one another, serving the same customers for the same types of trips. In a substitution scenario, an individual uses a TNC instead of public transit. Research Question In determining whether transit and TNCs are complements or substitutes, it is important to understand if they are spatially available to the same or different customer bases. Put differently, the more that transit and TNCs co-locate, the more likely they are to compete directly for the same people, enabling a stronger substitution effect. Thus the research question is whether area demographics for TNC supply match or contrast with area demographics for transit supply. For the purposes of the TNC data, the analysis uses the origin location rather than the desti- nation. This implies an assumption that the household or individual demographics of an area reflect the customer base for the transportation services there—that the people who live in an area are substantially the same as the people who use transportation services there. But in areas with a more commercial focus, this might not be the case (i.e., renters or homeowners might be only a A P P E N D I X C Transportation Network Company/Transit Exploratory Demographic Analysis

62 Broadening Understanding of the Interplay Among Public Transit, Shared Mobility, and Personal Automobiles small part of the transportation market in an area dominated by offices or restaurants and bars), and for this reason employment characteristics were also part of the analysis. Data and Constraints This study compares three main sets of data representing TNC trips, scheduled transit supply, and the characteristics of populations and in the areas where the transportation services operate: 1. TNC trip data for five metro areas. These data were provided by a major national TNC and were supplied to the researchers aggregated by origin and destination zip code and hour of week, using relative percent instead of counts. Further, the data sums all trips during May 2016 into a one-week time frame. The metro areas covered by the data are Chicago, Los Angeles, Nashville, Seattle, and Washington, D.C. See Appendices A and B for detailed descriptions of these data. 2. Transit stop and schedule data. For each metro area in the TNC data set all current General Transit Feed Specification (GTFS) data were downloaded for October and November 2016. This data is designed to contain all schedule and route information for all trips being per- formed by the transit agency, including detailed stop locations. 3. Demographic data. Information on population, race, income, car ownership, education, and employment is combined to create a demographic profile of each study area. Data was primarily collected from the American Community Survey (ACS) 2015 5-year estimates and Longitudinal Employment Household Dynamics (LEHD) databases. The TNC data provided the main constraints for the exploratory analysis. On the one hand, the data is composed of some 1.3 million observations across five different metro areas, a broad geographical scope. On the other hand, those data were summarized at the zip code level— a coarse spatial resolution for studying transportation, particularly public transit, since a single downtown zip code might contain multiple train stations and scores of bus stops on dozens of different routes. Since they were originally designed as a means for organizing mail delivery to a series of addresses, zip codes have a large variation in area, shape, population, and infrastructure distribution. Beyond the difficulty this produces for comparing across zip codes, this can also translate into different areas within a single zip code having widely varying transportation and demographic characteristics that might be obscured at a summary level. Thus, the research question had to be something which could be addressed using zip code level comparisons. Methodology The overall methodology for this analysis was to determine which demographic and trans- portation characteristics were associated with transit and TNC usage in different cities and at different points in the travel week. Geographic information systems (GIS), PostgreSQL/PostGIS (a spatially enabled database useful for storing and manipulating large quantities of data) and the Python programming language were used to facilitate spatial aggregation and provide a replicable, programmatic approach to the problem. The basic approach was to use ordinary least squares regression to explore associations between demographic and transportation characteristics with transit and TNC usage, and to contrast the variables associated with the two modes at a given portion of the travel week. • Frequency of TNC pick-ups versus transit stops. The first step was to estimate how often TNC pick-ups occurred in a given zip code and time frame versus how often transit stops

Transportation Network Company/Transit Exploratory Demographic Analysis 63 occurred in the same zip code and time frame. The TNC data was already organized by hour and zip code, so what remained was to create a similar matrix of scheduled transit stops by hour and zip code. It was necessary to transform the GTFS data from stop location and schedule information to spatial frequency information. Using an open source GIS tool called Better Bus Buffers, the data was processed to count how many times any transit vehicle would be scheduled to stop at each transit stop, and then that data was aggregated to the total number of transit vehicle stops in a given zip code by the hour of day. This number was also divided by the size of the zip code in square miles to get the density of stops per hour. • Demographic data. The background Census demographic data was also collected at or aggregated to the zip code level. The following were the data points used as independent variables in the regression: – Education: University Grads (%) – Households with no vehicle (%) – Housing tenure: Owner/Renter (%) – Job density (count per square mile) – Median household income – Population ages 15–24 (%) – Population ages 25–44 (%) – Population ages 45–64 (%) – Population density (count per square mile) – Population earning no more than twice the poverty line (%) – Unemployed population (%) – White/Non-white population (%) This data was used in an iterative process with the “Exploratory Regression Analysis Tool” available in ESRI ArcGIS. This tool will test a combination of independent variables against a dependent variable and return a log detailing the combinations with the highest explanatory power. All the demographic variables above were provided as potential independent variables. They were iteratively tested for their ability to predict dependent variables, which represented the two transport modes in each of the five study regions (plus a combination of them all) at six different “dayparts,” or combinations of time of day and day of week. Together, the two modes, six regions, and six dayparts produce a total of 72 possible combinations to be tested. • All day • AM Peak (6-10am) • Mid-day (10am-2pm) • PM Peak (4-8pm) • Late Night Saturday (10pm-3am) • Minimum Acve Service (9pm-12am, Sun) Daypart (weekday unless noted otherwise) • All regions • Chicago • Los Angeles • Nashville • Sea…le • Washington, DC Region • Transit • TNCMode

64 Broadening Understanding of the Interplay Among Public Transit, Shared Mobility, and Personal Automobiles Mode as Independent Variable In addition to the 12 demographic independent variables, TNC or transit supply were also used as independent variables when not being used as dependent variables. When predicting TNC usage (dependent variable) transit density per zip code and time of day was tested as an indepen- dent variable; when predicting transit usage (dependent variable) TNC density per zip code and time of day were tested as independent variables. This gave a total of 13 independent variables. Findings • Public transit supply. In most cases, greater supply of transit is associated with a combination of job density and zero-vehicle households. As a fixed route public service, transit is supplied based on the most consistent demand— explained by concentrations of jobs and of households without personal vehicles. Demographic factors such as income, housing tenure, race, and education had little noticeable association with the supply, which reflects the public orientation of transit service, and its mandate to serve as equitably as possible. • TNC use. TNC use, on the other hand, was more commonly associated with combination of population aged 25–44 and population density, with job density a factor in some of the metros. • Variations in demographic associations of TNC use. Though income, race/ethnicity, edu- cation, unemployment levels, and housing tenure were included as independent variables, they rarely emerged as associated variables, and in no case did one of these variables appear to be associated across multiple regions. In no other region did one of these factors appear across multiple dayparts.