First lauched in 2013, Divvy has now become one the most successful bike share programs in the United States. Imgae source: City of Chicago official website.

I. Research Problem

My thesis project seeks to explore the integration of bike share and public transit in Chicago. The existing literature on bike share attests that bike share can complement public transit by serving as a solution to “the last mile problem” (DeMaio, 2009; Fishman, Washington, & Haworth, 2013; Faghih-Iman & Eluru, 2015; Griffin & Sener, 2016). Zellner, Massey, Shiftan, Levine, & Arquero (2016) provides a concise formulation of the last mile problem: “Transit that offers frequent and rapid service along the main lines but leaves the travelers a mile from their destinations with poor connecting options is rarely the mode of choice” (2). Regarding this problem, researchers and policy makers often state that bike share offers convenient as well as affordable last-mile access from and to public transit stations, which, in turn, can allow existing public transit systems to serve more people with little additional cost. The empirical support for this statement often relies on survey methods rather than the actual bike share uses (See City of Chicago, 2014, and LDA Consulting, 2015; also, Martin & Shaheen, 2014 for a peer-reviewed study). Although studies using observational data exist, such studies tend to explore broader patterns in the bike share usage; for example, Faghih-Imani & Eluru (2015) seeks to “quantify the impact of various attributes on bicycle-sharing system destination choices (p. 55). My project is an attempt to address this gap. Using the per-trip data of Divvy, a bike share service local to the Chicago area, this project will conduct a focused investigation into the complementary effect of bike share on public transit through multi-modal trips. In doing so, I also seek to present an alternative measure for the multi-modal bike trips made in connection with public transit.

II. Previous Research

Although the idea of bike share existed as early as in the 1960s, its popularity and prevalence has significantly increased only in the last decade (Fishman, Washington, & Haworth 2013; Fishman 2016). Accordingly, bike share is a relatively new topic for academic research, and there have been few studies that concentrated on measuring quantitatively the integration between bike share and other, traditional modes of public transportation.

A recent paper by Greg Phillip Griffin and Ipen Nese Sener, “Planning for Bike Share Connectivity to Rail Transit” (2016), is one such study. The authors acknowledge that “there have been relatively few studies quantifying bike sharing’s potential impact in facilitating transit trips” (Griffin & Sener, 2016, p. 2), and seeks the address the gap using a multi-method approach that combines “descriptive statistics, plan evaluation techniques, and semi-structured interviews of bike share system planners” for two bike share programs in Chicago, Illinois (Divvy) and Austin, Texas (B-cycle) (p. 7). Grittin and Sener’s evaluation of Divvy and B-cycle data suggests that there is only a “weak relationship” between bike share activities and rail stations. Nonetheless, the authors’ method for analyzing bike share data may not be complex enough, for it involves hardly more than a comparison between the number of “bike share embarks” within 400 meters from a rail station and that of all others. In addition, their choice of 400 meters as the standard for proximity, which is simply drawn from “a maximum distance suggested by operators for spacing between bike share stations” (p. 11), is unreliable at best.

Ahmadreza Faghih-Imani and Naveen Eluru’s “Analysing bicycle-sharing system user destination choice preferences: Chicago’s Divvy system” (2015) is another study incorporating quantitative assessment of the impact of bike share on public transit. The authors’ analysis is statistically sound and considers a wider range of variables potentially relevant to Divvy users’ choice of destination, including user attributes, trip attributes and destination attributes (See p. 59 for more on the statistical model). With respect to the relationship between Divvy use and public transit use, Faghih-Iman & Eluru (2015) suggests that “for regular members, the BSS [bicycle-sharing system] is likely to complement existing public transit services, whereas, for daily customers, the BSS serves as a substitute for existing public transit services” (p. 61). The study’s spatial variable is based on the 300-meter “buffer,” which still seems unrealistic for measuring the impact of public transit on the Divvy user’s choice of destination. Despite its shortcomings, with respect to its overall methodological soundness and choice of data, this paper offers an excellent model as well as a useful reference for my own project.

While the two studies discussed above examine Divvy data, Ma, Liu, & Erdoğan (2015) examines another case of bike share program, namely, Capital Bikeshare (CaBi) in Washington, D.C. The variables of Ma, Liu, & Erdoğan’s choice are divided into four broader categories: transit services, CaBi ridership, built environment, and socio-demographic characteristics. The result of its OLS regression analysis suggests that the correlation between an increase in bike share trips to and from stations nearby public transit stops and an increase in public transit rides is statistically significant. However, the proximity standard in this paper (a quarter mile or approximately 400 meters) is again unsatisfying to me and, consequently, the conclusion is not convincing. In brief, the current literature is short of a more realistic measure.

III. Hypothesis

The principal hypothesis to be tested in my project is that bike share serves as a solution to the last mile problem in public transportation as described above. This hypothesis can be restated in the following, more practical terms: For all trips made from and to Divvy stations in proximity with CTA stops, the likelihood of trips that are potentially multimodal, i.e., made in connection with the public transit rides, is greater than by random chance. (See the “V. Research Design” section below for a discussion on the measures for proximity and multi-modality.)

IV. Significance

Understanding the state of integration of bike share and public transit and developing an effective measure for the former’s complementary effect on the latter, which are two major goals of this project, have practical policy implications while contributing to the relevant literature as suggested above. An accurate measurement for this multimodal transportation will enable the City of Chicago to make more informed decisions on where to install new stations and how to integrate the bike share to other public transit services, which will contribute to the use of public transportation and the greater mobility of its residents. Such informed decisions may also lead to other benefits, such as less traffic on the road due to the increased public transit ridership, greater productivity of workers as well as more consumption due to the enhanced mobility as well as less time and resources wasted in transit, and less carbon emissions. Other municipalities with similar bike share programs in operation—such as Boston (Hubway), Denver (B-cycle), New York (City Bike), Philadelphia (Indego), and Washington D.C. (Capital Bikeshare)—can also benefit from my study’s example and develop effective measures for their own services.

V. Research Design

Divvy offers its per-trip dataset since its launching in 2013 freely on its webpage. This project will take observations for the year of 2016, which amount to total 3,595,383 trips. For each trip, the dataset offers 1) trip start and end time, 2) trip start and end stations, 3) rider type (member or 24-hour pass user), and, if it is a member trip, 4) the member’s gender and year of birth. The Divvy webpage also offers a separate file specifying the geographic information on each station. The Divvy data will be augmented by CTA data, which include the scheduled timetables for bus routes and rail lines as well as the locations of all bus and rail stations. When combined, these two sets of data allow me to estimate the availability of public transportation for all CTA bus and rail stations.

Put simply, my project seeks to compare the trips that are likely made in connection with public transit uses (i.e., multimodal) to the other trips that are not likely made in connection with public transit uses. More specifically, I first identify all Divvy stations that are located in reasonable proximity with public transit stops, i.e., that can be utilized as transit locations for multimodal (Divvy to bus or rail, and vice versa) trips. For the proximity standard, I consider the Manhattan distance of 100 meters (approximately 330 feet) to be an appropriate choice. This distance is equivalent to a half of a standard block (200 m) (City of Chicago 2007), and takes a little more than one minute to travel by the average human walking speed (i.e., 5 kilometers per hour). I believe that this provides a more realistic measure for identifying bike share stations that can be used for multimodal trips; to be certain, however, I will use and compare multiple proximity standards: 50 meters, 100 meters, 200 meters, and 300 meters.

Once these “close” Divvy stations are marked, I will identify the trips made to and from such stations. In doing so, I will exclude Divvy trips that took too long (e.g. over 20 minutes) and, therefore, were unlikely to be made as part of multimodal trips. Then, I will distinguish between trips made close to the arrival of buses and/or trains (the “treatment” group) and all the other trips (the “control” group). For example, if a trip is made to a Divvy station A, which is close to a bus stop B, shortly before (e.g., within 5 minutes) a bus is scheduled to come, this trip may be counted as a treatment group observation. However, if a trip is made to the same station, A, more than 5 minutes after a bus is scheduled to pass by B, the trip will be assigned to the control group, for it is less likely that such a trip would be intended as part of a multimodal transportation. Comparison between the two groups of trips will illuminate the characteristics of potentially multimodal trips. Concerning what other variables to consider in my analysis, I will critically consult with the studies discussed above in the “II. Previous Research” section. To validate the results from the estimated model, I will set aside a random sample of observations to be used exclusively for model validation.