Bike-Sharing Station Usage and the Surrounding Built Environments in Major Texas Cities

Abstract

This study analyzes the effect of different built environments on bike-share usage in nascent dock-based systems in three Texas cities. Past research offers little insight as to whether elements associated with higher bicycle usage in major cities affect ridership in secondary, developing bike-share markets within lower density American cities. In Austin and Houston, a positive relationship emerges between bike-share usage and proximity to high-comfort bicycle facilities. All three cities demonstrated surprisingly minimal relationship between bike-share usage and other proven drivers of bicycling activity in urban areas, which may result from system design for leisure- and recreation-based trips.

Introduction

Houston, San Antonio, and Austin rank fourth, seventh, and eleventh, respectively, on the list of the most populated cities in the United States (U.S. Census Bureau 2016). Historically, these three Texas cities have developed in an auto-centric fashion where between 80 percent and 95 percent of the working population use an automobile to commute each day (American Community Survey [ACS] 2016). Today, much attention is paid to the urban renaissance of American cities, spurring the densification and revitalization of downtown districts. This is undisputedly occurring in these cities and with this change comes the inherent demand for travel modes alternative to spatially inefficient ways of getting around, such as single-occupant vehicles (SOV). Provision of space for automobiles in these medium-sized American cities in terms of movement (right-of-way) and storage (parking) serves as an inhibitor to densification and development of urban forms designed around principles of livability, safety, and human interaction. Transportation-demand management strategies in many major U.S. cities have focused on expanding transit options, building safer bicycle facilities, and the implementation of bike-share systems, among other elements.

Bike sharing plays a critical role in the transportation systems of many American cities, serving both as a recreational and utilitarian resource. Station placement and planning of bike-share systems is not necessarily always a scientific process—planners look at various indicators qualitatively and quantitatively to help to optimize what are usually minimal financial resources. This study employs a geographic information system (GIS)-based kernel density function and a forward-stepwise-regression-based statistical analysis to explore the impacts of various built-environment attributes on average daily bike-share station usage in three major Texas cities. In doing so, we seek to determine if built-environment indicators proven to drive higher levels of local bicycling activity in major U.S. cities (e.g., proximity to transit stations, high population/employment densities, etc.) contextually apply to bike-share ridership in secondary, developing bike-share markets in lower density American cities. Presently, bike-share systems are expanding rapidly in these areas, but the existing literature offers little guidance as to whether lessons learned in New York or Chicago can apply to places such as Houston, San Antonio, and Austin.

In Houston and Austin, the model indicates that a positive relationship exists between bike-share station usage and proximity to high-comfort bicycle facilities; however, theoretically important variables such as population and employment density surprisingly did not prove statistically significant. San Antonio’s models returned less statistically compelling results. We argue that because these systems particularly cater to leisure- and recreation-based trips, the application of a model based on indicators traditionally associated with commute-based transportation attributes is inappropriate. The strongest predictive model findings were in Austin, debatably the system best geographically laid out to accommodate regular riders and commuters.

Bike Sharing and the Surrounding Built Environment

Dock-based¹ bike-share systems allow customers to temporarily check out a bicycle from a station and return it to any other station in the network for a small user fee and/or annual membership price. They exist in more than fifty cities in the United States (National Association of City Transportation Officials 2017) and serve as a vital part of the travel-demand management toolbox to help serve recreational, tourist-based, commuter, and first-last mile trips in urban environments.

Our research examines three of these systems—Houston, San Antonio, and Austin B-cycle. The bike-share systems within this study all have less than sixty stations and limited geographic coverage across these sprawling auto-oriented cities. Low cultural (and political) acceptance of the bicycle as a particularly important travel mode in Texas (at least relative to the automobile) has resulted in relatively low quantities of high-comfort cycling infrastructure in these areas as well. These environmental factors combine to produce a built environment that is largely not conducive to regular use of a bicycle as a primary travel mode, which indicates that recreational or non-home-based trips may be contributing the lion’s share of bike-share trips in these three cities. As such, the findings from this study are unique from those conducted in well-established systems in dense urban environments such as Washington, DC, and New York City. Several studies using different methodologies have taken a lens to bike-share system activity relating to factors within the surrounding built environments (e.g., Buck and Buehler 2011; Fuller et al. 2011; Handy et al. 2002; Mahmoud, El-Assi, and Habib 2015; Ma, Liu, and Erdoğan 2015; Noland, Smart, and Guo 2015; X. Wang et al. 2015). Population, housing, retail and employment density, prevalence of safe bicycle infrastructure, sidewalks, and transit routes are all a part of or a result of the built environment.

A survey conducted of Montreal residents with respect to the BIXI bike-share system—at the time of study, the largest in North America—found that 14.3 percent of respondents who lived within a 250-meter walk from a docking station reported having used the system, whereas only 6 percent of those living farther than 250 meters from a station reported having done so (Fuller et al. 2011). This indicates that station proximity to residential units plays an important role in its use. In general, higher population density correlates with higher levels of bicycling (Handy et al. 2002), so it seems logical to suggest that the more people proximate to a bike-share station should correlate with more station use. In New York City, trips by annual members are associated with residential population, but casual trips have no association with more populated areas; however, both casual and subscriber trips tend to be positively correlated with areas with higher employment density (Noland, Smart, and Guo 2015).

Trip purpose is another important indicator to consider when comparing ridership levels across cities with different urban built environments. Some cities and systems may be better suited to accommodate commuter-based trips, while others may be specifically targeted at recreational riders. A survey of four of North America’s biggest bike-share systems conducted in 2012 found that commute travel to/from work or school to be the most common trip purpose (Shaheen et al. 2012). However, research conducted by the Rice Kinder Institute suggests that many of the bike-share system trips in Houston are likely to be recreational rather than commuter-based; with 56 percent of ridership activity occurring on weekends, and 82 percent of rides coming from nonmembers (Houston B-Cycle 2017), this seems probable. Meanwhile, a 2012 membership survey of the Capital Bike share system in Washington, DC, indicates that the top reasons for bike-share system trips came in the form of “social/entertainment” or “errands/personal appointments” (LDA Consulting 2012), indicating that station proximity to population, housing, retail/restaurant, and employment density may serve as important built-environment factors. Moreover, a system evaluation conducted in Minneapolis, Minnesota, found a positive correlation with food-related business density in that the presence of one additional restaurant near a station resulted in 1.7 percent more bike-share station ridership, according to their predictive model (X. Wang et al. 2015).

Safety concerns are a major barrier to any type of bicycling activity. Although expansion of urban bicycle networks has risen to a priority in many American cities, the fact remains that there is still a lack of sufficient protected bicycle infrastructure and comfortable riding facilities in most places. Research conducted in Washington, DC, in 2011, found a statistically significant relationship between bike-share system activity and the presence of bicycle lanes (Buck and Buehler 2011). A study conducted in Toronto, Canada, showed that “for a given origin-destination pair, the higher the percentage of bicycle infrastructure with respect to the total route length, the higher the corresponding ridership” (Mahmoud, El-Assi, and Habib 2015, 2). Meanwhile, in New York City, a 2016 study found a statistically significant relationship with bike-share system ridership and proximity to bike lanes for weekend trips, but not for weekday trips (Noland, Smart, and Guo 2015). The same study found bike lanes to be predictive of casual trips, but with no effect on annual member trips (Noland, Smart, and Guo 2015), indicating that recreational users of the system may be more concerned about perceived safety of cycling infrastructure than regular subscription users. Houston B-Cycle’s 2016 Annual Report states that their most popular stations are located adjacent to high-comfort bicycle facilities, specifically those located in the newly renovated Buffalo Bayou Park (Houston B-Cycle 2017). A 2015 study of the Nice Ride bike-share system in Minneapolis found that bike-share stations proximate to parks and central business districts tend to have higher levels of bicycle activity (X. Wang et al. 2015).

Bicycling as an access mode has been largely considered a tool that can be used to increase transit ridership; however, some transit agencies fear that bike-share trips are taking away choice riders, particularly bus-based systems (Campbell and Brakewood 2017; Graehler, Mucci, and Erhardt 2019). An origin-destination study of the bike-share system in Washington, DC (CaBi), found that users tended to use the bike-share system to replace shorter trips that would be comparably more inconvenient to make via public transit, but that users still relied on the subway for longer trips (Ma, Liu, and Erdoğan 2015). In fact, CaBi ridership was found to be positively associated with Metrorail ridership, with A.M.-peak frequency as the strongest predictive variable (Ma, Liu, and Erdoğan 2015). A statistical spatial analysis conducted in both Chicago and Austin found that bus-stop proximity to bike-share stations did not yield statistically significant results, but that the permanence of (and higher ridership associated with) rail stations tended to bode more significantly in both planning decisions for station placement and ultimate ridership levels (Griffin and Sener 2016).

It remains apparent that the built environment is undoubtedly a contributor to mode choice and transportation planning in the realm of cycling. Most of the past research has been conducted in relatively mature bike-share systems located in dense urban environments. While a certain recreational element is certainly a factor in these systems, the vast majority of trips are likely made primarily for transportation reasons—to get from place to place—with recreational joyrides and exercise coming as secondary.

The systems focused upon in this study represent relatively nascent systems with small geographic reach and access to urban densities much lower than the likes of Washington, DC, or Montreal. Overall cycling activity in Texas pales to levels seen in these cities. According to the 2016 five-year ACS estimates, only about 1.4 percent of people in Austin commute via bike. That proportion is even lower in Houston (0.5%) and San Antonio (0.2%; ACS 2016). The built environment and differences in bicycling infrastructure certainly contribute to these numbers. All the aforementioned large systems (e.g., Chicago, New York, etc.) hold awards of Silver or higher from the League of American Bicyclists.² Houston and San Antonio both received the lowest award of Bronze. Citing increased levels of pedestrian- and bicycle-related fatalities in recent years, local cycling activist groups including BikeHouston and Bike San Antonio are critical of their respective city’s progress toward Vision Zero, calling for more protected bicycle infrastructure (BikeHouston 2017; Bustamante 2019; J. Wang 2019). Meanwhile, the League awarded Austin a Gold-level award (The League of American Bicyclists 2015) and Austin’s Transportation Department is often lauded as an exemplar of progressive bicycle planning in the otherwise auto-dominated American South (Shilton and The Bicycling Magazine Editors 2018).

To date, few studies have investigated bike-share system ridership in recreation-heavy systems. As such, these operators have needed to rely on qualitative measures and public engagement to help build-out their system (Griffin and Jiao 2018, 2019). Our research intends to fill this gap and focuses on the relationship between built environments and bikeshare station usage in cities with small-scale systems and low bicycling-commute-share levels.

Data and Method

Bike-share systems in the three cities are administered by individual 501(c)(3) nonprofit organizations. They operate with funding from government grants and private industry sponsorships/advertising agreements plus user fee revenue. Aside from stations sponsored by businesses, these organizations conduct analysis and public engagement to site stations to achieve optimal ridership and community equity metrics. Bike-share system operators in the three cities agreed to provide trip data in a .CSV file format displaying daily check-in and checkout activities for each station in Austin, San Antonio, and Houston. These data are described below in Table 1.

Table 1.

Overview of Three Bike-Share Systems in Austin, San Antonio, and Houston.

	Austin	San Antonio	Houston
Inception date	2013/12	2011/03	2012/05
Study period	2013/12–2017/07	2014/01–2016/11	2015/10–2016/10
Number of stations (during study)	55	55	33
Number of bikes	400	600	400
Average annual trips	191,095	131,072	110,730
Average daily system ridership	523.6	359.1	303.4
Short-term	$12 for 24 hours	$12 for 24 hours	$3 per 30 minutes
Medium-term	$18 for 3-day access	n/a	n/a
Monthly membership	$11/month plus $15 one-time fee	$18/month plus one-time $5 card fee	$9/month
Annual membership	$80	$100	$99
Weekday weighted avg. trip duration	27 minutes	22 minutes	49 minutes
Weekend weighted avg. trip duration	36 minutes	29 minutes	57 minutes
Weekday weighted avg. activity (checkouts + check-ins) per station	11.7	12.2	11.6
Weekend weighted avg. activity (checkouts + check-ins) per station	23.0	18.4	27.3

Source: Austin B-cycle, Houston B-cycle, and San Antonio B-cycle websites.

Daily Bike-Share Ridership by Station

We used Python (see Supplemental Appendix A) to parse through the origin-destination datasets provided by the bikeshare system operators in each city and geocoded the origins and destinations in GIS using ESRI ArcMap.

Average daily station-level activities were calculated by dividing the total usage (bike checkouts plus returns) at each station by the number of days that station has been in continuous operation. Figures 1, 2, and 3 on the following pages represent the station distributions in each city. The size of each circle corresponds to the proportional average daily activity at each stop. The top ten stations in terms of daily activities are marked in the figures and presented in the following Table 2.

Figure 1.

Average daily bike-share system activity by station in Austin, Texas.

Source: Austin B-cycle data from December 2013 to July 2017.

Figure 2.

Average daily bike-share system activity by station in San Antonio, Texas.

Source: San Antonio B-cycle data from January 2014 to November 2016.

Figure 3.

Average daily bike-share system activity by station in Houston, Texas.

Source: Houston B-cycle data from October 2015 to October 2016.

Table 2.

Top Ten Bike-Share Stations by Average Daily Station Usage (12/13–07/17).

City	Austin (12/13–07/17)		San Antonio (01/14–11/16)		Houston (10/15–10/16)
Top 10 stations	Station name	Avg. daily usage	Station name	Avg. daily usage	Station name	Avg. daily usage
1	City Hall/Lavaca & 2nd	42.2	Blue Star	42	Sabine Bridge	86.8
2	Riverside @ S. Lamar	42	Mission San Jose	35.9	Hermann Park Lake Plaza	67.3
3	Zilker Park	41	Pearl Emma (Brewery)	34.3	Spotts Park	46.5
4	2nd & Congress	38.9	Mission Concepcion	29	La Branch & Lamar	38.7
5	Rainey St. @ Cummings	38.6	Concepcion Park	27.4	Jackson Hill & Memorial Dr.	31
6	4th & Congress	35.1	Mission San Juan	27	Stude Park	29.8
7	Convention Center/4th St. @ MetroRail	33.5	Ellis Alley	23.1	Market Square	29.5
8	5th & Bowie	32.9	Roosevelt Park	21.4	Lamar & Crawford	27.2
9	Davis at Rainey Street	31.8	423 Blue Star (Big Tex)	20.2	Smith & Capitol	20.9
10	South Congress & Barton Springs	29	La Villita	19.7	West Gray & Baldwin	18.7
System wide	Station Mean/S.D.	19.0/10.6	Station Mean/S.D.	13.1/8.7	Station Mean/S.D.	18.4/18.4

Source: Austin B-cycle, Houston B-cycle, and San Antonio B-cycle websites.

Looking at the Austin bike-share system ridership map, it becomes apparent that ridership is concentrated in the downtown area (1, 4, 6, and 7) and along the waterfront of Lady Bird Lake (2, 3, and 5), where high-quality bicycle infrastructure and trails connect many destinations including Zilker Park (2), central downtown (4) and South Congress Avenue (10), the Austin Convention Center (7) and the Rainey Street district (9), among others. The top three stations all featured average daily station usage counts above forty, and, on average, each station in Austin is used nineteen times per day. Weekend bike-share trip durations average just above thirty-five minutes, about nine minutes longer than the average weekday trip.

The configuration of the system provides a high density of station locations spread throughout the downtown and surrounding areas that allows for both north-south and east-west connections on high-quality bicycle infrastructure much of which has been improved in the last few years as a part of the Austin Bicycle Master Plan build-out. While, stations proximate to Zilker Park have relatively high daily ridership and a higher proportion of “round-trip” trips—an indicator of recreational or joyriding—the vast majority of trips are spread across the system and concentrated in downtown, indicating that point-to-point transportation—whether leisure- or commute-based—is likely the purpose of most trips in the system. In addition, station usage proximate to Zilker Park is likely skewed due to special events occurring there throughout the year. Austin City Limits music festival attracted 75,000 people per day over the six-day span of the event in 2016 (Anderson 2017) and accounts for 5.1 percent of all bike-share activity in Austin over the course of our study period (Ogura 2018). Other festivals such as the Trail of Lights and Fun Fun Fun Fest attract similar attendance numbers to the park area (Rogers 2015). During these large festivals, the Austin Police Department shuts down vehicular access to the park, and, as a result, bikeshare usage spikes during these periods—so much so that the bike-share system operator assigns staff to operate the stations near the venue in a valet-type fashion. Similarly, South by Southwest (“SXSW”), a three-week-long amalgamation of music, interactive media, and film festivals and conferences brought more than 400,000 people to downtown Austin in 2017 (Gallaga 2017). Due to peak congestion issues, traffic and transit detours, and other factors, many conference attendees choose to travel via bike share between events leading to spikes in bike-share ridership during the month of March, particularly around the Austin Convention Center. Overall, SXSW events account for 9.5 percent of all bike-share activity across our study period. Average daily bike-share usage amounts to more than 1,700 trips per day during this time, as opposed to 440 trips per day across the other eleven out of twelve months of the year (Ogura 2018).

San Antonio’s bike-share system is aligned along the San Antonio River connecting a more than thirteen-mile stretch of pathways running from Missions San Francisco de la Espanada and San Juan Capistrano in the South (2 and 6) all the way through downtown (7 and 10) to the Pearl Brewery (3) and San Antonio Zoo in the North. As Figure 2 shows, the majority of ridership exists along this linear corridor, which features little continuity in terms of surrounding built environment. Like Austin, the San Antonio system also features fifty-five stations; however, this system is designed predominantly to capture leisure and recreational trips along the riverfront with a secondary focus on commuter or utilitarian transportation trips. From a ridership perspective, the system is comparable with the bike-share system in Austin but features less average daily station usage, with thirteen versus the nineteen in Austin. Weekend bike-share trip durations in San Antonio average just above twenty-nine minutes, about seven minutes longer than the average weekday trip.

The Houston bike-share system features a handful of stations in downtown (4, 7, 8, and 9) with hotspots around transit stations and the museum district proximate to the George R. Brown Convention Center (4) and Discovery Green (8); however, the majority of ridership in Houston’s B-Cycle system takes place in peripheral stations located in public parks (1, 2, 3, 5, and 6). While the system has fewer stations than Austin and San Antonio, the average daily use per station of eighteen is similar to both of the other systems. Houston, however, also features the highest standard deviation among average daily station usage among the three cities, indicating a high level of variability between heavily used stations and those with little to no average daily usage. The system is also much more popular on weekends, averaging nearly three times the average station activity compared with weekday levels.

The most popular stations are those located at off-street locations in Buffalo Bayou Park (1, 3, and 5) and Stude Park (6) to the west of downtown and Hermann Park (2) at the very southern end of the system. These five stations, which alone constitute 43.1% of daily average system ridership, also feature large proportions of round-trip usage indicating recreational riding. This finding potentially owes to the renovation of Buffalo Bayou Park and its trail system completed in the fall of 2015. Furthermore, Houston’s average weekend and weekday bike-share trip duration of fifty-seven and forty-nine minutes, respectively, nearly doubles the values of Austin and San Antonio.

Comparative Discussion of Initial Findings

Upon initial comparison of the station usage results among the three cities, we found that Austin’s Zilker Park station, San Antonio’s linearly aligned Riverwalk station network, and Houston’s Buffalo Bayou, Stude, and Hermann Park stations all sit at the top end of popular stations in these cities, particularly when isolating weekend trips. Still, prevalence of high-quality bicycle infrastructure in Austin provides for a spread of popular stations throughout the downtown area, indicating that whether the trips are being taken for leisure or commute purposes, they are predominantly being taken to get from point-to-point in the general downtown area. Meanwhile, in San Antonio, station-usage levels are so highly concentrated along the linear thirteen-mile Riverwalk that it is difficult to make the case that people are using the system for any purpose other than recreation or sight-seeing. Similarly, in Houston, where such a large portion of system-wide activity is contained within parks, the same hypothesis arises. As such, traditional theoretical predictive built-environment indicators such as population, housing, and employment density may prove to be inconclusive predictive variables in these places.

Built Environments around Bike-Share System Stations

To better understand the bike-share system in Austin, San Antonio, and Houston, we analyzed the relationships between average daily bike-share station usage (dependent variable) and the built environments around these stations (independent variables). Presented in the following Table 3 are different built-environment variables and their sources for those cities.

Table 3.

Built-Environment Variables and Data Sources.

Built-environment variables	Austin	San Antonio	Houston
Population per Census BG	1	1	1
Housing units per Census BG	1	1	1
Non-SOV commuter mode count per Census BG	1	1	1
Ratio of renters to home owners by Census BG	1	1	1
Employment density	2	2	2
Restaurant density	3	no data available	no data available
Length of proximate high-comfort bikeways	6	7	4
Length of proximate high-comfort or medium-comfort bikeways	6	7	4
Length of proximate sidewalk	3	5	no data available
Area of proximate parkland acreage	3	5	4
Major road density	3	5	4
Minor road density	3	4	5
Walkscore	8	8	8
Transitscore	8	8	8
Bikescore	8	8	8
Transit route density	9	no data available	10

Source: I. U.S. Census—American Community Survey (2015—5 yr. estimates); 2. EPA Smart Location Database (2010); 3. City of Austin Open Data Portal—https://data.austintexas.gov/; 4. City of Houston Open Data Portal—http://data.houstontx.gov/; 5. City of San Antonio GIS Data—http://www.sanantonio.gov/GIS; 6. Austin Transportation Department—Active Transportation and Street Design Division; 7. Alamo Area Metropolitan Planning Organization GIS Data—http://www.alamoareampo.org/GIS/; 8. Walkscore.com; 9. Capital Metropolitan Transportation Authority—https://data.texas.gov/capital-metro; 10. Houston METRO—https://www.ridemetro.org/Pages/NewsDownloads.aspx.

Note: BG = Block Group; SOV = single-occupancy vehicle; EPA = Environmental Protection Agency; GIS = geographic information system.

Kernel Density Function

The relatively small spatial extent of these bike-share systems falls into only a handful of different Census block groups (CBG). Because CBGs are the most granular geography for which demographic data can readily be obtained, this posed an issue in that it limited the amount of dependent variable variation between stations. CBGs are statistical divisions of Census tracts typically containing between 600 and 3,000 people, with an optimal value of 1,500 (U.S. Census Bureau 2016). Because population density is relatively low, even the densest downtown districts are comprised of only a few CBGs, and, as such, most of the bike-share systems only span a handful of block groups in and around these downtown areas. As such, if a plurality of stations exists in just one CBG, the data will be limited in variation, attributing the same demographic indicator values to all station locations within this CBG.

To mitigate this issue and compare demographic data with other built-environment indicators examined in this study, we employed a kernel density function, a spatial analysis tool built into ESRI’s ArcMap software. This method’s applications are numerous, ranging from better estimating geographic customer densities for marketing purposes to predicting crime using Twitter data (Donthu and Rust 1989; Gerber 2014). Past studies have used kernel density to “calculate robust measures of exposure to one or more environmental features” (Thornton, Pearce, and Kavanaugh 2011, 6). Facing a similar data constraint challenge, Tessa Anderson (2009) weighed kernel density-based and Census area estimation-based methods for mapping road accident hotspots in London. Handy and Niemeier (1997) specifically employ kernel density to define a gravity model of accessibility. Similarly, we are looking to correlate proximate built-environment indicator exposure and accessibility to bike-share stations. The kernel density function takes either a point or a line dataset and fits a curved surface over these data points/lines. The surface has the highest value at the location of each point or line and diminishes eventually to zero with increasing distance from each point or line feature in the class. Thus, kernel functions will create continuous surfaces for variables based on the existing data and allow us to better measure the built environments around each bikeshare station.³

To input the polygon-based data into the kernel density calculation, a centroid point of each polygon group was created. Because line-based shape files can already be input into the kernel density function, they did not have to undergo any preprocessing. Each point or line can be given a weight value. For example, when running a kernel density calculation for population density, centroid points representing CBGs with more people would have a higher weight in the generated rasterized topography. Similarly, park area density is weighted by park acreage so larger parks have a larger effect than smaller ones. Restaurant locations and presence of bikeways, roads, sidewalks, and transit routes were analyzed on a binary scale with a weight of 1.

The specified search radius for each kernel density function was 400 meters in all cases except for high-comfort bikeways,⁴ high-comfort and medium-comfort bikeways,⁵ and parkland area. When compared with the other built-environment indicators in each downtown area, these three indicators were less dense than sidewalks, streets, and CBG-based data. These three indicators were hypothesized to represent heavy drivers of bike-share ridership based on theory and past research. As such, the search radius for these is 1,600 meters.

The resulting rasterized datasets present an interpolated estimate of what a particular density value would be for each indicator within cells that measure fifty by fifty meters across the study area. Using a function to extract the raster values for each indicator at each station location allows us to build-out our dataset of independent variables (see Supplemental Appendix B) to compare with our dependent variable of average daily bike-share system activity for the following statistical analysis.

Last, a few datasets in the analysis did not require a kernel density analysis. Walkscore, bikescore, and transitscore data are sourced from walkscore.com based on the address of each bike-share station. Linear distance to landmarks is generated using a linear distance function in ArcMap.

Stepwise Multiple Variable Regression

In comparing the average daily ridership of each individual station area with the myriad of indicator kernel density values generated at each location, we first generated a bivariate correlation table (see Supplemental Appendix C). Due to the high level of collinearity among the independent variable indicators, we employed a stepwise regression function that analyzes each candidate predictor variable in the model individually and returns only those that allow each individual independent variable’s coefficient included within the model to achieve a 95 percent level of confidence. Three sets of regression analysis for each city were carried out to model (1) all days, (2) weekends only, and (3) weekdays only. The following Table 4 presents these stepwise regressions for each city.

Table 4.

Austin, San Antonio, and Houston Model Results.

Dependent variable	Independent variables	Austin (N = 55)		San Antonio (N = 55)		Houston (N = 32)
		Coefficient	Sign.	Coefficient	Sign.	Coefficient	Sign.
Average bike-share activity by station (all days)	Intercept	41.34	n/a	21.03	n/a	−30.6	n/a
	High-Comfort Bikeway Density	3.36	0			15.36	0
	Distance to Convention Center	−0.01	0
	Sidewalk Density	−1.11	0
	Walkscore			−0.12	0.03
	Adjusted R 2	.5		.07		.41
Average bike-share activity by station (weekend only)	Intercept	82.08	n/a	35.17	n/a	−4.65	n/a
	High-Comfort Bikeway Density	2.86	0.08			17.48	0
	Distance to Convention Center	−0.01	0
	Sidewalk Density	−1.52	0
	Walkscore	−0.3	0.04	−0.25	0
	Park Density					1.17	0.04
	Adjusted R 2	.43		.16		.46
Average bike- share activity by station (weekday only)	Intercept	23.36	n/a	21.76	n/a	−2.86	n/a
	High-Comfort Bikeway Density	3.02	0			7.83	0
	High-Med. Comfort Bikeway Density	−6.08	0
	Walkscore	−0.14	0	−0.14	0
	Major Road Density	3.63	0
	Park Density					0.53	0.04
	Adjusted R 2	.45		.27		.48

Note: Italicized values are significant at 0.05 level.

Results

The Austin model for all days of the week generated an adjusted R² value of ~.5, which indicated a relatively good fit. Three independent variables were significant in the model. The average daily station activities were positively related to the density of high-comfort bikeways around the station and negatively correlated with the distance to convention center as well as the sidewalk density around the station. The weekend-only model generated an adjusted R² value of .43 and featured the same relationships along with a negative correlation with walkscore. Meanwhile, the weekday-only model returned an adjusted R² value of .45 and shows positive coefficients for high-comfort bikeway density and major road density and negative coefficients for walkscore and high-medium comfort bikeway density. The stepwise regression analysis eliminated all other built-environment indicators due to poor predictive abilities.

For the San Antonio model, most of the independent variables were not significant in the model except for walkscore, which was negatively correlated with the daily bike-share station usage. The model also had a very low adjusted R² value of .07, indicating a poor fit. Confining the independent variable by weekday versus weekend tended to marginally improve the model’s fit results to .27 and .16, respectively.

Houston’s model for all days had a better fit than the San Antonio model but the adjusted R² value was still relatively low for predictive purposes (.41) with only one statistically significant explanatory variable: high-comfort bikeway density. When segregating the model by weekday versus weekend trips, the model’s adjusted R² increased to .48 and .46, respectively, and displayed positive coefficients for park density and high-comfort bikeway density.

Overall, across all three models, indicators such as population density, housing unit density, employment density, proximity to transit, and bikescore—all values that past research and planning theory would uphold as valuable predictive factors of bicycle use—are not particularly telling factors in the case of these systems.

Discussion

Austin, Texas

Austin has gone to great lengths to expand its high-comfort bikeway network and build-out a system to accommodate all ages and abilities⁶ through expansion of on-street protected bikeways and off-street multiuse paths and trails. Based on this analysis, proximity to these high-quality bike facilities correlates positively with bike-share activity. This is a positive indication that street design does matter for bicycle ridership, especially bike-share systems catering to riders spanning all ages and abilities. In Figure 4 (left), darker shades of blue indicate a higher density of the presence of high-comfort bicycle infrastructure such as fully protected on-street cycle tracks (e.g., 3rd Street), off-street trails, and shared-use pathways. Visually, it is apparent that the larger station activities tend to be located in areas with higher kernel density values for this indicator.

Figure 4.

High-comfort bikeway (left) and sidewalk (right) kernel density in Austin.

The regression analysis also indicates a negative correlation with sidewalk density and walkscore. Sidewalk presence throughout most of the urbanized system area is relatively homogeneous. From the Figure 4 (right), we can see the density of sidewalk infrastructure is highest to the northeast of most of the bike-share system ridership. This could simply be a factor of high bike-share ridership in open spaces and park areas where sidewalk density is comparatively low (e.g., Zilker Park and along Lady Bird Lake) compared with the tight-square grid of downtown. This finding should not be construed to indicate that bike-share planners should specifically look for areas lacking sidewalks to locate stations. If people cannot safely access bikeshare stations on foot, this will lead to low station usage.

The analysis indicates a negative correlation with linear distance to the Austin Convention Center, which is a focal point of bike-share ridership in the system. As such, bikeshare ridership trends downward as the distance between the station location and this landmark increases. Austin has promoted its convention center as an attractor for conferences and other events, the largest of which occurs in March of each year: SXSW. This event transforms downtown Austin and the surrounding area for nearly a month’s time. During this time, the bike-share station adjacent to the convention center is operated in a valet-type fashion and manned with bike-share system employees due to the massive demand. Bike-share system ridership spikes as road closures, transit detours, and crippling congestion make it difficult to get around using other modes. This event alone may be the explanation for this factor rising to be part of the statistically significant model.

The prevalence of a positive correlation with major road density and the lack of a park-related independent variable correlation in the weekday model is also a telling finding for the Austin case. This indicates that bike-share trips are being made throughout the system and are not mostly confined to parks as the results from San Antonio and Houston would suggest. Perhaps the extensive network of high-comfort bikeways in Austin serves as a suitable parallel travel route within the 400-meter search radius of these major arterial roadways allowing for more utilitarian point-to-point travel on weekdays. Still though, median bike-share station activities are 78 percent higher on weekend days versus weekdays—with stations proximate to Zilker Park ringing in the top two spots. And average trip durations also tend to be higher on weekends—thirty-six minutes versus twenty-seven minutes on weekdays. While it is clear that the Austin system is heavily influenced by spikes in ridership related to special events and recreational riding occurring on weekends, the spread of overall activities throughout the system indicates that a portion of trips are made for utilitarian point-to-point transport. However, the lack of correspondence with traditional metrics of bicycling viability such as population and employment density shows that this system’s ridership patterns follow a different mold than those associated with major dock-based systems in the United States.

San Antonio, Texas

The low adjusted R² values of .07, .16, and .27 for San Antonio’s models indicate that only one indicator—walkscore—has a weak negative relationship with the dependent variable. This could perhaps be attributed to the fact that San Antonio’s bike-share system follows a riverfront park where address-based walkscore values are artificially low due to lack of development and proximity to high-speed and nonpedestrian-friendly roadways, among other factors. In addition, the way that the San Antonio bike-share system is laid out, such that it only really allows for north-south movement on a single corridor, is inherently limiting. This paired with relative lack of high-comfort bicycle infrastructure in the downtown area make using the system for point-to-point utilitarian travel inconvenient and uncomfortable. Therefore, much of the bike-share usage is confined to off-street trails heading north and south of the downtown area. These areas do not correspond with theoretical built-environment indicators such as population, employment, and housing density that traditionally lead to higher bike usage. In fact, the univariate Pearson correlation coefficients for all three of these indicators produced negative values of −.06, −.21, and −.23, respectively, indicating that perhaps recreational bike-share customers actually prefer less dense environments.

Like Austin, median average station activity increases by more than 50 percent on weekends versus weekdays, and average bike-share trip durations tend to be longer on weekends as well. All of this evidence points toward San Antonio’s bike-share system being patronized mostly for the purpose of leisure and recreational bicycle trips. The low level of the model’s ability to predict ridership based on these built-environment factors is a finding in and of itself. We hypothesize that the difficulty in predicting station activity in a system such as San Antonio’s stems from the fact that it is primarily tailored to recreational trips that may or may not involve the goal of point-to-point transport. In a situation where a rider checks out a bike and rides it up and down the riverfront before returning it at the original or another nearby station, it is inappropriate to apply a transportation-planning model based on theoretical correlations with built-environment indicators that would tend to correspond with utilitarian bicycling activity to try and predict where these trips will be made.

The regression results for this analysis show minimal relation to the built-environment variables that theoretically correspond with higher bicycle usage. As such, it is hypothesized that recreational riders are not necessarily concerned with many of the factors analyzed and tend to focus on the act of cycling more so than the destination or the journey. Thus, from the perspective of a bike-share system planner, determining where the next station in the system should go cannot be quantitatively determined solely from the use of a transportation theory-based model rooted in attributes commonly known to be closely related to bicycling activity in general. Instead, planners must look toward public engagement and other qualitative methods to achieve the best results.

Houston, Texas

The adjusted R² values for Houston’s models (.41, .46, and .48) indicate that two indicators—high-comfort bikeway density and park density—have statistically significant positive relationships with the dependent variables. This pairs with the fact that 43.1 percent of system ridership takes place at the top five most popular stations, which are located at off-street locations in parks with access to high-comfort, off-street pathways. Median bike-share activity by station in Houston increases by nearly fivefold on a given weekend compared with weekday values (see Supplemental Appendix D). Furthermore, average weighted trip duration by station in Houston tops all three cities at fifty-seven minutes on weekends and forty-nine minutes on weekdays. Activity spikes and longer trip durations on weekends paired with the popularity of parks-based stations indicate a high prevalence of leisure and recreational trips. Similar to the system in San Antonio, a majority of total system ridership is concentrated at stations within or along parkland. Thus, until the system build-out and bicycle lane infrastructure in the urbanized area increases to a level that results in more utilitarian or commute-based trips, traditional models looking at built-environment indicators that pertain to transportation planning will continue to fail to adequately predict station activity levels.

Applications to Planning Practice and Future Research

Based on the three Texas cities studied, fitting a predictive bike-share station usage model based on theoretical built-environment indicators does not work as well when compared with results found in other studies examining the country’s largest bike-share operations. The systems in San Antonio and Houston, situated in urban environments that are not particularly well-designed for safe and comfortable bicycle riding for people of all ages and abilities, tend to focus more on recreational and leisure trips in and around major parks. And even though Austin’s model indicates that a portion of weekday usage is certainly utilitarian-based, certain attributes of the data and resultant regressions indicate a high prevalence of leisure and recreational trips as well. As such, these types of recreationally focused operations require a different set of indicators altogether, perhaps more focused on tourism factors such as attractions, beauty, and proximity to hotels or available parking.

While both the models in Austin and Houston did generate a sufficiently reliable adjusted R² value indicating a positive correlation with high-comfort bicycle infrastructure, Houston’s ridership is so focused within parks that it is difficult to make the case that bike share is a viable commuting or utilitarian mode there. Unlike Houston and San Antonio, the League of American Bicyclists has heralded Austin with a Gold-level community award, influenced in part by the city’s robust plans for “All Ages and Abilities” bicycle networks and the transportation department’s aggressive implementation of high-quality bicycle infrastructure. The results of our study indicate that these facilities have been successful in that their presence attracts more bike-share usage. Therefore, despite the nascent maturity and small geographic scale of the system itself, our research indicates that the presence of a network of high-comfort bicycle facilities in Austin—not just a few disconnected sections and park trails as is the case in San Antonio and Houston—has been a positive element of higher bike-share ridership. While the construction of an entire network of high-comfort bicycle facilities is an expensive community decision, our research indicates that once a good portion of the network is in place, it functions to bolster bike-share system ridership. The planning community advocates that, ultimately, this positive relationship and the presence of more viable travel alternatives will lead to more sustainable travel patterns and healthier communities (see Griffin and Jiao 2015).

Conclusion

The most notable statistically significant finding from this study remains the positive relationship between bike-share station ridership and proximity to high-comfort bicycle facilities in Austin and Houston; however, this research also illustrates which built-environment indicators curiously do not hold substantial sway over bike-share activity in developing bike-share systems in Texas. The fact that population and employment density, proximity to transit, bikescore, and presence of non-SOV commuters did not factor into any of the nine models analyzed across Austin, San Antonio, and Houston is a finding in and of itself. Part of this, of course, is a fixture of the absence of precise, granular data, but the lack of findings also indicates that planning a relatively small-scale bike-share system in a medium-sized, auto-oriented American city is inherently very difficult. The standard indicators that help to guide active transportation and transit planners—things as elementary as population, housing, retail, and employment density—proved insignificant predictive variables in all three cities. Therefore, planning of these systems may require reliance on qualitative and tactile methods to perfect system balance and optimize station locations for ridership potential. In addition, systems focusing on recreational-type trips will need to employ an entirely different mode of quantitative planning, one involving experiential factors such as beauty, scenery, and parking availability, among other things, to develop a model focused on recreational bike trips, which inherently do not adhere to the principles of traditional point-to-point transportation planning.

Biographies

Louis G. Alcorn is a graduate student in the Community and Regional Planning (CRP) program at the University of Texas at Austin, pursuing a dual degree in Transportation Engineering. His research interests focus on leveraging data to produce more robust measures of transportation equity.

Junfeng Jiao is an associate professor and the founding director of the Urban Information Lab at University of Texas at Austin’s School of Architecture. His research focuses on shared mobility, smart cities, and transit access.