Infrastructure, Social Practice, and Environmentalism: The Case of Bicycle-Commuting

Abstract

Automobile travel is a major contributor to US greenhouse gas emissions. Reducing car travel could slow climate change, but this goal is made difficult by the constellation of economic, political, cultural, and infrastructural forces that form a mutually reinforcing “system of automobility.” This complexity suggests the need for an approach that attends to both the reciprocal relationship between infrastructure and practice and the moderating influence of cultural factors, but policies to encourage sustainable practices rely on asocial conceptions of behavior. With longitudinal data on bicycle-commuting in sixty-two large US cities from 2000 to 2014, I use fixed-effects regression and structural equation modeling to perform two strong tests of whether the expansion of bikeway facilities induces more bicycling, and whether this effect depends on the strength of environmentalism in a city. I also use structural equation modeling to perform a preliminary test of the reverse causal effect—whether bicycle-commuting can spur bikeway creation, again allowing for a moderating effect of environmentalism. My results add to existing evidence that bikeways induce bicycling, and provide preliminary support for the reverse effect, but also indicate that both effects are moderated by local environmentalism. Implications for the persistence of auto-dominance are discussed.

Automobile transportation is a significant contributor to global climate change. In the United States, transportation is the second largest source of greenhouse gas emissions, accounting for 28 percent of all emissions in 2006 (USDOT 2006). Almost two-thirds of these transportation emissions come from passenger cars, small trucks, SUVs, and vans.

Meeting emissions targets will require a combination of improved fuel efficiency, increased use of alternative fuel, and reductions in vehicle-miles traveled. However, automobile travel is bound up with other practices, cultural meanings, technology, infrastructure, and economic interests, which together constitute a mutually reinforcing “system of automobility” (Urry 2004). Automobile, petroleum, and related industries are politically influential, given their size and contribution to US economic activity. Americans are attached to their cars and the practice of driving in ways tied up with identity, status, and comfortable habit. US cities are sprawling, with development extending far from the city center, facilitated by freeway networks that make autonomous long-distance travel possible. Land uses are often segregated, lengthening trips to shopping and work.

Despite the complex social character of automobile travel and other carbon-intensive practices, policies meant to change these practices rely on an asocial conception of behavior (Shove 2010). As Szerszynski and Urry (2010) observe:

the argument of the science itself is that human practices are utterly central to [climate change] and that the only possible way of “mitigating” potentially catastrophic change, apart from vast and improbable geo-engineering projects, is through transformed human practices. So the social is both central and yet pretty well invisible. (3)

Efforts to move beyond automobility must understand the connections between its parts. Because the system is self-reproducing, it cannot be displaced through isolated change in any one part. However, Urry (2004) points to emerging developments that could combine to create a “tipping point” and usher in a new system. One such development is the shift away from transport-planning policies that built ever more roads in anticipation of increased driving. Instead, planners seek to slow demand for roads, including through improvement of facilities for walking and bicycling.

These “demand-reduction” policies offer an opportunity to examine the relationship between transportation infrastructure and travel practices. Transportation researchers have begun to show how the built environment affects travel behavior (Ewing and Cervero 2010; Cervero 2003), but little work investigates how built environments are shaped. Does infrastructure induce or rather respond to changes in practice, or are both effects operative?

The present study uses longitudinal data on bicycle-commuting and bicycle infrastructure in sixty-two large US cities to analyze the relationship between travel practices, infrastructure, and other place characteristics. The analysis uses fixed-effects regression to control for unmeasured, time-invariant characteristics of cities, as well as a technique proposed by Allison (2014) that controls for potential reciprocal effects between bicycle-commuting and bicycle infrastructure. In the next section, I review research on the relationship between infrastructure and behavior. This research suggests three hypotheses, which I state in the subsequent section. Next, I review methodological issues in research on transportation infrastructure and behavior, and describe the dataset and statistical analyses. Finally, I describe the results and conclude by discussing implications for automobility.

Past Work on Infrastructure and Social Practice

Past research on infrastructure and social practice spans the fields of urban sociology, social studies of technology, economics, transportation planning, and geography. Nonetheless, research on these topics coalesces around two opposing viewpoints. The demand-driven perspective is that infrastructure is shaped by consumer preferences and technological development. Consumers have desires, technological development enables the increasingly efficient satisfaction of those desires, and infrastructure is built to facilitate social practices that use the best available technology. The political perspective disputes the notion that infrastructure is shaped by consumer preferences, arguing instead that infrastructure is shaped by political processes, which often favor elite interests. Per this view, infrastructure is not a mere reflection of social practice, but is a factor influencing recruitment to and defection from practices. The demand-driven perspective predicts that infrastructure will be built to the extent that there is demand for it—bicycling will induce the creation of bicycle infrastructure. The political perspective predicts that the creation of bicycle infrastructure can induce bicycling, but that the influence of infrastructure on social practice—and vice versa—depends on contextual factors.

Infrastructure as Demand Driven

Demand-driven approaches to infrastructure are united by the assumption of strong biological imperatives. As living organisms, human beings have a fixed set of needs, and strive, individually and collectively, to satisfy those needs as efficiently as possible. Apparent variation in consumer tastes, between individuals and over time, is better understood as variation in prices and incomes relevant to satisfying tastes that are stable between individuals and over time (Stigler and Becker 1977).

Human ecologists apply this assumption of strong biological imperatives to the population level. A social system is primarily a “producer of sustenance…the entire system in all of its aspects is a producer of the wherewithal for life” (Hawley 1986, 8–9). Populations seek to grow “to the maximum size and complexity afforded by the technology of transportation and communication” they possess (Hawley 1986, 7). For economists and human ecologists, technological progress is a central force in social change, as new technologies allow more efficient satisfaction of needs and expansion of populations. The ends are fixed by individual- and population-level imperatives, and technology provides the means of achieving those ends.

This perspective is exemplified by Adams's (1970) influential analysis of urban morphology. By examining historical construction records, Adams shows that the boundaries of development in the Minneapolis urban area were concentric during the “walking and horse carriage” and “recreational automobile” eras of transportation technology, but exhibited irregular shapes during the “electric streetcar” and “freeway” eras, influenced by the locations of streetcar lines and freeways, respectively. Adams interprets this finding in terms of residents’ desire to move farther out of the city while still being able to access employment and amenities, with each successive transportation technology enabling the expansion of housing development.

In this narrative, market forces shape infrastructure. In fact, during the streetcar era, streetcar lines were built by private companies to enable housing development on the cheap land at the urban fringe (Muller 1995). As automobile ownership grew, road-building was increasingly publicly financed, complicating a market narrative. However, scholars in the demand-driven camp argue that publicly funded road construction did not create demand for automobiles, but merely facilitated the expression of that demand (Anas, Arnott, and Small 1998).

Much modern infrastructure is created by public agencies that may be insulated from market forces. From the demand-driven view, any political influence corrupts infrastructure planning, squandering resources on nonviable projects. Thus, the political attractiveness of rail transit in the 1970s and 1980s, or the “Desire Named Streetcar” (Pickrell 1992), led to costly new rail lines that failed to meet ridership projections (Meyer and Gomez-Ibanez 1981; Fielding 1995). The failure of these projects to attract sufficient usage is taken as evidence of their political origins. From the demand-driven perspective, infrastructure not designed in accordance with consumer preferences will be unsustainable, because infrastructure cannot change the preferences that drive behavior.

Given the fixed concept of human motivation undergirding the demand-driven perspective, technological development serves as the main source of social change. Social change follows the course of technological development along the lines of increasing logic and efficiency in satisfying human needs (Bimber 1994). Thus, technology structures social life, but technological development is structured by human needs—the technologies that most efficiently satisfy those needs win out and come into widespread use, creating demand for infrastructure that facilitates use of those technologies. Changes in infrastructure do not cause changes in practice, but rather result from them.

Infrastructure as Politically Driven

The demand-driven perspective on infrastructure rests on three propositions: (1) Behavior is driven by fixed human needs; (2) Technological development follows a path of increasingly efficient satisfaction of human needs; and (3) Infrastructure is designed to facilitate social practices that use the most efficient technologies available; infrastructure that does not do so is unsustainable. Proponents of the political perspective on infrastructure dispute all three of these propositions. They view the relationship between human needs and behavior as socially negotiated, and they view technological development and infrastructure design as contingent results of political struggle and thus shaped by power. These opposing propositions regarding motivation and technology lead to an opposing view of the relationship between infrastructure and behavior. Because behavior, technology, and infrastructure are contingent results of socio-political processes, infrastructure can have an independent and lasting influence on behavior.

Under the demand-driven perspective on infrastructure, human behavior is guided by internal states: the needs of the individual organism. Under the political perspective, behavior is a socially negotiated outcome. It is indisputable that humans have biological needs, but the way that these internal states are expressed in behavior is fundamentally mediated by social context. In fact, behavior is conceived as the performance of “social practices” (Schatzki 1996; Warde 2005; Shove, Pantzar, and Watson 2012; Bourdieu 1977). Understandings of why one performs a given practice and what the practice symbolizes, the cognitive and embodied knowledge of how to perform it, and the material artifacts and contexts with which one performs it are all socially learned and negotiated. Thus, behavior is political. Individuals have limited time and resources with which to perform social practices, and groups with investments in competing ideas or beliefs, skills, and technologies have interests tied to the performance of practices. “Recruitment” to and “defection” from social practices shape social life and allocation of resources (Shove, Pantzar, and Watson 2012; Watson 2013).

Like social practice theorists, behavioral economists dispute the notion that behavior is determined straightforwardly by needs and desires. Whereas social practice theorists might point to the importance of cultural “meaning-making” processes in linking behaviors to needs, behavioral economists emphasize “information constraints” and “logical fallacies” that influence the perceived costs and benefits of alternatives (Kahneman 2011; Thaler and Sunstein 2008). Though the terminology is different, both approaches imply that behavior is malleable. In their influential book Nudge (2008), Thaler and Sunstein argue that policymakers and other “choice architects” can influence behavior by making some options more visible and highlighting hidden costs and benefits of behaviors. Choice environments are never “neutral,” so those who design them must do so thoughtfully.

A logical consequence of this political understanding of behavior is a political understanding of technological development. Often, competing technologies embody different understandings of the meaning or purpose of a social practice. When there is disagreement over the purpose of a practice, technical efficiency cannot be the guiding force, because the question of “efficiency at achieving what end?” is unresolved.

The social construction of technology school of research, in particular the work of Bijker (1997) and Pinch (Pinch and Trocco 2009), rejects the notion that technology progresses according to natural laws of efficiency. These scholars argue that the success of a technology cannot be explained by its technical superiority. Rather, success results from political struggles between groups with competing visions of the meaning, purpose, and specifications of the artifact.

A political understanding of behavior and technological development implies a political understanding of infrastructure. From this perspective, infrastructure is not merely a reflection of the most efficient physical arrangement to satisfy objective social needs, or even the desires of a democratic majority. Rather, the shape of the built environment emerges from political struggles involving citizens, organizations, professional experts, and state bureaucrats. Moreover, as with technology, the settlement of these struggles and the emergence of a victorious design reverberates into the social world, as social relations and practices are rearranged.

Sociological research on transportation and infrastructure thus focuses on how infrastructure shapes social life, and on the political processes that shape infrastructure itself. Scholars studying infrastructural effects have demonstrated the influence of transportation infrastructure such as roads and airports on various aspects of social life. Highway expansion shapes migration patterns, especially in suburban areas, and road-building in the Amazon shapes development, with substantial environmental implications (Chi 2010; Perz et al. 2008). Airports have become influential in shaping economic dynamics, including regional talent share (Chen, Chi, and Chi forthcoming) and inter-city economic ties (Neal 2011), with some observers predicting a drastic reorientation of cities around airports (Appold and Kasarda 2013; Kasarda and Lindsay 2011).

Accounts of the political processes that shape infrastructure emphasize coalition-building and power. Despite low rates of automobile ownership in the United States before 1920, proponents of highway construction pushed through road projects in the first two decades of the twentieth century by building a coalition of automobile enthusiasts, Progressives concerned with urban overcrowding, professionalizing highway engineers, and farmers (Ling 1992). The success of these early road-building projects quite literally laid the groundwork for the subsequent explosion of auto-ownership and driving.

In the case of urban infrastructure, the political approach focuses on the question of who holds political power. Often throughout US urban history, the answer to this question has been large property owners and their allies with an interest in intensifying the use and profitability of their holdings (Feagin 1985; Gottdiener and Feagin 1988; Logan and Molotch 2007; Molotch 1976). These landowners pushed an agenda of growth—in population, size, and economic activity—by arguing that growth creates jobs and prosperity, despite subsequent evidence that rapid growth actually diminishes environmental quality and livability and has little effect on employment security (Molotch 1976).

Attention to powerful elites yields an alternative narrative of the automobile's rise to dominance. Historical evidence shows that actors in the highly concentrated automobile, oil, and rubber industries engaged in campaigns to undermine mass transit from the 1920s onward. These campaigns were facilitated by the lack of regional and national integration in mass transportation and limited city oversight (Whitt 2014; Whitt and Yago 1985; Yago 1984).

Scholarship on infrastructure from the political perspective suggests that the influence of coalitions favoring growth and urban sprawl has declined. By the 1970s, “anti-growth” political coalitions were gaining strength, and had begun enacting urban growth boundaries to check automobile-enabled sprawl (Molotch 1976). These coalitions were especially successful in places with “high amenity value” and strong environmental groups, such as Oregon, where Governor Tom McCall railed against “the ravenous rampage of suburbia in the Willamette Valley [that] threatens to mock Oregon's status as the environmental model for the nation” (Trimet 2015, 17). Oregon enacted urban growth boundaries for Portland in 1974, and the city rejected a freeway expansion in that same year.

This shifting perspective on growth and urban development also gained traction among “place professionals” (Gieryn 2000), who gradually pivoted from building roads to accommodate car travel to strategies of “demand reduction.” This new approach included public transportation improvements, and after the passage of the federal Intermodal Surface Transportation Efficiency Act (ISTEA) in 1991, improvements to walking and bicycling facilities.

ISTEA illustrates how infrastructural projects can be loosely coupled with local social practice, and thus potentially influence practice. By earmarking federal funds for bicycle and pedestrian facilities, ISTEA created leeway for localities to build such infrastructure even in the absence of local demand. The infrastructure built with federal funds may lend legitimacy to bicycling and reshape social practice locally (Law 1992; Winner 1993), but only if it is integrated with local elements, through the performance of practices, into “configurations that work” (Shove and Pantzar 2005; Shove, Pantzar, and Watson 2012). Characteristics of place influence whether and how imported infrastructures or technologies reshape social practice (Barley 1986; Gieryn 2000; Molotch, Freudenburg, and Paulsen 2000; Orlikowski 1992). Environmentalism helped shift the orientation of American transportation and land-use planning, but local levels of environmentalism vary. If environmental movement organizations cultivate an “ecological habitus” (Haluza-DeLay 2008), their local prevalence may moderate the influence of new infrastructure on practices seen as environmentally relevant, such as bicycling.

Hypotheses

The demand-driven and political perspectives on infrastructure suggest divergent hypotheses regarding the relationship between bicycling behavior and bicycle infrastructure. The demand-driven perspective suggests that infrastructure will reflect the preferred practices of the public, and thus will follow trends in behavior more than it determines them.

H1a: Increases in bicycle-commuting in a city will induce the creation of additional bicycle infrastructure.

The political perspective on infrastructure, in contrast, does not suppose that infrastructure will be built that reflects trends in behavior. Rather, this perspective views infrastructural decisions as results of contingent political processes, in which opinion and behavioral trends are but one factor. Thus, the influence of social practice on infrastructure will depend on political context. Given the role of environmentalism in shifting transportation planning toward “demand reduction,” and environmental movement organizations’ potential to cultivate an ecological habitus (Haluza-DeLay 2008), the prevalence of environmental organizations in a city may be an important contextual factor.

H1b: The effect of bicycle-commuting levels on subsequent creation of bicycle infrastructure will depend on the prevalence of environmental organizations in a city.

Whereas the demand-driven perspective views behavior as determined by fixed drives, the political perspective views preferences and behavior as malleable. Behavior is subject to political processes, wherein “practitioners” are “recruited” to new practices and “defect” from others. Changing infrastructure can influence these processes, making some practices more attractive and legitimate and others less so. Thus, the political perspective expects that infrastructural decisions will shape behavior as much as the reverse.

However, just as the political perspective views infrastructural decisions as contingent, so too does it view the effects of infrastructural changes as contingent on contextual factors. Infrastructure is an influence on social practice, but it is not the only influence. Whereas the demand-driven perspective views practice as a product of human needs and the technologies available to satisfy those needs, the political understanding implies that “demand” for practices is shaped by socio-cultural processes. Infrastructure can increase the visibility and legitimacy of a practice, but the extent of its influence may depend on whether proponents can connect the practice to locally resonant cultural frames. Again, environmental advocacy may play a role in this process.

H2: The effect of additional bicycle infrastructure on bicycle-commuting will depend on the prevalence of environmental organizations in a city.

Data and Methods

Empirical Research on the Built Environment and Travel

The empirical research on the built environment and travel, while extensive, is limited by two key methodological difficulties. The first is the possibility of spurious correlation driven by unmeasured factors. For instance, some cities may have a strong historical tradition of bicycling, and this tradition might explain both its bicycle facilities and its high bicycling rate. Such place characteristics are hard to quantify, but they can be partialed out using longitudinal data and fixed effects. Unfortunately, few studies of the built environment and travel use longitudinal data.

A second methodological difficulty is uncertainty over the direction of causality. Cross-sectional studies have found a positive correlation between the volume of bikeways and cycling rates (Buehler and Pucher 2012; Dill and Carr 2003; LeClerc 2002; Nelson and Allen 1997; Parkin, Wardman, and Page 2008). However, the direction of causality cannot be ascertained with cross-sectional data, and few studies make use of longitudinal data.

One of the few studies using longitudinal data found increased bicycle-commuting between 1990 and 2000 among residents living near newly constructed bikeways in Minneapolis–St. Paul, MN (Krizek, Barnes, and Thompson 2009). However, a second study that applied the same methodology to six additional US cities concluded that the effect of new bicycle infrastructure depended on publicity for the new facilities, their placement along common utilitarian travel routes, and the overall connectivity of the bikeway network (Cleaveland and Douma 2009). These mixed findings suggest that the effects of infrastructure depend on contextual factors.

Present Analysis

Data sources and variables

Table 1 describes the variables included in the analysis, and provides descriptive statistics. Variables are measured at the level of the city rather than metropolitan area whenever possible—exceptions are noted in the descriptions of variables below.

Table 1.

Description of Variables and Sources of Data, with Descriptive Statistics

Description	Source	Min	Q1	Median	Q3	Max
Pct commuters traveling by bicycle	US Census Bureau	0.00	0.31	0.63	1.21	7.79
Bicycle lanes & paths, miles/square mile	Dill and Carr 2003, Alliance for Bicycling and Walking Benchmarking Reports	0.00	0.33	0.65	1.25	5.76
Pop. density, residents/square mile	US Census Bureau	799.63	2447.38	3813.93	6425.72	28234.99
Public transit supply, regional annual vehicle miles per 1,000 population	National Transit Database (USDOT)	2.38	9.68	13.75	19.44	51.83
Gas price, dollars, state avg.	US Department of Energy	0.73	1.63	2.12	2.48	3.28
Cyclist fatalities per 10K bicycle commuters, statewide	NHTSA annual data	0.00	6.99	11.58	19.51	86.22
Average daily precipitation, inches	National Climatic Data Center	0.00	0.05	0.10	0.13	0.32
Pct days under 32°F	National Climatic Data Center	0.00	0.00	0.29	5.04	25.21
Pct days over 90°F	National Climatic Data Center	0.00	2.20	7.67	20.22	45.14
Pct adults w/bachelor's degree or more	US Census Bureau	9.12	23.41	28.58	34.97	59.80
Pct of pop. that is college students	US Census Bureau	3.82	6.68	7.95	9.62	18.13
Pct of pop. aged 25–34	US Census Bureau	9.30	14.80	16.03	17.67	24.80
Advocacy strength	Alliance for Bicycling and Walking Benchmarking Reports	−0.49	−0.44	−0.32	−0.01	12.05
City bicycle-ped staff per 100K pop.	Alliance for Bicycling and Walking Benchmarking Reports	0.00	0.16	0.39	0.85	5.58
Pct voting for Democratic presidential candidate	County-level voting data from CQ Voting and Elections Collection	17.09	46.32	55.78	65.73	92.95
Environmental nonprofits per 100K pop., 2000	National Center for Charitable Statistics	0.00	1.11	2.20	4.32	22.20

Bicycle-commuting and bikeways variables

The two main variables of interest in my analyses are the percent of commuters traveling by bicycle and the density of bikeways. Data on bicycle-commuting come from the Decennial Census and the American Community Survey, both administered by the US Census Bureau. Respondents were asked, “How did you usually get to work last week?” and were instructed to pick one mode as primary if they used more than one. I excluded individuals who reported working at home from the total in calculating the percent commuting by bicycle.

For this analysis, the term “bikeways” refers to two types of bicycle infrastructure: bicycle lanes and bicycle paths. Bicycle lanes are on-street facilities that demarcate road space for exclusive use of bicyclists with painted lane lines. Bicycle lanes may or may not be painted in a bright color to increase their visibility, and they may or may not be separated from car traffic by a buffer, vertical barriers, or a parking lane. Bicycle lanes are distinct from “bicycle routes,” which are designated only by street signs with no pavement markings, and from “sharrows,” which are designated by signs and pavement markings reminding motorists to share the road with cyclists, but which do not demarcate a lane for exclusive use of cyclists. Bicycle paths are off-street facilities designed for mixed use by cyclists and pedestrians. Bicycle paths offer more separation from car traffic than bicycle lanes, but are often built for recreation rather than transportation, and may be less accessible and offer less direct routes than on-street bicycle lanes.

Past research has found a positive association between bicycling rates and both bicycle lanes and paths, with similar effect sizes (Buehler and Pucher 2012; Krizek, Barnes, and Thompson 2009). For this analysis, I combine bicycle lanes and paths into a single “bikeways” variable because I am interested in the relationship between bicycle-dedicated infrastructure and bicycling in general, rather than in the relative influence of different types of bicycle infrastructure. In an auxiliary analysis, I found that bicycle lanes and paths have comparably sized effects on bicycle-commuting.

Data on extent of bikeways come from Dill and Carr (2003) for the year 2000, and from five Alliance for Bicycling and Walking (ABW) Benchmarking Reports (Alliance for Biking and Walking 2010, 2012, 2014, 2016; Thunderhead Alliance 2007). Together, these sources provide at least one measurement of bikeways for sixty-two large US cities, including all of the fifty most populous cities in the 2010 Census. Each source gathered data directly from city officials on the centerline miles of bicycle lanes and bicycle paths in their city. These sources provide measurements of the extent of bikeways in each city at the end of the years 2000, 2006, 2008, 2010, 2012, and 2014. To measure the density of bikeways, I divided the miles of bikeways by city land area in square miles. I obtained data on land area from the 2000 and 2010 Censuses, linearly interpolated values for the intervening years, and used the 2010 value for years 2011–2014.

There were some missing data for bikeways, mostly due to shifts in the set of cities present in Dill and Carr's study and each year of the ABW reports. The final dataset contained 286 city-year observations of bicycle lanes and 285 city-year observations of bicycle paths, out of 372 possible city-years (sixty-two cities at six time points). However, I deleted five city-year observations for bicycle lanes, and one for bicycle paths, because of the implausible year-to-year fluctuations they implied (e.g., a ninety-five-mile reduction in bicycle lanes over two years). This resulted in 281 valid observations for bicycle lanes and 284 for bicycle paths, which amounts to 24.4 percent missingness on lanes and 23.7 percent on paths.

Control variables

Other independent variables consisted of standard controls in studies of bicycle mode share (e.g., Buehler and Pucher 2012), as well as a few more novel control variables. Bicycling levels tend to be higher in densely populated areas with mixed land uses and grid-pattern streets (Ewing and Cervero 2010; Litman 2008; Parkin, Wardman, and Page 2008; Pucher and Buehler 2006; Zahran et al. 2008). To compute population density, I used data on land area described above, and interpolated population values for 2001–2004 based on figures from the 2000 Census and 2005 ACS. I allowed one exception to these rules for Louisville, KY, which merged with surrounding Jefferson County in 2003, quintupling its land area and doubling its population. For Louisville, I assumed constant population and land area from 2000 to 2002 (before the city-county merger), constant population from 2003 to 2005, and constant land area from 2003 to 2014.

The supply of public transportation may influence bicycling positively or negatively. Public transport may complement bicycling by allowing for combined bicycle-PT trips and offering a non-car backup mode in case of inclement weather (Brons, Givoni, and Rietveld 2009; Martens 2007; Pucher and Buehler 2009). On the other hand, public transport may compete with bicycling (Fietsberaad 2010; Heinen, van Wee, and Maat 2010; Pucher and Buehler 2007). To measure supply of public transport, I used data from the National Transit Database on the annual vehicle miles per 1,000 population at the metropolitan level, because the service areas of most transit agencies extend beyond central city boundaries.

Higher gas prices reduce driving, increase traffic safety, and influence residential relocation decisions (Buehler 2010; Chi and Boydstun forthcoming; Chi et al. 2010; Hanly, Dargay, and Goodwin 2002; Litman 2008), all of which could influence bicycle-commuting. Data on gas prices were not available at the city level, so I used state-level data on annual average gas prices from the US Department of Energy. Similarly, city-level data on bicyclist fatalities were unavailable. Instead, I used state-level data on the annual number of cyclist fatalities from the National Highway Transportation Safety Administration, standardized by the number of bicycle commuters in the state as measured by the Census Bureau. Perceived risk of injury and death influences cycling rates (Alliance for Biking and Walking 2010; Fietsberaad 2010; Jacobsen, Racioppi, and Rutter 2009; Pucher and Buehler 2008; USDOT 2010). However, according to the “safety in numbers” phenomenon, increased bicycling increases safety by raising drivers’ awareness of bicyclists (Elvik 2009; Jacobsen 2003; Robinson 2005). To ensure that any observed relationship runs from bicycle safety to bicycle rates, not vice versa, I used a lagged measure of cyclist fatality, the average for the previous two years, as a predictor of bicycle-commuting rates.

Car ownership is negatively related to bicycling (Heinen, van Wee, and Maat 2010; Pucher and Buehler 2006), so I include the percent of households not owning a car from Census and ACS data as a control variable. A number of studies have found that rain and cold or hot weather can deter bicycling (Baltes 1996; Dill and Carr 2003; Gatersleben and Appleton 2007; Heinen, van Wee, and Maat 2010; Winters et al. 2007). To control for the effect of weather, I included as controls the average daily precipitation, the percent of days with a high temperature under 32°F, and the percent with a high temperature over 90°F. These data come from the National Climatic Data Center.

I included several demographic control variables from the Census and ACS. College students have higher rates of cycling (Baltes 1996; Dill and Carr 2003; Ryley 2006), so I included percentage of college students in the population. I also included the percentage of adults aged 25 and over with a bachelor's degree or greater education and the percentage of a city's population aged 25–34 to capture in- and out-migration of young professionals. The work of Richard Florida on the “Creative Class” suggests that young professionals seek out cities with lifestyle amenities such as walkable and bikeable neighborhoods (Florida 2002, 2010).

Strong bicycle advocacy groups may boost a city's bikeway building efforts, while at the same time increasing recruitment to bicycling. This could create a spurious correlation between bikeways and bicycle-commuting. To control for this possibility, I included an index of advocacy strength that combines three measures of a city's advocacy organizations belonging to the ABW: number of members, number of staff, and organizational income. I combined these measures by standardizing each component, summing them, and standardizing the result, such that a one-unit change in advocacy strength reflects a change of one standard deviation. The number of city staff devoted to bicycle and pedestrian issues may also be associated with both bikeways and bicycling, and data on such staff are provided in the ABW reports, so I included this variable as well.

Finally, research on the history of US transportation policy suggests an association between environmentalism and the transportation policies of cities. The percent of voters in the surrounding county voting for the Democratic presidential candidate, interpolated between election years, and the number of environmental nonprofits in the year 2000, a time-constant predictor, are included as indicators of environmentalism. Political party identification is one of the most reliable predictors of concern for climate change (Hamilton et al. 2014). The measure of partisan voting serves as a control for within-city changes in environmentalism during the study period, whereas the prevalence of environmental organizations in 2000 captures between-city variation in environmentalism to be included as an interaction term to test hypotheses 1b and 2.

Figure 1 shows the trend in bicycle commute mode share and extent of bikeways for each city. For the imputation and analysis, I performed a logit transformation on variables with percentage units, and a natural-log transformation on variables with positively skewed distributions, to better satisfy assumptions of normality. I note transformations in the results tables.

Figure 1.

Plot of bicycle commute mode share and bikeways by city, 2000–2014

Imputation of missing data

Table 2 shows the year-by-year availability of each variable. The inconsistent availability of variables across the years of my study period (2000–2014), along with missing data for the three variables derived from ABW reports (bikeways, advocacy strength, and city bike-ped staff), posed a challenge for the analysis. Traditional approaches to missing data simply delete cases with missing values, or fill in those missing values with the sample mean, interpolated or regression-fitted values, or educated guesses. However, case-wise deletion leads to biased coefficient estimates, and single imputation of values biases estimates of uncertainty (Rubin 1987; Rubin and Little 2002). In recent decades, multiple imputation has emerged as the preferred method for dealing with missing data. In multiple imputation, the analyst uses all available data to construct a distribution for missing values conditional on the observed data, then takes multiple draws from those distributions to create several complete datasets. He or she then performs the analysis on each complete dataset, and combines the results. This process preserves the uncertainty caused by the missing values without discarding any available information.

Table 2.

Variable Availability by Year

	2000	2001	2002	2003	2004	2005	2006	2007	2008	2009	2010	2011	2012	2013	2014
Public transit supply	X	X	X	X	X	X	X	X	X	X	X	X	X	X	X
Gas price	X	X	X	X	X	X	X	X	X	X	X	X	X	X	X
Bicyclist fatality rate	X	X	X	X	X	X	X	X	X	X	X	X	X	X	X
Avg. precipitation	X	X	X	X	X	X	X	X	X	X	X	X	X	X	X
Pct cold days	X	X	X	X	X	X	X	X	X	X	X	X	X	X	X
Pct hot days	X	X	X	X	X	X	X	X	X	X	X	X	X	X	X
Bicycle rate	X					X	X	X	X	X	X	X	X	X	X
Population density	X					X	X	X	X	X	X	X	X	X	X
Pct adults w/bachelor's degree	X					X	X	X	X	X	X	X	X	X	X
Pct college students	X					X	X	X	X	X	X	X	X	X	X
Pct 25–34-year-olds	X					X	X	X	X	X	X	X	X	X	X
Lanes & paths	X						X		X		X		X		X
Advocacy strength							X		X		X		X
City bicycle-ped staff							X		X		X		X		X
Pct voting Dem. for pres.	X				X				X				X
Environmental nonprofits, 2000	X

In order to leverage the longitudinal structure of my dataset, I used the Amelia II package for R (Honaker, King, and Blackwell 2011). This program is designed for multiple imputation of panel datasets in which variables are thought to follow relatively smooth trends within cross-section units over time. The Amelia II package also allows the specification of priors for missing values based on substantive knowledge, so I used linear interpolation within cities over time to fill in priors. I then created five imputed datasets, ran my analyses with each dataset, and combined the results using the rules formulated by Rubin (1987). Additional details on my imputation of missing data can be found in the supplementary materials accompanying the online version of this article.

Methods of analysis

I used two methods of analysis. First, I computed a fixed-effects regression model. Fixed-effects regression only models change within cross-sectional units over time, allowing the analyst to control for the influence of time-invariant characteristics. However, fixed-effects regression cannot typically model a reciprocal causal relationship between two variables, such as that which may exist between rates of bicycle-commuting and bicycle infrastructure. To allow for a possible reciprocal relationship, I used structural equation modeling software to estimate a model with both fixed effects and a reciprocal relationship, as described by Allison (2014) and implemented by England, Allison, and Wu (2007).

Results

Fixed-Effects Model

Table 3 shows the results of the fixed-effects regression. I added predictors in three blocks. The first block consists of variables directly affecting the costs and benefits of bicycling in terms of time, money, and comfort. The second block is composed of demographic variables. The third block adds political and ideological variables and the interaction between environmental organizations and bikeway infrastructure. The main effect of environmental organizations is not included because this variable is time invariant, and fixed-effects regression only models change over time within cross-sectional units.

Table 3.

Fixed-Effects Regression of Logit Bicycle-Commute Mode Share, 2000–2014

	Cost-benefit	Demographics	Political	Reduced
Bikeways, miles/sq. milea	0.463 (5.379)**	0.244 (2.802)**	0.058 (0.552)	0.06 (0.58)
Pop. densitya	0.174 (0.993)	−0.039 (−0.222)	−0.102 (−0.545)
Public transit supply	−0.047 (−0.52)	−0.011 (−0.126)	0.003 (0.032)
Gas pricea	0.117 (4.166)**	0.113 (3.549)**	0.09 (2.353)	0.109 (3.392)**
Cyclist fatality rateb	−0.178 (−3.184)**	−0.078 (−1.442)	−0.076 (−1.396)*	−0.077 (−1.429)
Pct HHs w/o a cara	0.029 (0.256)	0.137 (1.278)	0.11 (1.061)
Avg daily precipitationb	−0.14 (−0.256)	−0.387 (−0.733)	−0.37 (−0.687)
Pct days under 32°Fb	0.051 (2.401)*	0.041 (1.987)*	0.039 (1.956)+	0.039 (1.932)+
Pct days over 90°Fb	0.01 (0.707)	0.007 (0.536)	0.009 (0.633)
Pct adults w/bachelor's degreeb		0.481 (3.502)**	0.381 (2.702)**	0.396 (2.886)**
Pct college students in popb		0.278 (2.527)*	0.25 (2.091)*	0.295 (2.743)**
Pct pop. aged 25–34b		0.718 (5.32)**	0.652 (4.628)**	0.654 (4.772)**
Pct voting for Democratic pres. candidatea			0.118 (1.087)
Bicycle-ped advocacy strengtha			0.058 (0.757)
City bicycle-ped staffa			0.083 (0.844)
Bikewaysa * environmental orgsa			0.135 (1.824)+	0.167 (2.239)*
Adjusted R2	0.272	0.331	0.34	0.334
F-Statistic	39.899**	40.158**	31.496**	60.916**

Note:t-scores shown in parentheses.

Signif: **p < .01 *p < .05 ⁺p < .1.

^aLog-transformed variable.

^bLogit-transformed variable.

To aid interpretation and comparison of effect sizes, I computed the expected change in the number of bicycle commuters caused by a shift of one interquartile range (IQR) in the predictor. For this computation, I held other predictors at their mean values, and I calculated effect sizes for a city with an average bicycle-commuting rate of 0.66 percent (the sample average) across the study period. I converted the bicycle-commuting rate to number of bicycle commuters by scaling to the sample average number of total commuters, which is 356,000. A 0.66 percent rate of bicycle-commuting with 356,000 total commuters amounts to 2,350 bicycle commuters. Effect sizes reported below can be interpreted as swings centered around this value.

Among cost-benefit variables, only bikeways, gas price, cyclist fatality, and percentage of days under 32°F have a statistically significant relationship to bicycle-commuting. In each case except cold weather, the relationship is in the expected direction: bikeways and gas prices are positively associated with bicycle-commuting, and the cyclist fatality rate is negatively associated with bicycle-commuting. Counter to expectations, bicycle-commuting is positively related to the number days having a high temperature less than 32°F. An increase of one IQR in bikeways is associated with 321 additional bicycle commuters, given the parameters described in the previous paragraph. For gas prices, a comparable increase is associated with 262 additional bicycle commuters. A one-IQR decrease in cyclist fatality predicts 213 additional bicycle commuters, and a one-IQR increase in days under 32°F predicts 43 additional bicycle commuters. The magnitude of these effect sizes points to the importance of using longitudinal data—previous cross-sectional studies estimated that adding one mile of bikeways per square mile was associated with an increase of a full percentage point in the bicycle mode share, which would amount to over 3,500 additional bicyclists in a city with 356,000 commuters (Dill and Carr 2003). Consistent with a recent cross-sectional study (Buehler and Pucher 2012), the supply of public transit and weather variables (excepting cold weather) were not significantly related to rates of bicycle-commuting. In contrast with past cross-sectional analyses, changes in residential population density were not significantly associated with bicycle-commuting, suggesting that the cross-sectional relationship may be caused by unmeasured characteristics of cities or a past causal relationship that no longer holds (England, Allison, and Wu 2007).

All three demographic variables in the second block were significantly associated with bicycle commute mode share, controlling for cost-benefit factors. A one IQR increase in the percent of adults with a bachelor's degree or greater is associated with 183 additional bicycle commuters. Equivalent increases in the percent of college students in the population and in the percent of the population aged 25–34 are associated with 136 and 222 additional bicycle commuters, respectively. The magnitude of the coefficients for bikeways and cyclist fatality rate decline noticeably with the inclusion of these demographic variables. This is consistent with the hypothesis that prospective college students and young, educated adults who enjoy bicycling are attracted to bicycle-friendly cities (Florida 2002, 2010). The magnitude of the coefficient for gas prices is not diminished when we control for these demographic characteristics.

Among political and ideological variables in the third block, support for the Democratic presidential candidate was positively associated with bicycle-commuting, but the coefficient was not statistically significant. As hypothesized, the interaction term between number of environmental organizations per 100,000 population in the year 2000 and growth of bikeways is positive, but is only marginally statistically significant. This suggests that in cities with a greater density of environmental nonprofits, installing bikeways has a greater effect on bicycle-commuting. This effect could occur through one or both of two mechanisms. On the one hand, environmental organizations may actively publicize new bikeways and promote their use. On the other hand, the density of environmental organizations may simply indicate high levels of environmental concern in a city, and people with greater environmental concern likely have a greater affinity for bicycling, given its low carbon footprint.

Surprisingly, changes in the strength of bicycle and pedestrian advocacy and in the number of city staff working on bicycle and pedestrian issues during this period are not significantly associated with changes in bicycle-commuting rates. There are several possible explanations for this lack of association. First, these factors might influence bicycling not in the short term but in the long term. We wouldn't expect an increase in advocacy organizations or city staff to immediately boost bicycling levels. Second, the influence of growing advocacy and municipal support for bicycling may occur mostly through the growth of the bikeway network, such that controlling for changes in bikeways masks the underlying effect of these factors. Additionally, there were substantial missing data for these variables. The high proportion of multiply imputed values for these variables may have dampened their association with bicycling by adding random noise.

The fourth column of table 3 shows a reduced model including only variables that are significantly associated with bicycle-commuting in at least one of the first three models. These coefficients are of similar magnitude to those in the saturated model, though the coefficient for the interaction term has a larger magnitude and is significant at a 0.05 alpha level.

Figure 2 illustrates the interaction between bikeways and environmental organizations, based on the reduced model. The y-axis shows number of bicycle commuters, scaled to the average number of commuters (356,000) of cities in the sample. The x-axis shows miles of bicycle lanes and paths, scaled to the average land area of cities in the sample (199 square miles). Thus, the figure shows how the effect of bikeways for an average city would differ if that city had a density of environmental organizations at the first or third quartile, fixing all other variables at the sample mean. The within-city IQR of bikeways was 0.46 miles per square mile, or 90 miles for a city of average size. For a city at the first quartile of environmental organizations, this change in bikeways results in 128 additional bicycle commuters. For a city at the third quartile of environmental organizations, the increase is 233 bicycle commuters. For a city with 3.5 million commuters (the number in New York City), the increase in bicycle commuters would be 1,261 versus 2,292 with environmental organizations at the first and third quartiles, respectively. In other words, the higher density of environmental organizations increases the gains from added bikeways by 82 percent.

Figure 2.

Difference in effect of bikeways depending on density of environmental organizations; increase in bicycle commuters when city shifts from first to third (within-city) quartile of bikeway extent, scaled for a city with average number of commuters (356,000) and land area (199 square miles)

Maximum-Likelihood Structural Equation Model

Fixed-effects models reduce the chance of omitted-variable bias by controlling for unmeasured time-invariant characteristics. However, fixed-effects models as typically estimated do not allow the analyst to control for lagged values of the dependent variable, nor to account for possible reciprocal effects between the dependent variable and future values of an independent variable. However, Allison (2014) proposes an estimation technique using maximum likelihood and structural equation modeling (hereafter, ML-SEM) that controls for unobserved, time-invariant characteristics (à la fixed effects), lagged values of the dependent variable, and reciprocal effects. The inclusion of lagged values of the dependent variable and modeling of reciprocal effects allows for a test of causal relationships that is arguably more conservative than a typical fixed-effects model, such as the one computed in the previous section. Moreover, in our case, it allows for a preliminary test of the causal effect of bicycle-commuting on bikeway construction, given the likely presence of reciprocal effects and the lack of past research on the factors influencing bikeway construction.

One trade-off of the ML-SEM approach is that it requires us to convert the data into wide form, with a separate variable for each year of each quantity measured. As the time series lengthens relative to the number of cross-sectional units, the number of variables in the model quickly approaches the number of observations, and convergence becomes less likely. For this reason, the results shown here include few control variables. This is a clear limitation, and prevents us from drawing strong conclusions from these models. However, given the advantages of these models described above, they provide additional evidence in evaluating the effect of bikeways on bicycle-commuting, and preliminary evidence on the effect of bicycle-commuting on bikeway construction.

Table 4 shows the results of ML-SEM predicting bicycle-commuting rate, with varying time lags for the predictors and dependent variable. In addition to the density of bikeways, predictors include average annual gas price (at the state level), number of environmental organizations per 100,000 residents in 2000 (time invariant), the interaction between bikeways and environmental organizations, and lagged values of the dependent variable. Because bikeways and gas prices could plausibly have an immediate effect on bicycle-commuting, I tested lag lengths ranging from zero to three years, while the dependent variable was lagged from one to four years in the corresponding models.

Table 4.

Maximum Likelihood Structural Equation Models of Logit Bicycle Rate with Varying Time Lags

	No lag	t − 1	t − 2	t − 3
Bikewaysa	0.477 (2.364)**	0.222 (1.818)+	0.072 (0.698)	−0.063 (−0.634)
Gas price	0.044 (0.959)	0.155 (3.34)**	0.202 (6.284)**	0.256 (7.946)**
Environmental orgs per 100,000 residents, 2000a	0.049 (0.167)	0.132 (0.437)	0.265 (0.229)	0.241 (0.859)
Bikewaysa * env. orgsa	0.318 (2.819)**	0.267 (3.717)**	0.256 (5.98)**	0.219 (3.406)**
Lagged bicycle rateb,c	0.173 (1.606)	0.196 (2.947)**	0.157 (3.217)**	0.15 (5.441)**

Note:z-scores shown in parentheses.

**p < .01 *p < .05 ⁺p < .1.

^aLog-transformed variable.

^bLogit-transformed variable.

^cBicycle rate is lagged one more year than other predictors, that is, 1–4 years instead of 0–3.

The results in table 4 provide additional support for the hypothesis that the effect of bikeways is moderated by the extent of environmental organizations in a city, with the interaction term significant and positive in all four models. These results also provide further evidence for the positive influence of gas prices on bicycle-commuting, and suggest that a change in gas prices takes at least a year to affect commuting behavior.

Table 5 shows the results of ML-SEM predicting bikeway construction by the lagged rate of bicycle-commuting. Changes in the rate of bicycle-commuting wouldn't affect bikeway construction immediately. A change in bicycle-commuting levels must be measured and recognized before it can affect bikeway construction, and new bikeways must undergo a planning process that takes months if not years. Similarly, any effect of change in the number of city staff working on bicycle and pedestrian issues is unlikely to occur in less than two years. Thus, I ranged the lag length of these predictors from two to five years, and used the same lag lengths for lagged values of the dependent variable.

Table 5.

Maximum Likelihood Structural Equation Models of Bikeways, Log-Miles per Square Mile, with Varying Time Lags

	t − 2	t − 3	t − 4	t − 5
Bicycle rateb	0.052 (0.657)	−0.012 (−0.248)	−0.087 (−1.23)	−0.071 (−0.828)
City bicycle-ped staff per 100,000 residentsa	0.063 (1.347)	0.11 (2.513)*	0.104 (2.292)*	0.076 (1.193)
Environmental orgs per 100,000 residents, 2000a	−0.151 (−0.326)	−0.061 (−0.224)	0.121 (0.521)	−0.108 (−0.501)
Bicycle rateb * env. orgsa	0.096 (1.684)+	0.095 (3.44)**	0.141 (3.589)**	0.112 (0.032)*
Lagged bikewaysa	0.564 (29.366)**	0.507 (5.466)**	0.473 (28.052)**	0.451 (7.033)**

Note:z-scores shown in parentheses.

**p < .01 *p < .05 ⁺p < .1.

^aLog-transformed variable.

^bLogit-transformed variable.

The results in table 5 suggest that bikeway construction is more responsive to bicycle-commuting in cities with more environmental organizations. An interaction term between bicycle-commuting rate and environmental organizations is positive and significant in three out of four models, and marginally significant in the other model. The effects of changes in bicycle-commuting rates and in city staff working on bicycle issues manifest most strongly after three or four years.

Figure 3 illustrates the interaction effect between bicycle-commuting rate and environmental organizations, based on the four-year lag model, in which the magnitude of the interaction term is greatest. The y-axis shows the miles of bikeways, scaled to the average square mileage (199) of cities in the sample. The x-axis shows the number of bicycle commuters, scaled to the average total number of commuters, 356,000. The within-city IQR of bicycle-commuting rate was 0.38 percent, which for a city with 356,000 commuters, and average bicycle rate at the sample mean (0.66 percent), constitutes an increase of 1,340 bicycle commuters (from 1,764 to 3,104). For a city at the first quartile of environmental organizations, this change in bicycle-commuting is associated with an increase of only three miles in bikeways. For a city at the third quartile of environmental organizations, in contrast, the same increase in bicycle-commuting is associated with an increase of twenty-five miles in bikeways. In other terms, a city with a higher density of environmental organizations is expected to build over seven times the extent of new bikeways as a city with fewer such organizations, in response to the same increase in bicycle-commuting.

Figure 3.

The effect of bike-commuting on subsequent bikeway construction is moderated by local environmentalism; increase in miles of bikeways when city shifts from first to third (within-city) quartile of bike commuters, scaled for a city with average number of commuters (356,000) and land area (199 square miles)

Discussion and Conclusion

The preceding analysis tested three hypotheses relating infrastructure and behavior. Hypothesis 1a, derived from the demand-driven perspective on infrastructure, posited that increases in bicycle-commuting would induce cities to build more bicycle infrastructure. Hypothesis 1b, derived from the political perspective on infrastructure, added the caveat that the effect of bicycle-commuting on bikeway construction would depend on the density of environmental organizations. The results of our ML-SEM analysis support hypothesis 1b. The effect of increased bicycle-commuting on bikeway construction is negligible in cities with few environmental organizations, but is substantial in cities with many such organizations.

Hypothesis 2 also sprang from the political perspective on infrastructure, predicting that the creation of bicycle infrastructure would induce increases in bicycle-commuting, but that this effect too would depend on the density of environmental organizations. Using fixed-effects regression and ML-SEM, I found a significant and positive interaction effect between bikeways and environmental organizations, providing support for hypothesis 2.

On the one hand, these results add to existing evidence, mostly cross-sectional, that bikeway infrastructure can induce bicycling, even when controlling for unmeasured, time-invariant characteristics and a host of material and demographic factors known to influence bicycling (in the fixed-effects model), and when controlling for lagged values of the dependent variable and possible reciprocal effects between bikeways and bicycle-commuting (in ML-SEM). On the other hand, the significant interaction effect with environmental organizations demonstrates that the effect of building bikeway infrastructure depends on the local political-cultural context.

In cities with relatively few environmental organizations, bikeway infrastructure was a minor influence on bicycle-commuting, relative to control variables, but in cities with many environmental organizations, bikeway infrastructure was among the largest influences on changes in bicycle-commuting from 2000 to 2014, second in influence only to gas prices. Similarly, increases in bicycle-commuting predicted substantial new bikeway construction only in cities with high densities of environmental organizations. Thus, this analysis offers strong support for a political conception of social practice and infrastructure. Infrastructure only affects practice when combined with other elements, such as concern for environmental problems, into “configurations that work” (Shove, Pantzar, and Watson 2012), and the influence of practice on infrastructure is mediated by political processes. For scholars interested in the factors shaping social practices more generally, these findings point to the importance of both material infrastructure and mediating political-cultural factors.

The effect of environmental organizations is consistent with the notion that movement organizations can cultivate an “ecological habitus” that teaches skills and encourages attention to the sustainability of daily practices (Haluza-DeLay 2008). Material elements such as infrastructure or bicycles themselves are crucial to bicycling as a social practice, but meanings that make sense of a practice and practical know-how are equally important, and environmental organizations may help connect these elements. This is especially relevant in the US context, where cycling is viewed primarily as a recreational activity (Pucher, Buehler, and Seinen 2011). To become a bicycle commuter, one must view cycling as a viable mode of transportation and acquire different skills and equipment than necessary for recreational cycling. Environmental organizations may introduce recreational cyclists to utilitarian cycling, and teach them about requisite skills and equipment.

Despite the encouraging conclusion that creating bicycle infrastructure can induce more bicycling, my analysis finds relatively small effect sizes for bicycle infrastructure, and the overall mode share of bicycle-commuting in US cities is still quite low. However, there are several reasons to be optimistic about the potential growth of bicycling in US cities. First, the isolated effect of bikeways may be small, but cities that have instituted a package of pro-bicycle policies have achieved rapid growth—in the case of Portland, OR, from 1.8 percent mode share in 2000 to 7.9 percent in 2014. Second, development of bikeway infrastructure in the United States is at an early stage, such that even in more bicycle-friendly cities, most trips cannot be completed exclusively on bicycle-dedicated infrastructure. Many potential cyclists may be dissuaded if they must share the road with car traffic at any point in their journey, so growth in bicycling may accelerate as bikeway-network connectivity increases. Third, planning preferences are shifting toward protected bicycle infrastructure, which could attract riders who demand separation from car traffic. Fourth, growth in bicycling may accelerate due to the “safety in numbers” phenomenon: bicycling becomes demonstrably safer as the number of bicyclists on the road increases (Jacobsen 2003). Finally, the promotion of bicycling in the United States has thus far proceeded mostly without efforts to restrict or increase the costs of car travel. Restricting car traffic through measures such as congestion charges may be doubly effective because it pushes some drivers to alternative modes, and simultaneously makes walking and bicycling safer and more pleasant by reducing car traffic (Pucher and Buehler 2008). If the political will can be mustered to make car travel less convenient (an admittedly big “if”), the growth potential of bicycling would increase substantially.

My results also highlight the influence of gas prices on travel behavior. Even measured at the state level, gas prices had a strong positive effect on bicycle-commuting. In fact, the largest single year of growth in bicycle-commuting occurred from 2007 to 2008, perhaps spurred by eight years of increasing gas prices. Following the drop in gas prices from 2008 to 2009, growth in bicycle-commuting plateaued from 2009 to 2010, before recovering as gas prices picked back up. These formal and descriptive findings add to past work on the influence of gas prices on travel behavior (Chi et al. 2010; Chi and Boydstun forthcoming).

Future research should address the limitations of this analysis. One limitation is the lack of data on key quantities of interest. This includes a general scarcity of longitudinal data on social practices, changes in infrastructure, and political-cultural trends relevant to practices of interest. The ACS provides an increasingly long, reliable, annual time series of data on bicycle-commuting, but data on infrastructure has spottier coverage and is less reliable due to reliance on city self-reports. The construction of bikeways likely increases bicycling for all purposes, but there are currently no city-level data on cycling for non-work purposes. Although the ACS provides data on commute times, it does not measure commute distance, which is doubtless an important determinant of travel mode choices. Moreover, longitudinal data on cultural beliefs, such as environmental concern, at the city or state level is virtually non-existent. Including a direct measure of environmental concern would have allowed us to separate out the effects of individual beliefs from the organizational resources provided by environmental organizations. Finally, studies using disaggregate data on travel behavior will continue to be an important complement to aggregate studies in understanding the effects of cultural beliefs and infrastructure on practice.

For most of the twentieth century, many factors aligned to tighten the grip of automobility in the United States. In recent decades, some of these factors have shifted in ways that could loosen that grip. Deindustrialization and urban revitalization have transformed cities from production to consumption centers, and the growth of suburbs has slowed relative to central cities (Frey 2012; Zukin 1989). Meanwhile, planners have come to view urban sprawl as unsustainable and have shifted from car-dependent fringe development to walkable and bikeable urban-infill development. Creating amenity-rich communities is now seen as crucial to attracting an educated, postindustrial workforce (Florida 2002). Concern about the impact of automobile emissions on climate change has been added to a list of automobile-related concerns such as air and water pollution, public safety, and foreign oil dependence. The results of the preceding analysis suggest that shifts in transportation infrastructure have a part to play in this larger constellation of changes that may eventually precipitate a “tipping point” and the emergence of a transportation system less dependent on the “steel and petroleum” automobile (Urry 2004).

About the Author

Derek Burk is a PhD candidate in sociology at Northwestern University. His research interests include environmental policymaking, social movements, social practice theory, and quantitative and comparative methods. His dissertation examines how local political climate, trends in the planning profession, and city branding efforts contribute to variation in the adoption of bicycle-friendly policies.

Supplementary Material

Supplementary data is available at Social Forces online.

Author notes

Thanks to Monica Prasad, Wendy Griswold, and James Mahoney for their guidance and feedback on this project and drafts of the paper, and thanks to members of the Culture and Society, SION, Social Movements/Social Enterprise, and Applied Quantitative Methods workshops at Northwestern University for their discussion and feedback. The author thanks Lincoln Quillian for advising him on the methods used in this paper, and Robert Brulle, Lawrence Hamilton, Paul Allison, and Ralph Buehler for feedback on specific aspects of the analysis. This paper greatly benefited from the feedback of three anonymous Social Forces reviewers. Thanks to Amiee Burk for steadfast support and encouragement. Any mistakes or omissions in this work remain those of the author. Please address correspondence concerning this article to Derek Burk, Northwestern University, Department of Sociology, 1810 Chicago Avenue, Evanston, IL 60208, USA. E-mail: .