Disparities in Bicycle Commuting: Could Bike Lane Investment Widen the Gap?

Abstract

Spanish

Chinese

This article examines how bicycle commuting is associated with bike lane access and sociodemographic advantage at the block group level in twenty-two U.S. cities. Using regression models with interaction terms, I find that associations between bike lanes and bicycle commuting are stronger among more advantaged block groups, due primarily to limited bicycle commuting in disadvantaged block groups even when bike lanes are present. This indicates the importance of considering heterogeneous effects in planning research and suggests that bike lane investment, in isolation, could widen sociodemographic disparities in cycling and its benefits if non-infrastructure barriers to cycling are not also addressed.

Keywords

transportation cycling equity bike lanes health disparities

Introduction

Cities have increasingly invested in their bike lane networks over the past decade, given evidence that bike lanes can facilitate cycling (Buehler and Pucher 2012; Dill and Carr 2003) and that cycling, in turn, is associated with potential benefits for the environment (Chapman et al. 2018; Keall et al. 2018; Neves and Brand 2019; Zhang and Mi 2018), the economy (Ma and Ye 2019; Poirier 2018), access to opportunity (Murphy and Owen 2019; Pritchard et al. 2019), and public health (Chapman et al. 2018; Pucher et al. 2010; Singleton 2019; Wanner et al. 2012). Recently, cycling advocates and researchers have suggested that bike lane investments tend to be made in places with greater sociodemographic advantage (Flanagan, Lachapelle, and El-Geneidy 2016; Hirsch et al. 2017; League of American Bicyclists 2013; Tucker and Manaugh 2018; Wang and Lindsey 2017), leading to lower access among low-income populations and people of color who are disproportionately burdened by the very environmental, economic, and health outcomes that cycling could influence (August and Sorkin 2011; Lee, Vojnovic, and Grady 2018; Li, Konisky, and Zirogiannis 2019; Loustaunau and Chakraborty 2019). These observations have led some to argue that investing in a more equitable distribution of bike lanes could help to reduce sociodemographic disparities in cycling and its potential benefits (Rachele et al. 2015), and have subsequently led to the development of new methods for prioritizing bike lane investments in places of lower sociodemographic advantage (Grisé and El-Geneidy 2018; Kent and Karner 2019).

Despite this growing emphasis, the relationship between bike lanes and social equity is complex and important questions remain unresolved. For instance, although bike lanes tend to be positively associated with cycling, few studies have considered whether these associations vary by sociodemographic characteristics such as race, ethnicity, and socioeconomic status (SES) (Smith et al. 2017). This is an important possibility to consider, given that inadequate infrastructure access is just one among many individual, environmental, social, and institutional barriers to cycling that low-income populations and people of color face (Golub 2016). Within this context, bike lanes could have the unintended consequence of increasing sociodemographic disparities in cycling and its benefits if advantaged populations are more likely to use them. Thus, examining potential heterogeneity in associations between bike lanes and cycling is critical for more fully understanding the equity implications of bike lane investment.

I address this research gap by examining cross-sectional, area-level associations between bicycle commuting, sociodemographic advantage, and bike lane presence in twenty-two large central cities in the United States. Using census block groups as the unit of analysis, I conduct a moderation analysis to assess whether associations between bike lane presence and bicycle commuting vary by sociodemographic context. In the sections that follow, I begin by summarizing how sociodemographic characteristics are related to cycling and describe how this relationship is typically treated in cycling research, establishing the value of a moderation approach. I then explain the data and methods used in this analysis and how they are equipped to address the relationships of interest. Finally, I present the results and discuss their implications, both for the broader planning literature and for cycling research in particular. Broadly, the findings illustrate the importance of considering social heterogeneity in the study of planning phenomena. Specifically, the findings provide insight into the complex relationship between bike lanes and social equity, suggesting the potential for infrastructure investment to widen the gap in cycling and its outcomes between places of higher and lower sociodemographic advantage if non-infrastructure barriers to cycling are not also addressed.

Sociodemographic Advantage and Cycling: Connections and Methods

Measures of sociodemographic advantage have recently been used to address two questions relevant to social equity in cycling: Who cycles, and who has access to cycling infrastructure? Answers to the first question have been mixed, with some studies finding levels of cycling to be lower among those with relatively low income and educational attainment (Adams 2010; Mäkinen et al. 2008; Rachele et al. 2015) and others finding levels of cycling to be higher in the context of individual- and neighborhood-level disadvantage (Jurakić, Pedišić, and Andrijašević 2009; League of American Bicyclists 2013; Van Dyck et al. 2010). Longitudinal analyses have provided further insight into the sociodemographic composition of cycling over time, indicating that growth in cycling over the past decade has tended to be fastest among low-income populations and people of color (League of American Bicyclists 2013).

Recent research has added depth and context to this work by considering the second question—that is, by examining sociodemographic disparities in access to cycling infrastructure. This research has generally found that access to bike lanes, urban trails, and low-stress cycling routes tends to be greater among more advantaged populations, with some exceptions and variations across cities (Braun, Rodriguez, and Gordon-Larsen 2019; Tucker and Manaugh 2018; Wang and Lindsey 2017; Winters et al. 2018). Similarly, longitudinal analyses have found that investments in bike lanes and trails tend to occur disproportionately in areas that are more advantaged to begin with or increasing in advantage (e.g., gentrifying) over time (Flanagan, Lachapelle, and El-Geneidy 2016; Hirsch et al. 2017). Others have observed sociodemographic disparities in access to bicycle sharing stations (Babagoli et al. 2019; Clark and Curl 2016; Hosford and Winters 2018), corresponding with related research that has found bicycle sharing program users to be disproportionately white, wealthy, and highly educated (Gavin et al. 2016; Heinen, Kamruzzaman, and Turrell 2018).

In light of these findings, advocates and scholars alike have recently called for more equitable patterns of investment in cycling infrastructure such as bike lanes (Babagoli et al. 2019; Grisé and El-Geneidy 2018; Hosford and Winters 2018; Kent and Karner 2019; League of American Bicyclists 2013; Rachele et al. 2015; Wang and Lindsey 2017; Winters et al. 2018). These calls have been based in part upon evidence that bike lanes are positively associated with cycling (Buehler and Pucher 2012; Dill and Carr 2003) and inversely associated with the risk of bicycle crashes (Marshall and Ferenchak 2019; Park et al. 2015), suggesting that a more equitable distribution of bike lanes could facilitate higher levels of cycling among disadvantaged populations while enhancing safety for those who already cycle.

Despite this potential, limited quantitative research has critically considered whether bike lane investment can adequately support cycling in diverse sociodemographic contexts. Indeed, most quantitative research on cycling behavior has treated sociodemographic characteristics as secondary control variables, allowing for greater focus on determinants such as bike lanes that are of more primary interest (Adkins et al. 2017; Cao, Mokhtarian, and Handy 2009). Few have instead considered the potential moderating influence of sociodemographic characteristics—that is, how associations between bike lanes and cycling themselves may vary by characteristics such as race, ethnicity, and SES. Notable exceptions include Goodman, Sahlqvist and Ogilvie (2013), who found that individuals with lower educational attainment and income were less likely to use new cycling routes in three U.K. cities; Goodman, Panter, et al. (2013), who found that the impacts of cycling interventions in eighteen U.K. cities tended to be stronger in socioeconomically deprived areas; and Parker, Gustat, and Rice (2011), who found that the ridership impacts of a new bike lane in New Orleans did not vary by race or ethnicity. There is value in expanding this area of the literature by continuing to foreground sociodemographic characteristics, as understanding how the determinants of cycling vary across sociodemographic groups could inform more equitable and contextually appropriate cycling promotion strategies.

A precedent for this type of moderation analysis can be found in recent research on the built environment and walking behavior. Adkins et al. (2017) reviewed seventeen studies and found that in thirteen, associations between the built environment and walking behavior were up to two times stronger among groups that were advantaged in terms of race, ethnicity, and educational attainment, relative to disadvantaged groups. To understand these differences, the authors noted that the impacts of the built environment on walking behavior may be stronger for advantaged groups for two primary reasons: (1) disadvantaged groups may walk more than advantaged groups in unsupportive environments and (2) disadvantaged groups may walk less than advantaged groups in supportive environments. These two factors, in combination with measurement error, could make it appear as though associations between the built environment and walking behavior are weaker in the context of sociodemographic disadvantage.

Figure 1 applies this concept to the case of cycling and forms the analytical framework for the present analysis. In this figure, a theoretically positive association between bike lanes and cycling is represented by the dashed black line. This association tends to be made stronger by observations (e.g., people, places) that fall into the two white quadrants that represent “expected” behavior (i.e., low levels of cycling where access to bike lanes is low, high levels of cycling where access to bike lanes is high). The association could be made weaker, however, by observations that fall into the two shaded quadrants that represent “unexpected” behavior—that is, by high levels of cycling even in the absence of bike lanes (quadrant (a)) and low levels of cycling even when bike lanes are present (quadrant (b)). If observations characterized by sociodemographic disadvantage are disproportionately likely to fall into either of these quadrants, associations between bike lanes and cycling could appear to be weaker for these observations relative to those with greater sociodemographic advantage.

Figure 1.

Conceptual framework for analysis.

There are several reasons to expect either of these possibilities. First, disadvantaged groups may cycle more than advantaged groups even in the absence of bike lanes (quadrant (a)) in Figure 1—due to constraints or preferences. Low-SES populations tend to have lower rates of vehicle ownership (Blumenberg and Pierce 2012) and may therefore be constrained to other travel modes such as cycling, even when bike lanes are not available. Alternatively, while cycling is not universally seen as a desired activity (Golub 2016), it is possible that disadvantaged populations prefer to cycle and will do so in the absence of environmental supports. As previously noted, some researchers have found levels of cycling to be relatively high among disadvantaged populations and communities (Jurakić, Pedišić, and Andrijašević 2009; League of American Bicyclists 2013; Van Dyck et al. 2010). These patterns, combined with evidence that bike lanes are less likely to be located in disadvantaged neighborhoods (Flanagan, Lachapelle, and El-Geneidy 2016; Hirsch et al. 2017; League of American Bicyclists 2013; Tucker and Manaugh 2018; Wang and Lindsey 2017; Winters et al. 2018), suggest that disadvantaged groups may cycle more than advantaged groups in the absence of supportive infrastructure, potentially contributing to attenuated associations between bike lanes and cycling.

Second, disadvantaged groups may cycle less than advantaged groups even in the presence of bike lanes (quadrant (b)) in Figure 1—due to non-infrastructure barriers to cycling. Adkins et al. (2017) found more empirical support for this explanation in their review of pedestrian research, attributing this finding in part to a lack of other supports for walking even in places that are objectively “walkable” according to traditional built environment measures. Applying this possibility to the present analysis, levels of cycling could be low even in the presence of bike lanes because inadequate infrastructure is just one of many barriers to cycling faced by low-income populations and people of color. Individual barriers may be financial and logistical in nature, as the costs of owning and maintaining a bicycle can be significant (Community Cycling Center 2012), disadvantaged groups may have lower access to secure bicycle parking (Lusk et al. 2017), and the suburbanization of poverty has led to longer commutes for low-income populations (Howell and Timberlake 2014). Other barriers may be environmental, as disadvantaged neighborhoods tend to experience disproportionate traffic safety risks (Ferenchak and Marshall 2019) and may lack other built environment features that support active transportation even if bike lanes are present (Kelly et al. 2007). Potential social barriers include fears of crime, discrimination, and racial profiling while cycling (Lubitow 2017), as well as social norms that often frame cycling as a predominantly white, high-SES mode of travel (Golub 2016; Sarrica et al. 2019). These barriers are related to institutional factors—such as the underrepresentation of low-income populations and people of color in decision making (Golub 2016) and the frequent framing of bike lane investment as an economic development strategy consistent with gentrification (Ibsen and Olesen 2018; Lubitow, Zinschlag, and Rochester 2016)—that lead to disadvantaged populations being excluded or disengaged from processes of bicycle planning and advocacy. Given these non-infrastructure barriers, bike lanes in isolation could appear to be a relatively weak determinant of cycling among disadvantaged groups.

To build upon this area of the literature, I ask the following primary research question:

Research Question: Do associations between bike lane presence and bicycle commuting vary by (i.e., are they moderated by) area-level measures of sociodemographic advantage?

Based on the findings of Adkins et al. (2017) with respect to the built environment and walking behavior, I expect to find that associations between bike lane presence and bicycle commuting are stronger among more advantaged block groups and correspondingly weaker among less advantaged block groups. I also use the conceptual framework presented in Figure 1 to add depth to these findings and to explore potential explanations in line with existing literature on social equity in cycling.

Data and Variables

Study Sample

I examined cross-sectional associations between bicycle commuting, sociodemographic advantage, and bike lane presence in twenty-two large central cities in the United States, with all variables aggregated to the census block group level. I began with the primary cities of the twenty-five most populous U.S. metropolitan areas (based on one-year estimates from the 2015 American Community Survey [ACS]) and excluded three cities—Baltimore, MD; St. Louis, MO; and Riverside, CA—for which current bike lane data could not be obtained. I selected all block groups whose centroids fell within the city limits of the remaining twenty-two cities (n = 23,003) and then excluded those with missing data on one or more variables of interest for this analysis, n = 1,251 (5%) excluded. The final study sample thus consisted of n = 21,752 block groups across the twenty-two cities displayed in Figure 2.

Figure 2.

Map of cities included in analysis.

Data Sources

Block group–level data on bicycle commuting and sociodemographic advantage were collected from the 2012–2016 ACS. Block group–level data on covariates were collected from the 2012–2016 ACS and the 2014 Longitudinal Employer–Household Dynamics data. Bike infrastructure data came from city-specific GIS shapefiles, which I collected from local and regional administrative sources (e.g., open data repositories, other government websites) in each city or through email correspondence with local planners when shapefiles were not available online. The dates of the twenty-two shapefiles correspond with the date range of the ACS data used for this analysis (i.e., 2012–2016), although the majority were dated between 2014 and 2016.

As each city used a unique classification system to describe its infrastructure, I reviewed each shapefile alongside Google Earth imagery to determine how various infrastructure types (e.g., trails, lanes, sharrows, routes) and statuses (e.g., planned vs. existing) were coded in the attribute fields. I used this information to consistently classify infrastructure types across all cities and to retain only existing on-street bike lanes (unprotected, buffered, or protected) from each shapefile. I spatially joined the resulting bike lane data to the block groups in my study sample, buffering block groups by ten meters to allow bike lanes along boundaries to be joined with all block groups sharing those boundaries.

Variables

Dependent variable: Bicycle commuting

The dependent variable was a measure of bicycle commuting derived from the ACS, which asked respondents to report the travel mode they generally used for their journey to work during the previous week; respondents who used multiple modes were asked to select the single mode used for the greatest distance. I used this information to measure the percentage of workers in each block group who reported commuting to work by bicycle. Given that a substantial proportion (69%) of block groups had no bicycle commuters, I defined the dependent variable as binary (1 if any bicycle commuters, 0 if none) in the main analysis and tested a continuous definition in a sensitivity analysis.

Primary independent variables: Bike lane presence and sociodemographic advantage

The primary independent variables of interest for this analysis were block group–level measures of bike lane presence and sociodemographic advantage. To measure bike lane access, I created a binary measure of bike lane presence indicating whether a block group contained any bike lanes (1 if any, 0 if none). For descriptive purposes and in a sensitivity analysis testing an alternative definition of bike lane access, I also created a continuous measure of bike lane density in meters per square mile of land area.

I measured a variety of sociodemographic characteristics to capture multiple dimensions of (dis)advantage. First, I measured the following five characteristics as separate variables: percentage non-Hispanic Black, percentage Hispanic or Latino, percentage below the federal poverty line, median household income, and percentage with a bachelor’s degree or more. Second, I estimated a composite SES index based on the methodology of Christine et al. (2015). This process involved performing a factor analysis on seventeen ACS variables related to race, ethnicity, income, educational attainment, poverty, housing, and employment. The first factor was weighted heavily (i.e., rotated factor loading > 0.4) on the nine variables listed in Table 1. I used z-scores to standardize each of these nine variables and then combined them into a single additive index, applying the rotated factor loadings as weights. The values of the resulting index ranged from approximately −14 to +16, with higher values reflecting higher SES.

Table 1.

Components of Additive Socioeconomic Status Index.

Variable	Weight^a
% Hispanic or Latino	−0.5509
Median household income	0.7667
% of households with income ≥ $50,000	0.7295
% of households with interest, dividend, or net rental income	0.7738
% with at least a high school diploma	0.7423
% with at least a bachelor’s degree	0.9489
% below federal poverty line	−0.5394
% of households with crowding greater than one person per room	−0.4303
% of workers with executive, managerial, or professional occupation	−0.9102

Weights based on rotated factor loadings from factor analysis of seventeen initial sociodemographic variables.

Covariates: Urban form and population age structure

To account for other factors that may influence bicycle commuting, I measured several covariates describing urban form and population age structure. Specifically, I accounted for urban form through measures of population density (persons/mile²), employment density (jobs/mile²), and distance to downtown (city hall); and for population age structure by measuring the percentage of block group residents between ages eighteen and thirty-four.

Method

Descriptive Analysis

First, I used descriptive statistics and corresponding statistical tests (analysis of variance [ANOVA], χ²) to examine the characteristics of the study sample and to identify potential differences between block groups with zero versus non-zero shares of bicycle commuting. Although this descriptive analysis did not adjust for covariates, it provided a preliminary understanding of potential disparities in bicycle commuting in the study sample.

Regression Analysis

Next, I estimated a set of logistic regression models to examine how bicycle commuting (binary dependent variable) was associated with sociodemographic advantage and bike lane presence (independent variables), adjusting for covariates. I accounted for the hierarchical structure of the data by estimating logistic multilevel mixed-effects (MLME) regression models with two levels (block groups nested within cities) and clustering all standard errors at the city level. This approach helped to account for the potential similarity, and thus spatial dependence, between block groups located within the same city.

The regression models included a categorical-by-continuous interaction term between bike lane presence (binary) and the sociodemographic characteristic of interest (continuous), which allowed for a moderation analysis—that is, the coefficients from these models could be used to assess whether associations between bike lane presence and bicycle commuting varied by level of sociodemographic advantage. I estimated a separate logistic regression model for each sociodemographic characteristic using the specification in equation (1):

\begin{array}{l} {BIKE}_{i j} = β_{0} + β_{1} {LANES}_{i j} + β_{2} {SD}_{i j} \\ + β_{3} ({LANES}_{i j} \times {SD}_{i j}) + β_{k} C o v_{i j} + u_{j} + ε_{i j}, \end{array}

(1)

where i indexes block groups, j indexes cities, and

${BIKE}_{i j}$ = binary indicator of whether a block group has bicycle commuters (1 if any, 0 if none); ${LANES}_{i j}$ = binary indicator of whether a block group contains bike lanes (1 if any, 0 if none); ${SD}_{i j}$ = continuous sociodemographic characteristic of interest (separate model for each characteristic); $({LANES}_{i j} \times {SD}_{i j})$ = interaction between bike lane presence and sociodemographic characteristic of interest; $C o v_{i j}$ = vector of continuous urban form and age covariates; $u_{j}$ = random intercept that shifts the regression line up or down for each city; and $ε_{i j}$ = error term.

Due to the interaction term, the primary coefficients of interest from these models— $β_{1}$ , $β_{2}$ , and $β_{3}$ —could only be meaningfully interpreted with reference to one another; indeed, by design, associations between $LANES$ and $BIKE$ varied by the level of $SD$ , while associations between $SD$ and $BIKE$ varied by the level of $LANES$ . To examine this interaction more closely, I used the regression coefficients from equation (1) to calculate predictive margins of bike lane presence—that is, the predicted probabilities of having any bicycle commuters for block groups with and without bike lanes—at varying levels of each sociodemographic characteristic. For characteristics measured as percentages, I evaluated predictive margins at ten percentage-point intervals ranging from 0 to 100. For median household income, I evaluated predictive margins at $25,000 intervals up to the maximum observed value ($250,000). For the SES index, I evaluated predictive margins at two-unit intervals across the observed range of values (–14 to +16). I presented the results in a set of graphs that showed the magnitude and statistical significance of the predictive margins across each of these variable-specific ranges.

Additional and Sensitivity Analyses

I conducted two additional analyses to further examine the regression results. First, to test alternative explanations for the observed regression results, I calculated the percentage of disadvantaged block groups that fell within the (a) and (b) quadrants of Figure 1 (i.e., block groups that had bicycle commuters but no bike lanes, and block groups that had bike lanes but no bicycle commuters). In this process, I defined disadvantaged block groups as those that fell below the median values of all six sociodemographic characteristics considered in this analysis, although I also tested alternative definitions framed around the median value of each separate sociodemographic characteristic. Second, I examined the random effects parameters and intraclass correlation coefficients (ICCs) from each regression model to more fully consider city-level variations in bicycle commuting.

I also conducted three sensitivity analyses to test the robustness of the results. To assess whether the findings were primarily applicable to cities with high levels of bicycle commuting, I tested an alternative version of the study sample restricted to seven cities in the data that were among the “top 10” cities with the highest bicycle commuting mode shares in 2016 (Alliance for Biking & Walking 2016). Next, I re-estimated the regressions using bike lane density as an alternative measure of bike lane access. Given that 57 percent of block groups had no bike lanes, I estimated these models for the full sample (predicting bicycle commuting as a function of binary bike lane presence [any vs. none]) and then for the subsample of block groups that contained any bike lanes (predicting bicycle commuting as a function of continuous bike lane density). Third, to incorporate information about the level (rather than the presence or absence) of bicycle commuting, I re-estimated equation (1) using two-part models, including (1) a logistic MLME regression modeling the likelihood of having any bicycle commuting (identical to the main models used in this analysis) and (2) a linear MLME regression modeling the amount of bicycle commuting among block groups that had any.

Results

Descriptive Analysis

On average, 1.4 percent of workers across all block groups commuted to work by bicycle (Table 2). This mode share is greater than the national average of 0.6 percent (U.S. Census Bureau 2016), reflecting higher levels of cycling in the large cities that were the focus of this analysis. Approximately 70 percent of block groups had no bicycle commuters; among the remaining block groups with any bicycle commuters, the average bicycle commuting mode share was 4.51 percent.

Table 2.

Descriptive Statistics, Total and by Bicycle Commuting Mode Share, 21,752 Block Groups in Twenty-Two U.S. Cities.

	All block groupsn = 21,752	By bicycle commuting mode share
	All block groupsn = 21,752	Zeron = 15,022	Non-zeron = 6,730	p ^a
	M (SD)^b	M (SD)^b	M (SD)^b	p ^a
Bicycle commuting
% of workers who commute by bicycle	1.40 (3.11)	—	4.51 (4.14)	—
Sociodemographic characteristics
% non-Hispanic black	23.09 (31.30)	27.14 (33.88)	14.04 (22.04)	.00***
% Hispanic	29.09 (28.77)	30.95 (29.72)	24.93 (26.08)	.00***
% below poverty line	20.43 (16.17)	21.35 (16.73)	18.36 (14.63)	.00***
Median household income ($1,000s)	60.73 (37.48)	57.25 (36.34)	68.50 (38.81)	.00***
% with bachelor’s degree or more	34.72 (25.27)	29.45 (23.10)	46.49 (25.93)	.00***
Composite SES index	0.00 (5.20)	−0.75 (4.99)	1.67 (5.25)	.00***
Bike lane access
Presence of bike lanes (yes), n (%)^b	9,355 (43%)	5,711 (38%)	3,644 (54%)	.00***
Bike lane density (100 m/mile²)	38.06 (79.87)	30.85 (71.60)	54.16 (93.82)	.00***
Covariates
Population density (1,000 persons/mile²)	28.15 (38.45)	26.98 (38.84)	30.75 (37.43)	.00***
Employment density (1,000 jobs/mile²)	9.54 (51.98)	7.45 (38.54)	14.19 (73.40)	.00***
Distance to downtown (km)	11.59 (7.30)	12.99 (7.15)	8.47 (6.63)	.00***
Age (% 18–34)	27.90 (12.47)	25.76 (10.67)	32.68 (14.69)	.00***

Note: SES = socioeconomic status; ANOVA = analysis of variance.

p-value for group differences based on ANOVA or χ² tests; significance not estimated for bicycle commuting, as this variable was used to create groups.

Values in the table are means (with standard deviations in parentheses), with the exception of the binary “Presence of bike lanes” variable, for which the number (and percent) of block groups with bike lanes are given instead.

***

Significant at 99 percent confidence.

Block groups with any bicycle commuting were significantly different from block groups with no bicycle commuting on all independent variables considered in this analysis. Compared to block groups with no bicycle commuting, those with any bicycle commuting had lower percentages of non-Hispanic Black and Hispanic residents; lower poverty levels; and higher median household income, educational attainment, and composite SES. Block groups with bicycle commuting were also more likely to have bike lanes; had a higher density of bike lanes; had higher population and employment densities and percentages of young residents; and were closer to downtown.

Regression Analysis

In adjusted regression models (Table 3), the coefficient on bike lane presence $(β_{1})$ generally indicates a positive association between bike lane presence and bicycle commuting. Similarly, the coefficient on each sociodemographic characteristic $(β_{2})$ generally indicates a positive association between sociodemographic advantage and bicycle commuting; specifically, higher percentages of non-Hispanic Black residents and residents below the poverty line were inversely associated with bicycle commuting, while median household income, educational attainment, and composite SES were positively associated with bicycle commuting. The coefficient on the interaction term $(β_{3})$ was statistically significant in four of the six regression models, suggesting that observed associations between bike lane presence and bicycle commuting were moderated by these four sociodemographic characteristics. Measures of overall fit were similar across the six models, although slightly better (i.e., lower Akaike’s information criterion values) in the models for percentage non-Hispanic Black (Model 1), percentage with a bachelor’s degree (Model 4), and composite SES (Model 6).

Table 3.

Adjusted Associations of Any Bicycle Commuting (Yes/No) with Sociodemographic (SD) Characteristics and Bike Lane Access (LANES), 21,752 Block Groups in Twenty-Two U.S. Cities.

	Associations with bicycle commuting mode share (separate models by SD characteristic)
	Model 1:SD = % black		Model 2:SD = % Hispanic		Model 3:SD = % poverty
	Coefficient (SE)	p	Coefficient (SE)	p	Coefficient (SE)	p
$β_{1}$ (LANES)	0.308 (0.060)	.00***	0.447 (0.123)	.00***	0.469 (0.106)	.00***
$β_{2}$ (SD)	−0.015 (0.003)	.00***	−0.002 (0.002)	.26	−0.012 (0.004)	.01***
$β_{3}$ (LANES × SD interaction)	0.001 (0.002)	.65	−0.005 (0.002)	.04**	−0.008 (0.004)	.02**
Pop. density (1,000 persons/mile²)	0.003 (0.001)	.75	0.013 (0.011)	.23	0.015 (0.012)	.21
Emp. density (1,000 jobs/mile²)	−0.006 (0.005)	.16	−0.003 (0.004)	.33	−0.005 (0.004)	.28
Distance to downtown (km)	−0.082 (0.030)	.01***	−0.083 (0.031)	.01***	−0.088 (0.031)	.01***
Age (% 18–34)	0.030 (0.003)	.00***	0.033 (0.004)	.00***	0.034 (0.004)	.00***
Constant	−0.684 (0.361)	.06*	−1.046 (0.428)	.01**	−0.850 (0.442)	.05*
Intraclass correlation coefficient	.18		.20		.19
Akaike’s information criterion	21,745.89		22,158.72		22,010.09
	Model 4:SD = % with bachelor’s		Model 5:SD = income		Model 6:SD = SES index
	Coefficient (SE)	p	Coefficient (SE)	p	Coefficient (SE)	p
$β_{1}$ (LANES)	0.043 (0.088)	.63	0.189 (0.091)	.04**	0.245 (0.058)	.00***
$β_{2}$ (SD)	0.016 (0.003)	.00***	0.007 (0.001)	.00***	0.057 (0.016)	.00***
$β_{3}$ (LANES × SD interaction)	0.005 (0.002)	.02**	0.002 (0.001)	.19	0.030 (0.012)	.01**
Pop. density (1,000 persons/mile²)	0.005 (0.014)	.74	0.012 (0.013)	.37	0.016 (0.014)	.26
Emp. density (1,000 jobs/mile²)	−0.017 (0.006)	.01***	−0.011 (0.006)	.06*	−0.013 (0.006)	.04**
Distance to downtown (km)	−0.077 (0.025)	.00***	−0.083 (0.028)	.00***	−0.084 (0.027)	.02**
Age (% 18–34)	0.027 (0.003)	.00***	0.036 (0.004)	.00***	0.032 (0.003)	.00***
Constant	−1.537 (0.323)	.00***	−1.581 (0.383)	.00***	−1.085 (0.362)	0.00***
Intraclass correlation coefficient	.18		.19		.18
Akaike’s information criterion	21,537.73		21,944.96		21,766.75

Note: All models estimated using multilevel mixed-effects (MLME) logistic regression with two-level clustering (block groups nested within cities). Shaded rows indicate coefficients of primary interest for the moderation analysis.

Significant at 90 percent confidence. **Significant at 95 percent confidence. ***Significant at 99 percent confidence.

These associations can be seen more clearly in Figure 3, which translates the coefficients from Table 3 into predicted probabilities of bicycle commuting. This exercise is important not only for understanding the nature of the significant interactions but also for more fully understanding the other regression coefficients; indeed, due to the interaction term, $β_{1}$ is only directly interpretable when the given sociodemographic characteristic is equal to zero, and $β_{2}$ is only directly interpretable in the absence of bike lanes (i.e., when $LANES$ = 0). In each panel of Figure 3, the solid black line represents the association between the given sociodemographic characteristic and the predicted probability of bicycle commuting among block groups with bike lanes, and the dashed black line shows the same association among block groups without bike lanes. The shaded region represents the range of each sociodemographic characteristic for which the association between bike lane presence and bicycle commuting—that is, the vertical distance or “gap” between the solid and dashed lines—is statistically significant at 90 percent confidence. Each panel thus presents three key pieces of information: (1) the overall association between sociodemographic advantage and bicycle commuting, (2) the overall association between bike lane presence and bicycle commuting, and (3) the way in which the second association varies by degree of sociodemographic advantage.

Figure 3.

Predicted probability of having a non-zero bicycle commuting mode share, by block group–level sociodemographic characteristics and bike lane presence, 21,752 block groups in twenty-two U.S. cities.

An interpretation of one panel of Figure 3 is as follows. In Model 1, the negative slopes of both lines indicate that higher percentages of non-Hispanic Black residents are associated with lower predicted probabilities of bicycle commuting, regardless of bike lane presence; this reflects the negative $β_{2}$ coefficient for this model in Table 3. The solid line is consistently above the dashed line, suggesting a positive association between bike lane presence and bicycle commuting and thus reflecting the direction of the $β_{1}$ coefficient for this model in Table 3. This association is statistically significant across the full range of the selected sociodemographic characteristic; however, this association does not vary (i.e., the distance between the solid and dashed lines does not significantly change) by the percentage of non-Hispanic Black residents, reflecting the non-significant coefficient on the interaction term $(β_{3})$ for this model in Table 3.

Interpreting all panels of Figure 3 in this way suggests that, in general, a block group’s probability of having bicycle commuters is positively associated both with sociodemographic advantage (shown by the negative slopes in Models 1–3 and the positive slopes in Models 4–6) and with bike lane presence (shown by the vertical gap between the dashed and solid lines). Moreover, the association between bike lane presence and bicycle commuting varies significantly across the range of four sociodemographic characteristics. For Models 2 and 3, the association between bike lane presence and bicycle commuting becomes smaller with increasing percentages of Hispanic residents and residents below the poverty line. As a result, associations between bike lane presence and bicycle commuting are only significant when the percentage of Hispanic residents is less than seventy (Model 2) and when the percentage of residents below the poverty line is less than fifty (Model 3). Beyond these thresholds of sociodemographic disadvantage, bike lane presence is not significantly associated with bicycle commuting.

For Models 5 and 6, the association between bike lane presence and bicycle commuting becomes larger with increasing educational attainment and composite SES. As a result, associations between bike lane presence and bicycle commuting are only significant when the percentage of residents with a bachelor’s degree is greater than twenty (Model 5) and when the value of the composite SES index is greater than −4 (roughly equivalent to the 25th percentile in the SES index distribution; Model 6). Below these thresholds of sociodemographic advantage, bike lane presence is not significantly associated with bicycle commuting.

Additional and Sensitivity Analyses

The results of the main analysis thus suggest that associations between bike lane presence and bicycle commuting are generally weaker in the context of sociodemographic disadvantage. As noted by Adkins et al. (2017) and illustrated in Figure 1, such an attenuation could arise from cycling in the absence of bike lanes (quadrant (a)), potentially reflecting cycling preferences or lack of alternative transportation options; or from a lack of cycling in the presence of bike lanes (quadrant (b)), potentially reflecting individual, environmental, social, and institutional barriers that keep people from cycling even when bike lanes are available. To test these alternative explanations for the regression results, I calculated the percentage of “disadvantaged” block groups that fell within each quadrant of this conceptual framework, defining “disadvantaged” as falling below the median value of all six sociodemographic characteristics considered in this analysis (n = 3,032, or 14% of the study sample). As shown in Figure 4, block groups meeting this definition of disadvantage were nearly three times as likely to fall in quadrant (b) (no bicycle commuting even in the presence of bike lanes) than in quadrant (a) (bicycle commuting even in the absence of bike lanes). Although this tendency was also observed for the full sample, it was weaker for the full sample—less than twice as likely to fall in (b) than in (a)—than for disadvantaged block groups. A similar pattern of results was found for all alternative definitions of disadvantaged block groups tested (data not shown). Thus, this additional analysis indicates greater support for the explanation represented by quadrant (b)—that is, that associations between bike lane presence and bicycle commuting tend to be lower in disadvantaged block groups due to a stronger lack of observed cycling even in the presence of bike lanes.

Figure 4.

Results of quadrant analysis examining potential explanations for attenuated associations between bike lane presence and bicycle commuting among disadvantaged block groups.

The estimated random effects parameters from Model 1 (percentage non-Hispanic Black) are summarized in Table 4; a similar pattern of results was found for the remaining five regression models, so for simplicity, only the parameters for Model 1 are shown. This table displays the overall constant for the model as well as the city-specific parameters that shift the regression line up (higher likelihood of bicycle commuting) or down (lower likelihood of bicycle commuting) for each city; the sum of the overall constant and city-level shift indicates each city’s random intercept. The cities are arranged in descending order from the highest to lowest random intercept. The results indicate that some cities tend to have higher levels of bicycle commuting regardless of sociodemographic characteristics, bike lane presence, and covariates (Portland, Washington, Seattle, Minneapolis), while others tend to have lower levels of bicycle commuting (San Antonio, Dallas, Charlotte); this ordering generally corresponds with city-level measures of bicycle commuting and bicycle-supportive policies (Alliance for Biking & Walking 2016). Such city-level differences can also be seen in the ICCs for the six regression models (Table 3). While the ICC values are somewhat low (ranging from .18 to .20), they indicate the presence of city-level variation, and likelihood ratio tests (data not shown) consistently suggested the appropriateness of the MLME regression structure.

Table 4.

Estimates of City-Level Random Effects Parameters From Multilevel Mixed-Effects Regression (Model 1), n = 21,752 Block Groups in Twenty-Two Large U.S. Cities.

City	Parameter
Overall constant	−0.684
City-level shifts
Portland, OR	1.837
Washington, DC	1.196
Seattle, WA	1.154
Minneapolis, MN	1.103
San Francisco, CA	0.729
Los Angeles, CA	0.568
Chicago, IL	0.318
Denver, CO	0.253
Boston, MA	−0.047
New York, NY	−0.130
Phoenix, AZ	−0.160
San Diego, CA	−0.161
Philadelphia, PA	−0.256
Tampa, FL	−0.261
Atlanta, GA	−0.276
Detroit, MI	−0.416
Miami, FL	−0.479
Orlando, FL	−0.532
Houston, TX	−0.555
Charlotte, NC	−1.141
Dallas, TX	−1.422
San Antonio, TX	−1.694

Note: All parameters estimated from Model 1 (% black); similar ordering of cities observed for other five models (not shown). The sum of the overall constant and city-level shift indicates each city’s random intercept.

When the sample was restricted to block groups in seven cities with high shares of bicycle commuting, the coefficients for bike lane presence $(β_{1})$ and sociodemographic characteristics $(β_{2})$ were in the same direction as those in the main analysis, although with occasional changes in statistical significance likely resulting from a smaller sample size (n = 3,934 block groups) (data not shown). However, the interaction terms in this subset of the sample were consistently non-significant, suggesting that associations between bike lane presence and bicycle commuting may not be moderated by sociodemographic advantage in cities with particularly high cycling mode shares. The coefficients on the interaction terms were also generally non-significant when bike lane density was substituted as the measure of bike lane access (data not shown); this suggests that the disparities identified in this analysis may not be moderated by the amount of bike lanes that a block group contains, but instead by whether it contains any bike lanes. When the regression models were re-estimated using two-part models, the majority of coefficients were statistically non-significant in the models predicting the amount of bicycle commuting among block groups that had any (data not shown). This suggests that the relationships identified in this analysis may be most relevant to the discrete distinction between places with versus without observed bicycle commuting, rather than to more subtle differences between places with varying (non-zero) shares of bicycle commuting. For this reason, I focus primarily on the results from the logistic portion of the two-part model.

Discussion

I observed significant associations between bicycle commuting, sociodemographic advantage, and bike lane presence at the block group level across twenty-two large U.S. cities. Although associations between bike lane presence and bicycle commuting were most commonly positive, the magnitude of these associations tended to increase with increasing levels of sociodemographic advantage; specifically, associations between bike lane presence and bicycle commuting tended to be stronger in places with lower percentages of Hispanic residents and residents below the poverty line, and with higher educational attainment and composite SES. Moreover, these associations were only statistically significant past a certain threshold of sociodemographic advantage. These results, which correspond with the findings of Adkins et al. (2017) with respect to the built environment and walking behavior, suggest that associations between bike lane presence and bicycle commuting could be higher in places characterized by greater sociodemographic advantage, and correspondingly lower in disadvantaged areas.

As previously described, there are several potential explanations for this finding. Associations between bike lane presence and bicycle commuting may be lower in disadvantaged block groups if the residents of those block groups tend to cycle even in the absence of supportive infrastructure (quadrant (a) of Figure 1), due to underlying preferences for cycling or to economic constraints such as low vehicle ownership. In this context, bike lane interventions could effectively serve an existing need for safe, convenient infrastructure, even if these interventions do not have a substantial impact on current cycling behavior.

While this explanation is plausible, an alternative explanation is that weaker associations between bike lane presence and bicycle commuting occur because residents of disadvantaged block groups tend to cycle less than those in advantaged block groups, even when bike lanes are present (quadrant (b) of Figure 1). As previously described, low-income populations and people of color may face disproportionate barriers to cycling that are individual (costs, commute distances), environmental (lack of other supportive built environment features), social (crime, discrimination, social norms), and institutional (inadequate representation, gentrification) in nature (Community Cycling Center 2012; Ferenchak and Marshall 2019; Golub 2016; Howell and Timberlake 2014; Ibsen and Olesen 2018; Kelly et al. 2007; Lubitow 2017; Lubitow, Zinschlag, and Rochester 2016; Lusk et al. 2017; Sarrica et al. 2019). Due to these non-infrastructure barriers, low-income populations and people of color may be less likely than more advantaged populations to use bike lanes when they are available, potentially contributing to increased disparities in cycling.

Similar to Adkins et al. (2017), I found evidence of both explanations but stronger support for the second—that associations between bike lane presence and bicycle commuting are weaker in disadvantaged block groups due to a lack of cycling even when bike lanes are present. This possibility is illustrated in Figure 4, which shows that disadvantaged block groups (i.e., those that fell below the median value of all sociodemographic characteristics considered in this analysis) were nearly three times as likely to fall in quadrant (b) (no bicycle commuting even in the presence of bike lanes) than in quadrant (a) (bicycle commuting even in the absence of bike lanes). The results therefore suggest that weaker associations between bike lane presence and bicycle commuting in disadvantaged places could stem largely from the multifaceted non-infrastructure barriers to cycling documented in past research.

These findings raise important questions about bike lane investment as an equitable tool for cycling promotion. If disadvantaged populations tend to cycle less than their more advantaged counterparts even when bike lanes are available, bike lane investment, in isolation, could lead to counterintuitive outcomes. Indeed, if advantaged populations face fewer non-infrastructure barriers to cycling and are thus more likely to use bike lanes when they are provided, investing in bike lanes could unintentionally widen sociodemographic disparities in cycling and its potential benefits. This possibility points to the value of complementary approaches to cycling promotion that address non-infrastructure barriers, particularly in low-income communities of color.

Several recent studies provide a precedent for this approach, underscoring the value of complementing cycling infrastructure investments with other types of supportive interventions. Some of this work has suggested that investments in infrastructure such as bike lanes and bicycle sharing stations, in isolation, are insufficient to encourage notable shifts to cycling, particularly when these investments are made in the otherwise unsuitable or unsupportive environments that tend to be disproportionately faced in low-income communities of color (Dill et al. 2014; Wang and Lindsey 2019). Other research has focused on the role of perceptions and attitudes, finding these and other forms of social support to be important determinants of cycling—particularly among low-income populations and people of color—that warrant consideration alongside infrastructure investments (Barajas 2019; Biehl et al. 2019; Dill et al. 2014). Pritchard et al. (2019) note that some structural barriers to cycling, such as the socio-spatial distribution of employment opportunities and corresponding commute distances, may be challenging to overcome. However, the findings of a randomized controlled trial conducted by Schneider et al. (2018) in low-income communities of color suggest that other common barriers, such as perceived traffic stress and limited cycling experience, may be effectively addressed through education and encouragement programs. These studies fit within a larger body of research emphasizing the need for complementary individual, social, and spatial interventions to adequately support active transportation (Gotschi et al. 2017). Moreover, they reinforce the findings of the present analysis, suggesting that a more comprehensive definition of “access” to bike lanes—one that incorporates both infrastructure and non-infrastructure barriers to cycling—is needed if cycling is to be equitably promoted among diverse populations, and if the potential benefits of bike lane investment are to be adequately realized in disadvantaged communities.

Strengths and Limitations

This study is among the first to examine how associations between bike lanes and cycling are moderated by sociodemographic advantage. The findings are based on a geographically extensive sample of block groups across twenty-two large U.S. cities, allowing for interpretation beyond the context of a single geographic setting.

Several caveats should be considered when interpreting the results. The data are cross-sectional and thus do not indicate causality, nor do they provide information about changes in cycling or in cycling disparities when bike lanes are introduced. Rather, they apply to differences between places with different sociodemographic and infrastructure characteristics at the same point in time. Further longitudinal work on behavioral responses to bike lane interventions could add depth to the conclusions of this study. Similarly, the results do not indicate which types of barriers (e.g., individual, environmental, social, institutional) are the most influential; qualitative work on the barriers to cycling faced by marginalized groups would be a valuable extension to this cross-sectional, quantitative analysis. Due to the aggregate unit of analysis (i.e., block groups), the findings apply to places rather than to people. Indeed, it is impossible to tell from the data used who cycles, but rather the types of places where commuters cycle. While individual-level analyses are needed, research showing that neighborhood-level disadvantage is an important determinant of travel behavior (Turrell et al. 2010) suggests that this aggregate analysis is a valuable first step in understanding the contextual influences of cycling disparities. In addition, it is important to keep in mind that the results of this analysis apply specifically to bicycle commuting and to the physical presence of bike lanes; different findings may be observed for other trip purposes and infrastructure characteristics (e.g., bike lane quality, trails). The journey-to-work data collected by the U.S. Census Bureau via the ACS are also subject to relatively large margins of error, which could partially explain the attenuated associations between bike lane presence and bicycle commuting observed in this analysis; this limitation attests to the value of conducting similar analyses with alternative sources of cycling data, including those that incorporate non-commuting trip purposes. Finally, as previously noted, the findings may be primarily applicable to the distinction between block groups with versus without bicycle commuting rather than to the level of bicycle commuting among those with any, and may be less prominent in large cities with very high shares of bicycle commuting.

Conclusion: Widening the Gap?

This study provides evidence that associations between bike lane presence and bicycle commuting, while generally positive, tend to be stronger in more advantaged block groups and correspondingly weaker in less advantaged block groups. This finding could reflect the influence of non-infrastructure barriers to cycling among low-income populations and people of color, creating a situation in which advantaged populations are better positioned to make use of cycling infrastructure when it is available. Thus, bike lane investment could have the unintended consequence of widening the gap in cycling—and potentially in related environmental, economic, access, and health outcomes—between places of high versus low sociodemographic advantage if non-infrastructure barriers to cycling are not also addressed.

The implications of these results for planning practice and research are twofold. First, in designing interventions and studying their outcomes, planners and researchers should recognize that the potential impacts of infrastructure such as bike lanes are likely to vary across different population subgroups. This recognition is an important starting point for ensuring that sociodemographic differences are brought to the foreground of both plans and studies, allowing for a more careful consideration of who stands to benefit from infrastructure investments. This lesson, which Biehl et al. (2019) describe as the identification of distinct “subcommunities” within a broad population of interest, can be extended to other areas of travel behavior research and to planning research more broadly. Planners should critically consider research that presents single estimates of an “effect” or relationship of interest, keeping in mind that such relationships may be nuanced—and potentially even counterintuitive—for different population subgroups. Foregrounding sociodemographic characteristics to consider this possibility is an important direction for more equitable, inclusive planning research and practice.

Second, a more comprehensive definition of bike lane “access” should be consistently sought and emphasized in planning research and practice, building upon the past work reviewed in this analysis. This work suggests that the physical availability of bike lanes may not capture all elements of access that are needed to support cycling among diverse populations. Importantly, cycling interventions and studies should be designed around this comprehensive definition, ensuring that bike lane investments are complemented by other strategies that address the non-infrastructure barriers to cycling that low-income populations and people of color disproportionately face. These strategies might include bicycle ownership and maintenance programs; partnerships with groups addressing crime, gentrification, and inequitable policing; broader efforts to promote traffic safety; education and outreach programs designed to influence attitudes, preferences, and cycling skills; and inclusive, grassroots planning and advocacy processes that seek to fully engage traditionally marginalized and underrepresented populations. Through these multifaceted approaches, bike lane investments could have a greater potential to address cycling disparities and thus to advance social equity goals.

Footnotes

Acknowledgements

I would like to sincerely thank Dr. Daniel A. Rodriguez and Dr. Penny Gordon-Larsen for their mentorship and guidance during the data collection process.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Lindsay M. Braun

Author Biography

Lindsay M. Braun is an assistant professor in the Department of Urban and Regional Planning at the University of Illinois at Urbana-Champaign. Her research addresses the connections between the built environment, active transportation, public health, and social equity.

References

Adams

Jean.

2010. “Prevalence and Socio-Demographic Correlates of ‘Active Transport’ in the UK: Analysis of the UK Time Use Survey 2005.” Preventive Medicine 50:199–203. doi:10.1016/j.ypmed.2010.01.006.

Adkins

Arlie

Makarewicz

Carrie

Scanze

Michele

Ingram

Maia

Luhr

Gretchen

. 2017. “Contextualizing Walkability: Do Relationships between Built Environments and Walking Vary by Socioeconomic Context?” Journal of the American Planning Association 83:296–314. doi:10.1080/01944363.2017.1322527.

Alliance for Biking & Walking. 2016. “Bicycling & Walking in the United States.” https://bikeleague.org/sites/default/files/2016BenchmarkingReport_web.pdf.

August

Kristin J.

Sorkin

Dara H.

2011. “Racial/Ethnic Disparities in Exercise and Dietary Behaviors of Middle-Aged and Older Adults.” Journal of General Internal Medicine 26:245–250. doi:10.1007/s11606-010-1514-7.

Babagoli

Masih A.

Kaufman

Tanya K.

Noyes

Philip

Sheffield

Perry E.

2019. “Exploring the Health and Spatial Equity Implications of the New York City Bike Share System.” Journal of Transport & Health 13:200–209. doi:10.1016/j.jth.2019.04.003.

Barajas

Jesus M.

2019. “Perceptions, People, and Places: Influences on Cycling for Latino Immigrants and Implications for Equity.” Journal of Planning Education and Research. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3436544.

Biehl

Alec

Chen

Ying

Sanabria-Veaz

Karla

Uttal

David

Stathopoulos

Amanda

. 2019. “Where Does Active Travel Fit within Local Community Narratives of Mobility Space and Place?” Transportation Research Part A: Policy and Practice 123:269–287. doi:10.1016/j.tra.2018.10.023.

Blumenberg

Evelyn

Pierce

Gregory

. 2012. “Automobile Ownership and Travel by the Poor: Evidence from the 2009 National Household Travel Survey.” Transportation Research Record: Journal of the Transportation Research Board 2320:28–36. doi:10.3141/2320-04.

Braun

Lindsay M.

Rodriguez

Daniel A.

Gordon-Larsen

Penny

. 2019. “Social (In)Equity in Access to Cycling Infrastructure: Cross-sectional Associations between Bike Lanes and Area-level Sociodemographic Characteristics in 22 Large U.S. Cities.” Journal of Transport Geography 80:102544. doi:10.1016/j.jtrangeo.2019.102544.

10.

Buehler

Ralph

Pucher

John

. 2012. “Cycling to Work in 90 Large American Cities: New Evidence on the Role of Bike Paths and Lanes.” Transportation 39:409–432. doi:10.1007/s11116-011-9355-8.

11.

Cao

Xinyu (Jason)

Mokhtarian

Patricia L.

Handy

Susan L.

2009. “Examining the Impacts of Residential Self-Selection on Travel Behaviour: A Focus on Empirical Findings.” Transport Reviews 29:359–395. doi:10.1080/01441640802539195.

12.

Chapman

Ralph

Keall

Michael

Howden-Chapman

Philippa

Grams

Mark

Witten

Karen

Randal

Edward

Woodward

Alistair

. 2018. “A Cost Benefit Analysis of an Active Travel Intervention with Health and Carbon Emission Reduction Benefits.” International Journal of Environmental Research and Public Health 15:962. doi:10.3390/ijerph15050962.

13.

Christine

Paul J.

Auchincloss

Amy H.

Bertoni

Alain G.

Carnethon

Mercedes R.

Sánchez

Brisa N.

Moore

Kari

Adar

Sara D.

, et al. 2015. “Longitudinal Associations between Neighborhood Physical and Social Environments and Incident Type 2 Diabetes Mellitus: The Multi-Ethnic Study of Atherosclerosis (MESA).” JAMA Internal Medicine 175:1311. doi:10.1001/jamainternmed.2015.2691.

14.

Clark

Julie

Curl

Angela

. 2016. “Bicycle and Car Share Schemes as Inclusive Modes of Travel? A Socio-Spatial Analysis in Glasgow, UK.” Social Inclusion 4:83–99. doi:10.17645/si.v4i3.510.

15.

Community Cycling Center. 2012. “Understanding Barriers to Bicycling Project: Final Report.” https://communitycyclingcenter.org/wp-content/uploads/2012/07/Understanding-Barriers-Final-Report.pdf.

16.

Dill

Jennifer

Carr

Theresa

. 2003. “Bicycle Commuting and Facilities in Major U.S. Cities: If You Build Them, Commuters Will Use Them.” Transportation Research Record: Journal of the Transportation Research Board 1828:116–123. doi:10.3141/1828-14.

17.

Dill

Jennifer

Mohr

Cynthia

Liang

. 2014. “How Can Psychological Theory Help Cities Increase Walking and Bicycling?” Journal of the American Planning Association 80:36–51. doi:10.1080/01944363.2014.934651.

18.

Ferenchak

Nicholas N.

Marshall

Wesley E.

2019. “Equity Analysis of Proactively- vs. Reactively-Identified Traffic Safety Issues.” Transportation Research Record 2673:596–606. doi:10.1177/0361198119841296.

19.

Flanagan

Elizabeth

Lachapelle

Ugu

El-Geneidy

Ahmed

. 2016. “Riding Tandem: Does Cycling Infrastructure Investment Mirror Gentrification and Privilege in Portland, OR and Chicago, IL?” Research in Transportation Economics 60:14–24. doi:10.1016/j.retrec.2016.07.027.

20.

Gavin

Kristin

Bennett

Aronté

Auchincloss

Amy H.

Katenta

Anna

. 2016. “A Brief Study Exploring Social Equity within Bicycle Share Programs.” Transportation Letters: The International Journal of Transportation Research 8:177–180. doi:10.1080/19427867.2015.1126065.

21.

Golub

Aaron

2016. Is the Right to Bicycle a Civil Right? Synergies and Tensions between the Transportation Justice Movement and Planning for Bicycling. In Bicycle Justice and Urban Transformation: Biking for All? edited by Golub

Hoffmann

Lugo

Sandoval

, 20–31. New York: Routledge.

22.

Goodman

Anna

Panter

Jenna

Sharp

Stephen J.

Ogilvie

David

. 2013. “Effectiveness and Equity Impacts of Town-Wide Cycling Initiatives in England: A Longitudinal, Controlled Natural Experimental Study.” Social Science & Medicine 97:228–237. doi:10.1016/j.socscimed.2013.08.030.

23.

Goodman

Anna

Sahlqvist

Shannon

Ogilvie

David

. 2013. “Who Uses New Walking and Cycling Infrastructure and How? Longitudinal Results from the UK iConnect Study.” Preventive Medicine 57:518–524. doi:10.1016/j.ypmed.2013.07.007.

24.

Gotschi

Thomas

de Nazelle

Audrey

Brand

Christian

Gerike

Regine

. 2017. “Towards a Comprehensive Conceptual Framework of Active Travel Behavior: A Review and Synthesis of Published Frameworks.” Current Environmental Health Reports 4:286–295. doi:10.1007/s40572-017-0149-9.

25.

Grisé

Emily

El-Geneidy

Ahmed

. 2018. “If We Build It, Who Will Benefit? A Multi-Criteria Approach for the Prioritization of New Bicycle Lanes in Quebec City, Canada.” Journal of Transport and Land Use 11:1. doi:10.5198/jtlu.2018.1115.

26.

Heinen

Eva

Kamruzzaman

Md.

Turrell

Gavin

. 2018. “The Public Bicycle-Sharing Scheme in Brisbane, Australia: Evaluating the Influence of Its Introduction on Changes in Time Spent Cycling amongst a Middle- and Older-Age Population.” Journal of Transport & Health 10:56–73. doi:10.1016/j.jth.2018.07.003.

27.

Hirsch

Jana A.

Green

Geoffrey F.

Peterson

Marc

Rodriguez

Daniel A.

Gordon-Larsen

Penny

. 2017. “Neighborhood Sociodemographics and Change in Built Infrastructure.” Journal of Urbanism: International Research on Placemaking and Urban Sustainability 10:181–197. https://doi.org/10.1080/17549175.2016.1212914.

28.

Hosford

Kate

Winters

Meghan

. 2018. “Who Are Public Bicycle Share Programs Serving? An Evaluation of the Equity of Spatial Access to Bicycle Share Service Areas in Canadian Cities.” Transportation Research Record 2672:42–50. doi:10.1177/0361198118783107.

29.

Howell

Aaron J.

Timberlake

Jeffrey M.

2014. “Racial and Ethnic Trends in the Suburbanization of Poverty in U.S. Metropolitan Areas, 1980–2010.” Journal of Urban Affairs 36:79–98. doi:10.1111/juaf.12030.

30.

Ibsen

Mikkel E.

Olesen

Kristian

. 2018. “Bicycle Urbanism as a Competitive Advantage in the Neoliberal Age: The Case of Bicycle Promotion in Portland.” International Planning Studies 23:210–224. doi:10.1080/13563475.2017.1402675.

31.

Jurakić

Danijel

Pedišić

Želko

Andrijašević

Mirna

. 2009. “Physical Activity of Croatian Population: Cross-sectional Study Using International Physical Activity Questionnaire.” Croatian Medical Journal 50:165–173. doi:10.3325/cmj.2009.50.165.

32.

Keall

Michael D.

Shaw

Caroline

Chapman

Ralph

Howden-Chapman

Philippa

. 2018. “Reductions in Carbon Dioxide Emissions from an Intervention to Promote Cycling and Walking: A Case Study from New Zealand.” Transportation Research Part D: Transport and Environment 65:687–696. doi:10.1016/j.trd.2018.10.004.

33.

Kelly

Cheryl M.

Schootman

Mario

Baker

Elizabeth A.

Barnidge

Ellen K.

Lemes

Amanda

. 2007. “The Association of Sidewalk Walkability and Physical Disorder with Area-Level Race and Poverty.” Journal of Epidemiology & Community Health 61:978–983. doi:10.1136/jech.2006.054775.

34.

Kent

Margaret

Karner

Alex

. 2019. “Prioritizing Low-Stress and Equitable Bicycle Networks Using Neighborhood-Based Accessibility Measures.” International Journal of Sustainable Transportation 13:100–110. doi:10.1080/15568318.2018.1443177.

35.

League of American Bicyclists. 2013. “The New Majority: Pedaling Towards Equity.” https://content.sierraclub.org/press-releases/2013/05/new-majority-pedals-towards-equity

36.

Lee

Jieun

Vojnovic

Igor

Grady

Sue C.

2018. “The ‘Transportation Disadvantaged’: Urban Form, Gender and Automobile versus Non-automobile Travel in the Detroit Region.” Urban Studies 55:2470–2498. doi:10.1177/0042098017730521.

37.

Zhengyan

Konisky

David M.

Zirogiannis

Nikolaos

. 2019. “Racial, Ethnic, and Income Disparities in Air Pollution: A Study of Excess Emissions in Texas.” PLoS ONE 14:e0220696. doi:10.1371/journal.pone.0220696.

38.

Loustaunau

Michel G.

Chakraborty

Jayajit

. 2019. “Vehicular Air Pollution in Houston, Texas: An Intra-Categorical Analysis of Environmental Injustice.” International Journal of Environmental Research and Public Health 16:2968. https://doi.org/10.3390/ijerph16162968

39.

Lubitow

Amy

2017. “Narratives of Marginalized Cyclists: Understanding Obstacles to Utilitarian Cycling among Women and Minorities in Portland, Oregon.” No. NITC-SS-994. Portland: Transportation Research and Education Center, Portland State University.

40.

Lubitow

Amy

Zinschlag

Bryan

Rochester

Nathan

. 2016. “Plans for Pavement or for People? The Politics of Bike Lanes on the ‘Paseo Boricua’ in Chicago, Illinois.” Urban Studies 53:2637–2653. doi:10.1177/0042098015592823.

41.

Lusk

Anne C.

Anastasio

Albert

Shaffer

Nicholas

Juan

Yanping

. 2017. “Biking Practices and Preferences in a Lower Income, Primarily Minority Neighborhood: Learning What Residents Want.” Preventive Medicine Reports 7:232–238. doi:10.1016/j.pmedr.2017.01.006.

42.

Liang

Runing

. 2019. “Does Daily Commuting Behavior Matter to Employee Productivity?” Journal of Transport Geography 76:130–141. doi:10.1016/j.jtrangeo.2019.03.008.

43.

Mäkinen

Borodulin

Laatikainen

Fogelholm

Prättälä

2008. “Twenty-Five Year Socioeconomic Trends in Leisure-Time and Commuting Physical Activity among Employed Finns: Socioeconomic Trends in Physical Activity.” Scandinavian Journal of Medicine & Science in Sports 19:188–197. doi:10.1111/j.1600-0838.2007.00739.x.

44.

Marshall

Wesley E.

Ferenchak

Nicholas N.

2019. “Why Cities with High Bicycling Rates Are Safer for All Road Users.” Journal of Transport & Health 13:285–301. doi:10.1016/j.jth.2019.03.004.

45.

Murphy

Brendan

Owen

Andrew

. 2019. “Implementing Low-Stress Bicycle Routing in National Accessibility Evaluation.” Transportation Research Record 2673:240–249. doi:10.1177/0361198119837179.

46.

Neves

Andre

Brand

Christian

. 2019. “Assessing the Potential for Carbon Emissions Savings from Replacing Short Car Trips with Walking and Cycling Using a Mixed GPS-Travel Diary Approach.” Transportation Research Part A: Policy and Practice 123:130–146. doi:10.1016/j.tra.2018.08.022.

47.

Park

Juneyoung

Abdel-Aty

Mohamed

Lee

Jaeyoung

Lee

Chris

. 2015. “Developing Crash Modification Functions to Assess Safety Effects of Adding Bike Lanes for Urban Arterials with Different Roadway and Socio-Economic Characteristics.” Accident Analysis & Prevention 74:179–191. doi:10.1016/j.aap.2014.10.024.

48.

Parker

Kathryn M.

Gustat

Jeanette

Rice

Janet C.

2011. “Installation of Bicycle Lanes and Increased Ridership in an Urban, Mixed-Income Setting in New Orleans, Louisiana.” Journal of Physical Activity & Health 8:S98–S102.

49.

Poirier

Joseph A.

2018. “Bicycle Lanes and Business Success: A San Francisco Examination.” Transportation Research Record 2672:47–57. doi:10.1177/0361198118792321.

50.

Pritchard

John P.

Tomasiello

Diego B.

Giannotti

Mariana

Geurs

Karst

. 2019. “Potential Impacts of Bike-and-Ride on Job Accessibility and Spatial Equity in Sao Paulo, Brazil.” Transportation Research Part A: Policy and Practice 121:386–400. doi:10.1016/j.tra.2019.01.022.

51.

Pucher

John

Buehler

Ralph

Bassett

David R.

Dannenberg

Andrew L.

2010. “Walking and Cycling to Health: A Comparative Analysis of City, State, and International Data.” American Journal of Public Health 100:1986–1992. doi:10.2105/AJPH.2009.189324.

52.

Rachele

Jerome N.

Kavanagh

Anne M.

Badland

Hannah

Giles-Corti

Billie

Washington

Simon

Turrell

Gavin

. 2015. “Associations between Individual Socioeconomic Position, Neighbourhood Disadvantage and Transport Mode: Baseline Results from the HABITAT Multilevel Study.” Journal of Epidemiology and Community Health 69:1217–1223. doi:10.1136/jech-2015-205620.

53.

Sarrica

Mauro

Alecci

Eleonora

Passafaro

Paola

Rimano

Alessandra

Mazzara

Bruno M.

2019. “The Social Representations of Cycling Practices: An Analysis of Symbolic, Emotional, Material and Bodily Components, and Their Implication for Policies.” Transportation Research Part F: Traffic Psychology and Behaviour 64:119–132. doi:10.1016/j.trf.2019.04.019.

54.

Schneider

Robert J.

Kusch

Jennifer

Dressel

Anne

Bernstein

Rebecca

. 2018. “Can a Twelve-Week Intervention Reduce Barriers to Bicycling among Overweight Adults in Low-Income Latino and Black Communities?” Transportation Research Part F: Traffic Psychology and Behaviour 56:99–112. doi:10.1016/j.trf.2018.03.023.

55.

Singleton

Patrick A.

2019. “Walking (and Cycling) to Well-Being: Modal and Other Determinants of Subjective Well-Being during the Commute.” Travel Behaviour & Society 16:249–261. doi:10.1016/j.tbs.2018.02.005.

56.

Smith

Melody

Hosking

Jamie

Woodward

Alistair

Witten

Karen

MacMillan

Alexandra

Field

Adrian

Baas

Peter

Mackie

Hamish

. 2017. “Systematic Literature Review of Built Environment Effects on Physical Activity and Active Transport—An Update and New Findings on Health Equity.” International Journal of Behavioral Nutrition and Physical Activity 14:158. doi:10.1186/s12966-017-0613-9.

57.

Tucker

Bronwen

Manaugh

Kevin

. 2018. “Bicycle Equity in Brazil: Access to Safe Cycling Routes across Neighborhoods in Rio de Janeiro and Curitiba.” International Journal of Sustainable Transportation 12:29–38. doi:10.1080/15568318.2017.1324585.

58.

Turrell

Gavin

Haynes

Michele

Burton

Nicola W.

Giles-Corti

Billie

Oldenburg

Brian

Wilson

Lee-Ann

Giskes

Katrina

Brown

Wendy J.

2010. “Neighborhood Disadvantage and Physical Activity: Baseline Results from the HABITAT Multilevel Longitudinal Study.” Annals of Epidemiology 20:171–181. doi:10.1016/j.annepidem.2009.11.004.

59.

U.S. Census Bureau. 2016. “American Community Survey: 2012–2016 Five-Year Estimates.” https://data.census.gov/cedsci/

60.

Van Dyck

Delfien

Cardon

Greet

Deforche

Benedicte

Sallis

James F.

Owen

Nevil

De Bourdeaudhuij

Ilse

. 2010. “Neighborhood SES and Walkability Are Related to Physical Activity Behavior in Belgian Adults.” Preventive Medicine 50, S74–S79. doi:10.1016/j.ypmed.2009.07.027.

61.

Wang

Jueyu

Lindsey

Greg

. 2017. “Equity of Bikeway Distribution in Minneapolis, Minnesota.” Transportation Research Record: Journal of the Transportation Research Board 2605:18–31. doi:10.3141/2605-02.

62.

Wang

Jueyu

Lindsey

Greg

. 2019. “Do New Bike Share Stations Increase Member Use: A Quasi-Experimental Study.” Transportation Research Part A: Policy and Practice 121:1–11. doi:10.1016/j.tra.2019.01.004.

63.

Wanner

Miriam

Götschi

Thomas

Martin-Diener

Eva

Kahlmeier

Sonja

Martin

Brian W.

2012. “Active Transport, Physical Activity, and Body Weight in Adults: A Systematic Review.” American Journal of Preventive Medicine 42:493–502. doi:10.1016/j.amepre.2012.01.030.

64.

Winters

Meghan

Fischer

Jaimy

Nelson

Trisalyn

Fuller

Daniel

Whitehurst

David G. T.

2018. “Equity in Spatial Access to Bicycling Infrastructure in Mid-Sized Canadian Cities.” Transportation Research Record 2672:24–32. doi:10.1177/0361198118791630.

65.

Zhang

Yongpin

Zhifu

. 2018. “Environmental Benefits of Bike Sharing: A Big Data-Based Analysis.” Applied Energy 220:296–301. doi:10.1016/j.apenergy.2018.03.101.