UBC Faculty Research and Publications

Bike Score®: Associations between urban bikeability and cycling behavior in 24 cities Winters, Meghan; Teschke, Kay; Brauer, Michael; Fuller, Daniel Feb 11, 2016

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


52383-12966_2016_Article_339.pdf [ 916.39kB ]
JSON: 52383-1.0308593.json
JSON-LD: 52383-1.0308593-ld.json
RDF/XML (Pretty): 52383-1.0308593-rdf.xml
RDF/JSON: 52383-1.0308593-rdf.json
Turtle: 52383-1.0308593-turtle.txt
N-Triples: 52383-1.0308593-rdf-ntriples.txt
Original Record: 52383-1.0308593-source.json
Full Text

Full Text

RESEARCH Open AccessBike Score®: Associations between urbanbikeability and cycling behavior in 24 citiesMeghan Winters1*, Kay Teschke2, Michael Brauer2 and Daniel Fuller3AbstractBackground: There is growing interest in designing cities that support not only walking, but also cycling. Bike Score®is a metric capturing environmental characteristics associated with cycling that is now available for over 160 US andCanadian cities. Our aim was to determine if Bike Score was associated with between and within-city variability incycling behavior.Methods: We used linear regression to model associations between Bike Score and journey to work cycling mode share(US: American Community Survey, 2013 or 2012 5-year estimates; Canada: 2011 National Household Survey) for 5664census tracts in 24 US and Canadian cities.Results: At the city level, the correlation between mean Bike Score and mean journey to work cycling mode share wasmoderate (r = 0.52). At the census tract level, the correlation was 0.35; a ten-unit increase in Bike Score was associatedwith a 0.5 % (95 % CI: 0.5 to 0.6) increase in the proportion of population cycling to work, a meaningful differencegiven the low modal shares (mean = 1.9 %) in many North American cities. Census tracts with the highest Bike Scores(>90 to 100) had mode shares 4.0 % higher (β = 4.0, 95 % CI: 2.9 to 5.0) than the lowest Bike Score areas (0–25). Cityspecific analyses indicated between-city variability in associations, with regression estimates between Bike Score andmode share ranging from 0.2 to 3.5 %.Conclusions: The Bike Score metric was associated bicycle mode share between and within cities, suggesting its utilityfor planning bicycle infrastructure.Keywords: Active transport, Cycling, Built environment, Multi-level modeling, Bike ScoreBackgroundConsidering both risks and benefits, active travel carriesa net benefit on all-cause mortality [1–3]. In ecologicalstudies, areas with higher levels of active travel are asso-ciated with lower traffic fatality risk, [4] higher levels ofphysical activity, [5] and lower rates of obesity and dia-betes [5]. Many studies have documented links betweenneighborhood design and active transportation [6, 7]. Inearly research project-specific measures of walkabilitywere developed, limiting comparability between studies[8, 9]. More recently, researchers have used Walk Score®(www.walkscore.com), [10–13] a web-based tool thatscores neighborhood walkability based on proximity tovarious destinations. The popularity of the Walk Scoremetric is likely due in part to its extensive coverage ofNorth America, providing consistent methodology acrosssettings at relatively low cost. Walk Score is correlated withother measures of access to destinations and walkability[14], and with walking for transportation [11, 15, 16] at alevel comparable with other walkability measures [15].In contrast with walking, cycling has received lessattention in the neighborhood design literature. Cyclingis currently underused as a transportation mode inNorth America, although there may be increased uptakewith improvements in infrastructure amid health, environ-mental, and mobility considerations [17, 18]. There aresimilarities in features that constitute a walkable or abikeable neighborhood, however, certain environmentalcharacteristics such as cycling-specific infrastructure andtopography are additional factors relevant for cycling[19–22]. In order to promote a shift to active transpor-tation for trips of moderate distance, beyond distances* Correspondence: mwinters@sfu.ca1Faculty of Health Sciences, Simon Fraser University, Blusson Hall Rm 11522,8888 University Drive, Burnaby, BC V5A 1S6, CanadaFull list of author information is available at the end of the article© 2016 Winters et al. Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, andreproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link tothe Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.Winters et al. International Journal of Behavioral Nutritionand Physical Activity  (2016) 13:18 DOI 10.1186/s12966-016-0339-0suitable to walking, metrics specific to cycling are use-ful for guiding neighborhood design.In 2012 we partnered with Walk Score to incorporatefindings from our empirical research on cycling and urbanform, [19] which led to their development of “Bike Score”in North American cities. Bike Score is based on environ-mental characteristics consistently associated with cycling:density and quality of cycling infrastructure, topography,desirable amenities and road connectivity. As of 2015 BikeScore was available for over 160 US and Canadian cities.City- and neighborhood- rankings have been publicizedby Walk Score, [23] but its correlation with cycling be-havior has not been explored. We assess the extentwhich Bike Score predicts cycling behavior, both betweenand within cities, through an analysis of 5664 census tractsacross 24 Canadian and US cities.MethodsIn this ecological analysis, cities and census tracts arethe units of analysis. Census tracts in both the US andCanada represent populations of 2000–8000 and approxi-mate neighborhoods.Bike Score dataWe obtained Bike Score data shapefiles (2012) for nineCanadian cities and 15 US cities directly from WalkScore (now RedFin Real Estate). We were provided withpoint files (100 m grid) with attributes for Bike Score andeach of its components for each city that was included inthe original Bike Score launch. The methodology is here:https://www.walkscore.com/bike-score-methodology.shtml.In brief, Bike Score ranges from 0 to 100, and is comprisedof 3 environmental components: a Bike Lane Score, a HillScore, and a Destinations and Connectivity Score, eachranging from 0 to 100 where 100 describes the most bike-able (more bicycle facilities, flat topography, and more des-tinations and connectivity, respectively). The Bike LaneScore is derived from cycling infrastructure data providedby municipal governments. It captures painted bicyclelanes, off-street trails, cycle tracks, and residential bike-ways; it does not include sharrows (shared-lane markings),or other cycling initiatives such as bicycle parking or bikeshare programs. The score weights separated facilitiestwice as much as on-street facilities, and uses a distancedecay function to favour proximity. The Hill Score isbased on the steepest grade in a 200 m radius area. TheConnectivity Score is equivalent to Street Smart WalkScore®. For US cities, the public version of Bike Score alsoincludes a fourth “social” component: the proportion ofthe population that cycles to work. Given that our aimwas to assess how Bike Score predicts cycling behavior, weused the 3 component version of Bike Score for bothCanadian and US cities. The summary metric, Bike Score,is a weighted sum of the Bike Lane Score (50 %), HillScore (25 %), and Destinations and Connectivity Score(25 %). As indicated above, the authors (MW, KT, MB)contributed to the development of Bike Score.Spatial summary of Bike Score dataFor city level analyses, we used the mean city-wide BikeScore as provided by Walk Score. For within-city ana-lyses, we used ArcGIS 10.2 to calculate a mean BikeScore for each census tract from Bike Score shapefiles.In brief, we imported the Bike Score point data for eachcity into ArcGIS and merged it into one file per country.We obtained census geography files for Canadian (2011)and US (2010) census tracts, and developed a Model-Builder toolbox to summarize Bike Score and compo-nent score values for each census tract. This processinvolved (1) attributing each Bike Score point to theappropriate census tract and calculating the average BikeScore for each census tract and (2) calculating the cover-age of Bike Score data for each census tract (area withBike Score data/total area of census tract; using Mollweideprojection). We excluded census tracts where Bike Scoredata coverage was less than 80 % by area.Cycling mode share dataWe sought to use the most accurate, comparable andup-to-date sources of cycling data. For Canada, we ex-tracted journey to work mode share data from the 2011National Household Survey, [24] available at both citylevel and census tract level. For the US, we drew datafrom American Community Survey, [25] using the 20131-year estimates for city level analyses, and 5-year esti-mates (2012) for census tract analyses. While censusjourney to work data is the best available, it must benoted that it represents only cycling trips for work pur-poses, it does not capture multi-modal trips, and thespatial information is linked to the residential location(not route or destination).Statistical analysisAll statistical analyses were conducted in R (Additionalfile 1 includes R code and output) [26]. In the city levelanalysis we report correlations using both Pearson r andand Spearman ρ, and use linear regression to examinethe relationship between Bike Score and cycling modeshare. Within each city there was substantial variabilityin both Bike Score and cycling mode share, and thus weconducted census tract level analyses to understand therelationship at the neighborhood-level. We calculateddescriptive statistics across all census tracts and stratifiedby city.We used linear regression to analyse the associationbetween Bike Score and cycling mode share, with censustract as the unit of analysis. We calculated unadjustedassociations between Bike Score, or its components, andWinters et al. International Journal of Behavioral Nutrition and Physical Activity  (2016) 13:18 Page 2 of 10cycling mode share. Given the wide range of Bike Scoreand its components (0–100), we present coefficients forthe effect of a 10-unit change on cycling mode share,and also for Bike Score in five categories (0 to 25, >25 to50, >50 to 75, >75 to 90, and >90 to 100), as has beenused in previous work [11, 13]. We report adjusted asso-ciations from fixed effect regression models, with adummy variable for city. The fixed effect approach ac-counts for the clustered nature of census tracts withincities, and provides an average estimate of the associ-ation across all census tracts in all cities. We also rancity-specific models to provide city-specific estimates ofthe association, which may be of value to future studiesfocused on a particular location. We first modeled theoutcome of cycling mode share with Bike Score and cityas independent variables (model 1). In model 2, the BikeScore components (Bike Lane Score, Hill Score, Destina-tions and Connectivity Score) and city were independentvariables. In model 3, we included the categorized BikeScore variable and city as independent variables. Weused New York City as the reference city for all models,given that it has the largest number of census tracts andthe lowest mean cycling mode share across census tracts.Finally, we also performed a multilevel model of the associ-ation between Bike Score and cycling modes share, usingrandom slope models which allow for different magnitudesin the association across cities. We ran null models, con-sidering city, or city and country, as random effects, andrelated models with Bike Score as a fixed effect.ResultsCity level analysisAcross the 24 cities, the city-wide mean Bike Scores andmode shares ranged from 20 to 73 and 0.3 to 12.3 %, re-spectively (Additional file 2: Table S1). In cities withhigher mean Bike Score, more people cycled to work(Fig. 1). At the city level, the correlation between meanBike Score and mean journey to work cycling modeshare was moderate (Pearson’s r = 0.52 and Spearman’sρ = 0.56). The association between Bike Score and cyc-ling mode share was positive and significant (β = 1.5 %for a 10-unit change in Bike Score, 95 % CI 0.4 to 2.6 %)with Bike Score explaining 27 % of the variation incycling mode share between cities.Fig. 1 Scatter plot and estimated regression line for City-wide cycling mode sharea and City-wide Average Bike Score®a.b Linear regressionestimated association between Bike Score and cycling mode share. We show the full possible range of Bike Score, however, city-wide averagesfor the study cities do not cover this full range. aCity-wide journey to work cycling mode share (% commutes by bike for workers aged 16 yearsand older) -American Community Survey, 2013 1-year estimates, U.S. Census Bureau; CA data is 2011 National Household Survey, Journey to workBicycle Mode Share for population aged 15 years and older with a usual place of work. bCity-wide Bike Score and components provided directlyfrom the company (now Redfin Real Estate), using 3 components (Bike Lane Score, Hill Score, Destinations and Connectivity Score), May 2012 releaseWinters et al. International Journal of Behavioral Nutrition and Physical Activity  (2016) 13:18 Page 3 of 10Census tract level analysesThe analysis dataset included 5664 census tracts nestedin 24 cities, with a range of 15 (Moncton, New Brunswick)to 2164 (New York City) tracts per city. Across all censustracts, the mean Bike Score was 67.0, with a range from5.9 to 100. Cycling mode share had a mean of 1.9 % andrange of 0.0 to 34.0 %. Table 1 provides the descriptive sta-tistics for cycling mode share and Bike Score (overall andcomponents) stratified by city. Across all census tracts, thecorrelation between Bike Score and cycling mode share wasmoderate (Pearson’s r = 0.35 and Spearman’s ρ = 0.40).In unadjusted analyses, Bike Score and each of the BikeScore components - Bike Lane Score, Hill Score, Destina-tions and Connectivity Score - were significantly associ-ated with cycling mode share (Table 2). The unadjustedassociation for Hill Score (higher scores mean flatter top-ography, which would be hypothesized to promote cyc-ling) was negative - opposite to expectation.In multiple linear regression adjusting for city (Table 2,Model 1), a ten-unit increase in the Bike Score of a cen-sus tract was associated with a 0.5 % increase in theproportion of population cycling to work (β = 0.5, 95 %CI: 0.5 to 0.6). Adjusting for city improved the modelfit (Model 1 AIC = 11035, versus unadjusted modelAIC = 12627; R2 = 0.35 versus R2 = 0.12). In the modelincluding all Bike Score components (Table 2, Model 2)each of the components had significant associationsand the Hill Score was positively associated with cyc-ling mode share, as expected. The change in the direc-tion between unadjusted and adjusted models suggestsTable 1 Descriptive characteristics for 5664 census tracts in 24 study citiesCity, State/Province Number ofCensus TractsCycling Mode Sharea (%)(mean, (SD))Bike Score®b(mean, (SD))Bike Lane Score(mean, (SD))Hill Score(mean, (SD))Destinations andConnectivity Score(mean, (SD))Ann Arbor, Michigan 33 3.6 (2.7) 76.4 (13.9) 79.9 (18.2) 91.2 (9.0) 55.6 (26.9)Austin, Texas 164 1.8 (2.8) 48.3 (17.4) 26.3 (24.5) 85.8 (17.5) 55.3 (27.2)Boston, Massachusetts 179 1.6 (2.4) 73.4 (19.1) 57.7 (32.0) 88.6 (14.5) 90.4 (17.2)Calgary, Alberta 221 1.2 (1.8) 74.4 (13.0) 84.0 (19.7) 88.3 (13.3) 42.0 (26.0)Chicago, Illinois 768 1.2 (2.0) 60.5 (13.6) 25.9 (24.6) 100.0 (0.4) 90.9 (15.0)Eugene, Oregon 31 10.6 (7.2) 77.9 (18.4) 83.4 (18.4) 82.2 (29.5) 63.6 (24.6)Fort Collins, Colorado 33 7.8 (6.0) 83.6 (10.7) 93.4 (10.5) 97.8 (5.9) 50.7 (23.8)Halifax, Nova Scotia 25 3.9 (3.9) 67.4 (14.6) 60.9 (22.2) 71.5 (14.4) 76.9 (22.0)Madison, Wisconsin 53 5.9 (5.0) 67.4 (19.8) 58.5 (27.5) 91.2 (8.3) 62.0 (26.6)Minneapolis, Minnesota 115 3.9 (3.3) 77.6 (15.2) 65.8 (27.0) 96.4 (6.3) 82.8 (18.4)Moncton, New Brunswick 15 0.4 (0.8) 49.3 (15.3) 29.1 (25.6) 94.2 (3.8) 45.5 (30.9)Montréal, Québec 320 4.8 (4.6) 78.8 (17.7) 64.4 (33.3) 97.8 (9.3) 89.2 (21.9)New York, New York 2164 0.7 (1.4) 64.8 (18.3) 36.4 (35.7) 95.4 (11.4) 91.6 (19.2)Portland, Oregon 137 6.3 (5.6) 69.5 (20.3) 58.7 (25.9) 80.5 (26.7) 80.8 (23.1)San Francisco, California 196 3.1 (3.4) 77.8 (17.3) 84.3 (24.4) 53.8 (32.5) 89.8 (21.1)Saskatoon, Saskatchewan 45 2.2 (2.4) 78.7 (13.1) 84.5 (20.0) 98.4 (3.3) 48.2 (27.9)Seattle, Washington 132 3.3 (2.6) 60.9 (19.4) 51.2 (31.8) 65.0 (16.9) 77.1 (25.4)St. John’s, Newfoundlandand Labrador26 0.0 (0.0) 44.8 (16.7) 30.9 (24.9) 62.2 (23.0) 55.9 (33.3)Tempe, Arizona 37 4.1 (4.4) 76.2 (12.4) 70.1 (22.8) 99.2 (3.4) 66.2 (15.3)Toronto, Ontario 544 2.0 (3.8) 66.9 (16.4) 45.7 (30.9) 96.8 (6.4) 80.2 (19.9)Tucson, Arizona 115 2.6 (3.8) 74.4 (19.2) 72.3 (26.6) 98.8 (5.5) 55.0 (26.7)Vancouver, BritishColumbia115 4.1 (3.7) 78.0 (14.8) 71.2 (27.4) 79.3 (15.2) 91.1 (13.6)Victoria, British Columbia 17 11.5 (4.3) 74.3 (17.1) 54.2 (32.4) 95.6 (6.2) 93.8 (5.7)Washington, DC 179 2.5 (3.0) 66.5 (20.9) 52.2 (33.6) 79.5 (19.0) 82.8 (20.0)Total 5664 1.9 (3.3) 67.0 (18.5) 46.5 (35.7) 91.9 (16.6) 83.6 (24.3)aCensus tract level Journey to work Bicycle Mode Share (% commute by bike for workers aged 16 years and older) -American Community Survey, 5-year estimates(2012 5-year estimates), U.S. Census Bureau, 2013 American Community Survey; CA data is 2011 National Household Survey, Census tract level Journey to workBicycle Mode Share for population aged 15 years and older with a usual place of workbBike Score spatial data provided from Walk Score (May 2012 release); analysis here includes 3 components (Bike Lane Score, Hill Score, Destinations and ConnectivityScore); spatial data aggregated to the census tract in ArcGIS 10.2Winters et al. International Journal of Behavioral Nutrition and Physical Activity  (2016) 13:18 Page 4 of 10Table 2 Results of linear regression models estimating associations between Bike Score® and componentsa, and cycling mode sharebModel 1 Model 2 Model 3Unadjusted estimates Bike score + City Term Bike Score Components + Bike Score CategoricalCity Termβ (95 % CI) β (95 % CI) β (95 % CI) β (95 % CI)Intercept −2.6 (-2.9 to -2.3) −4.4 (−5.1 to −3.8) −1.4 (−2.4 to −0.3)Bike Score (10-unit change) 0.6 (0.6 to 0.7) 0.5 (0.5 to 0.6)Destinations/Connectivity Score(10-unit change)0.2 (0.1 to 0.2) 0.4 (0.3 to 0.4)Bike Lane Score (10-unit change) 0.3 (0.3 to 0.3) 0.2 (0.2 to 0.2)Hill Score (10-unit change) −0.1 (−0.2 to −0.1) 0.1 (0.1 to 0.2)Bike Score (categorical)0 to 25 0 (Reference) 0 (Reference)>25 to 50 −0.2 (−1.3 to 1.0) 1.1 (0.1 to 2.2)>50 to 75 0.8 (−0.3 to 2.0) 1.8 (0.8 to 2.9)>75 to 90 2.0 (0.8 to 3.1) 2.6 (1.5 to 3.6)>90 to 100 3.5 (2.3 to 4.7) 4.0 (2.9 to 5.0)CityNew York, New York Reference Reference ReferenceAnn Arbor, Michigan 2.3 (1.3 to 3.2) 3.4 (2.5 to 4.4) 2.5 (1.5 to 3.4)Austin, Texas 2.0 (1.5 to 2.4) 2.8 (2.3 to 3.2) 1.8 (1.3 to 2.2)Boston, Massachusetts 0.5 (0.4 to 0.9) 0.6 (0.2 to 1.0) 0.5 (0.1 to 0.9)Calgary, Alberta 0.0 (−0.4 to 0.4) 1.5 (1.0 to 1.9) 0.3 (−0.1 to 0.6)Chicago, Illinois 0.7 (0.5 to 0.9) 0.7 (0.5 to 0.9) 0.7 (0.5 to 1.0)Eugene, Oregon 9.3 (8.3 to 10.2) 10.3 (9.3 to 11.2) 9.4 (8.4 to 10.3)Fort Collins, Colorado 6.2 (5.3 to 7.1) 7.5 (6.6 to 8.4) 6.3 (5.3 to 7.2)Halifax, Nova Scotia 3.1 (2.0 to 4.2) 3.7 (2.6 to 4.8) 3.3 (2.2 to 4.4)Madison, Wisconsin 5.0 (4.3 to 5.8) 5.9 (5.1 to 6.6) 5.1 (4.4 to 5.8)Minneapolis, Minnesota 2.6 (2.0 to 3.1) 3.0 (2.5 to 3.5) 2.7 (2.2 to 3.2)Moncton, New Brunswick 0.5 (−0.9 to 1.8) 1.5 (0.1 to 2.9) 0.3 (−1.1 to 1.6)Montréal, Québec 3.4 (3.1 to 3.7) 3.7 (3.3 to 4.0) 3.4 (3.1 to 3.8)Portland, Oregon 5.4 (4.9 to 5.9) 5.8 (5.4 to 6.3) 5.4 (5.0 to 5.9)San Francisco, California 1.7 (1.3 to 2.1) 2.1 (1.6 to 2.5) 1.8 (1.4 to 2.2)Saskatoon, Saskatchewan 0.8 (0.0 to 1.6) 2.1 (1.3 to 3.0) 1.0 (0.2 to 1.8)Seattle, Washington 2.8 (2.4 to 3.3) 3.3 (2.8 to 3.8) 2.8 (2.3 to 3.3)St. John’s, Newfoundland and Labrador 0.3 (−0.8 to 1.4) 1.1 (−0.2 to 2.1) 10.6 (−0.2 to 21.4)Tempe, Arizona 2.8 (1.9 to 3.7) 3.6 (2.8 to 4.5) 3.1 (2.2 to 3.9)Toronto, Ontario 1.2 (1.0 to 1.5) 1.5 (1.3 to 1.8) 1.3 (1.0 to 1.5)Tucson, Arizona 1.4 (0.9 to 1.9) 2.4 (1.9 to 3.0) 1.4 (0.9 to 1.9)Vancouver, British-Columbia 2.8 (2.3 to 3.3) 3.0 (2.5 to 3.6) 2.9 (2.4 to 3.4)Victoria, British-Columbia 10.4 (9.1 to 11.6) 10.4 (9.2 to 11.7) 10.3 (9.0 to 11.6)Washington, DC 1.7 (1.3 to 2.2) 2.1 (1.6 to 2.5) 1.7 (1.3 to 2.1)Adj-R2 0.12 (Bike Score, unadjusted) 0.35 0.36 0.34AIC 12627 11035 10870Data for 5664 Census Tracts in 24 Cities. Coefficients represent % mode share. Boldface indicates statistical significance (p < 0.05)aBike Score spatial data provided from Walk Score (May 2012 release); analysis here includes 3 components (Bike Lane Score, Hill Score, Destinations andConnectivity Score); spatial data aggregated to the census tract in ArcGISbCensus tract level Journey to work Bicycle Mode Share (% commute by bike for workers aged 16 years and older) -American Community Survey, 5-year estimates(2012 5-year estimates), U.S. Census Bureau, 2013 American Community Survey; CA data is 2011 National Household Survey, Census tract level Journey to workBicycle Mode Share for population aged 15 years and older with a usual place of workWinters et al. International Journal of Behavioral Nutrition and Physical Activity  (2016) 13:18 Page 5 of 10confounding of the association by one or more of theother variables in the model (Destinations and Con-nectivity Score, Bike Lane Score, city). Across the threecomponents, the Destination and Connectivity Scorehad the largest adjusted estimate (β = 0.4, 95 % CI: 0.3to 0.4), suggesting a slightly stronger relationship withFig. 2 Scatter plot and estimated regression line for cycling mode sharea and Bike Scoreb, for 5664 census tracts in 24 study cities (Panel a), andstratified by city (Panel b) * City-specific regressions with significant slope estimates (see Table 3 for estimate values). a Census tract level Journeyto work Bicycle Mode Share (% commutes by bike for workers aged 16 years and older) -American Community Survey, 5-year estimates (2012 5-year estimates), U.S. Census Bureau, 2013 American Community Survey; CA data is 2011 National Household Survey, Census tract level Journey towork Bicycle Mode Share for population aged 15 years and older with a usual place of work. b Bike Score spatial data provided from Walk Score(May 2012 release); analysis here includes 3 components (Bike Lane Score, Hill Score, Destinations and Connectivity Score); spatial dataaggregated to the census tract in ArcGIS 10.2Winters et al. International Journal of Behavioral Nutrition and Physical Activity  (2016) 13:18 Page 6 of 10journey to work mode share as compared with the BikeLane Score (β = 0.2, 95 % CI: 0.2 to 0.2) or Hill Score(β = 0.1, 95 % CI: 0.1 to 0.2).Across all census tracts there was substantial variabilityin cycling mode share across Bike Score values (Fig. 2a):there were census tracts with 0 % mode share across allBike Score values, and at the maximum Bike Score, therewere census tracts with mode shares ranging from 0 toabove 20 %. Moreover, the distribution of Bike Score wasnot normal (median = 65, 10th percentile = 46; 90th per-centile = 94), and the scatterplot indicated a ceilingeffect (140/5664 census tracts have a Bike Score of 100).Given the potential for non-linear effects we also cate-gorized the Bike Score into five categories, reflectingvisual breaks in the data. This model (Table 2, Model 3)showed consistent increases in cycling mode share acrossthe increasing categories of Bike Score. Compared withcensus tracts with Bike Scores of 0–25, those with BikeScores of >75 to 90 had mode shares 2.6 % higher (β = 2.6,95 % CI: 1.5 to 3.6), and the highest Bike Score censustracts (>90 to 100) had mode shares 4.0 % higher (β = 4.0,95 % CI: 2.9 to 5.0).Figure 2b shows city-specific scatterplots and regres-sion lines, highlighting differences in the underlying dataand the nature of the association between Bike Scoreand cycling mode share. Certain cities have no censustracts with low Bike Scores (e.g., Tempe; Saskatoon,Victoria), while others have none with high Bike Scores(e.g., Moncton, St John’s). The strength of the associationvaries between cities, with cities such as Madison and FortCollins showing steeper gradients. City-specific regressioncoefficients (Table 3) were significant for 18 of 23 cities(the model for St John’s had no fit), ranging from a high ofa 3.5 % change in mode share for 10 unit change in BikeScore (Fort Collins), to a low of 0.2 % (Boston, Calgary,and New York).Table 3 City-specific linear regression results for cycling mode share and Bike Score (5664 census tracts in 24 study cities)Intercept Bike score coefficient Adjusted R2(10-unit change)β (95 % CI)Ann Arbor, Michigan −3.3 0.9 (0.3–1.5) 0.19Austin, Texas −2.5 0.9 (0.7–1.1) 0.30Boston, Massachusetts 0 0.2 (0.0–0.4) 0.02Calgary, Alberta −0.5 0.2 (0.0–0.4) 0.02Chicago, Illinois −1.6 0.4 (0.3–0.5) 0.09Eugene, Oregon −0.5 1.4 (0.1–2.7) 0.10Fort Collins, Colorado −21.8 3.5 (2.0–5.0) 0.38Halifax, Nova Scotia −1 0.7 (−0.4−1.8) 0.04Madison, Wisconsin −7.2 1.9 (1.4–2.4) 0.55Minneapolis, Minnesota 0.9 0.4 (0.0–0.8) 0.02Moncton, New Brunswick −0.8 0.2 (−0.1−0.5) 0.13Montréal, Québec −7.7 1.6 (1.4–1.8) 0.36New York, New York −0.6 0.2 (0.2–0.2) 0.07Portland, Oregon −2.3 1.2 (0.8–1.6) 0.20San Francisco, California −3.2 0.8 (0.5–1.1) 0.15Saskatoon, Saskatchewan 0.5 0.2 (−0.5−0.9) −0.01Seattle, Washington 1.4 0.3 (0.1–0.5) 0.05St. John’s, Newfoundland and Labrador No fita - -Tempe, Arizona −8.3 1.6 (0.6–2.6) 0.19Toronto, Ontario −1.8 0.6 (0.4–0.8) 0.06Tucson, Arizona −5 1.0 (0.7–1.3) 0.24Vancouver, British-Columbia −2.1 0.8 (0.4–1.2) 0.09Victoria, British-Columbia 19.9 −1.1 (−2.2−0.0) 0.15Washington, DC −1.4 0.6 (0.4–0.8) 0.16Bold indicates coefficient is statistically significant at p < 0.05aCycling mode share was 0 % for all census tracts in St. John’sWinters et al. International Journal of Behavioral Nutrition and Physical Activity  (2016) 13:18 Page 7 of 10As we were primarily interested in interpretability andthe magnitude of the association (versus its variance) wehave focused on the fixed effect models. However, wedid also fit multilevel models. These showed similarmagnitude in the association between Bike Score andcycling mode share (Additional file 3: Table S2). Whenwe compared the relative fit of multilevel models usingthe likelihood ratio test between nested models, andfound that random slope models fit better than randomintercept models.DiscussionThe development of Bike Score created the first oppor-tunity to conduct between and within city comparisonsbetween cycling mode share and a widely available metricfor measuring the cycling environment. Across 24 cities,there was a moderate correlation between Bike Score andjourney to work mode share. Prior ecological studies havelooked at the environmental, climate, and social influencesassociated with cycling mode share using cities or healthregions as the unit of analysis [27–29], but these maymask important variability in cycling rates and conditionswithin cities. Given the high resolution Bike Score data wewere able to do a census tract level analysis and found thata 10 unit increase in Bike Score was associated with a0.5 % increase in journey to work mode share, a meaning-ful difference given the low cycling mode shares acrossmuch of North America. This work confirms that BikeScore is associated with cycling mode share, and suggeststhis metric has utility for research and practice to aid withplanning bicycle infrastructure and increasing bicyclemode share.We found a significant association across all cities, how-ever, our within-city analysis identified important nuanceson the association for specific cities. Eighteen cities hadsignificant associations between mode share and BikeScore with estimates varying from modest (0.2 % per 10unit change in Bike Score) to dramatic (3.5 %) in city-specific models. We conclude that Bike Score shows utilityfor national or multicity studies, but closer inspection maybe needed prior to its application for city-specific analysisand planning in certain locations.The development of Bike Score was based on environ-mental factors consistently related to cycling in theliterature [19, 30]. We found that Bike Lane Score, HillScore, and the Destinations and Connectivity Score wereall independently associated with cycling mode share.The Destination and Connectivity Score had a marginallystronger association than the other components in theadjusted model (Model 2). This score is equivalent tothe Walk Score’s ‘Street Smart Walk Score’, and thusthis finding highlights synergies between promotingwalking and cycling. Topography is arguably more of abarrier for cycling than for walking. The study citiesincluded very hilly areas (San Francisco, Seattle) andalso very flat areas (Saskatoon), and the Hill Scoremaintained an independent effect: areas with fewer hillshad higher cycling mode shares in the adjusted model.Of the three components, the Bike Lane Score may bethe most actionable component for local and regionalgovernments in the short term. The importance of cycling-specific facilities has been emphasized for promoting safeand comfortable cycling [21] as well as attracting newcyclists [31]. The data for the Bike Lane Score was pro-vided directly from city governments, with only the follow-ing cycling infrastructure included: bike lanes, residentialstreet bikeways (combined as on-street), and cycle tracksand off-street paths (combined as off-street). The BikeLane Score could be considered an indicator of safety,given that these infrastructure types are safer than majorstreets [32, 33] and are preferred by cyclists, [34] especiallywomen and those new to bicycling [31, 35]. Subsequentwork may evaluate correlations between Bike Score andcycling safety, although obtaining consistent and compar-able safety data across countries, cities and census tractswill be a challenge.The metrics developed by Walk Score (Walk Score,Bike Score, and Transit Score) are intuitive, easy to use,and available online, fulfilling many of the recommenda-tions for making built environment measures relevant topractice [36]. The Bike Score methodology was informedby empirical research, however, the specific algorithmsand decay functions are proprietary. We recommendedthat users think critically about the quality of the under-lying data sources. The Hill Score is based on the widely-used National Elevation Data set from the US GeologicalSurvey [37]. The Bike Lane Score is based on data providedby local governments in 2012 and again in 2015 accordingto standardized criteria. In the future bicycle facility datamay be derived from open sources (e.g., Open Street Maps)although this brings concern around consistency acrosscities. The Destination and Connectivity Score is the StreetSmart Walk Score, for which destinations are identifiedusing a proprietary search strategy across diverse databases.We cannot know if there is spatial bias in the completenessof the amenity data. The metrics are constantly updated, astrength for current research but a challenge for longitu-dinal studies. Researchers should ensure they report thecalculation date for the data used, and in the long term,data archiving may be needed. We observed that BikeScore also has a ceiling effect (many census tractsscoring high), so that Bike Score may have limitedsensitivity for tracking change in already bikeableareas. Similarly, as Bike Score was created based onNorth American cities. Given the relatively low prevalenceof cycling infrastructure in many North Americancities, the score may not be calibrated for other loca-tions, especially those with extensive infrastructure.Winters et al. International Journal of Behavioral Nutrition and Physical Activity  (2016) 13:18 Page 8 of 10Finally to note that we used a Bike Score includingthe three environmental components, to analyse howthis predicts mode share; data for all components areavailable from the company, but the version of BikeScore visible on the website for US cities includes themode share component.This is the first study to use Bike Score, and covers 24cities across two countries. Several limitations of thecurrent work should be acknowledged. We used journeyto work mode share data from national surveys, the onlycomparable data across study cities. For the AmericanCommunity Survey, we used 2012 5-year averages to in-crease the stability of the estimates for census tracts. Weused the 2011 Canadian National Household Survey fortemporal alignment, but 5-year means do not exist. Inaddition, this is a voluntary survey and may carry highernon-response error than the Census data that wasformerly available. Journey to work mode share data isspatially located to work trip origins (home locations),and is not necessarily an indicator of the areas with thehighest cycling volumes or destinations. Of note, the 24study cities included here are those in the initial BikeScore launch; while the Canadian cities are diverse, manyof the US cities were selected because they had highcycling rates. The 9 Canadian cities included comprise6.7 million people, or 20.1 % of the Canadian popula-tion, whilst the 15 US cities comprise 16.6 millionpeople, or 2.1 % of the US population. The includedcities have great variety in terms of size and climate(Additional file 2: Table S1). Subsequent research mayinvestigate how neighbourhood composition impactsassociations between Bike Score and cycling modeshare. The city-wide Bike Score used in this analysisalso includes some proprietary population weighting,which explains why the city-wide mean Bike Scores donot match with the mean of the census tracts for thatcity. City-wide averages are also sensitive to adminis-trative city boundaries, which include surroundingsuburbs in some cases but not others, and it ispossible that these may differ between Canadian and UScities. Finally, this is a cross-sectional analysis of a newmetric. Planners can use Bike Score to prioritize whereto locate new infrastructure, and subsequent researchmay assess if changes in Bike Score are associatedwith changes in mode share.ConclusionsThe new Bike Score index predicts some of the variabilityin cycling to work mode share, and can be used for re-search with similar utility to the popular Walk Scoremetric [15, 16, 11]. Given the demonstrated significantand meaningful association across neighborhoods indiverse US and Canadian cities, Bike Score may be avaluable tool to aid with research and with planningfor bicycle infrastructure and increasing bicycle modein large studies. Further, our city-specific analysesshowed some city level variation, suggesting thatstudies within a city should further investigate thesuitability of this score and its component scores fortheir setting.Additional filesAdditional file 1: R code and output for all analyses. (PDF 1350 kb)Additional file 2: Table S1. Characteristics of 24 study cities (city-levelvariables). (DOCX 15 kb)Additional file 3: Table S2. Results of linear multilevel models.(DOCX 13 kb)Competing interestsThe authors have no financial or non-financial competing interests to declare.Authors’ contributionsMW, KT and MB conceptualized the study. MW compiled the data. DF ledthe statistical analyses. MW drafted the manuscript, and all author contributedmeaningfully to the interpretation of analyses and revisions of the manuscript.All authors have read and approved the final manuscript.AcknowledgementsWe gratefully acknowledge Josh Herst and Matt Lerner at Walk Score (nowRedfin Real Estate) for their collaboration in the development of Bike Scoreand the provision of data for this research. We also recognize Melissa Nunesand Moreno Zanotto for their assistance in preparation of the cycling andGIS data. This research was supported by a Canadian Institutes of HealthResearch Meetings, Planning and Dissemination Grant (Knowledge Translation),KTB-112789.Author details1Faculty of Health Sciences, Simon Fraser University, Blusson Hall Rm 11522,8888 University Drive, Burnaby, BC V5A 1S6, Canada. 2School of Populationand Public Health, University of British Columbia, 2206 East Mall, Vancouver,BC V6T 1Z3, Canada. 3Department of Community Health and Epidemiology,University of Saskatchewan, Health Science Building, 107 Wiggins Road,Saskatoon, Saskatchewan S7N 5E5, Canada.Received: 12 August 2015 Accepted: 30 January 2016References1. Woodcock J, Edwards P, Tonne C, Armstrong BG, Ashiru O, Banister D et al.Public health benefits of strategies to reduce greenhouse-gas emissions:urban land transport. Lancet. 2009;374(9705):1930-43. doi:10.1016/S0140-6736(09)61714-1.2. Woodcock J, Tainio M, Cheshire J, O’Brien O, Goodman A. Health effects ofthe London bicycle sharing system: health impact modelling study. BMJ.2014;348. doi:10.1136/bmj.g425.3. Teschke K, Reynolds C, Ries F, Gouge B, Winters M. Bicycling: Health risk orbenefit?. UBC Med J. 2012;3(2):6–11.4. Jacobsen PL. Safety in numbers: more walkers and bicyclists, safer walkingand bicycling. Inj Prev. 2003;9(3):205–9.5. Pucher J, Buehler R, Bassett DR, Dannenberg AL. Walking and cycling tohealth: a comparative analysis of city, state, and international data. Am JPub Health. 2010;100(10):1986-92. doi:10.2105/AJPH.2009.189324.6. Handy S. Health and community design: The impact of the builtenvironment on physical activity. J Am Plann Assoc. 2004;70(3):375–6.7. Saelens BE, Handy SL. Built environment correlates of walking: a review.Med Sci Sports Exerc. 2008;40(7 Suppl):S550-66. doi:10.1249/MSS.0b013e31817c67a4.8. Frank LD, Sallis JF, Conway TL, Chapman JE, Saelens BE, Bachman W. Manypathways from land use to health - Associations between neighborhoodWinters et al. International Journal of Behavioral Nutrition and Physical Activity  (2016) 13:18 Page 9 of 10walkability and active transportation, body mass index, and airquality. J AmPlann Assoc. 2006;72(1):75-87.9. Owen N, Cerin E, Leslie E, duToit L, Coffee N, Frank LD et al. NeighborhoodWalkability and the Walking Behavior of Australian Adults. Am J Prev Med.2007;33(5):387-95. doi:10.1016/j.amepre.2007.07.025.10. Chudyk AM, Winters M, Moniruzzaman M, Ashe MC, Gould JS, McKay H.Destinations matter: The association between where older adults live and theirtravel behavior. J Transp Health. 2014; 2(1):50–57. doi:10.1016/j.jth.2014.09.008.11. Hirsch JA, Moore KA, Evenson KR, Rodriguez DA, Roux AVD. Walk Score®and Transit Score® and Walking in the Multi-Ethnic Study of Atherosclerosis.Am J Prev Med. 2013;45(2):158–66. doi:10.1016/j.amepre.2013.03.018.12. Yew A, Lear S, editors. The Association of Walkability with Physical Activityat Baseline and 3-Year Follow Up. Active Living Reseach Conference; 2012;San Diego, CA.13. Thielman J, Rosella L, Copes R, Lebenbaum M, Manson H. Neighbourhoodwalkability: differential associations with self-reported transport walking andleisure-time physical activity in Canadian towns andcities of all sizes. PrevMed. 2015;77: 174-180.14. Carr LJ, Dunsiger SI, Marcus BH. Validation of Walk Score for estimatingaccess to walkable amenities. Br J Sports Med. 2011;45(14):1144-8. doi:10.1136/bjsm.2009.069609.15. Manaugh K, El-Geneidy A. Validating walkability indices: How do differenthouseholds respond to the walkability of their neighborhood? Transp Res DTransp Environ. 2011;16(4):309–15. doi:10.1016/j.trd.2011.01.009.16. Carr LJ, Dunsiger SI, Marcus BH. Walk Score™ As a Global Estimate ofNeighborhood Walkability. Am J Prev Med. 2010;39(5):460-3. doi:10.1016/j.amepre.2010.07.007.17. Pucher J, Buehler R, Seinen M. Bicycling renaissance in North America? Anupdate and re-appraisal of cycling trends and policies. Transp Res Part APolicy Pract. 2011;45(6):451-75. doi:10.1016/j.tra.2011.03.001.18. Pucher J, Dill J, Handy S. Infrastructure, programs, and policies to increasebicycling: An international review. Prev Med. 2010;50:S106-S25. doi:10.1016/j.ypmed.2009.07.028.19. Winters M, Brauer M, Setton E, Teschke K. Mapping bikeability: a spatial toolto support sustainable travel. Environ Plann B Plann Des. 2013;40(5):865–83.20. Moudon AV, Lee C. Walking and bicycling: An evaluation of environmentalaudit instruments. Am J Health Promot. 2003;18(1):21–37.21. Reynolds CCO, Harris MA, Teschke K, Cripton PA, Winters M. The impact oftransportation infrastructure on bicycling injuries and crashes: a review ofthe literature. Environ Health. 2009;8. doi:10.1186/1476-069X-8-47.22. Van Dyck D, Cerin E, Conway TL, De Bourdeaudhuij I, Owen N, Kerr J et al.Perceived neighborhood environmental attributes associated with adults’transport-related walking and cycling: Findings from theUSA, Australia andBelgium. Int J Behav Nutr Phys Act. 2012;9(70):1-14. doi:10.1186/1479-5868-9-70.23. Musiker R. Minneapolis Tops San Francisco, Portland as Most Bikeable Cityof 2015. Walk Score Blog, Seattle, WA. http://blog.walkscore.com/2015/05/minneapolis-tops-san-francisco-portland-as-most-bikeablecity-of-2015/#.VbgFuvlVhBc. Published 2015. Accessed May 23,2015.24. Statistics Canada. 2011 National Household Survey. https://www12.statcan.gc.ca/nhs-enm/2011/dppd/prof/index.cfm?Lang=E. Accessed May15, 2014.23. US Census Bureau. American FactFinder. http://quickfacts.census.gov/qfd/index.html. Accessed May15,2014.26. R Core Team. R: A language and environment for statistical computing. RFoundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0.http://www.R-project.org/.27. Winters M, Friesen MC, Koehoorn M, Teschke K. Utilitarian bicycling: amultilevel analysis of climate and personal influences. Am J Prev Med. 2007;32(1):52-8. doi:10.1016/j.amepre.2006.08.027.28. Dill J, Carr T. Bicycle Commuting and Facilities in Major U.S. Cities: If YouBuild Them, Commuters Will Use Them – Another Look. Transp Res Rec.2003;1828:116–23.29. Alliance for Biking and Walking. Bicycling and Walking in the U.S.: 2014Benchmarking Report; 2014.30. Heinen E, van Wee B, Maat K. Commuting by Bicycle: An Overview of theLiterature. Trans Rev 2010;30(1):59-96. DOI:10.1080/01441640903187001.31. Winters M, Teschke K. Route Preferences Among Adults in the Near Marketfor Bicycling: Findings of the Cycling in Cities Study. Am J Health Promot.2010;25(1):40-7. doi:10.4278/ajhp.081006-QUAN-236.32. Teschke K, Harris M, Reynolds C, Winters M, Babul S, Chipman M et al. Routeinfrastructure and the risk of injuries to bicyclists- a case-crossover study.Am J Public Health. 2012.33. Lusk AC, Furth PG, Morency P, Miranda-Moreno LF, Willett WC, DennerleinJT. Risk of injury for bicycling on cycle tracks versus in the street. Inj Prev.2011;17(2):131-5. doi:10.1136/ip.2010.028696.34. Winters M, Babul S, Becker H, Brubacher J, Chipman M, Cripton P et al. SafeCycling: How Do Risk Perceptions Compare With Observed Risk? Can JPublic Health. 2012;103(9):eS42-eS7.35. Garrard J, Rose G, Lo SK. Promoting transportation cycling for women: therole of bicycle infrastructure. Prev Med. 2008;46(1):55-9. doi:10.1016/j.ypmed.2007.07.010.36. Eyler A, Blanck H, Gittelsohn J, Karpyn A, McKenzie T, Partington S et al.Physical Activity and Food Environment Assessments: Implications forPractice. Am J Prev Med. 2015;48(5):639–45.37. US Geological Survey. National Elevation Dataset, Digital Elevation Model.http://ned.usgs.gov/index.html.•  We accept pre-submission inquiries •  Our selector tool helps you to find the most relevant journal•  We provide round the clock customer support •  Convenient online submission•  Thorough peer review•  Inclusion in PubMed and all major indexing services •  Maximum visibility for your researchSubmit your manuscript atwww.biomedcentral.com/submitSubmit your next manuscript to BioMed Central and we will help you at every step:Winters et al. International Journal of Behavioral Nutrition and Physical Activity  (2016) 13:18 Page 10 of 10


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items