Guidelines for Analysis of Investments in Bicycle Facilities (2006)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

B-1 APPENDIX B BICYCLING DEMAND AND PROXIMITY TO FACILITIES The Effect of Bicycle Facilities and Retail on Cycling and Walking in an Urban Environment INTRODUCTION Urban planners and public health officials have been stead- fast in encouraging active modes of transportation over the past decades. While the motives for doing so differ somewhat between professions—urban planning to mitigate congestion, public health to increase physical activity—both have ardently aimed to increase levels of walking and cycling among the U.S. population. Deci- sions to walk or bike tend to be the outcome of myriad factors. Conventional thinking suggests two dimensions are important: for cycling, this includes the proximity of cycling-specific infrastruc- ture (i.e., bicycle lanes or off-street paths); for walking, this includes the proximity of neighborhood retail (i.e., places to walk to). This study focuses on two modes of active transportation—walking and cycling—and two different elements of the physical environ- ment that are often discussed in policy circles, respectively, neigh- borhood retail and bicycle facilities. Our work aims to answer the following questions: (a) does having a bicycle lane/path close to home increase the propensity to complete a cycling trip and (b) does having neighborhood retail within walking distance increase the propensity to complete a walk trip from home? The primary advan- tage of this work is that it carefully analyzes these relationships for an urban population employing detailed GIS/urban form data and a robust revealed preference survey. The study uses multivariate modeling techniques to estimate the effect of features of the built environment on outcomes related to bicycling and walking. We first briefly review directly relevant literature to this pursuit and describe some issues that limit the utility of previous research. We explain the setting for this application, the travel data, and the detailed urban form data. We then report the results of our analysis in two different tracks: one to estimate the odds of cycling; another to estimate the odds of walking. The final section discusses these results and offers policy implications. EXISTING LITERATURE AND THEORY Attempts to document correlations between active transportation and community design have been a focus of much of the recent urban planning and public health literature. Available literature underscores the importance of this research (143, 144), establishes a common language for both disciplines (79, 145), helps refine approaches for future studies (146), and comprehensively reviews available work (147). Existing research, however, varies in geo- graphical scope, the manner in which it captures different dimensions of active transportation, and the strategies used to measure key fea- tures of the built environment. Some of the difficulty in tackling the literature on community design and physical activity—namely walk- ing and cycling—is that the bulk of the literature aggregates these two modes. Some known work coming from the non-motorized com- munity on the environmental determinants of bicycling and walk- ing dually considers both modes (148, 149). For abstract or general purposes this may suffice; the two modes are almost always aggre- gated in transportation research. In terms of daily use, facility plan- ning, and community design, however, bicycling and walking dif- fer substantially. The following paragraphs point the reader to some of the salient literature related to walking and cycling and assesses some of the theoretical differences between the two modes. Pedestrian travel and infrastructure have the following distin- guishing characteristics. First, all trips—whether by car, rail transit, or bus—require pedestrian travel because they start and end with a walk trip. Second, sidewalks and other pedestrian related infrastruc- ture (e.g., crosswalks, public spaces) are now often incorporated into zoning codes. Third, pedestrian concerns typically apply to con- fined travel-sheds or geographic scales (e.g., city blocks). Finally— and most germane to the analysis presented herein—pedestrian travel usually tends to be influenced by a broad array factors that go beyond simply sidewalks and other infrastructure. This means that both the attractiveness of features along the route (e.g., interesting facades, a variety of architecture, the absence of long, blank walls) and destinations (e.g., close-by stores) are important. Early research on pedestrian travel underscored the importance of neighborhood retail in creating inviting pedestrian environments (150–152). Several studies offer detailed strategies using these mea- sures (153–157). Much of the empirical work matches measures of pedestrian behavior with assorted place-based destinations in their work (158, 159) or even select measures of retail (160–166). How- ever, much of the available work on pedestrian behavior vis-à-vis retail tends to lack detailed spatial attributes or be specific to urban design features such as benches, sidewalk width or other streetscape improvements (167); few studies examine such behavior over an entire city with detailed measures of retail activity. Bicycle travel and facilities, on the other hand, apply to longer corridors, and fail to be used as frequently as walking facilities. Such trips are usually considered more discretionary in nature. Where pedestrian planning applies to a clear majority of the popu- lation (nearly everyone can walk), bicycle planning applies to a con- siderably smaller market of travelers—those who choose to own and ride a bicycle. During the summer months in most of the U.S., this includes just over a quarter of the American population (47). Bicycle travel has a longer travel shed and most of the population has a heightened sensitivity to potentially unsafe conditions (e.g., shared facilities with autos speeding by). The quality of the facility is often paramount. Such facilities along a route include wide curb lanes, and on-street or off-street bike paths. Similar concerns pervade available literature on cycling and the provision of cycling-specific infrastructure. There is consider- able enthusiasm about the merits of bicycle trails and paths to induce use (55, 168–170). Little work, however, has rigorously tested such claims. Existing studies have examined the use of particular trails

(171–173), cycling commute rates vis-à-vis bicycle lanes (174, 175) or their impact on route choice decisions (176). Again, there exists a dearth of empirical knowledge about the merits of such cycling infrastructure using disaggregate data for individuals who may live across entire cities. SETTING AND DATA Our research is based on the Twin Cities of Minneapolis and St. Paul, Minnesota, which border one another and are roughly the same geographic size (approximately 57 square miles each). The separate central businesses districts for each city are less than ten miles from one another. According to the 2000 Census, Minneapo- lis has roughly 100,000 more residents than St. Paul (382,618 ver- sus 287,151). The setting of these cities proves to be almost ideal for several reasons. Both Minneapolis and St. Paul are well-endowed with both on-street and off-street bicycle paths. Figure 9 shows a combined 60 miles of on-street bicycle lanes and 123 miles of off- street bicycle paths. Furthermore, the population comprise residents who appear to cherish such trails, particularly in the summer months. Minneapolis ranks among the top in the country in percentage of workers commuting by bicycle (175). For the walking query, each B-2 city also has a wide distribution of retail activity across the city (see the top half of Figure 10) and many homes with close proximity to neighborhood retail.v Our knowledge of who walked and cycled is derived from a home interview survey known as the 2000 Twin Cities Metropolitan Area Travel Behavior Inventory (TBI). This survey captures household travel behavior and socio-demographic characteristics of individuals and households across the 7-county metropolitan area, encompassing primarily the urbanized and suburbanized parts of Twin Cities of Minneapolis and St. Paul metropolitan area. The TBI data were orig- inally collected via travel diaries in concert with household telephone interviews.vi Participants were asked to record all travel behavior for a 24-hour period in which they documented each trip that was taken, including the origin and destination of the traveler, the mode of travel, the duration of the trip, and the primary activity at the destination, if one was involved.vii Household characteristics and household loca- tion were attributed to each individual. We additionally linked house- holds with neighborhood spatial attributes relative to their reported home location. We selected all subjects from the TBI diary database who were residents of Minneapolis or St. Paul and 20 years of age or older (n = 1,653).viii A key feature of this investigation is that it applies to two entire central cities, rather than precise study areas or specific corridors of interest. Figure 9. Map of study area showing bicycle facilities and home location of cyclists.

B-3 Figure 10. Maps of study area showing location of retail establishments (top) and home location of walkers (below).

POLICY VARIABLES OF INTEREST Our policy variables of interest differ for each mode and are based on distance which is often mentioned as a suitable measure of impedance (178). For cycling, our analysis examines the prox- imity of bicycle facilities in the form of on-street bicycle lanes and off-street bicycle paths (Figure 11). Three continuous distance mea- sures were calculated using GIS layers furnished by the Minnesota Department of Transportation, with separate map layers for on- street and off-street trails. Marrying this data with precise household locations, we calculated the straight-line distance in meters to the nearest on-street bicycle lane, the nearest off-street trail, and the nearest bike facility of either type. Four distinct categories represent the distance from one’s home to the nearest bicycle trail as < 400 meters (one-quarter mile), 400–799 meters, 800–1,599 meters, and 1,600 meters or greater (greater than one mile). For walking, we measure neighborhood retail in a detailed and rig- orous manner. We first obtained precise latitude and longitude infor- mation for each business in Minneapolis and St. Paul.ix Relying on the North American Industrial Classification System (NAICS),x we included businesses such as general merchandise stores, grocery stores, food and drinking establishments, miscellaneous retail and the sort.xi These types of businesses were retained because they would likely attract walking trips for neighborhood shopping and be representative of good walking environments that would likely gen- erate non-shopping walking trips. We again combined this informa- tion with household location data. Finally, we calculated the network distance between the home location and the closest retail satisfying the above criteria. For analysis, we used the distance variables to classify subjects into one of four categories. The four categories rep- resent the distance from home to the nearest retail establishment as < 200 meters (one-eighth mile), 200–399 meters, 400–599 meters, and 600 meters or greater.xii To provide the reader with visual repre- sentations of the retail “catchment” areas for varying distances, we provide Figure 12 showing a home location (in the center) and retail establishments within varying walking distances from the home. When measuring each of the policy relevant variables, a four- level ordinal variable is advantageous over the continuous distance measure in two respects. First, the categorical measure allows us to relax the strong linearity assumption that underlies continuous mea- sures.xiii Second, the four-level categorical measure allows flexibil- ity relative to ease of presentation and intuitive interpretation. Given that we used distance cut-points with relatively simple interpreta- tion, it provides a compelling way to grasp the reported findings in B-4 terms of comparing individuals who live within 400 meters of a bike trail and those who live more that 1,600 meters from a bike trail.xiv COVARIATES We identify several covariates to represent individual, household, and other characteristics. These covariates represent factors that may differ across exposure levels and thus could potentially con- found our effect estimates. To help free our estimates from con- founding explanations we use these covariates to statistically equate subjects on observed characteristics across exposure groups; there- fore, the only measured difference between them is the proximity to either the retail or the bicycle facilities. For individual characteristics, we use age, gender, educational attainment (college degree or not), and employment status (employed or not). For household characteristics, we use household income (five categories), household size, and whether the household had any children younger than 18 years old. We also use two other measures: household bikes per capita and household vehicles per capita. We calculate these by dividing the total number of bicycles by household size and dividing the total number of vehicles by household size. Spatial measures and other attributes of the built environment in this study are limited to proximity to retail or bicycle facilities. Focusing on a sample contained within the boundaries of Min- neapolis and St. Paul helps to control for other variation among spa- tial measures by nature of the research design. For example, our sample has little variation in density,xv regional accessibility, and access to open space. Other than a few small bluffs—where there are few known residences—there is little variation in topography. And, almost every neighborhood street in Minneapolis and St. Paul has sidewalks on both sides, thereby controlling for an often cited important urban design feature. RESULTS AND INTERPRETATION Overall, our sample was nearly evenly split on gender (52% female vs. 48% male) and two-thirds (67%) were residents of Minneapolis (as opposed to St. Paul). Most subjects were employed (83%) and had at least a 4-year college degree (63%). The majority lived in house- holds with no children (80%) and reported household incomes less than $50,000 per year (36%). Figure 11. Representative photographs of off-street trail and on-street bicycle lane (respectively).

We first used descriptive techniques (i.e., chi-square and t-tests) to characterize our sample by proximity to each type of facility. We explored the distributions of individual and household characteris- tics for subjects at each level of exposure. Subjects living within dif- ferent proximity levels to bike facilities or retail differ somewhat with respect to many of the individual and household characteris- tics. For example, subjects living in close proximity to any bicycle facility are more likely to be 40 or older, have a college degree, and live in households with no children than subjects living farther away from a bike facility. Different covariate patterns emerge depending upon which proximity measure we examine. The outcomes of interest in this application are twofold; both were operationalized in a dichotomous manner. The first is whether the respondent completed a bicycle trip as documented in the 24-hour travel behavior diary. A total of 86 individuals from our 1,653 indi- viduals reported doing so (5.2 percent).xvi While this rate is higher than both the larger TBI sample and national average—which tend to hover around two percent of the population (179)—one needs to recognize this is a relatively small number. However, a close look at this population showed that they did not have exceptional or pecu- liar characteristics (e.g., the majority of them showing up near the university). The second outcome of interest was if the respondent completed a walking trip from home, 12.4 percent of our sample (n = 205).xvii Because our outcome measures are dichotomous, we use multiple logistic regression models to examine the effect of facilities and retail on bicycling or walking. For each proximity measure (e.g., distance to any trail, distance to on-street trail, distance to off-street trail, dis- tance to retail), we conduct a series of analyses; they build from a sim- ple logistic regression of the exposure on the outcome to a multiple B-5 200 meters 400 meters 600 meters 800 meters 1200 meters 1600 meters Lake Calhoun Lake Calhoun Lake Calhoun Lake Calhoun Lake Calhoun Lake Calhoun Lake Harriet Lake Harriet Lake Harriet Lake Harriet Lake Harriet Lake Harriet Figure 12. Retail “catchment” areas for an example home (shown in the center) and varying network walking distances. Dots represent retail locations; polygon shaded areas represent a catchment area.

logistic regression fully adjusted for all subsets of covariates (Mod- els 1 and 2 in Table 12). Because our data are hierarchically structured—individuals are nested within households—we use robust standard errors to account for the effects of this clustering. Subjects who reside in the same household are more alike within a household than they are with sub- jects residing in other households. Accordingly, less independent information is contributed by individuals from the same household, which may artificially decrease the standard error of the estimate.xviii Bicycling Our first models explore the odds of bicycle use and proximity to any type of bicycle facility. From the simple logistic regression B-6 model to the fully adjusted model, the odds of bike use did not dif- fer significantly by proximity to any bike facility (this includes either on-street bicycle facility or off-street bicycle trails). Our model suggests that there is no effect of proximity to any bike facil- ity on bike use. We therefore used a separate model to estimate the effect of proximity to off-street facilities on the odds of bike use. Examining the simple logistic regression model to the fully adjusted model for off-street bicycle facilities, the odds of bike use did not differ significantly by proximity to a trail. We detected no effect of proximity to off-street bike facilities on bicycle use. Finally, we examined the effect of proximity to on-street bicycle facility on the odds of bike use. In the simple logistic regression model (Model 1a in Table 12), subjects living within 400 meters of an on- street bicycle facility had significantly increased odds of bike use com- pared with subjects living more than 1,600 meters from an on-street Bicycle Use Walk use Model 1a Model 1b Model1c Model 2a Model 2b Model 2c Distance to nearest on-street bicycle path Distance to nearest retail establishment < 400 meters 2.933 3.101 2.288 < 200 meters 3.098 3.060 2.348 (3.11)** (3.21)** (2.23)* (3.41)** (3.36)** (2.51)* 400 – 799 m 2.108 2.012 1.511 200 – 399 m 1.653 1.616 1.316 (2.05)* (1.89) (1.07) (1.48) (1.41) (0.80) 800 – 1599 m 1.390 1.361 1.163 400 – 599 m 1.448 1.422 1.288 (0.88) (0.81) (0.39) (1.02) (0.97) (0.69) >= 1600 m referent referent referent >= 600 m referent referent referent Individual Characteristics Male subject 2.015 2.160 0.760 0.787 (2.96)** (3.12)** (1.80) (1.57) College 1.753 2.840 1.113 1.271 (2.15)* (3.47)** (0.68) (1.42) Employed 0.783 1.187 0.771 0.901 (0.71) (0.43) (1.24) (0.49) 40-59 years 0.520 0.623 1.004 1.112 (2.73)** (1.83) (0.03) (0.64) >=60 years 0.081 0.115 0.769 0.752 (3.49)** (2.98)** (1.03) (1.10) Household Characteristics $15,000 - $49,000 0.402 0.874 (2.30)* (0.41) $50,000 - $74,999 0.293 0.704 (2.83)** (1.00) >= $75,000 0.206 0.880 (3.33)** (0.35) Income missing 0.172 0.886 (3.00)** (0.32) Household w/ kids 0.640 0.790 (2.21)* (2.08)* HH bikes per capita 2.463 0.892 (7.85)** (0.80) HH vehicles per capita 0.114 0.300 (5.29)** (4.56)** Wald chi-square = 137.65 Wald chi-square = 55.61 Log pseudolikelihood = -262.34 Log pseudolikelihood = -583.69 Pseudo R-square = 0.224 Pseudo R-square = 0.058 # of observations in all models = 1653 Odds ratios, robust z statistics in parentheses. * significant at 5 %; ** significant at 1 % TABLE 12 Models comparing the effect of distance to an on-street bicycle facility on odds of bike use (Models 1a-1c) and the effect of distance to neighborhood retail on odds of making a walk trip from home (Models 2a–2c)

bike facility. As expected, those that lived within 400 to 799 meters of an on-street bike facility also had significantly increased odds of bike use compared with subjects living more than 1,600 meters from an on- street bike facility, although the odds of bike use were slightly lower than for those living closest to an on-street facility. After adjusting for individual and household characteristics, the effects were somewhat attenuated (see Models 1b and 1c). Subjects living in close proximity to an on-street facility (< 400 meters) still had statistically significantly increased odds of bike use compared with subjects living more than 1,600 meters from an on-street bike facility. However, subjects within 400 to 799 meters still tended toward increased odds of bike use, however this failed to reach the level of statistical significance. Consistent with prevailing theories of bicycle use, our models show that cyclists tend to be male, from older populations, and from households with children. Walking We employed a similar approach to examine walking behavior vis-à-vis retail and discovered similar results. In the simple logistic regression model (Model 2a in Table 12), subjects living within 200 meters of a retail establishment had significantly increased odds of making a walk trip compared with subjects living more than 600 meters away from retail. Households living between 200– 400 meters and 400–600 meters of retail, however, failed to reach a level of statistical significance. Again, after adjusting for individual and household characteris- tics, the effects were somewhat attenuated (see Models 2b and 2c). Subjects living in close proximity to retail (< 200 meters) still had statistically significantly increased odds of walking. Interestingly enough, however, household with children was the only household characteristic variable that was significant. In each model, the results suggest that distance to bicycle facilities and retail is statistically significant; however, the relationship is not linear. The most important point is that close proximity matters, which challenges conventional wisdom that people are willing to walk up to one-quarter mile as well as analogous cycling specific hypotheses. CONCLUSION This research reports the results of individual level models predict- ing bicycling and walking behavior and correlations with proximity to bicycle paths and neighborhood retail, respectively. We do so focus- ing on particular behavior—whether an individual biked or walked from home—and robustly measuring policy relevant dimensions of the built environment. The travel, bicycle facility, and the retail data we employed are the most precise among city-wide measures for a metropolitan area in the U.S. The primary merits of this exercise focus specifically on measuring the exposure measures, each of which have direct policy relevance. To our knowledge, this question has not previously been asked or answered across an entire city. We separated facilities into two categories: off-street bicycle trails and on-street bicycle lanes. For the former group of facilities, there is no effect of proximity to off-street bike facilities on bicycle use. For on-street bicycle lanes, subjects living within 400 meters of a bike facility had significantly increased odds of bike use com- pared with subjects living more than 1,600 meters from an on-street B-7 bike facility. Walking use increases if retail is within 200 meters. While not the focus of this analysis, our study reaffirmed that many of the socio-demographic and economic variables used in other studies are important. Some officials have supported the use of community design to induce or enable physical activity, but this analysis suggests that the argument is more complex. First, our results underscore the fact that we are addressing fringe modes and rare behavior (180). Even among the urban population, only five percent cycled and twelve percent walked. And, the criteria for satisfying this measure were generous—any cycling or walking trip from home that was reported by the individual over a 24-hour period.xix Second, the research sup- ports the theory that the built environment matters; however, it sug- gests that one needs to live very close to such facilities to have an statistically significant effect (i.e., less than 400 meters to a bicycle trail for bicycling and less than 200 meters to retail for walking— approximately the length of two football fields). While the odds- ratios for longer distances failed to reach levels of statistical signif- icance, it is important to mention that in all model estimations, they were always in decreasing orders of magnitude and always in the assumed direction. Planners need to be aware of such distance considerations when designing mixed land use ordinances (181). The results, however, need to be viewed in the following light. The first consideration is that the analysis is reported for only an urban and adult population. Conventional wisdom suggests children (84), rural or suburban residents (182) may value different features of the built environment.xx The second is that the original TBI survey was the result of a complex sampling design which needs to be taken into account.xxi Being based on cross-sectional analysis, these results cannot be used to infer causal relationships (183). We can conclude that respondents living very close to bicycle paths or retail bike or walk more than their counterparts further away. However, consistent with emerging theories about travel behavior, the decision to live in close proximity to such features is likely endogenous (184). There are likely attitudes, preferences, or other attributes motivating such bicycling or walking behavior (185, 186). Such attributes are not directly captured in this analysis—and, strictly using the results from this research, we would be remiss to conclude that adding retail or bicycle paths would directly induce such behavior. This investigation makes progress by using focused research and carefully measured variables. The work raises a number of important data, measurement, and methodological issues for future researchers endeavoring to predict levels of walking or bicycle use for entire cities or metropolitan areas. We make headway in learning that dis- tance matters—particularly close distance. Relative to the larger picture of travel behavior, however, our understanding remains unclear. The evidence suggests that features of the built environ- ment matter, although it is hardly compelling. Statistical analysis like ours needs to be complemented with more direct sampling as well as qualitative modes of analysis to shed light on different fac- tors and attitudes as well as sorting out the issue of residential self- selection. Further work will inevitably allow planners and modelers to better understand relationships between cycling and walking infrastructure and physical activity. Continued and thorough under- standing will therefore assist policymakers to construct better informed policies about using features of the built environment to induce physical activity, namely walking and cycling.