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.
7 Study Methodology In this study, investigations center on the revealed preferences of existing cyclists and the stated preferences of potential cyclists through a multi-method data collection effort. First, focus groups were held to explore current and potential usersâ needs and preferences and to aid in survey design in later steps. The study then continued with a before-and-after survey methodology. The first-wave survey was distributed among a sample of current and potential users in the study areas to evaluate their personal attitudes, preferences, and behaviors before the opening of planned bicycle facilities. A second-wave survey was also administered to all respon- dents of the first-wave survey to assess changes in these factors, particularly for perceptions of and preferences for bicycle facilities and frequency of bicycling. Study Site Selection The choice of specific study areas was driven by the timelines of the new bicycling facility projects, as well as the expected date of entry into service of the newly built infrastructure. Six neighborhoods (study areas) were included in this study: â¢ Three âtreatmentâ neighborhoods in which bicycling facilities were scheduled to be opened between Fall 2016 and Fall 2017. â¢ Three âcontrolâ neighborhoods in which no bicycling facilities were planned to open over the same time frame. Each control neighborhood was chosen to serve as an approximate match with one treatment area based on demographics and land use characteristics. The neighborhoods were chosen according to selection criteria based on the relevance and extent of the bike-infrastructure projects and the type of bicycling facility to be built. Previous research has predominantly been conducted in communities where cycling is widely accepted and automobile drivers are conditioned to the presence of cyclists. In contrast, this study focuses on communities in Alabama and Tennessee, where cycling for transportation is some- what new and rapidly expanding. The three project sites include the following: â¢ Roadway diets in Opelika, Alabama, that planned to turn a four-lane road into a three-lane road with buffered and protected/separated bike lanes for roughly a mile. â¢ Plans to add 4 miles of protected/separated bike lane extensions in Chattanooga, Tennessee. â¢ A network of short bike lanes and sharrows in Anniston, Alabama, that designated space for cyclists around the center of town. The original timeline and scope of the project in each treatment neighborhood changed as the research progressed: â¢ In Opelika, Alabama, roughly a mile of two-way unprotected cycletracks was implemented in Spring 2017. C H A P T E R 2
8 Bicyclist Facility Preferences and Effects on Increasing Bicycle Trips â¢ In Chattanooga, Tennessee, about 2 miles of traditional (unprotected) bike lanes was imple- mented in Summer 2018. â¢ In Anniston, Alabama, a network of sharrows was added to several downtown streets in Fall 2016. Google Street View images for the roadways before and after the bicycle treatments in Opelika and Chattanooga are presented in Figure 2.1, Figure 2.2, Figure 2.3, and Figure 2.4. Research Design Literature Review There is a sizeable amount of literature regarding bicycle facility preferences and the impact of bicycle facilities on bicycling rates; however, there are substantial gaps in the literature to date. This section includes an overview of research designs for forecasting the effects of bicycling facilities, along with their limitations. This review has informed the design of the study that is the subject of this report. Early studies employed cross-sectional methods, where measurements at a single point in time are used to establish associations between observed behaviors (such as bicycling rates) and possible factors (such as bicycle facilities) influencing such behavior (Krizek et al. 2009b). The Figure 2.1. Representative view of street in Opelika before treatment (Credit: Google Street View). Figure 2.2. Representative view of street in Opelika after treatment (Credit: Google Street View).
Study Methodology 9 first major study of this nature, Nelson and Allen (1997), evaluated data from 18 major U.S. cities, and found a loose correlation at the city level between miles of bicycle infrastructure and cycling rates. Other aggregate-level studies followed suit in efforts to explain inconsistencies observed throughout different cities. These studies increased the number of cities and variables, with Dill and Carr (2003) using data from 43 large U.S. cities and Buehler and Pucher (2012) using data from 90 of the 100 most populous U.S. cities. Both studies confirm a correlation between infrastructure availability and bicycle-commute mode share at the city level. On the more granular census-tract level, Teschke et al. (2017) performed a study in Vancouver and Montreal, identifying that living in tracts near bikeways, especially cycletracks, was associated with a greater probability to bike. Although cross-sectional methods are informative, they reveal only correlationânot causality. Consequently, these studies fall short of adequately answering the question of whether bicycling preceded infrastructure or vice versa. Other studies have taken a disaggregate approach to identifying the relation between bicy- cling rates and infrastructure. Such studies use individualized data rather than data aggregated Figure 2.3. Representative view of street in Chattanooga before treatment (Credit: Google Street View). Figure 2.4. Representative view of street in Chattanooga after treatment (Credit: Google Street View).
10 Bicyclist Facility Preferences and Effects on Increasing Bicycle Trips on the city or census-tract level. Several studies have found a positive link between a personâs propensity to cycle and the presence of dedicated facilities (Moudon et al. 2005; Krizek and Johnson 2006; Akar and Clifton 2009; Handy and Xing 2011; dellâOlio et al. 2014; Stinson et al. 2014). Others have found cyclists near bicycling facilities made more trips (Dill and Voros 2008; Stinson et al. 2014). Xing et al. (2010) also found a correlation between the presence of bicy- cling facilities and the miles a cyclist will ride. Disaggregate studies are useful in investigating relationships between individual-level data, though without a time element, there is still a weak case of determining causality rather than solely correlation. Many researchers have recommended time-based studies to counteract the major flaw of cross-sectional studies in failing to identify time-based trends (Nelson and Allen 1997; Dill and Carr 2003; Pucher et al. 2010; Buehler and Pucher 2012). Repeated cross-sections have been conducted for major cities to measure bicycle commuting rates at two points in time (before and after infrastructure investments). Krizek et al. (2009a) used a repeated cross-sectional design with data from two consecutive decennial censuses to show that traffic analysis zones near new bicycling facilities showed increased cycling rates as compared with traffic analysis zones outside of the buffer zone. However, Cleaveland and Douma (2008) used a similar methodology in six other major U.S. cities and showed varying effects. Parker et al. (2013) conducted an aggregated count-based study along a corridor in New Orleans before and after the implementation of a bike lane with two parallel control streets, showing that more users biked along the corridor after implementation and fewer biked along the control streets; however, the scale was not large enough to truly assess the changes throughout the neighborhood. Although repeated cross-sections are an improvement over the basic cross-sectional design, they still have their own limitations, resulting in a need for experimental and quasi-experimental studies. Such studies employ surveys at two or more points in time to measure changes in prefer- ences or behavior individually, as opposed to two aggregate measures. In a truly experimental survey design, samples of the population are randomly assigned to the treatment and control groups and intervention is administered to the treatment group. However, as Krizek et al. (2009b) pointed out, when studying the effect of bicycle facilities on bicycle ridership, it is not possible to randomly grant members of the population access to the intervention, as would be required in a true experimental design. Instead, quasi-experimental methods may be used, in which residentsâ behavior is measured before and after the intervention, controlling for factors other than the intervention that may influence the behavior. This behavior is then compared with the behavior of residents from a neighborhood without a similar intervention, with all other measurable variables being as similar as possible. Quasi-experimental studies on bicycle facility implementation have been rare. Researchers have differed in describing what constitutes a quasi-experimental design, with Mitra et al. (2016) calling a repeated cross-section study quasi-experimental, and Heesch et al. (2016) calling a repeated cross-section a natural experiment. Heinen et al. (2015) conducted a 4-year quasi- experimental panel study finding that UK commuters were likely to begin bicycling on a new multi-use path for trips they already make. Although this disaggregate study was able to quan- tify use of the new facility, there was no control for users diverting from existing infrastructure, so it could explain only the overall trends in the neighborhood, without being able to separate the infrastructure effects from any other environmental effects. A similar study was performed by Song et al. (2017), also in the United Kingdom, using three waves of panel data collected from residents near urban cycle facilities. They found no general change in mode choice in the aggregate but did find that those who started using the facility had significant mode shifts, particularly away from private automobile use. Sahlqvist et al. (2015) similarly conducted a panel survey for residents near multi-use paths in cities throughout the United Kingdom. They found that measures related to positive perceptions of walking and bicycling generally
Study Methodology 11 improved after the implementation of new bicycling facilities, though they lacked an analysis to describe differences in preferences or use. Rissel et al. (2015) performed a similar study in Australia, which used bike counts in addition to survey data, finding that bike counts increased after the treatment. However, the self-reported cycling rates did not change significantly, likely because of riders shifting from other routes or individuals outside the study area using the routes more often. Thus, there is a need for more extensive studies of the quasi-experimental nature with the specific purpose of analyzing the effects of infrastructure on propensity to cycle, including (1) disaggregate data of individual measures, (2) repeated observations before and after treatment implementation, and (3) a control group that closely mirrors the treatment group with the only difference being the treatment itself. These considerations were key in the design of the study for this NCHRP project. Data Sources One challenge in determining the causal effect of infrastructure on bicycling behavior is the number of possible confounding variables, which requires collecting accurate data on many such potential variables, including from nonbicyclists. This section includes a summary of necessary data sources and potential collection methods. Qualitative methods, such as interviews and focus groups, are critical to understanding infrastructure needs (Handy et al. 2014). These qualitative methods can support quantita- tive methods, such as statistical models, by suggesting new variables to be tested (Clifton and Handy 2003; Spencer et al. 2013). Focus groups can provide insights into attitudes, perceptions, preconceptions, and factors that might prompt changes in behavior. Chatterjee et al. (2013) used interviews to discover the impact of life events as triggers for change, while the interviews of Steinbach et al. (2011) and Aldred (2013) and the focus groups of Daley and Rissel (2011) revealed the connections between cycling and social stigma. While qualitative data sources are difficult to analyze statistically, they do give researchers a chance to get immediate feedback and ask specific questions geared toward the research subject, as Emond and Handy (2012) and Underwood et al. (2014) did in their investigations of teenagersâ cycling patterns. For these reasons, focus groups were included in the first phase of this NCHRP project. Quantitative data is commonly obtained on the aggregate level from existing sources, such as the Census, the American Community Survey, or the National Household Travel Survey, which allows for large-scale studies comparing different geographic areas (Dill and Carr 2003; Cleaveland and Douma 2008; Parkin et al. 2008; Krizek et al. 2009b; Buehler and Pucher 2012; Jones 2012; Schoner and Levinson 2014; Stinson et al. 2014). Similar data sets are used abroad, such as the National Travel Survey in Denmark (Nielsen and Skov-Petersen 2018) and the Bicycle Ridership survey in Canada (Cabral et al. 2018). These types of data sets have been used for cross-sectional and repeated cross-sectional designs, though they cannot be used to describe individual changes based on treatment. For this, and other reasons, Cabral et al. (2018) demonstrated that using exclusively public data falls short of the mark. Repeated observations can be difficult to collect, particularly on the individual level, because this requires substantial, consistent data collection from the same individuals over a sustained period (Nelson and Allen 1997). However, such data are necessary for a quasi-experimental design and the associated implications of causality. Researchers have typically used surveys as the primary data collection instrument for panel studies. Xing and Handy (2014) warned that the survey platform itself might influence the rep- resentativeness of the sample. Intercept surveys can be used to collect data from bicyclists (Mitra et al. 2016; Thakuriah et al. 2012). Yet such surveys are not capable of measuring the general population, and repeat observations from the same individuals are more difficult to solicit.
12 Bicyclist Facility Preferences and Effects on Increasing Bicycle Trips Forsyth et al. (2010) pointed out the importance of using household-based surveys to collect data on noncyclists, and that surveys targeting only current cyclists may lead to results that do not represent the population in general. The flexibility of designing a custom survey allows researchers to redefine the variable of interest. Typical measures of actual cycling are ideal, though, particularly for noncyclists, measures of intentions or perceptions can be informative in their own right. These can be measured by presenting respondents with hypothetical bicycle infrastructure and asking ques- tions about their perceived comfort or safety or their willingness to try bicycling on such a facility. Sanders (2014) presented respondents with digitally manipulated images, using the images to isolate small changes while keeping the rest of the environment the same. Since then, others have used images in their surveys (Mertens et al. 2016; Abadi and Hurwitz 2018; Ghekiere et al. 2018). Griswold et al. (2018) similarly used videos embedded in the survey to produce an added level of relatability. Although more advanced forms of media allow for more realistic experiences, these methods require specialized survey mediums, which may limit the ability to obtain a representa- tive sample, particularly from those who have reduced access to the appropriate technology. Even more advanced data sources have also been utilized. Blanc and Figliozzi (2016) employed crowdsourced data and an app to gather self-reported measures of cyclistsâ comfort. Oh et al. (2017) used instrumented bike data to estimate objective cyclist comfort rather than stated or perceived comfort. MarquÃ©s et al. (2015) even used bikeshare data to estimate overall bike- ridership data. However, the difficulty in applying these methods for data collection from the general public is substantial. For that reason, the methodology for this NCHRP project is limited to household printed and online surveys. This project was conducted in areas of the United States where bicycling for transportation was not presently widespread. For this reason, focus groups were held before the design of quan- titative data collection to probe for key considerations for the study population. Further, mail- home surveys were chosen to facilitate the collection of multiple observations from the same respondents. Although online surveys are often much more convenient, this survey medium would have been too narrow to collect an adequately representative sample of the population, perhaps through exclusion of those who are not comfortable with technology or have reduced access to it. For these reasons, the surveys for this NCHRP project were administered with both an online and a printed mail-back option. Covariates for Bicycling While this study seeks primarily to quantify the effects of infrastructure, it is also important to control for other factors. Studies have found a vast array of individual, environmental, cultural, and political factors associated with cycling. Individual characteristics are commonly found to relate to propensity to bike. Handy et al. (2014) summarized common elements, such as gender, income, and age. Characteristics such as education, income, and vehicle ownership are often correlated and can present difficulty in discerning effects. Education level is typically viewed as a positive indicator of bicycling in the United States (Krizek and Johnson 2006). Krizek and Johnson (2006) found an inverse relation- ship between income and bicycling. Handy and Xing (2011) found that age, income, and educa- tion level were not significant on their own, though homeownership is, which could serve as a proxy for the combined effects of all three. These characteristics, along with other commonly collected demographics, were controlled for in the survey used in this NCHRP project. The key to interpreting the interaction between these collinear variables may lie in attitudes. Handy and Xing (2011) identified important attitudes related to mode preference, such as
Study Methodology 13 bicycling comfort, liking bicycling, needing a car, limiting driving, liking transit, the need to run errands on the commute, the need to drive, and a preference for living in a bikeable community. Other individual factors include exercise habits (Moudon et al. 2005) and good health (Emond et al. 2009). Emond et al. (2009) also found that liking cycling increased propensity to cycle, while liking transit and the perception that cyclists are poor are negatively associated with cycling. Many attitudinal items were designed into the survey used in this study, including some that were bicycle-specific, others that were general to transportation, and more still that related to time use and technology adoption. Parkin et al. (2008) pointed out that ethnic origin is likely a contributor based on its repre- sentation of different cultures that may influence cycling behavior. Hankey et al. (2012) found that whites are less likely to cycle. However, Parker et al. (2013) found ethnicity insignificant as a predictor of changing behavior based on infrastructure investments. Bike ownership has been shown to be a significant enabling factor (Cervero and Duncan 2003; Moudon et al. 2005; Krizek and Johnson 2006; Akar and Clifton 2009). Handy et al. (2010) further analyzed the predictors of bicycle ownership, suggesting that improving peopleâs perceptions and attitudes toward bicycling will increase bicycle ownership and use. These and other characteristics can have major implications for how an individual uses and perceives bicycle facilities. Further, the heterogeneity of preferences and perceptions among different groups can be controlled for in several ways (Rossetti et al. 2018). FÃ©lix et al. (2017) discussed some of these modeling strategies, including segmented models and latent class models. They found most useful segmentations include classes similar to proficient riders, those willing but not convinced, and noncyclists. Sanders and Judelman (2018) used frequency-based segments while Wang and Akar (2018) used attitudinal and frequency-based classes of regular cyclist, potential cyclist, pro-drive noncyclists, and pro-walk noncyclists. Dill and McNeil (2013) segmented the population into four different cyclist types based on confidence level: strong and fearless, enthused and confident, interested but concerned, and no way, no how. Handy et al. (2010) segregated based on both bike ownership and bicycling frequency. One difficulty in studying the effects of facilities is the difficulty in separating the facility from the environment. Dill et al. (2014) suggested that the built environment influences cycling behavior through its effects on attitudes and perceived behavioral control. Urban form appeared to be significant in multiple studies (Pucher and Buehler 2006; Parkin et al. 2008; Buehler and Pucher 2012; Stinson et al. 2014). Cole-Hunter et al. (2015) found that vegetation along the route is associated with more cycling, while changes in elevation are associated with less cycling. Holle et al. (2014) found that vegetation can make cycling infrastructure more inviting to cyclists and noncyclists. Slope and elevation differences have also been identified as deterrents to cycling (Nelson and Allen 1997; Dill and Voros 2007; Parkin et al. 2008; Cole-Hunter et al. 2015). Many of these characteristics can be difficult to include in a survey. Communities with environments similar to those of the treatment communities should therefore be chosen as controls. Addi- tionally, stated preference questions should be designed to isolate the effects of any nonfacility environmental characteristics from those of the facilities themselves. Piatkowski et al. (2017) identified behavioral patterns of interactions between automobile users and cyclists, finding that drivers behave in a way to enforce their perceived norms, based on personal experience rather than standardized and existing laws. Thompson et al. (2017) further corroborated these findings and suggested that for engineering solutions to change this danger- ous dynamic, bicycle facilities must physically separate cyclists from motorists. The effects of perceived safety on cycling have been backed up by analysis of cycling rates and actual safety (Pucher and Buehler 2006; Buehler and Pucher 2012), so measures that improve the safety of the cycling environment can jointly serve both interests by also encouraging cycling. A more comprehensive analysis of bicycle safety was performed by DiGioia et al. (2017).
14 Bicyclist Facility Preferences and Effects on Increasing Bicycle Trips The needs, behaviors, and preferences of cyclists may vary based on trip type. Trip purpose likely plays at least a small part in explaining inconsistencies between studies in this regard. Many studies only considered commute trips because of the ease of obtaining aggregate commuting data, which may miss valuable data from other trip purposes (e.g., Nelson and Allen 1997; Dill and Carr 2003; Cleaveland and Douma 2008; Parkin et al. 2008; Krizek et al. 2009a; Buehler and Pucher 2012; Jones 2012). Others built disaggregate bicycle-commute mode choice models (Handy and Xing 2011; dellâOlio et al. 2014). Some studies accounted for all trip types with no specification of purpose (Hankey et al. 2012; Parker et al. 2013; Heinen et al. 2015). Other studies accounted for differences in behavior between commuting/utilitarian trips and recreational trips, and their models included separate considerations for each (Dill and Voros 2007; Xing et al. 2010; Stinson et al. 2014). Buehler and Pucher (2012) suggested that separate facilities would see more use along utilitarian routes than along recreational routes. Thus, the survey for this NCHRP project was designed to collect cycling frequencies and distances separately based on trip purpose. Although the needs and preferences of cyclists are often studied, there is a shortage of research involving current and potential cyclists from places in the United States more representative of the typical cycling scene, in which cycling is an emerging activity. The few studies that do explore stated preferences from the general population do not link these preferences back to characteristics about the type of cyclists. The research conducted in this project is designed to confirm findings from studies conducted in cycling hubs to assess if these trends hold across dif- ferent communities. In addition, the research has incorporated questions regarding cyclist type to explain the differences in preferences among different types of cyclists. Focus Group Methodology The first stage of the project involved using focus groups intended to provide insight to the unique problems cyclists face. The focus groups were intended to provide valuable perspective on their own and were also used to refine the survey. The focus groups were held in the three treatment communities (Anniston, Opelika, and Chattanooga). Six focus groups were organized with two sessions in each location in Spring 2016. Recruit- ment efforts, including e-mails to community groups and advertising on Facebook, were made to gather opinions from regular cyclists as well as from those who did not currently bike. The only requirement to be considered was to be physically capable of riding a bike. Participants were offered a modest incentive of $40 to attend a 90-minute focus group held at a venue located near local planned bicycle facility improvements. Characteristics of focus group participants are included in Table 2.1. Gender Ethnicity Rider Frequency Age Kids at Home Total M al e Fe m al e W hi te M in or ity N ev er O cc as io na l Fr eq ue nt 18 â2 4 25 â4 4 45 â6 4 65 + Ye s N o Opelika 1 6 4 9 1 1 6 3 0 3 6 1 5 5 10 Opelika 2 6 2 8 0 1 5 2 1 4 3 0 3 5 8 Anniston 1 5 5 10 0 0 3 7 1 8 1 0 7 3 10 Anniston 2 1 1 2 0 0 0 2 0 1 1 0 1 1 2 Chattanooga 1 7 6 7 6 2 2 9 0 9 3 1 3 10 13 Chattanooga 2 5 4 8 1 1 4 4 2 4 3 0 1 8 9 Total 30 22 44 8 5 20 27 4 29 17 2 20 32 52 Table 2.1. Summary of demographics of focus group participants.
Study Methodology 15 As the focus groups began, participants were prompted to share background involving their experiences as cyclists and how they viewed the bicycling conditions of their communities. Respondents were also asked to share their thoughts on things that make them feel comfort- able when bicycling. Those who had children at home were also invited to consider their views toward their children bicycling. For the second stage, images of various facility types created in Adobe Photoshop were presented to respondents. One common roadway setting was chosen as a base image to control for urban environment, weather, and other contextual variables. Variations were made based on different types of bicycle facilities, the presence or absence of on-street parking, and the number of vehicular lanes. Each scenario exhibited a moderate amount of vehicular traffic that would allow for near-free-flow conditions with a reasonable amount of opportunity for vehicle-to-cyclist interactions. The images were designed such that the background scenery would be recognizable by both urban dwellers as an in-town neighborhood and rural dwellers as a small town. Nineteen total images were prepared. The images for on-street facilities are presented in Figure 2.5. These facility types included sharrows, bike lanes, and buffered bike lanes. For scenarios with a parking lane, both parking-side buffers and traffic-side buffers were included. Each facility treatment was shown for four roadway sections: two-lane, two-lane with parking, four-lane, and four-lane with parking. All four images with a sharrow marking were presented and discussed first, followed by all bike lane images, then buffered bike lane images. After this set of images, separated bicycling facility images, including protected/separated bike lanes and a shared-use path, were shown (Figure 2.6). In the interest of time, not all lane and parking configurations were shown for the protected/separated facilities. Protected/separated bike lane scenarios were built using the same urban environment as before, but the nature of a shared-use path required a separate built environment. Participants were invited to indicate their comfort level with riding in each presented environment as very comfortable, somewhat comfortable, somewhat uncomfortable, or very uncomfortable. Participants were invited to explain their reasoning and express their concerns about each scenario. Although the focus groups were intended to center around the concept of route-based attri- butes, it was anticipated that many respondents would want to express concerns about intersec- tions as well. An additional set of images about bike boxes, two-stage turn queues, and protected intersections was presented, although this section was briefer. Respondents generally liked the more protected intersections but had not experienced enough alternatives to share substantive opinions about the infrastructure. Additional images were presented of green bike lanes and neighborhood greenways (residential streets, sometimes called bike boulevards, using traffic- calming measures to give preference to bikes) but again, these images were not intended as much to solicit opinion as to generate general discussion. Participants liked the more designated facilities that showed where bikes belonged, but neighborhood greenways were not given enough time to solicit a response. Wave 1 Survey The predominant study method centered on a two-wave panel survey, the first wave of which was distributed in Summer and Fall 2016 to a sample of residents living in the study areas. The focus groups allowed an exploration of attitudes and obstacles to cycling in a variety of settings and provided a forum to pretest images and descriptions of various bicycle facilities to determine which are most effective in helping potential users evaluate their likelihood of using the proposed
Not Applicable Four Lanes, No Parking Not Applicable Sharrow Bike Lane Buffered Bike Lane (Traffic-side Buffer) Buffered Bike Lane (Parking-side Buffer) Two Lanes, No Parking Two Lanes, On-street Parking Four Lanes, On-street Parking Figure 2.5. Combinations of on-street bicycle facilities used in focus groups.
Study Methodology 17 infrastructure. On completion of the focus groups, the first-wave survey was finalized and distributed. Although the ultimate goal of the first-wave survey was to serve as a baseline for comparison of second-wave measures, the survey was also designed to provide quality cross-sectional data that allow this phase to stand on its own. Key dependent variables measured in the survey include preference for facility types and revealed amounts of actual cycling. The survey was also designed to control for several explanatory variables, including â¢ Individual sociodemographic characteristics; â¢ Personal attitudes, personality traits, lifestyles and preferences; â¢ Household characteristics and living arrangements; and â¢ Current travel behavior patterns for both commuting and leisure trips. Addresses in each of the six study locations were purchased from DirectMail. In late Summer 2016, residents received postcard invitations that included a URL to take the survey online. In early Fall 2016, those who had not responded were mailed a printed version of the survey along with a prepaid return envelope. The survey was designed through extensive writing, debating, and rewriting over six months to identify and refine survey questions. The goal was to produce a survey instrument that took no more than 30 minutes to complete. This allowed a balance between gathering a thorough data set and asking a limited time commitment from participants. To reduce potential response biases, the survey content was purposefully broader than cycling to ensure that participants remained interested and did not quit the survey if they did not recognize themselves as the âbicycling type.â To the extent practical, questions from previous surveys were reused, both to Two Lanes, On-street Parking, Planter-separated Two-way Bike Lane Shared Use Path Four Lanes, On-street Parking, Planter-separated Bike Lane Two Lanes, No Parking, Planter-separated Bike Lane Four Lanes, On-street Parking, Bollard-separated Bike Lane Figure 2.6. Combinations of physically separated bicycle facilities used in focus groups (shared-use path image courtesy of Atlanta BeltLine Inc.).
18 Bicyclist Facility Preferences and Effects on Increasing Bicycle Trips rely on previously tested and vetted questions and to maximize opportunities for cross-study comparisons of results. The resulting survey contained six sections: A. Attitudes B. Technology usage C. Home D. Daily travel E. Bicycling experience F. Demographics Feedback from the focus groups was used to edit the bicycle infrastructure images. Sixteen images were prepared with all combinations of four bike-facility classes (sharrows, bike lanes, buffered bike lanes, and protected/separated bike lanes), presence or absence of on-street parking, and two versus four traffic lanes. The base background image used to create the images for the focus groups was reused for the survey, though it was edited to represent springtime. It was impractical to ask each respondent to rate all 17 images, so four different versions of the survey were prepared, using a modified factorial design that gives each respondent six images to evaluate. Each respondent was presented with one image from each of the four types of on-street facilities (sharrows, bike lanes, buffered bike lanes, and protected/separated bike lanes) for the same roadway characteristics, as well as at least one additional image from among those four types. The additional image differed either in whether parking was present or not, or in whether the street was two lanes or four. The sixth image was either another âdoubleâ from among the four facility types or one that portrayed a multi-use path as shown in Figure 2.7. These combinations ensured that across the entire sample, specific comparisons of interest could be made. All 17 images were tested in the focus groups and some modifications were applied. Figure 2.8 displays the images used for the 16 on-street infra- structure configurations. The survey was pretested with graduate students, the NCHRP panel, and members of the public. Both an online version and a paper version were prepared. All four versions of the final survey are attached to this report in Appendix B, and are available by contacting the study team. The survey is intended to be generic enough to use across the country for future comparison of results. The survey was initially deployed in August 2016 and responses were collected throughout late summer. A letter printed on cardstock was sent to about 24,000 possible participants with instructions to either return a postcard for a hard copy of the survey or to complete it online using a code. E-mail addresses and a toll-free number were established to answer questions. Figure 2.7. Image for multi-use paths used in survey.
Figure 2.8. Images of infrastructure configurations for different roadway layouts used in survey. Cycletrack Buffered Bike Lane Bike Lane Sharrow Two Lanes, No Parking Two Lanes, On-street Parking Four Lanes, No Parking Four Lanes, On-street Parking
20 Bicyclist Facility Preferences and Effects on Increasing Bicycle Trips Incentives of either a free treat at a local store or a $2 bill were offered to those who responded. Unfortunately, despite the best efforts of the team, the responses were far fewer than anticipated, with only 175 online responses and 276 postcards returned. In retrospect, the double barrier of requiring respondents to either return a postcard or visit a survey online (in turn requiring a computer and Internet connection) may have been too high. To remedy this problem, the entire survey was printed and mailed to the full list of residents who had not yet responded to the survey in October 2016. It was hoped that by receiving an actual paper copy after the initial letters, the residents would be more likely to respond to the survey. The 1-800 number and e-mail address were still used to respond to survey questions, and responses were assembled and entered from November to February. Each paper survey was entered (coded) twice, and the two data sets were compared to ensure no coding errors were introduced in data entry. Wave 2 Survey The data collection with the second-wave survey commenced in Spring 2018. All respondents who completed the first-wave survey were invited to complete the follow-up second-wave online or mail-back survey. For the treatment communitiesâthose undergoing the addition of new bicycle infrastructureâthis served as the after survey, which was used to measure any changes in travel behavior reported after the opening of the new planned bicycle infrastructure. For the other communities, the survey acted as a control for background changes in attitudes and demographics that may be confounded with the influence of the new infrastructure. This before-and-after-with-control-group approach is considered to be a robust quasi-experimental design that protects against several common threats to validity, providing strong evidence for the impacts of various infrastructure improvements on cycling behavior. The structure of the second survey was similar to that of the first survey. The survey was shortened somewhat to ease the burden placed on the respondents, but many questions were repeated verbatim to provide identical measurements for two points of time. Examples include bicycling distances and frequency, from which changes in behavior can be inferred. Additionally, respondents were shown two new questions that asked for their perceptions and use of any new infrastructure improvements. These questions were designed to measure perceived changes that could be self-contained in the second wave. The survey was pretested with graduate students and members of the public. Both an online version and a paper version were prepared. The resulting survey (which can be found in Appendix C) was nine pages, took approximately 20 minutes to complete, and contained four sections: A. Attitudes B. Daily travel C. Bicycling experience D. Demographics With the ever-changing nature of some transportation systems, the researchers wanted to gauge the general perceptions of changes in each neighborhood, including for automobiles, transit, walking, and bicycling. This survey design also helped the research team avoid leading respondents about specific changes, and provided a reasonable basis for comparing perceptions of bicycle infrastructure. A general question on perceptions of changes in transportation in the community was included to fulfill this purpose, with statements relating to perceived changes in quality of roadways, transit, bicycling, walking, and ridehailing.
Study Methodology 21 In addition to general perceptions, the research team also wanted to measure recognition of changes in bicycle facilities. The goal was to measure recognition of the addition of any bicycling facility, as well as to properly identify what facility was added. Recognition in treatment sites was compared with that in the respective control sites (which had not received bicycling facilities during the study period). Parallel to the questions of recognition, the researchers also asked respondents whether they have used the new bike facilities and if they like them. The differences in the responses between treatment and control pairs were to be noted. The second-wave survey was deployed in May 2018 and responses were collected throughout the summer, with the exception of Chattanooga and Birmingham, where, because of delays in the Chattanooga projects of interest, the survey deployment was intentionally postponed until fall. The invitation list for the second-wave survey was composed of all respondents from the first wave. Respondents were again offered incentives of $2 bills (as obtaining coupons for free treats was difficult in some locations) for repeating this wave of the survey. Printed versions of the survey were mailed to all on the list. Additionally, e-mail invitations with a URL to take the survey online were sent to all subjects who had provided an e-mail address. As with the first- round survey, the research team provided a 1-800 number and an e-mail address to field ques- tions or comments from respondents. Each paper survey was entered (coded) twice, and the two data sets were compared to ensure no coding errors were introduced in data entry.