Bicyclist Facility Preferences and Effects on Increasing Bicycle Trips (2020)

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.

34 Expressed Comfort, Safety, and Willingness to Try Cycling One goal of the first-wave survey was to measure preference for facility types with regard to comfort, perceived safety, and a willingness to try cycling. This chapter explains these preferences through average responses and regression models. Regression models presented include infra- structure characteristics, as well as controls for sociodemographics, rider types, and attitudes. Analysis of Expressed Comfort, Safety, and Willingness to Try Cycling The images presented to respondents were created in Adobe Photoshop using one common roadway setting as a base image to control for urban environment, weather, and other con- textual variables. Variations were based on different types of bicycle facilities, the presence or absence of on-street parking, and the number of automobile lanes (one versus two in each direc- tion). Each scenario exhibited moderate automobile traffic that would allow for near-free-flow conditions with a reasonable opportunity for automobile-to-cyclist interactions. The images were designed such that urban dwellers could relate to the background scenery as an in-town neighborhood and rural dwellers, as a small town. Seventeen total images were prepared, shown in Figure 5.1 and Figure 5.2. The bicycling facilities include sharrows, bike lanes, buffered bike lanes, and barrier-protected bike lanes. Two of the protected/separated bike lanes were one-way, while the other two were two-way. An image for a multi-use path was also created, though owing to the nature of this type of infrastructure, a different road environment had to be used. For each image, respondents were given the prompt: “Bicycling on a road [trail] like this is . . .,” with the sentence being completed in each of three ways (perceptions): “comfortable,” “safe,” and “something I’d try.” For each perception, they were asked to choose the most appro- priate response on a five-point Likert-type scale (completely disagree, disagree, neutral or no opinion, agree, or completely agree). Respondents were randomly assigned one of four versions of the survey, each of which had a different combination of infrastructure images. Each version had a base road configuration (e.g., two lanes with on-street parking, four lanes with no park- ing) for which a sequence of all four on-street facilities was shown. Thus, each version had four “in-sequence” configurations for the sharrow, bike lane, buffered bike lane, and protected/separated bike lane, each sharing a common lane configuration. Two other images were also included, from among the other road configurations and multi-use trails, so that each respondent was presented with six infrastructure combinations, and several were repeated between surveys. Figure 5.3, Figure 5.4, and Figure 5.5 show the distribution of responses for comfort, safety, and willingness to try, respectively. These figures are grouped so that for each lane combina- tion, each row is progressively more separated from traffic. The agreement with each perception C H A P T E R 5

Version 1 (Two Lanes with Parking) Sharrow Sharrow Two Lane No Parking Bike Lane Bike Lane Bike Lane Buffered Bike Lane Two Lane No Parking Buffered Bike Lane Four Lane No Parking Buffered Bike Lane Buffered Bike Lane One-way Protected Bike Lane Two-way Protected Bike Lane Multi-Use Path In-sequence image Version 2 (Two Lanes, no Parking) Figure 5.1. Combinations of bicycle infrastructure used in survey versions 1 and 2.

Version 3 (Four Lanes with Parking) Sharrow Sharrow Two Lane Parking Bike Lane Bike Lane Bike Lane Four Lane Parking Bike Lane Buffered Bike Lane Buffered Bike Lane One-way Protected Bike Lane Two-way Protected Bike Lane Multi-Use Path Multi-Use Path In-sequence image Version 4 (Four Lanes, no Parking) Figure 5.2. Combinations of bicycle infrastructure used in survey versions 3 and 4.

Expressed Comfort, Safety, and Willingness to Try Cycling 37 Note: The number in parentheses is the number of responses for the associated configuration. SH = sharrow, BL = bike lane, BB = buffered bike lane, 1C = one-way protected cycletrack, and 2C = two-way protected cycletrack. Figure 5.3. Distribution of agreement to statement “Bicycling on a road like this is comfortable” for each image. Figure 5.4. Distribution of agreement to statement “Bicycling on a road like this is safe” for each image. Note: The number in parentheses is the number of responses for the associated configuration. SH = sharrow, BL = bike lane, BB = buffered bike lane, 1C = one-way protected cycletrack, and 2C = two-way protected cycletrack.

38 Bicyclist Facility Preferences and Effects on Increasing Bicycle Trips clearly increases with each degree of separation from traffic, decreases with the addition of on-street parking, and is much subtler for the number of lanes, though more rigorous analysis is necessary to address inconsistencies. The perceived levels of comfort, safety, and willingness to try the presented infrastructure were converted to numeric values, with “completely disagree” equal to 1 and “completely agree” equal to 5. The average ratings for comfort, safety, and willingness to try are presented in Figure 5.6, Figure 5.7, and Figure 5.8, respectively, for all scenarios with on-street facilities Note: The number in parentheses is the number of responses for the associated configuration. SH = sharrow, BL = bike lane, BB = buffered bike lane, 1C = one-way protected cycletrack, and 2C = two-way protected cycletrack. Figure 5.5. Distribution of agreement to statement “Bicycling on a road like this is something I’d try” for each image. Note: Axis does not start at 0. Figure 5.6. Average expressed comfort levels for each lane/parking configuration by bicycle facility type.

Expressed Comfort, Safety, and Willingness to Try Cycling 39 (i.e., multi-use paths are not included in these figures). As mentioned previously, each version of the survey focused on the continuum of four bicycling facility types within the same traffic lane and parking lane combination, plus two additional images duplicated from the other survey versions. To avoid the potential framing effects introduced by the insertion of these additional images “out of sequence,” only the responses for the four in-sequence images are included in the descriptive analysis presented here (sample sizes of between 266 and 308 responses for each mean). However, all responses were included in models, which were designed to account for these effects, as explained in the following section. The characteristics of the bicycle infrastructure portion of the roadways for the sharrow, bike lane, and buffered bike lane cases were consistent between roadway configurations. However, protected/separated bike lanes had two variations, one way and two way, only one of which was presented for a given configuration to limit the number of images presented. The broken Note: Axis does not start at 0. Figure 5.7. Average expressed safety levels for each lane/parking configuration by bicycle facility type. Note: Axis does not start at 0. Figure 5.8. Average expressed level of willingness to try for each lane/parking configuration by bicycle facility type.

40 Bicyclist Facility Preferences and Effects on Increasing Bicycle Trips lines on the graphs show the point in the progression of bicycle facility at which barrier protec- tion is introduced, and two different protected/separated bicycle infrastructure types are por- trayed. The two-lane with no parking and four-lane with parking configurations had one-way protected/separated bike lanes (indicated by the dotted lines), while the four-lane with no park- ing and two-lane with parking images had two-way protected/separated bike lanes (indicated by the dash-dot lines). Given the close clustering of the four means for this facility type, the figures indicate that the differences in ratings between protected/separated bike lane scenarios may be unrelated to roadway characteristics. Ratings for comfort, safety, and willingness to try tended to follow similar patterns. However, the mean ratings exhibit some small differences, indicating that, at least to some extent, respon- dents were able to distinguish between the different perceptual dimensions (comfort versus safety versus willingness to try) for each image. For example, the protected/separated facilities had mean responses up to 4.2 for safety, but only 4.1 for comfort and 3.9 for willingness to try. On the other hand, facilities with sharrows had the lowest mean responses for safety (2.4) followed by willingness to try (2.6) and comfort (2.7). The implications are that the ranges for perceived safety are somewhat larger, with the greatest apparent differences between the least protected and the most protected facilities being for safety, while comfort tends to be slightly higher on average. Comfort, safety, and willingness to try all improved for each increased degree of separation provided by the bicycling facilities, indicating a positive benefit associated with separation from moving and parked cars. Each version of the survey began the infrastructure image section with a sharrow configuration, which allows the sharrow infrastructure layouts to serve as a base mea- surement for each lane configuration. In each version, the sharrow configurations received the lowest ratings, and the existence of any sort of spatial separation was influential in increasing each perception measure. Average ratings for each traditional bike lane scenario were higher than those for sharrows on the same roadway configuration. The difference is more pronounced for bicycle lanes without adjacent curb parking, which supports the earlier focus group finding of the disutility of combining bike lanes with on-street parking. Buffered bike lanes received higher average ratings than traditional bike lanes, and saw the same disutility with parking lanes. As previously mentioned, two different protected/separated bike lane scenarios were tested in the survey. Table 5.1 shows the average ratings for each protected/separated bike lane scenario along with the multi-use path. As shown previously, the presence of the barrier was effective in overcoming the obstacles created by the inclusion of parking or extra traffic lanes. Focus group participants suggested that one-way protected/separated bike lanes would be preferable to two- way lanes. The primary determinant for preferring protected/separated bike lanes would be based on whether the lane is one-way or two-way. In contrast, parking and traffic lane character- istics were more influential in shaping perceptions of the more vulnerable layouts. The multi-use One-way Protected Two-way Protected Multi-use Path Two-Lane/ No Parking Four-Lane with Parking Two-Lane with Parking Four-Lane/ No Parking Comfort 4.12 4.13 3.89 3.97 4.14 Safety 4.14 4.24 3.89 4.01 4.15 Willingness to Try 3.78 3.89 3.63 3.68 3.95 Table 5.1. Average agreement ratings for comfort, safety, and willingness to try for protected/separated bike lanes and multi-use paths (from 1 = completely disagree to 5 = completely agree).

Expressed Comfort, Safety, and Willingness to Try Cycling 41 path received ratings comparable to those of the one-way and two-way protected/separated bike lanes. These observations are among those statistically tested in the following section. Regression Models While the preceding descriptive analysis is useful, it is also desirable to control for covariates whose effects might otherwise be confounded with those of facility type and roadway configura- tion. Linear regression models were built using the multiple responses by 1,178 respondents for each of the three dependent variables (comfort, safety, and willingness to try), as presented in Table 5.2. Dummy variables for each infrastructure type, along with the presence of on-street parking and additional lanes of traffic, were included in the models. Cluster-robust standard errors were used to account for correlations of unobserved characteristics influencing the mul- tiple responses from each respondent. Although linear regression models have limitations for application to Likert-type data, they can serve as a reliable approximation with four or more ordinal response levels with “little worry” (Bentler and Chou 1987). Much empirical research over the years has used Likert-type data with parametric methods such as regression, and a review of the progression of this research assures scholars that such methods can be employed in these cases “with no fear of ‘coming to the wrong conclusion’” (Norman 2010). Therefore, the linear regression model was carried further because of easier interpretability. Appendix E includes more detail about alternative model testing. An issue resultant from the survey design was the emergence of a framing effect. Each ver- sion of the survey had a logical sequence of four images based on a common lane configuration, along with two out-of-sequence images. Each out-of-sequence image, which was a repeat of an Variable Comfort Safety Willingness to Try Coefficient P Coefficient P Coefficient P Constant 2.90 *** <0.001 2.62 *** <0.001 2.82 *** <0.001 Bicycle Facility Types Bike Lane (BL) 0.37 *** <0.001 0.45 *** <0.001 0.30 *** <0.001 Buffered BL (BBL) 0.73 *** <0.001 0.89 *** <0.001 0.57 *** <0.001 One-way Protected 1.34 *** <0.001 1.68 *** <0.001 1.12 *** <0.001 Two-way Protected 1.16 *** <0.001 1.45 *** <0.001 0.96 *** <0.001 Multi-use 1.24 *** <0.001 1.53 *** <0.001 1.12 *** <0.001 Roadway Characteristics Parking −0.27 *** <0.001 −0.26 *** <0.001 −0.17 *** <0.001 Four Lanes 0.02 0.477 0.05 0.103 −0.02 0.500 Framing Effects BL—No Parking 0.42 *** <0.001 0.50 *** <0.001 0.41 *** <0.001 BBL—No Parking 0.22 *** <0.001 0.33 *** <0.001 0.22 ** 0.002 BL—Two Lanes 0.28 *** <0.001 0.35 *** <0.001 0.22 * 0.015 # of Responses 6,743 6,723 6,664 R2 0.175 0.232 0.093 *P < 0.05, **P < 0.01, ***P < 0.001 Table 5.2. Linear regressions for expressed comfort, safety, and willingness to try, with infrastructure characteristics only.

42 Bicyclist Facility Preferences and Effects on Increasing Bicycle Trips image displayed in another version of the survey, appeared either before or after the most con- ceptually similar image of the sequence. Five roadway images were repeated in another version (bike lane with two automobile lanes and no parking, buffered bike lane with two automobile lanes and no parking, buffered bike lane with four automobile lanes and no parking, bike lane with two automobile lanes and parking, and bike lane with four automobile lanes and parking). Each image received different responses based on the version in which it appeared. Specifically, these images attracted different responses when they were out of sequence (e.g., the “two-lane/ no parking bike lane” image in Version 1 of Figure 5.1) than when they were in sequence (the same image in Version 2). The multi-use path appeared in three versions and had consistent scores in each version. Dummy variables were included in the regression to capture the variation because of the framing effects introduced by the interruption of the natural sequence of each version. Most images, when compared with the preceding image, changed only one variable (bike-facility type, parking, or automobile lanes). Conversely, each time the sequence is broken, two variables must be changed at once, either to break the sequence or to return to the sequence. For example, the bike lane with two automobile lanes and parking in Version 3 (in Figure 5.2) breaks the sequence, changing the number of automobile lanes (from four to two) and the bike-facility type (from sharrow to bike lane) from the previous image; however, the (out-of-sequence) bike lane with four automobile lanes and parking in Version 4 only changes one variable from the preceding image (the addition of parking), while the subsequent image changes two variables at once, the change of bike lane to buffered bike lane and the removal of parking. Three dummy variables were created and applied to the appropriate images when their appearance involved changing two variables at once: Bike Lane (BL)—No Parking, Buffered Bike Lane (BBL)—No Parking, and BL—Two Lanes. The BL—No Parking variable was set to 1 for the second image in Version 1, which added a bike lane and removed parking compared with the preceding image; the BBL—No Parking variable was set to 1 for the two-lane buffered bike lane image in Version 1 along with the four-lane buffered bike lane in Version 4, both of which added a buffer to the bike lane and removed parking compared with the preceding image; and the BL—Two Lanes variable was set to 1 for the second image in Version 3, which introduced a bike lane and removed the additional lanes of traffic compared with the preceding image. A fourth dummy variable was also considered for the two-lane one-way protected/separated bike lane without parking image in Version 2, however this variable was eventually excluded because it undermined the stability of the model. This instability was perhaps because of empirical collinearity issues related to one- way protected/separated bike lanes appearing in only one version of the survey. The dummy variables for each facility type were significant. The coefficients for each on-street facility variable (BL, BBL, and Protected/Separated Lanes) were also significantly different from each other, supporting the earlier finding that greater separation of cyclists from cars increases all three measures of effectiveness. The multi-use dummy coefficient was not substantially different from the protected/separated bike lane coefficients; however, it was still included separately in the model because the multi-use images excluded the effects of roadway characteristic variables. The framing effect terms were significant in each model. These variables showed sensitivity to the comparative removal of a perceived negative aspect (parking or an additional travel lane) that is not explained by the variables indicating the absence of that aspect alone. For example, when an image without parking was presented after an image with parking, it tended to receive a higher rating than if it were preceded by an image that also had no parking. While the framing variables picked up the influence of multiple simultaneous changes from image to image, the Parking and Four Lanes variables represented the effects of roadway char- acteristics. The parking variable was significant in all models, indicating that the overall effect of parking was still significant, even after accounting for the strong impact of the removal of

Expressed Comfort, Safety, and Willingness to Try Cycling 43 parking in the few images affected by framing. Interestingly, the variable for the number of traffic lanes alone was not significant in any of the models. This is consistent with the ambigu- ous influence of number of lanes expressed by focus group participants: roads with more lanes generally contain traffic with higher speeds, which is a negative, but also potentially gives drivers more flexibility to avoid cyclists, which is a positive. However, the significance of the BL—Two Lanes framing variable indicates at least a (positive) situational effect when the number of lanes in the figure changes from four to two. Regression Models with Sociodemographic Data Sociodemographic data were also collected using the survey instrument. The previous linear models were supplemented with sociodemographic data, as presented in Table 5.3. As explained previously with regard to imputing data, for the few cases in which this information was not reported, data obtained from targeted marketing sources were used as an estimate. In all three models—comfort, safety, and willingness to try—age and education were significant, with con- sistent signs between models. However, both coefficients were comparatively larger in the will- ingness to try model. Older individuals tended to express lower perceived comfort and safety, and even more so for willingness to try. Individuals with higher levels of education tended to express greater perceived comfort and safety, and even more so for willingness to try. The number of vehicles per licensed driver (at the household level, capped at 1.0) was sig- nificant in the comfort and willingness to try models. This variable measures individuals’ access to an automobile in their homes, and its negative coefficients indicate that those with greater access tend to view a given infrastructure as less comfortable and as something they would be less willing to try. Variable Comfort Safety Willingness to Try Coefficient P Coefficient P Coefficient P Constant 3.09 *** <0.001 2.55 *** <0.001 3.59 *** <0.001 Bicycle Facility Types Bike Lane (BL) 0.40 *** <0.001 0.47 *** <0.001 0.32 *** <0.001 Buffered BL (BB) 0.77 *** <0.001 0.90 *** <0.001 0.59 *** <0.001 One-way Protected 1.39 *** <0.001 1.69 *** <0.001 1.15 *** <0.001 Two-way Protected 1.21 *** <0.001 1.47 *** <0.001 1.03 *** <0.001 Multi-use 1.30 *** <0.001 1.55 *** <0.001 1.19 *** <0.001 Roadway Characteristics Parking −0.27 *** <0.001 −0.25 *** <0.001 −0.16 *** <0.001 Four Lanes 0.03 0.477 0.04 0.103 −0.03 0.441 Framing Effects BL—No Parking 0.41 *** <0.001 0.50 *** <0.001 0.44 *** <0.001 BB—No Parking 0.23 *** <0.001 0.34 *** <0.001 0.26 *** <0.001 BL—Two Lanes 0.26 *** <0.001 0.31 *** <0.001 0.19 * 0.038 Sociodemographics Age (in decades) −0.04 *** <0.001 −0.04 *** <0.001 −0.10 *** <0.001 Education 0.04 *** <0.001 0.03 ** 0.001 0.09 *** <0.001 Vehicles Per Driver −0.16 ** 0.003 −0.38 *** <0.001 Driver’s License 0.18 *** <0.001 Child in Home −0.08 * 0.033 Female −0.29 *** <0.001 African American −0.08 * 0.047 # of Responses 6,159 6,529 6,086 R2 0.201 0.248 0.153 *P < 0.05, **P < 0.01, ***P < 0.001 Table 5.3. Linear regression for expressed comfort, safety, and willingness to try by infrastructure and sociodemographic characteristics.

44 Bicyclist Facility Preferences and Effects on Increasing Bicycle Trips The coefficients for driver’s license and child in home were significant only in the safety model. The positive coefficient for driver’s license may indicate that those with a license feel more control over the safety of the roadway in general. The child in home coefficient was negative, which indicates that those who have children in their homes tend to view bicycling infrastructure as less safe than those who do not. This could be the result of considering cycling with their children or of an increased attention to safety owing to the responsibilities of raising children. Coefficients on the female and African American variables were significant only in the willing- ness to try model. Their significance and negative values indicate that these two subsets of the population may be less willing than others, all else equal, to try a given infrastructure configura- tion, and may serve as the basis for further analysis. Sociodemographic characteristics seemed to play a larger role in the willingness to try model than for the other two perceptions, as seen by the 0.06 increase in the R2 value from 0.093 (Table 5.2) to 0.153 (compared with increases of 0.026 and 0.016, respectively, for the other two models). This indicates that individual characteristics are more influential to potential users’ stated inclinations to use a certain type of infrastructure than to their perceptions of whether it is comfortable or safe in general. Rider Segmentations and Segmented Models Another key aspect of understanding infrastructure development and the potential to attract new cyclists is the riders that would be attracted to various infrastructure types. A segmented model was developed to investigate how the influence of the other explanatory variables differs by rider group. The criteria for inclusion in one of these categories comes from responses to questions regarding bicycling confidence, cycling distances for recreation/utilitarian purposes, and cycling trip frequency for commuting/other purposes. There are four segments and their criteria are 1. Potential cyclist (N = 700)—those who report zero miles of cycling per month, but report being able to ride a bike, regardless of confidence level. 2. Recreational cyclist (N = 166)—those who bike a nonzero distance per month, but bike less than once a month and less than a mile a week, on average, for utilitarian purposes. 3. Utilitarian cyclist (N = 84)—those who bike at least once a month or at least a mile a week, on average, for utilitarian purposes. 4. Cannot bike (N = 163)—those who state that they cannot ride a bicycle. The “potential cyclist” group was chosen for the base, as this is the largest group and the target population. The segmented models are presented here as main effects and incremental- difference effects for segments with coefficients significantly different from the base group. This means that the main-effects section of the models can be used to describe potential cyclists’ responses, with the incremental effects describing deviations between potential cyclists and other segments. Not all segments were significantly different from the base in each model. Each segmented model started from the previously reported regression models for comfort, safety, and willingness to try, respectively. Dummy variables were introduced for the “recre- ational,” “utilitarian,” and “cannot bike” segments, using the “potential cyclists” as the base. The incremental effects for each segment were estimated using interaction terms between the main effect explanatory variables and the segment dummy variables, piecewise removing insig- nificant variables (constraining them to be 0). Insignificant variables were included in cases with borderline significance, in which a main effect was insignificant but an associated interaction effect was significant, and in cases in which the coefficient is necessary for interpretation of a similar variable, such as for different types of bicycle facilities. Although the addition of multiple

Expressed Comfort, Safety, and Willingness to Try Cycling 45 interaction terms has the potential to introduce collinearity into regression models, the effect in this case is not much cause for concern, given the ability of the large sample to minimize the potential inflations in variance brought on by collinearity. The segmented models for expressed comfort, safety, and willingness to try are presented in Table 5.4 and Table 5.5. The primary differences uncovered by the expressed comfort model are the incremental effects for utilitarian cyclists. Compared with the rest of the population, utilitarian cyclists were less likely to express discomfort because of the presence of parking. Age is a net-positive coeffi- cient for the utilitarian group, implying that older utilitarian cyclists are more likely than others to rate infrastructure as comfortable. Like the previous model, most of the differences in expressed safety come from the utili- tarian group. Each bicycling facility variable is positive for utilitarian cyclists, indicating that although all groups see each added degree of protection as an increase in safety, the group that cycles most perceives an even greater increase in safety. The parking coefficient was posi- tive for utilitarian cyclists, with a similar magnitude to the (negative) base parking coefficient, indicating that utilitarian cyclists do not view on-street parking as significantly unsafe like the rest of the sample does. The coefficients for the variable measuring the presence of children in the home for the utilitarian and unable groups were significantly (or borderline significantly) negative, while the base coefficient became insignificant in this model, indicating that the nega- tive impact on perceived safety associated with the presence of a child in the home is driven by these two groups. Constant 3.14 *** <0.001 2.64 *** <0.001 3.74 *** <0.001 Recreational (2) 0.17 *** <0.001 0.15 *** <0.001 −0.65 ** 0.004 Utilitarian (3) −0.54 * 0.012 −0.86 *** <0.001 −0.20 0.428 Unable (4) −0.09 * 0.031 −0.01 0.858 −1.89 *** <0.001 Bicycle Facility Types Bike Lane (BL) 0.40 *** <0.001 0.47 *** <0.001 0.32 *** <0.001 Buffered BL (BB) 0.76 *** <0.001 0.90 *** <0.001 0.59 *** <0.001 One-way Protected 1.39 *** <0.001 1.72 *** <0.001 1.15 *** <0.001 Two-way Protected 1.22 *** <0.001 1.50 *** <0.001 1.02 *** <0.001 Multi-use 1.30 *** <0.001 1.56 *** <0.001 1.19 *** <0.001 Roadway Characteristics Parking −0.29 *** <0.001 −0.28 *** <0.001 −0.21 *** <0.001 Four Lanes 0.02 0.438 0.05 0.099 −0.05 0.152 Framing Effects BL—No Parking 0.41 *** <0.001 0.51 *** <0.001 0.44 *** <0.001 BB—No Parking 0.24 *** <0.001 0.36 *** <0.001 0.25 *** <0.001 BL—Two Lanes 0.24 ** 0.001 0.29 *** <0.001 0.18 * 0.043 Sociodemographics Age (in 10s of years) −0.03 *** <0.001 −0.04 *** <0.001 −0.09 *** <0.001 Education 0.03 ** 0.002 0.02 * 0.032 0.03 * 0.012 Vehicles per Driver −0.23 *** <0.001 −0.17 ** 0.007 −0.48 *** <0.001 Child in Home −0.05 0.236 Driver’s License 0.22 * 0.015 Female −0.19 *** <0.001 African American −0.16 *** <0.001 P < 0.1, *P < 0.05, **P < 0.01, ***P < 0.001 Variable Comfort Safety Willingness to Try Coefficient P Coefficient P Coefficient P Table 5.4. Linear regression for expressed comfort, safety, and willingness to try including incremental effects of cyclist segments (main effects).

46 Bicyclist Facility Preferences and Effects on Increasing Bicycle Trips For the willingness to try model, the only roadway characteristics to be significant in any segmentation were the Parking and Four Lanes variables for those unable to bike. Both were positive, with higher magnitudes than the negative base coefficients. This segregation likewise allows the coefficient for the rest of the population to be more negative for the parking vari- able, while the four-lane variable inches closer to significantly negative. This implies that the stated responses of those who cannot bike may contradict those of the rest of the population in willingness to try cycling in the presence of parking and additional traffic lanes. Although the change in sign for these coefficients may seem unexpected, the rather large magnitude of the negative constant term indicates that this group is still substantially less willing to try cycling in comparison with the other groups. The coefficients for age are all significant and have similar magnitudes, with only the base being negative. This indicates that age is a deterrent for those in the potential cyclist group but does not have a significant effect among the recreational, utilitarian, and unable groups. The main-effects portion of the models is similar to the previous models from Table 5.3, though with a few adjustments. The difference is that the incremental effects help correct for major discrepancies for particular segments, allowing the main effects to account for the rest of respondents. In particular, the coefficient for vehicles per licensed drivers got more negative in each model, with the difference being enough to make the variable significant in the safety model. Such significance indicates that by allowing the coefficient of utilitarian cyclists to differ, the influence of access to vehicles is an even stronger deterrent for perceptions. Likewise, the coefficients for parking are more negative in the main-effects model after accounting for dif- ferences from the utilitarian group and the group that cannot bike. The coefficient for African Americans also became more negative after accounting for the influence of those who cannot Variable Comfort Safety Willingness to Try Incremental Effects Coefficient P Coefficient P Coefficient P Utilitarian Segment (3)*Bike Lane (BL) 0.31 * 0.041 (3)*Buffered BL 0.36 * 0.026 (3)*One-way Protected 0.44 * 0.030 (3)*Two-way Protected 0.40 0.072 (3)*Multi-use 0.59 ** 0.002 (3)*Parking 0.20 * 0.046 0.29 ** 0.009 (3)*Age (in 10s of years) 0.09 ** 0.008 0.06 0.095 0.09 * 0.032 (3)*Vehicles/Driver 0.44 ** 0.010 0.44 * * * 0.014 0.50 * 0.016 (3)*Child in Home −0.27 * * 0.036 Cannot Bike Segment (4)*Child in Home −0.25 0.082 (4)*Parking 0.37 *** <0.001 (4)*Four Lanes 0.24 * 0.014 (4)*Age (in 10s of years) 0.09 * 0.015 (4)*African American 0.62 *** <0.001 (4)*Vehicles/Driver 0.40 * 0.026 Recreation Segment (2)*Age 0.008 ** 0.009 (2)*Education 0.15 *** <0.001 # of Responses 6,038 5,982 5,966 R2 0.212 0.268 0.206 Adj R2 0.210 0.265 0.203 P < 0.1, *P < 0.05, **P < 0.01, ***P < 0.001 Table 5.5. Linear regression for expressed comfort, safety, and willingness to try including incremental effects of cyclist segments (incremental effects).

Expressed Comfort, Safety, and Willingness to Try Cycling 47 bike, who were also overrepresented in that group. Interestingly enough, even though the coeffi- cient for females was not significantly different for any segment in any models, the coefficient in the main effects shifted closer to zero from -0.29 to -0.19. This may be a result of the correction of other effects associated with gender. Analysis of Attitudes While the impact of sociodemographics and rider type on perceptions and preferences is an indication that individual characteristics are significant, many other characteristics are not as easily observed. This section uses information gained from 38 attitudinal questions regarding mode and travel perceptions, technology, time use, and other concepts to discuss the impact of attitudes on bicycle facility perceptions and preferences. Factor Analysis The attitudinal questions in the survey were intentionally written so that several items would pertain to various aspects of a single construct. A standard method called factor analysis was used to combine related items, which allows a given attitudinal construct to be more robustly mea- sured than would be possible with a single item alone. Before the factor analysis, small amounts of missing data (five or fewer items per subject) were imputed with expectation maximization. The complete data set was then analyzed with exploratory factor analysis. A key decision in exploratory factor analysis is the number of factors, or attitudinal constructs, to select. Solutions with 8 to 13 factors and all 38 items were considered, using guidelines (including for interpretability and parsimony) found in the literature. Items with weak loadings were considered for removal. A key tool for interpreting the meaning of the factors is the factor pattern loading matrix, in which the cell represents the degree of association of the item (row) with the factor (column). The final (rotated) factor loading matrix, including 28 of the original 38 items and 10 factors, is presented in Table 5.6 (loadings below 0.25 have been suppressed for clarity). See Part A of the survey instrument in Appendix B for exact question wording. To improve the generalizability of the factor loading matrix, the research team combined attitudinal data from this survey with similar responses from a reuse of the authors’ survey to assess infra- structure in Atlanta, GA, to inform the estimation of the factor loading matrix. However, scores for this analysis were restandardized for the sample to present a more familiar interpretation of coefficients of factors in models. The factor correlation matrix is presented in Table 5.7. Although rotation of the factor load- ing matrix was allowed to be oblique for maximum flexibility, correlations among factors are generally small, indicating that each factor explains a substantively distinctive dimension of the common factor space. Regression Models Including Attitudinal Factors Models were estimated including attitudinal characteristics from both the 10 factors and the 10 single-item (standardized) variables dropped from the factor space. The cleaned version of each model is presented in Table 5.8. With the addition of attitudes, sociodemographics dropped entirely out of the comfort model. In all models, the variable for vehicles per licensed driver dropped out completely, largely because of the car dependence factor. Holding a driver’s license and having a child at home were still significant in the safety model, as were age, education, gender, and race in the willingness to try model. The R2 value went up for each model, with more modest increases in comfort (0.201 to 0.246) and safety (0.248 to 0.290), while the improvement seen in the willingness to try model was much more substantial (0.153 to 0.335).

48 Bicyclist Facility Preferences and Effects on Increasing Bicycle Trips Car Dependence Bike Enjoyment Active Travel Travel-Time Usefulness Utilitarian Travel Anti- Exercise Time Pressure Risk- Taking Tech Adoption Rarity of Cycling 3. Car as a symbol of freedom 0.666 17. Like traveling by car 0.620 25. Rent rather than own car −0.531 35. Cars first priority of roads 0.431 0.298 31. Dream to own big house/yard 0.297 0.257 37. Like biking 0.692 32. Would bike more w/ friends 0.623 22. Bike/walk instead of car 0.373 26. Improve sidewalks 0.473 29. Like transit −0.282 0.382 7. Too far for environment −0.374 0.273 1. Like to walk to store 0.360 13. Environmentally friendly 0.354 20. Travel time is wasted −0.641 11. Able to use travel time 0.635 27. Only travel for destination 0.675 8. Enjoy traveling 0.292 −0.321 2. Exercise is overrated 0.654 30. Exercise is important −0.506 21. Too busy to do everything 0.616 6. Often in a hurry 0.503 23. Need to use every minute 0.440 5. Taking risks 0.509 15. Like to try new things 0.483 12. Like to be first to try tech 0.539 36. Phone as important as body 0.409 9. Bicyclists are seen as odd 0.505 4. Drivers don’t notice cyclists 0.346 Note: Shaded cells indicate the items most closely associated with each factor. Table 5.6. Factor pattern loading matrix of 10 factors on 28 items. Car Dependence Bike Enjoyment Active Travel Travel-Time Usefulness Utilitarian Travel Anti- Exercise Time Pressure Risk- Taking Tech Adoption Rarity of Cycling Car Dependence 1.00 –0.05 –0.10 0.15 <0.01 0.02 0.07 0.02 0.03 –0.02 Bike Enjoyment 1.00 0.17 0.05 0.02 –0.16 0.09 0.18 0.09 0.01 Active Travel 1.00 0.05 0.02 –0.07 0.01 0.07 0.03 –0.01 Travel-Time Usefulness 1.00 0.03 0.02 –0.03 –0.12 –0.04 –0.14 Utilitarian Travel 1.00 0.12 0.03 0.04 <0.01 –0.02 Anti- Exercise 1.00 0.04 <0.01 0.02 <0.01 Time Pressure 1.00 0.09 0.09 0.11 Risk-Taking 1.00 –0.03 0.05 Tech Adoption 1.00 0.11 Rarity of Cycling 1.00 Table 5.7. Factor correlation matrix for 10 factors.

Expressed Comfort, Safety, and Willingness to Try Cycling 49 Three single-item attitudes were significant in these models. The stubbornness question, “It’s pretty hard for my friends to get me to change my mind,” was significant and negative in each model. Coupled with most respondents not being current bicyclists, this indicates that one barrier to cycling is simply the difficulty of changing some people’s minds. The perception that cyclists are unsafe, measured by, “Many bicyclists appear to have little regard for their personal safety,” was also significantly negative in each model, indicating that a perception of the lack of safety of cycling and a perception of negative characteristics with those considered bicyclists is associated with more negative perceptions of the infrastructure configurations presented. The perception that cyclists are poor, as measured by, “Most bicyclists look like they are too poor to own a car,” is likewise negative, but only for the comfort and willingness to try models. This further supports the argument that the identification of negative or undesirable traits with bicyclists affects an individual’s general perception of comfort and expressed willingness to try bicycling, though the evidence for perceived safety is only strong for perceptions related to safety. Four attitudinal factors were significant in each model. Car dependence was negative in each model, indicating that those who depend on automobiles respond with lower degrees of perceived comfort, safety, and willingness to try bicycling. Bike enjoyment was positive with Table 5.8. Linear regression for expressed comfort, safety, and willingness to try by infrastructure, sociodemographic, and attitudinal characteristics. Variable Comfort Safety Willingness to Try Coefficient P Coefficient P Coefficient P Constant 2.89 *** <0.001 2.43 *** <0.001 3.03 *** <0.001 Bicycle Facility Types Bike Lane (BL) 0.36 *** <0.001 0.46 *** <0.001 0.30 *** <0.001 Buffered BL (BB) 0.74 *** <0.001 0.91 *** <0.001 0.60 *** <0.001 One-way Protected 1.35 *** <0.001 1.71 *** <0.001 1.14 *** <0.001 Two-way Protected 1.16 *** <0.001 1.47 *** <0.001 0.96 *** <0.001 Multi-use 1.23 *** <0.001 1.54 *** <0.001 1.14 *** <0.001 Roadway Characteristics Parking −0.26 *** <0.001 −0.25 *** <0.001 −0.17 *** <0.001 Four Lanes −0.007 0.477 0.02 0.566 −0.07 * 0.024 Framing Effects BL—No Parking 0.44 *** <0.001 0.50 *** <0.001 0.43 *** <0.001 BB—No Parking 0.20 *** <0.001 0.32 *** <0.001 0.18 ** 0.002 BL—Two Lanes 0.28 *** <0.001 0.30 *** <0.001 0.20 * 0.012 Sociodemographics Age (in 10s of years) −0.036 *** <0.001 Education 0.029 ** 0.004 Driver’s License 0.18 *** <0.001 Child in Home −0.082 * 0.020 Female −0.21 *** <0.001 African American −0.13 *** <0.001 Attitudes A18. Stubborn −0.053 *** <0.001 −0.036 ** 0.009 −0.033 * 0.025 A24. Cyclists unsafe −0.087 *** <0.001 −0.11 *** <0.001 −0.15 *** <0.001 A28. Cyclists poor −0.030 * 0.036 −0.31 * 0.037 Car Dependence −0.033 * 0.013 −0.055 *** <0.001 −0.14 *** <0.001 Bike Enjoyment 0.20 *** <0.001 0.14 *** <0.001 0.47 *** <0.001 Active Travel 0.11 *** <0.001 0.090 *** <0.001 0.11 *** <0.001 Utilitarian Travel −0.058 *** <0.001 −0.039 ** 0.008 Risk-Taking 0.051 *** <0.001 0.043 ** 0.002 0.069 *** <0.001 Travel-Time Usefulness −0.071 *** <0.001 Anti-Exercise −0.069 *** <0.001 # of Responses 6,762 6,578 6,557 R2 0.246 0.290 0.335 Adj R2 0.244 0.288 0.332 P < 0.1, *P < 0.05, **P < 0.01, ***P < 0.001

50 Bicyclist Facility Preferences and Effects on Increasing Bicycle Trips comparatively large coefficients, as would be expected. Active travel was also positive, though to a lesser extent. Risk-taking was also positive, indicating that those who are more prone to taking risks are more open to the concept of cycling, while risk-averse persons are less open. Utilitarian travel was significantly negative for comfort and safety only, indicating that those who do not like traveling have somewhat lower perceptions of comfort and safety. Travel-time usefulness was significant only for willingness to try, as was anti-exercise, both of which were negative. This indicates that those who currently view the time they spend traveling as useful are less likely to be willing to try cycling, which for many would be a new mode. Some respondents may view bicycling as making it more difficult to make good use of travel time: it generally takes longer, it is not hands-free, and listening to music may not be as satisfying as in an enclosed vehicle. If such respondents also are not interested in the exercise benefits of cycling, they may downplay “getting exercise” as being a good use of the time. The anti-exercise attitude has the expected negative coefficient, which has essentially the same magnitude (and sign) as that of the travel- time usefulness perception. This indicates that a pro-exercise attitude can more or less counteract a sense that bicycle time is wasted, while an anti-exercise attitude magnifies the latter effect. Summary This chapter discussed the findings of the first-wave survey (N = 1,178) deployed in six com- munities in Alabama and Tennessee, where cycling is not widely adopted. Consistent with pre- vious studies, the survey results suggest similar patterns between perceived comfort and safety and willingness to try bicycling. Respondents rated bicycling facilities having a higher degree of separation from drivers more positively, with protected/separated bike lanes and multi-use paths being the best. Parking was a clear deterrent for all measures of preference, while the effects of the difference between two and four traffic lanes were mixed. Protected/separated bike lanes seemed effective in reducing the negative effects of roadway characteristics. User characteristics were significant in modeling respondents’ perceptions of being comfort- able, safe, and willing to try bicycling. Sociodemographic information was more influential in pre- dicting willingness to try than comfort and safety, indicating that even when safety and comfort are similarly perceived across population segments, willingness to try can differ. On average, older respondents responded more negatively on all measures as opposed to younger respondents. The segmented models indicate that perceptions of infrastructure characteristics can be sub- stantially different among different rider types. Utilitarian cyclists overwhelmingly viewed sepa- rated facilities as safer than sharrows, and even more so than the rest of the sample, but were less fazed by the presence of parking. Occasional/recreational cyclists’ preferences were surprisingly similar to those of potential cyclists, with no significant difference for perceived comfort and safety, and only education and age being significant for willingness to try. Those who are not able to bike did not differ significantly from the base of potential cyclists except in the willingness to try model, though this segment may be of less interest. The limitations of the analysis of user preferences contained in this chapter include the hypo- thetical nature of stated preference studies. Although Birmingham and Chattanooga have a reasonable amount and variety of bike infrastructure and Northport has several trails, most residents in the sample likely would not have seen many of the presented facility types, adding to the hypothetical challenges in this study. The findings from this study bring new insights from areas of the United States without a strong cycling presence. This study includes only a handful of locations from one geographic region, but the survey was written for general application in other locations. Examples of projects already using aspects from this survey include an extension of this same project by the authors in Atlanta and Sanders and Judelman (2018) in Michigan.