Complete results from the models described here can be obtained from the authors with reasonable request.
The user survey aims to identify the drives and constraints on rural cycling, either for commuting or leisure, for different user types. The results will indicate which variables are more relevant to users to cycle (or not) on rural highways. Based on the results, we will recommend future research on improving and revising existing HCM’s BLOS evaluation and Green Book design guidance to accommodate users’ revealed preferences.
The survey was designed in LimeSurvey and contained the following main sections:
The first section has the introductory message to the respondent that includes the purpose of the survey, the research project, the research team, and their contact information. We also included the expected time to complete the survey. The survey is targeted at bicycle users. The welcome message includes the age and US resident explicitly: We are conducting a 15-minute survey to identify the preferences and characteristics of cyclists that use the road infrastructure in the US. If you are a US resident, and 18 or older, we kindly invite you to participate. For this survey, respondents do not require any kind of previous/current experience in bicycle facilities - your input is important for conducting this research.
It was also indicated that the survey was used exclusively for research purposes and approved by the University of Florida Institutional Review Board. Further, it was explicit that the record of survey responses will not contain any identifying information. Survey respondents had the option to provide their contact information on a separate survey, and the responses of both surveys cannot be attached. Therefore, the contact information cannot be associated with the primary survey responses.
The second section follows a rule-based decision tree from Félix et al. (2016). In this seminal paper, they propose typologies of urban cyclists and market segmentation methods. The purpose of this section is to classify the respondents based on their cycling experience, as non-cyclist, potential cyclist, or regular cyclist. Regular cyclists have cycled anytime in the past 12 months for commuting or leisure, cycled in the past month for commuting, and will consider cycling again in the next month. Potential cyclists answered “no” to some of the previous questions but expressed their interest in cycling anytime in the next 12 months. Finally, non-cyclists indicated they are not interested in cycling anytime in the next 12 months.
The bicyclist segmentation was used to direct the respondents to the next sections, as follows:
The third section started with the instructions for the choice tasks. This first page contains one screenshot of an example choice task and guides the respondent on how it should be completed.
On the next page, the respondent had four choice tasks. Respondents were asked to imagine riding a bicycle on a rural road in each choice task. Then, four images with different cycling conditions were. The respondent needed to choose the Which cycling conditions would you feel serves better your needs as a bicyclist. Only one image per task could be selected. If they do not feel like riding in any of them, they could select “None of them”.
A total of 16 images were generated with varying speed limits, presence of shoulder, grade, context classification, automobile traffic volumes, and pavement quality. All 16 images were presented to the respondents in four separate choice tasks. However, choice tasks were randomly generated to minimize the bias in the model. After completing the four choice tasks, respondents enter another page with a second set of four choice tasks. This second set of choice tasks was
introduced to increase the sample size of responses while keeping the time of the overall survey relatively short.
Specifically, the following attributes and attribute levels are considered:
Table G-1. Attributes and attributes level of the choice tasks

The fractional design was applied to reduce the potential number of alternatives to present from 324 to 16 (4 choice tasks with four options each). The final set of other options is summarized in the following table.
Table G-2. Attributes of the choice task scenarios

The fourth section collects more details on their cycling experience. For respondents classified as “leisure rural cyclists,” collected information on their typical cycling leisure rides include:
The closing section starts with a ranking task. In this task, respondents must order the different groups of variables based on how important they are when riding their bicycles. This question was also adapted from the practitioners’ survey and included the following elements:
The second question in the closing section consisted of stated cycling comfort levels at different cycling environments. The scenarios were developed following the description of Dill and Mc Neil (2016) that was used to classify cyclists by their LTS. The original typology was developed by Geller (2008) and was based on the individuals’ stated comfort level to ride in different bicycle infrastructures. Specifically, they included:
To reduce the survey time, the descriptions of bicycle infrastructure were replaced by cross sections and adapted to have all the different configurations with bike lanes, striped bike lanes, and no bike lanes on two-lane streets and major four-lane streets. A total of nine bicycle infrastructure scenarios were presented, as follows:
Respondents were presented with cross sections of different bicycle infrastructures and were asked How comfortable would you cycle in the following scenarios. Four answers were allowed: Very uncomfortable - Uncomfortable - Comfortable - Very comfortable.
Finally, we include sociodemographic questions regarding gender, age, income, and how they found out about the survey. In the end, it is possible to leave further comments, and contact data in a separate study in case they want to know more about the research.
A pretest was conducted with the research team members and panel members to refine overall understanding, time management, and variables included. Afterward, it was submitted to the University of Florida Institutional Review Board (UF-IRB), which oversees research studies involving human subjects. UF-IRB approved the survey in November 2022.
The survey is targeted to bicycle users and will be distributed to the American League of Cyclists and any other cyclist associations; also, the survey was distributed through social platforms such as Facebook groups, LinkedIn groups, Twitter, and University Transportation Centers (UTC). Although the survey is not intended for practitioners, it was shared on the same distribution channels as the practitioners’ survey to increase its reach within their communities. The survey was distributed through the following channels:
A total of 1650 persons entered the survey and 1049 completed the survey (63.57 percent of respondents).
Two hundred twenty-six persons left the survey in the welcome section. After the welcome section, respondents were asked about their cycling experience in the last month and year to segment the responses between regular, potential, and no cyclists. One hundred sixty left the survey after this section. Afterwards, 157 persons did not complete the choice tasks, and only 58 respondents left during the specific questions on cycling experience.
Table G-3. Number of responses by the last section of the survey

Almost all respondents lived in the United States (1032 respondents), with the notable exception of 13 responses from Germany (4 responses), France (1 response), Netherlands (1 response), and Togo (1 response). A total of four persons did not include their country of residence. Given that the analysis is focused on the United States, only 1032 complete responses from individuals living in the United States are used.
Table G-4. Country of residence

Responses from 51 states were received. The highest number of responses by state is 95, from California, followed by Pennsylvania (92 responses), Michigan (85 responses), Illinois (80 responses), and North Carolina (58 responses).
Absolute values were divided by the number of inhabitants in 2015 to obtain the density of responses per state. The six states with more responses per inhabitant are Montana (26.14 responses per million inhabitants), Vermont (17.57 responses per million inhabitants), Alaska (14.9 responses per million inhabitants), Wisconsin (10.05 responses per million inhabitants),
Wisconsin (10.05 responses per million inhabitants) and Wisconsin (10.05 responses per million inhabitants).
The respondents are primarily males (N = 701, 67.93%), aged 60 or more (N = 513, 49.71%), with high annual household income (N = 460, 44.57%) and commuting by car (N = 422, 40.89%) or bicycle (N = 174, 16.86%). Many respondents do not commute for work or education (N = 345, 33.43%), which relates to the high share of individuals over 60. Only 21.71% of the respondents were younger than 45 years old.
Table G-5. Survey descriptive statistics - gender.

Table G-6. Survey descriptive statistics – age.

Table G-7. Survey descriptive statistics – household income.

Table G-8. Survey descriptive statistics – main commute mode.

Most of the respondents found the survey via email (N = 424, 41.09%) or social media (N = 196, 18.99%), with Facebook as the most used channel (N = 160, 15.5%). Distribution on Twitter and LinkedIn had a lower impact, with 20 and 16 respondents, respectively (3.49% of responses). Interestingly, 167 of participants (16.18%) were referred to the survey from a colleague.

In the first section of the survey, respondents were asked about their cycling experience in the last year and month and if they would consider cycling again in those time frames. The questions were designed to classify the respondents into three groups:
The classification from Félix et al. (2009) uses the following logic.
However, this classification does not account for seasonality. Based on the frequency of bicycle use those potential cyclists indicated, we re-defined the categories as follows:
Most respondents were all-year-round (N = 685, 66.38%) or seasonal cyclists (N = 200, 19.38%). Only 72 and 68 individuals were classified as non-cyclists (6.98%) or potential cyclists (6.59%). Non-cyclists were males over 60 (N = More than 60, 24%) and females between 46 and 60 years old (N = 46 - 60, 15%).
Table G-10. Cyclist type by seasonal cycling frequency

All-year-round cyclists were mainly men (75.33%), contrasting with 22.34% of females. Gender distribution was more balanced on other cyclist types: 57% males and 38% females. There were no significant variations in the age distributions for regular and potential cyclists and non-cyclists.
Table G-11. Cyclist type by seasonal cycling frequency – gender

Table G-12. Cyclist type by seasonal cycling frequency – age

Not surprisingly, all-year-round cyclists commuted more by bicycle (23.21%) than potential cyclists (6%) and non-cyclists (0%); and used more E-Bicycles (2.63% compared to 0.5% and
0%). However, the differences between all-year-round and seasonal cyclists were more minor. The main commute mode for potential cyclists and non-cyclists was auto: 50% and 63.24%, respectively.
Table G-13. Cyclist type by seasonal cycling frequency – main commute mode

All-year-round, seasonal, and potential cyclists were asked about their typical bicycle trip. Most regular cyclists cycled at least three times per week or once per week, and less than 2% cycled less than once per month. Not surprisingly, seasonal, and potential cyclists cycled less than regular cyclists. Initially, almost 75% of potential cyclists cycled at least once per week, as all individuals who did not cycle in the last month were classified as potential cyclists. The survey was conducted in December and January, which could have affected the cycling experience of the previous month and neglected seasonality.
This suggests that the rule-based decision tree cycling experience on the last year and month may not be too accurate for areas where cycling may be less frequent in winter. Surveying in another season of the year may lead to different results, as one respondent indicated in the open question: “One thing these surveys do not consider or measure very well is seasonal variability. While I ride nearly 100% of the time in summer, I tend to drive much more often in the peak of winter”.
Therefore, seasonal cyclists were segmented from potential cyclists if they cycled at least once per week for their typical cycle. The following table summarizes that the highest difference between all-year-round cyclists and others is winter: less than 6% of seasonal and potential cyclists’ cycle in winter, compared to 70.66% of all-year-round cyclists.
Table G-14. Cyclist type by seasonal cycling frequency – frequency

Table G-15. Cyclist type by seasonal cycling frequency – seasons

Regarding the time of week, most cyclists cycle on Saturdays and Sundays (between 65 and 80%). High frequencies were also observed for workdays, off-peak hours, for all-year-round and seasonal cyclists. Among cyclist types, the group that cycled most during peak hours were all-year-round cyclists (43.1%), compared to 29.5% of seasonal cyclists and 17.7% of potential cyclists—around one-quarter of all-year-round cyclists cycled during nighttime.
Table G-16. Cyclist type by seasonal cycling frequency – day of week and time period

Weather conditions showed the highest differences among cyclist types. Specifically, we observed that all-year-round cyclists were more comfortable with all adverse environmental conditions compared to seasonal and potential cyclists: all-year-round cyclists cycled between two and 10 times more than potential cyclists under adverse environmental conditions. For
example, 19.71% of all-year-round cyclists cycle with snow, compared to 2.5% of seasonal cyclists and 1.47% of potential cyclists. The highest differences were for fog (25.11% vs. 10% and 5.88%), rain (49.93% vs. 41% and 22.06%), and wind (74.74% vs. 64.5% and 30.88%).
Table G-17. Cyclist type by seasonal cycling frequency – weather conditions

At the closing section, participants were asked to select which phrase would describe themselves better as a cyclist:
The four phrases align with the four cyclist types that are considered in the LTS theory.
Most of the participants (N = 723, 70.06%) considered that “enthused and confident” described them best, followed by “interested but concerned” (N = 196, 18.99%). Eighty-eight participants described themselves as “strong and fearless” (8.53%). The distribution differs from the general population of cyclists described by Geller (2006) and Dill and McNeil (2013), on where the shares of “strong and fearless,” “enthused and confident,” and “interested but concerned” were equal to 4, 9 and 56%, respectively. The survey sample over-represented individuals described as “enthused and confident,” which could be caused by the focus on cyclists who ride on rural highways. For those individuals, the confidence level is expected to be higher than the overall population of cyclists.
Table G-18. Stated cyclist type - distribution.

Women are most likely to describe themselves as “interested but concerned,” with a share of 47.45% in the category. Other categories are dominated by males, with shares between 72.34% and 76.14%. Age distributions are distinct for each type of cyclist. “Enthused and confident” and “interested but concerned” cyclists are dominated by individuals over 60 (52.84% and 44.9%), followed by persons between 46 and 60 years old (26.83% and 23.98%). The age distribution is like the overall age distribution of the entire sample. However, strong, and fearless presents a different distribution, with a higher share of persons aged 45 and less (30.68%) than for the overall sample (21.8%).
Table G-19. Stated cyclist type – gender.

Table G-20. Stated cyclist type – age.

The most used mode for “strong and fearless” cyclists is bicycle (30.68%), compared to auto for “enthused and confident” (37.9%) and “interested but concerned” (58.16%). For those cyclists, only 18.67% and 5.61% commute by bicycle. “Strong and fearless” also has the highest use of ebikes, with 3.41%, and the least number of persons that do not commute for work or education (23.86%).
Table G-21. Stated cyclist type – main commute mode.

We also compared the share of cyclists that cycled in the different seasons, times of the week, and under adverse environmental conditions. It could be observed that cycling frequencies were consistently higher for “strong and fearless” cyclists compared to other groups. Further, the differences between “strong and fearless” and “enthused and confident” were more minor than between “enthused and confident” and “interested but concerned”. The highest difference was found for cycling in winter, with a drop from 65% (for “strong and fearless”) to 29.1% (for “interested but concerned”). As indicated before, the difference between “enthused and confident” was more minor (53.1%).
The time of week patterns did not present significant differences, only for the share of nighttime rides (between 13.8 and 35.2%) and peak hour rides (between 26.02 and 61.36%).
As for cycling frequency, cycling under adverse environmental conditions significantly differed across groups. “Strong and fearless” were generally more comfortable cycling in adverse weather than others, followed closely by “enthused and confident.” The highest differences in percentage were observed in snow: “strong and fearless” cycled 70% more than “enthused and confident” (26% compared to 15%) and seven times more “interested but concerned” (26% compared to 3.6%). Cycling under fog was also limited to “interested but concerned.” The environmental conditions most tolerated by cyclists were high temperatures, wind, and low temperatures. Only 60% of “strong and fearless” cycled when raining (compared to 24.5% of interested but concerned).
Time of week patterns were similar across comfort levels, with higher shares on Saturday, Sunday, and workday, off-peak hours.
Table G-25. Stated cyclist type – frequency.

Lastly, a comparison between the two classifications was performed. It could be observed that almost all the individuals who indicated agreement with being defined as “strong and fearless” were all-year-round cyclists (86.4%), followed by seasonal cyclists (6.8%). The share of potential and non-cyclists was below 3%. On the other hand, most enthused and confident cyclists were classified as all-year-round (70.9%) or seasonal (20.2%). Not surprisingly, interested but concerned cyclists presented more significant shares of potential and non-cyclists. However, most of them could be classified as all-year-round or seasonal cyclists.
Table G-26. Comparison between cycling frequency and stated cyclist type.

In this survey section, choice tasks were presented to the participants. In each choice task, respondents were asked to imagine that he/she is riding a bicycle on a rural road, and four images were presented to them. They needed to select the image with the cycling conditions that served better their needs as bicyclists, if any. Specifically, the question was: Imagine you are riding a bicycle on a rural road. Which cycling conditions would you feel serves better your needs as a bicyclist? Please choose only one answer. If they do not feel like riding in any of them, they could select “None of them.”
A total of 16 combinations of speed limit, presence of shoulder, grade, context classification, automobile traffic volume, and pavement quality were presented to the participants.
Preliminary analysis consisted of comparing the frequency of a specific element being presented to participants and being selected by participants. Firstly, an individual analysis by element was carried out, followed by an analysis of all the scenarios.
Individual analysis by element included the number of times each category was selected and the number of times it was presented to participants. Fractional design produced an uneven distribution of types; therefore, analyzing the absolute number of responses may need to be more accurate. Thus, the absolute and relative differences between the number of selected and presented responses were computed.
Participants were presented with scenarios with two different speed limits: low or high. Specific speed limit depended on the context classification:
Not surprisingly, most respondents selected scenarios with low speed limit (N = 5213, 63.16%), compared to scenarios with high speed limit (N = 1850, 22.41%). However, there were more scenarios with low speed limit in the choice set due to the fractional design. Therefore, analyzing the absolute number of responses may be misleading, and the difference between the number of selected and presented responses should be used instead. The low speed limit was selected 13.16 percent points more than presented to the participants (an increase of 26.31%). The high speed limit was strongly affected, with a 25.28% difference between the selected and presented.
Table G-27. Percentage of times selected by speed limit.

Two variations of grade were used: high and low grade. Images with high grades had more challenging scenery, with more trees on the sides to represent rolling terrain, while images with low grades presented flat conditions.
The two categories were presented the same number of times (N = 3302, 40%); however, scenarios with low grades were selected more (47.36%) than scenarios with high grades (38.21%). Participants represented a higher propensity of choosing scenarios with low grades (7.36 percent point difference) and slightly lower propensity of selecting scenarios with high grade (-1.79 percent point difference).
Table G-28. Percentage of times selected by grade.

Three different categories of automobile traffic volume were presented in the scenarios:
The results of this element were also intuitive: the higher the automobile traffic volume, the lower the propensity of being selected: -2.8 percent point difference for high volume, compared to 3.79 percent point difference for no vehicles. As for speed limit, the distribution of the three categories was uneven, and therefore, the analysis of the absolute number of responses may be misleading. High volume was selected more times (N = 2658) than no vehicles (N =2029).
Table G-29. Percentage of times selected by automobile traffic volume.

Three different configurations of shoulder were generated:
Wide shoulder scenarios outnumbered other categories: they were selected by 4,398 (53.28%), compared to being presented 2,064 times (25%). This represented an absolute difference of 28.28 percent points, the highest difference across all individual elements.
On the other hand, having no shoulder presented one of the highest negative differences across all individual elements, with a -20.11 percent point difference (-57.46 relative difference). The narrow shoulder had a slightly lower propensity of being selected (-2.6 percent point difference).
Table G-30. Percentage of times selected by presence of shoulder and shoulder width.

Pavement conditions were divided into three categories:
Like shoulder width, one of the categories outnumbered the others. Good pavement conditions had a high propensity of being selected, with a difference between selected and presented equal to 14.22 percent points (40.64%). On the other hand, deteriorated pavement conditions had little propensity of being selected: -8.78 percent point difference (-43.91). Participants selected fewer scenarios with acceptable pavement conditions (N = 2074, 25.13%), with only a 0.13 percent point difference.
This is the second highest (positive and negative) differences across all elements.
Table G-31. Percentage of times selected by pavement quality.

Finally, the differences in context classification were analyzed. Three different context classifications were included: rural, rural–town, and suburban. Even though suburban segments are not usually considered rural highways, they were included in the sample to determine whether rural conditions differ substantially from more urbanized suburban settings.
The results indicate that suburban and rural–town context classifications were more likely to be selected, with absolute differences equal to 7.31 percent points (36.52) and 3.8 percent points (12.68), respectively. Conversely, rural context presented a negative difference, like -5.54 (- 18.46%). It should be noted that this difference was not the highest negative across all elements. Therefore, other factors might strongly influence the participants' choices.
Table G-32. Percentage of times selected by context classification.

After analyzing each element individually, all scenarios were compared to each other. The following table indicates the frequency with which each scenario was selected and its attributes.
The most frequent scenario was number 14, which was selected 1386 times (16.79%). This scenario had a low speed limit (20 mph), low grade, wide shoulder, low traffic volume, good pavement conditions, and suburban context. It was selected three times more than presented (16.79% compared to 5%). Other scenarios that were set more times than given include scenarios number 10 (11.93%), 8 (10.23%), 3 (9.64%), 13 (8.29%), 1 (5.8%), and 7 (5.22%). These scenarios mainly were suburban or rural–town settings, with wide shoulders, low speed limits, and good pavement conditions. Nevertheless, some combinations also presented high speed limits, no shoulders, rural conditions, and the range of traffic volumes.
On the other side of the spectrum, the least selected scenario was number 11, which presented a high speed limit (30 mph), high grade, no shoulder, low traffic volume, deteriorated pavement, and rural–town context. Surprisingly, this scenario was selected by some participants (N = 13, 0.16%). Participants' three least selected scenarios (scenarios 4, 15, and 11) had in common that no shoulder was present, and that speed limit was high. However, they varied in the other elements: grade (low and high), traffic volume (low and high), pavement conditions (acceptable and deteriorated), and context classification (rural, suburban, and rural–town). These scenarios were selected less than 1.2%.
It should be noted that the second most selected scenario was “None of them” (N = 1191, 14.43%). This scenario was presented to the participants 20% of the time, once for each task choice scenario.
Table G-33. Percentage of times selected by scenario.

This section describes the statistical analysis that was carried out to evaluate each element's contribution to each scenario's utility. A binary logit model was used to estimate the utility of each component. This model predicts the probability of the scenario being selected (yes or no), given a set of independent variables. The alternative “no” was selected as the reference level.
The independent variables considered in the model were the attributes of the scenarios (speed limit, grade, automobile traffic volume, shoulder width, pavement conditions, context classification) and the individual characteristics of the participants (type of cyclist by cycling frequency and stated cyclist type, age, gender, income, commute mode). Correlation among independent variables was tested, and only variables that were not highly correlated (50% or less) were included in the model.
Stepwise selection was carried out to reduce the number of independent variables. The final model had a pseudo R2 equal to 39.7%. Only statistically significant variables at 95% confidence
were included in the model (p-value below 0.05). The model parameters' coefficients, standard error, t-test, and p-values are summarized in the following table.
Table G-34. Choice model – statistical summary

All six attributes of the scenarios (speed limit, grade, automobile traffic volume, shoulder width, pavement conditions, and context classification), the interaction between context classification and the presence of shoulder, and two attributes of the participants (stated cyclist type and age) were statistically significant. The signs of the coefficients and values were plausible and agreed with analyzing individual elements.
The model's coefficients reflect the change in the log-odds of the selected scenario associated with a one-unit change in that variable. Positive coefficients indicate that the predictor variable is associated with an increase in the probability of being selected, while negative coefficients indicate a decrease. The magnitude of the coefficient indicates the strength of the relationship
between the predictor variable and the outcome. Odds ratios were also calculated to provide a more intuitive interpretation of the coefficients. This parameter is defined as the ratio of the odds of the outcome occurring for a one-unit increase in the predictor variable compared to the odds for a one-unit decrease. An odds ratio greater than 1 indicates that the predictor variable is associated with increased odds of being selected, while an odds ratio of less than 1 indicates a decrease. For interaction terms, a negative coefficient indicates that the direction of the effect is opposite for different groups.
The presence of a shoulder had the highest impact on the probability of being selected: including a narrow or wide shoulder increased the log-odds of being chosen by 0.53 and 2.26, respectively. This represents an odds ratio of 1.7 and 9.6, respectively. This indicates that if a narrow shoulder was present, the odds of being selected increased by a factor of 1.7 compared to not having a shoulder. When having a wide shoulder, the estimated odds of selecting that scenario were 9.6 times higher than for scenarios without a shoulder.
The interaction between shoulder and context class was significant, indicating that the effect of having a shoulder on the likelihood of selecting to cycle in the scenario depended on context class. Still, the direction of the product was the opposite for rural and rural–town contexts. Specifically, as the interaction term is negative, it suggests that rural areas without shoulders were less likely to be selected compared to rural–town areas without shoulders. However, for scenarios with shoulders, the effect of context was weaker, indicating that context was less of a factor if shoulders were present. Similar results were produced between rural and suburban context classes, although the differences were lower than between rural and rural–town contexts.
Scenarios in rural towns were more likely to be selected than those in rural contexts by a factor of 1.9 (log-odds of 0.68). The difference between rural and suburban conditions was slightly more significant. Interestingly, the difference between rural towns and suburban contexts was minor. However, the interactions between context classification and shoulder presence were significantly different.
Two other important variables were automobile traffic volume and pavement conditions. Automobile traffic volume had negative coefficients, indicating that scenarios with more traffic were less likely to be selected by respondents than those with no traffic. Similarly, good pavement increased the log-odds by 0.84 compared to deteriorated pavement (factor of 2.3). The differences between acceptable and deteriorated pavement conditions were lower, with a log-odds difference of 0.18 (factor of 1.2).
Surprisingly, the coefficient for high grades was positive. This indicated that scenarios with higher grades were more likely to be selected than those with lower grades. However, the individual grade analysis showed respondents preferred low-grade scenarios more frequently. This difference between the model estimation and preliminary analysis could be caused because scenarios with low grades were chosen due to other elements and not precise grades. Grades could have been interpreted differently by participants and used as a proxy for scenery quality. Images with high grade corresponded to scenery with trees and rolling terrain, compared to images with low grade in flat terrain, which could be perceived as monotonous (and therefore less attractive) to cycle.
Lastly, enthused and confident individuals were more likely to select a scenario, compared to selecting “none of them” and individuals who were strong and fearless. As expected, the
likelihood of being selected was higher for strong and fearless than for enthused and confident. Interactions between the cyclist type and scenario elements were tested but were not statistically significant. This indicated that the effect of the elements did not depend on cyclist types but rather that the more confident groups generally had a more substantial likelihood of selecting to cycle in each scenario. Age was also statistically significant: individuals aged 60 or older had an increased chance of selecting the scenario compared to individuals younger than 60. Other sociodemographic attributes were not statistically significant.
In the first question of the closing section, respondents were asked to rank these elements according to how important they are when riding your bike. Specifically, eight groups of factors were presented:
Firstly, we analyzed the frequency of each response. To facilitate the visualization, we merged the responses in four ranks: “1st – most important or 2nd”, “3rd or 4th”, 5th or 6th” and “7th or 8th – least important”, and ordered the elements based on their median rank.
Clearance-related variables, which include shoulder width or lane width, were the most critical variables, followed by automobile volume-related (e.g., number of vehicles, presence of trucks) and speed-related (e.g., speed limit, automobile average speed). Most respondents ranked them as “1st or 2nd” (between 40% and 60%). The next group of variables was activity volume-related, such as commercial areas or driveway density and pavement condition. The two groups of variables received 1st to 4th rank by almost half of the respondents. The third cluster of variables is environmental-related, with more responses in the second half of the rank (5th to 8th) than in the first half (1st to 4th). This group includes adverse weather conditions. Finally, the least important groups of variables were terrain-related and other cyclists. We found that most responses were equal to “7th or 8th”, although, for 20% of the respondents, they ranked among the four most important variables.
Secondly, all variables were compared and arranged by median and mean values in descending order. The aggregated summary included IQR as a measure of dispersion. IQR represents the range of the 50% central responses. The higher the IQR, the more dispersion on central values. This metric was preferred over standard deviation because Likert scales are ordered variables, and answers are not normally distributed around a central value.
The analysis of the median and IQR shows similar trends. Most respondents indicated agreement with the idea that clearance-related is the most critical variable when riding their bicycle (median = 2, IQR = 2), followed closely by automobile volume-related and speed-related variables (median = 2, IQR <= 3). Neutral opinions were found for activity volume-related, pavement condition, and environmental-related, with a median value equal to 5 (IQR >= 3). However, there was a high dispersion in the responses, especially for environmental-related variables (IQR = 4). Most respondents indicated that the least important variables why cycling are terrain-related, such as elevation or grade, and the presence of other cyclists (median = 7, IQR = 2).
Table G-35. Ranking – summary of statistics

In this last section, a multinomial logit model was used to estimate the ranked choices of the different elements. Given a set of independent variables, this model predicts the probabilities of the first-ranked variables. The set of independent variables considered for the model are the sociodemographic attributes of respondents (age, gender, income), their cycling behavior (main commute mode, cycling on rural highways), type of cyclist by cycling frequency (regular, potential, or non-cyclist) and their stated cyclist type (strong and fearless, enthused and confident, interested but concerned).
One of the alternatives is selected as reference level to estimate a multinomial logit model. We chose “Other cyclists” as the reference level in this case due to the lower ranking. Therefore, parameter estimates are relative to the expected difference between each level (e.g., “Clearance-related”) and “Other cyclists.” The parameter estimates of the resulting model are summarized in the following table.
Table G-36. Ranking model – statistics summary

Alternative specific constants (ASC) capture the average effect on the utility of all factors not included in the model. The model predictions are driven by the ASCs, as ASCs (greater than 0.84) are higher than most individual-specific variables in the range of -0.8 to 0.7. This suggests that there are unobserved factors on the independent variables that affect the difference between “other cyclists” and the other groups of variables. Specifically, the multinomial logit for clearance-related relative to other cyclists is 2.56 units higher, given that all other predictor variables in the model are constant. The second variable group is volume-related and speed-related, with ASCs between 2.24 and 2.01. The third variable group is activity volume and pavement quality, with ASC between 1.27 and 1.35. Their ASC is slightly higher than environment and terrain, with ASC between 0.84 and 1.08.
As for the impact of individual-specific variables, the significant attributes are age (equal or over 60), bicycle commuting, cycling for leisure on rural roads, the stated cyclist type, and whether the respondent is a regular cyclist. Specifically, individuals over 60 have a negative volume-related and speed-related coefficient, indicating that respondents are likelier to rank them lower than others. On the other hand, they are more likely to rank higher clearance-related and environment-related variables. Individuals who commute by bicycle tend to rank higher in clearance, volume, and speed, while they do not influence other variables. Interestingly, participants who cycle on rural roads provide a higher rank to the environment, and strong and fearless, and enthused and confident cyclists are less challenged by terrain. Finally, all-year-round regular cyclists rank lower in clearance, pavement quality, environment, and terrain than cyclists or non-cyclists.
The second question in the closing section stated cycling comfort levels in different cycling environments. Respondents were presented with nine cross sections of different bicycle infrastructures and were asked How comfortable would you cycle in the following scenarios. Four answers were allowed: Very uncomfortable - Uncomfortable - Comfortable - Very comfortable.
The preliminary analysis of the results included an individual analysis of the frequency of each response by stated cyclist type (interested but concerned, enthused and confident, strong and fearless). Afterward, the mean comfort level at the different infrastructures was compared. Lastly, a statistical analysis was carried out.
The following figures show the proposed bicycle infrastructure and the results by stated cyclist type. The figure shows in the x axis the percentage of responses by each category. Labels with the number of responses by category were added as well. The interested but concerned group generally felt less comfortable than the other two groups. Strong and fearless group felt the most comfortable in most environments, and the difference to enthused and confident group were smaller than the difference between enthused and confident group and interested but concerned group. The comfort level decreased as the infrastructure lacked dedicated bike infrastructure or increased the number of traffic lanes.
In path or trails, separated from motorized traffic, 98% of respondents felt were very comfortable or comfortable. A few respondents (less than 3 per stated cyclist type) indicated that they felt uncomfortable or very uncomfortable.
The number of very uncomfortable and uncomfortable responses increased on the infrastructure scenario with a bicycle lane on a two-lane street: 32% of interested but concerned, 15% of enthused and confident, and 10% of strong and fearless expressed their discomfort. Nevertheless, the percentage of very uncomfortable was below 4%.
Increasing the number of lanes from two to four increased the level of discomfort by 11-12 percent points to interested but concerned (from 32 to 45%) and to enthused and confident (from 15% to 26%) and by 7 percent points to strong and fearless (from 10 to 17%). In this environment, more than half of the interested but concerned were comfortable.
The following environment replaced the bicycle lane with a striped bike lane next to motorized traffic. This change increased even more the discomfort of respondents. Specifically, more individuals from the interested but concerned group expressed discomfort (68%) than comfort (32%), and half of the individuals from the enthused and confident group were uncomfortable or very uncomfortable (50%). For strong and fearless, the percentage of comfortable and very comfortable was more than half.
Modifying the configuration from two lanes to two lanes per direction with a center divider increased around 4 percent points the discomfort for the enthused and confident group and strong and fearless group. In this environment, more than half of those enthused and confident were uncomfortable or very uncomfortable (58%), compared to a third of strong and fearless (38%). The effect was more significant for those interested but concerned, whose level of discomfort increased by 8 percent points, from 68 to 76%.
Removing the center divider increased slightly the level of discomfort, by around 4 percent points for all groups. The strong and fearless group had more responses, comfortable or very comfortable (62%) than uncomfortable (28%) or very uncomfortable (10%). The percentage of responses as very uncomfortable increased to one-third (33%) on interested but concerned and to one-sixth (17%) on enthused and confident.
Removing the bicycle lane substantially increased the share of respondents who felt uncomfortable. Specifically, 83% of interested but concerned responded very uncomfortable or uncomfortable, closely followed by 75% of enthused and confident. For the first time, more than half of strong and fearless expressed discomfort in an environment (55%).
As in the case of the striped bike lane, modifying the configuration from two lanes to two lanes per direction with a center divider did not alter the results significantly for all groups, with an increase between 3 and 5 percent points on the level of discomfort.
Minor differences were found between a major street with four lanes and a major road with two lanes per direction and a center divider. The increase in discomfort was around 1% for all groups. Notably, 72% of strong and fearless felt uncomfortable or very uncomfortable, of which 33% were very uncomfortable.
After the individual analysis of the responses, mean comfort levels were calculated and compared across bicycling environments. The results on mean comfort level agreed with the individual analysis: strong and fearless presented a higher level of comfort by environment, followed by enthused and confident. Interested but concerned had the least mean comfort level across groups. This result agrees with the LTS theory.
The highest differences were produced due to the presence of a bicycle lane and its typology. Environments with trails were very comfortable for all users. Bike lane environments were the only additional ones that produced a positive mean comfort level for the interested but concerned group and the enthused and confident group. Once striped bike lanes were introduced, the mean comfort level decreased significantly for the interested but concerned group, with values around - 0.5 (uncomfortable). The decrease is lower for enthused and confident, who presented values close to zero (somehow uncomfortable). Strong and fearless, on the other hand, still showed a positive mean comfort level. Without bike lanes, all users had a negative mean comfort level.
Minor differences in the mean comfort level could be observed due to the number of lanes and the presence of a center divider. Generally, increasing the number of traffic lanes decreased the comfort level, with the differences more pronounced with bike and striped bike lanes. Removing the center divider slightly decreased the mean comfort level, with differences below 0.05.
The same comparison was made by type of cyclist by cycling frequency. There cannot be observed a clear distinction by type of cyclist. On bike lanes, the mean comfort level of regular cyclists was higher than that of potential cyclists and non-cyclists. However, this trend was reversed in environments with no bike lanes. This inclusive result can reinforce this classification's misuse in a winter survey.
Lastly, an ordered logit model was used to analyze ordered categorical response variables. The response variable is assumed to have an inherent ordering or ranking, such as in the question: very uncomfortable, uncomfortable, comfortable, and very comfortable. This model estimates the probability that a response falls into each category based on a set of independent variables.
Like the previous statistical analyses, the independent variables considered in the model were the attributes of the scenarios (bicycle lane configuration, number of lanes, presence of center divider) and the individual characteristics of the participants (type of cyclist by cycling frequency and stated cyclist type, age, gender, income, commute mode). Correlation among independent variables was tested, and only variables that were not highly correlated (50% or less) were included in the model.
The following table summarizes the final model. Only statistically significant variables at the 95% confidence interval were included. The fraction of correct choice predictions is 51.5%.

The coefficients of the variables represent the change in the log-odds of being in a higher category (e.g., uncomfortable instead of comfortable) for a one-unit change in the explanatory variable. A positive coefficient indicates that an increase in the explanatory variable is associated
with an increase in the log-odds of being in a higher category. A negative coefficient indicates that an increase in the explanatory variable is associated with a decrease in the log-odds of being in a higher category.
The bicycle infrastructure presented the highest impact on the level of comfort: the log-odds of being in a higher comfort category (i.e., moving from very uncomfortable to uncomfortable or from uncomfortable to comfortable) increased by 5.7 if the bicycle lane was segregated (path or trail), and by 2.79 if there was a bicycle lane. For striped bike lanes, the increase was more moderated (by 1.3). The interpretation of the center divider and number of lanes also made sense, although their magnitude was more moderated.
The stated cyclist type was also significant. This variable was modeled as a linear variable, as observed in comparing mean comfort values. The variable was equal to 0 for no way, no how, 0.33 for interested but concerned, 0.67 for enthused and confident, and 1.0 for strong and fearless. Being enthused and confident increased the log-odds by 1.51 (equal to 0.67 times 2.25), while strong and fearless were more comfortable by 2.25, compared to no way, no how.
Other sociodemographic attributes were also significant in this model. Females were generally more uncomfortable than males, as well as individuals younger than 31. Interestingly, commuters by bicycle tend to be more uncomfortable with a given environment than by transit or walking.
Respondents could provide their comments on the last page of the survey in a free-text box. More than 400 respondents introduced specific remarks. With the text, we generated a word cloud with the top 90 most mentioned words, as follows.

We can observe that the words mentioned more than 100 times were, in descending order: roads, shoulder, rural, bike, traffic, cyclists, drivers, cycling, speed, lane, and rumble strips. Also, there were mentions of pavement, debris, or gravel, as in the following excerpt of comments:
“Most of the rural highways around here are higher speed like 55 mph and very little shoulder….. mostly try to avoid them and choose more of a county type road in a more rural farmland area with car speeds more in the 30-45 mile an hour range. Although little shoulder, they are very low volume. I like rural areas for riding… and mostly try to avoid the highways or the thoroughfares.. I think a blinking light and a reflective vest really help…. and a mirror…. I keep a close look on cars approaching me from behind (especially if little to no shoulder) and give myself 2 feet from the edge and then try move over a little as they pass.”
“I rode my bicycle about 3,500 miles this year, including a trip all the way around Lake Michigan. Some of my riding was on bike paths, and some was on urban/suburban streets, but the majority was on rural roads. The most important safety feature to me is a wide shoulder, or other way to feel separated from motor vehicle traffic. I am comfortable riding on roads with little or no shoulder as long as traffic is light and visibility is good. ”
“Since 2010, I’ve completed six bicycle tours of between 200-600 miles, crossing through 5 US states (MN, WI, IN, IL, MI, VT) and two Canadian Provinces (ON, QC). Personally, the biggest barrier to feeling safe/comfortable on rural roads is a lack of separation between a vehicle and myself [e.g., either no shoulder, or a small shoulder (2’-3’)], even on low-speed roads (35 or 40-mph road). I’d prefer to ride on a 55-mph road if it either had a 6’ shoulder or had a 3’ shoulder and rumble strips that separated the vehicle travel lane and the shoulder. Overall, I am most uncomfortable when I feel like drivers may be thinking that I am riding in “their” space or “very close” to their space. By having a large enough shoulder to ride where I can ride on the far right of, I feel that I’m completely out of the driver’s space. Additionally, although I may be close a driver’s space when riding on a 3’-wide shoulder, if there are rumble strips in between the travel lane and shoulder, it is more definitive that the shoulder belongs to me/a cyclist. I can also be more relaxed because I’m less concerned I will be hit from behind. ”
“I train where I live, in the suburbs. I ride long distance in rural areas. I don’t ride in cities. I prefer a shoulder that is at least 3 feet clean space. If rumble strips, the shoulder must maintain that 3 feet of usable space and be cleaned. If it’s full of debris and pebbles (usually caused by the rumble strip that wasn’t cleaned well), then I’m forced to ride on the road. If it’s a very low volume road, like many rural roads in the UP, then I don’t need a shoulder at all. To me, volume is the most important element of determining where shoulder work is necessary. It’s also important to provide “routes” or “connections” for bikers in rural areas. Bikers can travel hundreds of miles and a state needs to be aware of the most likely and safest passages (roads) bikers will use to get from A to B, then to C and D. etc. ”
“I mainly cycle on multi-use trail systems. Cycling on rural roads scares me a little as I often cycle alone and I’m afraid I won’t be seen by drivers. I prefer wide shoulders with lots of clearance to feel safe. ”
“We need shoulders to support those that ride for fun, leisure and exercise or training. I am not comfortable riding on a shoulder or in the lane if vehicles are going faster than 40MPH. I prefer separated pathways and if traffic is slow and cooperating, the lane. ”
“My primary concerns are pavement quality and road width (could be travel lane width(s) or shoulder width). Somewhat less important are traffic volume and % of trucks. Assuming satisfactory pavement quality, I would prefer a wide road with heavy traffic over a narrow road with light or occasional traffic. ”
“The four Geller categories are bogus because each of them combines two different characteristics which are not necessarily related. I am old, not a strong rider but I am fearless, because I am skillful. Enthused and confident? Many bicycling advocates are enthused, but not confident. Interested but concerned? many are concerned, but not interested (they always worry about my safety), and vice versa (they would ride if it were more convenient). No way, no how: this is a slang expression describing aversion but in fact there are people from whom there is no way, due to disability, and others for whom there is no how, due to inconvenience. By using the slang expression, Geller leaves out both of these important deterrents to bicycling. ”
All comments are listed in the appendix. Comments are not edited.
The elements that survey respondents felt were most important when riding their bicycles were clearance, automobile traffic volume, and speed. Among them, clearance was the most relevant in the three analyses. Clearance-related elements presented the largest log odds in the stated rank of elements, the presence of the shoulder and its width had the highest impact on the probability of being selected in one scenario in the choice task, and bicycle infrastructure produced the highest log-odds to move to a higher comfort category in the stated cycling comfort evaluation. All-year-round regular cyclists provided lower importance to clearance than other cyclists, confirming that this group may cycle in more challenging conditions, with higher bicycling frequency throughout the year and under adverse environmental conditions. Furthermore, wide shoulder and lateral clearance were among the most mentioned terms in the final comments. Moreover, the effect of having a shoulder was larger for rural context class compared to rural–town and suburban. The differences between rural and suburban contexts were more minor than between rural and rural–town contexts. This suggests that the relative importance of clearance depends on context classification. Therefore, the analysis procedure for shared and bicycle lanes (paved shoulders) should be distinct, as their sensitivity to variables differs. The results of this survey validated the conclusions from the practitioner survey: respondents felt that the assessment of quality of service by rural bicyclists would differ in shared and bicycle lanes.
Automobile traffic volume-related and speed-related were the second most important group of variables. Automobile traffic volume was slightly more likely to be ranked more critical than speeds. Also, the log-odds were higher, indicating that scenarios with more traffic were less likely to be selected by respondents compared to scenarios with no traffic in a more pronounced way than scenarios with high-speed limit compared to scenarios with low speed limit. It should be noted that the respondents who commute by bicycle ranked higher volume and speed than other respondents, suggesting that frequent cyclists were more aware of the impact of these variables on their comfort. Survey respondents included multiple comments regarding speed and volume. Mainly, they indicated that high speeds were tolerable on routes with very low volume and wide shoulders.
The three top elements agree with previous research and the practitioners' survey. In the practitioners' survey, the aspects that survey respondents felt that bicyclists are most sensitive to in assessing the quality of service on rural highways were related to speed, clearance, automobile
traffic volume, and pavement conditions. They all affect safety conditions, some of which are either design elements or automobile traffic characteristics, which were ranked as the most critical variable groups by practitioners. More specifically, speed limit, lateral clearance, directional automobile volume, snow, fog, and rainfall were believed to be very sensitive to assessing bicyclists' service quality in shared lanes.
In their analysis of bicyclists' perceived risk when overtaken by motor vehicles on rural highways, Llorca et al. (2017) concluded that lateral clearance was the most significant factor. They pointed out that lateral clearance and speed correlated better with average risk perception as a proxy to measure aerodynamic forces between overtaking. When compared with the US HCM (TRB 2022) analysis procedure for the bicycle mode, the results largely agree with the five main criteria to evaluate performance: average effective width of the outside through lane, motorized vehicle volumes, motorized vehicle speeds, heavy vehicle volumes, and pavement conditions. Further, HCM-LOS is highly sensitive to shoulder width and heavy vehicle percentage (around 2 LOS letter grades) and moderately sensitive to lane width (around 1 LOS letter grade). However, field data comprised less than 2% of heavy vehicle percentages.
On the other hand, most bicycle facility design guidelines use motor vehicle volume and speed limit to determine a preferred bikeway type and required clearance (FHWA, 2019; ODOT, 2020; MnDOT, 2007; WSDOT, 2015). Generally, the higher the speed and traffic volume of a road, the more protective the recommended bikeway, either by segregating the facility or increasing its width. FHWA Bikeway Selection Guide (2019) recommends shared or striped bike lanes for the lowest speeds and volumes, while bike lanes are recommended for low speeds and low to moderate volumes. For rural highways, it is recommended to have shared lanes, paved shoulders, and shared use paths. Shoulder width was a key element to accommodate bicyclists, and guidance was provided to select shoulder width based on the posted speed limit and traffic volume (ADT). Similar criteria are used in design guidelines from some state DOTs, with recommendations of shoulder widths depending on expected traffic volume and posted speed limit (MnDOT, 2007), facility type based on ADT and motor vehicle target speed (WSDOT, 2015) or LTS-based criteria based on ADT, shoulder width and posted speed limit (ODOT, 2020). It should be noted that in the case of rural facilities with a posted speed limit below 45 mph, the ODOT Multimodal analysis manual recommends using the methodology for urban segments based on the number of lanes, ADT, functional class, and posted or prevailing speed.
The other two core elements for respondents to select to cycle in each scenario were pavement conditions and context classification. Respondents ranked these two elements fourth and fifth in the explicit ranking, although their coefficients in the choice model were even higher than for traffic volume and speed limit.
More specifically, scenarios with deteriorated pavement, such as large potholes and deep cracks, were less likely to be selected than scenarios with acceptable or good pavement conditions. This agrees with the practitioner survey, as most respondents indicated agreement with the statement that rural cyclists perceptions of quality of service were very sensitive to pavement conditions in shared lanes and sensitive to bicycle lanes. The HCM-LOS is moderately sensitive to pavement conditions (difference around 1 LOS letter grade). Nevertheless, other design guidelines do not include pavement conditions under the criteria or as a consideration to modify segment
rideability. A notable exception is the ODOT Multimodal analysis manual that indicates that the presence of debris in the shoulder and/or poorly placed rumble strips may limit or prevent shoulder use and suggests that it should be coded as no shoulder if long-term debris is present in the shoulder (ODOT, 2015). This aligns with the comments to the user survey, where multiple respondents indicated rumble strips, gravel and debris as main issues while cycling on rural roads.
The impact of context classification was evaluated in the choice tasks. The experimental design included scenarios on rural, rural–town and suburban context to derive differences between suburban and more “pure” rural contexts. The impact of context classification was significant: rural conditions were less likely to be selected than rural–town and suburban conditions. However, the difference between rural–town and suburban contexts was lower than between rural and rural–town. This suggests that the current HCM analysis procedure for rural highways, which was adopted from arterials, may be debatable; and that different thresholds should be proposed for rural and rural–town context. In the current HCM, context classes are not accounted for, and the only element that related to activity-volume is percent of on-street parking. The HCM-LOS is moderately sensitive to on-street parking (around 1 LOS letter grades), which aligns responses of practitioners to the influence on with the moderate sensitivity of practitioners' responses to on-street parking, commercial areas.
Surprisingly, terrain-related elements, such as elevation, grade, or sequence of curves, ranked low. Furthermore, the results of the choice task for grade were mixed. While the individual analysis showed a lower propensity to select high-grade scenarios, the estimated binomial logit model indicated that the higher the grade, the higher the log-odds of choosing that scenario, given that all other attributes remained constant. This could be explained as grade was not explicitly mentioned in the scenarios and may not have been well distinguished by respondents and used as a proxy of scenery attractiveness: scenarios with high grade usually had rolling terrain with trees, compared to scenarios with low grade that were flat in semi-desertic conditions. Therefore, respondents could have interpreted grade as more pleasant scenery rather than challenging terrain and low-grade scenarios, which could be perceived as monotonous (and therefore less attractive) to cycle. On the other hand, interested, more confident cyclists (either enthused and confident or strong and fearless) provided lower importance in the ranking to terrain and all-year-round regular cyclists.
This result does not agree with the practitioner survey, where grade, sight distance, and curvature were believed to be sensitive to the quality of service for rural cyclists, in the respondents’ opinion. Furthermore, field studies indicated that grade length and radii of horizontal curves play an essential role in bicyclists’ speed selection (Llorca et al., 2015) and comfort level (Noel et al., 2003). The Oregon DOT Multimodal analysis manual indicates that a “high frequency of sharper curves and short vertical transitions can increase stress, especially on roadways with less than 6’ shoulders. Engineering judgment may be needed to determine what impact this will have on the BLTS level on a particular segment” (ODOT, 2020).
Environmental conditions received mixed opinions. At the same time, some respondents found environmental conditions among the most essential elements when cycling; the majority ranked them in the bottom half. The practitioner survey responses were also divided about
environmental conditions: the influence on perceived quality of service of environmental-related elements was very sensitive, but environmental conditions were considered within the least important groups. The latter aligns with the current evaluation methodologies (HCM, LTS, BCI), where environmental conditions are not accounted for. Moreover, weather conditions may not be available to most agencies and administrations and including them in the analysis should be considered with caution. Also, the typical bicycle trip characteristics showed that even all-year-round cyclists do not cycle much under adverse weather conditions, with a third of all-year-round cyclists cycling in the rain and less than 16% cycling with snow or fog. The frequencies of potential or seasonal regular cyclists are even lower, with less than 3% cycling with snow or fog and 12% with rain.
This survey used two classifications of cyclists: 1) cyclist type by cycling frequency and 2) stated cyclist type. The first classification follows the logic from Félix et al. (2017) that uses past cycling frequency (whether they cycled in the last year and month) and if they consider cycling again in the next month/year to classify them. Three classes are distinguished: regular cyclists, potential cyclists, and non-cyclists. The second classification is based on the stated level of confidence and follows the four types of cyclists defined by Geller (2006) and refined by Dill and McNeil (2016): no way no how, interested but concerned, enthused and confident, and strong and fearless. Survey respondents were asked which one of them defined better as cyclist.
Interestingly, cycling frequency captured more differences in the ranks of the elements. All-year-round cyclists ranked lower clearance, pavement quality, environment, and terrain than seasonal regular cyclists, potential cyclists, or non-cyclists. This is consistent with a higher tolerance to cycle under adverse weather conditions, which can indicate more experience and higher cycling skills. These criteria are used by the new edition of the AASHTO Green Book (2018) to segment design cyclists: comfort level, bicycling skill, and experience.
Individuals who stated to be enthused and confident were more likely to select to cycle in each scenario, compared to selecting “none of them,” and individuals who were strong and fearless. It can be observed that the likelihood of strong and fearless was higher than for enthused and confident, as expected. However, the highest difference was between interested but concerned and enthused and confident. Furthermore, the sensitivity to the attributes of the scenario was equal across all groups, as interaction terms were not statistically significant. Cycling frequency was not statistically significant, indicating that, for the choice task, stated cyclist types were a better predictor to evaluate the likelihood of cycling in each scenario.
Stated cyclist types aligned better with the reported comfort level in different cycling environments. More specifically, the results show apparent differences by stated cyclist type, with an increased level of comfort for the enthused and confident group compared to the interested but concerned group and for the strong and fearless group compared to the enthused and confident group. This difference was linear. Contrarily, cyclist types by cycling frequency provided mixed results: while regular cyclists were more comfortable in bike lanes and trails, non-cyclists were more comfortable in environments without bike lanes. Moreover, the differences among groups were minor and could result from the low sample size of non-cyclists. In this sense, the classification by confidence level better captures the comfort of cyclists while cycling in urban environments in contrast with their cycling frequency.
The distribution of cyclists was compared to the results from Dill and McNeil (2016). While Dill and McNeil (2016) represented the entire spectrum of cyclists in urban areas, this study targeted cyclists on rural highways. Most of the respondents were enthused and confident cyclists.
Afterward, we compared the gender and commute mode distribution across the cyclist groups. To avoid the original bias resulting from the sampling process, we first calculated the percentage of being in each cyclist type for each category (e.g., male and female) under that variable and then compared the difference between different sociodemographic categories across cyclist types.
The percentage of males in the strong and fearless group and enthused and confident group was larger than females in both surveys, while other categories were the opposite. However, the gap between males and females in each category was more apparent in this survey. This could be explained by cycling on rural roads more appealing to males than females. In the survey, the percentage of strong and fearless and enthused and confident cyclists was larger than in other modes, while walking dominated the survey from Dill and McNeil. For the other two categories, the percentage of bicycles is much smaller than other modes in this survey, but only a little smaller than other modes in 2015’s survey. One possible reason could be that people living in metropolitan areas usually have shorter commute distances. This also indicates that the cyclist type significantly impacts commute mode for those who cycle on rural roads.
After comparing the sample distribution, the level of comfort from survey respondents was compared with the level of comfort collected by Dill and Mc Neil (2016) and the proposed LTS-based criteria from Furth et al. (2016). In their work, Furth et al. (2016) proposed that scenarios with trails, bike lanes, or striped bike lanes in residential streets are compatible with interested but concerned cyclists, while scenarios with striped bike lanes in major streets or scenarios without bike lanes in residential streets would be compatible with enthused and confident cyclists. Major streets without bike lanes would only be compatible with strong and fearless cyclists. The following table collects the percentage of respondents that indicated that a given scenario was comfortable or very comfortable for them by stated cyclist type.

In the current survey, the percentage of respondents being comfortable with a scenario decreased as the scenario reduced the cycling facilities and increased the number of lanes. The results were consistent across all cyclist groups. However, in the results from Dill and McNeil, the trends were unclear: the percentage of strong and fearless cyclists comfortable on trails was lower than that in striped bike lanes in major arterials. Furthermore, in scenarios with bike lanes, the percentage of comfortable responses was higher for enthused and confident than for strong and fearless. The criteria by Furth et al. (2016) agrees with which scenarios were comfortable for at least 50% of cyclists: scenarios without bike lanes in major arterials were comfortable for 100% of strong and fearless, but for 40% of enthused and confident. Similarly, scenarios without bike lanes were compatible with enthused and confident but not interested but concerned. This survey reflects other results: while most of interested but concerned were comfortable in scenarios with a bike lane, only one-third was comfortable in minor streets with striped bike lanes. Moreover, between 42% and 50% of enthused and confident were comfortable in scenarios with striped bike lanes; compared to 84% to 100% in the study from Dill and McNeil. Similarly, their strong and fearless cyclists were more comfortable in major arterials with four lanes, but in the current survey, only 28% to 45% were comfortable.
The differences suggest that the criteria for the LTS in urban areas should be revisited for rural cyclists. Rural bicyclists may be more aware and likely to tolerate higher stress levels than urban cyclists. On the other hand, rural bicyclists may not be used to cycling in low-speed environments, where traffic disruptions may be more frequent and therefore may generate higher stress levels, as one participant indicated in the open question: “PS: those StreetMix scenarios aren’t very rural. Given the size of those buildings, one is in a large city and roads are going to
be busy. Even with a protected bike lane, I would still feel uncomfortable due to the volume of traffic and all of the distracting stimuli assaulting the drivers and leaving less cognitive room for them to be looking out for me. Hence all my answers are” “very uncomfortable”.
“The four Geller categories are bogus because each of them combines two different characteristics which are not necessarily related. I am old, not a strong rider but I am fearless, because I am skillful. Enthused and confident? Many bicycling advocates are enthused, but not confident. Interested but concerned? many are concerned, but not interested (they always worry about my safety), and vice versa (they would ride if it were more convenient). No way, no how: this is a slang expression describing aversion but in fact there are people from whom there is no way, due to disability, and others for whom there is no how, due to inconvenience. By using the slang expression, Geller leaves out both of these important deterrents to bicycling.”
“I’m a vehicular cyclists and feel comfortable riding with traffic however I find myself less enthusiastic about riding on roads since the Covid19 pandemic. I have been leading multi-day bicycle tours for 20 years and have observed that traffic is traveling faster and faster on rural roads previously with low traffic volumes. Increased speed and traffic volume has made me look at rural roads more critically when considering a rural route for a bicycle tour. I judge roads, rural or otherwise, by my comfort level and more importantly the comfort level of my participants/guest.”
An online survey was conducted to identify the drives and constraints on rural cycling for commuting or leisure for different user types. More than 1000 complete responses were collected across the United States. The results suggest which variables are more relevant to users to cycle (or not) on rural highways and which are highly valued. Moreover, it collected cycling comfort levels in different cycling urban environments. The influence of individual-specific attributes, such as age, gender, income, commute mode, or cyclist type, on the results was tested. For these, two classifications of cyclist types were used in the analysis: by cycling frequency and by stated cycling comfort (or level of traffic stress-based). Results were compared with previous research and design guidelines.
The main conclusions of the study are:
width are ADT and posted speed limit. High speed limits could be tolerated by cyclists if automobile traffic is very low, reinforcing the importance of a combined analysis of speed and traffic volume.
The conclusions of the study are limited to the designed questionnaire and the collected responses. The survey was targeted at individuals who cycle on rural highways, and therefore, the extrapolation of these results to the overall cycling population or urban cyclists in specific geographical areas should be undertaken with caution. Based on these conclusions, the recommendations of the study are:
This research points to the need for future discussions on the role of bicycle user segmentation for operational analyses and the need for a planning methodology. If the HCM analysis procedure aligns with the current LTS-based design criteria, different LOS thresholds should be considered for bicycle users. The most important characteristics have been identified, but specific criteria and thresholds still need to be defined. On the other hand, a planning methodology for bicycle mode based on the LTS should be developed based on users’ perceptions and incorporated into NCHRP Report 825: Planning and Preliminary Engineering Applications Guide to the Highway Capacity Manual (Dowling et al. 2016). Even though this
effort was intended for rural highways, it could also provide a sound basis for a potential review of cycling performance evaluation on urban streets.
Further comments on bicycle use in rural areas. You might like to comment on your cycling experience, your experience with other vehicles on the road and/or road conditions, points of interest, safety issues, etc. If you are not a cyclist, you can tell us what must change, or needs to improve for you to start cycling in rural areas.
Comments
traffic. I feel safer with flashing lights front and back and I think the “locals” look out for me. I’d be MUCH more comfortable with a bike path.
on their minds that watching out for bicycles. If there is a wide shoulder next to a road, it does not bother me.
maintain that 3 feet of usable space and be cleaned. If it’s full of debris and pebbles (usually caused by the mumble strip that wasn’t cleaned well), then I’m forced to ride on the road. If it’s a very low volume road, like many rural roads in the UP, then I don’t need a shoulder at all. To me, volume is the most important element of determining where shoulder work is necessary. It’s also important to provide “routes” or “connections” for bikers in rural areas. Bikers can travel hundreds of miles and a state needs to be aware of the most likely and safest passages (roads) bikers will use to get from A to B, then to C and D. etc.
recovered 2 years later. If I can’t find a multi-use path, I ride my mountain bike on the trails in the woods.
road decisions. PennDOT verbalizes all the right things but when construction occurs it’s usually the same as in the past. There are wins for active transportation users but usually it’s because of pressure. Active user accommodations should be automatic.
car travel may be enjoyable for leisure but won’t do much to reduce reliance on cars. I also found driver experience and respect for cyclists to be a more important factor than the actual infrastructure in many cases. For example, I traveled through Kansas almost exclusively via US 36, which did not have great clearances for cycling, however, I found drivers to be more attentive and respectful of my presence on the road than anywhere else I have ever ridden. While I took great care to watch my back and be cautious to stay safe, I found the stretch quite enjoyable. Lower speed limits can help, but often make little difference if drivers think they are too low, they will often drive faster and more recklessly, especially when they encounter cyclists that they may see as just another hindrance. Careful placement of rumble strips is another important consideration. To allow for safe cycling, there needs to be enough space to the right of the strip for safe biking, without having to frequently cross the strip. There should be breaks in the strip to allow for safe crossing for bikes as well. In areas with few options for providing adequate clearance, such as tunnels, I found buttons that activate a signal alerting drivers of the presence of cyclists do help.
ratchets up to meet the same BLTS as previously. Instead of an increase in traffic volume however, it would likely translate into increased speeds.
curve when they have no clue if a car is approaching from other directions, drivers who pass riders on a hill and have no clue if a car is approaching. A friend and I were almost taken out last summer by a car passing on a hill and ignoring the no passing line markings—a car was coming the other direction and we bailed off the shoulder. All I can say is guardian angel—blessing we were not hit.
has only gotten busier, which has significantly reduced the number of my trips. It is rare to see a cyclist on the roadway today, because of the safety factor.
a traffic light. I will ride occasionally out of my way to avoid traffic, for a change of scenery, or to use separated trails/pathways. But with a family, I have less available time to take side-trips. Cycling is a convenient way to fit a moderate amount of exercise into my routine. And I’ve saved thousands of dollars in our gas budget over the years, reducing wear and tear on the car, using that mileage for longer trips where a vehicle is more efficient than around town.
other vehicle expenses, reduces my negative impact on our environment, and I enjoy the challenge of grocery shopping on my bike or pedaling in my office attire.
especially poison ivy, or that it may block me from a passing car). Too few trees are also a major issue, limiting riding after about 10:30 am in many regions in warm weather.
mostly keep to local highways with large shoulder areas so I can see and be seen while biking.
deaths due to drivers not looking out for cyclists. This, I believe, is the greatest risk to the lives of cyclists that use public roads for cycling.
my mapping earlier in the survey has a degraded paved shoulder that is less than 6 feet in width. City and County planning allowed a softball complex to be built north of Fargo and created a very dangerous situation in which motorists use a section of highway 81 to get to the softball field complex but did not make improvements to this section of county roadway before or while the 28-diamond softball complex was built, and usage began. I imagine no survey was done to see how Cass County Road 81 is used by bicyclists prior to approving the plans for the softball field complex. A definite disconnect between city/county planning and users of this roadway.
detriment. There are still a few random vehicles that will pass somewhat close or rev their engine when passing.
of their tour was, yet I saw that they kept that same stretch in their Death Valley Tour later in 2022 and in their 2023 tour. Go figure.
pass a cyclists/group of cyclists when there is possible/unseen oncoming traffic is a regular occurrence—this would be the biggest concern.
foot into the shoulder itself and for the shoulder not being raw chip seal or gravel. Cars should have room to give cyclists space, without a firm barrier on the inside of the lane. Parked cars opening doors are terrifying.
my few feet on the roadway with cars and trucks driving 55 mph! I seek out the few off-road trails where I’m near t afraid of being mowed down.
can cause motorist to dislike cyclist on the road. I live in a college town and the college students would benefit from understanding that they need to abide by traffic laws.
but it’s the combination of the hazards and the rider’s assessment of that combination that sets the rider’s comfort level. For casual users with little experience, separate bike paths/lanes are best, although they come with their own hazards.
if the conditions are too constrained. Impatience with waiting for a safe time to pass results in tailgating, close passing, or unpredictable behavior. Enhancements to the road shoulder or building side paths can help alleviate these issues, but education and enforcement are also pieces that need to be considered. Cycling routes through calmer back roads are usually preferred and can be comfortable to ride with few enhancements needed.
fog/shoulder lines (increases vehicle speed, poor vehicle lane position, difficulty selecting lane/shoulder position in inclement weather).
render many shoulders dangerous and discouraging for cyclists and often force us to ride in the main lanes of auto traffic.
has a crossing signal and has the right of way. The primary focus of US road design should not be on the LOS for drivers, but instead on the safety of nonmotorized users.