Click for next page ( 79


The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 78
D-1 APPENDIX D USER MOBILITY BENEFITS Trails, Lanes, or Traffic: Valuing Bicycle Facilities with an Adaptive Stated Preference Survey INTRODUCTION Wardman (195) investigated the demand for cycling facilities using stated preference in a route choice context. They found that individ- If bicycling is to be a viable mode of transportation, it must have uals were willing to pay a premium to use facilities that are deemed appropriate facilities. Evaluating what is appropriate requires an safer. The authors argue that increasing safety is likely more impor- understanding of preferences for different types of cycling facilities. tant than reducing travel time to encourage bicycling. In this study we explore and provide a quantitative evaluation of indi- Abraham et al. (196) also investigated cyclist preferences for vidual preferences for different cycling facility attributes. This under- different attributes using a SP survey again in the context of route standing can be incorporated into an evaluation of what facilities are choice. Respondents were given three alternate routes and their attrib- warranted for given conditions. utes and were then asked to rank the alternatives. The responses were The facilities considered here are A) Off-road facilities, B) In- analyzed using a logit choice model. Among other variables that were traffic facilities with bike lane and no on-street parking, C) In- of interest to their study, the authors found that cyclists prefer off- traffic facilities with a bike lane and on-street parking, D) In-traffic street cycling facilities and low-traffic residential streets. But the facilities with no bike lane and no on-street parking, and E) In-traffic authors also claim that this may be due to an incorrect perception of facilities with no bike lane but with on-street parking. The aim is to safety on the part of the respondents, and education about the safety understand what feature people desire by quantifying how many of off-road facilities may change the stated choice. additional minutes of travel they would be willing to expend if Shafizadeh and Niemeier (197) investigate the role that proximity these features were to be available. This added travel time is the to an off-road bicycle trail plays in route choice decisions. Using inter- price that individuals are willing to pay for the perceived safety and cept surveys along the Burke-Gilman trail in Seattle, they find that comfort the attributes provide. among people who reported origins near the off-road facility, travel A computer based adaptive stated preference survey was devel- time gradually increases as they are further from trail to a point and oped and administered to collect data for this study. To understand if then decreases, leading them to speculate that there may be a 0.5 to the value that people attach to attributes of facilities is systematically 0.75 mile "bike shed" around an off-road bike path, within which indi- related to different individual and social characteristics, the study has viduals will be willing to increase their travel time to access that facil- also collected demographic, socioeconomic, household, and current ity and outside of which a more direct route seems to be preferred. travel mode information from each participant. This information is Aultmann-Hall, Hall, and Beatz (198) use GIS to investigate then used to build an empirical model to evaluate relationships bicycle commuter routes in Guelph, Canada. While comparing the between these independent variables and the additional travel time shortest path to the path actually taken, they found that people that people are willing to expend for different attributes of cycling diverted very little from the shortest path and that most bicycle com- facilities. In addition to giving a measure of the appeal of the attrib- muters use major road routes. They found little use of off-road trails. utes under discussion, the model also highlights the social and indi- While this may be due to the location of the trails and the O-D pair vidual factors that are important to consider in evaluating what they connect, even in five corridors where comparably parallel off- facilities to provide. road facilities do exist to in-traffic alternatives, they found that Interest in studying bicyclists and cycling environments is grow- commuters used the in-traffic facilities much more often. Only the ing. Recent papers by a number of authors have investigated pref- direct highest quality off-road facility (one that is "wide with a good erences of cyclists and the bicycling environment as well as the quality surface and extends long distance with easy access points") relationship between the supply and use of facilities. Availability seemed to be used relatively more. of cycling facilities and the type and quality of a cycling facility Stinson and Bhat (199) using data from a web based stated pref- are important determinants of how well they are used. Studies by Dill erence survey estimate a logit model to understand important attrib- et al. (55) and Nelson et al. (174) have shown that there is a positive utes for commuter cyclist route choice. They find that respondents correlation between the number of facilities that are provided and the preferred bicycling on residential streets to non residential streets, percentage of people that use bicycling for commuting purposes. likely because of the low traffic volumes on residential streets. While While both studies state that causality cannot be proved from the data, their model showed that the most important variable in route prefer- Nelson and Allen (174) state that in addition to having bicycle facil- ence was travel time, the facility was also significant. It was shown ities, facilities must connect appropriate origins and destinations to that cyclists preferred in-traffic bike lanes more than off-road facil- encourage cycling as an alternative commuting mode. ities. Both facility types had a positive effect on utility but the for- Bovy and Bradley (194) used stated preference (SP) to analyze mer added more to utility than the latter. In addition they find that bicycle route choice in the city of Delft. Their work looked at facility cyclists try to avoid links with on-street parking. Another study by type, surface quality, traffic levels and travel time in route choice. Taylor and Mahmassani (200), also using a SP survey to investigate They found that travel time was the most important factor in route bike and ride options, finds that bike lanes provide greater incen- choice followed by surface type. Another study by Hopkinson and tives to inexperienced cyclists (defined as those with a "stated low

OCR for page 78
D-2 to moderate comfort levels riding in light traffic") as compared with interest. In this type of survey each option is presented based on more experienced cyclists, with the latter group not showing a sig- choices the respondent has already made. This allows for the pre- nificant preference to bike lanes over wide curb lanes. sentation of choices that the individual can actually consider The results from these papers seem somewhat mixed. Though while removing alternatives that the respondent will surely not some of the research has shown a stated preference and revealed consider. This methodology has been adopted in a number of con- preference with some constraints for off-road facilities, others have texts, including value of time for commercial vehicle operators shown that cyclists generally prefer in-traffic cycling facilities with (201), in mode choice experiments (202), and in evaluating tran- bike lanes. Especially in revealed preference cases, the apparent pref- sit improvements (203). erence for in-traffic routes may be due to their ability to connect to many destinations in a more direct fashion and therefore leading to a lower travel time. In addition, route choice may be restricted by facil- SURVEY INSTRUMENT, DESIGN, ity availability, geographic features or missing information. It may AND ADMINISTRATION also be that for people who regularly bicycle, who are most likely the subjects of the revealed preference studies, travel time and not per- All respondents of the ASP survey were given nine presentations ceived safety are likely of utmost importance, as these individuals are that compared two facilities at a time. Each presentation asks the more likely to be conditioned to the cycling environment. The actual respondent to choose between two bicycle facilities. The respondent preference therefore may not be for the in-traffic facility; however, it is told that the trip is a work commute and the respective travel time may be the best alternative available to the cyclists. they would experience for each facility is given. Each facility is pre- Commuter choices are clearly limited by facilities that are avail- sented using a 10-second video clip taken from the bicyclists' per- able to them. Understanding preferences and behavior is crucial to spective. The clips loop three times and respondents are able to providing choices that people desire. This can be best accomplished replay the clip if they wish. when the value of any given improvement in facility attribute is Each facility is compared with all other facilities that are theoret- known. Valuation of facility attributes can be done by considering ically of lesser quality. For example, an off-road facility (A) is what people are willing to pay for using these facilities. In this study compared with a bike lane no on-street parking facility (B), a bike we try to uncover this value by measuring how much additional lane with parking facility (C), a no bike lane and no parking facil- time individuals would be willing to spend bicycling between a ity (D), and a no bike lane with parking facility (E). Similarly, the given origin and destination if alternate facilities with certain attrib- four other facilities (B, C, D and E) are each compared with those utes were available to them. facilities that are theoretically deemed of a lesser quality. The less In the next section we present the methodology in detail. This is attractive of the two facilities is assigned a lower travel time and the followed by a description of the survey instrument and design. The alternate (higher quality) path is assigned a higher travel time. The analysis methodology is presented, and then the results. respondent goes through four iterations per presentation with travel time for the more attractive facility being changed according to the previous choice. The first choice set within each presentation assigns METHODOLOGY the lesser quality facility a 20-minute travel time and the alternate (higher quality) path a 40-minute travel time. Travel time for the The methodology we follow to extract this valuation of attrib- higher quality facility increases if the respondent chose that facility, utes uses an Adaptive Stated Preference (ASP) survey. While both and it decreases if the less attractive facility was selected. A bisection revealed and stated preference data can be used to analyze prefer- algorithm works between 20 and 60 minutes either raising or lower- ences, there are certain advantages to using the latter method in this ing the travel time for the alternate path so that it becomes less attrac- case. In using consumer revealed preference, often a limitation arises tive if it is chosen or more attractive if the shortest path is chosen. By because only the final consumer choice is observed. This makes it dif- the fourth iteration, the algorithm converges on the maximum time ficult to ascertain how consumers came to their final decision. This difference where the respondent will choose the better facility. This complication arises because the number of choices that are available way the respondent's time value for a particular bicycling environ- to each consumer may be very large, and information on those ment can be estimated by identifying the maximum time difference alternatives that went into an individual's decision may not be fully between the two route choices that he/she will still choose the more known. Even in cases where all possible alternatives are known, it is attractive facility. Pictures of these facilities are shown on Figure 13. difficult to assess whether the decisionmakers considered all avail- Figure 14 maps the locations of the facilities where the videos were able alternatives. In addition, the exact tradeoff of interest may not be taken in St. Paul, Minnesota. readily available. Even in cases where the tradeoffs seem to be avail- The procedure used to converge on the time trade-off for the par- able, one cannot be certain that the consumer is acting out his prefer- ticular facility is illustrated as follows. If the subject first chose the ence for the attributes we are observing. The lack of appropriate data longer option, then the next choice set assigns a higher travel time for can pose a major challenge in this respect. the higher quality path (raised from 40 minutes to 50 minutes). If the Stated preference surveys overcome these complications because respondent still chooses the longer option, the travel time for that the experimenter controls the choices. In SP settings, the experi- choice increases to 55 minutes and the choice is posed again. If on the menter determines the choices and the respondent considers. While other hand, the 50-minute option is rejected and the respondent chose this may not reflect the actual market choice that individual would the 20-minute route, the bisection algorithm will calculate a travel make because of the constraints the survey places on the choice set, time that is between the now rejected option and the previously it allows us to measure attribute differences between the presented accepted option, in this case 45 minutes. By the time the respondent alternatives. Further, by using specialized forms of SP such as ASP, makes a fourth choice, the survey will have either narrowed down the one can measure the exact value individuals attach to attributes of respondent's preference to within 2 minutes or the respondent will

OCR for page 78
D-3 (A) Off-road bicycle facility (B) Bike lane, no parking (C) Bike lane, on-street parking (D) Bike lane, no parking (E) No bike lane, on-street parking Figure 13. Cycling facilities used in the study. Figure 14. Location of facilities used in the Adaptive Stated Preference survey. (Note: (A) off-road facility; (B) bicycle lane, no parking facility; (C) bicycle lane, on-street parking facility; (D) no bicycle lane, no parking facility; (E) no bicycle lane, on-street parking facility.)

OCR for page 78
D-4 TABLE 16 Facility pairs compared in the ASP survey Base Route E B C D No bike lane Bike lane, no Bike lane with on- No bike lane, no with on-street parking street parking parking parking A off-road T1 T2 T3 T4 B Alternate routes Bike lane, no parking N/A T5 T6 T7 C Bike lane with on-street parking N/A N/A N/A T8 D No bike lane, no parking N/A N/A N/A T9 Ti represents the average additional travel time users are willing to travel. have hit the maximum travel time that can be assigned to the longer survey and another 91 people participated in the summer survey, trip, which is 58.5 minutes. Table 16 shows the pairs of comparisons making a total of 181 people. Among these, 13 people had to be that were conducted and used in the analysis. Table 17 shows a removed due to incomplete information, leaving 168 people. Of sample series of travel time presentations and Figure 15 shows these 168, 68 people indicated that they have bicycled to work at least sample screenshots of the survey instrument. once in the past year. Thirty-eight of these 68 identified themselves The survey was administered in two waves, once during winter and as regular bicycle commuters at least during the summer. Also, 127 once during summer. The winter and summer respondents were of the 168 people said they have bicycled to somewhere, including shown video clips that reflected the season at the time of the survey work, in the past year. Further demographic information on the taken at approximately the same location. Our sample for both waves respondents is given in Table 18. comprised employees from the University of Minnesota, excluding students and faculty. Invitations were sent out to 2,500 employees, randomly selected from an employee database, indicating that we MODEL SPECIFICATION AND RESULTS would like them to participate in a computer based survey about their commute to work and offering $15 for participation. Participants Switching Point Analysis were asked to come to a central testing station, where the survey was being administered. A total of 90 people participated in the winter The adaptive nature of the survey allows us to extract the actual additional minutes each individual is willing to travel on an alternate facility. In the context of the survey, this is the maximum travel time TABLE 17 Choice order for a sample beyond which the respondent would switch to use the base facility. presentation For each pair of facilities that are compared during the summer and the winter, the averages of this switching point are computed and Facility Travel plotted in Figure 16. On average, individuals are willing to travel Time more on an alternate facility when the base facility is E (undesig- Presentation Route 1 Route 2 Choice nated with on-street parking), followed by D (no bike lane without Route parking) and C (bike lane with parking). For example, individuals are willing to travel further on facility B when the base facility is E, choice set 1 40 min 20 min 2 as opposed to D or C. Route Figure 16 shows the hierarchy between facilities clearly--each of choice set 2 30 min 20 min 1 the lines plotted connects the average additional travel time that indi- Route viduals are willing to bicycle over the 20 minutes that they would have bicycled if they had chosen the base facility. For example, look- choice set 3 35 min 20 min 1 ing at the winter data, the top solid line connects the average addi- Route tional time individuals say they would travel on an alternate facility choice set 4 37 min 20 min 2 when the base facility is E (in-traffic with parking at 20 minutes). The alternate facilities are as shown on the horizontal axis. For exam- Ti 36 min ple, on average respondents are willing to travel about 22 additional

OCR for page 78
D-5 Figure 15. Top: comparing designated bicycle lanes with no parking with in-traffic bicycling with no parking. Bottom: same presentation three iterations later.

OCR for page 78
D-6 TABLE 18 Demographic distribution of respondents Model Number of subjects 168 We start with the economic paradigm of a utility maximizing indi- Sex vidual, where given a bundle of goods the individual chooses that % Male 34.5% % Female 65.5% bundle which results in the highest possible utility from the choice set. In the current context then, given two alternatives, the chosen Age Mean (Std. deviation) 44.19 (10.99) alternative is the one that the respondent derives a higher utility from. We can then break down each bundled alternative to its components Us u al mode (Year ro u nd) to understand what amount each contributes to utility. This will %Car 69.7% enable us to extract the contribution of each feature of the facility in %Bu s 18.5% the choice consideration of the individual. Mathematically, we would %Bike 9.2% state this as alternative A is selected if UA is greater than UB, where %Walk 2.6% A and B are the alternatives and U is the utility function. We hypothesize that the utility a user derives from using a bicycle Bike commu ter facility depends on the features of the facility and the expected travel All season 9.2% time on the facility. Choices are also affected by individual char- Su mmer 22.6% acteristics that we may not directly observe but can try to estimate using individual specific variables such as income, sex, age etc. As HH income discussed earlier, each individual records a response over various < $30,000 8.3% alternatives, and therefore the data reflects the repeated choices over $30,000 - $45,000 14.3% the same respondent. This implies that the errors are no longer inde- $45,000 - $60,000 19.6% pendently distributed. To overcome this problem one can use a gen- $60,000 - $75,000 15.5% eralized linear mixed model which would estimate a random effect $75,000 - $100,000 20.2% for the between-subject effect, thus separating the within-subject and $100,000 - $150,000 17.9% between-subject errors. Both subject random effects are assumed to > $150,000 4.2% have a normal distribution with zero mean and separate variances. The error term of the utility's linear component is assumed to have HH Size a Gumbel distribution. The model's linear utility component is spec- 1 25.0% ified as follows: 2 32.7% 3 16.7% U = f (Facility, Travel Time, Season, Individua al Variables) 4 20.8% > 4 4.8% The utility of a particular alternative can be written as follows: U iA = ViA + iA minutes if an off-road bike path is available if the alternative is to ViA = 0 + 1WiA + 2O iA + 3BiA + 4 PiA + 5TiA + 6SiA + 7 A iA bike in traffic. We can further describe the data by employing tech- + 8 I iA + 9 H iA + 10 CiA niques such as the non-parametric bootstrap. The bootstrap approx- imates the sampling distribution of the mean by repeatedly sampling Where: with replacement from the original data. We employ the nonparam- etric bootstrap where no prior assumptions are made on the distrib- W = Weather (winter = 1, summer = 0) ution of the statistic. The bootstrap approach was first developed by O = dummy indicating whether the facility is off-road (1 = Yes, Efron in 1979 (204). 0 = No) Consider the histogram shown in Figure 17, it reflects the addi- B = dummy indicating whether the facility has a bike lane (1 = tional travel times individuals in the sample said they would travel Yes, 0 = No) between facilities A (off-road) and C (in traffic with parking). It is dif- P = dummy indicating whether on-street parking is absent or ficult to make any distributional assumptions based on the observed present (1 = absent, 0 = present) sample. Employing the nonparametric bootstrap on this data with T = Expected travel time on the facility being considered 5,000 resamples (Figure 18), we can see that the bootstrap distrib- S = Sex (Male = 1, Female = 0) ution of the mean is very close to normal, and hence a normal inter- A = Age val can be built around it. The bootstrap distributions of all nine pairs I = Household Income (Inc/1000) of comparisons lead to very symmetric distributions that show no evi- H = Household Size (>2 = 1, Otherwise = 0) dence of non-normality. The percentile confidence interval based on C = Cyclist at least during summer (Yes = 1, No = 0) the actual 2.5% and 97.5% values of the bootstrapped mean are also = estimable coefficient computed. The bootstrap also allows us to estimate the bias of the Gumbel (0, ) sample mean. The sample mean, the estimate of the bias and the con- fidence interval (CI) using the normal distribution and the percentile To interpret the model appropriately it is important to note how of the bootstrap are reported in Table 19 for each pair of comparisons the dummy variables are coded (Table 20). Variable B represents both for the combined and season specific data. whether a facility has a designated bike lane, O represents whether

OCR for page 78
D-7 Maximum Additional Travel Time Between Facility Pairs (Combined Data) 25.00 Base 20.00 Facility B Travel Time 15.00 C D 10.00 E 5.00 0.00 A B C D Alternate Facility Maximum Additional Travel Time Between Facility Pairs (Winter data) 25.00 Base 20.00 Facility Travel Time B 15.00 C D 10.00 E 5.00 0.00 A B C D Alternate Facility Maximum Additional Travel Time Between Facility Pairs (Summer Data) 30.00 Base 25.00 Facility 20.00 Travel Time B C 15.00 D 10.00 E 5.00 0.00 A B C D Alternate Facility Figure 16. Hierarchy of facilities. (Note: (A) off-road facility; (B) a bike lane, no parking facility; (C) a bike lane, on-street parking facility; (D) a no bike lane, no parking facility; (E) a no bike lane, on-street parking facility.) the facility is off-road, and P represents whether a facility has no odds of choosing the theoretically better facility. Choices depend on parking adjacent to it. This would allow separately valuating bike the attributes of the facilities, the travel time the user experiences on lanes as well as being off-road. It should be observed that `O' is not the facilities, and individual characteristics. The signs of the esti- equivalent to an off-road trail. `B' and `O' together constitute an mated parameters are as expected. The travel time is negative show- off-road trail. ing an aversion to longer trips. The improvements (off-road, bike lane The parameter estimates of binomial logit model are given in and no parking) all have a positive and significant influence on choice Table 21. The model is estimated such that the results indicate the of different magnitudes. Of these three, a bike lane improvement

OCR for page 78
D-8 indicating that preferences are not dictated by experience at least in 35 this SP context. The model also tells us that older individuals have 30 higher odds of choosing the better quality facility. Also, individuals whose household size is greater than two have lower odds of choos- ing the better quality, longer travel time facility. This may be because 25 these individuals have higher constraints on their time than individ- uals who live in single or two person households. Frequency 20 The estimates of a linear utility model can be used to determine the value of an off-road facility, a bike lane facility and a facility with no 15 parking in terms of the time cost of travel. These are derived using the marginal rate of substitution between each of the facility features and 10 travel time (Table 22). These values are derived based on SP ques- tions that have a 20-minute base travel time, and should be interpreted 5 as such. Accordingly, a bike lane improvement is valued at 16.3 min- utes, a no parking improvement is valued at 8.9 minutes and an off- 0 road improvement is valued at 5.2 minutes. This is to say, keeping 0 10 20 30 40 utility at the same level, one can exchange the off-road improvement for 5.2 minutes of travel time, a bike lane for 16.3 minutes of travel time time and a no parking improvement for 8.9 minutes of travel time. This says that the most value is attached to having a designated bike Figure 17. Distribution of additional travel time lane. While having an off-road facility would certainly increase the for facility C over facility A. utility of the individual, most of the gains of an off-road facility seem to be derived from the fact that such facilities provide a designated bike lane. The absence of parking is also valued more than taking the increases the odds much more than a parking elimination or that of facility off-road. an off-road improvement alone. An alternate specification of the model looks at time as a depen- The season variable is negative and significant, indicating that dent variable and features of the facility as independent variables people have lower odds of choosing the higher travel time facility along with demographic covariates. This specification also employs during winter than during summer. Looking at the individual covari- a mixed models approach to account for the repeated measurements ates that are used, income and sex are not significant at the 0.10 level; taken over the same subject. The dependent variable is the switching however the signs seem to indicate that women have a higher ten- point travel time minus the base facility travel time. This approach dency to choose the facilities that are perceived safer (better quality) yields similar patterns in the order of valuation of the different attrib- than men (p-value = 0.11); and higher incomes seem to be associated utes of the facilities and the expected directions of the parameter with a tendency to choose the better quality facility (p-value = 0.11). estimates. A side by side comparison of the two model coefficients The cyclist variable, which indicates if the respondents use bicycling is not possible; however, we can compare the values derived for dif- as their main mode at least during summer, is highly insignificant; ferent facility pairs based on our logit model and the linear model (Table 23). This is given in Table 24 and Figure 19. As can be seen, most comparisons are very close to one another in magnitude. As Figure 19 shows, the results derived from the logit model more 400 closely replicate what is observed in the raw data, even though that is not always the case across the nine comparisons. The overall assessment of the models suggests that designated 300 bike lanes seem to be what are desired the most. It is also important to consider that both the linear and logit models found no evidence Frequency against the possibility that preferences between cyclists and non- 200 cyclists are the same. This is encouraging in many respects, because it avoids the dilemma of which interest to serve. The policy impli- cation is that by addressing this common preference, we can ensure 100 cyclists receive the facilities they prefer and non-cyclists get the facilities that they could at least consider as a viable alternative. 0 CONCLUSION 14 16 18 20 This appendix analyzes preferences for different cycling facilities time using a computer-based adaptive stated preference survey with first person videos. Using the survey on 168 randomly recruited individ- Figure 18. The bootstrapped mean for the uals, we derive the values that users attach to different cycling facil- additional travel time between facilities A and C ity features and expose which are most important. The choice data (based on 5,000 resamples). were collected based on individual preferences between different

OCR for page 78
D-9 TABLE 19 Mean additional travel time between facility pairs and confidence interval of the bootstrapped distribution of the mean Original Normal 95% Percentile 95% Fac1 Fac2 Mean Bias Standard Error CI CI Combined Data A B 14.21 0.0223 0.962 (12.30, 16.08) (12.41, 16.17) A C 16.00 0.0136 0.964 (14.10, 17.88) (14.16, 17.92 ) A D 18.46 -0.0160 0.984 (16.55, 20.41) (16.58, 20.40) A E 23.14 -0.0051 0.939 (21.30, 24.98) (21.26, 24.94) B C 10.13 0.0092 0.973 (8.21, 12.03) (8.25, 12.06) B D 13.73 -0.0008 0.957 (11.85, 15.61) (11.90, 15.62) B E 20.87 0.0245 0.956 (18.97, 22.72) (19.09, 22.84) C E 19.65 -0.0033 0.950 (17.79, 21.51) (17.79, 21.49) D E 18.25 0.0211 1.002 (16.27, 20.20) (16.35, 20.22) Winter Data Original Normal 95% Percentile 95% Fac1 Fac2 Mean Bias Standard Error CI CI A B 15.33 0.0208 1.335 (12.69, 17.92) (12.78, 18.00) A C 13.69 0.0339 1.327 (11.06, 16.26) (11.21, 16.40) A D 17.57 -0.0252 1.344 (14.96, 20.23) (14.99, 20.19) A E 20.66 -0.0025 1.319 (18.08, 23.25) (18.16, 23.28) B C 6.17 -0.0064 1.197 (3.83, 8.52) (3.97, 8.57) B D 10.86 -0.0244 1.180 (8.57, 13.19) (8.58, 13.25) B E 17.45 -0.0101 1.248 (15.02, 19.91) (15.02, 19.91) C E 17.39 -0.0097 1.264 (14.92, 19.87) (14.98, 19.92) D E 15.72 0.0074 1.270 (13.22, 18.20) (13.22, 18.22) Summer Data Original Normal 95% Percentile 95% Fac1 Fac2 Mean Bias Standard Error CI CI A B 13.04 -0.0051 1.338 (10.43, 15.67) (10.49, 15.74 ) A C 18.43 0.0146 1.353 (15.76, 21.07 ) (15.84, 21.16 ) A D 19.40 0.0079 1.434 (16.58, 22.20 ) (16.58, 22.25 ) A E 25.73 -0.0071 1.292 (23.21, 28.27 ) (23.18, 28.27 ) B C 14.28 0.0154 1.397 (11.53, 17.01 ) (11.63, 17.10 ) B D 16.75 -0.0128 1.481 (13.86, 19.66 ) (13.89, 19.68 ) B E 24.46 -0.0072 1.332 (21.85, 27.07 ) (21.78, 27.06 ) C E 22.03 0.0013 1.403 (19.27, 24.77 ) (19.30, 24.82 ) D E 20.92 -0.0055 1.485 (18.01, 23.83 ) (17.96, 23.82 )

OCR for page 78
D-10 TABLE 20 Coding for facility features Facility O B P A (Off-road) 1 1 1 B (Bike lane, No parking) 0 1 1 C (Bike lane, on-street parking) 0 1 0 D (In traffic, No parking) 0 0 1 E (In traffic, on-street parking) 0 0 0 TABLE 21 Logit model Random effects: Group Variance Std.Dev. subject 1.550 1.245 Fixed effects: Variable Description Estimate Std. Error z value Pr(>|z|) (Intercept) 0.620 0.472 1.315 0.1885 Season (1 = winter, W 0 = summer) 0.627 0.207 3.028 0.0025 ** T Travel time 0.051 0.004 12.685 0.0000 *** Offroad O Improvement? 0.264 0.060 4.386 0.0000 *** Parking P Improvement? 0.456 0.065 7.067 0.0000 *** Bikelane B Improvement? 0.831 0.067 12.475 0.0000 *** A Age 0.021 0.010 2.126 0.0335 * Sex S (1 = M, 0 = F) 0.350 0.223 1.567 0.1171 I Income 0.005 0.003 1.584 0.1132 HHsize ( if>2, 0 H otherwise) 0.594 0.229 2.589 ** Cyclist (1 = atleast C summer, 0 = No) 0.133 0.253 0.524 0.6003 TABLE 22 Time values of facility attributes Marginal Rate of Attribute Substitution (minutes) O Off street improvement 5.20 P Parking improvement 8.98 B Bike lane improvement 16.36

OCR for page 78
D-11 TABLE 23 Linear model Random Effects (Intercept) Residual StdDev: 8.98 8.01 Fixed effects: Description Value Std. Error t-stat p-value (Intercept) 7.24 3.377 2.143 0.032 Yes=1 W Season Winter? No=0 4.13 1.485 2.782 0.006 * * * Offroad Yes=1 O Improvement? No=0 2.38 0.429 5.540 0.000 * * * Parking Yes=1 P Improvement? No=0 3.50 0.456 7.673 0.000 * * * Bikelane Yes=1 B Improvement? No=0 5.98 0.456 13.127 0.000 * * Yes=1 A Age No=0 0.15 0.071 2.092 0.038 * Male=1 S Sex Female=0 3.36 1.604 2.093 0.038 * I Inc/1000 0.03 0.021 1.475 0.142 >2=1 H Household Size 2=0 3.75 1.645 2.278 0.024 * Yes=1 C Summer cyclist? No=0 2.22 1.818 1.221 0.224 Significance ***0.001 **0.01 *0.05 +0.1 TABLE 24 Comparison of travel time values between facilities using the linear model and the logit model Comparison Facility 1 Facility 2 Logit Linear Mean (raw data) 1 A B 5.2 9.6 13.0 2 A C 14.2 13.1 18.4 3 A D 21.6 15.6 19.4 4 A E 30.5 19.1 25.7 5 B C 9.0 10.7 14.3 6 B D 16.4 13.2 16.7 7 B E 25.3 16.7 24.5 8 C E 16.4 13.2 22.0 9 D E 9.0 13.2 20.9

OCR for page 78
D-12 35.0 30.0 25.0 Travel Time Logit 20.0 Linear 15.0 Raw data 10.0 5.0 0.0 1 2 3 4 5 6 7 8 9 Comparision (See Table 24) Figure 19. Comparison of the estimates of the additional time willing to travel between facility pairs based on logit model, linear model, and the raw data. facilities having different travel times, but the same origin and desti- users are willing to pay the highest price for designated bike lanes, nation. From the raw data we have demonstrated that a hierarchy followed by the absence of parking on the street and by taking a bike exists between the facilities considered, and we have extracted a mea- lane facility off-road. In addition, we are able to extract certain indi- sure of how many additional minutes an individual is willing to vidual characteristics that are indicative of preferences such as age expend on an alternate facility if it were available and provided cer- and household structure and make loose connections with sex and tain features that were not available on the base facility. The data were household income. Such an understanding can be incorporated into then used to fit a random parameter logit model using a utility maxi- the planning process to help planners make appropriate recommen- mizing framework. A linear model was also estimated and compared dations and investment decisions in developing bicycle facilities that with the results from the mixed logit model. The results show that are more appealing to the public.