Crossing Solutions at Roundabouts and Channelized Turn Lanes for Pedestrians with Vision Disabilities (2011)

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

43 This chapter is devoted to the results of the field trials with blind study volunteers at the selected treatment sites. It pres- ents a summary of findings that are most pertinent to the understanding of the specific crossing challenges at the site, as well as the effect of the installed crossing treatments. Appendix A contains more detailed results for each site, and the reader is encouraged to refer to that material for a more in-depth dis- cussion. The results are presented sequentially for channelized turn lanes, single-lane roundabouts, and two-lane round- abouts. The analysis uses the analysis framework, event defini- tions, and performance measures defined in Chapter 4. The chapter concludes with a summary comparison of all field trials and some discussion items related to the results. Channelized Turn Lane Study Overview The field study at the CTL location focused on two crossing treatments. The treatments were (1) sound strips (SS) that were intended to increase the awareness of pedestrians of approaching vehicles and (2) sound strips in combination with a pedestrian-actuated flashing beacon (FB) that was intended to increase driver yielding behavior. In the following discussion, the crosswalks will be identified by treatments installed as SS-ONLY and SS+FB, respectively. Both turn lanes were further supplemented with lane delineators that were intended to prevent late merges into the turn lane. All treat- ments, including the lane delineators, were installed between pretest and posttest. A more detailed description of the site and treatments is given in Chapter 3. Crossing Performance Results The evaluation of pedestrian crossing performance used the measures defined in Chapter 4: the availability of crossing opportunities in the form of yields and crossable gaps, the rate of utilization of these opportunities, pedestrian delay, and the rate of O&M interventions. The field evaluation at the CTL location generally showed that participants experienced significant delay and risk. Despite the fact that only a single lane needed to be crossed, the com- bination of background noise at the busy intersection and high approach speeds in the turn lane caused high delays and frequent interventions. A comparison of the two crossing loca- tions in the pretest did not show significant differences for most of the measures, giving confidence that the two loca- tions allow for a valid comparison of the different treatments installed. The starting order of participants was randomized as to which crossing they started on. No consistent and signif- icant differences were identified between crossing attempts from the curb versus from the island, although some partici- pants stated that crossing from the island was harder due to the traffic noise behind them. The pretest study was completed in May 2008, and a total of 16 blind travelers participated. Fourteen of the original 16 participants returned for the posttest study in November 2008. The treatments were installed in early October 2008, allowing six weeks for driver adaptation. Sound-Strips-Only Treatment (SS-ONLY) Table 1 summarizes the crossing performance for the CTL crossing that only had sound strips and lane delineators installed. The figures were obtained by averaging the mean crossing performance of each of the study participants at this location. Table 1 shows that the installation of the sound strips by themselves did not have a large or significant impact on yield and CG crossing opportunities. This confirms the underlying research hypothesis since the treatment was primarily intended to aid with the utilization of opportunities. The probability of utilizing a yield surprisingly decreased from 50.8% to 40.5%, although this difference was not statistically significant. A closer C H A P T E R 5 Results

look at the yield utilization illustrates that the percentage of yield events that were forced by the pedestrian decreased from 11.3% to 6.4% (not shown in Table 1 but discussed in Appen- dix A). A forced yield occurs when a pedestrian steps out onto the roadway prior to the driver initiating the yielding process. The degree of risk associated with these events depends on the relative position and speed of the vehicle at the time of cross- ing initiation. Forced yield events should therefore not neces- sarily be interpreted as poor or risky decisions. However, the reduction in the occurrence of forced yields (though not sig- nificant) may suggest that the sound strips were successful in assisting the pedestrians to distinguish yielding and non- yielding vehicles. However, this hypothesis should be confirmed with additional research with this treatment. The SS-ONLY treatment further increased the rate of cross- able gap utilization by approximately 8%. However, similar to other measures, this increase was not statistically significant. In general, all observed measures experienced a large degree of variability across participants, making it challenging to sta- tistically validate small changes in a performance measure. Overall, the installation of the sound strip and lane delineator treatments at the SS-ONLY turn lane did not have a large impact on most of the availability and utilization performance measures when aggregated for all participants. The treatments did seem to reduce the average and 85th percentile pedestrian delay, as well as the delay over minimum. Similar to the effects above, these decreases were not statistically significant but sug- gest promise for the sound strips with some modifications or in combination with another treatment. This CTL crossing had the highest observed occurrence of interventions of any of the test locations, with the highest intervention rate per participant reaching 30% (six interven- tions in 20 crossings). The pretest intervention rate corre- sponded to 30 individual interventions recorded among 10 of the 16 participants. Of these 30 interventions, seven were asso- ciated with a stopped car on top of the crosswalk. For these cases, vehicles were queued back from the downstream merge point and did not leave the crosswalk unoccupied. Participants seemingly were not able to hear the fact that a car was stopped due to the high ambient noise at the busy intersection. For the remaining 23 interventions, vehicles were moving and there- fore represented situations were a potential collision would have been more severe. After treatment installation, only eight interventions were observed for five of the 14 posttest participants. This corre- sponds to a significant reduction of O&M interventions, from 9.4% to 2.9%, which is noteworthy and further establishes that the treatment has some potential. The reduction of inter- ventions supports the hypothesis that the sound-strip treat- ment improved the ability of the study participants to audibly interpret the vehicle patterns in the turn lane. However, even an intervention rate of 2.9% remains risky, as it corresponds to one potentially dangerous event in 34 crossing attempts. In the pretest condition the intervention rate corresponded to a risk chance of approximately one in 11 crossing attempts. The analysis does suggest that additional treatments may be necessary at this location, due to a combination of high volumes, high speeds, background noise, and driver disregard of crosswalk laws and pedestrians. Sound Strip plus Flashing Beacon Treatment (SS+FB) Table 2 summarizes the crossing performance for the CTL crossing that had sound strips, lane delineators, and a pedestrian-actuated flashing beacon installed. The figures were obtained by averaging the mean crossing performance of each of the study participants at this location. No consistent and significant differences were identified between crossing attempts from the curb versus from the island. As hypothesized, Table 2 shows that the supplemental installation of the flashing beacon resulted in an increase in yielding from 15.2% to 22.0%. While statistically significant, the resulting posttest yielding rate of 22.0% is still considered low. The observed yielding rate with the flashing beacon installed is generally within the range of findings from a national survey of the yield performance of overhead flashing beacons at unsignalized midblock crossings (Fitzpatrick et al. 2006). 44 Performance Measure Pre Post Difference p-value Yield Opportunities 18.40% 18.60% 0.20% 0.2728 CG Opportunities 34.90% 41.20% 6.30% 0.1666 Yield Utilization 50.80% 40.50% –10.30% 0.2878 CG Utilization 60.30% 68.20% 7.90% 0.4238 Average Delay (s) 26.2 18.5 –7.7 0.1898 Delay>Min (s) 15.6 11.7 –3.9 0.4224 85th Percentile Delay (s) 40.9 32.7 –8.2 O&M Interventions 9.40% 2.9% –6.5% 0.0204 – Table 1. Crossing performance summary, pretest and posttest, at SS-ONLY CTL.

In a synthesis of results from 17 sites, the authors recorded an observed range in the average yielding rates of from 13% to 91%, which suggests that other contributory factors (speeds, driver population, regional differences) affect yielding behavior. Consistent with the research hypotheses, the treatment did not have a significant impact on the availability of crossable gaps. The rates of utilization for yield and crossable gap oppor- tunities both saw a significant increase of 11.5% and 26.2%, respectively. Only the CG utilization is statistically significant, but both trends are surprising in light of the fact that the SS-ONLY treatment did not show the same results. A possible explanation is proposed: It is hypothesized that the auditory message that accompanied the flashing of the beacon gave pedestrians additional confidence. The primary intent of the treatment was to alert the approaching driver to the pedestrian’s crossing intent and increase yielding. However, the audible message provided an auditory confirmation to the pedestrian that the beacon was flashing and appears to have contributed in some way to the willingness of pedestrians to utilize cross- ing opportunities. However, it’s not clear that the pedestrians were able to confirm the opportunity auditorily. The O&M specialist who was monitoring participants was concerned that the auditory message masked some vehicular sounds and could encourage crossings that were more risky because some participants seemed to rely more on the vehicles yielding, with- out confirmation, when the flasher was present. In the tested combination, the SS+FB treatments had a large effect on the pedestrian delay as well as a reduction in O&M interventions. Average delay and the delay over mini- mum were both reduced significantly, by 10–11 s, and the 85th percentile delay was reduced by over 20 s. O&M inter- ventions were reduced significantly, from 5.6% to 1.4%, which reduced the likelihood of a potentially dangerous event from one in 18 decisions to one in 71. The raw count of interven- tions was 14 in the pretest and four in the posttest conditions, and both counts are lower than at the SS-ONLY crossing. It is unclear what the specific reason for this difference in inter- ventions is between the two crossing locations. Of the 14 pretest interventions, only one was associated with a stopped car on the crosswalk, while the SS-ONLY crossing resulted in seven of these events. Given that the two locations had similar traffic volumes and that the starting order for participants was ran- domized, a likely contributory explanation for the difference in intervention rates is related to sound patterns at the crossing. Appendix A gives additional results for the CTL crossings and treatments, including a discussion of inter-participant variability. Participant Feedback Following each of the pretest and posttest studies, partici- pants were asked a series of questions about their perceptions of the crossing and their level of confidence in their crossing decisions. The blank survey forms are provided in Appendix G. While participants seemed somewhat more confident during the treatment conditions, their feedback generally indicated some uncertainty in crossing at the CTL locations. However, this level of uncertainty didn’t seem to quite reach the level that may be expected from the high delay and intervention rates at this site. The average of the responses on the confidence ques- tion (“How would you rate your confidence in your ability to cross here safely on a scale of 1 to 5, with 1 being not at all and 5 very confident?”) was 3.66 in the pretest, and 4.0 and 4.1 in the SS-ONLY and SS+FB posttests, respectively. These numbers seem to indicate a false sense of participant confidence in their ability, if measured by the rate of interven- tions. One participant who rated her confidence in crossing safely at 4 said: “but maybe it’s too dangerous. I knew you were behind me to grab me.” Another said: “It’s a little risky; I could make mistakes.” Although the team instructed participants to cross only if they would do so if they were alone, without an O&M specialist, it seemed that they felt they should try to cross to help us in our research. Participants repeatedly were reminded that even the fact that they reached the 2-min time- out or decided not to cross were valuable research findings. However, all participants attempted to cross and very few peo- ple actually timed out. It is possible that the 2-min time limit was longer than participants were willing to wait. 45 Performance Measure Pre Post Difference p-value Yield Opportunities 15.20% 22.00% 6.80% 0.0363 CG Opportunities 44.70% 49.20% 4.50% 0.444 Yield Utilization 53.10% 64.60% 11.50% 0.2769 CG Utilization 63.20% 89.30% 26.10% 0.0011 Average Delay (s) 23.4 12.2 –11.2 0.0453 Delay>Min (s) 14.9 4.9 –10.0 0.0342 85th Percentile Delay (s) 38.6 17.9 –20.7 – O&M Interventions 5.6% 1.40% –4.2% 0.0112 Table 2. Crossing performance summary, pretest and posttest, at SS+FB CTL.

While participants said yes in the CTL pretest to a question about whether they would use these crossings if they were on the route home from work (14 yes, 2 no), their responses often had qualifying remarks such as one person who said yes, then added “but I’d probably try to find another way, but most of the corners around here are like this.” In the SS-ONLY and the SS+FB conditions, the responses were roughly the same, with 10 yes, 3 no, and 1 no answer in the SS-ONLY and 9 yes, 2 no, and 3 no answer in the SS+FB condition. From the pretest, the participants’ responses to the question “Do these crossings need anything to increase safety and usabil- ity? If so, what would you suggest?” may provide more insight into their confidence than their ratings on the confidence ques- tion. Unlike the responses at the Charlotte single-lane round- about (same city and some of the same participants), almost everyone made a suggestion regarding detecting gaps or yields, which included: • “Tripper that lets you know that cars are coming, like old gas station tubes”; • “Better indications to the drivers to yield, or a cop to bust people for not yielding”; • “Flow indicators like a signal box to tell you if something is coming, but have to know how fast though so not sure how well that would work”; • “Audible signals to tell me when to cross, but don’t know how that would be done for the whole thing though because that would delay traffic”; • “Would be helpful to have auditory cue that cars were rolling across further down the lane, like a single speed bump or something”; • “Something to make them slow down”; • “Modify to push button to change light and make them stop for that lane too; very short sound/message to know its working”; and • “Audible signal that lets you know when cars are yielding; can’t trust them all to stop for you.” To the same question in the posttest, several individuals responded very positively to the sound strips, mainly suggest- ing that the sound strips needed to be louder and more con- sistent in providing a sound when vehicles were moving at a slower speed. An interesting comment from one individual was that she was more confident in the pretest, but she’d real- ized during the posttest that she’d just not heard some of the cars the first time. In the SS condition, she heard the vehicles approaching from further away and realized that she’d been stepping out in front of cars in the channelized lane in the pretest condition. This realization was accurate and matches the performance data for this participant. In the pretest, a car sometimes entered the CTL just as this participant started crossing. Since she walked very quickly, and since drivers noticed her, she was able to complete most of these crossings in the pretest without an intervention. In this participant’s case, the delay went up in the SS-ONLY posttest, attributed to a self-diagnosed increased awareness of traffic patterns due to the sound strips. Another interesting issue is that eight participants said they could hear better to make crossing decisions from the curb as opposed to from the island. They stated that the sound of traffic behind them made the crossing decision more difficult when waiting on the island. That is an interesting point because it is expected that pedestrians are more visible to drivers when on the island because drivers are predominantly focusing to their left and on the traffic they will be merging into. Several participants also commented that they could hear the sound strips better on the SS-ONLY corner than they could on the SS+FB corner. This was also noted by the O&M specialist and other members of the research team. From the testing it is unclear what caused this perceived sound difference. Traffic volumes (and presumably noise) were higher at the SS-ONLY corner, which suggests that the perceived sound difference is unlikely to be traffic related. One possible expla- nation is the fact that the SS-ONLY approach was on a slight down slope in the approach of the crosswalk, while the SS+FB corner was on an upslope. Assuming that sound is projected horizontally as vehicles traverse the sound strips, the SS-ONLY sound waves would be projected more directly toward the participants’ ears. Due to the upslope at the SS+FB crosswalk, the sound waves would be projected toward the pavement, which may have contributed to the perceived difficulty of hear- ing the sound strips at that approach. Participants in the posttests were asked several questions about the sound strips and about the push-button locator tone and audible information on the flashing beacon (Table 3). For these questions, participants were asked to rate the extent of their agreement, with 1 being strongly disagree and 5 strongly agree. The questions about the sound strips were about both cor- ners, so the problems some participants had in hearing the strips on the southeast corner can be expected to have influ- enced their ratings. Two qualified their responses about the sound strips by “when I could hear them.” Others commented on not hearing slower cars on the strips. Overall, the responses to the treatments were mixed, as was noted in the performance measures. Impact on Vehicular Traffic The channelized turn lane enabled vehicular traffic to bypass the busy intersection of a four-lane and a six-lane arterial. Both CTL locations featured long 275- and 300-ft deceleration lanes measured from the beginning of the crosswalk. These allowed right-turning traffic to avoid any through-movement queues 46

that may have formed at the main intersection. Consequently, the average CTL vehicle delay (in the absence of pedestrians) was low compared to other movements, and drivers expected largely free-flowing operations. The observed vehicle free-flow speeds in the pretest con- dition were high at both CTL locations, with an average 32.8 mph upon entering the turn lane and 21.5 mph as vehi- cles crossed the plane of the crosswalk, determined from laser speed measurements. The crosswalk speeds were lower during signal phases that resulted in traffic moving downstream of the CTL (opposing left-turn and cross-street through phases). During those phases, average free-flow speeds at the cross- walk were 18.5 mph, while the average during other phases was 22.4 mph (p < 0.0001). This difference points to the fact that drivers are well aware of downstream traffic conditions. This behavior further causes concerns regarding driver atten- tiveness to pedestrians at the crosswalk since the distance between the back of the crosswalk and the downstream merge point is only approximately 50 ft. It is one possible explana- tion for the low observed yielding rate at this site. With the installation of the sound strip and flashing beacon treatments, no significant effects on speeds were detected in most segments of the turn lanes. A slight increase in speeds was measured just upstream of the crosswalk (increase from 21.5 to 22.5 mph, p = 0.0536), but this change is not considered to be a notable impact on driver behavior. Generally, no large queuing impacts were detected as a result of drivers yielding to pedestri- ans. In some cases, queues from the downstream merge point spilled back across the crosswalk in both pretest and posttest conditions, resulting in a few crossing attempts in between stopped cars. In several instances, participants were not aware of a stopped vehicle on the crosswalk, resulting in O&M inter- ventions as discussed above. Given the low yielding behavior, the pedestrian presence and treatment installation were not considered to have a notable effect on traffic operations. Single-Lane Roundabout Study Overview The scope of NCHRP Project 3-78A originally included only one single-lane roundabout site for a pretest and posttest study, consistent with other locations. However, following the pretest, the research team was directed not to pursue with treatment installation and a posttest at the site. The site was not considered suitable for treatment installation due to par- ticipant responses, low intervention rate, and low traffic vol- umes during the pretesting. Consequently, the research funds were reallocated to revisit available video data collected at a single-lane roundabout in Raleigh, NC, and to study an addi- tional single-lane roundabout in Golden, CO. The prior data collection at the Raleigh site was comparable to the NCHRP Project 3-78A studies since the same general data collection protocol was used. The Golden site was selected for supple- mental study due to its proximity to the studied two-lane roundabout. The site was studied twice (concurrent with the two-lane roundabout pretest and posttest), yet no treatments were installed. This approach served two objectives: (1) the ability to compare the crossing ability of the same participants at a single-lane and two-lane roundabout, and (2) to test for a learning effect of the same participants repeating the study without any treatment installation. This section presents the crossing performance results sequentially for all three locations. It then discusses the impact on vehicle traffic and a summary of participant survey responses concurrently for all three locations. The Charlotte, NC, single-lane roundabout (DAV-CLT) was studied in the fall of 2007, and data from a total of 10 blind participants were used in the analysis. Even though there were a total of 19 participants in the study, the remaining data were not available for analysis due to video malfunction. For the 47 Rating questions Average of responses (N=13) 1=strongly disagree, 5=strongly agree The sound strips helped me know when vehicles were approaching. 4.38 The sound strips helped me know when vehicles were slowing down. 3.54 The sound strips helped me know when vehicles had yielded. 2.92 The sound strips made me confident that I was starting to cross at a safe time. 3.46 Where there were beacons installed, I’d push the button each time I wanted to cross. 4.08 Knowing the beacon was flashing made me more confident that I was starting to cross at a safe time. 3.83 The speech message didn’t interfere with my ability to hear traffic. 4.33 The locator tone on the beacon helped me know I was coming to the crosswalk. 3.77 The locator tone helped me go straight across the crosswalk. 2.64 The locator tone helped me know I was approaching the end of the crosswalk. 2.83 Table 3. Participant responses to treatment effectiveness at CTL.

Golden single-lane roundabout, the pretest was completed in July 2008 with a total of 18 blind travelers (GOL-PRE). Thirteen of the original 18 participants returned for the posttest study in September 2008 (GOL-POST). Again, no treatments were installed at this roundabout, so the underlying hypothesis is that overall performance in pretest and posttest conditions is the same. The Raleigh site was originally studied in a separate research project (NIH 2010) in 2004, with video recordings re- analyzed as part of this project using the NCHRP Project 3-78A analysis framework (PS-RAL). The analysis included a total of 12 blind study participants. Appendix A gives additional results for these locations, including a discussion of between-participant variability. Crossing Performance Results Charlotte, NC, Single-Lane Roundabout The field evaluation at the Charlotte single-lane roundabout location (DAV-CLT) yielded mixed results. On the one hand, participants experienced very little risk in terms of O&M inter- vention rates; on the other hand, delays experienced by the participants were high. Table 4 summarizes the crossing per- formance for the DAV-CLT site. The figures were obtained by averaging the mean crossing performance of each of the study participants at this location. Table 4 shows a very low occurrence of yield opportunities at only 6.3%. The rate of encountering crossable gaps was higher, at 28.8%, which is explained by generally low traffic volumes at the site except for some peak-hour traffic surges. The rate of utilization of both yields and crossable gaps was about two thirds of all encounters, suggesting that despite low volumes and associated low levels of ambient noise, partic- ipants had some difficulties at this location. From the rate of utilization it is unclear whether these opportunities were missed (i.e., the pedestrian failed to detect them) or rejected. In fact, several participants indicated that they were aware of the presence of, for example, a yielding vehicle, but nonetheless chose to wait. At low conflicting volumes, the additional wait time was generally rewarded by an all-quiet period that made crossing very comfortable. This type of behavior may further explain why the observed delay was relatively high at an average of 25.3 s. The delay over minimum was also high, at 18.0, due the aforementioned inef- ficiency in utilizing crossing opportunities. The 85th percentile delay was expectedly even higher at 35.4 s. Further, these delay figures mask the fact that some pedestrians experienced even higher delays. The maximum average delay experienced by a single pedestrian was 74.0 s over 12 trials. The maximum average delay over minimum was 59.4 s. The site further exhibited an overall low rate of O&M inter- ventions. In fact only two interventions were observed over the 120 crossings at the entry and exit leg, corresponding to the stated rate of 0.8%. The two interventions (one at entry and one at exit) were recorded for two different participants. The remaining participants therefore did not experience any interventions at this site. This analysis points to mixed results in terms of the ques- tion of roundabout accessibility. Clearly, a low intervention rate speaks to a safe crossing. However, the elevated levels of delay are indicative of significant inconvenience for pedestrians who are blind and further may lead people to avoid this cross- ing location. A further cause for concern is that traffic volumes may increase and reduce the availability of crossable gaps while presumably increasing the level of ambient noise. It is unclear whether yielding behavior would be affected with higher volumes or would stay the same. As discussed, no treatments were tested at this location, but at higher volumes, a treatment intended to increase driver awareness and propensity to yield may be appropriate. Raleigh, NC, Single-Lane Roundabout The field evaluation at the Raleigh single-lane roundabout location (PS-RAL) was intended to test the hypothesis that the findings at the DAV-CLT site were biased by low conflicting traffic volumes. The PS-RAL exhibits higher traffic volumes and is further located in a campus environment with more frequent pedestrian activity. It was hypothesized that driver expectations of encountering pedestrians would be elevated, which would result in an increased propensity to yield. Table 5 summarizes the crossing performance for the PS-RAL site. The figures were obtained by averaging the mean crossing performance of each of the study participants at this location. Table 5 in fact shows a relatively high occurrence of yield opportunities at 33.0%. The rate of encountering crossable gaps was lower than at DAV-CLT, at 19.1%. Combined, over half of all vehicle encounters corresponded to crossing opportunities. At higher traffic volumes, this results in relatively frequent crossing opportunities over time. The rate of utilization of yields was very high at this site, at 85.4%. This may be explained by 48 Performance Measure Mean Std. Dev. Yield Opportunities 6.3% 4.9% CG Opportunities 28.8% 6.8% Yield Utilization 67.4% 42.3% CG Utilization 63.3% 19.3% Average Delay (s) 25.3 13.8 Delay>Min (s) 18.0 12.8 85th Percentile Delay (s) 35.4 – O&M Interventions 0.8% 2.6% Table 4. Crossing performance summary at DAV-CLT.

slower speeds that result in an overall reduced level of ambient noise. The rate of crossable gap utilization was just below 60%, which is in the same region observed at the DAV-CLT site. A combination of frequent crossing opportunities and high utilization rates expectedly results in relatively low delay estimates. The average delay was 11.1 s, and the delay over minimum was only 5.8 s. The 85th percentile delay was some- what higher at 19.6 s. As with other sites, individual participants experienced higher average delays, with the maximum average observed delay at 34.2 s (24.7 s delay over minimum). However, the site exhibited a higher rate of O&M interven- tions, at 3.9%. This rate corresponds to a total of 15 interven- tions, 11 of which were observed at the exit portion of the crossing. The interventions were further distributed across nine of the 12 participants, suggesting that some risk was evident for most of the participants. One participant experienced five interventions across 32 crossing attempts, equivalent to a rate of 15.6%. In interpretation of these intervention statistics, it is important to highlight that a different O&M specialist accom- panied the participants in this prior study. Therefore, while the experimental protocol was generally the same, it is possible that individual differences of the O&M specialist contributed to the difference in interventions. The analysis shows the opposite trend of the previously studied DAV-CLT site. Due to the relatively frequent availabil- ity of yields and good utilization of crossing opportunities, the observed delays at PS-RAL were relatively low and further do not seem too different from what a sighted pedestrian might have experienced (see Delay>Min). However, the crossing decisions were characterized by a high intervention rate, which raises concerns for pedestrian safety. Since this study was per- formed as part of a different research project, no treatments were tested. Golden, CO, Single-Lane Roundabout The Golden single-lane roundabout was studied concur- rently with a nearby two-lane roundabout. Two rounds of field testing were performed at the same time as the single-lane roundabout pretest and posttest treatment conditions. Even though no treatments were installed at the two-lane round- about, the two test phases are referred to as GOL-PRE and GOL-POST, respectively. Table 6 summarizes the crossing per- formance for the site. The figures were obtained by averaging the mean crossing performance of each of the study partici- pants at this location. Table 6 shows no significant differences in the operational and safety performance during the pretest and posttest condi- tions. It can therefore be concluded that no measurable partic- ipant learning effect took place between the two studies and further that driver behavior and traffic patterns were similar. Both conditions showed a high occurrence of yield opportuni- ties (over 40% of vehicle encounters) along with some cross- able gap opportunities (over 20%). The utilization rates for both yields and crossable gaps were high, at or above 80%. As a result, the average delay experienced by participants was low (around 12 s), and only a very small portion of that delay was due to missed opportunities (delay over minimum was about 3 s). The variability in delay was higher at this site, and the highest average delay experienced by a participant was 51.4 s. The intervention rate shows an apparent drop from 1.4% to 0.5% between pretest and posttest conditions, but this slight difference is not significant. Overall, five interventions were recorded, four of which occurred in the pretest condition. 49 Performance Measure Mean Std. Dev. Yield Opportunities 33.0% 16.6% CG Opportunities 19.1% 9.2% Yield Utilization 85.4% 17.3% CG Utilization 57.8% 34.4% Average Delay (s) 11.1 7.8 Delay>Min (s) 5.8 6.4 85th Percentile Delay (s) 19.6 – O&M Interventions 3.9% 5.8% Table 5. Crossing performance summary at PS-RAL. Performance Measure Pre Post Difference p-value Yield Opportunities 40.4% 43.8% 3.4% 0.6398 CG Opportunities 23.5% 21.3% –2.2% 0.5800 Yield Utilization 79.4% 84.7% 5.3% 0.4897 CG Utilization 85.1% 81.2% –3.9% 0.6368 Average Delay (s) 11.9 12.1 0.2 0.9544 Delay>Min (s) 2.8 3.1 0.3 0.8156 85th percentile Delay (s) 22.2 21.7 –0.5 – O&M Interventions 1.4% 0.5% –0.9% 0.2651 Table 6. Crossing performance summary at GOL-PRE and GOL-POST.

There may be some evidence that participant decision mak- ing in terms of risk taking improved; however, this claim is not supported by the other performance statistics (mostly the utilization rates). Since interventions are very rare events, it is likely that this apparent reduction is the result of random variability in this measure. At the PS-RAL and DAV-CLT single-lane crossings, the queues that formed behind the yielding driver were safely contained behind the stopped car. However, a wide exit lane at the Golden single-lane roundabout caused some drivers to overtake the stopped car, creating a multiple-threat situation for the pedestrian. This behavior was also observed in relation to buses stopped at a downstream bus stop at this exit-lane crossing. While the intervention rate at this roundabout was low, this particular aspect of driver behavior causes concern for pedestrian safety. The issue of multiple-threat situations, where a near-lane yielding vehicle blocks the visual (and audi- tory) connection between the pedestrian and a far-lane driver, was hypothesized for two-lane roundabouts. Due to the wide lane widths, it also proved a concern at this exit-lane crossing. No multiple-threat events were observed at the entry leg. The findings from the GOL-PRE and GOL-POST studies first and foremost give confidence that any observed treat- ment effect at the nearby two-lane roundabout was largely unrelated to a participant learning effect. Since the same par- ticipants performed the studies at both sites, the lack of an effect at this single-lane location supports the notion that the tested two-lane treatments indeed improved the crossing task, as is discussed below. In comparison with the other single-lane roundabouts, the GOL-PRE/POST site exhibits similar performance to the PS- RAL site in the accessibility criteria of crossing opportunities, opportunity utilization, and delay. However, its safety per- formance is more similar to the DAV-CLT site, with a gener- ally low intervention rate. Overall, this site appears to be the most accessible of the three tested single-lane roundabouts with low delay and risk. Participant Feedback Following each of the single-lane roundabout studies, par- ticipants were asked a series of questions about their percep- tion of the crossing and the level of comfort in their crossing decisions. Participant feedback from the Raleigh roundabout was not available to the research team. The blank survey forms for the Charlotte and Golden roundabouts are provided in Appendix G. Feedback from the Charlotte and Golden single-lane round- abouts indicated that participants generally felt comfortable with the crossing task at those locations in each round of testing. The average of the responses on the confidence ques- tion (“How would you rate your confidence in your ability to cross here safely on a scale of 1 to 5, with 1 being not at all and 5 very confident?”) was 4.40 in Charlotte, 4.41 for the Golden pretest, and 4.46 for the Golden posttest. While Golden par- ticipants seemed more confident to the research team at the single-lane location in the posttest round, that impression was not supported by the participant survey results. On the question “Would you use these crossings if they were on the most direct route home from work?” (slightly dif- ferent wording in Charlotte), nine participants said yes and one said “no, would find another way” in Charlotte; for the Golden pretest 15 said yes, one no, one “maybe, depending on time of day,” and one did not answer that question; and in the Golden posttest, all 12 said yes, including one who had said no in the Golden pretest. In Charlotte, participants were asked about yielding vehicles and their reasons for crossing or not crossing when vehicles had yielded. (This question was not separated from the ques- tion about cues used for crossing in the Golden debriefing questionnaires.) Most comments indicated that participants found it difficult to decide if the drivers were yielding for them or something else (possibly other vehicles at entry lanes) and were concerned about how long drivers would stay stopped. Some stated that they waited for a person to actually roll down the window and talk to them before they would cross in front of a yielding vehicle. Some said they were unaware that a vehi- cle had yielded until after they had completed their crossing and heard the vehicle move behind them. In other words, they had not heard the vehicle at all and thought they were crossing in a gap in traffic. In response to a question about modifications that might be needed to improve accessibility, the main concerns were with wayfinding, although a couple of participants suggested a signal of some sort to indicate when vehicles had yielded. Several said the roundabout was pretty accessible as is, or that it didn’t need anything. The Charlotte participant who said she wouldn’t use the crossings stated that she was still con- fused about finding the crosswalks and lining up correctly and not so concerned about the traffic and crossing. Golden participants commented on the narrow splitter island and the fact that no detectable warnings were installed. These geo- metric features raised concerns about being sure they stopped on the island before crossing the next lane of traffic. Impact on Vehicular Traffic Since no treatments were tested at the single-lane round- abouts, any impact on vehicle traffic is strictly attributable to the normal interaction between pedestrians and drivers. While no formal delay studies were performed at these loca- tions, the research team noted no significant queuing impacts that resulted from the presence of the blind study participants. 50

The largest pedestrian-induced vehicle delay impact was evi- dent at the PS-RAL site due to a combination of high traffic volumes and frequent background pedestrian (student) traffic. Background pedestrian traffic was present at all three locations but was much lower at the other two. Thus, the overall delay impact is marginal relative to vehicle delay upon entering the roundabouts. At all locations, the team would occasionally observe a deter- mined yielding driver, who would wait 10 or more seconds to allow the pedestrian to cross, which would then cause some vehicle delay. Any queues that formed as a result of pedestrian presence dissipated quickly once the pedestrian was out of the crosswalk. In fact, buses that stopped downstream of the exit portion of the tested crosswalks at PS-RAL and DAV-CLT fre- quently caused more significant queuing than the pedestrians. Two-Lane Roundabout The field study at the two-lane roundabout location focused on two crossing treatments. The treatments were (1) an RCW that was intended to reduce vehicle speeds and encourage yielding behavior, and (2) a PHB, also known as a HAWK sig- nal. The PHB is intended to stop traffic at a red signal indica- tion while minimizing vehicular delay because it is not a full signal, and to supply auditory information to the pedestrian via APS. All treatments were installed between the pretest and posttest. A more detailed description of the site and treatments is given in Chapter 3. Crossing Performance Results The evaluation of pedestrian crossing performance used the measures defined in Chapter 4: the availability of crossing opportunities in the form of yields and crossable gaps, the rate of utilization of these opportunities, the 85th percentile pedes- trian delay, and the rate of O&M interventions. As discussed in that chapter, the analysis approach for two-lane crossing was revised slightly to reflect the fact that pedestrians are faced with two conflicting lanes. For ease of understanding, this chapter combines the three yield classifications (rolling, stopped, and forced) and only discusses differences where necessary. The reader is encouraged to refer to Appendix A of this report for a more detailed evaluation of the results. Concurrent with the two-lane roundabout data collection, participants also crossed at a nearby single-lane roundabout. The use of this comparison site was intended to allow for a direct comparison between the same participants’ abilities to cross at a single-lane versus a two-lane roundabout. It further allowed the team to test for a learning effect since no treatments were installed at the single-lane roundabout. The results of the single-lane roundabout study were discussed above, and no such learning effect was evident. The starting order of partici- pants was randomized as to which roundabout participants crossed at first. For crossings at the two-lane roundabout, the order of which of the two crosswalks was tested first was also randomized. The pretest was completed in July 2008, and a total of 18 blind travelers participated in the study. The treatment was installed following the pretest, and 13 of the original 18 par- ticipants returned for the posttest in September 2008. Raised Crosswalk Figure 17 summarizes the encountered and utilized events at the RCW location during the pretest. As discussed, events are shown for the condensed matrix of near lane and far lane that combines the different yield types. The raw event data is contained in Appendix A. The results in Figure 17 show that a total of 686 pedestrian– vehicle interaction events were observed in the pretest and that 443 of these (64.5%) were non-utilized events. This sug- gests pedestrian delay since blind pedestrians wait longer on average before utilizing an opportunity. The events are further divided into three vehicle states in the near lane in the categories of yield (183 events, 26.7%), crossable gap (175, 25.5%), and 51 Figure 17. RCW pretest event utilization. 0.0% 10.0% 20.0% 30.0% 40.0% 50.0% 60.0% Fa r L a n e= Y Fa r La ne =C G Fa r La n e = no nC G Fa r L a n e= Y Fa r La ne =C G Fa r La n e = no nC G Fa r L a n e= Y Fa r La ne =C G Fa r La n e = no nC G Near Lane = Y (n=183) Near Lane = CG (n=175) Near Lane = non- CG (n=328) Pe rc e n t o f T o ta l E ve n ts (n = 68 6) RCW-Pretest (n=686) Non-Utilized (n=443) Utilized (n=243) Y=Yield, CG=Crossable Gap, non-CG = Non-Crossable Gap This figure shows a bar chart of all observed pedestrian–vehicle events during the RCW pretest. The graph shows a total of nine event categories, representing all combinations of event outcomes yield, crossable gap, and non-crossable gap for two conflicting lanes (near lane and far lane).

non-crossable gap (328, 47.8%). For each near-lane category, the events are broken down by vehicle state in the far lane as well as by whether the particular event combination was utilized or not. Overall, the majority of events fell into the category of a non-crossable gap in both lanes, which can also be interpreted as drivers who did not yield to pedestrians. Expectedly, almost all of these events were not utilized by the pedestrians. The figure shows that some of the event combi- nations correspond to crossing opportunities (yield or cross- able gap) in both lanes. The utilization statistics show that most of those opportunities involving at least one crossable gap were utilized, but that almost a third of the dual-yield events were not. Special attention should be paid to the events with a near- lane yield and a far-lane non-crossable gap since these are related to the multiple-threat condition. This event combi- nation represents 7.6% of all events, and 5.8% of these events actually were utilized by pedestrians (0.4% of all events). These potentially risky events need to be interpreted in addition to the O&M interventions that are discussed below. Overall, 1.3% of events were utilized events with a non-crossable gap in one or both lanes. Figure 18 shows the corresponding event distribution for the RCW posttest. It shows a reduction in the number of events encountered by the pedestrian from 686 in the pretest to only 269 events. Accounting for the number of participants in the pretest (18) and posttest (12) studies, the resulting aver- age number of events was reduced from 38 to only 22 events over 16 crossing attempts per participant. Consequently, par- ticipants in the posttest had much fewer non-utilized events (62) than in the pretest (443). This points to much improved decision-making efficiency as well as more courteous and/or compliant driver behavior. A closer look at potentially risky events shows 3.0% of events that correspond to a potential multiple-threat situation with a near-lane yield and a far-lane non-crossable gap. Partic- ipants utilized 12.5% of these events, which represents 0.4% of all crossing events. Overall, 2.6% of events were utilized events with a non-crossable gap in one or both lanes. This suggests some potential risk at the crossing, in addition to the O&M interventions discussed below. A comparison of Figure 17 and Figure 18 further shows that with the installation of the RCW, the relative percentage of yield events increased, and pedestrians also encountered more yields in both lanes. In return, the percentage of non- crossable gap opportunities decreased, suggesting that fewer drivers proceeded through the crosswalk without stopping. The utilization of dual crossing opportunities also increased, which is surprising since no audible message was associated with this particular treatment. Following the discussion in Chapter 4, the nine event cat- egories were converted to the probabilities of encountering and utilizing a crossing opportunity in both lanes, PA_Dual and PU_Dual, as well as in only one of the lanes PA_Half and PU_Half. By definition, the remaining events correspond to non-crossable events in both lanes (PA_No and associated utilization rate PU_No). Table 7 summarizes these statistics for the RCW analysis and also presents the associated delay and intervention statistics as defined previously. Table 7 indicates that the availability of dual crossing oppor- tunities increased significantly, from 56.0% to 76.9%, after RCW installation. The utilization of these events was already high during the pretest condition (88.3%) but increased fur- ther to 98.1%. This rate of utilization is rather extraordinary considering that the blind participants make crossing deci- sions about two conflicting lanes based on auditory infor- mation alone. Due to the reduction in non-crossable events that was discussed in relation to Figure 18, the occurrences of PA_Half and PA_No events both decreased. The utilization rates of these opportunities did not change significantly. With more frequent and better utilized dual crossing oppor- tunities, the delay experienced by participants decreased sig- nificantly, from 17.0 s to 8.0 s. The maximum average delay experienced by a single participant also decreased, from 84.9 s 52 Figure 18. RCW posttest event utilization. 0.0% 10.0% 20.0% 30.0% 40.0% 50.0% 60.0% Fa r La ne =Y Fa r La ne =C G Fa r L a n e= no n CG Fa r La ne =Y Fa r La ne =C G Fa r L a n e= no n CG Fa r La ne =Y Fa r La ne =C G Fa r L a n e= no n CG Near Lane = Y (n=138) Near Lane = CG (n=80) Near Lane = non- CG (n=51) Pe rc en t o f T ot al E ve nt s (n = 2 69 ) RCW-Posttest (n=269) Non-Utilized (n=62) Utilized (n=207) Y=Yield, CG=Crossable Gap, non-CG = Non-Crossable Gap This figure shows a bar chart of all observed pedestrian–vehicle events during the RCW posttest. The graph shows a total of nine event categories, representing all combinations of event outcomes yield, crossable gap, and non-crossable gap for two conflicting lanes (near lane and far lane).

to 18.2 s. The delay improvements are also mirrored in the Delay>Min and 85th percentile delay statistics. From a crossing risk perspective, the installation of the RCW decreased the rate of O&M interventions from 2.8% to 0%. The 2.8% pretest interventions corresponds to a risk of 1 in 36 crossing attempts. The rate further represents eight individual observations distributed across six of the 18 par- ticipants. No interventions were observed for the 12 returning posttest participants. It is important to emphasize that four of these 12 did experience at least one intervention in the pretest condition. While the reduction of interventions to 0% is statis- tically significant, this number in all likelihood does not repre- sent an absolute zero. It should therefore not be interpreted as the RCW resulting in zero risk to blind pedestrians. It is very unlikely that any treatment would result in zero risk to blind pedestrians or sighted pedestrians, as evident by pedestrian injuries and deaths at intersections across the country. The 0% intervention rate therefore should only be interpreted as the fact that no interventions were observed during the 16 cross- ing attempts by each of the 13 participants during the posttest. Overall, the installation of the RCW resulted in drastic improvements in crossing performance at this location, both in terms of delay and safety. It is emphasized here that these find- ings are only representative of this one crossing location. How- ever, the results do seem to suggest that the RCW treatment shows a lot of promise at two-lane roundabout approaches. Prior research (Geruschat and Hassan 2005) has linked lower speeds to an increased likelihood of yielding, which likely explains the effect observed here. More surprising is that while the RCW seemingly improved the utilization of crossing opportunities, no treatment was installed that was intended to provide more information to blind pedestrians. Two possible explanations are offered: (1) The RCW gave pedestrians increased confidence in utiliz- ing crossing opportunities. It therefore may not have improved the ability of pedestrians to detect opportunities (yields), but rather increased their willingness to step in front of yielding vehicles. (2) Driver stopping behavior may have been altered with the RCW installation. In the pretest condition, many drivers were observed to slowly coast to a stop and to approach the crosswalk cautiously. This form of a rolling yield is difficult to distinguish auditorily from background noise at the round- about. In the posttest condition, driver yielding behavior seemed to be more rapid. In other words, cars quickly decel- erated to a stop in proximity of the crosswalk, much like they would stop at a signalized intersection or a stop sign. This mod- ified deceleration rate seemed to be more discernable auditorily, which presumably aids utilization. Pedestrian Hybrid Beacon Figure 19 summarizes the encountered and utilized events for the PHB location in the pretest. As discussed previously, results are shown for the condensed matrix of near-lane and far-lane events. The raw event data is contained in Appendix A. Figure 19 shows similar trends at this crossing to those that were observed at the RCW pretest. More than half of the events (325 out of 603) represent non-utilized opportunities and indicate delay. Similar to the other studied crosswalk, over 30% of events correspond to non-crossable gaps in one or more lanes, signifying drivers that did not yield to the pedestrians. Among those events that do represent crossing opportunities in both lanes, the rate of utilization is again lowest for dual- yield events. The figure further shows that 8.1% of events are associated with a potential multiple-threat situation with a near-lane yield and a far-lane non-crossable gap. Participants utilized only 2.0% of those events at this crossing, which represents 0.2% of all crossing events. Overall, 1.0% of events were utilized events with a non-crossable gap in one or both lanes. Similar to the RCW, the results in Figure 20 show a reduc- tion in events encountered by the pedestrian, from 603 to 242 53 RCW Performance Measure Pre Post Difference p-value Dual Opportunities (PA_Dual) 56.0% 76.9% 20.9% 0.0003 Single Opportunities (PA_Half) 12.5% 7.8% –4.7% 0.0842 None Opportunities (PA_No) 31.5% 15.3% –16.2% 0.0016 Dual Utilization (PU_Dual) 88.3% 98.1% 9.8% 0.0062 Single Utilization (PU_Half) 12.9% 15.2% 2.3% 0.7980 None Utilization (PU_No) 2.0% 7.6% 5.7% 0.3257 Average Delay (s) 17.0 8.0 –9.0 0.0434 Delay>Min (s) 3.4 2.3 –1.1 0.2117 85th Percentile Delay (s) 29.8 12.9 –16.9 – O&M Interventions 2.8% 0.0% –2.8% 0.0230 Table 7. Crossing performance summary at RCW, pretest and posttest.

events. Again, this reduction is primarily attributable to the greatly reduced number of non-utilized events, from 325 in the pretest to only 31 in the posttest, signifying improved decision- making efficiency and more courteous (signal compliant) driver behavior with installation of the PHB. Figure 20 fur- ther shows that with the installation of the PHB, most encoun- tered events fell into the yield category, which corresponds to vehicles stopping at the red signal indication. Similar to the RCW, very few non-crossable gap events were encountered in the posttest condition. The figure also shows that 4.5% of posttest events are asso- ciated with a potential multiple-threat situation and that par- ticipants utilized 9.1% of these events (0.4% of all crossing events). Overall, only 0.4% of events were utilized events that were associated with a non-crossable gap in one or both lanes. Following the discussion in Chapter 4, the nine event cat- egories may be converted to the probabilities of encountering and utilizing a crossing opportunity in both lanes, PA_Dual and PU_Dual, as well as in only one of the lanes PA_Half and PU_Half. Table 8 summarizes these statistics for the PHB analysis and also presents the associated delay and intervention statistics as defined previously. Table 8 shows an increased probability of encountering dual crossing opportunities, from 55.5% to 89.3%. The rate of utilization of these events was high in the pretest condition (91.6%) and was increased further in the posttest (98.3%). Similar to the RCW, the rate of PA_Half and PA_No opportu- nities decreased, which is explained by fewer non-crossable gap events and generally less non-utilized events (see Figure 20). Accordingly, the average pedestrian delay statistics improved significantly with PHB installation, from 16.0 to 5.8 s. This trend is mirrored in the Delay>Min and 85th percentile delay times. The single highest average delay was 46.5 s in the pretest and 14.6 s in the posttest. The PHB also improved pedestrian safety performance and reduced interventions from 2.4% to 0%. In the pretest, seven interventions were distributed among six of the 18 participants. Five of those six participants returned for the posttest, and no interventions were observed for them or the other seven posttest participants. A 2.4% intervention rate corresponds to a risk of 1 in 42 crossing attempts. Similar to the RCW, it is important to emphasize that while the reduction of interven- tions to 0% is statistically significant, this number in all likeli- hood does not represent an absolute zero. It should therefore 54 Figure 20. PHB posttest event utilization. This figure shows a bar chart of all observed pedestrian–vehicle events during the PHB posttest. The graph shows a total of nine event categories, representing all combinations of event outcomes yield, crossable gap, and non-crossable gap for two conflicting lanes (near lane and far lane). 0.0% 10.0% 20.0% 30.0% 40.0% 50.0% 60.0% Fa r La ne =Y Fa r La ne =C G Fa r La ne =n o n CG Fa r La ne =Y Fa r La ne =C G Fa r La ne =n o n CG Fa r La ne =Y Fa r La ne =C G Fa r La ne =n o n CG Near Lane = Y (n=185) Near Lane = CG (n=40) Near Lane = non- CG (n=17) Pe rc en t o f T ot al E ve nt s (n = 2 42 ) PHB-POST (n=242) Non-Utilized (n=31) Utilized (n=211) Y=Yield, CG=Crossable Gap, non-CG = Non-Crossable Gap Figure 19. PHB pretest event utilization. This figure shows a bar chart of all observed pedestrian–vehicle events during the PHB pretest. The graph shows a total of nine event categories, representing all combinations of event outcomes yield, crossable gap, and non-crossable gap for two conflicting lanes (near lane and far lane). 0.0% 10.0% 20.0% 30.0% 40.0% 50.0% 60.0% Fa r La ne =Y Fa r L a n e = CG Fa r La n e= no n CG Fa r La ne =Y Fa r L a n e = CG Fa r La n e= no n CG Fa r La ne =Y Fa r L a n e = CG Fa r La n e= no n CG Near Lane = Y (n=194) Near Lane = CG (n=173) Near Lane = non- CG (n=236) Pe rc en t o f T ot al E ve nt s (n = 6 03 ) PHB-Pretest (n=603) Non-Utilized (n=325) Utilized (n=278) Y=Yield, CG=Crossable Gap, non-CG = Non-Crossable Gap

not be interpreted as the PHB resulting in zero risk to blind pedestrians. It is very unlikely that any treatment would result in zero risk to blind pedestrians or sighted pedestrians, as evi- dent by pedestrian injuries and deaths at intersections across the country. The 0% intervention rate therefore should only be interpreted as the fact that no interventions were observed during the 16 crossing attempts by each of the 13 participants during the posttest. Overall, the installation of the PHB resulted in signifi- cant improvements in crossing performance from a delay and safety perspective. In the interpretation of the delay times, it needs to be highlighted that the study design used in the PHB study assumed that pedestrians arrived at the crossing after the minimum green time had elapsed. Effectively, this design min- imized the potential delay encountered by pedestrians. At a phase cycle length of 38 s, this assumption is valid for pedes- trian volumes of less than 90 per hour (which would result in pedestrian headways of 40 s assuming uniform arrivals). For heavier pedestrian flows, this assumption does not hold because additional pedestrians are increasingly likely to arrive just after the “Walk” phase has elapsed and higher delays are antic- ipated. However, as demonstrated in a simulation-based sen- sitivity analysis of roundabout PHB installations (Schroeder et al. 2008), additional pedestrians will also be more likely to join existing signal calls. The pedestrian delay therefore plateaus with increasing pedestrian demands, where the magnitude of this terminal delay is a function of the signal phasing. Participant Feedback Following each the pretest and posttest two-lane round- about studies, participants were asked a series of questions about their perception of the crossing and the level of com- fort in their crossing decisions. The blank survey forms are provided in Appendix G. The participant feedback at the two-lane roundabout pres- ents a somewhat different picture than the feedback at the single-lane roundabouts or the CTL. The average of the responses on the confidence question (“How would you rate your confidence in your ability to cross here safely on a scale of 1–5, with 1 being not at all and 5 very confident?”) was 4.0 in the pretest, 4.83 at the PHB installation posttest, and 4.58 in the RCW crossing posttest. On the question, “Would you use these crossings if they were on the most direct route home from work?” in the pretest, 15 participants said yes, and 1 said “no, would find another way.” In the posttest, with the PHB, 12 said yes and no one said no, and with the RCW, 11 said yes and one said no. While the numbers were positive in the pretest and posttest conditions, the comments that were noted along with the responses changed considerably. In posttest debrief- ing, comments at the PHB were generally more enthusiastic, with additions like “definitely!” or “would go out of my way to use them.” A couple of questions were added for the Golden debriefing that provide a little more insight into the participants’ think- ing. Participants were asked if they considered the crossings more risky, less risky, or of about the same risk as an intersec- tion with a signal and four lanes of traffic. Table 9 shows the frequency of responses and percentages in parentheses. In the pretest, 17% considered the two-lane roundabout less risky, and about 40% each thought it was the same or more risky than a signalized intersection. In the posttest, at the PHB, 54% considered it less risky and 31% about the same risk. Only one person thought it was more risky, compared to seven in the pretest. At the RCW, 23% considered it less risky, 62% about the same risk, and again only one person (8%) more risky than a signal. Presumably, this suggested greater confidence and comfort in participants’ perception of the PHB relative to the RCW. Both treatments showed perceived benefits over the base two-lane roundabout case, as well as the (hypothetical) signalized intersection. 55 PHB Performance Measure Pre Post Difference p-value Dual Opportunities (PA_Dual) 55.5% 89.3% 33.8% <.0001 Single Opportunities (PA_Half) 15.0% 4.1% –10.9% 0.0001 None Opportunities (PA_No) 29.5% 6.6% –23.0% <.0001 Dual Utilization (PU_Dual) 91.6% 98.3% 6.7% 0.0062 Single Utilization (PU_Half) 8.8% 8.3% –0.5% 0.9468 None Utilization (PU_No) 0.0% 0.0% 0.0% – Average Delay (s) 16.0 5.8 –10.2 0.0007 Delay>Min (s) 3.2 1.4 –1.8 0.0044 85th Percentile Delay (s) 29.5 7.7 –21.8 0.0001 O&M Interventions 2.4% 0.0% –2.4% 0.0112 Table 8. Crossing performance summary at PHB, pretest and posttest.

As discussed in the previous sections on participant feed- back, the participants’ suggestions of intersection modifica- tions after their crossings in the pretest may provide additional information about their concerns. Some suggested signals, even though some of those same individuals said that they were not fans of audible signals, and suggested that it would be good if the signal could be used optionally. One suggested a light that signaled to motorists that pedestrians were waiting to cross. Some suggested that detectable warnings needed to be present on both crosswalks (for some reason detectable warn- ings were only installed on the north crosswalk during pretest). One suggested some way to know that it was a roundabout when approaching as a pedestrian, and another suggested mov- ing the exit crosswalks further from the roundabout circu- latory roadway. In the posttest, participants commented about the uneven surface of the raised crosswalk and the need to have a smoother transition. Several made positive comments about the signal and push-button locator tone; one said: “if they’re going to use locator tones, those were pretty good.” At posttest, participants seemed generally happy with the modifications, particularly the pedestrian hybrid beacon. Several stated that the raised crosswalk didn’t make a difference, while others felt it made a big difference in drivers’ willingness to yield. Participants in the posttest were also asked several questions specific to the PHB, the push-button locator tone, and the audible message (see Table 10). For the questions in the table, participants were asked to rate the extent of their agreement, with 1 being strongly disagree and 5 strongly agree. The responses suggest some hesitation in the expected use and effectiveness of the PHB. While most said they would use the APS device, only a few said they would wait for the audible “Walk” message (and thereby the “Walk” signal phase). Most respondents acknowledged the benefit of the push-button locator tone to identify the crossing locations, but responses on initial alignment and maintaining alignment during cross- ing were mixed. It needs to be acknowledged here that the locator tones were generally too quiet in this installation to be audible across the width of the street. Therefore, participants generally couldn’t hear the far-side message or locator tone until a little past the middle of the crossing. The far-side APS did seem to help people know when they were about to reach the end of the crosswalk. Some participants negatively com- mented on the fact that the push buttons on the island were installed on top of a wall used to contain landscaping, which they didn’t expect. This part of the installation clearly wasn’t ideal and was partly related to the temporary nature of this PHB installation. Compliance with Signal Indications Pedestrian Compliance The analysis of pedestrian crossing performance does not consider an important aspect of the behavior at the PHB: the signal phase during which pedestrians chose to cross. The analysis results remain valid since they describe pedestrian actions in terms of actual driver behavior. For example, most drivers are expected to stop before the “Walk” phase comes on. Similarly, a “Walk” indication is no guarantee of perfect driver compliance and is associated with a risk of red-light running events. Consequently, the analysis above initially ignored the signal phase. This also ensures that the results are directly com- 56 Less Risky About the Same Risk More Risky Not Answered Pretest (n=18) 3 (17%) 7 (39%) 7 (39%) 1 (6%) PHB (n=13) 7 (54%) 4 (31%) 1 (8%) 1 (8%) RCW (n=13) 3 (23%) 8 (62%) 1 (8%) 1 (8%) Table 9. Participant survey comparison of two-lane roundabout treatments. Rating Questions Average of Responses (N=10) 1=strongly disagree, 5=strongly agree If there were signals like these, I’d push the button each time I wanted to cross 3.7 If there were signals like these, I would always wait to cross until I hear “walk sign is on.” 2.0 These signals helped me know I was coming to the crosswalk. 4.2 These signals helped me align to cross. 2.4 These signals helped me go straight across the crosswalk. 2.4 These signals helped me know I was approaching the end of the crosswalk 3.4 Table 10. Participant survey response to PHB installation.

parable to the pretest condition as well as crossing performance at the RCW crossing. Nonetheless, the installation of the PHB is associated with some legal implications for when pedestrians should cross. At the PHB, the vehicle signal display rests in a “Dark” mode pending a pedestrian’s pressing of the APS push button. While the language in the MUTCD (FHWA 2009) allows for a PHB to rest in “Dark” for the pedestrian mode, this was not the case at the tested installation. The pedestrian display for the signal rested in “Don’t Walk,” as a conventional signal would. Conse- quently, pedestrians were expected to push the button and wait to cross until the onset of the “Walk” phase. In this project, par- ticipants were informed of the phase sequence and intended behavior. However, they were always instructed to “cross when they are ready” and to “rely on their own judgment” when mak- ing crossing decisions. Participants were not given any specific information regarding the legal issue of beginning crossing during “Walk” at pedestrian signals. In other words, pedestri- ans were not told that they had to cross in the “Walk” phase. With the PHB, pedestrians encountered a signal indicating that the signal phase is either “Walk” (W), “Flashing Don’t Walk” (FDW), or “Don’t Walk” (DW). Blind pedestrians heard a push-button locator tone during the DW and FDW phases and a speech message during the W phase. Figure 21 shows the frequency of crossing initiation for the (blind) pedestrian relative to PHB signal phases. More details on the PHB phase sequence and timing parameters were given in Chapter 3. The results show that only 36.7% of pedestrians crossed in the intended “Walk” phase and that many (39.0%) actually initiated the crossing just before the “Walk” phase (and the APS alert) in the vehicular solid yellow. In other words, they began to cross following pressing the call button but prior to the audible message. Further, 11% crossed even earlier, during the vehicle “Flashing Yellow” phase, and 13.3% didn’t cross until the flashing “Don’t Walk” phase. Overall, only three times did pedestrians not cross in the first crossing phase and have to reactivate the signal. From anecdotal observation, participants appeared to cross whenever they first heard cars stop. The actual pedestrian signal display at that time seemed to be less impor- tant to the participants. As discussed above, observations fur- ther suggested that driver deceleration behavior was more rapid and the stop location was closer to the crosswalk, which seemed to make these yields more distinguishable for the blind par- ticipants. It is re-emphasized here that participants were not instructed that they had to cross during the “Walk” phase. These findings suggest that the study participants rely heav- ily on their own personal judgment, even with the signal beacon in place. Pedestrians tended not to cross in “Walk” if they were unsure about whether vehicles had in fact stopped. Even when the audible message confirmed to the blind pedestrian that a “Red” signal indication was being presented to an approaching driver, some would still not cross until they were confident that it was safe to do so. Similarly, they would readily cross before the “Walk” phase if they perceived a crossing opportunity. Driver Compliance In the evaluation of the PHB, an important question of inter- est to traffic engineers is the behavior of drivers relative to the signal phases. In particular, the PHB is intended to reduce 57 Figure 21. Blind pedestrian crossings at PHB by signal phase (% of all crossings). 11.0% 39.0% 36.7% 13.3% 0 10 20 30 40 50 60 70 80 90 0 5 10 15 20 25 30 35 40 45 Flashing Yellow Yellow Red / Walk Flashing Red / Flashing Don't Walk Fr eq u en cy o f E ve nt s Pe rc en ta ge o f E ve nt s Signal Phase n=211 This figure shows a bar chart of pedestrian crossing behavior in regard to the four PHB signal displays: “Flashing Yellow/Don’t Walk,” “Yellow/Don’t Walk,” “Red/Walk,” and “Flashing Red/Flashing Don’t Walk.” The results are discussed in the text.

vehicular delay by allowing drivers to proceed during the “Flashing Red” phase. The following analysis is intended to capture driver understanding of and compliance with the sig- nal indication. Driver understanding of and compliance with the PHB can be evaluated by relating the driver stopping behavior to the indicated signal phase. Figure 22 plots two categories of driver behavior for each signal phase: (1) vehicles stopped or stop- ping, and (2) vehicles proceeding through the crosswalk. The figure shows four signal phases that correspond to the PHB phasing sequence for vehicles: “Flashing Yellow,” “Solid Yel- low,” “Solid Red,” and “Flashing Red.” The figure shows that 34.1% of drivers proceeded through the crosswalk in “Flashing Yellow,” which is permitted behav- ior. As the signal changed to “Solid Yellow,” 11.4% of drivers proceeded through the crosswalk, which is allowable if the vehicles were too close to the crosswalk to come to a stop. How- ever, even during the “Solid Red,” 12.6% of observed vehicles proceeded through the crosswalk. This statistic is a concern, since drivers are legally required to stop for the red signal indi- cation and because pedestrians expect a crossing opportunity. Driver behavior during “Flashing Red” shows that almost half of the drivers (48.2%) remained stopped, suggesting some inefficiency in driver behavior in response to the PHB. These findings raise some concerns that the PHB traffic control device may not have been properly understood by drivers or that the PHB display was ignored. An education campaign by the city of Golden, using web and news media outlets, informed citizens of the PHB installation at the test roundabout and discussed appropriate behavior. However, it is unclear how frequently the device was actually used in the 6-week driver adaptation period prior to the posttest. Further, frequent tourist and non-commuter traffic in the area may have contributed to driver confusion. Given the apparent lack of understanding of the PHB, it seems that the flashing red indication of the traffic control was not intuitive to drivers. Impact on Vehicular Traffic Raised Crosswalk A pretest and posttest speed study was performed at the RCW installation to estimate the impact of the treatment installation on free-flow vehicle speeds. All speeds were col- lected from video observations using known reference dis- tances from roadside markers. The speeds correspond to the average speed just upstream of the crosswalk, measured over a distance of approximately 100 ft at exit and 160 ft at the entry leg. The study included only free-flowing vehicles that passed through the crosswalk in the absence of pedestrians. A sample size of approximately 100 vehicles was collected for the entry and exit leg of the roundabout in both pretest and posttest conditions. The total dataset of 405 observations was collected for different times of day and on different days of the week. The results show an average entering speed approaching the crosswalk of 25.3 mph, which was significantly reduced to 20.5 mph with the installation of the raised crosswalk (p < 0.0001). While this speed reduction is as anticipated, the posttest speeds were still relatively high given the RCW treatment. This is attributed to the relatively low vertical height and gentle slope transition that was used in the RCW design. The design therefore results in a relatively low impact on vehicle speeds in the absence of pedestrian and vehicle 58 Figure 22. Evaluation of driver behavior at PHB. 65.9% 88.6% 87.4% 48.2% 34.1% 11.4% 12.6% 51.8% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Flash Yellow Solid Yellow Solid Red Flash Red Signal Phase Proceeding through Crosswalk Stopped or Stopping This figure shows a bar chart of driver behavior in regard to the four PHB signal displays: “Flashing Yellow/Don’t Walk,” “Yellow/Don’t Walk,” “Red/Walk,” and “Flashing Red/Flashing Don’t Walk.” The results are discussed in the text.

platoons, while still showing beneficial impacts on pedestrian crossing performance as discussed above. At the exit leg, the average pretest speed of 17.7 mph was reduced to 16.1 mph (p < 0.0001). While statistically signifi- cant, the practical implication of this speed reduction is mar- ginal in terms of vehicle delay. But again, as the analysis above showed, the RCW had a very positive impact on pedestrian crossing performance, despite the relatively low speed impact. Pedestrian Hybrid Beacon A special queuing study was performed at the PHB installa- tion to address concerns that the signal would cause extensive queuing and that those queues could spill back into the circu- lating lane. The analysis measured the maximum queue length for each pedestrian crossing in all pretest and posttest trials. The maximum queue was defined as the longest pedestrian-induced queue length measured in vehicles. Queues were measured rel- ative to the crosswalk and therefore do not include additional vehicles that were waiting to enter the roundabout downstream of the crosswalk (at the entry). Vehicle queues were combined for both lanes since no significant difference was observed between queues in the inside and outside lanes. The results showed that the average maximum queue length increased from 2.3 to 5.0 vehicles at the entry and from 1.5 to 3.9 vehicles at the exit over both approach lanes. The increases in average maximum queues are significant at p < 0.0001. With available queue storage of two vehicles (one per lane) at the exit leg, it is evident that the maximum queue sometimes spilled back into the circulating lane. With the installation of the PHB, that proportion of maximum queues greater than two vehicles increased from 29.8% to 69.2%. However, the average queue is expected to be much lower, so that the overall effect of the PHB installation on vehicle queues is considered to be mar- ginal. In fact, a determined yielder may be stopped for 10 or more seconds waiting for the pedestrian to cross and is likely to cause similar if not more delay to a driver waiting at the PHB, as evident by some long queues observed in the pretest. More detail on the queuing analysis is given in Appendix A. Summary of Results This chapter presented findings from 12 rounds of field studies performed at a total of seven pedestrian crossings: two channelized right turn lanes, three single-lane roundabout crossings, and two two-lane roundabout crossings. While no treatments were tested at the single-lane roundabout loca- tions, the remaining crossings and studies included pretest and posttest comparisons of the effectiveness of different treatments in improving accessibility of these locations. All 12 studies were performed using the same experimental protocol. The study for one of the single-lane roundabouts (PS-RAL) was performed as part of an earlier research project (NIH 2010). The remaining data collection occurred under the auspices of NCHRP Project 3-78A. For these 11 rounds of testing, the same O&M specialist was used to establish con- sistency and uniformity in the experimental design. This point is especially important in light of the rate of O&M interven- tions, an important safety measure used in the analysis. The analysis framework was discussed in detail in Chapter 4 and uses a four-pronged approach for assessing the accessibil- ity of a crossing through (1) the availability of crossing oppor- tunities, (2) the rate of utilization of these opportunities, (3) the delay experienced by the pedestrian, and (4) the level of risk associated with the crossing. The first two measures are largely intended to find an explanation for the latter two. This section summarizes some of the delay and risk measures to allow a comparison across the different test locations. Figure 23 summarizes the average delay across all partic- ipants for the 12 locations. The graph further contains error bars at 1 standard error of the estimate of the mean. The stan- dard error is calculated by dividing the standard deviation of the estimate by the square root of the sample size (number of participants). The summary chart shows highest delays observed at the two CTL locations in the pretest treatment con- dition as well as for the DAV-CLT single-lane roundabout. Interestingly, these three data points represent very differ- ent traffic conditions, but were all located in Charlotte. At DAV-CLT, volumes were low and participants were delayed despite ample gap crossing opportunities as many waited for all-quiet periods. At the two CTL crossings, traffic volumes were much higher, and the delay was exacerbated by a very high level of ambient noise from the main intersection. All three of these locations were characterized by a low propensity of drivers to yield, which may be characteristic of the local driv- ing culture or may be coincidence. With treatment installation, the average delay at CTL crossings was reduced. No treatment was tested at the single-lane roundabout. Of the remaining sites, the two-lane roundabout crossings exhibited the highest average delay and the highest variabil- ity across participants. The two-lane roundabout delay was reduced significantly with installation of either crossing treat- ment (RCW and PHB) to levels that were below the remaining single-lane roundabout delays. It is important to emphasize that the studies at the two-lane and single-lane roundabouts in Golden were performed concurrently with the same partici- pants. Consequently, these sites can be directly compared. The comparison shows that delays were higher at the two-lane roundabout and there was greater variability across partici- pants. The assessment of the PHB and RCW treatments fur- ther highlights that pedestrian delay and risk were effectively improved over the base condition of the single-lane round- about comparison site. Figure 24 shows the comparative results of O&M interven- tions across the 12 test conditions. Again the figure shows the 59

Figure 24. Summary of O&M interventions for all sites. 9.4% 2.9% 5.6% 1.4% 0.8% 3.9% 1.4% 0.5% 2.4% 0.0% 2.8% 0.0% 0% 2% 4% 6% 8% 10% 12% 14% SS -O NL Y PR E SS -O NL Y PO ST SS +F B PR E SS +F B PO ST D AV -C LT PS -R AL * G O L- PR E G O L- PO ST PH B PR E PH B PO ST RC W P R E RC W PO ST Channelized Turn Lane Single-Lane RBT Two-Lane RBT O&M Interventions This figure shows a bar chart of the summary of the O&M intervention results at all test sites and all study conditions. The graph shows the average pedestrian delay observed for each of 12 studies as well as the standard error of that estimate. The results are discussed in the text. Error bars shown at one standard error *Data collection performed in prior research Figure 23. Summary of 85th percentile delay for all sites. 26.2 18.5 23.4 12.2 25.3 11.1 11.9 12.1 16.0 5.8 17.0 8.0 0 5 10 15 20 25 30 35 SS -O NL Y PR E SS -O NL Y PO ST SS +F B PR E SS + FB PO ST D AV - CL T PS -R AL * G O L- PR E G O L- PO ST PH B PR E PH B PO ST RC W P R E RC W PO ST Channelized Turn Lane Single-Lane RBT Two-Lane RBT Average Delay (s) This figure shows a bar chart of the summary of the delay results at all test sites and all study conditions. The graph shows the average pedestrian delay observed for each of 12 studies as well as the standard error of that estimate. The results are discussed in the text. Error bars shown at 1 standard error *Data collection performed in prior research

average observed over all participants and the standard error of the estimate. The intervention comparison shows very high risk at the CTL crossings in the pretest condition along with a very high variability across participants. Intervention rates at the CTL locations were reduced with treatment installation, but some interventions remained even in the posttest condition. For the roundabouts, the highest intervention rate was observed at PS- RAL, although this study was not performed as part of NCHRP Project 3-78A. With the use of different O&M instructors there is the potential that individual differences may be a contribut- ing factor to the difference in intervention rates. Since the remaining 11 data points all involved the same O&M specialist, greater comfort exists that the results are comparable. Conse- quently, the two-lane roundabout pretest conditions exhibited the greatest level of risk among the roundabouts and the largest variability across participants. With treatment installation, no interventions were observed at these locations. 61