IDENTIFYING HAZARD MITIGATION BEHAVIORS THAT LEAD TO DIFFERENCES IN THE CRASH RISK BETWEEN EXPERIENCED AND NOVICE DRIVERS
A Dissertation Presented
by
JEFFREY W. MUTTART
Submitted to the Graduate School of the University of Massachusetts Amherst in partial fulfillment of the requirements for the degree of
DOCTOR OF PHILOSOPHY
May 2013
Mechanical and Industrial Engineering
UMI Number: 3589108
All rights reserved INFORMATION TO ALL USERS The quality of this reproduction is dependent upon the quality of the copy submitted. In the unlikely event that the author did not send a complete manuscript and there are missing pages, these will be noted. Also, if material had to be removed, a note will indicate the deletion.
UMI 3589108 Published by ProQuest LLC (2013). Copyright in the Dissertation held by the Author. Microform Edition © ProQuest LLC. All rights reserved. This work is protected against unauthorized copying under Title 17, United States Code
ProQuest LLC. 789 East Eisenhower Parkway P.O. Box 1346 Ann Arbor, MI 48106 - 1346
© Copyright by Jeffrey W. Muttart, 2013
All Rights Reserved
IDENTIFYING BEHAVIORS THAT LEAD TO CRASH RISK BETWEEN EXPERIENCED AND NOVICE DRIVERS
A Dissertation Presented by
JEFFREY W. MUTTART
Approved as to style and content by: _______________________________________ Donald L. Fisher, Chair _______________________________________ Jenna Marquard, Member _______________________________________ Alexander Pollatsek, Member ____________________________________ Donald L. Fisher, Department Head Mechanical and Industrial Engineering
DEDICATION
I started out as an empty shell and with the careful craftsmanship of parents, teachers, friends, co-workers, my children (Kristina and Matthew) and my wife Pam; I am able to accomplish a goal that I thought was only a dream. I really do not know who deserves each of the letters “P h and D”, while I certainly worked very hard to get to this point, I look back with the realization that without the support of nearly every person in my life, this would not be possible. Without my parents, I would not have had the confidence. Without my friends, I would not have had the sense of urgency and fear that I might let them down. Without Kristina, Matthew, and Pam, I would not have been able to keep my sense of humor. Without Professor Fisher and all those in the lab, I would not have been able to program the simulator, or the eye tracker, or take the next step toward being a professional researcher. Without good teaching, I would not have had the motivation to continue, or the knowledge that was a prerequisite to accomplish what I have done. I would also like to thank my late father-in-law, Russ Cavanaugh for his support, I miss him. But most of all, without the help of my wife Pam, nothing was possible. Pam not only supported me through words, she also helped compile data, acted as a blind scorer, ran macros on the data, and she is the love of my life.
ACKNOWLEDGMENTS I have to thank all those in the Human Performance Laboratory for the help I have received. Gautam Divekar and Mathew Romoser taught me how to program the driving simulator. Hasmik Mehranian assisted me with trouble-shooting problems. Tracy Zafian went above and beyond in her constant support, from help with parking permits, to IRB approvals, routine assistance, and encouragement. Siby Samuels was a spectacular blind scorer, he is the fastest and most accurate I have ever seen. Also, Erin Lyman from Crash Safety Solutions has been a god-send and has always been there for me. Steve Guderian and Dan Coughlin helped with blind scoring and C. Greg Russell helped with the development of the ACT program. I would also like to thank former lab mates, particularly Anuj Pradhan who was always there to help, many times at the expense of his work. Also, I would like to thank Mary Ellen Paciorek and Lucas Kendall from the Pioneer Valley Driving School who worked tirelessly both recruiting and scheduling drivers for this research and included many early mornings and late nights. I have to also thank all the researchers that have worked in this area, there are so many to name, but I want you all to know that I enjoyed learning from your work. Relative to learning, several former lab mates express how special it is to have professors Donald Fisher and Alexander (Sandy) Pollatsek as advisors, I agree. The insight we receive from them cannot be compared. I do not know of a single person who can better construct a study than Don Fisher. Regarding Sandy, I do not know of a single person who can look at spreadsheets of data and instantly analyze the trends and weaknesses. What is even better is that one is a top-down thinker and the other a bottom-up and all of us in the lab benefit. The third person on my committee, Jenna Marquardt was selected for her knowledge of decision making. Given that the goal is to develop a driver training program, an understanding of driver decision making and the reasoning of experts versus novices is vital, which made her input both essential and appreciated. I would not be here if it were not for all my prior teachers at Montville High School, Eastern Connecticut State University, University of Hartford, and the University of Massachusetts. Lastly, I have to thank the Groton Town Police Department for giving me the opportunity to pursue my dream of being an investigative researcher of crashes… it all started there.
v
ABSTRACT IDENTIFYING BEHAVIORS THAT LEAD TO CRASH RISK BETWEEN EXPERIENCED AND NOVICE DRIVERS MAY 2013 JEFFREY W. MUTTART, B.A., EASTERN CONNECTICUT STATE UNIVERSITY M.A., UNIVERSITY OF HARTFORD Ph.D., UNIVERSITY OF MASSACHUSETTS AMHERST Directed by: Professor Donald L. Fisher The three most common crash types for drivers under age 18 are run-off-the-road crashes, left turn at intersection crashes, and rear end crashes. Previous literature points to novice drivers being less likely to anticipate hazards or maintain attention to the forward roadway and as a result failing to mitigate hazards by slowing adequately. In two experiments using a fixed-based high fidelity driving simulator, two groups of drivers were evaluated in potential hazard scenarios. Anticipatory glances, slowing behaviors, and lane position of experienced drivers with exemplary records and newly licensed 16 – 18 year old drivers were compared at the two curves, two intersections, and two straight segments that were most heterogeneous relative to tasks and risk of all those negotiated. In Experiment 1, experienced drivers were significantly more likely to make anticipatory (glance), and mitigation (slowing and lane keeping) responses when approaching locations of greatest risk. Experienced drivers crashed nine times and novice drivers crashed 23 times. Overall, experienced drivers began to slow approximately eight seconds before the incidents, slowed to target speed when within three seconds of the incident and selected safer lane positions than did novice drivers. In Experiment 2, the ACT (Anticipate, Control, and Terminate) computer program was developed and utilized to train one group of novice drivers. The other group received placebo training. The ACT Program was designed to teach novice drivers to slow for HRECCS (pronounced wrecks). HRECCS is an acronym that explains the reasons a driver should slow (hidden obstacles, roadside hazards, no escape route, closing with no option to pass, curves, and traffic signals). Each participant completed a pretest, training, saw the responses made by the experienced drivers, was offered mediated training (shown their mistakes and correct responses), and finally, completed a posttest. Placebo trained drivers had the same routine but rather than rules training and mediation, they received training related to street signs. ACT trained drivers made many more anticipatory glances, slowed to target speed more often and selected safer lane positions than did the placebo trained drivers. ACT trained drivers crashed eight times compared to 22 for the placebo trained drivers.
vi
EXECUTIVE SUMMARY Crash statistics demonstrate that newly licensed teenage drivers face a high risk of crashing. Research efforts have been made to identify the underlying reasons. The literature reveals that novice drivers exhibit poor hazard anticipation skills. Hazard anticipation is not the only skill that is important when a hazardous situation arises; it is also important that the driver mitigate the potential hazard. For example, if a pedestrian steps off the crosswalk and stops after taking just one step, a driver might respond by placing his or her foot over the brake. Yet, if the pedestrian keeps walking and a collision appears imminent, the response necessary to avoid the crash might be to brake or steer sharply. In either situation, the ultimate goal is to reduce the probability of a crash by anticipating and mitigating the hazard. Currently it is not known what hazard mitigation strategies novice drivers have in their repertoire, or whether their hazard mitigation strategies actually reduce their likelihood of crashing. This dissertation will seek to address these gaps in our understanding. In two experiments, two groups of drivers were evaluated in potential hazard scenarios (no crash could occur) and crash scenarios (a crash could potentially occur). An understanding of behavior in potential hazard (no-crash) scenarios is important to identify the hazard mitigation strategies of drivers when they are not hypersensitive to crashes. In this study, drivers were exposed to infrequent potential hazards, which could develop into an immediate near-crash or crash situation. Near-crash events were infrequent to keep the drivers from becoming hypersensitive in crash scenarios and changing their hazard mitigation strategies. The two experiments were completed using a driving simulator. The use of a driving simulator allows for the examination of riskier scenarios that might otherwise be impossible in the field. An example of one such scenario is referred to in the literature as the truck midblock crosswalk scenario (Pradhan et al., 2006; Fisher, 2008). Drivers approached a marked, midblock crosswalk with a truck in the parking lane stopped immediately before the crosswalk and obscuring pedestrians that might be entering the crosswalk. Hazard anticipation and hazard mitigation behaviors can be analyzed here. For example, as drivers anticipate the hazard and glance towards the front corner of the truck, a hazard might appear as they approach the truck from behind, and then as they pass the truck (before entering the crosswalk). Drivers mitigate the hazard when they slow, take their foot off the accelerator and/or position their foot over the brake. Mitigation would also include simply choosing a slower speed when approaching the stopped truck. Altogether, there were a total of 18 scenarios on which participants’ hazard anticipation and hazard mitigation were evaluated in Experiments 1 and 2. Experiment 1. In Experiment 1, the hazard mitigation strategies of novice and experienced drivers were compared. In the current experiment, there was an equal number of experienced and novice drivers which allows for a simple measure of hazard mitigation. The hazard mitigation strategy to be evaluated in this study with respect to the above scenario was based on whether a driver slows as he or she approaches within six seconds of the truck. Suppose, as was the situation in Experiment 1 that 78 percent of experienced drivers and 43 percent of novice drivers respond by slowing. It would appear that experienced drivers are 1.8 times as likely to mitigate a hazard as novice drivers. However, hazard mitigation can only occur if the driver is aware of the hazard, vii
and much of the research to date documents that novice drivers typically glance toward fewer hazards than do experienced drivers (Pradhan et al., 2005). Thus, it could be the case in the above example that novice drivers are actually more likely to mitigate a hazard than are experienced drivers, given that they glance toward it. To fairly determine the likelihood that a novice and experienced driver mitigate a hazard, I estimated the conditional probability that a driver (novice or experienced) mitigates the hazard, given that the driver glances towards the area of the potential hazard. I also estimated the conditional probability that a driver mitigates the hazard, given that the driver did not glance towards the area of the potential hazard. Behaviors at curves, intersections and straight road segments were analyzed. Curves. Experienced drivers were much more likely to mitigate the hazard than novice drivers. Among all drivers who mitigated the hazard, the experienced drivers were more likely to mitigate the hazard earlier than were the novice drivers. A driver was assumed to mitigate the hazards of a curve before entering the curve when the projected target speed at curve entry is equal to the threshold speed at curve entry. For example, if a driver was traveling 30 mph and is decelerating 2 mph per second when five seconds from the curve, that would mean the driver was on track to slow to 20 mph at the curve entry. In this example, the driver’s target speed would be 20 mph. If the safe curve entry speed were 20 mph, then this driver would have successfully mitigated the curve and done so at 5 s before the curve. For this same driver, if he or she were traveling faster, or slowing less, the driver would not be on-target to slow to 20 mph. Consider the results from just the sharp curve to the right. (a) Five seconds before entering this curve, some 65% of the experienced drivers had glanced to the far extent of the curve whereas only 13% of the novice drivers had glanced to the far extent of the curve. Furthermore, 70% of the experienced drivers mitigated the hazard whereas only 53% of the novice drivers mitigated the hazard. Finally, among those who mitigated the curve, some 22% of the experienced drivers did so at 5 s whereas only 3% of the novice drivers did so at 5 s. These findings indicate that novice drivers mitigate the hazard less well than experienced drivers. However, the data cannot be used alone to determine whether the failures of mitigation among the novice drivers are due to failures of anticipation or instead to failures of mitigation even when the hazard is anticipated. (b) Thus, the next question is whether there are differences in mitigation behaviors of novice and experienced drivers, conditional on a glance being made. Among those drivers who glanced, 90% of the experienced drivers mitigated the hazard compared to only 67% of the novice drives. This evidence indicates that novice drivers are inferior mitigating a hazard even when they are inferred to anticipate the hazard. (c). Finally, among those drivers who did not glance, 63% of the experienced drivers mitigated the hazard compared to only 47% of the novice drives. This type of response is consistent with ambient processing and was more common with experienced drivers. (d) The above differences in the behaviors of novice and experienced drivers should be evidenced in crashes as well as glances and slowing. In fact, novice drivers crashed six times whereas only two experienced drivers crashed. Intersections. At intersections, just as at curves, experienced drivers were more likely to mitigate a hazard than were inexperienced drivers. For example, consider the results from a scenario at a signalized, through four-way intersection. Drivers traveled in the right lane through a green traffic signal. A truck was stopped in the left lane to make a left turn. The truck in the adjacent left lane obstructs the view of oncoming traffic in viii
the opposing lane across the intersection. (a) Experienced drivers had significantly more glances towards the front of the truck both seven and three seconds before the intersection. The experienced drivers had reduced their speed to 34 mph at the intersection. The novice drivers traveled through the intersection at speed in excess of the posted speed limit (44 mph), while at the same location. Both groups were traveling nearly the same speed nine seconds earlier, but the experienced drivers reduced speed by an average of 6.6 mph, the novice drivers did not. (b) The next question, as above, is whether the differences in hazard mitigation are entirely a consequence of differences in hazard anticipation. While 69% of the experienced drivers slowed at least 3 mph, given that they glanced, only 36% of the novice drivers slowed at least 3 mph, given that they glanced. Thus, novice drivers are definitely worse mitigating the hazard even when it was anticipated. There is evidence that experienced and novice drivers who anticipate the hazard are doing so from as far back as seven seconds or more, though it was not possible from the glance behavior to determine such since at that distance the entire intersection is in view and all drivers are looking ahead at the intersection. (c) While 50% of the experienced drivers slowed at least 3 mph, given that they did not glance, 71% of the novice drivers slowed at least 3 mph, given that they did not glance. Why novice drivers are more likely to slow than experienced drivers, given that they did not glance, is discussed in more detail at the end of Experiment 1. (d) The above differences in the behaviors of novice and experienced drivers should be evidenced in crashes as well as glances and slowing. In fact, novice drivers crashed twelve times compared to six for the experienced drivers. Straight Road Segments. Experienced drivers were more likely to mitigate a hazard than were inexperienced drivers on straight road segments scenarios. First, consider that on a straight road segment, the hazardous situation was easily visible and had two cues: (a) the movement of the lead vehicle, and (b) the possible movement of another hazard. For example, consider the results from a scenario at a mid-block crosswalk. Drivers traveled in the left lane through a mid-block crosswalk. A bus was stopped in the right lane immediately in front of the crosswalk. A lead vehicle moves from the right lane and passed the bus in the left lane. The bus in the adjacent right lane obstructs the view of possible pedestrians in the crosswalk. Although the difference in the speed or likelihood of a glance between the experienced and novice drivers at any given second did not reach significance, the experienced drivers showed consistently safer behaviors. (a) Experienced drivers had nearly half again more glances towards the front of the truck when they were two to three seconds before the crosswalk. Interestingly, two seconds earlier (four and five seconds before the crosswalk), the experienced drivers had reduced their speed to an average of 23.6 mph, while the average speed in the same interval for the novice drivers was 5.2 mph greater. The novice drivers compensated for the difference by slowing more in the last second or two before the crosswalk, many times to avoid a dangerous situation that manifested. Both groups were traveling nearly the same speed when entering the crosswalk. (b) The next question, as above, is whether the differences in hazard mitigation are entirely a consequence of differences in hazard anticipation. When five seconds from the crosswalk 70% of the experienced drivers slowed to a target sped of 23 mph if they glanced to the near extent of the sightline. Whereas 61% of the novice drivers slowed at least 3 mph, given that they glanced. Thus, novice drivers did not mitigate as often even when the hazard was ix
anticipated. (c) While 83% of the experienced drivers slowed at least 3 mph, given that they did not glance, and 85% of the novice drivers slowed at least 3 mph, given that they did not glance. Why novice drivers are nearly as likely to slow as experienced drivers, given that they did not glance is discussed in more detail at the end of Experiment 1. (d) The above differences in the behaviors of novice and experienced drivers should be evidenced in crashes as well as glances and slowing. In fact, novice drivers did crash six times compared to two for the experienced drivers. Experiment 2. In Experiment 2, I wanted to determine whether hazard mitigation skills can be trained using a program that I developed. This program, ACT (Anticipate, Control, Terminate), was developed so that it could be run as a standalone program on a PC or be downloaded from the web, similar to the existing hazard anticipation (RAPT; Pollatsek, et al., 2006) and attention maintenance (FOCAL; Pradhan et al, 2011) training programs. In Experiment 2, the effectiveness of two driver training programs was compared on a driving simulator. One set of participants was assigned to placebo training and one to ACT training (hazard anticipation and hazard mitigation). The same virtual world utilized in Experiment 1 was also utilized in Experiment 2. Since Experiment 2 was focused on the effects of training only on novice drivers, there was no need to retest the experienced driver group in Experiment 2. The ACT training program was used to teach novice drivers first to anticipate hazards and then to make proper hazard mitigation behaviors. With hazard anticipation (RAPT), the simple lesson was that one needed to look for latent hazards (hazards that could materialize, either ones that are hidden – e.g., a pedestrian in front of the truck that suddenly emerges into the path of the driver -- or ones that are visible – e.g., a car in a long line of vehicles in a left turn lane that suddenly pulls into the travel lane occupied by the driver). With attention maintenance (FOCAL), the simple lesson was to glance down no more than two seconds in every six. With hazard mitigation (ACT), the simple lesson was to slow at least eight seconds before arriving at a hazard. Specifically, drivers should slow for a HRECC (Hidden obstacles, Roadside hazards, locations where there is no Escape route, when Closing on a lead vehicle with no passing, and at Curves). Drivers’ mitigation can consist of a driver removing his or her foot from the accelerator or possibly moving within his or her lane to prepare for and neutralize potential hazards. The training included scenarios that are associated with increased crash rates for novice drivers: horizontal curves (neutralize by slowing before entering the curve, moving to the inside of the curve), straight road segments (neutralize by slowing when pedestrians or traffic can suddenly enter) such as the truck blocking crosswalk scenario described earlier, and intersection scenarios (neutralize by slowing, keeping a buffer space) when no initial stop is involved. The concept of slowing for a HRECC (pronounced wreck) was addressed at nine traffic scenes in ACT. After training, a participant received guided training on his or her pretest responses before a posttest on the PC and evaluation drive in the simulator. Curves. Consider again the sharp right curve. (a) Five seconds before entering the curve, some 57% of the trained novice drivers had glanced to the far extent of the curve whereas only 15% of the untrained novice drivers had glanced to the far extent of the curve. Furthermore, 77% of the trained novice drivers mitigated the hazard whereas only 63% of the untrained novice drivers mitigated the hazard. Finally, among those who x
mitigated the curve, some 16% of the trained novice drivers did so at 5 s whereas only 4% of the untrained novice drivers did so at 5 s. (b) The next question is whether there are differences in mitigation behaviors of trained and untrained novice drivers, conditional on a glance being made. Among those drivers who glanced, 89% of the trained drivers mitigated the hazard but 100% of only three untrained drivers mitigated the hazard. It appears that glancing is linked to slowing, regardless of training. (c) Finally, among those drivers who did not glance, 71% of the trained drivers mitigated the hazard but only 63% of the untrained drivers mitigated the hazard. This type of response is consistent with ambient processing and was more common with trained novice drivers. (d) The above differences in the behaviors of novice and experienced drivers should be evidenced in crashes as well as glances and slowing. In fact, placebo trained drivers did crash five times compared to only once for the ACT trained drivers. Intersections. Consider intersections next, in particular the four-way intersection. (a) Six and seven seconds before entering the intersection, some 28-32% of the ACT trained drivers had glanced to the near extent of the sight line whereas only 13% of the placebo trained drivers had glanced to the near extent of the sight line. The near extent is the farthest one can look toward the front of the truck in the adjacent lane. Furthermore, 68% of the ACT trained drivers mitigated the hazard (slowed to a target speed of 37 mph) whereas 29% of the placebo trained drivers mitigated the hazard. (b) The next question is whether there are differences in mitigation behaviors of placebo trained and ACT trained drivers, conditional on a glance being made. Among those drivers who glanced, 71% of the ACT trained drivers mitigated the hazard compared to only 40% of the placebo trained drivers. (c). Finally, among those drivers who did not glance, 50% of the ACT trained novice drivers mitigated the hazard while only 38% of the placebo trained drivers mitigated the hazard. This type of response is consistent with drivers who process the driving environment ambiently (without direct glances) which was more common with experienced drivers. (d) The above differences in the behaviors of placebo trained and ACT trained drivers should be evidenced in crashes as well as glances and slowing. In fact, placebo trained novice drivers crashed thirteen times compared to six for ACT trained novice drivers. Straight Road Segments. ACT trained drivers were more likely to mitigate a hazard than were placebo trained drivers. First, consider that on a straight road segment where there was a lead vehicle changing lanes and a bus blocking a crosswalk. The bus in the adjacent right lane obstructs the view of a pedestrian in the crosswalk (when materialized). Likely because of the several cues available to the untrained as well as the trained drivers, glances and speed at any given second did not reach significance. Nevertheless, the ACT trained drivers showed consistently safer behaviors. (a) Most (56%) of the ACT trained drivers had already slowed to a target speed of 23 mph when ten seconds from the crosswalk. At the same time, only 44% of the placebo trained drivers had slowed to the target speed when ten seconds before the crosswalk. While the speed choices did not differ significantly, the ACT trained drivers consistently traveled an average speed of more than one mph slower than did the placebo drivers. The peak difference was at nine seconds before the crosswalk when the 33% fewer placebo trained drivers slowed to the target speed of 23 mph than did experienced drivers. Both groups were traveling nearly the same speed when entering the crosswalk. (b) The next question is whether the differences in hazard mitigation are entirely a consequence of differences xi
in hazard anticipation. When five seconds from the crosswalk 67% of the ACT trained drivers slowed to a target speed of less than 23 mph, given that they glanced, exactly the same percent of the placebo trained drivers slowed at least 3 mph, given that they glanced. This suggests that placebo trained drivers mitigated as often if they anticipated. (c) However, 75% of the ACT trained drivers slowed to the target speed of 23 mph, given that they did not glance, compared to 52% of the placebo trained drivers slowed to the target speed, given that they did not glance. Overall, if the driver slowed or not, ACT trained drivers were more likely to slow (69% vs. 58%) and glance (67% vs. 42%) to the near extent. (d) The above differences in the behaviors of placebo trained and ACT trained drivers was evident in crashes as well as glances and slowing. Placebo trained drivers crashed four times compared to one for the ACT trained drivers. Comparing Experienced and ACT Trained Novice Drivers. The goal of the research was to teach a group of novice drivers to select speeds more like experienced drivers. The evidence suggests that this training did that. The conditional probabilities suggest that ACT trained drivers performed nearly as well as the experienced drivers. At a sharp curve right experienced drivers glanced slightly more (68% of ACT trained novice drivers as compared to 70% of experienced drivers), and they slowed similarly (77% to 78%). However, if given a glance, more experienced slowed (90% of experienced drivers compared to 89% of ACT trained novice drivers). Furthermore, experienced drivers attained the target speed on average three seconds before the event; while ACT trained novice drivers attained the target speed on average one second later. While one full second could be very valuable to the driver, consider that the novice drivers in Experiment 1 and placebo trained novice drivers in Experiment 2 never reached an average target speed. At intersections, both the ACT trained novice drivers and experienced drivers glanced toward the near extent slowed 67% of the time. When there was not a glance, ACT trained drivers slowed 89% compared to the experienced drivers 50%. Again, the ACT trained and experienced drivers slowed an average of more than 6 mph in the seven seconds before the intersection, while both the novice and placebo groups slowed less than 2 mph. General Discussion. With our four groups, across both experiments (six scenarios), when drivers from each group glanced, the probability of reaching target speed or slowing averaged across the four groups was 65%. Regardless of training, the results show that glancing toward the extent of the sightline is an effective tactic for drivers. It is clear that training has an effect on novice drivers, both on hazard anticipation and on hazard mitigation. In Experiment 2, on the sharp curve placebo trained novice drivers glanced 37% of the time and slowed 63% of the time while ACT trained novice drivers glanced 68% of the time and slowed 77% of the time. The differences are greater at the curve left where more than two times as many ACT trained drivers slowed to the target speed and half the ACT trained drivers has glanced to the far extent when seven seconds before the curve. Conversely, not until the drivers were entering the curve did half the novice drivers make a far extent glance. As evidence that the hazard mitigation training on curves was successful, one needs to condition on a glance. Doing this, on curves novice drivers without ACT training (novices from Experiment 1 and placebo trained in Experiment 2) glanced in the target zone 11% and 17% respectively, yet with ACT training, 56% made a glance to the far extent, approximately four-fold increase. In xii
Experiment 2, at intersections placebo trained novice drivers glanced 56% of the time and slowed 39% of the time while ACT trained novice drivers glanced 78% of the time and slowed 67% of the time. As evidence that the hazard mitigation training on intersections was successful, one needs to condition on a glance. With a glance, the probability of ACT trained novice drivers slowing to the 37 mph target speed or less before entering the intersection was 0.71 and the probability of placebo trained novice drivers slowing to target speed or less was 0.40. Without a glance, the probability of an ACT trained novice driver slowing below the target speed before entering the intersection was 0.50 and the probability of placebo trained novice driver slowing to a speed below the target speed was 0.38. In general, at non-hazardous locations, the speed choices of drivers could not be differentiated based upon experience or training, but when a hazard was present, experienced drivers both selected lower speeds and earlier deceleration at each of the locations. ACT trained drivers selected slower speeds at every second from five seconds until the event location compared to placebo trained or novice drivers at both the sharp curve and dangerous obstructed intersection. The speed choices were clearly related to crashes. In Experiment 1, all drivers who reached a target speed before the curve avoided a crash and no driver who crashed obtained the target speed. In Experiment 2, no driver who crashed had a reduction in speed of less than 3 mph. Finally, glances were also clearly related to crashes. ACT trained drivers crashed eight times, while novice drivers and placebo trained drivers each crashed 22 times, 13 at intersections, four at the bus blocking the crosswalk, and five at the sharp curve right. Of the fourteen crashes at the sharp curve by the four groups, only two drivers glanced across the curve. If a glance was the only way to mitigate a hazard, we would expect that no driver would slow unless they glance, which was not the case. However, glances across the curve were strongly associated with favorable speed management, while failing to glance is not. Finally, the ACT training program that taught rule oriented behavior (i.e., slow for a HRECC) both improved novice drivers’ ability to anticipate hazards and to manage their speed. ACT training might prove even more effective if it were to follow RAPT training. Clearly, glance behavior is shown to have an influence on speed management behaviors at risky areas such as sharp curves and four-way intersections. ACT training supplements that training by teaching anticipatory slowing that has been shown in this research to improve speed management both with and without a glance.
xiii
TABLE OF CONTENTS Page ACKNOWLEDGMENTS ...................................................................................... v ABSTRACT .......................................................................................................... vi EXECUTIVE SUMMARY ................................................................................. viii LIST OF TABLES ................................................................................................ xx LIST OF FIGURES ........................................................................................... xxix CHAPTER 1. HAZARD MITIGATION BEHAVIORS: DIFFERENCES BETWEEN EXPERIENCED AND NOVICE DRIVERS AND TRAINING PROGRAMS TO DECREASE THE DIFFERENCES.................................................................................... 1 1.1 Introduction .................................................................................................... 1 1.2 Crash Statistics Show Novice are Overrepresented ....................................... 2 1.3 Crash Types for Novice Drivers .................................................................... 2 1.4 Causes of the Crashes: Implications of the Crash Types ............................... 3 1.5 Problem Statement ......................................................................................... 6 1.5.1 Experiment 1: Understanding the Shortcomings of Novice Driver Hazard Mitigation Strategies .................................................................. 6 1.5.2 Experiment 2: Developing a Hazard Mitigation Training Program ........ 7 2.
LITERATURE REVIEW ........................................................................... 10 2.1 Hazard Anticipation Behaviors .................................................................... 10 2.1.1 Knowledge or Workload Handling Deficits? ........................................ 13 2.1.2 Hazard Anticipation Training: Simulator Evaluation ........................... 15 2.1.3 Hazard Anticipation Training: Evaluation in the Field ......................... 17 2.2 Hazard Mitigation and Hazard Anticipation Behaviors ............................... 18 2.2.1 Horizontal Curves ................................................................................. 19 2.2.1.1 Hazard Anticipation (Glance Behaviors) at Horizontal Curves ..... 24 2.2.1.2 Hazard Mitigation: Speed Choice when Approaching or Negotiating a Horizontal Curve ......................................................................... 26
xiv
2.2.1.3 Hazard Mitigation: Lane Position when Approaching or Negotiating a Horizontal Curve ......................................................................... 30 2.2.2 Intersections: Vehicular Obstructions and Intrusions ........................... 31 2.2.2.1 Hazard Anticipation: Glance Behaviors When Approaching and Within Intersections ....................................................................... 33 2.2.2.2 Hazard Mitigation: Speed Choice and Positioning When Approaching and Within Intersections35 ....................................... 35 2.2.3 Straight Road Segments: When there are Hazards Near or within The Road on Straight Road Segments ......................................................... 36 2.2.3.1 Hazard Anticipation (Glance Behaviors) when there are Hazards Near or within The Road on Straight Road Segments ................... 37 2.2.3.2 Hazard Mitigation: Speed Choice When there are Hazards near or within The Road on Straight Road Segments ................................ 38 2.2.3.3 Hazard Mitigation: Lane Position in when there are Hazards Near or within The Road on Tangent Road Segments ................................ 42 2.3 Summary of the How Driving Speed Changes at Horizontal Curves, Straight Road Segments and Intersections ............................................................... 44 3.
RESEARCH GOALS AND SIGNIFICANCE ........................................... 49 3.1 Hypothesis related to glance behaviors........................................................ 49 3.2 Hypotheses related to speed choice ............................................................. 54 3.3 Hypotheses related to conditional probabilities ........................................... 55
4. EXPERIMENT 1: FINDING THE DIFFERENCES BETWEEN NOVICE AND EXPERIENCED DRIVER BEHAVIOR ................................................................ 56 4.1 Participants ................................................................................................... 57 4.2 Internal Review Board Approval ................................................................. 58 4.3 Equipment .................................................................................................... 58 4.4 Experimental Design: The Virtual Drives and Order of Exposure .............. 60 xv
4.4.1 Geometry: Horizontal Curves ............................................................... 64 4.4.1.1 Horizontal Curve Arrangements ..................................................... 68 4.4.1.2 Horizontal Curve Measures ............................................................ 74 4.4.2 Geometry: Intersections ........................................................................ 75 4.4.2.1 Intersection Arrangements .............................................................. 75 4.4.2.2 Intersection Measurements ............................................................. 82 4.4.3 Geometry: Straight Road Segments ...................................................... 83 4.4.3.1 Straight Road Arrangements ........................................................... 83 4.4.3.2 Straight Segment Measurements..................................................... 90 4.5 Procedure ..................................................................................................... 91 4.6 Data Collection and Analysis....................................................................... 93 4.7 Results .......................................................................................................... 97 4.7.1 Geometry: Horizontal curves ................................................................ 97 4.7.1.1 Aggregate Glance, Braking and Slowing Behaviors, and Lane Position ........................................................................................... 97 4.7.1.2 Conditional Braking, Slowing and Crashes .................................. 109 4.7.2 Geometry: Intersections ..................................................................... 116 4.7.2.1 Aggregate Glance, Braking and Slowing Behaviors .................... 119 4.7.2.2 Conditional Braking and Slowing and Crashes ............................ 132 4.7.3 Geometry: Straight Segments ............................................................. 137 4.7.3.1 Aggregate Glance, Braking and Slowing Behaviors .................... 138 4.7.3.2 Conditional Braking and Slowing and Crashes ............................ 148 4. 8 Discussion. ................................................................................................ 153
5. EXPERIMENT 2: EVALUATION OF THE RISK MITIGATING BEHAVIORS OF TRAINED AND UNTRAINED NOVICE DRIVERS USING ANTICIPATION-CONTROL & TERMINATION [ACT]............................................ 156 xvi
5.1 Development of ACT Training Program ................................................... 156 5.1.1 ACT: Basic Elements of the Training Program -- Modules and Responses ........................................................................................... 157 5.1.2 ACT: Practice Module ........................................................................ 160 5.1.3 ACT: Pre-Test Module ....................................................................... 160 5.1.4 ACT: The Theory Behind the Development of the Training Module -Error Based Training: Mistakes, Mediation, and Mastery (3M) ........ 163 5.1.5 The ACT Training Module ................................................................. 166 5.1.6 ACT: The Post-Test Module ............................................................... 170 5.1.7 ACT: Recording User Responses ........................................................ 170 5.2 Development of the Placebo Training Program......................................... 171 5.3 Participants ................................................................................................. 171 5.4 Experimental Design .................................................................................. 171 5.5 Procedure ................................................................................................... 172 5.6 Data Collection and Analysis..................................................................... 173 5.7 Results ........................................................................................................ 174 5.7.1 PC Posttest Performance of ACT Trained Versus Controls (Placebo Trained) .............................................................................................. 174 5.7.2 Geometry: Horizontal curves .............................................................. 178 5.7.2.1 Aggregate Glance, Braking and Slowing Behaviors, and Lane Position ......................................................................................... 179 5.7.2.2 Conditional Braking, Slowing and Crashes .................................. 189 5.7.3 Geometry: Intersections ..................................................................... 196 5.7.3.1 Aggregate Glance, Braking and Slowing Behaviors, and Lane Position ......................................................................................... 198 5.7.3.2 Conditional Braking and Slowing and Crashes ............................ 211 5.7.4 Geometry: Straight Segments ............................................................. 215 5.7.4.1 Aggregate Glance, Braking and Slowing Behaviors .................... 215 5.7.4.2 Conditional Braking and Slowing and Crashes ............................ 226 5.8 Pattern of Glances, Speed Loss, and Lane Position ................................... 231 xvii
6.
DISCUSSION ........................................................................................... 235 6.1 Experiment 1 ............................................................................................ 236 6.1.1 Curves.................................................................................................. 237 6.1.2 Intersections ........................................................................................ 243 6.1.3 Straight Segments ................................................................................ 247 6.1.4 Implications of All Findings in Experiment 1 ..................................... 251 6.2 Experiment 2 .............................................................................................. 254 6.2.1 ACT Training Program ....................................................................... 254 6.2.2 Simulator Evaluation of ACT ............................................................. 255 6.2.3 Training Implications of the Research ................................................ 260 6.2.4 Other Implications. .............................................................................. 261 6.3 Comparison of Experiment 1 to Experiment 2 .......................................... 262 6.4 Applications and Generalizability of This Research .............................. 263
APPENDICES A. INFORMED CONSENT ............................................................................ 266 B. HUMAN PERFORMANCE LABORATORY POST-STUDY QUESTIONNAIRE .................................................................................... 270 C. INSTRUCTIONS TO PARTICIPANTS: SIMULATOR DRIVES............ 271 D. EYE GLANCE SCORING TRAINING ..................................................... 273 E. ACT TRAINING ......................................................................................... 281 REFERENCES ................................................................................................... 307
xviii
LIST OF TABLES Table
Page
1: Three most common crash types from the NMVVCS data involving 677 teens who were involved in serious crashes in 2005-2006 from NMVVCS Query. ..................................................................................... 2 2: Percentage of trials on which risky features of scenarios was fixated.................. 11 3. The results from Pradhan et al (2005) categorized by obstruction type ............... 13 4: Crashes per 10,000 miles traveled based upon the FARS 2009 crash history database and NHTSA yearly mileage data for various age groups. MV refers to motor vehicles. http://wwwfars.nhtsa.dot.gov/QueryTool/QuerySection/SelectFields.aspx ............. 20 5: Research listing the amplitude of speed loss and factors that are correlated to speed loss. ........................................................................................... 40 6: A summary of the topographical feature and how drivers’ speed choices change due to a variation in that feature. ................................................ 46 7: Counter-balancing of materialized hazards. (“I” refers to intersection, S refers to straight road segments (Straight road); C refers to curves. 1, 2, and 3 refer to the nominal identifier. If highlighted and an asterisk, a potential hazard was materialized.)........................................ 64 8: Description of the two curves where driver behaviors were recorded in context with the previously stated research. One might say that the two curves represent an easy curve left and a more dangerous curve to the right. .................................................................................... 73 9: Description of the two intersections where driver behaviors were recorded in context with the previously stated research. ....................................... 81 10: Description of the two straight segments where driver behaviors were recorded in context with the previously stated research. ........................ 89
xix
11: Comparison of experienced and novice drivers’ glances within each 1second epoch before a curve. (Numbers in bold yellow indicate significant differences.)......................................................................... 100 12: Cumulative percentage of experienced and novice drivers who glanced toward the far extent of the sight line in each second while approaching a sharp curve to the right. ................................................. 102 13: Percentages of experienced and novice drivers that slowed to target speed of 20 mph when approaching a sharp curve to the right. (Numbers in bold yellow indicate significant differences.) ................................... 106 14: Percentages of experienced and novice drivers that slowed to target speed of 34 mph when approaching a sharp curve to the right. (Numbers in bold yellow indicate significant differences.) ................................... 107 15: Average lane position of experienced and novice drivers when approaching a sharp curve right. (Distances refer to the position left (negative in parentheses) or right (positive) of the center of the lane.)...................................................................................................... 109 16: Average lane position of experienced and novice drivers when approaching a moderate curve left. (Distances refer to the position left (negative in parentheses) or right (positive) of the center of the lane. Bold yellow signifies that experienced and novice drivers selected significantly different lane positions.) .................................... 110 17: Number of experienced and novice drivers that slowed to target speed after glancing to the far extent when five to eight seconds before a sharp curve right. .................................................................................. 111 18: Percentage of experienced and novice drivers that slowed if they had glanced to the far extent or had not glanced to the far extent. (Numbers in bold yellow indicate a significant difference.) ................. 112 19: Number of experienced and novice drivers that slowed to target speed after glancing to the far extent when five to eight seconds before a moderate curve left. .............................................................................. 113 20: Percentage of experienced and novice drivers that slowed if they had glanced to the far extent or had not glanced to the far extent when approaching a curve left. (Numbers in bold yellow indicate significant differences.)......................................................................... 114
xx
21: Percentage of drivers who crashed, slowed before crashing, slowed when not crashing or glanced and slowed without crashing. Numbers in bold yellow suggest a significant difference between experienced and novice drivers. ................................................................................ 116 22: Comparison of the percentage of experienced and novice drivers that crashed based upon whether the driver slowed to the target speed or did not slow to the target speed. ....................................................... 116 23: Percentage glances toward the near extent in each second as experienced and novice drivers approached an intersection that was obstructed by a left turning truck. (Numbers in bold yellow indicate significant differences.)......................................................................... 122 24: The cumulative percentage of ACT trained and placebo trained drivers that completed both side road glances in each second (0 – 9 s) before the curve. Also, the percentage of ACT trained and placebo trained drivers that made secondary glances, and finally (bottom row), the percentage drivers that completed the necessary side road glances and a secondary glance before turning (Numbers in bold yellow indicate that experienced drivers were significantly more likely to make a glance toward in that second.) ................................... 125 25: Comparison of experienced and novice drivers’ Glances within Each 1second epoch when approaching an intersection with its view obstructed by a left-turning truck. (Numbers in bold yellow indicate significant differences.) ........................................................... 127 26: Cumulative percentage of experienced and novice drivers to slow to a target speed of zero mph in the nine seconds before arriving at a busy intersection. (Numbers in bold yellow indicate significant differences.) .......................................................................................... 129 27: Average lane position of experienced and novice drivers when approaching an intersection with a left turning truck obstructing the view. (Distances refer to the position left (negative) or right (positive) of the center of the lane. Numbers in bold yellow indicate that experienced and novice drivers selected significantly different lane positions.) ....................................................................... 131 28: Average lane position of experienced and novice drivers when approaching a busy four-way intersection. (Distances refer to the position left (negative) or right (positive) of the center of the lane. Numbers in bold yellow indicate that experienced and novice drivers selected significantly different lane positions.) ........................ 132
xxi
29: Number of experienced and novice drivers that slowed to target speed and glanced to the far extent when two to five seconds before an intersection with a left turning truck. .................................................... 133 30: Percentage of experienced and novice drivers that slowed if they had glanced to the near extent or had not glanced to the near extent when approaching an intersection that was obstructed by a left turning truck. ......................................................................................... 134 31: Number of experienced and novice drivers that slowed to target speed of 0-mph and glanced toward all three legs of the intersection before turning left ............................................................................................. 135 32: Percentage of experienced and novice drivers that slowed if they had glanced to the far extent or had not glanced to the far extent. .............. 135 33: Percentage of drivers who slowed before crashing, or did not slow before crashing. ................................................................................................ 137 34: Comparison of the percentage of experienced and novice drivers that crashed based upon whether the driver slowed to the target speed or did not slow to the target speed. (Numbers in bold yellow indicate significant differences.) ........................................................... 137 35: Percentage of experienced and novice drivers that glanced toward the near extent when approaching a bus stopped in front of a crosswalk........... 140 36: Cumulative percentage that glanced both toward a pedestrian along the roadside left, and a truck parked perpendicularly along the roadside right before passing the truck. ............................................................... 142 37: Percentage experienced and novice drivers that slowed to a target speed of 23 mph in each second when approaching a bus stopped at a crosswalk............................................................................................... 144 38: Cumulative percentage to slow to a target speed of 23 mph when approaching a roadside pedestrian and truck. ....................................... 146 39: Average lane position of experienced and novice drivers when approaching a bus stopped at a crosswalk. (Distances refer to the position left (negative in parentheses) or right (positive) of the center of the lane.)................................................................................. 147 40: Average lane position relative to the center of the lane when approaching a roadside truck to the right and pedestrian to the roadside left. (Numbers in bold yellow indicate significant differences.) .................. 148 xxii
41: Number of experienced and novice drivers that slowed to 23 mph based upon whether they glanced to the near extent two to four seconds before the crosswalk. ............................................................................ 149 42: Percent of experienced and novice drivers that glanced to the near extent five to eight seconds before the crosswalk and slowed to a target speed of 23 mph as they approached a bus parked in front of a crosswalk............................................................................................... 150 43: Number of experienced and novice drivers that slowed to 23 mph and glanced to roadside pedestrian and truck two to four seconds before obstacle. ..................................................................................... 151 44: Percent of experienced and novice drivers that glanced and slowed as they approached a roadside pedestrian and truck ......................................... 151 45: Percentage of experienced and novice drivers that slowed to a speed of 23 mph based upon whether they crashed or not.(Numbers in bold yellow indicate significant differences.) ............................................... 152 46: Percentage of drivers that crashed based upon whether they slowed to a speed of 23 mph or not. (Numbers in bold yellow indicate significant differences.)......................................................................... 153 47: Screen shots from the ACT program and a description of each training scenario, as well as the simulator evaluation scenario that best matches each training scenario. Circles in the photographs depict the mediation training that showed the trainees the selections made by exemplary experienced drivers ........................................................ 161 48: The four training slides for ACT Scenario 1, van blocking crosswalk scenario. Here, the drivers were taught about the concepts of hazard anticipation (glancing toward latent hazards), Slowing for HRECCS (Hidden hazards and roadside obstacles), maintaining a Safety Bubble by changing lane position (third slide), and using the horn when possible (when not steering) to keep others out of the Safety Bubble. ................................................................................. 168 49: Comparison of ACT trained and Placebo trained Drivers’ Glances within Each 1-second epoch before a curve. (Numbers in bold yellow indicate significant differences.) ........................................................... 180 50: Cumulative percentage of ACT trained and placebo trained drivers who glanced toward the far extent of the sight line in each second while approaching a sharp curve to the right. ................................................. 182
xxiii
51: Percentages of ACT trained and placebo trained drivers that slowed to target speed of 20 mph when approaching a sharp curve to the right. ...................................................................................................... 186 52: Percentages of ACT trained and placebo trained drivers that slowed to target speed of 34 mph when approaching a sharp curve to the right. (Numbers in bold yellow indicate significant differences.) ....... 186 53: Average lane position of ACT trained and placebo trained drivers when approaching a sharp curve right. Distances refer to the position left (negative in parentheses) or right (positive) of the center of the lane ........................................................................................................ 189 54: Average lane position of ACT trained and placebo trained drivers when approaching a moderate curve left. Distances refer to the position left (negative in parentheses) or right (positive) of the center of the lane. ....................................................................................................... 189 55: Number of ACT trained and placebo trained drivers that slowed to target speed after glancing to the far extent when five to eight seconds before a sharp curve right. .................................................................... 191 56: Percentage of ACT trained and placebo trained drivers that slowed if they had glanced to the far extent or had not glanced to the far extent. (Numbers in bold yellow indicate significant differences.) .................. 191 57: Number of ACT trained and placebo trained drivers that slowed to target speed after glancing to the far extent when five to eight seconds before a moderate curve left. ................................................................ 192 58: Percentage of ACT trained and placebo trained drivers that slowed if they had glanced to the far extent or had not glanced to the far extent when approaching a curve left. (Numbers in bold yellow indicate significant differences.) ........................................................................ 193 59: Percentage of ACT and Placebo trained drivers who crashed, slowed before crashing given that they crashes, slowed given that they did not crash, glanced and slowed given that they crashed and glanced and slowed given that they did not crash .............................................. 195 60: Comparison of the percentage of ACT trained and placebo trained drivers that crashed based upon whether the driver slowed to the target speed or did not slow to the target speed. . (Numbers in bold yellow indicate significant differences.) ............................................... 195
xxiv
61: Comparison of experienced and novice drivers’ Glances within Each 1second epoch when approaching an intersection with its view obstructed by a left-turning truck. (Numbers in bold yellow indicate significant differences.) ........................................................... 202 62: Cumulative percentage of ACT trained and placebo trained drivers that completed glances all three glances to all three legs of the intersection. The cumulative percentage of ACT trained and placebo trained drivers that completed both side road glances in each second (0 – 9 s) before the curve. Also, the percentage of ACT trained and placebo trained drivers that made secondary glances and finally (bottom row), the percentage drivers that completed the necessary side road glances and a secondary glance before turning. ...................................................................................... 204 63: Comparison of percentage of ACT trained and placebo trained drivers’ Glances within Each 1-second epoch when approaching an intersection with its view obstructed by a left-turning truck. (Numbers in bold yellow indicate significant differences.) .................. 206 64: Cumulative percentage of ACT trained and placebo trained drivers to slowed to a target speed of zero mph in the nine seconds before arriving at a busy intersection. (Numbers in bold yellow indicate significant differences.) ........................................................................ 208 65: Average lane position of ACT trained and placebo trained drivers when approaching an intersection with a left turning truck obstructing the view. (Distances refer to the position left (negative) or right (positive) of the center of the lane. Numbers in bold yellow indicate significant differences.) ........................................................... 209 66: Average lane position of ACT trained and placebo trained drivers when approaching a busy four-way intersection. Distances refer to the position left (negative) or right (positive) of the center of the lane. ..... 210 67: Number of ACT trained and placebo trained drivers that slowed to target speed after glancing to the far extent when two to five seconds before an intersection with a left turning truck. .................................... 211 68: Percentage of ACT trained and placebo trained drivers that slowed if they had glanced to the near extent or had not glanced to the near extent when approaching an intersection that was obstructed by a left turning truck. ......................................................................................... 212
xxv
69: Number of ACT trained and placebo trained drivers that slowed to target speed of 0-mph and glanced toward all three legs of the intersection before turning left. ............................................................. 213 70: Percentage of ACT trained and placebo trained drivers that slowed if they had glanced to the far extent or had not glanced to the far extent. ....... 213 71: Percentage of drivers who slowed before crashing, or did not slow before crashing at an intersection with the view obstructed by a left turning truck. ......................................................................................... 214 72: Comparison of the percentage of ACT trained and placebo trained drivers that crashed at an intersection with a left turning truck based upon whether the driver slowed to the target speed or did not slow to the target speed. (Numbers in bold yellow indicate significant differences.) .......................................................................................... 215 73: Percentage of ACT trained and placebo trained drivers that glanced toward the near extent when approaching a bus stopped in front of a crosswalk. (Numbers in bold yellow indicate significant differences.) ......................................................................................... 217 74: Cumulative percentage that glanced both toward a pedestrian along the roadside left, and a truck parked perpendicularly along the roadside right before passing the truck. ............................................................... 219 75: Percentage ACT trained and placebo trained drivers that slowed to a target speed of 23 mph in each second when approaching a bus stopped at a crosswalk. (Numbers in bold yellow indicate significant differences.) ......................................................................................... 222 76: Cumulative percentage to slow to a target speed of 23 mph when approaching a roadside pedestrian and truck ........................................ 224 77: Average lane position of ACT trained and placebo trained drivers when approaching a bus stopped at a crosswalk. Distances refer to the position left (negative in parentheses) or right (positive) of the center of the lane. (Numbers in bold yellow indicate significant differences.) .......................................................................................... 225 78: Average lane position relative to the center of the lane when approaching a roadside truck to the right and pedestrian to the roadside left. (Numbers in bold yellow indicate significant differences.) .................. 226
xxvi
79: Number of ACT trained and placebo trained drivers that slowed to 23 mph based upon whether they glanced to the near extent two to four seconds before the crosswalk. ............................................................... 227 80: Percent of ACT trained and placebo trained drivers that glanced to the near extent when two to four seconds before the crosswalk and slowed as they approached a bus parked in front of a crosswalk. ........ 227 81: Number of ACT trained and placebo trained drivers that slowed to 23 mph based upon whether they glanced to the near extent two to four seconds before the roadside pedestrian and truck. ................................ 228 82: Percent of ACT trained and placebo trained drivers that glanced and slowed as they approached a roadside pedestrian and truck ................. 229 83: Percentage of ACT trained and placebo trained drivers that slowed to a speed of 23 mph based upon whether they crashed or not. (Numbers in bold yellow indicate significant differences.) .................. 230 84: Percentage of drivers that crashed based upon whether they slowed to a speed of 23 mph or not.......................................................................... 230
xxvii
LIST OF FIGURES Figure
Page
1: NMVCCS Crash configurations for the 2nd most teenage serious crashes (2) and the 7th most teenage serious crashes (1) and the 4th most teenage serious crashes (7)...................................................................... 24 2: Eye glance of a young driver. Left: licensed for two months Right: Licensed for ten years. ............................................................................ 25 3: Of all crash types, the LTAP configurations represent the first (scenario 68) and 6th greatest (scenario 69) serious crash situations for novice drivers according to NMVCCS, 2008. Red represents the subject driver and blue is the threat. .................................................................... 31 4: The 3rd greatest crash configuration for serious teenage crashes (82), and the 8th greatest serious crash configuration involving teenage drivers (NMVCCS, 2008). Red represents the subject driver and blue is the threat. ..................................................................................... 32 5: Driving simulator at the University of Massachusetts-Amherst showing the truck near crosswalk scenario. ................................................................ 59 6: Examples of far extent, near extent and ahead-opposite glances. ....................... 67 7: Largest radius curve that drivers will negotiate, a curve to the left. Aerial sketch was shown to the right and screen-shots from the RTI simulator showing the driver’s view was shown to the left. The truck in the background is parked along the opposite side of the road and a car (red) moves out from behind the truck, crossing the center lines by one-foot. The blue car depicts a lead vehicle that moves through the curve slightly more than ten seconds before our driver. ...................................................................................................... 69 8: Curve 3, a tightening curve to the right. Right depicts an aerial sketch and left shows screen-shots from the RTI simulator for this scenario. The first three screen-shots are taken as the driver moves through the curve. ................................................................................................. 71
xxviii
9: Negotiating a horizontal curve to the right with a queue of traffic at a stop sign beyond the curve. Aerial sketch to the right and screen-shots from the RTI simulator on the left. The blue car slows and turns left, leaving the right lane open for our driver to pass. ........................... 72 10: Truck left turn scenario or NMVCCS scenario 69. (The traffic signal remains green. The approaching driver was able to see a vehicle travel through the intersection. After the oncoming blue car passes by, the truck in the next lane slows to a stop. When the driver gets to a position near the rear of the truck, an oncoming car starts to turn left across the path of the driver. ..................................................... 76 11: Opposing left turning truck in the left (higher speed) lane. (The view of traffic in the lane to the right of the truck was obstructed. To the right is a sketch of the scenario from above. To the left is a screenshot from the RTI simulator.) ................................................................. 77 12: Side road traffic to the right, an oncoming vehicle approximately 8 ½ seconds away and a previously unseen car approaching from the left when materialized. (A pedestrian was standing on the far left street corner (behind the A-pillar). The aerial view is shown to the right and screen-shots from the RTI simulated scenario is shown to the left.) ................................................................................................... 80 13: Bus blocking crosswalk scenario. (The approaching driver is able to see a lead vehicle pass the bus and travel through the crosswalk. After that car clears a pedestrian might suddenly emerge into the path of the lead vehicle from a position in front of the bus.) .............................. 85 14: Pedestrian in the work zone. When negotiating a right lane closure a pedestrian might or might not move into the drivers’ travel lane. (To the right is a sketch of the scenario from above. To the left are screen-shots from the RTI simulator showing this scenario.)................. 86 15: Side road car can be to the right. (This vehicle will move quickly toward this position and stop when depicted if not materialized. The sketch to the right shows an overhead view, the screen-shots to the right are taken from the RTI simulated scenario.) .................................. 87 16: The cumulative percentage of experienced and novice drivers that made a glance to the far extent (across the curve) as they approached a sharp curve to the right. (*The asterisk - The difference between the experienced and novice drivers’ glances were significant.) ............. 99
xxix
17: The cumulative percentage of experienced and novice drivers that made a glance to the far extent (across the curve) as they approached a curve to the left. .................................................................................... 101 18: Average speed of experienced and novice drivers in the ten seconds before reaching a sharp curve to the right. (* The asterisk refers to times when significantly more experienced drivers slowed to a target speed of 20 mph.) .................................................................................. 103 19: Average speed of experienced and novice drivers in the ten seconds before reaching a moderate curve left. (* The asterisk refers to times when significantly more experienced drivers slowed to a target speed of 34 mph.) .................................................................................. 104 20: Percentage of experienced and novice drivers that glanced to the near extent left in each one-second periods when approaching an intersection with the left lane obstructed by a turning truck. (* Asterisk signifies that experienced drivers were much more likely to make a glance to the near extent in that second.) ................... 121 21: Cumulative percentage of drivers that completed both glances to the left and right side road in the nine seconds before turning. Also, the percentage of drivers that made a secondary glance toward oncoming traffic before turning, and lastly, the percentage of drivers that both completed the side road glances and the secondary glance before turning. (* The asterisk signifies that experienced drivers were significantly more likely to make a glance toward the side road in that second.) ......................................... 123 22: Average speed of the experienced and novice drivers during each second when approaching an intersection with an obstructed view due to a left turning truck in the adjacent left lane. (* The asterisk suggests that significantly more experienced drivers slowed to the target speed of 37 mph in that second.) .......................................................... 126 23: Average speed of the experienced and novice drivers during each second when approaching a busy four-way intersection. ................................. 128 24: Cumulative percentage of experienced and novice drivers that made a glance toward the near extent in each second when approaching a bus stopped and partially blocking the view of a crosswalk. ................ 139 25: Cumulative percentage of experienced and novice drivers that glanced toward both a roadside pedestrian to the left and perpendicular truck to the right in each second when approaching the truck. ............. 141
xxx
26: Average speed of experienced and novice drivers when approaching a bus stopped in front of a mid-block crosswalk. ........................................... 143 27: Average speed of experienced and novice drivers when approaching a roadside pedestrian to the left and truck to the right. ............................ 145 28: The highlighted regions represent the times where experienced drivers outperformed the novice drivers at the three locations where drivers crashed. ..................................................................................... 154 29: Screen shot from the ACT training program. (The pedals at the bottom right represent the throttle - green, “off-throttle” – small gray pedal, and brake –large gray pedal. The five arrows represent positions left and right of lane-center. The horn icon shows where the horn is located and users may click anywhere in the forward view to indicate where they wish to glance.) ........................................ 160 30. Safety bubble involves staying away from other obstacles by slowing or horn use (forward arrow), or moving left or right within the lane. ....... 167 31: Number of Correct responses for ACT trained drivers in the pretest and posttest compared to the correct responses by the placebo trained drivers in the posttest. ........................................................................... 176 32: The cumulative percentage of ACT trained and placebo trained drivers that made a glance to the far extent (across the curve) as they approached a sharp curve to the right. (* The Asterisks - The difference between the ACT trained and placebo trained drivers’ glances were significant.)...................................................................... 179 33: The cumulative percentage of ACT trained and placebo trained drivers that made a glance to the far extent (across the curve) as they approached a curve to the left. .............................................................. 181 34: Average speed of ACT trained and placebo trained drivers in the ten seconds before reaching a sharp curve to the right. .............................. 183 35: Average speed of ACT trained and placebo trained drivers in the ten seconds before reaching a moderate curve left. (*The Asterisks refers to times when significantly more ACT trained drivers slowed to a target speed of 34 mph.) .................................................... 185 36: Percentage of ACT trained and placebo trained drivers that glanced to the near extent left in each one-second periods when approaching an intersection with the left lane obstructed by a turning truck ................. 202
xxxi
37: Cumulative percentage of drivers that completed both glances to the left and right side road in the nine seconds before turning. Also, the percentage of drivers that made a secondary glance toward oncoming traffic before turning, and lastly, the percentage of drivers that both completed the side road glances and the secondary glance before turning ........................................................... 204 38: Average speed of the ACT trained and placebo trained drivers during each second when approaching an intersection with an obstructed view due to a left turning truck in the adjacent left lane. * The Asterisks - suggests that significantly more ACT trained drivers slowed to the target speed of 37 mph in that second. ............................................ 205 39: Average speed of the ACT trained and placebo trained drivers during each second when approaching a busy four-way intersection. (* The Asterisks - suggests that ACT trained drivers were significantly more likely to slow to the target speed of zero mph in that period.) .... 207 40: Cumulative percentage of ACT trained and placebo trained drivers that made a glance toward the near extent in each second when approaching a bus stopped and partially blocking the view of a crosswalk. *The Asterisks - indicates that the ACT trained drivers made significantly more glances than did the placebo trained drivers in that second. ........................................................................... 216 41: Cumulative percentage of ACT trained and placebo trained drivers that glanced toward both a roadside pedestrian to the left and perpendicular truck to the right in each second when approaching the truck. ............................................................................................... 218 42: Average speed of ACT trained and placebo trained drivers when approaching a bus stopped in front of a mid-block crosswalk.............. 220 43: Average speed of ACT trained and placebo trained drivers when approaching a roadside pedestrian to the left and truck to the right. .... 221 44: The highlighted regions represent the times where ACT trained drivers outperformed the placebo trained drivers at the three locations where drivers crashed. .......................................................................... 232
xxxii
45. Glancing, slowing, and average lane position of experienced and novice drivers when approaching a sharp curve to the right. [Red Lines: Proportion of drivers who glance to the far extent right. Black Lines: Proportion of drivers who slowed to the target speed of 32 km/h (20 mph). Blue Lines: Average lane position of drivers when approaching the sharp curve to the right (negative is left). Each time represents the time before the curve.] ........................................... 239 46. An example of delayed apexing versus a sweeping turn that starts much earlier and how each turn tactic remains within the zone where it is most likely to be in conflict with other vehicles or pedestrians. (Yellow highlight is the conflict area; green highlight is the safer region). .................................................................................................. 247
xxxiii
CHAPTER 1 HAZARD MITIGATION BEHAVIORS: DIFFERENCES BETWEEN EXPERIENCED AND NOVICE DRIVERS AND TRAINING PROGRAMS TO DECREASE THE DIFFERENCES
1.1 Introduction The transition from being a teen novice driver to a skilled, experienced driver involves a process that may take months or years. Novice drivers are generally defined as drivers in their teens with a solo license (either restricted or unrestricted) for six months or less, though this definition varied from study to study. The greater death rate for novice drivers during the first six months of solo licensure shows that the transition to being an experienced driver is costly in terms of lives lost. In an attempt to shorten the painful transition from novice to experienced driver, knowledge of how a safe and experienced driver behaves differently from a novice driver can potentially be of use. Once we gain knowledge of how safe and experienced drivers behave when compared with a novice driver, the novice driver can potentially be trained to act like experienced drivers. A program of research has been successful with hazard anticipation (Pollatsek, et al., 2006) and attention maintenance (Chan et al., 2010; Fisher et al., 2010; Pradhan et al., 2011). The current research will attempt to gather the information to create and test the effective hazard mitigation-training program.
1
1.2 Crash Statistics Show Novice are Overrepresented For drivers ages 15 through 24 years, traffic crashes account for more deaths than suicide, homicide and other accidents combined (OECD, 2006). The death rate has been worse for newly licensed drivers ages 16 and 17. These young novice drivers are involved in more traffic crashes than the number of experienced drivers in Nova Scotia (Mayhew et al, 2003), Europe (Vlakveld, 2005; OECD, 2006), and the U.S. (Braitman, et al., 2008; McCartt, Shabanova and Leaf, 2003; Masten and Foss, 2010).
1.3 Crash Types for Novice Drivers A query was performed of the data collected in the National Motor Vehicle Crash Causation Survey [NMVVCS] (Winston, 2011). Specifically, the query compiled all serious (tow away) crashes involving drivers ages 16-18 in 2005 and 2006. The most frequent crash
types included turn-across-path crashes, the next most frequent crash type for this age group was road departures and rear end crashes. All drivers were out of the license permit phase. These three most common crash types for these teenage drivers account for more than 70% of the serious 677 teen-involved crashes (See Table 1). Table 1: Three most common crash types from the NMVVCS data involving 677 teens who were involved in serious crashes in 2005-2006 from NMVVCS Query.
Crash Type
Percent of Serious Crashes
Road Departures
26.4%
Turn across path
29.7%
Rear ender
15.4%
2
Mayhew et al. (2003) indicated that for 16-19 year olds single vehicle run-off-theroad (ROR) crash involvement decreased by approximately 70% after eleven to twelve months of licensure. Braitman, et al. (2008) interviewed 260 sixteen-year-old drivers who crashed and were licensed for less than eight months. Braitman, et al. reported the most common crash types were run-off-the-road crashes (30%), rear-enders (35%), and right-of-way crashes (23%). Conditional probabilities were calculated to determine how specific crash types differ with age (Abdel-Aty, M. A., Chen, C. A., Radwan, E., 1999). Younger drivers were found to be overinvolved in crashes on curves, crashes that resulted in an overturn, head-on, or ROR. Furthermore, novice drivers are overrepresented in ROR crashes on two-lane lowvolume roads, compared to more experienced drivers. (Chen, Ivers, Martiniuk, Boufous, Senserrick, Woodward, Stevenson, Williamson, Norton, 2009). The crash data overall exhibits that the greatest crash risk to novice drivers has been on rural roads with higher speed limits (Chen et al., 2009; Glennon et al, 1985). Again, the problem for novice drivers has been the high-speed roads, not a driver traveling above the speed limit. Of the 8,821 crashes on rural roads with a speed limit of 55 mph, most of these fatalities involved novice and young drivers (Glennon et al., 1985; FARS, 2009).
1.4 Causes of the Crashes: Implications of the Crash Types The conclusion from several researchers (Mayhew et al, 2003; Braitman et al, 2008; Winston, et al, 2011; California DMV, 2008) is that poor vigilance, poor anticipation, and inability to manage speed and lane position rather than risky behavior are the most likely causes of the major types of crashes among novice drivers. Poor anticipation refers to an inability to recognize potentially dangerous situations. Examples 3
of potentially dangerous situations include scenarios with a restricted sight line due to road curvatures, slippery roads, or nighttime conditions. The slippery road scenario offers an opportunity to explain the general problems encountered with novice drivers. In this dissertation, the primary focus was hazard mitigation, an act that follows hazard anticipation. While anticipation can occur without mitigation, the mitigation might be less likely without anticipation. To better explain the concepts, consider a slippery road crash. A slippery road crash is a symptom of both poor anticipation and poor speed management. If a driver fails to recognize the road as being slippery, and he or she does not slow, and the driver may find that he or she is unable to implement an effective avoidance response. A potential hazard was defined as a hazard that did not require an emergency response. An emergency situation may be averted when a driver recognizes a potential hazard, and reduces his or her speed before arriving at the slippery area (or area of potential danger). A slippery road crash may not involve high speeds, but implies driver speeds are too fast for the conditions. A speed too fast for the conditions implies the driver may not have anticipated or prepared for that situation beforehand. An open question in this research was why do novice drivers fail to mitigate hazards?
Based upon the crash literature cited above, one could form one of two
conclusions. The first possible conclusion is that teenage drivers are risk takers. While the second possible conclusion is that teen drivers recognize the hazards and fail to implement safety precautions because they are ignorant of the potential ways to mitigate the hazards. The majority of the evidence suggests that, teenage drivers, fail to recognize
4
the hazards (Pradhan et al, 2005; 2009) and then fail to mitigate them successfully (Fisher et al., 2002) The evidence against the first conclusion above -- that novice drivers are especially reckless -- came from several sources. As an example, consider how likely alcohol is associated with novice driver crashes. In California, alcohol was identified as a cause of the crash in 5.1% of the fatal crashes involving 16 to 19 year-olds. However, alcohol was involved in only 1.8% of all crashes involving 16-year-old drivers. (http://www.dmv.ca.gov/teenweb/more_btn6/traffic/traffic.htm#). Mayhew, et al (1986) came to a similar conclusion when they found that 16 and 17 year olds had the fewest incidences of high blood alcohol concentrations and the lowest average BAC of any age group involved in crashes. Risky behaviors such as driving intoxicated were less prevalent among 16-year-old drivers than among other age groups. The above suggests that the high crash rate of novice drivers was not due to a willingness to engage in especially risky behaviors if those behaviors are defined by alcohol consumption or high-speed driving. More generally, as the literature review will make clear, the available evidence strongly suggests that novice drivers are not adequately prepared to mitigate the hazards because they fail to anticipate the hazards. Simply put, novice drivers are not careless, but clueless as coined by McKnight and McKnight (2003). These inadequate behaviors do not appear to be related to risk taking behaviors.
The behaviors are strongly indicative of lack of knowledge of the
prerequisites of safe driving.
5
1.5 Problem Statement It was not known which hazard mitigation skills novice drivers lack. In the first experiment, I will compare novice and experienced drivers’ hazard mitigation skills using a driving simulator. This research will build upon the research by Fisher et al. (2002). Fisher et al. examined the hazard mitigation response behavior of teenage and experienced drivers in potentially hazardous and nonhazardous situations. At the time of the research by Fisher et al., eye-tracking equipment was not readily available. In my research, I will use an eye tracker in addition to collecting vehicle and driver measures. Currently, no widely available programs are focused on the training of novice drivers’ hazard mitigation skills. In the second Experiment, I will develop and evaluate such a program. The ACT hazard mitigation program was written for a PC. The evaluation of ACT was accomplished using a driving simulator. Part of the problem may be that younger drivers may not be aware of their inadequate skills and knowledge. Training specially designed to raise younger drivers’ awareness of expected problems and provide response strategies may decrease the crash rate for this age group.
The training should target proper scanning strategies and
response tactics to neutralize hazards. 1.5.1 Experiment 1: Understanding the Shortcomings of Novice Driver Hazard Mitigation Strategies
In the first part of this research, Experiment 1, novice and experienced drivers will navigate through traffic scenarios associated with crash risk. When driving through
6
these scenarios, hazards may or may not materialize. The goal was to find the risk mitigation tactics used by experienced drivers that are not used by novice drivers. Crash risk mitigation entails the ability to navigate, guide and control the vehicle (Alexander and Lunenfeld, 1975). Glance location and speed choices were recorded during this research. Specifically, I wanted to know whether experienced drivers slow more or less often than do novice drivers and under which circumstances. In more complex scenarios, experienced drivers may have learned to compartmentalize and mitigate the complexity of the scenes they face. It seems unlikely that a novice driver would be able to perform well as the difficulty of the scenario escalates. 1.5.2 Experiment 2: Developing a Hazard Mitigation Training Program The second goal of this research was to develop and evaluate on a driving simulator and in the field, the third component of a novice driver-training program. Fisher et al. (2008) suggested that there are three components necessary for an effective novice driver-training curriculum: attention maintenance, hazard anticipation and lastly, speed and lane position management. Two of the training components have been developed and tested; these include FOCAL and RAPT-3, both of which will be explained in detail later in this research. FOCAL addressed attention maintenance and RAPT-3 addressed hazard anticipation. The current research will attempt to develop the third component of the curriculum and will be referred to as ACT, which stands for anticipation, control and terminate. Hazard mitigation training cannot easily be separated from hazard anticipation training (since learning to mitigate a hazard without anticipating it makes little sense. Hazard mitigation training can easily be separated from attention maintenance training 7
(since the driver should stop all glances inside the cabin of the automobile) (Fisher et al., 2010). An important prerequisite to a crash risk mitigation response by a driver is the driver’s ability to anticipate and look for hazards. Drivers might not properly reduce speed unless they perceive a change in the driving environment. Before isolating the problem as being speed or lane position related, there must be proof that novice drivers are able to anticipate hazards. The development of the ACT program was based in-part upon the results from part one of the research. High-risk crash types for novice drivers were identified from crash data. These high-risk crash scenarios were reproduced in a virtual world. Novice and experienced drivers were compared. The incidences where novice behaviors deviate from those of the exemplary experienced drivers were identified. These behaviors were addressed in the ACT program and the ACT program was utilized in part two of the experiments. The ultimate goal was to develop an effective training program that teaches novice drivers to mitigate hazards like safe drivers. The construct here was the definition of a good driver. Consequently, rather than randomly recruiting experienced drivers, attempts were made to recruit exemplary experienced drivers. The training program involves testing, as such; any test or assessment should attempt to maintain construct validity (Walsh and Betz, 2000). Construct validity means that the test measures what it is intended to measure. Here, the goal is to measure novice drivers against the performance of drivers who we know have exemplary safety records. In summary, research in Experiment 2 attempted to determine if a cause of crashes was the failure of novice drivers to respond appropriately to situations associated
8
with crash risk. If this was the case, then novice drivers trained to mitigate crashes should have fewer crashes.
9
CHAPTER 2 LITERATURE REVIEW Hazard mitigation is especially difficult to study in novice drivers because so much of hazard mitigation depends on anticipating a hazard and so few novice drivers, comparatively speaking, actively anticipate. Below, the literature on hazard anticipation was reviewed, with an eye towards the differences between novice and experienced drivers. Additional detail was provided about hazard anticipation training programs. This was followed by a discussion of what was known about the differences between the hazard mitigation skills of novice and experienced drivers and of programs to train these skills.
2.1 Hazard Anticipation Behaviors In 2005, a simulator experiment examined whether there were differences among older drivers’ (24 drivers aged 60-75), younger drivers’ (24 drivers aged 18-26) and learner’s permit drivers’ (24 drivers aged 16-17) abilities to attend to (scan) the visual field selectively for information, that could potentially reduce their risks (Pradhan, Hammel, DeRamus, Pollatsek, Noyce, and Fisher, 2005). Eye movements were utilized as the prerequisite to safe driving behavior, with speed choice being used as secondary measure of safe driving behavior, and also lane position. For example, if a driver fails to glance toward a potential hazard, and subsequently slows, the slowing may not be associated with the potential hazard but may be by mere chance. Drivers may use
10
ambient vision to monitor their environments (and that was considered in the results). However, it was much more likely that a driver who looks toward a hazard and immediately slows was likely responding to that hazard. Pradhan et al. wanted to know whether novice drivers recognize the risks less often than do young or experienced drivers. After exposure to 14 potentially hazardous traffic scenarios, the novice drivers engaged in behaviors indicative of their recognition of the potential for risk 35.1% of the time, the younger drivers engaged in such behaviors 50.3% of the time, and the older drivers, 66.2% of the time. (See Table 2). Table 2: Percentage of trials on which risky features of scenarios was fixated
Generally, there was remarkable improvement from novice to young drivers. (See Table 2). I noticed that if I eliminated two scenarios from Pradhan et al (scenarios 4 and 10) and categorize the scenarios by obstruction type we see a significant improvement in performances for scenarios were there was an obstruction and a large improvement for scenarios with anticipated movements (Table 3). The two anomalous scenarios (scenarios 4 and 10) had a trait in common in that neither scenario leads directly to a potential for a crash. The first scenario
11
(scenario 4) measured whether a driver glanced at a sign that indicated a signal ahead. For this to be a hazard, the driver would also have to neglect the signal. In essence, it was a warning of a warning. The second anomalous scenario (scenario 10) measured whether the drivers fixated on a pedestrian on the west (left) side of the road that may or may not make a north to south crossing on the side road. A lead vehicle signals a left turn but made a sudden slowing, presumably yielding to the pedestrian, before turning left into the side road. Here, a reason experienced drivers would not perform better than younger drivers might be that the hazard was the lead vehicle not the pedestrian. Hence, if the driver simply maintains a reasonable following distance, no crash occurs where there is a glance toward the pedestrian or not. Experienced drivers likely focus more on the lead vehicle and possibly slow more than a novice driver even with fewer relative glances toward the pedestrian (but recall that experienced drivers still made more glances toward the pedestrian).
12
Table 3. The results from Pradhan et al (2005) categorized by obstruction type
Scenario Type
Novice
Young
Older
26.5%
34.0%
51.0%
33.8%
49.2%
72.0%
48.5%
66.8%
Environmental Obstructions (2) (S4 excluded) Traffic Obstructions (6) (S10 excluded)
Anticipated Movements with No Obstruction (4) 30.0%
In summary, drivers with only a learner’s permit, both as a group and individually, were worse at anticipating hazards than younger non-novice drivers and considerably worse than much more experienced drivers. The absolute differences were often impressive, upwards of 56 percentage points in one scenario (S9, Table 2). And when compared relatively, experienced drivers were up to six times more likely to look for information in a scenario, which could reduce their likelihood of a crash (S13). Also, at locations that might lead to greater crash risk, experienced drivers made significantly more glances than did novice drivers, yet when there was a reduced threat or the object was not directly related to a crash threat, glances between the groups became more similar. 2.1.1 Knowledge or Workload Handling Deficits? There are three common reasons why drivers fail to discern hazards: (1) the hazard was below the threshold for discernment; (2) the driver did not know to look for 13
the hazard, and (3) the driver may be devoting the necessary mental resources toward other objects or tasks, such as was the case with inattentional blindness (Simons, 1999). Novice drivers were overrepresented in nighttime crashes despite their ability to see more information at night than their older, more experienced counterparts. Additionally, novice drivers were overrepresented in run-off-the-road crashes and generally, lane keeping was not affected at night (Liebowitz, et al., 1991; 1998). This leaves as explanations of novice driver higher involvement in such crashes as either a lack of knowledge of what to look for, or an inability to handle the vehicle and to scan the road ahead for hazards. Garay-Vega and Fisher (2005) evaluated the hypothesis that novice drivers failed to recognize a threat was unfolding because the vehicle handling task was capturing their attention. When a risk was foreshadowed several seconds before the novice driver encounters a scenario, the foreshadowing element should break the set and alert the novice driver to the risk and cause them to glance in the appropriate location at the appropriate time. Apparently, this was not true for many of the novice drivers. As an example, a truck was parked along the roadside immediately before a crosswalk. A pedestrian appears on the sidewalk, behind the truck (the foreshadowing element) when the driver was three seconds from the crosswalk. As the driver approaches, the pedestrian disappears behind the truck. If a driver saw the pedestrian (the foreshadowing element), the driver should then be alerted to the danger presented in the scenario by a hidden pedestrian and be much more likely to glance to the right as he or she passed in front of the truck.
14
On the basis of the results from Garay-Vega et al., 2005), vehicle handling deficits alone cannot explain the differences between novice and experienced drivers. If cued to a potential hazard, such as a pedestrian on the sidewalk near a crosswalk, someone might expect drivers who anticipate the hazard to look for that pedestrian if the truck should obscure it. Of all those who fixate upon the pedestrian when on the sidewalk, 85.0% of the experienced drivers scanned toward the crosswalk in front of the truck, compared to only 47.5% of the novice drivers. Of those drivers who did not fixate upon the pedestrian on the sidewalk, 60.7% of the experienced drivers scanned the critical area in front of the truck compared to 33.7% of the novice drivers. More than half the novice drivers failed to recognize that the situation was a hazardous one. Take note that this study did not answer the question of whether novice drivers are more engaged in the driving task. However, it did indicate that even when novice drivers see elements that define a potential hazard, they do not understand that in order to drive safely they need to scan the scenario for information, which could reduce their likelihood of a crash. 2.1.2 Hazard Anticipation Training: Simulator Evaluation Knowing whether there are differences in the hazard mitigation skills of novice and experienced drivers was not the only goal of this research, but also knowing whether the differences that do exist can be reduced. As discussed above, there are very few hazard mitigation-training programs. However, the situation was considerably more robust for hazard anticipation training programs. The work by Fisher et al. (2002) compared the vehicle behaviors of novice (learner’s permit) drivers trained to anticipate hazards with untrained novice drivers and 15
more experienced drivers. They found statistically significant differences in the vehicle behaviors of the two groups of drivers. But it was not possible to determine whether the difference in the vehicle behaviors of experienced and novice drivers was due to better hazard anticipation skills of experienced drivers, better hazard mitigation skills, or some combination of both. Eye movements are a direct way of evaluating the effect of training on hazard anticipation skills. Using a standalone PC-based training program, RAPT (Risk Awareness and Perception Training), novice drivers were trained to recognize situations where there was information that could reduce their crash risk and to visually attend to that information (Fisher, Pollatsek and Pradhan, 2006). In the pretest phase, participants were asked to note areas of the roadway that should be continually monitored and, more importantly, areas that could contain information hidden from view which was relevant to making a response that could reduce the driver’s risks. After responding to a pretest scenario, the training phase for that scenario began. The trainees were encouraged to identify where to look for information that could reduce their risks. If the threat was caused by risks which were potentially hidden from view, coaching was offered with the use of diagrams to indicate what would be hidden from view and plausible risks such as vehicles obscured by bushes, or pedestrians obscured by vehicles were pointed out. One of the first tests of the PC-based training program (Pollatsek, Narayanaan, Pradhan, and Fisher, 2006 involved novice (16-17 year old) drivers that were divided into two groups, trained and control. When the effects of the PC training were evaluated, the trained participants were 80% better at identifying areas they should continuously 16
monitor and were 280% better identifying areas of potential hazards. In subsequent simulator research, it was shown that the training generalized on the simulator not only to scenarios which had been trained on the PC (near transfer scenarios) but also to scenarios which were unlike those trained on the PC but still required the exercise of hazard anticipation skills (far transfer scenarios). The improvement in far transfer scenarios proved that drivers were learning and extrapolating what they learned to other scenarios. The training effect was significant and about the same in the far-transfer (20 percentage point improvement) and near-transfer scenarios (25 percentage point improvement). In summary, it appears that when novice drivers are given knowledge relevant to the judgment of where and when to scan selectively for information that could reduce their likelihood of a crash, they are able to use this information to their advantage on a driving simulator. 2.1.3 Hazard Anticipation Training: Evaluation in the Field In Experiment 2 I will focus on the question of whether a PC-based training program can improve novice drivers’ ability both to anticipate and mitigate hazards on a driving simulator. One can ask whether there is reason to believe that the effects of such a training program could generalize to the field. In a field studies, Pradhan et al. (2006; also, Taylor et al., 2011) compared untrained drivers and drivers who completed RAPT-3 PC-based training program. The results from Pradhan et al. are encouraging in that trained drivers made glances into areas that contained potentially relevant threat information 64.4% of the time, while the untrained drivers looked into the same areas only 37.4% of the time. Similarly favorable results were shown in far-transfer scenarios (scenarios on which the drivers were not trained). The results by Pradhan et al. (2006 and 17
Taylor et al., 2011) buoy the suggestion that the anticipation effect of ACT training could be as successful in the field. If the mitigation effect were successful on the driving simulator, then one would hope that it would translate to the field as well. Of course, further experimentation would be required to show that this was the case.
2.2 Hazard Mitigation and Hazard Anticipation Behaviors The literature mentioned earlier in this paper informs us that many novice drivers fail to anticipate hazards. The literature reveals that drivers’ abilities to anticipate hazards improve with time and training. Without training, it may take years to improve the level of a novice driver to the level of an experienced driver. One might ask at this point one of two questions: (1) once novice drivers begin to anticipate hazards, do they institute a risk mitigating response? And (2) was their failure to institute a risk mitigating response associated with greater risk taking and sensation seeking or due to a lack of knowledge regarding the proper response? When driving, there are three common road geometries. A driver could be traveling a straight road, also referred to as a straight road segment, they might be traveling through an intersection, or they might be negotiating a curve. Although there are both vertical and horizontal curves, after a literature review, the most significant crash issues have been with horizontal curve (McElheny, Blanco and Hankey, 2006). The three parts of this section are devoted to studies of drivers’ speed choice, and glancing behaviors in horizontal curves, straight road segments, and intersections. The reason for choosing these three types of scenarios was that they represent the crash scenarios that teen drivers are most involved (Winston, 2011): right-of-way violations (intersections), run-off-the-road crashes (horizontal curves), and rear-enders (straight road segments). 18
The current research will focus on the two most common crash types, single-vehicle ROR and left-turn-across-path [LTAP] crashes. The studies in this section are grouped as studies of hazard mitigation and anticipation since the two are inextricably linked. For my purposes, hazard mitigation will refer to vehicle or driver behaviors exclusive of eye movements. Hazard anticipation and monitoring will refer solely to driver eye behaviors. Hazard anticipation occurs at that point in time when a driver realized that a threat or potential threat could materialize or was actually in progress. Hazard monitoring occurs once a driver anticipated a hazard and decided that an area of the environment needs to be resampled at various closely spaced intervals. Note that hazard monitoring was typically part of the hazard mitigation process, but was not considered an actual behavior that mitigated a threat. Instead, it was a behavior that guided an action, which controls the vehicle and thereby mitigated the threat. Studies discussing the differences between novice and experienced drivers’ hazard mitigation behaviors in specific scenarios are discussed. 2.2.1 Horizontal Curves Successful curve negotiation depends upon the choice of appropriate approach speed and adequate lateral positioning through the curve. Loss-of-control crashes result from an inability to maintain lateral position through the curve because of excessive speed, excessive steering, and inadequate deceleration in the approach zone, or a combination of these factors. Based upon the NMVVCS crash data (Winston, 2011), of the 677 serious crashes involving teenage drivers, 32% involved a driver leaving his or her lane and 19% (almost one of every five) involved a driver leaving the road.
19
To corroborate the findings from the query performed by Winston (2011), I performed a query of the Fatal Accident Reporting System [FARS] for all 2009 crashes. I compared only 16year-old drivers with 35-year-old drivers. Thirty-five year old drivers were more likely to be experienced and less likely to suffer from age-related declines. The 16-year-olds are much more likely to be novice drivers with less than eight months of driving experience. Each age was at the bottom of an age range reported by NHTSA. NHTSA reported on their web page (2011: http://www.fhwa.dot.gov/ohim/onh00/bar8.htm) that 16 to 19 year olds travel an average of 7,624 miles each year. Drivers age 35 travel an average of 15,291 miles each year. From the comparison, we can see distinct differences in the hazard mitigation responses by the groups, as well as the crash types. For all crash situations, 8.1 of every one thousand 16 year olds drivers die in a crash compared to only 4.9 in a thousand for 35 year olds. But for run-off-the road crashes, the crash rate is much worse. Indeed, 16-year-old drivers were more than three times more likely to crash off the road and to be in a rollover crash than were drivers age 35 (See Table 4).
Table 4: Crashes per 10,000 miles traveled based upon the FARS 2009 crash history database and NHTSA yearly mileage data for various age groups. MV refers to motor vehicles. http://wwwfars.nhtsa.dot.gov/QueryTool/QuerySection/SelectFields.aspx
Age 16 Crash Avoidance Maneuver
Steering Other Avoidance Maneuver Not Reported Total Age 35 Steering Other Avoidance Maneuver Not Reported
Most Harmful Event Off Rollover Road 111 56 0 0 100 104 211 160
38 1 31
9 0 39
20
MV on Roadway 59 1 241 301
TOTAL 226 1 445 672
36 3 127
83 4 197
Total
70
48
166
284
These crash statistics offer evidence that novice drivers were six times more likely to steer before leaving the road, presumably over-steering late in an event which will result in loss of control (rollovers) and traveling off the road. This data might lead one to believe that 16 year old drivers were less likely to keep their eyes on the road (more failed to respond at all), less likely to anticipate hazards such as curves (more crashes off the road), and less likely to respond safely (more steering responses before leaving the road) rather than braking earlier in the event. While all ROR crashes do not occur on curves, curves are repeatedly linked to ROR events while ROR crashes are likely the result of other factors such as following too close or late decisions to turn on straight road segments (McLaughlin, Hankey, Klauer and Dingus, 2009). Roads with a curvature with radii less than 400 m (1,300 ft.) have a particularly high crash rate (McLean et al, 1981) and the crash rate increased as the radius of the road curvature decreased (Choueiri and Lamm, 1987). Glennon, Neuman, and Leisch (1985) found that drivers tend to overshoot the curve radius, or steer too wide around the turn. When a driver steers too wide, he or she must re-correct back to his or her lane, which required a driver to negotiate a sharper turn than the road dictates. The tendency to overshoot was independent of speed. The authors observed that the straight segment immediately in advance of the curve was the critical region of operations, because at about 61 m (200 ft.) before the beginning points of the curve (or approximately 3 s driving time); drivers begin to adjust both their speed and path. Such adjustments are particularly large on sharper curves.
21
Zegeer, Stewart, Reinfurt, Council, Neuman, Hamilton, Miller, and Hunter (1990) determined the first maneuver made by drivers at crashes that occurred on two-lane horizontal curved roads. In 77% of fatal crashes, the driver’s first maneuver was toward the outside of the curve (those who brake hard usually go straight). In a study focused on 600 crash sites (and 600 comparison sites) involving fixed objects, crash locations were best discriminated from comparison locations by a combination of a curvature greater than nine degrees and a downhill gradient steeper than three percent; and, for the fatal fixed-object crash population, the crash locations were best discriminated from comparison locations by a combination of curvature greater than 6 degrees and downhill gradient steeper than 2 percent (Wright and Robertson, 1979). Glennon et al (1985) offered a useful multiple linear regression equation that may be used to evaluate the crash risk at a horizontal road curvature. They showed that horizontal road curvatures with a high discrimination score D were associated with greater crash risk and roads with a low discrimination score were associated with low crash risk. Glennon et al. developed an equation for D that was based upon the degree of curvature, the length of the curve, and the shoulder width (Equation 1): -
(1)
Where DC was the curvature of the road in degrees, LC was the length of the curvature in miles, and SW was the shoulder width in feet. This equation identified 76% of the high crash locations (Glennon, et al). The discrimination score can be used to determine the probability that the site was a high crash location by utilizing a table in his publication. I have taken the entries from the table and found that the probability of a scene being a crash location may be calculated using Equation 2: 22
Pcrash 0.004D3 0.003D2 0.191D 0.553
(2)
Anderson and Krammes (2000) took the next step; they developed two models that estimate the odds ratio of crash risk for a site. The first model was based upon the speed reduction of the 85th percentile driver from the straight road segment, to the apex of the curve. The second model estimated the crash risk due to the degree or curvature. After an examination of 563 curves, Anderson and Krammes (2000) found a relationship between speed reduction (∆V85), the degree of curvature (Deg.) and the crash rate (CR) at curves.
CR 0.54 0.27V85
(3) CR 0.18 0.23Degrees
(4)
Therefore, crash risk (CR) was influenced by the curvature of the road (Degrees) and the speed loss (∆V85) exhibited by the 85th percentile drivers when negotiating the curve. According to the NMVCCS (National Motor Vehicle Crash Causation Survey) data (Winston, 2011), in 2005-2006, 677 teens were involved in serious crashes, 60 were single vehicle off the road right, which was the second most frequent crash type. The fourth and seventh most frequent crash types were also single vehicle off the road crashes. 53 were single vehicle off the road right due to loss of control (4th most frequent crash type) and 36 were single vehicle off the road left (7th most frequent crash type). (See Figure 1).
23
Figure 1: NMVCCS Crash configurations for the 2nd most teenage serious crashes (2) and the 7th most teenage serious crashes (1) and the 4th most teenage serious crashes (7).
2.2.1.1 Hazard Anticipation (Glance Behaviors) at Horizontal Curves Hazard mitigation behaviors are possible only if the driver correctly anticipated the curvature of the roadway. Land and Lee (1994) suggested that drivers’ glances might be used to forecast the curvature in the road ahead. To illustrate what Land and Lee meant, note the crosshairs in the figure below. The crosshairs indicate where this experienced (28 year old) driver was looking. Note how the glance cuts across the inside of the turn and, as predicted by Land and Lee, forecasts the curvature of the road. On the same curve, Pradhan et al. (2005) collected eye glance behaviors of 18 and 19 year old drivers. While that age group was slightly more mature relative to driving skills, many drivers still exhibited poor glance techniques. A case in point was the glance behavior of a novice driver depicted in Figure 2. Note the crosshairs are much closer to the front of the vehicle and do not forecast the curvature of the road ahead.
24
Figure 2: Eye glance of a young driver. Left: licensed for two months Right: Licensed for ten years.
The tendency for experienced drivers to make anticipatory glances across the inside of the curve was not uncommon. Suh, Jin Park, Ho Park, and Chon (2006) came to a similar finding. Suh et al. showed driver glances on a straight road segment and a curve to the right. On the straight road, the glances were near but slightly above the horizontal plane (neither too far up nor down) and slightly more left than right (recall they drive on the left). Yet when negotiating a curve to the right, most glances were above the horizontal plane and between five and ten degrees to the right. Also note that most glances are above the horizontal plane. The glance distribution shown by Suh et al was consistent with the glances from the 28-year-old driver depicted in Figure 2. Conversely, the 18-year-old driver’s glance location shown in Figure 2 was inconsistent with the findings by Suh et al, (2006) and Land and Lee (1994). The available research evaluating drivers’ anticipatory glances on curves suggests that most presumably experienced drivers will make a glance at the far extent of the inside of a curve. Such a glance offers them the greatest sightline and time to mitigate the potential hazard created by curves. Potential hazards of curves include greater lateral forces, reduced sightline and more difficult lane keeping by the driver and by oncoming 25
traffic. Maintaining a safe speed will reduce lateral forces and leave resources available to a driver who might need to respond to an oncoming vehicle or unforeseen hazard. Maintaining a proper lane position will extend the sightline and may also reduce the lateral forces on the vehicle. 2.2.1.2 Hazard Mitigation: Speed Choice when Approaching or Negotiating a Horizontal Curve AASHTO (2001) defined the effective speed of a road based upon the radius of the curve. The effective speed was the maximum safe speed for the radius, friction, and superelevation of the road. Superelevation is banking the roadway to help offset tangential pulling that a vehicle experiences as it negotiated the radius of a curve. The effective speed was defined below as a function of the superelevation, radius, and road friction:
VEff .(inmph ) 15r ( e)
(5)
The parameter e was set to the superelevation, r was the radius, and μ was side friction. Side friction was also known as lateral friction. The quantity V (Eff) was the maximum speed for the radius [r] and superelevation [e] of a given curve such that the resultant side friction factor (µ) did not exceed the maximum side friction factor for that curve. AASHTO simplified the equation. Daily, Shigemura and Daily (2006) made the assumption that drivers will negotiate a curve by traveling along the centerline of the road. It was likely that experienced drivers will flatten (increase) the radius of a curve by optimizing lane position (Bonneson and Pratt, 2009; Emmerson, 1969). For instance, when negotiating a right curve, if a driver started out near the centerline, moved toward
26
the fog line and ends near the center line, the vehicle will travel a path with a greater radius than that if measured at the center of the lane. Bonneson and Pratt (2009) developed a model that predicts average free-flow speed in a curve based upon the speed of traffic on the straight segment (well before the curve), the superelevation (banking), and curve radius. Bonneson and Pratt developed an 85th percentile speed model and found that the average speed was 90% of the 85th percentile speed. The reader may recognize the Bonneson and Pratt equation as being similar to the equation by AASHTO.
Vcurve.avg. (
15Rp (0.112 0.00066Vt .avg 0.000091Vt 2.avg 0.0108trk e / 100 1 0.00136Rp
)0.5
(6)
Where Vcurve.avg was the average speed of traffic in mph, Rp was the radius of the travel path, Vt.avg was the velocity on the straight (straight road) segment, trk was an indicator variable if estimating truck speed and e was the superelevation (banking of the road) as a percent. When analyzing the curve speed of a truck, the V term should be the average speed of trucks on the straight. If the average speed of trucks is not known, the user may enter 97% of the average speed of cars. This equation accounted for 93% of the variance. The radius of the curvature has been cited in research as a significant influence upon vehicle speed and crash rate (Lamm & Choueiri, 1987; Fitzpatrick, Krammes, and Fambro, 1997). Fitzpatrick et al. (1999) established that daytime vehicle speeds on twolane non-sharp curves (radius greater than 450 m) would not differ from the speed on straight sections of the same highway. Yet, when the radius of the curve was 200 m (650 feet) eighty-five percent of drivers traveled slower than 42 mph (67 km/h) and the eighty-
27
fifth percentile speed increased to 53 mph (85 km/h) as the radii of the road curvatures increased to 3280 feet (1000 m). (Fitzpatrick et al., 1997). According to Hulse, et al. (1989) and Cullinane and Green (2006), the inverse of the radius and the mathematical log of the sight distance (in meters) are better indicators of a drivers’ behavior in a curve. Cullinane and Green (2006) measured the throttle release time to a lead vehicle that started 118 ft. (36 m; 1.8 seconds) ahead and slowed at 0.2 Gs for four to five seconds from a speed of 45 mph (72.4 km/h). A vehicle slowing at this rate from this distance and time will not lead to a crash. The authors indicated "Curve radius had very little effect on response time to a lead vehicle braking with the mean response times for the 200 m and 400 m..." When the curve radius was 100 m throttle release time increased by a factor of 5.4 times the inverse of the radius of the curve. They also found that response time decreased by 1.3 times the Log of the sight line in meters. When the radius of the curve was smaller (sharper), speeds decreased and the start of slowing and braking began earlier (Mikolajetz, et al., 2009). Mikolajetz et al. offered two models, one that predicts where the average driver began to slow in response to a curve of less than 24 ft. (80 m) radius, and the other that predicts where braking began. From the results of this study, the minimum speed may be modeled based upon whether the curve was to the passenger’s side or driver’s side, the length of the curve and the radius of the curve:
Dslow 3.75(Vt Vc ) 37.79
(7)
Dbrake 2.21(Vt Vc ) 19.73
(8)
28
Min.Speed km / h 0.08dr.side 0.411R 0.107L 27377
(9)
Vt refers to the straight (straight segment) road speed and Vc was the speed in the curve in km/h. Dslow and Dbrake was the distance in meters that a driver begins to slow and brake before reaching the start of a curve. Min Speed was the minimum speed in the curve, dr.side refers to curves to the left in the US and was coded as a one if true, zero otherwise. R was the radius of the curve in meters and L was the length of the curve in meters. The Mikolajetz models show that we can compare novice and experienced driver behavior by comparing how soon before a curve we might expect slowing and braking to occur. This type of model will likely be useful in that prior studies (Fisher et al., 2002; McElheny, et al., 2007) showed that experienced drivers began to slow earlier before a curve than did young or novice drivers. In summary, the speed in a curve was based upon the sight line, the radius of the curve, the banking, angle of the curve, and length of the curve.
The distance where a
driver began to respond to a curve will be dependent upon the speed loss that was typical of that curve. For each of the factors associated with speed loss, novice drivers are expected to respond later (slower) and slow less early (from farther away). These models offer a baseline from which we can evaluate the experienced and novice drivers. The Bonneson model was used as a baseline for the expected speed loss in each curve and the Mikolajetz model was used to compare where pre-curve slowing and braking began. Essentially, crash risk increased as the difference between the negotiation speed and the safe curve (design) speed increased (Tate and Turner, 2007; Anderson and Krammes, 2000).
29
2.2.1.3 Hazard Mitigation: Lane Position when Approaching or Negotiating a Horizontal Curve Naturalistic observation research (Bonneson & Pratt, 2009; Emmerson, 1969; Zegeer, et al., 1990) shows that at many curves, drivers tend to move laterally inward while cornering. This inward lateral movement enlarged the radius of the curve. Emmerson reported the average lateral shift to be approximately 3.0 ft. (0.9 m). This shift inward on the curve has significant implications. A driver who starts near the outside of the curve and moved to the inside of the curve will maintain a greater sightline and more control of the vehicle at the same speed. Drivers are able to maintain more control with an inward shift and reduce the lateral acceleration on the vehicle. The lateral acceleration was usually greater in curves with smaller radii (tighter turns) (van Winsum, 1996).
Steering errors increased linearly
with required steering wheel angle (Glennon et al., 1985). On roads with smaller radii the steering angle must be larger, and the likelihood of an under or over-steer is greater (Glennon et al.). Steering angle was also affected by driver competence, with experienced drivers making fewer errors and demonstrating less variation in steering wheel amplitude than inexperienced drivers (van Winsum, 1996). Overall, the research tells us that inexperienced drivers were more likely to make steering errors than experienced drivers (Maeda, Irie, Hidaka, Nishimura, 1977). These steering errors will likely be associated with greater lateral accelerations on the vehicle. Rather than uniformly moving to the inside of the curve as do most drivers, inexperienced drivers were more likely to position themselves randomly within the lane.
30
When there are tighter curves and when steering variability increases, inexperienced drivers are more likely to fail to maintain control of the vehicle. Conversely, those who moved toward the outside of the curve were associated with more crashes. 2.2.2 Intersections: Vehicular Obstructions and Intrusions According to the NMVCCS data (2008), in 2005-2006, 677 teens were involved in serious crashes; 67 were left turn across the path of oncoming traffic (scenario 68), the most of any crash type. Forty-eight of the crashes involved teenage drivers crashing into vehicles that were turning left across their path, the 6th most of any crash configuration (scenario 69) and accounting for 7% of all serious crashes in the NMVVCS data. (See Figure 3).
Figure 3: Of all crash types, the LTAP configurations represent the first (scenario 68) and 6th greatest (scenario 69) serious crash situations for novice drivers according to NMVCCS, 2008. Red represents the subject driver and blue is the threat.
31
According to the NMVCCS data (2008), teens were involved in 55 serious crashes after they turned left across the path of a vehicle on a main road, the 3rd most likely serious crash configuration among all crash configurations in the NMVVCS data (scenario 82). Also a common serious crash configuration involved teenagers crashing into drivers turning left across their path (scenario 83). There were 30 such serious teenage crashes in 2005-2006. (See Figure 4).
Figure 4: The 3rd greatest crash configuration for serious teenage crashes (82), and the 8 th greatest serious crash configuration involving teenage drivers (NMVCCS, 2008). Red represents the subject driver and blue is the threat.
Along with left turns across the path of opposing traffic, an intersection scenario that has led to 84 of the 677 NMVVCS crashes entailed a driver traveling straight through an intersection. Such an event can be dangerous to a driver who did not anticipate hazards that may not be initially visible. An example of this was a driver who believed it was safe to travel through an intersection believing he or she maintains the right-of-way. When a young driver believes that he or she is in the correct, he or she may be less inclined to believe that others may violate the right-of-way. Noted developmental psychologist Lawrence Kohlberg (1958) addressed this type of behavior as “Stage 4: Law 32
and order”. In this stage of development, young adults exhibit rule-oriented behavior, but such behavior was not likely to address exceptions to the rule. Indeed, pedestrians of ages 11 to 16 are most likely struck when in a crosswalk (Read, 1969; Agran et al., 1994; Agran, et al., 1996). The greater crash rate exhibited by teen drivers suggests that they exhibit similar behaviors when driving through crosswalks as when walking through crosswalks. This research will explore the some of the most common intersection crash configurations, particularly, a driver turning left across the path of oncoming traffic. This will include a vehicle making a left turn across the path of our driver, as well as our driver making a left turn across the path of opposing traffic. A driver who believes he has the right of way traveling straight through an intersection and responding to a situation where a side road driver may not yield might also expose novice drivers’ inabilities to anticipate and mitigate the hazard. 2.2.2.1 Hazard Anticipation: Glance Behaviors When Approaching and Within Intersections Recall that in this research, we are interested in conditional responses. Specifically, do drivers who anticipate a hazard then mitigate the hazard? A prerequisite for risk mitigation was an anticipatory glance. Pradhan et al. (2005) and Taylor et al. (2011) examined the ability of novice and experienced drivers to anticipate hazards in intersection scenarios. Pradhan et al. equated hazard anticipation with driver glances toward the potential hazard. In that study, experienced drivers were more likely to glance into an area of potential hazard in each of the two intersection scenarios. In particular, Pradhan et al. observed the glance and 33
stopping behaviors of drivers at two intersection configurations. At one stop signcontrolled intersection there was a pedestrian crosswalk beyond the stop line (Scenario six). A hedge to the right hid the sidewalk to the right, which led to the crosswalk. The researchers recorded driver glance and stopping behavior. Specifically, three measures were collected: (1) whether the driver glanced toward the hidden sidewalk, presumably for a pedestrian or bicyclist that might emerge from the obstructed view to the right, (2) if the driver came to a complete stop before the crosswalk, and (3) if the driver made a secondary glance toward the areas of conflicting traffic to the left as he or she started forward into the intersection. The second intersection configuration involved the driver making a left turn across a conflicting lane of through traffic at a site involving a limited sight line ahead (Scenario nine). Hypothetically, a vehicle traveling at a high speed could emerge from the limited sight line ahead and present the turning driver with a dangerous situation. If a driver failed to make a neutralizing secondary glance toward the area of conflicting traffic, opposing traffic had additional time to approach and strike the left turning vehicle. Every second a driver glanced toward his or her intended turning destination (typically to the left) without checking toward the short sight line ahead, gave opposing traffic an additional second to encroach and possibly crash with the turning driver’s vehicle. A glance immediately before or immediately after the start of a turn reduced the potential that an oncoming vehicle will sneak up and strike a turning vehicle without the turning driver detecting it. In the research by Pradhan et al, (2005) at the through intersection with the hidden sidewalk, novice drivers made anticipatory glances toward the area of greatest threat only 34
5% of the time. On the other hand, experienced drivers made anticipatory glances in 29% of instances. When making a left turn across the path of potential oncoming traffic at a short sightline, novice drivers made secondary glances during 14% of the turns, while experienced drivers made secondary glances in 70% of the turns. In both instances, the percentages of glances by experienced drivers were better by an order of five or more. 2.2.2.2 Hazard Mitigation: Speed Choice and Positioning When Approaching and Within Intersections Speed factors at intersections differ from free-flow speed choice. Speed within an intersection is more a factor of jockeying the vehicle so as to create the greatest sightlines and buffer space, both ahead of the vehicle and to the sides. Thus, both speed and lane positions are intertwined when evaluating a driver’s behavior within an intersection. When traveling through an intersection that might include obstructed views of traffic within an intersection, proper anticipation of hazards required glances toward all directions from which traffic might emerge. The time necessary for a glance left to a glance right (or vice-versa) took one second in light traffic and longer in heavy traffic (Robinson, Thurston and Clark, 1972). If approaching a four-way intersection, there is a need to make at least three glances. If each needs one second, we can see a need to spend more time glancing before turning. The only way to increase the time to search each direction in a busy intersection and to assure that traffic in all directions has been neutralized with a glance might be to stop before turning. When making a left turn across the path of opposing traffic, the driver of a turning vehicle was likely to make the decision (not) to encroach on the through vehicle approximately 8.9 m (29 ft.) from the centerline of the side road if the speed limit was 40 35
km/h (25 mph). If the speed limit on the road were 50 km/h, the decision position started 8 to 12.5 m (26 to 41 ft.) from the centerline of the side road (Smith, Thome, Blåberg, and Bärgman, 2009). This research might be applied to my research as a means of evaluating the safety of a turn. If a driver begins the turn considerably earlier, it would suggest that the turning driver was not traveling a safe speed leading into the turn. Specifically, an earlier turn would likely involve a greater speed, a longer turn and fewer resources available to the turning driver. The reduced resources include reduced friction, greater stopping distance, and less buffer space for traffic that may be emerging from the side road. A driver who finally stops much later in the intersection may block the intersection as a result of a failure to respond and stop earlier. 2.2.3 Straight Road Segments: When there are Hazards Near or within The Road on Straight Road Segments Straight (tangent) road segments have been cited as the location of several different crash types. Among them nearly half of the 677 serious NMVVCS crashes (332) were traveling straight and 144 of the crashes involved a vehicle decelerating or stopped. Typical straight road crash types include mid-block crosswalk crashes, work zone crashes, and rear end crashes. Several studies have shown that the probability of a rear-end crashes is greater on straight road segments rather than on a curved road (Muttart, Messerschmidt and Gillen, 2005). Straight roads offer the driver little context of the closing speed of a lead vehicle. Also, many studies have shown that most rear end crashes involve a stopped lead vehicle, more so than decelerating lead vehicles (Dingus et al, 2006; FARS, 2009; Knipling et al., 1993; McGehee et al., 1997; Muttart et al., 2003, 2005; Sorock et al., 1996). 36
Finally, on higher volume multi-lane roads, having a marked crosswalk without other substantial improvements was associated with a higher pedestrian crash rate (after controlling for other site factors) compared to an unmarked crosswalk (Zegeer, et al, 2002).
2.2.3.1 Hazard Anticipation (Glance Behaviors) when there are Hazards Near or within The Road on Straight Road Segments There are two types of visual processes utilized when driving, focal vision and ambient vision (Horrey and Wickens, 2004; Leibowitz and Owens, 1986; Resnick, 2004). Ambient vision, or guidance mode, allows us to identify where objects are and guide us through environments. Focal vision, or recognition, mode allows us to identify what we are looking at. Experienced drivers are able to maintain lane position without focal vision, while novice drivers need to make periodic glances ahead to keep the vehicle in its lane (Summala et al., 1996). Such a requirement is likely to reduce the resources available to recognize objects ahead. As suspected, Mourant and Rockwell (1972) showed that experienced drivers were more likely to make glances downstream, while novice drivers were more egocentric in their search pattern. With novice drivers making fewer glances downstream, and more glances immediately in front of their vehicles, novice drivers are more likely to miss hazards that develop downstream. . Along with reduced downstream glances, novice drivers may also fail to recognize hazardous situations. For example, Garay-Vega et al (2005) cued drivers to a potential hazard, such as a pedestrian on the sidewalk near a crosswalk. A truck obscured the pedestrian later. Of all those who fixate upon the pedestrian when on the sidewalk, 37
85.0% of the experienced drivers scanned toward the crosswalk in front of the truck, but only 47.5% of the novice drivers scanned toward the area of the crosswalk in front of the truck. Of those drivers who did not fixate upon the pedestrian on the sidewalk, 60.7% of the experienced drivers scanned the critical area in front of the truck and only 33.7% of the novice drivers scanned the area in front of the truck. 2.2.3.2 Hazard Mitigation: Speed Choice When there are Hazards near or within The Road on Straight Road Segments Polus et al. (1999) developed a series of models to predict operating speeds on straight road segments for two-lane rural roads. These models were based upon the tangent length (TL), and the radius of the curves to the start (R1) and end (R2) of each straight section. The Polus model accounts for the amplitudes and frequency of negotiating curves relative to the straight road segments, other factors influence the speed choice of drivers when traveling straight road segments, including optical flow, sight line, roadside obstacles and other factors. Optical flow is the amount of information or visual noise that emanates from the downstream horizontal extent and radiates past the driver. Nearly every other study that cites an influence on drivers’ speeds on a straight road segment could be considered a measure of optical flow. As an example of optical flow as part of a study by Shinar, McDowell, and Rockwell (1977) drivers were told to travel 60 mph (85 km/h) with the speedometer covered. When driving on a tree-lined road, the average speed of the drivers was 53 mph (85 km/h). On an open road, the average speed was 57 mph (91 km/h). The trees along the side of the road created more optical flow and a greater sense of speed as the trees moved past the peripheral view of the drivers. Conversely, drivers tend to overestimate 38
their speed when optical flow is reduced, such as when driving in fog or in stark environments (Snowden, et al, 1998; Trick, 2009; Shinar et al., 1977). Optical flow could be considered the amount of information per second that a driver is processing. At greater speeds drivers process more information per second, assuming the environment does not change. If drivers reduce their speed to be better able to process the information they receive, experienced drivers would be expected to reduce their speed for increased traffic density, work zones, and mid-block crosswalks as well. Practical application of the optical flow concept can be seen in traffic publications and in research. Optical flow can be described by using a tunneling metaphor. If the roadside trees are closer (Shinar et al., 1977), drivers slow more. Note that when a lane is narrowed (HCM, 2000) -- drivers slow; when we place roadside obstacles (O’Leary et al., 2006; Thompson et al., 1985; Lisle and Hargrove, 1980) – drivers slow; and when we take away the center divider (HCM, 2000) – drivers slow (See Table 5).
39
Table 5: Research listing the amplitude of speed loss and factors that are correlated to speed loss.
Source
Stimulus
Speed Loss
HCM, 2000
Obstacle within 6 ft. (1.8 m) to the right or 2 ft. (1.2 m) to the left
Speed loss begins
HCM, 2000
Undivided highway
2.0 mph (3.2 km/h)
O’Leary et al., 2006
Roadside Pedestrian
0.5 mph (0.8 km/h)
O’Leary et al., 2006
Roadside pedestrian and vehicle
1.5 mph (2.4 km/h)
O’Leary et al., 2006
Roadside vehicle
1.6 mph (2.6 km/h)
Thompson, Fraser, & Howarth, 1985
Ten roadside pedestrians
1.0 mph (1.6 km/h)
Lisle & Hargrove, 1980
Portable concrete barriers at laneside
2.0 mph (3.2 km/h)*
HCM, 2000
Lane narrowing
ROADSIDE
4%
1 ft. (0.3 m) VISUAL NOISE HCM, 2000
Traffic volume 1,300 vehicle/lane/hour
Speed loss begins at a rate of 1.0 mph (1.6 km/h) / 200 vehicles.
Fitzpatrick, Krammes, & Fambro, 1997
Driveway cut outs
Approximately 2.5 mph (4.2 km/h) per cut-out / km for up to 4 cut-outs (minimally thereafter)
Fuller, 1984 Construction zone signs 50% will slow * Lisle and Hargrove (1980) found that when portable concrete barriers were placed on the median at a road curvature to the right, average speeds dropped by 6.6%, compared with 2.3% on left curves and 3.1% on straight sections of the road.
Lane width may be associated with speed changes, but did not have a strong association with crash rate (Glennon et al., 1985; Choueiri and Lamm, 1987). Glennon et al. claimed that the crash rates was more sensative to the shoulder width than to lane width. As can be seen in Table 4, many of the research studies related to speed loss involve a reduction in the available shoulder area. When considered in context with the research by Glennon et al and Choueiri et al, drivers might slow due to optical flow (an ambient vision task) but might now slow enough to mitigate dangers when there were roadside obstacles (a focal vision task and more associated with crash risk). From this 40
information we might expect drivers who heed the available information to slow more when the road is narrowed by traffic or obstacles. In before-and-after studies involving the installation of rumble strips near a midblock crosswalk, after 90 days, the average speed was slightly less than baseline. However, after one year, speeds returned to their pre-installation average. Of the ten locations studies, five reported greater average speeds and five reported lower average speeds. During this period, there were two more crashes than for the same period before installation (Cynecki, et al., 1993). With rumble stripes, there are no obstacles or increased information density, nor is there a safety-related need to reduce speed, both likely reasons for the failure to slow over time. Varhelyi (1998) found that 5% of traffic yielded to pedestrians at non-signalized midblock crosswalks, 25% slowed and 75% of drivers maintained the same speed or accelerated. Drivers used high-speed behavior to signal pedestrians that they did not plan to give way. The speed limit for this study was 50 km/h. Fisher et al (2002) reported that young drivers slowed an average of 1.2 mph (0.4 m/s) when responding to a pedestrian preparing to cross in a marked midblock crosswalk. The average drivers first showed evidence of slowing 33 ft. (10 m) before the crosswalk. Experienced drivers braked more strongly near a STOP AHEAD sign placed 50 ft. (16 m) before an obscured stop sign than did younger drivers. However, experienced drivers were traveling faster. Younger drivers exhibited peak braking very late 5.7 ft. (1.75 m) before the stop sign and braked with greater force at that time. This suggests that younger drivers may exhibit compensatory and harder braking due to the delayed response.
41
Yielding behavior was monitored for drivers approaching a midblock crosswalk with shark's tooth advanced yield markings and with standard yield markings (Fisher et al., 2011). A vehicle was parked near the crosswalk to restrict the driver's view of the sidewalk. A pedestrian entered the crosswalk, emerging from behind the parked vehicle. In the standard crosswalk condition, 21.2% of drivers yielded to a pedestrian in the crosswalk and 29.5% yielded with advanced yield markings. Harrell (1994) monitored the braking responses of drivers when there was a pedestrian in a midblock crosswalk near a university hospital. Harrell found that drivers were more likely to stop for a roadside pedestrian in a crosswalk with bright clothing. Also, pedestrians with bright clothing did not have to wait as long. A street sign placed approximately 150 feet before the location of a pedestrian had no significant effect upon the motorists yielding to the pedestrians under daytime conditions. In summary, the literature tells us that roadside obstacles and objects that narrow the lane or increase visual noise are associated with decreased driver speed. When there is no narrowing of the available road (roadway plus buffer area), drivers’ speeds do not decrease as much, if at all. 2.2.3.3 Hazard Mitigation: Lane Position in when there are Hazards Near or within The Road on Tangent Road Segments O'Leary et al (2006) compared the lane position of drivers with no roadside obstacle, with a pedestrian only, with a pedestrian and vehicle, with a vehicle only, and with a vehicle with flashing lights. With a roadside pedestrian, drivers moved 0.9 ft. (0.27 m) away from the objects. When a pedestrian and vehicle were present, drivers
42
moved away 1.7 ft. (0.52 m) and when there was only a roadside vehicle, drivers moved away 1.2 ft. (0.37 m). Summala et al. (1996) measured the lane position and speed of vehicles approaching oncoming vehicles at several locations. They also measured the lateral position of the vehicles at positions up to ten seconds before and after passing. This experiment showed that there was no discernible lateral movement in car following situations and only a smooth 0.65 ft. (0.2 m) average movement in the opposite direction in open road driving. Triggs (1980) indicated that lateral movement away from oncoming traffic commenced at 2.4 and 3.6 seconds on average at two separate road locations. He found that drivers do not drive towards objects. When drivers were required to pass a parked truck, experienced drivers crossed the centerline of the road 39 ft. (12 m) before young untrained drivers (Fisher et al., 2002). An earlier movement around the truck allows the driver to anticipate the presence of oncoming traffic. If there is oncoming traffic, an earlier glance will allow the driver to move back behind the truck and stop in his lane if it is not clear to pass. All drivers slowed a little more than 13.4 mph (21.6 km/h). From the available research experienced drivers would be expected to move within their lane in response to a potential hazard. Related to the specific crash scenarios at mid-block crosswalks, work zones and variable speeds of traffic, it appears that when there are obstacles in and near the road, many experienced drivers will move within the lane to increase sightline and increase buffer space.
43
2.3 Summary of the How Driving Speed Changes at Horizontal Curves, Straight Road Segments and Intersections The studies mentioned thus far have shown that the speed of experienced drivers was influenced by several factors. When an experienced driver reduced his or her speed, it might suggest that the driver was anticipating some hazard. Given that novice drivers are not as adept at anticipating hazards, it is unlikely that novice drivers will reduce their speed as much, or as often, as have experienced drivers. In a complex environment, the amplitude of the speed loss is likely to be greater. When speed loss was greater, specifically, if a driver stops before turning, drivers give themselves more time to survey a complex environment. Hence, a complex environment was most likely associated with a need for an earlier anticipation and a greater response, which would suggest that there would be two reasons for novice drivers to slow less than experienced drivers. Based upon the literature discussed earlier,
44
Table 6 shows how several variables have influenced drivers to reduce their speed most. For example, sightline restrictions and traffic density increase driving difficulty and one can expect greater speed reductions. It was anticipated that novice drivers will not slow as much or as early, particularly when factors associated with difficulty are present.
45
Table 6: A summary of the topographical feature and how drivers’ speed choices change due to a variation in that feature.
TOPOGRAPHICAL FEATURE
LEAST OR NO SPEED LOSS
GREATEST SPEED LOSS
Approach speed
Low speed
High speed
Radius
450 m (1500 ft.)
100 m (330 ft.)
Curvature
6 degrees
6 degrees
Length of curve
100 m (330 ft.)
200 m (650 ft.)
Shoulder
Wide
Sight Line
Banking
Banked curve
No banking or negative
Right or left
Left
Right
Obstacle present on curve
None
Present
Pedestrian crosswalk
None
Present
Sightline restriction
None
Present
Foreshadowing event
None
Present
Traffic density
None
Present
Straight length
Long
Short
Radius before & after Straight
Shoulder
Wide
Road width
Road obstacles: Size
Small
Large
Road obstacles: Position
Right
Left
Oncoming traffic/traffic density
None
Present
HORIZONTAL CURVES
None 450 m (1500 ft.)
70 m (230 ft.)
INTERSECTIONS
STRAIGHT ROAD 450 m (1500 ft.)
100 m (330 ft.)
None 3.6 m (12 ft.)
3.6 m (12 ft.)
As a measure of behavior when traveling through an intersection, we can learn a lesson from the manner in which emergency-vehicle drivers are taught to negotiate through an intersection. Imagine that an emergency vehicle driver must arrive expeditiously for the purpose of saving life and limb. Clearly, normal drivers don’t need to travel that fast, but if an emergency vehicle driver must make efforts to slow before 46
traveling through an obstructed intersection, so should us normal drivers. For instance, Connecticut State law (CGS 14 – 283 (b) (2)) stated “…(2) proceed past any red light or stop signal or stop sign, but only after slowing down or stopping to the extent necessary for the safe operation of such vehicle…” . Also, “The provisions of this section shall not relieve the operator of an emergency vehicle from the duty to drive with due regard for the safety of all persons and property.” As a result, emergency vehicle drivers are taught to verify the intersection is clear before traveling through. The law in Connecticut regarding traveling through traffic signals is as follows: (CGS 14-299 b (1)) Circular green alone: Vehicular traffic facing a green signal may proceed straight through or turn right or left unless a sign or marking at such place prohibits either such turn or straight through movement, except that such traffic shall yield the right-of-way to pedestrians and vehicles lawfully within a crosswalk or the intersection at the time such signal was exhibited; pedestrians facing the green signal, except when directed by separate pedestrian-control signals, may proceed across the highway within any marked or unmarked crosswalk. Emergency vehicles drivers recognize that traveling through an obstructed intersection at full speed could be very dangerous in some instances. The law requires drivers to yield to vehicles and pedestrians that are already within an intersection… even when they have a green traffic signal. Therefore, all drivers should heed this same warnings and concerns. Drivers must yield to all vehicles within or an immediate hazard to the intersection before entering the intersection. What this means is that drivers should 47
monitor traffic in every direction before attempting a turn. Also, a driver would verify that the intersection is clear of vehicles and pedestrians before entering the intersection. If the driver is unable to verify that the intersection is clear, he or she should glance to the extent of the sight line and slow, after all, if an ambulance responding to a lifethreatening event would slow in that situation, so should we.
48
CHAPTER 3 RESEARCH GOALS AND SIGNIFICANCE It might not be possible to claim that a driver failed to mitigate a hazard appropriately unless it was known that he or she anticipated that hazard. The hazard anticipation research demonstrated that many novice drivers fail to anticipate hazards. Thus, it could be difficult to determine whether novice drivers have the appropriate hazard mitigation skills simply because they responded or did not respond in a measurable way to a potential hazard. However, the hazard anticipation research has clearly shown that novice drivers can be trained to look for hazards (Fisher, et al., 2008; Pradhan et al., 2005, Taylor et al., 2011). The next step was to determine whether hazard anticipation training combined with hazard mitigation training would get novice drivers to the point where they could mitigate hazards as well as would an experienced driver. In the virtual environment of the simulator, experienced drivers are expected to respond differently than novice drivers (Experiment 1) and trained novice drivers are expected to respond differently than untrained novice drivers (Experiment 2). The detailed hypotheses are listed below.
3.1 Hypothesis related to glance behaviors A goal of the research is to determine if anticipation, via a glance leads to slowing (mitigation). For example, if a driver traveled on a main road and saw a stop sign on the side road to the right, he or she may glance toward the side road to determine whether a vehicle could potentially pose a threat on the side road if it did not stop. This glance was referred to as an anticipatory glance since it was made with the foreknowledge that a 49
potential threat could materialize. Subsequent to the anticipatory glance, the driver can initiate behaviors to mitigate the threat such as changes in speed or lane position. Hypothesis G1: In Experiment 1, at curves, straight roads, and intersections experienced drivers are more likely to make proper anticipation glances and novice drivers will be less likely to make proper anticipation glances. In Experiment 2, at curves, straight roads, and intersections novice drivers who receive ACT training will be more likely to make anticipatory glances than the untrained novice drivers. A proper glance was defined differently for curves, intersections and straight segments. At a curve, a proper glance was defined as a glance to the far extent. The far extent was a location that was more than five degrees to the right. A far extent glance is not near the immediate inside of the curve but cuts across the inside of the curve. While several authors have noted where drivers glance in curves, they have been silent regarding when those glances occurred. Many research papers describe launch zones and target zones. Experiment 1 will examine all glance behaviors from a curve, straight road segment, or intersection up to ten seconds before that feature. As an example, suppose (as was the case) exemplary experienced drivers make considerably more glances than the novice drivers to the far extent of the sight line right at a sharp curve to the right when five to eight seconds before the curve. If so, a proper glance will be defined as a glance across the curve at a time earlier than five seconds before the curve (the launch zone would then be the area of the roadway in which the driver was between five and ten seconds from the curve). For intersections, the launch zone typically begins when a driver is near the back of an obstructing vehicle, or at some time near three seconds before the front of the 50
obstructing object. Again, the open question in Experiment 1 was to define where exemplary drivers glance and to note if there was a difference between them and the novice drivers. In the data collection, records were kept of all glances and one of the bins (areas of glances) included glances to the near extent, specifically, whether the driver glanced toward the extent of the sightline adjacent to an obstruction. (e.g., the right front of a vehicle stopped to the turn left in the lane to the left). When attempting a left turn at a busy intersection, glances were recorded in three categories. First, while approaching the intersection glances to the side roads were compared. Specifically, blind scorers recorded when each driver made an anticipatory glance to both the left leg and the right leg and which one-second epoch both side road glances were completed. Secondly, as the driver made a left turn, there was an oncoming vehicle. The blind scorers recorded whether the driver made a secondary glance toward the oncoming vehicle immediately after starting the left turn. As the name implies, this secondary glance was a last chance to avoid a crash. Lastly, the blind scorers recorded the percentage of drivers in each group that made both the side road glances, as well as the secondary glance. Ultimately, each driver was evaluated on this measure. For straight segments, the launch zones were similar to the intersection measures. At straight segment 1, the bus stopped near a crosswalk, the configuration was very much like the left turning truck at the intersection, but here, the large stopped vehicle was to the right rather than to the left. Again, near and far extent glances were recorded in each of the ten seconds when approaching the crosswalk. At the roadside obstacle scenario, where a pedestrian was along the left side of the road and a truck was to the right, glances were recorded similarly to the busy intersection. 51
The question here was whether the driver made anticipatory glances toward both the pedestrian to the left and the truck to the right. The hypothesis here is that experienced drivers (in Experiment 1) and ACT trained drivers (in Experiment 2) would be more likely to make neutralizing glances to all three legs of an intersection before attempting a turn and would be more likely to make anticipatory glances when entering an intersection with a view obstruction due to a left turning truck. Hypothesis G2: In Experiment 1, at curves, experienced drivers were hypothesized to be more likely to make earlier anticipatory glances toward the extent of the sightline. In Experiment 2, at curves, novice drivers who receive ACT training would be more likely to make earlier anticipatory glances toward the extent of the sightline. In Experiment 1, attempts were made to conditionalize the time and location of the glances with subsequent slowing. While we do not know what a driver was thinking and whether the glance was associated with anticipatory thoughts, if the speed loss can be tied to whether a driver made an anticipatory glance or not, we can argue that the glances were associated with anticipatory actions. Hence, one hypothesis was that glances to the extent of the sightline would be associated with subsequent speed loss. Hypothesis G3: In Experiment 1, at intersections where the driver does not need to stop before entering the intersection, experienced drivers will be more likely to make anticipatory and secondary glances when approaching the intersection. An anticipatory glance was a glance to the right or left side road traffic when approaching the intersection. The goal of the anticipatory glance was to assure that traffic hazards are neutralized before a turn is commenced. A secondary glance was a glance toward the 52
next most threatening area at the moment the turn begins or sometime immediately after the start of turning. Specifically, in this research, the secondary glance was a glance toward an oncoming vehicle. And in Experiment 2, at intersections, I hypothesized that ACT trained novice drivers would be more likely to make anticipatory and secondary glances. Therefore, a correct glancing pattern was assessed in two ways. First, did the driver make a secondary glance toward oncoming traffic at the moment or sometime after the start of lateral movement? Second, did the driver glance toward each of the three road segments in the three seconds before the start of lateral movement? Such a check is to make sure the driver is controlling the situation and terminating the potential risk. Specifically, was traffic at all three legs stable, stopped, or not threatening before the turn begins? If not, the driver should slow or stop, as described earlier relative to the way emergency vehicles are allowed to travel through and intersection only after slowing enough to assure that the intersection is clear The hypothesis here is that experienced drivers (in Experiment 1) and ACT trained drivers (in Experiment 2) would neutralize potential for hazards by making anticipatory glances as described here. Hypothesis G4: Straight Segments. An anticipatory glance at a Straight Segment was a glance to the roadside traffic along the right or left side of the road when approaching the straight segment incident or where there was a sight line obstruction, to the extent of the sight line. The goal of the anticipatory glance was to assure that traffic hazards were neutralized before passing roadside obstacles. For intersections in Experiment 2, it was hypothesized that ACT trained novice drivers would be more likely to make anticipatory and secondary glances. Therefore, a correct glancing pattern was evaluated in two ways. First, did the driver make glances to both sides of the road before 53
passing the roadside obstacle? Specifically, was traffic to the left and right stable, stopped, or not threatening before passing through the area? If not, the driver should slow or stop. The hypothesis here was that experienced drivers (in Experiment 1) and ACT trained drivers (in Experiment 2) would neutralize potential for hazards by making earlier anticipatory glances as described here.
3.2 Hypotheses related to speed choice Hypothesis S1: For each of the three environments listed earlier, it was hypothesized that experienced drivers in Experiment 1 would be more likely to do each of the following more than novice drivers in Experiment 1: a. begin to slow earlier than novice drivers, b. strike other vehicles less frequently, and c. drive out of his or her lane or leave the road less frequently. Drivers that struck another vehicle or completely left the travel lane were considered to have crashed. I hypothesized that all drivers would eventually brake or slow for a sharp curve. If a driver failed to slow early in the event, harder braking would be necessary later. Therefore, the probability of braking before a curve might not be a measure that can be used to differentiate safe from unsafe behavior. Instead, the thought was that gradual slowing earlier in the event might offer greater insight into the differences between novice and experienced drivers.
54
Hypothesis S2: In Experiment 2, drivers who receive ACT training would perform better than the untrained novice drivers for (a), (b) and (c) in all three environments.
3.3 Hypotheses related to conditional probabilities Hypothesis S|G1: As was mentioned earlier, I expected experienced drivers to make considerably more anticipatory glances than the novice drivers. Also, I expected experienced drivers to slow more often and earlier than novice drivers. A question remains unanswered, “If the novice drivers made a glance, would they slow as much and as often?” In Experiment 1, experienced drivers were still expected to slow more often than the novice drivers, even when a glance occurs. Similarly, it was hypothesized that more ACT trained drivers would slow if they glance than would novice drivers. Hypothesis S|nC1: While glances are a vital part of a driver’s situational awareness, slowing avoids crashes is a more direct act. In Experiment 1, it was hypothesized that experienced drivers who did not crash were going to be much more likely to slow to a target (safe) speed than novice drivers. Also, in Experiment 2, it was hypothesized that ACT trained drivers would be more likely to have slowed more often if they avoided the crash than placebo trained drivers. As part of this hypothesis, the flip side is that those who crash will not be likely to have slowed.
55
CHAPTER 4 EXPERIMENT 1: FINDING THE DIFFERENCES BETWEEN NOVICE AND EXPERIENCED DRIVER BEHAVIOR
The literature in Chapter 2 tells us where drivers, particularly experienced drivers, manage their speed, lane position, and fixation locations in each of two geometric environments that are most associated with crashes by novice drivers. In all the situations cited, the movement left or right, or the slowing cannot be attributed solely to the geometry. Clearly, however, the average experienced driver slowed, or moved, or both, in response to the geometries and arrangements of signs, shoulder availability, traffic density, and other factors. The purpose of Experiment 1 described below was to examine the speed management and anticipatory glances of novice drivers. Specifically, where do novice drivers fail to act similar to experienced drivers with much safer driving histories? The literature mentioned in Chapter 2 was a clue that allows us to know where to look for probable differences between novice and experienced drivers. The literature shows us specific geometries and arrangements where most experienced drivers slow. Examining the novice drivers at these geometries and arrangements was the best starting point based upon the available evidence. Driving simulators do not replicate the lateral forces experienced by a driver. However, this research addresses hazard anticipation and driver response to the presence of curves and obstacles, not the drivers’ responses to the curve itself. Given the stated crash risks mentioned earlier, the simulated environment was the best option for the current research. Drivers were exposed to potential hazards that, if ignored, might 56
develop into an immediate hazard and crash situation. To be able to expose drivers to life-like hazards without endangering the participants, a simulated environment was required.
4.1 Participants Eighteen newly licensed drivers were recruited through youth organizations and local driving schools. The average newly licensed driver had been licensed for 2.6 months and was 17.0 years old. Drivers ranged in age from 16 to 18. Eighteen experienced drivers were recruited from State Police, crash reconstruction organizations and community organizations. Every effort was made to recruit exemplary drivers. The requirements for inclusion in the experienced group were holding a license for at least ten years and no crashes or citations within the previous ten years, or be a police crash investigator, driving more than 20,000 kilometers per year. The average experienced driver was licensed 29.4 years and was 45.8 years old. The goal is to compare the novice drivers with a group of very good drivers to gather the information necessary to develop a hazard mitigation training program. The goal is to teach novice drivers to behave like a good driver. I don’t want novices to act like random experienced drivers, some of which could be safe or undafe drivers. Recruiting random experienced drivers would not allow me to reach that goal The local driving schools were offered nominal monetary incentives to assist in the recruiting of novice drivers. Throughout this paper, I have referred to novice drivers or newly licensed drivers. Each research study referred to novices in a slightly different manner. Generally, a novice driver was a16 or 17 year old driver with less than one year 57
driving experience. In this paper, I sometimes refer to novice drivers as newly licensed. My reasoning is to clarify that these experiments were addressing the behavior of not only young drivers, but young drivers with very limited driving experience. Participants in Experiment 1 received $40 compensation. Every effort was made to have an equal number of participants. To obtain an equal number of participants, twenty-three adults and twenty-one novice drivers were tested. Due to simulator sickness, the resulting participation was eighteen of each group.
4.2 Internal Review Board Approval This study has been reviewed and approved by the University of Massachusetts Amherst IRB, Federal Wide Assurance # 00003909. Approval was granted with the understanding that investigator(s) are responsible for modifications, consent forms, adverse event reporting and completion Reports. Protocol Title: Impact of a Speed Management Skills Training Program on Younger Drivers Tested on a Driving Simulator Protocol ID: 2012-1256 Approval Date: 04/20/2012 Expiration Date: 04/19/2013
4.3 Equipment In a virtual world, drivers were exposed to situations where a hazard may materialize. Materializing hazards in a field study would be unethical and dangerous. Thus, to address our initial research questions a driving simulator was selected instead of 58
a field study where participants maneuvered an actual car on the open road. On a driving simulator, neither the participant driver nor other drivers are a risk to themselves or others. (See Figure 5).
Figure 5: Driving simulator at the University of Massachusetts-Amherst showing the truck near crosswalk scenario.
A fixed-based driving simulator in the Arbella Insurance Human Performance Laboratory at the University of Massachusetts Amherst was utilized for this study (Figure 5). The simulator makes use of a Saturn sedan. The forward driving scene is displayed across three screens that encompass a visual horizontal field of 150 degrees and a vertical field of 30 degrees. The images are displayed at a resolution of 1024 X 768 dpi in each screen with a refresh rate of 60 Hz. The simulator also broadcasts road and engine noises with a Bose surround sound audio system. The ASL MobileEye eye tracker was employed to monitor eye movements of the driver. The MobileEye sampled eye movements and the forward view at 60 Hz. It contained both a scene camera (pointed ahead of the driver) that records 30 frames per second and infrared optics aimed at the 59
eye that also records at 30 frames per second. In the video that was reconstructed from the infrared and scene data, a crosshair representing the direction of gaze of the driver was superimposed onto the forward scene view. The resulting eye tracking video shows where the driver was looking in the environment by depicting crosshairs on the forward scene view. Among other things, this allows one to determine whether drivers made glances into any particular area of their forward field.
4.4 Experimental Design: The Virtual Drives and Order of Exposure A scenario was a specific combination of geometry, arrangement, and threat potential. There are three geometries: LTAP (Left Turn across Path) intersections, curves, and straight road sections. There were three arrangements of each geometry, meaning that there were three types of curves, three types of LTAP configurations, and three straight road segments. Drivers faced each of the three curves, LTAP intersections and straight road arrangements three times (three repetitions). Hence, each driver negotiated through nine curve scenarios (three different curve arrangements displayed three times each), nine LTAP intersection arrangements, and nine straight road segments (twenty-seven total scenarios). The three repetitions of each the three arrangements in each of the three different geometries were not exact duplicates of one another. In one of the three repetitions of each curve arrangement, LTAP arrangement, and straight segment arrangement, (twentyseven scenarios) a potential threat was materialized. In two of the three repetitions for each arrangement, no threat was displayed. Thus, there were a total of nine scenarios in which potential hazards could materialize during approximately 36 minutes of driving.
60
For the most part, a driver would have to be extremely careless for an immediate hazard to develop. Drivers who slowed early in the event did not experience an immediate hazard or emergency event, so some drivers might not experience any hazard that required hard braking. Conversely, failure to anticipate a potential hazard would lead to an emergency response situation and some drivers could experience as many as nine emergency response situations. The three LTAP intersection arrangements included NMVCCS scenarios 68, 69, and 82. Specifically, in one scenario the driver was turning left and a hazard could encroach from the left (scenario 82), in a second scenario the driver was traveling straight and a vehicle traveling in the opposite direction vehicles turned left across the driver’s path (scenario 69), and in a third scenario the driver was taking a left turn at the same time oncoming vehicles were traveling straight through the intersection (Scenario 68). In each of the LTAP intersection scenarios, a vehicle initially obstructed the driver’s view of an intruding vehicle when materialized. The three curve arrangements included a gradual curve left, a moderate curve right followed by a stop sign, and a sharp curve right when already negotiating a moderate curve to the right. For each of the curve arrangements, a vehicle was placed near the end of the curve at a location where the car was visible if the driver initially looked to the farthest extent of the sightline. Yet, in each of the curves, the vehicle, when materialized, became obstructed as the driver moved toward the curve. The drivers negotiated several straight segments as well. The three straight road arrangements included three instances where the driver was following a lead vehicle. In the first scenario the driver was traveling straight when a truck that was parked 61
perpendicular to the road started to move across the road. This caused the lead vehicle to stop for the truck. There was a pedestrian standing at the roadside to the left. In a second scenario the lead vehicle drifted out of its lane in a work zone and stopped suddenly when a pedestrian emerges from the work zone, and in a third scenario the lead vehicle passed a slow moving bus in the right lane. A pedestrian who emerged from in front of the bus at a midblock crosswalk caused the lead vehicle to stop suddenly. In each of the straight road scenarios, a lead vehicle slowed suddenly. In the first two scenarios the event developed slowly and directly in front of the driver. Yet in the third scenario, the lead vehicle was not obscured while it changed lanes into the path of the driver, but the pedestrian that caused the lead vehicle to stop was obstructed by the bus in the right lane. The counterbalancing of the scenarios in each drive is shown in Table 7. A letter and a number were used to identify each scenario. The letter refers to the geometry of the environment -- “c” is horizontal curve, ”s” is straight (straight) road, and ”i” is intersection. The number represents the arrangement. This arrangement number was nominal and without significance other than to differentiate one arrangement from another arrangement. An exception to this was that for curves, “3” refers to the arrangement with the smallest radius curve and “1” refers to the arrangement with the largest radius curve. The radius of the curve has been shown to be a difficulty factor, but that was not the intent with the numbering of the arrangements. The presence of an asterisk indicates that a secondary hazard materialized. For instance, “S-2 *” would be the second straight (straight) road segment; the asterisk indicates that a hazard materialized. “C-3” indicates the driver would be faced with the tightest radius curve; no hazard was materialized. 62
The counter-balancing meets the following six rules: 1. Across the three drives, every geometry appears exactly once as the first scenario, the second scenario, and so on; 2. Across the drives, a hazard was materialized exactly once in the first scenario, exactly once in the second scenario, and so on; 3. Across the drives, each of the nine combinations of geometry and arrangement (three curve arrangements, three intersection arrangements and three straight road segment arrangements) appears once with the hazard materializing and twice without the hazard materializing (for a total of twenty-seven different scenarios); 4. Within each drive, each of the nine combinations of curve, intersection and straight geometry and arrangement appears exactly once; 5. Within each drive, a hazard was materialized exactly three times, once for each of the three different geometries; and 6. Within each drive, exactly two scenarios in which a hazard does not materialize are sandwiched between the three scenarios in which a hazard does materialize. Road lengths were placed between each of the scenarios below such that the driver was equally likely to emerge from straight segments and intersections. No arrangement was preceded directly by a curve in that Polus et al. (1999) showed that straight road speed choice was influenced by curves placed before or after a straight road segment. One additional curve was added to the end of each drive and at three other locations pedestrians were placed near crosswalks or at the roadside without consequence. There was a concern that too many curves, in conjunction with left turns at intersections, would cause drivers to have a greater incidence of simulator sickness. 63
Hence, curves and turns had to be carefully balanced such that they are not salient to the driver, but not frequent enough to cause discomfort. . (See Table 7). Table 7: Counter-balancing of materialized hazards. (“I” refers to intersection, S refers to straight road segments (Straight road); C refers to curves. 1, 2, and 3 refer to the nominal identifier. If highlighted and an asterisk, a potential hazard was materialized.)
Drive 1
Scenario Materialize I-1 C-1 * S-1 C-2 S-2 * I-2 S-3 I-3 * C-3
Drive 2
Scenario Materialize S-3 I-3 C-3 * I-1 C-1 S-1 * C-2 S-2 I-2 *
Drive 3
Scenario Materialize C-2 * S-2 I-2 S-3 * I-3 C-3 I-1 * C-1 S-1
The order in which the drives were assigned to the participants was also counterbalanced. Drivers were equally likely to experience each drive first, second, or third. Also, each of the three drives was equally likely to follow any other drive. 4.4.1 Geometry: Horizontal Curves In summary, the root cause of many single vehicles crashes at curves appears to stem from a driver’s speed selection before entering the curve. When a driver enters a curve too fast, the tangential friction begins to approach or exceed the tire-road friction. This unstable situation results in an overreaction, particularly by inexperienced drivers. In many single-vehicles crashes, drivers under steered, a likely result from braking sharply (Zegeer, et al, 1990), or over steering (Glennon, et al., 1985; Maeda, et al. 1977), producing a turn that was sharper than the highway curve and further reducing the available friction. Under steering might cause a driver to lose control off the outside of 64
the curve (Go too straight at the curve). Over steering might cause a driver to lose control off the inside of the road. Over steering could be, and many times is, the result of an initial under steer (Maeda et al. 1977) or from a startling event such as slipping off the edge of the road (Hallmark et al., 2006) or being suddenly alerted from fatigue (Boyle et al. 2008) or a tire blow-out (Blythe et al. 1998) . For either problem, the cause could be attributed to the absence of slowing by the driver on the straight segment immediately in advance of the curve (Glennon et al, 1985; Mikolajetz, 2009). Glennon et al. referred to the region 3-seconds before the curve as the critical region of operations. In the critical region, drivers begin to adjust both their speed and path. Such adjustments are particularly large on sharper curves. Curves are naturally locations where there has been much greater crash risk (Zegeer et al., 1990), but there are two curve types that are of greatest interest and were examined. One curve type is associated with a relatively lower crash risk, and the other a very high crash risk. Curves to the left induced lesser speed loss and were associated with reduced crash risk when compared to curves right (Mikolajetz, et al, 2009). Right curves, longer curves and curves with smaller radii have been associated with greater crash risk (Anderson, et al., 2000; Glennon, et al., 1986). This research examined driver glance and speed mitigation behaviors for a gradual curve left and a longer tightening curve to the right. During each drive, participants negotiated through three types of curves (of which the behaviors where analyzed in only the above two types of curves, the gradual curve left and the longer, spiral curve to the right). Each of the three curves had different radii. A curve-ahead sign was placed along the side of the road consistent with the conditions 65
set forth in the MUTCD (2009), in this case 125 ft. in advance of the curve. Hazard anticipation was recorded if a driver glanced across the curve toward the road ahead (the target zone). Glance and speed behaviors were measured beginning ten seconds before the curve and marked the start of ten one-second launch zones. The hazard, when materialized, was visible before entering the curve should a driver look ahead across the curve, but was not visible in the immediate moments before the driver was rounding the curve due to vegetation. After receiving an average of 90 minutes of training, blind scorers compiled crash, road departure, and glance results. The glance results reported for each second represent the glances made during the preceding second. At each curve, there were three types of glances recorded: (1) Glances to the far extent across the curve (beyond a sightline obstruction); (2) Glances to the near extent of the sightline obstruction; and (3) Glances ahead or opposite the direction of the curve. For instance, if a driver reached the apex of the curve at 2:15 (two minutes and 15 seconds into the simulation), a bin was created for each second from 2:15 and back to 2:05 (a ten second window with ten onesecond epochs). First, please refer Figure 6 shown below. The glances were recorded in a matrix. Imagine each row represents the three areas shown in Figure 6. These areas are glances ahead or opposite (left on a curve right), glance to the near extent, or glances to the far extent (far right on a curve right or far left on a curve left). A glance ahead was defined as a glance that landed anywhere below the tree line and left of the right fog line. A glance to the near extent was recorded if the crosshairs from the eye tracker landed in the region depicted in blue which was from the right fog line to the tree along-side the road. A glance to the far extent was 66
recorded if the driver glanced across the curve toward the roadway ahead. This area is depicted with a green tint in Figure 6. Glances to the far extent (far right in this example) were particularly interesting in that they are indicative of the greatest forethought on behalf of the driver.
Figure 6: Examples of far extent, near extent and ahead-opposite glances.
I will illustrate how the scoring was performed. The scorer first identified the apex (peak) of the curve and recorded the time on the video. The video was reversed ten seconds and played frame-by-frame. If a driver reached the apex of the curve at 2:00.00, and if the eye tracking system shows the driver glanced both ahead (the red area) and to the far right extent (the green area) between 2:05 and 2:06, in row t = -6, the three cells (bins) would read 1 0 1: A one for a glance ahead, a zero because there was no glance in the area between the right road edge and the trees, and a one in the column designated for the far extent. The columns match the orientation of the glances (left and right) and the 67
rows match the orientation of the approach if moving down the page, farthest away (10seconds) is at the top, and the curve (t = 0 seconds), is the bottom row (Refer to Appendix D to see a copy of the scoring sheets). 4.4.1.1 Horizontal Curve Arrangements Curve 1. There were three curve arrangements. When I refer to the first through third arrangement, I am referring to arrangement types, not the order a driver encountered each arrangement. The three curves have different features. The first such arrangement (Curve 1) was the largest radius curve, which the literature would suggest was the easiest and safest to negotiate. In the example shown below, a curve to the left with a warning sign was depicted. Traffic on the back side of the curve was visible before entering the curve which is not the case if the driver was farther upstream of the curve. If a hazard materialized, a car (shown as a red car here, but it was actually a green car in the simulator) in the oncoming lane moved out around the truck that was stopped in the breakdown lane and intrudes one foot into the driver’s lane. (See Figure 7).
68
Figure 7: Largest radius curve that drivers will negotiate, a curve to the left. Aerial sketch was shown to the right and screen-shots from the RTI simulator showing the driver’s view was shown to the left. The truck in the background is parked along the opposite side of the road and a car (red) moves out from behind the truck, crossing the center lines by one-foot. The blue car depicts a lead vehicle that moves through the curve slightly more than ten seconds before our driver.
Curve 2. The second curve arrangement was a tightening spiral curve. An example of a materialized hazard is depicted in the third snapshot of the sequence. Again, 69
as in Curve 1, there was an oncoming vehicle in the area of the parked vehicle if present. Figure 8 depicts three sequential snapshots showing the smallest radius curve (top left, middle left, bottom left). The third sequential screenshot shows a parked vehicle partially in the driver’s lane. The parked vehicle represents the materialized hazard. The fourth screen shot on the right is an aerial view of the scene as the driver approached the stopped vehicle.
70
Figure 8: Curve 3, a tightening curve to the right. Right depicts an aerial sketch and left shows screenshots from the RTI simulator for this scenario. The first three screen-shots are taken as the driver moves through the curve.
Curve 3. In addition to a vehicle being downstream from the curve as with Curve 1, Curve 2 involves a horizontal curve to the right with a hidden intersection beyond the curve. This scenario was referred to as the “Curve-Stop-Ahead scenario” in prior experiments (Pradhan et al, 2005, 2009; Fisher et al., 2008). The Curve-Stop-Ahead scenario was a moderate horizontal curve to the right. The materialized hazard was a stopped car positioned at the far side of the curve. The car was stopped in a queue of traffic at a stop sign that was located beyond the curve. There was a symbolic sign warning drivers of the stop sign ahead before entering the curve. The thought behind this curve was that when a driver took the curve too quickly, there would be a need to slow rapidly in order to avoid the crash. Essentially, this scenario was very much like curve 1. In the scene shown below, a stopped vehicle at the far end of the curve can be seen. A left turning vehicle in the left lane encouraged drivers to remain in the right lane before entering the curve. The figure below is an example of a materialized hazard. As with Curve 1, the materialized vehicle will appear in only one of three occasions in which the participant negotiates Curve 3. Drivers who are maintaining a glance location at the far inside of the curve will detect the stopped vehicle earlier than will a driver who was randomly glancing ahead. While negotiating the curve, there was vegetation that limited the view ahead, but stopped traffic was visible under the umbrella of the leaves. (See Figure 9). 71
Figure 9: Negotiating a horizontal curve to the right with a queue of traffic at a stop sign beyond the curve. Aerial sketch to the right and screen-shots from the RTI simulator on the left. The blue car slows and turns left, leaving the right lane open for our driver to pass.
In summary, Curve 1 was a relatively stark curve to the left with the largest radius. Curve 2 had the smallest radius turn which occurred after an initial gradual curve right. Curve 3 was the same as curve 1 but curved to the right rather than the left. All three curves had the same theme - if a driver anticipates and heeds the earlier cues offered by the scene, a dangerous situation will never materialize, with or without the hazard downstream present. At all three curves, a driver was able to see the materialized hazard before entering the curve (it was visible in the far extent). Once near the curve, the sightline was restricted by an obstacle at the inside of the curve. After emerging from the curve, the driver was exposed to the materialized hazard when present. Two of these curves, curves 1 and 2 were sites where driver behavior was recorded. The two curves represented a routine and normal curve to the left and a rather sharp curve to the right. In the context of the literature mentioned earlier, curves 1 and 2
72
are described in more detail in Table 8. As you might notice, curve 1 is a site where few crashes were expected. The only reason for an increase in crash risk at all is because it is not a straight road and curves by themselves are associated with elevated crash risk. Bonneson’s equation predicts a speed near 34 mph at the apex of the curve. Curve 2 is a different situation entirely; this was a very dangerous curve. Someone might refer to this curve as “dead man’s curve” if it were in his home town. Nearly every small town has a curve like this, and as you read this I am sure you are thinking of a curve in your town. Curve 2 was associated with nearly every attribute that was associated with crash risk. These include a narrow shoulder, small radius, curve right, obstructed view, and a tightening (spiral) curve (the curve starts out normal and gets sharper). Table 8: Description of the two curves where driver behaviors were recorded in context with the previously stated research. One might say that the two curves represent an easy curve left and a more dangerous curve to the right.
Curve Description Straight Segment speed Radius of Curve Curve deflection Length of Curve Bonneson & Pratt (2009) Estimate of traffic speed in curve Mikolajetz et al. (2009) Estimate of where drivers begin to slow Estimated of crash risk Anderson & Krammes (2000) (Specific to Characteristics) Zegeer, et al. (1990) (all curves)
73
Curve 1 Left
Sharp Curve 2 Right
40 mph 656 ft. 90 deg. 656 ft.
40 mph 150 ft. 90 deg. 804 ft.
32 mph
20 mph
4.9 sec
16.7 sec.
2.2 x 1.5 - 4 x
9.7 x 1.5 - 4 x
4.4.1.2 Horizontal Curve Measures My interest here was whether the driver was anticipating the curvature or roadside ahead and at what time this occurred. For both curve scenarios, I reported the cumulative percentage of drivers that glanced toward the farthest extent of the sightline and the time before the curve that the glance occurred. The performance of novice drivers and experienced drivers was recorded for later comparison. Specifically of interest was the slowing and anticipation behavior of the two different cohorts. The following information was recorded for each driver at each curve:
Distance from curve at which a driver slowed to a target speed. Target speed was the projected speed if the driver slowed at the same rate as he did in the previous second. Target speed was equal to the initial speed minus the deceleration experienced in the previous second projected (see equation 10). Region where the driver glanced during each of the ten seconds before arriving at the curve. Lane position during each of the ten seconds before arriving at the curve.
Results were also recorded so that the following graphs could be developed:
Percent slowing to some threshold speed as defined by the Bonneson and Pratt speed loss in curves model (experienced versus novice); Percent who have glanced to the far extent by the time they closed to that time from the curve (experienced versus novice); Conditional situations that account for slowing if there was a glance or not.
Target _ Speed Vo At ,
(10)
where Vo is the original velocity, ΔA is the change in speed during the previous second in mph/second, and t is the time from the current location to the apex of the curve.
74
You may recall that these measures were conditionalized, the condition being whether the driver looks across the curve. Differences in the behaviors of experienced and novice drivers who had, and had not anticipated the hazard, would point to raw hazard mitigation skills for the experienced drivers that the novice drivers did not have (assuming both that the probability that an experienced driver slowed given that he or she glanced is higher than this probability for a novice driver and that the probability that an experienced driver slowed given that he or she did not glance is higher than this probability for a novice driver). Even still, the real proof of safety is not whether a driver thinks, anticipates or glances, but whether they mitigate the hazard. Accordingly, crashes were also recorded. 4.4.2 Geometry: Intersections There were three different LTAP intersection arrangements. Again, when I refer to arrangement one, two or three, I am referring to a different scenario, not the order in which each appears. 4.4.2.1 Intersection Arrangements Intersection 1. The first of the three intersection arrangements was referred to as the truck left turn scenario or NMVCCS scenario 69. The truck left turn scenario involves a driver in the right lane of two through lanes in that direction. There was a permissive green signal (green ball) at the intersection ahead. As the driver approaches, the signalized four-way intersection, he or she was able to see opposing traffic traveling through the intersection in the opposite slow lane. A truck in the left lane next to the driver comes to a stop and signals a left turn. It was unclear why the truck was stopping
75
for a green signal until a car in the opposing lane of traffic does or does not make a left turn from a position in front of the truck and crosses the path of the driver. In the truck-left-turn scenario, anticipation was registered if the driver looked at the front right of the lead truck in the left lane when approaching it and then when passing the truck. (See Figure 10).
Figure 10: Truck left turn scenario or NMVCCS scenario 69. (The traffic signal remains green. The approaching driver was able to see a vehicle travel through the intersection. After the oncoming blue car passes by, the truck in the next lane slows to a stop. When the driver gets to a position near the rear of the truck, an oncoming car starts to turn left across the path of the driver.
Intersection 2. The second scenario was referred to as the “Opposing left turning truck scenario” or NMVCCS scenario 68. In this scenario a sports utility vehicle (SUV) 76
in the opposing high speed lane blocks the driver’s view of traffic in the opposing slow lane to the right of the SUV. A car in the slow opposing travel lane (i.e., the lane to the right of the SUV) did or did not appear. Anticipation was registered if the driver looks to the lane to the right of the SUV when turning left in front of the SUV in the opposing lane. Essentially, intersections 1 and 2 are the same scenario viewed from the opposite directions. (See Figure 11).
Figure 11: Opposing left turning truck in the left (higher speed) lane. (The view of traffic in the lane to the right of the truck was obstructed. To the right is a sketch of the scenario from above. To the left is a screen-shot from the RTI simulator.)
Intersection 3. This scenario was referred to as the “Intersection Path Intrusion Scenario” and replicates NMVCCS scenario 82. When approaching a two-way stop controlled intersection, our driver had the right of way and thus no obligation to stop. However, as our driver approached the intersection, traffic was approaching the intersection from the right, there was an oncoming vehicle approaching ahead, and there was a large truck stopped in the slow movement lane to the left. The side road was a four-lane road; hence, the approaching driver was unable to see if there was traffic in the 77
high speed lane on the road to the left. To add to the visual noise, a pedestrian can be seen on the sidewalk to the side road left. The oncoming vehicle was programmed to approach at a speed that would create approximately an 8 ½ second gap between the driver and that vehicle at the intersection. Several studies suggest that 50% of drivers will accept a 6 ½ second gap, meaning that half of all drivers will attempt a turn if the oncoming vehicle was 6 ½ seconds away (IDRR, 2012). Given an 8 ½ second gap, most drivers would turn, but given the visual noise and instability, the most prudent action would be to stop or nearly stop to allow for proper glances in each of the directions before attempting the turn. Side road traffic should stop, but as we know, some drivers do not make complete stops or stop beyond the stop line. The materialized hazard was a car approaching the intersection in the high speed lane from the left of the driver. The approaching car was initially hidden by the stopped truck on the driver’s left in the right lane. A driver that failed to look for traffic in the high speed lane might find it necessary to modify the turning path when the previously hidden car approaches from the left. Other, more prudent drivers glanced toward each area a vehicle might emerge, thus neutralizing each hazard before making the turn. For these prudent drivers, the approaching car from the left was not a hazard at all. The degree of difficulty rises sharply based directly upon the mitigation behaviors implemented earlier. As an example, looking toward the approaching cars to the left, right and ahead, as well as the pedestrian to assure that each did not enter the intersection, takes time that was not available when the driver made a sweeping left turn. When the
78
driver’s stopped or slowed, it allowed them to become assured that all hazards were neutralized before turning. Also, Happer, et al. (2009) showed that drivers making left turns at intersections made tighter (sweeping) turns if they did not stop and moved farther upstream before the turn if they stopped. Drivers that stop or slow will be more able to anticipate hazards due to the additional time and the wider turn. Anticipation was registered once the driver completed glances toward both the left and right side road traffic when approaching the intersection. Also, anticipation was recorded with secondary glances. A secondary glance is a glance toward the oncoming vehicle at or immediately after the start of the turning motion. Presumably, a secondary glance gives the driver a last chance to modify the turn should the approaching driver be traveling too fast. When the hazard was materialized a vehicle in the side road to the left failed to stop at the stop sign and attempted to drive across the intersection. This configuration has been shown to be a common crash configuration (Choven, et al., 1994). The materialized vehicle (the blue car) is pictured to the far left in the driver’s view. The intruding blue car is depicted as a red (materialized) car in the aerial view in Figure 12.
79
Figure 12: Side road traffic to the right, an oncoming vehicle approximately 8 ½ seconds away and a previously unseen car approaching from the left when materialized. (A pedestrian was standing on the far left street corner (behind the A-pillar). The aerial view is shown to the right and screen-shots from the RTI simulated scenario is shown to the left.)
In summary, Intersection 1 was a traffic controlled intersection displaying a green (permissive) ball-shaped signal. The view of traffic within the intersection was obscured by a stopped truck in the left lane. Intersection 2 was essentially the same configuration as Intersection 1, but viewed from the perspective of a left turning vehicle that was facing the stopped truck, but there was a large SUV rather than the truck. Intersection 3 was again the perspective of a left turning driver but at a busy four-way intersection with traffic on each of the other three legs. All three intersections had the same theme - if a driver anticipated and heeded the earlier warnings offered by the scene, a dangerous situation will never materialized. At all three intersections, a driver was able to see the potential (latent) hazard before entering the intersection. Once near the intersection, the sightline toward pedestrian and vehicle traffic was restricted by a vehicle on one of the four legs of the intersection. After entering from the intersection, the driver was exposed to the materialized hazard when present.
80
Two of these intersections, intersection 1 and 3 were sites where driver behavior was recorded. The two curves represented a straight through movement and a left turn movement. In the context of the literature mentioned earlier, Intersection 1, replicated scenario 69 in the NMVCCS database. Intersection 2 addressed scenario number 69 in the NMVCCS database (straight through movement with a left turning principle-othervehicle [POV]), while Intersection 3 addressed both scenario 68 (a left turn across the path of another vehicle) and scenario 82 (a vehicle approaching from the driver’s left when attempting a left turn) of the NMVCCS database. Intersections 1 and 3 were selected for analysis for several reasons. I wanted to include one through movement and one left turn movement. I also wanted to evaluate drivers’ behaviors at locations that address crash risk. Intersection 2 and Intersection 3 both address NMVCCS scenario 68, but Intersection 3 also addresses NMVCCS scenario 82. Also, Intersections 1 and 2 were essentially the same scenario from different perspectives. Consequently, Intersection 2 was somewhat redundant, with Intersections 1 and 3 offering the greatest contrast, in that they address both type movements (through and turning), and address the greatest overall crash risk. Table 9: Description of the two intersections where driver behaviors were recorded in context with the previously stated research.
Intersection Description Intersection 1 Initial speed 40 mph Movements Through Sightline restriction Ahead to left
Intersection 3 40 mph Left Turn Left Side Road From Left Side Materialized hazard Opposing Left turn Road Anticipatory Target Speed Preparing to stop Slowing Ranking as Most Dangerous 81
NMVCCS
6th Most
1st & 3rd Most
4.4.2.2 Intersection Measurements The performance of novice drivers and experienced drivers was recorded for later comparison. Specifically of interest was the slowing and vehicle positioning behaviors of the two cohorts. Preparatory slowing was of particular interest when moving near a hazard. The following information was recorded for each driver at each LTAP intersection: Straight movements through LTAP intersections: Distance from intersection the driver began to slow; and Percentage of drivers that glance toward the near extent for each second before the center of the intersection. Lane position during each of the ten seconds before arriving at the intersection. Conditional situations that account for slowing if there was a glance or not. Left turning LTAP at intersections: Distance from intersection the driver began to slow; Percentage of drivers who make anticipatory glances toward all three potential hazards regions (traffic left and right, and oncoming vehicle). Conditional situations that account for slowing if there was a glance or not. At intersection 1 each glance was categorized into one of four regions. For straight movements through an intersection, the four regions were view left, view toward the near extent, view ahead and view right. When driving straight through the intersection, view left was a glance to the left of the obstructing truck in the next lane. The near extent was the location between the dashed line (along the left side of the driver’s vehicle) and the right edge of the obstructing vehicle. There were some glances ahead and some glances to the road to the right. The percentage glances for each second 82
was reported rather than the cumulative percentage. At this scenario, my interest was whether the driver made an anticipatory glance toward the front of the truck while passing it. This glance was presumably toward oncoming traffic or a crossing pedestrian that was previously hidden by the left turning truck. I was particularly interested in an attempt to determine if the glances were intentional and anticipatory. To gain such knowledge, recording the specific epoch of the glance was necessary. Reporting cumulative glances did not offer the necessary specificity. When making a left turn (intersection 3), there were three forward regions, left, right and ahead. Imagine traveling toward a busy intersection with potential hazards approaching from each leg of the intersection. I was interested in the cumulative percentage of drivers that made glances to both the right and left side road as they approached the intersection. Specifically, when did they make neutralizing glances to the left and right legs of the intersection? Next, I was interested in whether the driver made a secondary glance toward oncoming traffic at the moment or immediately after the start of the turn. I also reported the percentage of drivers that made both side road glances and secondary glances. 4.4.3 Geometry: Straight Road Segments There were three different straight road arrangements. Again, when I refer to arrangement one, two or three, it refers to a different scenario, not the order in which each appears. 4.4.3.1 Straight Road Arrangements Straight Road 1. The first tangent road segment arrangement is referred to as the midblock crosswalk on a multi-lane road scenario. A bus in the right lane ahead slowed 83
to a stop as our driver and a lead vehicle approach the bus, the lead vehicle moves out of the right lane and begins to pass the bus as the bus settled to a stop immediately before a mid-block crosswalk. When the hazard was materialized, a pedestrian emerged from in front of the bus which caused the car (lead vehicle) to stop suddenly. Anticipation is registered if driver glances in the area to the right far extent (between the bus and hedge alongside the road right) when five to ten seconds from the crosswalk and in the area of the near extent (left edge of the bus) when one to five seconds from the crosswalk. Essentially, we are asking if the driver was looking to see (anticipating) if there was pedestrian traffic in the area of the stopped bus. In particular, was the driver able to appreciate that a pedestrian might cause the lead vehicle to stop? The figure below shows an example of the scenario. There is a red lead vehicle, a bus stopped in the right lane, and when materialized, a pedestrian crossing from right to left in front of the bus and red (lead) car. The crosses depict the three glance regions. There was a warning sign that advises motorists of the crosswalk both at, and 100 ft. before, the crosswalk in accordance with the MUTCD guidelines. When the hazard was not materialized, the lead vehicle drove past the bus without slowing. (See Figure 13).
84
Driver’s View Figure 13: Bus blocking crosswalk scenario. (The approaching driver is able to see a lead vehicle pass the bus and travel through the crosswalk. After that car clears a pedestrian might suddenly emerge into the path of the lead vehicle from a position in front of the bus.)
Straight Segment 2. The second tangent road segment arrangement is referred to as pedestrian in the work zone. As the driver negotiates through the right lane closure, he or she will pass several workers within the work zone. The work zone is in the right lane and our driver will be in the left lane. When a hazard was materialized, a pedestrian from within the work zone walked into the open travel way in front of a lead vehicle. Our driver will be following this same lead vehicle. The lead vehicle steers left away from the pedestrian and then stops suddenly in the median shoulder. When the hazard is not materialized, the lead vehicle will continue moving forward without slowing and the pedestrian walks within the work zone but does not move toward the travel lane. There are construction barrels that are lined up along the dashed line. A reason why the lead vehicle steered left was to assure that the pedestrian would remain with the drivers’ sight lines. (See Figure 14).
85
Figure 14: Pedestrian in the work zone. When negotiating a right lane closure a pedestrian might or might not move into the drivers’ travel lane. (To the right is a sketch of the scenario from above. To the left are screen-shots from the RTI simulator showing this scenario.)
Straight Segment 3. The third tangent road segment scenario will be referred to as the backing truck scenario. Our driver is following a lead vehicle. Ahead of the lead vehicle, a truck backs from a private driveway and across the road. Oncoming traffic prevents the lead vehicle from driving around the truck ahead. Also, there is a pedestrian, presumably a truck-helper, standing along the left road edge. When materialized, the truck begins to move into the road when the driver is approximately ten seconds away. This scenario suggests that the lead vehicle likely will be slowing soon. If the driver is attending to traffic downstream, he should be prepared to slow. When the hazard is not 86
materialized, the pedestrian is standing along the left road edge and the truck is parked perpendicular to the road but slightly protruding into the road. Anticipation is registered if the driver glances at the truck ahead of the lead vehicle and toward the pedestrian as the driver approaches the area of the truck. This scenario is not unlike scenario 10 by Pradhan et al. (2006). Rather than the lead vehicle turning left into the path of a pedestrian, here, the lead vehicle moves toward a vehicle that crosses the road. When the hazard does not materialize, the truck lurches forward about one foot which allowed traffic to move through the area without stopping. (See Figure 15).
Pedestrian
Figure 15: Side road car can be to the right. (This vehicle will move quickly toward this position and stop when depicted if not materialized. The sketch to the right shows an overhead view, the screen-shots to the right are taken from the RTI simulated scenario.)
87
In summary, Straight Segment 1 was a bus stopped in front of, and partially obstructing the view of, a crosswalk ahead. The view of pedestrians on the sidewalk and entering the crosswalk were obscured by a stopped bus in the right lane. Straight Segment 2 was a right lane closure work zone where, when materialized, a pedestrian walked into the path of the lead vehicle. Straight segment 3 was similar to straight segment 1 but did not involve a work zone. There was a pedestrian along the roadside to the left and a perpendicularly parked truck at the road edge to the right. The lead vehicle slowed for the emerging truck when materialized. All three straight road segments had the same theme - if a driver anticipates that the lead vehicle might stop quickly, a dangerous situation will never materialize. At all three straight segments, a driver was able to see the potential hazard before reaching the area. In each of the three straight segment scenarios the immediate hazard, if materialized, could be seen developing slowing. At Straight Segment 1, the lead vehicle passed a bus that recently stopped at the crosswalk. At Straight Segment 2, the pedestrian took several seconds to walk from the far right road edge, across the closed lane, and into the drivers’ travel lane. At Straight Segment 3, the truck starts into the road nearly ten seconds before arrival. Two of these straight segments, Straight Segments 1 and 3 were sites where driver behavior was recorded. The two straight segments represented a latent hazard (The potential that a pedestrian might emerge from the front of the bus). Straight segments 2 and 3 represent scenarios where the hazard was not latent and was long developing. In scenario 3, the lead vehicle stopped for the crossing truck before the arrival of the driver. In the context of the literature mentioned earlier, Straight segment 1 was a midblock crosswalk that Zegeer et al (2002) indicated was a location of increased pedestrian crash 88
risk and involved a decelerating lead vehicle. In Straight segment 2 there was a decelerating lead vehicle at a long developing situation. At Straight Segment 3, it was also a long developing situation, but if differed from Straight Segments 1 and in that the lead vehicle stopped before the arrival of the driver. Straight Segments 1 and 3 were selected for analysis for several reasons. I wanted to include one decelerating lead vehicle scenario and one stopped lead vehicle scenario. I wanted to include a latent hazard and an unconcealed hazard. I also wanted to evaluate drivers’ behaviors at locations that address crash risk. According to the NMVCCS data, nearly half of all teen-involved crashes occurred when traveling straight and 21% of the crashes involved a vehicle decelerating or stopped. Also, Straight segment 2 and Intersection 3 involve similar issues, but Straight segment 3 involved a stopped lead vehicle, while in scenario 2, the lead vehicle moved out of the travel lane. Consequently, Straight segments 1 and 3 offered the greatest contrast, in that when combined I was able to addressed latent (most difficult) and unconcealed (Least difficult) hazards created by roadside obstacles. To continue with the issue of difficulty, drivers approached Straight Segment 1 at a greater speed (near 40 mph), than when approaching Straight Segment 3 (near 25 mph). The lead vehicle at Straight Segment 1 stopped suddenly, while the lead vehicle at Straight Segment 3 stopped when well ahead. Table 10: Description of the two straight segments where driver behaviors were recorded in context with the previously stated research.
Straight Segment Description Initial speed LV Movements Sightline restriction Materialized hazard
Straight Segment 1 40 mph Slowed Suddenly Ahead to left Latent 89
Straight Segment 3 25 mph Stopped Left Side Road Unconcealed
Crash Risk Zegeer et al., 2002 Greater Pedestrian Risk Several studies Stopped vehicles 4.4.3.2 Straight Segment Measurements In the straight road arrangements, the way glancing is recorded will be markedly different from what it is in the other two arrangements. Imagine driving down a road with a truck edging into the road on the right and a pedestrian standing on the left. Clearly, a driver should assure that both potential hazards are neutralized before driving through the area. Therefore, in Straight Segment 3, the specific glance measure was the cumulative percentage of driver glances toward both the pedestrian to the left and truck to the right. On the other hand, when approaching a bus blocking a crosswalk with a clear sightline ahead and to the left, clearly, the most dangerous area will be to the right. Glances to the extent of the sightline are best; however, the bus blocking crosswalk scenario is very dynamic. As was the case with the left turning truck at an intersection, reporting cumulative glances toward the front of the obstacle serve little purpose when they offer the driver little or no information. Such is the case when a driver is eight and ten seconds from the crosswalk. However, a glance toward the front of the truck at a time that is associated with a specific glance for a pedestrian, tells us much more about the anticipatory behavior of the driver. Consequently, the percentage of drivers that glanced each second was reported at the crosswalk scenario. The performance of novice drivers and experienced drivers was recorded for later comparison. Specifically of interest was the slowing and glancing behaviors of the two 90
cohorts. The following information was recorded for each driver at each of the two straight segments which were analyzed:
Speed of the vehicle in the ten seconds before reaching the bus (segment 1) or truck (segment 3);
time from obstacle at which the driver slows,
percentage of the drivers that slow to a safe (target) speed,
glances toward the pedestrian (left) and truck (right) at segment 3, and glances toward the near extent, toward the front of the bus at segment 1, and, lane position during each of the ten seconds before arriving at the incident.
Conditional probabilities were also computed. In particular, for those participants who glanced toward the potential hazards to the left and/or right, the percentage of drivers that reached target speeds in each second as they approached the position of the truck will be determined.
4.5 Procedure Upon arrival, participants were asked to read and sign a consent form following University policy (see Appendix A). By signing the document, the participants indicated their understanding of the experiment and their willingness to continue. This document also informed the participants that compensation will be provided at the end of the experiment. Experienced and novice drivers each received a pre-experimental questionnaire. The pre-experimental questionnaire was given to the drivers to record whether they use corrective lenses and demographic information such as age and gender (see Appendix B). Both in the recruiting materials and on the questionnaire the experienced driver group 91
was asked if they have been involved in a crash, or received any traffic citations in the previous ten years. Since this was necessary qualifying information for experienced drivers, novice drivers were asked as well. All experienced drivers had to have an unblemished driving record for ten years as a prerequisite for inclusion in this research. [As noted earlier, the goal in Experiment 2 was to train novice drivers to respond as would an excellent (experienced) driver.] Thus, I wanted a suitable baseline of experienced drivers in Experiment 1. The remaining questions allowed for posttest analysis to determine whether individual differences, such as health and miles driven each year, had a significant impact on the responses by any particular participant. All instructions were written down and read aloud to each driver to assure that each participant received the same instructions, delivered in the same way. These instructions can be seen in Appendix C. A four minute practice drive was completed before testing. Based upon the research by McGehee et al. (2005), a practice drive of three minutes or more was associated with the time necessary to reach a point that behavior was no longer influenced by unfamiliarity with simulator controls and virtual world. During the practice drive (and before the final instructions were read), the drivers were encouraged to “test” the vehicle’s steering, brakes and acceleration capabilities. The drivers were given as much time as they believed was necessary to become comfortable with the handling of the vehicle during the practice drive. There were hazards and traffic during the practice drive and the drivers learned during the practice drive that all vehicles stop for stop signs and other traffic acts as was typical of most drivers. At the end of the
92
practice drives the participant was asked to assume he was driving a friend’s vehicle that handled just as the simulator was handling and to drive accordingly. Participants were asked to travel at 40 mph through mixed road types that included four-lane suburban business arterials and two-lane rural and residential roads. Drivers were asked to drive as they normally would if it was necessary to get to an important meeting on time. Specifically, each driver was told that if he or she drives an average speed near that of the posted speed limit, the driver would arrive at the appointment on-time.
4.6 Data Collection and Analysis Data was recorded from the driving simulator, and eye-tracking systems. When collecting driving simulator information, the simulated world records all vehicle speeds, foot pedal use and lane position. The current research focuses on the speed before the apex of the curve and speed loss as well as the speed when approaching an obstructed intersection or roadside obstacles when following a lead vehicle on straight segments. Every effort was made to avoid recording any identifying information for any driver. No recordings were made of the driver. Once the data collection process was completed, the results of Experiment 1 were analyzed and documented. In each of the scenarios the locations where data were collected were defined as the launch and target zones for a proper glance. The launch and target zones coincided with the window at which speed and risk management behaviors were implemented. Unless otherwise stated in Chapter 4 of this dissertation,
93
the data collection window was ten seconds prior to the point of interest (apex of the curve, or intersection, or location of the roadside obstacle). Step 1. The hypotheses listed earlier were evaluated in Experiment 1. The first step in data collection was to collect eye glance data. For each of the scenarios, drivers’ eye glances and responses were measured and recorded during the ten seconds prior to an intersection, curve or straight segment roadside obstacle. A ten second window was selected for four primary reasons: (1) first, in pilot studies experienced drivers were subjectively driving “safer” but had similar initial and final speeds as the novice drivers. Examination of the speed data revealed that novice and experienced drivers differed in when they begin to slow, which was as early as 10 seconds before the point of interest. (2) Second, using the eye tracking equipment, the ability to determine exactly where a driver was glancing becomes reduced when a driver was as far as eight to ten seconds away. This means that the collection tools were accurate only to eight to ten seconds before the point of interest. (3) Third, the intersection research by Akcelik and Biggs (1987) shows that drivers begin to slow when nearly ten seconds before an intersection. (4) Forth, and lastly, Fisher et al. (2002) showed that drivers began slowing 50 m (164 ft.) to 110 m (361 ft.) before the obstacles. If traveling 30 mph, 361 feet would be over eight seconds away. Mikolajetz et al., (2009) also showed that drivers begin to slow hundreds of feet before a curve and the exact distance was dependent upon the sharpness of the curve. When the curve was gradual slowing occurred later or not at all. When the curve was more drastic, slowing began over ten seconds before arrival. Two reference points define the boundaries of each risky area. The first boundary defines the earliest area on the roadway in which a driver’s glance was considered as 94
relating to the latent hazard (Launch zone). The second boundary was usually at the location of the potential hazard, or the last point at which the potential hazard was still not visible. Glances toward a potential hazardous area were the surrogate measure of anticipation. Also, ten seconds before the event was at the extent of the capability of the eye tracking system. When ten seconds or more from an obstacle, it became more difficult to determine the exact object upon which the driver fixated. To remove subjectivity, blind scorers were utilized. A glance was recorded as occurring in a particular one second interval t if it occurred at a particular time x in the one second interval t < x < t + 1. For example, the glance epoch of time zero included all glances that occurred between 0.00 and 1.00 seconds before the curve. Speed was measured in a similar way. For example, time zero for speed includes the average of all speeds in the interval [t to t + 0.05]. At times the driving simulator might report an anomalous reading in one cell of the matrix output. To avoid the influence of an anomalous reading, the average of three consecutive readings was taken. The simulator reports information at a rate of 60 samples per second, or 0.0167 seconds for each sample. Three samples comprised 0.05 seconds. The open question was whether experienced drivers anticipate the need to slow earlier than novice drivers. This question has three parts. First, do experienced drivers glance earlier toward the most threating areas? Second, if given a glance, or no glance, did the experienced drivers slow earlier and, if so, by how much? And third, how much earlier before the point of interest did the glances and slowing begin? At most scenarios, there were three glance regions that were analyzed that encompassed the entire forward view as depicted in Figure 6. At the left turning truck 95
scenario (Intersection 1) and the bus stopped at the crosswalk (Straight Segment 1) there were four regions. View left, view toward the near extent, view ahead and view right. Glances during every one-second period were recorded. The results were in matrix form with each column representing a glance region (g1 gn) and each row representing a one-second interval (t-9, t-8… t0). A glance region (1 n) represented a sector of the drivers’ forward visual field. For instance, when approaching a curve right, the three regions would be 1. Left/Ahead, 2. Near extent right, 3. Far extent right. Initially each arrangement had more regions which included signs and traffic signals. However, particularly when more than six seconds from the event, the blind scorers had little agreement regarding traffic control glances. It was not uncommon for one scorer to record several sign glances and another to record none for the same driver at the same location. Hence, this research focused on glances toward the larger more general areas where there was very good agreement between scorers. Step 2. Step two involved an evaluation of the hypothesis that among drivers who made a proper glance, experienced drivers were likely to initiate a risk mitigating response earlier than were novice drivers. The distance at which the slowing begins was measured in seconds from the location of the potential hazard (lead vehicle in a line, crosswalk, apex of a curve, or stop line). Whether the glance at the hazard occurred before or after the driver’s first response was recorded as well. Therefore, there are four possible outcomes: (1) The driver does not make a glance towards the target zone and does not respond; (2) The driver does not make a glance toward the target zone and does respond in some way; (3) The driver makes a proper glance and then responds; and (4) The driver makes a proper glance and does not respond. 96
The analysis of driver performance in the experimental scenarios will answer two main questions: (1) whether the driver looks at the risk (or area where a risk could emerge) and (2) if the driver looks, did the driver diminish the potential risk by slowing?
4.7 Results This analysis will speak to the linked hazard anticipation and hazard mitigation behaviors of novice and exemplary experienced drivers at: (a) the two curves, a routine curve left (curve 2) and a long tightening curve to the right (curve 1); (b) the two intersections, the truck left turn scenario (intersection 1) and the path intrusion scenario at a busy four-way intersection (intersection 3); and (c) the two straight road segments, a bus blocking a crosswalk (straight segment 1) and a lead vehicle slowing for a crossing truck (straight segment 3). 4.7.1 Geometry: Horizontal curves According to the model proposed by Bonneson and Pratt (2009), most drivers would reduce speed to near 34 mph on the curve left and 20 mph on the sharp curve right. Using the Bonneson model as a benchmark, the speed loss and glance behaviors of the experienced and novice drivers were compared. First, the aggregate glance, slowing, and lane position choice behaviors were compared. Second, the slowing behaviors were compared conditional on an anticipatory glance being made (or not made). 4.7.1.1 Aggregate Glance, Braking and Slowing Behaviors, and Lane Position The principal relationships being examined were whether drivers made anticipatory glances when approaching curves and the possible link to subsequent speed loss before entering a curve. The open question here was whether the novice drivers 97
lacked some tactical response that the experienced drivers exhibited after controlling for anticipatory glances. Glances. The prerequisite to anticipatory slowing was anticipatory glances. The percentage of experienced drivers that glanced at each second when approaching a sharp curve and a routine curve were compared. Significantly more experienced drivers made glances across the curve (to the far extent) at five to six seconds before the curve when compared to the novice drivers. When seven and eight seconds from the curve, more than four times as many experienced drivers had made a glance toward the far extent than had novice drivers. Figure 16 and Table 11 also shows that experienced drivers were 52% more likely to glances across the curve than novice drivers when five seconds from the curve.
98
Figure 16: The cumulative percentage of experienced and novice drivers that made a glance to the far extent (across the curve) as they approached a sharp curve to the right. (*The asterisk - The difference between the experienced and novice drivers’ glances were significant.)
The statistical significance of the difference between the glance proportions of the experienced and novice drivers for each second at the sharp curve is shown in Table 11. At every one-second bin, experienced drivers were more likely to glance across the curve. When six and five seconds from the curve, 43% and 65%, respectively, of the experienced drivers had made a glance across the curve to the far extent. At the same time, fewer than 13% of the novice drivers made an anticipatory glance toward the far extent when five seconds from the curve.
99
Table 11: Comparison of experienced and novice drivers’ glances within each 1-second epoch before a curve. (Numbers in bold yellow indicate significant differences.)
100
At the moderate curve left, experienced drivers were more than 12% more likely to make glances to the far extent than were the novice drivers when three to seven seconds from the curve. (See Figure 17).
Figure 17: The cumulative percentage of experienced and novice drivers that made a glance to the far extent (across the curve) as they approached a curve to the left.
The statistical significance between the cumulative percentage of glances by the experienced and novice drivers for the moderate curve left during each second is shown in Table 12. Of interest, half of the experienced drivers had already made a glance toward the far extent when as far as seven seconds from the curve. At the moderate curve the largest difference in the groups was when they were three to five seconds before the curve where experienced drivers were at least 18 percentage points more likely to make a glance across the curve. 101
Table 12: Cumulative percentage of experienced and novice drivers who glanced toward the far extent of the sight line in each second while approaching a sharp curve to the right.
Speed. The speed profiles of the experienced and novice drivers show that experienced drivers had already slowed considerably from an initial speed near 40 mph when nine seconds before the curve when approaching the sharp curve. In addition to slowing very early, the experienced drivers began to slow at a greater rate when five seconds before the curve, while the novice drivers did not begin to brake more sharply until within three seconds from the curve. The asterisks refer to times when significantly more of the experienced drivers had slowed to a target speed of 20 mph, meaning that they were on-track to slow to 20 mph.
102
Figure 18: Average speed of experienced and novice drivers in the ten seconds before reaching a sharp curve to the right. (* The asterisk refers to times when significantly more experienced drivers slowed to a target speed of 20 mph.)
When approaching the sharp curve both the experienced and novice drivers had already reduced speeds from the speed limit when twelve seconds from the curve. At the moderate curve to the left, drivers were moving from a straight segment compared to the sharp curve right, where they were negotiating a moderate curve right before approaching the sharper curve. Even still, differences between the experienced and novice drivers were similar at both curves. The experienced drivers were slowing when nine seconds from the curve, yet we do not see a downturn in speed by the novice drivers until nearly six seconds before the sharp curve. In Figure 19, an asterisk refers to times when
103
significantly more of the experienced drivers had slowed to a target speed of 34 mph, meaning that they were on-track to slow to 34 mph.
Figure 19: Average speed of experienced and novice drivers in the ten seconds before reaching a moderate curve left. (* The asterisk refers to times when significantly more experienced drivers slowed to a target speed of 34 mph.)
If you recall, because the left curve was a larger radius than the sharp curve, less speed loss was necessary for safe operation according to the Bonneson model. According to the equation by Bonneson and Pratt (2009), we would expect an average curve speed to be 20 mph at the sharp curve and 34 mph at the moderate curve left. Also, I was interested in measuring anticipatory behaviors. A speed at a given spot did not describe the drivers’ behaviors as did a target speed. A target speed considers the current speed as well as the deceleration. Target speed refers to a speed that the driver is “on-target” to reach. For example, a driver might be preparing to stop, but if stopping is not necessary, 104
he or she might never stop. The valuable information is to know that the driver anticipated the need to stop at one time. The target speed is the current speed minus the speed loss in the previous second projected forward in time until the location of the curve. At the sharper curve, experienced drivers were significantly more likely to reduce speed to a target speed below 20 mph farther from the curve than were the novice drivers. When two seconds before the curve, more than half the experienced drivers had slowed to a target speed of less than 20 mph, yet fewer than half of the novice drivers slowed to a target speed of 20 mph. (See Table 13).
105
Table 13: Percentages of experienced and novice drivers that slowed to target speed of 20 mph when approaching a sharp curve to the right. (Numbers in bold yellow indicate significant differences.)
At the moderate curve left, the curve was not as sharp and the speed threshold was different, but the way experienced and novice drivers slowed to target speed was very similar to the speed choices at the sharp curve. Experienced drivers were significantly more likely to slow to a target speed of 34 mph when three and four to seven seconds before the curve than were the novice drivers. Just as with the sharp curve, fewer than half the novice drivers slowed to target speed, yet more than half the experienced drivers had slowed to target speed when six seconds before the curve.(See Table 14).
106
Table 14: Percentages of experienced and novice drivers that slowed to target speed of 34 mph when approaching a sharp curve to the right. (Numbers in bold yellow indicate significant differences.)
Also, novice drivers exhibited harder braking in the last second before the curve. An issue to consider is the effect of speed variance. The standard deviation for the novice drivers was 8.42 mph at the curve compared to only 5.14 mph for the experienced drivers. Essentially, the novice drivers’ speed selections were less reliable as well as being less anticipatory than the experienced drivers. The groups were compared relative to the extent at which the drivers slowed to the target speed. For instance, if Driver A slowed to target speed five seconds from the curve and Driver B slowed to the target speed one second before the curve, the average for these two drivers would be 3.0 seconds. Using this same methodology, the average time epoch at which the drivers in each group slowed to the target speed was compared. 107
At the sharp curve right, experienced drivers on average slowed to target speed 3.2 s. (SD = 2.3 s.) before the curve. Novice drivers did not reach target speed until 0.6 s. (SD = 3.0 s.) before the curve, a difference which was significant [t (29) = 2.78, P = 0.01]. At the moderate curve left experienced drivers slowed to target speed 4.6 s. (SD = 4.0 s.) before the curve, compared to novice drivers who did not reach target speed until 1.0 s. (SD = 2.3 s.) before the curve [t (17) = 2.40, P = 0.03]. Lane Positioning. Speed differences could be in-part a result of lane positioning. Let’s assume a driver is negotiating a curve to the right. If the driver starts at a position to the left of center of the lane and moves to the right of center of the lane at the apex of the curve, the driver could flatten the radius of the curve and be able to travel a greater speed, or maintain more control at the same speed. Table 15 shows that experienced drivers started left of center of the lane, moved to a position 1.5 foot left of center at two seconds before the curve and eased into a position on the inside of the curve two feet right of center when at the curve. These findings are similar to those by Bonneson et al. (2009) who indicated that the average lateral shift was three feet. The novice drivers had no real plan of attack. The average drivers maintained a position near the center of the lane and experienced the full radius of the road until the last second when they moved to the inside of the curve (1.7 feet right of the center of the lane). The lane position results suggest that experienced drivers not only approached at slower speeds, but were also in more control of the vehicle than novice drivers. By starting toward the outside of the curve, the sight line was also slightly lengthened for the experienced drivers. 108
Table 15: Average lane position of experienced and novice drivers when approaching a sharp curve right. (Distances refer to the position left (negative in parentheses) or right (positive) of the center of the lane.)
While the magnitude of the offset at the moderate curve left was not as great, the significant differences remained. Again, experienced drivers followed a path that took them from the outside of the curve (right of center of the lane on a curve left) to the inside of the curve at the apex. The novice drivers moved from right of center of the lane to a position near the center of the lane when at the apex of the curve. As can be seen in Table 15 and Table 16, experienced drivers selected significantly more optimal lane positions throughout many of the time bins as they approached a curve.
109
Table 16: Average lane position of experienced and novice drivers when approaching a moderate curve left. (Distances refer to the position left (negative in parentheses) or right (positive) of the center of the lane. Bold yellow signifies that experienced and novice drivers selected significantly different lane positions.)
4.7.1.2 Conditional Braking, Slowing and Crashes Glancing and Slowing to Target Speed. The conditional probability that a driver glanced, given that he or she had glanced when five to eight seconds before the curve, was computed. A comparison was then made between these conditional probabilities for experienced and novice drivers. Table 17 and Table 18 show the conditional outcomes of the eighteen drivers the first time they drove through the sharp curve. Many more experienced drivers made a glance to the far extent and slowed than did the novice drivers.
110
Table 17: Number of experienced and novice drivers that slowed to target speed after glancing to the far extent when five to eight seconds before a sharp curve right.
Of greatest interest is whether the experienced or novice drivers reduced speed more or less if they had glanced or not. The experienced drivers were three times more likely to make a glance to the far extent (10 vs. 3) and more than 50% more likely to slow to the target speed of 20 mph (14 vs. 9). If the driver glanced across the curve to the far extent, 90% of the experienced drivers slowed (9 out of 10), but more remarkably, the novice drivers that slowed increased to 67% (2 out of 3). If the novice driver did not make an anticipatory glance, the probability reduced to 47%. Likewise, only 63% of the experienced drivers that failed to make a glance to the far extent slowed to the target speed.
111
Table 18: Percentage of experienced and novice drivers that slowed if they had glanced to the far extent or had not glanced to the far extent. (Numbers in bold yellow indicate a significant difference.)
There are interesting comparisons to be made between the sharp curve right and the moderate curve left for the experienced drivers. The same percentage of experienced drivers glanced to the far extent at the curve left as at the curve right. The percentage that slowed to target speed was similar (78% at the sharp curve, 67% at the curve left). However, the experienced drivers’ probability of slowing, given that they glanced, was markedly different in the two curves. Of the experienced drivers that glanced to the far extent on the sharp curve right, 90% slowed to target speed; however, on the moderate curve left only 55% slowed to target speed (Table 19 and Table 20). A similar relationship can be seen with the novice drivers in that 67% slowed to target speed at the sharp curve and 44% slowed to target speed at the moderate curve left. While glancing across the curve appears to be predictive of speed loss on the sharp curve right, such does not appear to be the case for the moderate curve left for the experienced drivers. On the sharp curve right, 90% of the experienced drivers reached target speed given that they glanced whereas only 63% of the experienced drivers reached target speed given that they did not glance. However, on the moderate curve left,
112
only 55% of the experienced drivers reached target speed, given that they glanced, whereas fully 93% slowed given that they did not glance. On the other hand, novice drivers saw an improvement in the probability of slowing if they had glanced to the far extent for both the sharp curve right and the moderate curve left. Of those who glanced to the far extent on the sharp curve right, 67% slowed, while 47% slowed if they had not glanced. Of those who glanced to the far extent on the moderate curve left, 44% slowed, while 39% slowed if they had not glanced. These results show that novice drivers were more tied to processing information by looking at it compared to the experienced drivers who seemed to be exhibiting an ability to process the environment ambiently (See Table 19 and Table 20). Table 19: Number of experienced and novice drivers that slowed to target speed after glancing to the far extent when five to eight seconds before a moderate curve left.
113
Table 20: Percentage of experienced and novice drivers that slowed if they had glanced to the far extent or had not glanced to the far extent when approaching a curve left. (Numbers in bold yellow indicate significant differences.)
Conditional Outcomes in Crashes. Lastly, a reason for this research was to investigate reasons for crashes among novice and experienced drivers at curves, hopefully leading to methods for reducing such crashes. Blind scorers identified drivers that drove entirely off the road, out of the proper lane or struck another vehicle. All drivers that drove through the sharp curve for the first time, and who drove through the curve when the hazard was materialized were evaluated in this analysis. Recall that the materialized hazard here was an SUV parked along the right side of the road after the curve. The SUV protruded into the travel lane by 3 feet, which left adequate width to remain in the lane. The materialized hazard at the curve left was an oncoming car that crossed the center line by approximately 1.5 feet and returned to its lane. The driver experienced the head-on car driving through the peak of the curve. While there were no crashes at the curve to the left, there were eight crashes at the sharp curve to the right. Additional information can be learned by examining how each occurred.
114
As the crash statistics suggest, the most likely crash type was run-off-the-road [ROR] crashes. Novice drivers had five such incidents, while two experienced drivers drove off the road. Also, one novice driver struck a car at the curve. I considered both ROR incidents and struck vehicle incidents as crashes. Therefore, experienced drivers had two crashes and novices had six crashes. We can learn from each of these crashes. Of the experienced drivers, one glanced to the far extent, but both crashes were obviously speed related in that neither slowed to a speed less than 26 mph at any time. Of the novice drivers, two made a glance to the far extent. One of these drivers never slowed to a speed less than 35.1 mph. The other novice driver that glanced to the far extent also slowed to the target speed. The cause of this driver’s crash can be traced to poor lane keeping in that the driver started from a position 3.7 feet right of center and was 5.3 feet left of center of the lane when at the apex of the curve. The remaining four drivers that crashed failed to make a glance to the far extent and none of these drivers slowed to target speed. When there was not a crash, experienced drivers were still much more likely to both glanced to the far extent and slow to target speed than were the novice drivers (Experienced 9 out of 16 vs. Novices 1 out of 12). Also, of those who did not crash, experienced and novice drivers slowed 88% and 42% of the time. No driver that crashed had slowed to target speed before the crash.
115
Table 21: Percentage of drivers who crashed, slowed before crashing, slowed when not crashing or glanced and slowed without crashing. Numbers in bold yellow suggest a significant difference between experienced and novice drivers.
I compared the probability of a crash for those who slowed versus those who did not slow. Both experienced and novice drivers were much less likely to crash if they had slowed to the target speed. (See Table 22). Clearly, those who slowed to a target speed before the sharp curve were much less likely to be involved in a crash. Table 22: Comparison of the percentage of experienced and novice drivers that crashed based upon whether the driver slowed to the target speed or did not slow to the target speed.
4.7.2 Geometry: Intersections In the above analysis of novice and experienced drivers, the question was whether and when the two sets of drivers reached the target speed.
116
Recall that the target speed is
the current speed minus the speed loss in the previous second projected forward in time until the location of the intersection. In order to compute a driver’s target speed, one needs to know how the appropriate speed when the driver enters a given scenario or the speed loss necessary before arriving at an incident. The speed that needs to be lost at a curve has been studied relatively extensively. So, the target speed can easily be computed. However, the speed that needs to be lost at intersections has been studied less extensively. Thus, it can be more difficult to know a-priori what target to use as a threshold when comparing the two sets of drivers. The speed that needs to be lost from the posted speed before the intersection will depend on the geometry of the intersection, the traffic, and the built and natural environment. At Intersection 1, (Left turning truck), there is a large truck in the adjacent lane and a raised curbing to the right. According to the several authors (O’Leary, 2006; Lisle et al., 1980; Thompson et al., 1985), drivers reduced speeds by 1.0 to 2.0 mph when an obstacle was placed alongside the driver’s lane. These studies suggested a speed loss of approximately 1.5 mph without a sightline obstruction when entering an intersection. I assumed drivers would reduce speed by at least twice the amount. Three mph was selected by extrapolating from the literature as well as considering how speed is lost. If a driver coasts, average rolling resistance would result in a speed loss of approximately 0.3 mph to 2.2 mph in each second (Warner et al., 1983). Thus a speed loss of greater than 3 mph would more likely be a result of a conscious choice by the driver. I predicted at each one second interval before Intersection 1 based on current speed and speed loss in the preceding second, the percentage of the novice and experienced drivers who would 117
have reached a target speed of 37 mph when they entered the intersection (a 3.0 mph speed reduction from the posted speed limit.) The geometry and traffic is different at Intersection 3. When approaching a left turn at a busy intersection, drivers should be prepared to stop. While a stop might not be required, certainly, it could be necessary. Thus, drivers should be prepared to be ontarget to slow to a stop until they deem it is not necessary to do so. The percentage of novice and experienced drivers who reached a target speed of zero mph when approaching a left turn at a busy intersection was compared at each second prior to entering the intersection. This intersection is referred to as a busy intersection in that there is traffic approaching on all three legs. Target speed, glance behaviors and target speed of the experienced and novice drivers were compared at each one second interval as they approached the intersection. First, the aggregate glance, braking and slowing behaviors were compared. Second, these behaviors were compared conditional on an anticipatory glance being made (or not made). Each group was compared when driving through the intersection once when there was not a materialized hazard and once when there was a materialized hazard. Recall that the materialized hazard at the left turning truck was a previously obstructed opposite direction car that made a left turn into the path of the driver (Intersection 1). The intruding vehicle was previously obstructed by a truck in the left lane. In the last two seconds, the left turning car could be seen by drivers after passing the stopped leftturning truck. At the busy intersection, the materialized hazard was a car entering the intersection in the left lane on the side road to the left of the driver (Intersection 3). 118
4.7.2.1 Aggregate Glance, Braking and Slowing Behaviors Glances. When the drivers approached a curve, the glances toward the threatening area were followed by slowing. When approaching an obstructed intersection or left turn across path situation, slowing had to precede the measureable glance. Here, glancing was confirmation that the slowing behaviors related to anticipatory actions rather than random behaviors. Clearly, there must have been some anticipatory actions that occurred before slowing began, but such glance behaviors could not be recorded with the available equipment. Two intersection configurations were addressed. At a busy intersection the driver must slow for a left turn (Intersection 3), and at the intersection with a view obstruction due to a left turning truck the driver travels straight through (Intersection 1). Because the intersection movements were much different, glances had to be recorded differently as well. When approaching a left turning truck in the left lane, the sight line toward traffic within the intersection is obstructed (Intersection 1). When approaching the left turning truck, a driver cannot see in front of the truck until two to five seconds before the intersection. A glance to the right edge of the stopped truck is not strong evidence that the driver was anticipating traffic in front of the truck when as far back for six to ten seconds. However, if a driver made a glance toward the front of the truck as he or she was passing it, it suggests that the glance was anticipatory. Therefore, for the intersection with the left-turning truck, I reported the percentage that glance in each second.
119
At the busy intersection, the potential hazard could emerge from any of the three remaining legs of the four-way intersection (Intersection 3). Imagine traveling toward the intersection. First drivers need to assure that traffic on the side roads has stopped or will not enter the intersection. With side road glances, I was interested in when they occurred as the driver approached the intersection. Therefore, side road glances were reported as the cumulative percentage of drivers who glanced to the side roads in each second. I use the plural, “roads”, meaning that a driver must glance toward both the left and right legs of the intersection. After assuring that the side road traffic is neutralized, a driver must then assure that the turn is safe. With an oncoming vehicle that may be traveling fast or slow, a driver should make a secondary glance. A secondary glance is a glance toward oncoming traffic at the moment or immediately after the turn starts. The secondary glance offers the driver a last chance to abort the turn. Secondary glances were reported as a percentage of drivers that made the glance. Lastly, the percentage of drivers that both made the necessary side road glances and the secondary glance were reported. At Intersection 1, experienced drivers were much more likely to make glances toward the near extent. The differences were quite noticeable even when as far back as nine seconds from the intersection. (See Figure 20).
120
Figure 20: Percentage of experienced and novice drivers that glanced to the near extent left in each one-second periods when approaching an intersection with the left lane obstructed by a turning truck. (* Asterisk signifies that experienced drivers were much more likely to make a glance to the near extent in that second.)
The statistical significance of the difference between the percentage of experienced and novice drivers who glanced to the near extent for each second before the intersection is shown in Table 23. At nearly every one-second bin, experienced drivers were much more likely to glance toward the near extent of the sightline (Recall that the near extent is at the right front of the stopped truck.) When three, five, and nine seconds before the intersection, experienced drivers were significantly more likely to make a glance to the near extent.
121
Table 23: Percentage glances toward the near extent in each second as experienced and novice drivers approached an intersection that was obstructed by a left turning truck. (Numbers in bold yellow indicate significant differences.)
At Intersection 3, when approaching a busy intersection, the differences in glances between the experienced and novice drivers were not as obvious until the moments immediately before turning. Experienced drivers were more likely to make glances to both side roads in the two seconds before turning. Also, experienced drivers were more likely to make a secondary glance toward the oncoming traffic before turning. Logically, experienced drivers were also more likely to both glance toward the side roads, and make a secondary glance ahead before turning. In Figure 21 the asterisks refer to times when experienced drivers were significantly more likely to have completed glancing toward left and right side roads. 122
Each of the novice and experienced drivers that completed both side road glances also made a secondary glance. The result was that the percentage drivers that made both glances equal the percentage that made secondary glances for both groups.
Figure 21: Cumulative percentage of drivers that completed both glances to the left and right side road in the nine seconds before turning. Also, the percentage of drivers that made a secondary glance toward oncoming traffic before turning, and lastly, the percentage of drivers that both completed the side road glances and the secondary glance before turning. (* The asterisk signifies that experienced drivers were significantly more likely to make a glance toward the side road in that second.)
Novice drivers (33%) were more likely to have glanced toward both the left and right side roads than experienced drivers (24%) when six seconds from the intersection. However, in the next few seconds, nearly two-thirds of the experienced drivers completed the side road glances, while only an additional 17% of the novice drivers completed the 123
glances in the last six seconds. Significantly more experienced drivers (88%) completed the glances to both side roads in the last two seconds before turning than did the novice drivers (50%). The difference here is not related to if there was a glance, but more so when the glance occurred. Experienced drivers were also much more likely to make a secondary glance toward oncoming traffic before turning. Not surprisingly, experienced drivers were more than 1.5 times more likely to have glanced toward all three legs of the intersection before starting the turn. All three legs refers to a driver who looked to both side roads and also made the secondary glance ahead before turning into the opposite lane. (See Table 24).
124
Table 24: The cumulative percentage of ACT trained and placebo trained drivers that completed both side road glances in each second (0 – 9 s) before the curve. Also, the percentage of ACT trained and placebo trained drivers that made secondary glances, and finally (bottom row), the percentage drivers that completed the necessary side road glances and a secondary glance before turning (Numbers in bold yellow indicate that experienced drivers were significantly more likely to make a glance toward in that second.)
. Speed Loss. When approaching the obstructed intersection (left turning truck), experienced drivers made significantly more near extent glances at seven and three seconds and we see a corresponding decrease in the speed profile of the experienced drivers at six and three seconds before the intersection (Intersection 1). In Figure 22, the asterisks refer to moments when significantly more experienced drivers had slowed to the target speed of 37 mph (i.e., their predicted speed at the intersection based on their
125
current speed and their speed loss during the previous second was 37 mph). Of note is that the speed choice of the experienced and novice drivers did not differ initially. Only as the two groups approached the intersection did the speed between the two groups begin to differ.
Figure 22: Average speed of the experienced and novice drivers during each second when approaching an intersection with an obstructed view due to a left turning truck in the adjacent left lane. (* The asterisk suggests that significantly more experienced drivers slowed to the target speed of 37 mph in that second.)
The actual percentages at the straight movement intersection of experienced and novice drivers who reduced their speed to a target speed below 37 mph earlier are displayed in Table 25. When two to four seconds before the intersection, more than half the experienced drivers had slowed to a target speed of less than 37 mph, yet only 29% of novice drivers ever slowed to a target speed of 37 mph.
126
Table 25: Comparison of experienced and novice drivers’ Glances within Each 1-second epoch when approaching an intersection with its view obstructed by a left-turning truck. (Numbers in bold yellow indicate significant differences.)
At the left turning truck scenario, cognition, or anticipation was necessary by the drivers to recognize a need to slow. At a busy intersection where the goal was to turn left, the speed differences between the experienced and novice drivers were not as noticeable (Intersection 3). Again, experienced drivers slowed more than novice drivers and also started into the turn at a lower speed. (See Figure 23).
127
Figure 23: Average speed of the experienced and novice drivers during each second when approaching a busy four-way intersection.
If you recall, the two intersections had much different target speeds. The target speed of 37 mph (3 mph less than the posted speed limit) was selected because it is the point at which speed loss becomes a conscious decision rather than an arbitrary variance. Warner et al (1983) showed that drivers that coast and experienced rolling or engine resistance typically experienced a 0.02 to 0.10 G deceleration, which equates to a speed loss of 0.5 to 2.5 mph per second. Therefore, a three mph per second speed loss indicates that the driver reduced his or her speed by more than a routine amount. A target speed of zero was selected for the busy intersection. Although stopping might not be necessary, a driver does not know that beforehand and should approach the intersection as if stopping
128
were necessary. Also recall that the target speed is the current speed minus the speed loss in the previous second projected forward in time until the location of the intersection. When within six seconds of a busy intersection, experienced drivers were approximately 9% more likely to slow to a target speed of 0 mph. When seven to nine seconds before the intersection, more novice drivers slowed to the target speed of 0 mph than experienced drivers, but again, there was no follow up on that early slowing in the form of additional glances or significant increases in slowing that would show evidence of a mindful anticipatory tactical action. (See Table 26). Table 26: Cumulative percentage of experienced and novice drivers to slow to a target speed of zero mph in the nine seconds before arriving at a busy intersection. (Numbers in bold yellow indicate significant differences.)
129
At the intersection where the truck obstructed the view, experienced drivers slowed to target speed significantly earlier. Experienced drivers slowed to the target speed of 37 mph and average of 4.6 s. (SD = 4.0 s.) before the intersection while novice drivers did not slow to target speed until 1.0 s. (SD = 2.3 s.). [t (35) = 3.15, P = 0.00]. At the busy intersection, experienced drivers slowed to target speed earlier, but not significantly so. Experienced drivers slowed to a target speed of 0 mph when 3.2 s. (SD = 3.7 s.) from the intersection compared to 2.8 s. (SD = 4.1 s.) for the novice drivers [t (17) = 0.22; P=0.83]. While the difference is unremarkable, if we compare the groups with an 18 mph target speed, experienced and novice drivers slowed to the target speed nearly identically. The average turning speed for drivers in the turn was 18 mph. This result suggests that the novice drivers likely slowed for the purpose of the turn without much thought that additional slowing might be necessary. Lane Position. Some drivers might choose to move within the lane and offer additional buffer space as a way to mitigate the need to slow as much. At the obstructed intersection (where there was a left turning truck), the truck is in the adjacent lane to the immediate left. Moving to the right would allow for a greater sightline to the front of the truck, and would add buffer space between objects that may emerge from the front of the truck. Initially, when eight and nine seconds from the intersection, experienced drivers began significantly left of center in the lane compared to novice drivers. As the drivers passed the truck (2 to 3 s before the intersection), experienced drivers moved to a position right of the novice drivers, but the amplitude of the difference appears to be
130
unimpressive, only 0.25 ft. It would be difficult to claim that this difference changes the likelihood of a crash. (See Table 27). Table 27: Average lane position of experienced and novice drivers when approaching an intersection with a left turning truck obstructing the view. (Distances refer to the position left (negative) or right (positive) of the center of the lane. Numbers in bold yellow indicate that experienced and novice drivers selected significantly different lane positions.)
At the busy intersection, there were significant differences, but again, the amplitude of the differences was so small, so it would be difficult to claim these results are notable. What I can say is that drivers generally moved approximately 1-foot left of center (See Table 28). The only significant difference of note is that the novice drivers began to turn earlier than did the experienced drivers. The novice drivers were more likely to make a longer arching turn that left them exposed to conflicting traffic longer. At the start of the intersection, the average novice driver was 4 feet left of center. Given a vehicle of average width, that places the vehicle over the center line of the road. The lane position results means that the experienced and novice drivers remained approximately 1-foot left of center until making the turn and the novice drivers began the turn earlier than did the experienced drivers. Instead, the experienced drivers moved 131
farther into the intersection before turning and apparently used that extra distance (and time) to make more glances. Table 28: Average lane position of experienced and novice drivers when approaching a busy four-way intersection. (Distances refer to the position left (negative) or right (positive) of the center of the lane. Numbers in bold yellow indicate that experienced and novice drivers selected significantly different lane positions.)
Post test data analysis revealed that five novice drivers and two experienced drivers crossed the centerline before entering the intersection (The elongation of the closest edge of the side roads). Further, of these drivers who started the turn early, only one experienced driver and one novice driver made glances to all three legs of the intersection before the turn. Comparing these data with the percentages that glanced to all three legs (71% vs. 44%, for the experienced and novice drivers respectively) suggests that moving farther into the intersection before turning was associated with a greater likelihood of glancing to all three legs of the intersection. 4.7.2.2 Conditional Braking and Slowing and Crashes Consider next Intersection 1, the left-turning truck intersection. Both experienced and novice drivers were nearly 20 percentage points more likely to make anticipatory glances at this intersection that was obstructed by the left turning truck 132
(Intersection 1) than they were the busy intersection (Intersection 3). Moreover, experienced drivers were nearly significantly more likely to make glances to the near extent than were novice drivers (89% compared to 62%). (See Table 29). Table 29: Number of experienced and novice drivers that slowed to target speed and glanced to the far extent when two to five seconds before an intersection with a left turning truck.
When comparing the conditional probabilities, the experienced drivers were slightly more likely to slow if they had glanced (69%) and less likely to slow if they did not make a glance to the near extent (50%). However, the novice drivers were actually less likely to slow if they had glanced (36%) and more likely to slow if they did not glance (71%). These counter-intuitive results might suggest that the glances by the novice drivers were not mindful or anticipatory. Given a glance, experienced drivers were 1.92 times more likely to slow than were the novice drivers. This difference was nearly significant. (See Table 30).
133
Table 30: Percentage of experienced and novice drivers that slowed if they had glanced to the near extent or had not glanced to the near extent when approaching an intersection that was obstructed by a left turning truck.
At the busy intersection, a comparison was made of the percentage of novice and experienced drivers who slowed, based upon whether he or she glanced to both side roads and toward oncoming traffic before entering the opposite lane (Intersection 3). The conditional outcomes of the eighteen drivers when confronted with the materialized hazard were compared. Experienced drivers were more likely to slow (62% versus 50%) and more likely to glance (67% versus 44%) than novice drivers (see Table 31).
134
Table 31: Number of experienced and novice drivers that slowed to target speed of 0-mph and glanced toward all three legs of the intersection before turning left.
Of greatest interest is whether the experienced or novice drivers reduced speed more or less if they had glanced or not. Both groups were similarly likely to slow if they had made glances to all three legs of the intersection (Experienced 67%; Novices 63%). If they had not glanced, half the experienced drivers slowed compared to only 40% of the novice drivers (see Table 32). Table 32: Percentage of experienced and novice drivers that slowed if they had glanced to the far extent or had not glanced to the far extent.
135
Conditional Outcomes in Crashes. Lastly, the reason for this research was to investigate reasons for crashes among novice and experienced drivers at high crash risk locations. Blind scorers identified drivers that struck another vehicle while negotiating through these intersections. No driver crashed at the busy intersection. However, several drivers crashed at the intersection where the truck was making a left turn. Recall that the materialized hazard here was a car turning left across the path of the driver. The view of the car had previously been obstructed by a large truck that was waiting to turn left. Novice drivers were involved in twelve crashes, out of the possible 18 drivers. Six experienced drivers also crashed. Of those who did not slow, experienced and novice drivers crashed 83% and 100% of the time. Yet for those who slowed, experienced drivers crashed only 8% and novices crashed 33%. Clearly, the experienced drivers who slowed to a target speed before the intersection were much less likely to be involved in a crash. The crash rate of the novice drivers did not seem to be tied to glances, suggesting that the glances were not anticipatory. Also, although two novice drivers (out of four that glanced and slowed) glanced and slowed before the crash, both were traveling well above target speed only one second before the intersection and slowed in an emergency manner (hard last second braking), not an anticipatory manner (gradual slowing earlier in the event). (Table 33).
136
Table 33: Percentage of drivers who slowed before crashing, or did not slow before crashing.
I also compared within the experienced and within the novice driver groups the probability of a crash for those who slowed versus those who did not slow. Both experienced and novice drivers were significantly less likely to crash if they had slowed to the target speed. Table 34: Comparison of the percentage of experienced and novice drivers that crashed based upon whether the driver slowed to the target speed or did not slow to the target speed. (Numbers in bold yellow indicate significant differences.)
4.7.3 Geometry: Straight Segments Similar to the intersection arrangements, the target speed thresholds were based upon the research related to speed loss when approaching roadside hazards (O’Leary, 2006; Lisle et al., 1980; Thompson et al., 1985) as well as the coasting speed loss research (Warner et al., 1983). When discussing speed loss on straight segments, the 137
selection of target speeds was governed more by Straight Segment 3 than Straight Segment 1, so forgive me as I take the scenarios out of order for a moment. When approaching the roadside truck (right) and pedestrian (left) (Straight Segment 3), the initial speed of the drivers was usually no more than 26 mph when 12 seconds from the truck (Straight Segment 3). A three mph speed loss would make the target speed 23 mph, which is the threshold that was used as a comparison between groups of drivers. While the initial speed was greater at the bus blocking crosswalk, nearly half of all drivers reduced speeds to 23 mph (Straight Segment 1). For this reason and for consistency, a 23 mph threshold was used as the speed comparison at the bus blocking crosswalk scenario as well (Straight Segment 1). Using these target speed thresholds as a benchmark, the speeds and speed loss (to compute the target speed) along with the glance behaviors of the experienced and novice drivers were compared at each second as they approached the critical point. First, the aggregate glance, braking and slowing behaviors were compared. Second, these behaviors were compared conditional on an anticipatory glance being made (or not made). 4.7.3.1 Aggregate Glance, Braking and Slowing Behaviors Glances. The glancing results here for the bus blocking crosswalk scenario were similar to the left turning truck at the intersection scenario. Other than the similarity of a large vehicle in the adjacent lane, the glance patterns are also similar. Recall that at the intersection with the left-turning truck, experienced drivers had significantly more glances toward the near extent at three, seven and nine seconds before the intersection. Somewhat remarkably, the greatest differences between the experienced and novice 138
drivers at the bus blocking crosswalk scenario were also at three, seven and nine seconds before the crosswalk where 31%, 25%, and 25% more experienced drivers glanced toward the near extent than did the novice drivers.
Figure 24: Cumulative percentage of experienced and novice drivers that made a glance toward the near extent in each second when approaching a bus stopped and partially blocking the view of a crosswalk.
The difference in the likelihood of a glance to the near extent between the experienced and novice drivers did not reach significance. Yet, in Table 35 we can see that on average there was a 20 percentage point difference between the experienced drivers and novice drivers glancing toward the near extent at each of the above one second epochs (3, 7 and 9 s).
139
Table 35: Percentage of experienced and novice drivers that glanced toward the near extent when approaching a bus stopped in front of a crosswalk.
Next, consider Straight Segment 3. When approaching a roadside pedestrian on the left and perpendicular truck on the right, almost 25% fewer novice drivers had glanced toward both the pedestrian (left) and truck (right) in the two to six seconds before arriving at the location of the truck. (See Figure 25 and Table 36).
140
Figure 25: Cumulative percentage of experienced and novice drivers that glanced toward both a roadside pedestrian to the left and perpendicular truck to the right in each second when approaching the truck.
The difference in the likelihood of glances between the experienced and novice drivers was not significant yet the differences are noticeable. In the time from four to seven seconds before the truck, experienced drivers were nearly half again more likely to complete both the glances to the left and right.
141
Table 36: Cumulative percentage that glanced both toward a pedestrian along the roadside left, and a truck parked perpendicularly along the roadside right before passing the truck.
Speed Loss. In both straight segment scenarios, experienced and novice drivers had already slowed to speeds well below the posted speed limit of 40 mph. A likely reason for this is that there were cues to slow when much farther back than ten seconds from the incident locations. In both scenarios, the drivers slowed to a minimum speed and then accelerated to a speed near 25 mph when passing the incident area. First, consider Straight Segment 1. When approaching the bus stopped at the crosswalk, the experienced drivers had slowed to a speed of less than 35 mph by the time they were ten seconds from the crosswalk (Straight Segment 1). At the same time, the novice drivers were traveling 37 mph. The target speed was 23 mph. Experienced
142
drivers had slowed to an average speed near 21 mph when three seconds from the crosswalk, well below the target speed. Also, at three seconds before the crosswalk, the novice drivers were traveling 25 mph, still above the target speed.
Figure 26: Average speed of experienced and novice drivers when approaching a bus stopped in front of a mid-block crosswalk.
When approaching the bus, drivers also responded to the presence of a lead vehicle. To differentiate when the slowing occurred, I analyzed the percentage of experienced and novice drivers that slowed to the target speed of 23 mph in each second while approaching the crosswalk. While the speed differences are not as pronounced as was the case at the intersection with the left-turning truck, or at curves, we can still see that the difference between the percentage of novice and experienced drivers who slowed
143
to a target speed of 23 mph between four to six seconds before the crosswalk varied between 8 and 17 percentage points. Table 37: Percentage experienced and novice drivers that slowed to a target speed of 23 mph in each second when approaching a bus stopped at a crosswalk.
Next consider Straight Segment 3. The bus stopped near a crosswalk creates issues that are both obvious and latent. The presence of the lead vehicle and bus are obvious to an approaching driver. However, the possibility of a pedestrian emerging from in front of the bus is a latent hazard. At the bus-crosswalk scenario, there appears to be a difference in the speed choice between the experienced and novice drivers, although that relationship was not significant. When approaching two roadside obstacles, the pedestrian to the left and truck to the right, there were no hidden (latent) hazards. The
144
issue when approaching the roadside obstacles was whether the driver prepared for the possible intrusion of either obstacle or the sudden slowing of the lead vehicle. At the roadside obstacle situation, experienced drivers and novice drivers slowed nearly the same way. The speed choices of the experienced and novice drivers were very similar in Figure 27.
Figure 27: Average speed of experienced and novice drivers when approaching a roadside pedestrian to the left and truck to the right.
When approaching the roadside obstacles, the speed choices between the experienced and novice drivers were very similar, as was the percentage that slowed to target speed. At no point did the two groups differ by more than 6 percentage points and
145
that was when the drivers were six to ten seconds before the location of the truck. (See Table 38). Table 38: Cumulative percentage to slow to a target speed of 23 mph when approaching a roadside pedestrian and truck.
When approaching the bus stopped in front of the crosswalk, experienced drivers slowed to target speed only slightly earlier but the difference was not significant (Experienced 10.5 s. (SD = 4.0 s.); Novice 9.9 s. (SD = 4.3 s.). [t (35) = 0.07, P = 0.94). When approaching the roadside obstacles, experienced drivers and novice drivers slowed to target speed at similar times. (Experienced 8.2 s. (SD = 5.5 s.); Novice 8.1 s. (SD = 5.6 s.). [t (35) = 0.05, P = 0.95] Lane Position. When approaching the stopped bus that was to the driver’s right, being farther left would offer greater buffer space and sightline for what might lie ahead of the bus. Experienced drivers maintained a position slightly more left of the lane 146
throughout the approach toward the bus, but both the experienced and novice drivers drove left of center of the lane. The average position was one to two feet (0.3 to 0.6 m) left of center of the lane. Initially when approaching the area, the lead vehicle changes lanes from the right lane to the left lane. If a driver is following close enough, conceding distance offers more buffer space earlier. When the driver approaches the bus, it is immediately to the right of the driver. Table 39 shows that drivers started left of center and moved farther left by about two feet when passing the bus. The novice drivers also started near the center of the lane when nine seconds back but moved only about 1.3 feet left of the center of the lane when passing the bus. Table 39: Average lane position of experienced and novice drivers when approaching a bus stopped at a crosswalk. (Distances refer to the position left (negative in parentheses) or right (positive) of the center of the lane.)
When approaching the roadside obstacles, recall that the truck protrudes into the road on the right and there is a roadside pedestrian left as well as an oncoming vehicle coming from the opposite direction. Experienced drivers maintained a position over a 147
foot and a half left of center when approaching the truck, but when passing the truck, both experienced and novice drivers were nearly two feet left of the center of the lane. Here, the difference was that the experienced drivers did not wait until the last second to move left as the novice drivers did. (See Table 40). Table 40: Average lane position relative to the center of the lane when approaching a roadside truck to the right and pedestrian to the roadside left. (Numbers in bold yellow indicate significant differences.)
4.7.3.2 Conditional Braking and Slowing and Crashes The conditional probability that a driver slowed given that the driver had glanced toward the near extent when passing the bus can tell us if there was anticipation associated with the drivers’ actions in Straight Segment 1. First, consider the unconditional probabilities. When within two to four seconds of the crosswalk, a time when a driver is first able to look in front of the bus, 83% of the experienced drivers glanced toward the near extent compared to only 64% of the novice drivers. Novice drivers (69.4%) and experienced drivers (72.2%) slowed in similar amounts.
148
Table 41: Number of experienced and novice drivers that slowed to 23 mph based upon whether they glanced to the near extent two to four seconds before the crosswalk.
Both groups slowed more if they did not glance toward the near extent (See Table 41 and Table 42). For experienced drivers, 70% slowed given that they glanced, but 84% slowed given that they did not glance. For novice drivers, 61% slowed given that they glanced but 85% slowed given that they did not glance. This might be due to the presence of the lead vehicle. As noted above, while a similar number of experienced and novice drivers slowed, 20% more of the experienced drivers glanced to the near extent. If the reason that fewer novice drivers glanced to the near extent was the lead vehicle, this result suggests that more experienced drivers were able to process two potential hazards simultaneously (the possible pedestrian and the lead vehicle) than were the novice drivers (just the lead vehicle). A logical way to reduce the need for having to process two hazards at once would be to slow so each could be processed serially.
149
Table 42: Percent of experienced and novice drivers that glanced to the near extent five to eight seconds before the crosswalk and slowed to a target speed of 23 mph as they approached a bus parked in front of a crosswalk.
At Straight Segment 3, a roadside pedestrian and truck scenario, the exact same number of experienced and novice drivers slowed to a speed of 23 mph. Here, the hazards are not latent, but directly ahead, so anticipatory slowing is not surprising. However, there is still the need for drivers to process the roadside pedestrian and truck, and 16% more experienced drivers glanced toward each before arriving at that location than did the novice drivers. At the roadside obstacles scenario, the experienced drivers slowed similarly if they glanced or not (72% versus 73%; Table 43). The novice drivers were less likely to slow if they glanced to both the pedestrian and truck as if those glances were distracting rather than anticipatory or informative. Of the novice drivers, 88% slowed if they did not make the glances to the pedestrian and truck, but only 58% slowed if they did (Table 44).
150
Table 43: Number of experienced and novice drivers that slowed to 23 mph and glanced to roadside pedestrian and truck two to four seconds before obstacle.
Table 44: Percent of experienced and novice drivers that glanced and slowed as they approached a roadside pedestrian and truck
Conditional Probability related to Crashes. When approaching the roadside truck and pedestrian, no drivers crashed, but six drivers crashed at the location where the bus was stopped in front of the crosswalk. Of the six that crashed, there was one experienced
151
driver and five novice drivers. Only one experienced driver failed to slow and only one experienced driver crashed (a rear-ender into the lead vehicle). Only one experienced driver crashed and that driver also slowed and glanced. The cause of that crash was a failure to anticipate the sudden slowing by the lead vehicle and following too close. Of the crashes by the novice drivers, all of the drivers made a glance, and three of the five also slowed. Among the novices, the theme of the crashes was failure to exhibit anticipatory slowing. While three of five reached target speed, two of those did so due to emergency braking before the crash, not anticipatory slowing to avoid the crash. (See Table 45). Table 45: Percentage of experienced and novice drivers that slowed to a speed of 23 mph based upon whether they crashed or not.(Numbers in bold yellow indicate significant differences.)
The relation between speed and crashing is clear cut for the novice drivers. The likelihood that a novice driver crashed, given that he or she slowed, was 8%, whereas the percentage that a novice driver crashed, given that he or she did not slow, was 67% (Table 46). For experienced drivers, the probability that the experienced driver crashed given that he or she slowed as also very small (8%), but so too was the probability that the novice driver crashed given that he or she did not slow (0%). 152
Table 46: Percentage of drivers that crashed based upon whether they slowed to a speed of 23 mph or not. (Numbers in bold yellow indicate significant differences.)
4. 8 Discussion Overall, experienced drivers made more glances towards the appropriate area(s), reached the target speed sooner, and maintained a safer position within the lane. When the drivers made the appropriate glance, experienced drivers were more likely to slow to the target speed at all six scenarios. When we consider the results at all the scenarios, nearly twice as many experienced drivers slowed to target speed and made the glance (78 vs. 40). There was not a significant difference in the number of experienced and novice drivers that slowed to target speed at the straight road segments. However, if we look at only the curves and intersections (The locations associated with the greatest crash risk), many more experienced drivers slowed to target speed than did novice drivers (61 vs. 41). The probability that the experienced drivers crashed, given that they slowed, is less in all three of the scenarios where crashes occurred. Glances are tied to slower speeds, which in turn are tied to reductions in crashes. Next consider Figure 28. The above differences speak to a tactical plan of action where anticipatory glances relate the mitigating behaviors that are not prevalent with novice drivers. The areas highlighted in yellow refer to times when experienced drivers were ½ a foot or more in a more optimal position than the novice drivers. Optimal 153
position would be a position that offers the greatest buffer space between the driver and surrounding obstacles. At curves, an optimal position would be to the outside of the curve (left of the center of the lane on a curve right), and to move to the inside of the curve (right of center of the lane at the curve). As an example, referring to Figure 28, where there is yellow in the 9s, 8s, 2 s, and 0 s columns, this means that when we compare the performances of the experienced and novice drivers at all six scenarios, the experienced drivers maintained a position that was at least ½ foot better than did the novice drivers at those times. Red refers to times when experienced drivers were at least an average of 20% more likely to be slowing than novice drivers at the six scenarios. There are two shades of green; dark green refers to all six scenarios in the research. The light green refers only to the three scenarios where crashes occurred (Sharp curve right, left turning truck at intersection, and a bus at a crosswalk). At times represented by the dark green experienced drivers glanced toward the potential hazard or extent of the sight line at least an average of 20% more often than novice drivers. Light green refers to times when at least 20% more experienced drivers made glances to the far extent at the scenarios where a crash occurred. 9 s.
8 s.
7 s.
6 s.
5 s.
4 s.
3 s.
2 s.
1 s.
0 s.
Glance Slow Lane Pos. Figure 28: The highlighted regions represent the times where experienced drivers outperformed the novice drivers at the three locations where drivers crashed.
154
When comparing the experienced and novice drivers’ performances at the three locations where no crashes occurred, there was not as robust a difference, although experienced drivers continued to perform better overall. At the three sites where crashes occurred, experienced drivers were more likely to glance to the extent much earlier before the incident than novice drivers.
155
CHAPTER 5 EXPERIMENT 2: EVALUATION OF THE RISK MITIGATING BEHAVIORS OF TRAINED AND UNTRAINED NOVICE DRIVERS USING ANTICIPATIONCONTROL & TERMINATION [ACT]
The information gained in Experiment 1 was used to develop a PC-based program to train novice drivers how to better neutralize hazards. Included in the training were nine scenarios associated with the eight most common crash types in which teens age 1618 were involved in serious crashes (NMVCCS, 2008). These eight crash types account for more than 60% of all serious teen crashes. One group of novice drivers was asked to complete the ACT training program. The ACT training program consisted of a practice module, a pre-test module, a training module, and a post-test module. A second group was exposed to a placebo training program. The participants in this group were administered the ACT post-test module. After both groups had been trained and completed the ACT post-test module, the effects of the training programs were evaluated on the driving simulator.
5.1 Development of ACT Training Program Experiment 1 and the previous literature formed the basis from which a PC-based computer program was developed. In Experiment 1, exemplary drivers were compared to novice drivers at scenarios that are known to be associated with teen crashes. The specific second-by-second behaviors of each age group were compared. Furthermore, the
156
conditional probabilities that a driver mitigated a hazard, given that they engaged in appropriate glance behaviors, were analyzed. From Experiment 1 it was clear that a training program is needed that ties hazard anticipation, specifically anticipatory glances, with subsequent slowing (risk mitigation) for curves and other locations where hazards need to be mitigated, and ties hazard anticipation, specifically slowing, with subsequent preparatory glances (risk mitigation) for intersections. Part of the training task has already been established with RAPT-3, a PC-based program that teaches novice drivers how to better anticipate hazards. Also FOCAL was developed to teach drivers to better maintain attention toward the roadway ahead. A new program, ACT, which stands for Anticipate-Control-Terminate, is an extension of the RAPT research (Pollatsek et al., 2006) and is a supplement to these programs. 5.1.1 ACT: Basic Elements of the Training Program -- Modules and Responses The ACT training program consisted of a practice module, a pretest module, a training module, a mediation module, and a posttest module. Each of the five modules showed participants a series of snapshots leading up to a traffic scene. The eight most common crash configurations were included in the nine scenarios (the same nine scenarios) administered each of the pre-test, training, mediation, and post-test modules. Scenarios: What the Participant Sees. The ACT modules were designed to have a format similar to the RAPT modules so trainees could easily transition from one scenario to the next. For each scenario, ACT, like RAPT, shows the user a series of snapshots played at real time speed. For instance, if the vehicle is traveling 30 mph (44 feet per second), each snapshot would be taken approximately 44 feet apart. When the snapshots are shown at a rate of one-per-second, the illusion would be that the observer is 157
moving forward at 44 feet per second, or 30 mph. RAPT allows the observer to see each snapshot for three seconds. ACT users are given an opportunity to practice before the pretest to become accustomed to the faster frame rate (the practice module). During initial testing, each slide was shown for three seconds, as is the case with the RAPT program. However, a one-frame-per-second speed was selected for five reasons. 1. Hazard anticipation, not reaction time is being taught. Hence, users should be anticipating rather than reacting. 2. Users enjoyed the faster pace in that it was more of a challenge, which avoided boredom. Every user reported the one-frame per second method to be more of a challenge but more enjoyable. 3. The faster pace shortened the length of the training. 4. Response choices did not change significantly with the faster pace. In the ten users that were pilot tested, there was no noticeable difference between those who had three seconds for each slide and those who had only one second per slide. 5. One of the primary purposes of this training is to teach an understanding of speed choice. Showing the approach to a potential hazard at actual speed is more representative of real life and depicts vehicle speed. When played at a slower rate, there is no sense of speed.
Scenarios: How the Participant Responds. In all modules, participants could make one of four responses. Specifically, the tasks the participants were asked to perform included: (1) hovering a mouse over an area where a threat could potentially materialize and clicking on the area; (2) indicating where (time-wise) a driver should slow, accelerate or brake; (3) responding that the horn should be sounded; and (4) indicating where within the width of the road a driver should position the vehicle for each of the nine scenarios. The participants were told that they needed to identify
158
potential threats by indicating where they would be glancing with the mouse. The forward view presented in each snapshot was always straight ahead; the steering options and horn were near a pictured steering wheel in the lower center of the screen (Figure 29). The foot pedals positions (brake, accelerator, location off the brake and accelerator) were depicted at the bottom right of the screen and were selected with the mouse.
159
Figure 29: Screen shot from the ACT training program. (The pedals at the bottom right represent the throttle - green, “off-throttle” – small gray pedal, and brake –large gray pedal. The five arrows represent positions left and right of lane-center. The horn icon shows where the horn is located and users may click anywhere in the forward view to indicate where they wish to glance.)
The drivers were taught not only to anticipate hazards in ACT much as they did in RAPT (by moving the mouse over an area and clicking where threats could materialize), but to also show how they would mitigate the hazard with slowing, steering or horn use. The RAPT training program has been described elsewhere (Pollatsek et al., 2006). The mitigation tasks will be described here. Specifically, after any selection, users were given feedback in the form of a stimulus to an appropriate response (e.g., the sound of accelerating or braking after the participant selected the accelerator or brake) and the 160
pedal or arrow changed color. The appropriate arrow changed color after a steering selection. A crosshair appeared in the forward view every time the user selected a glance location. Finally, the horn was sounded after it was selected. 5.1.2 ACT: Practice Module In the practice module, participants were first shown scenarios with the snapshots appearing at the rate of one frame per second. Once accustomed to seeing sequences of snapshots at the rate of one per second, participants were introduced to the use of foot pedal position selectors. The foot pedal position selectors allowed the user to indicate when they would take their foot off the throttle, put their foot on the brake, or accelerate. After practicing with the foot pedal position selectors, users were introduced to the lane position (steering) selectors and allowed to practice again. As with RAPT, the user also had the option of making anticipatory glances (mouse clicks on the forward scene) and sounding the horn. 5.1.3 ACT: Pre-Test Module The drivers were shown nine driving scenarios, three curves, three intersections and three straight road sites (Table 47).
161
Table 47: Screen shots from the ACT program and a description of each training scenario, as well as the simulator evaluation scenario that best matches each training scenario. Circles in the photographs depict the mediation training that showed the trainees the selections made by exemplary experienced drivers
Driver’s View of Training Scenario
Description
1. Van blocking crosswalk. Most Like: Straight Segment 1 (Bus blocking crosswalk) Driver travels past a parked van on right with crosswalk ahead.
2. Curve right with roadside pedestrians left Most like: Curve 1 (Curve left) Drivers negotiate a curve right with possibility of oncoming vehicle crossing centerline.
3. Left turn across obstructed intersection Most like: Intersection 2 (Left turn across opposing SUV) Drivers turn left at an intersection where oncoming traffic might be obstructed.
4. Following a lead vehicle in a work zone Most like: Straight Segment 2 (Pedestrian in work zone) Drivers travel straight through work zone, after a transition right by the lead vehicle, workers can be seen ahead.
162
5. Sharp left curve Most like: Curve 2 (Sharp right curve) Drivers negotiate a sharp curve to the left that leads to a narrowed road.
6. Busy Intersection Most like: Intersection 2 (Busy Intersection) Drivers make a left turn at an intersection where traffic is approaching from all three remaining legs. 7. Sudden stop by lead vehicle Most like: Straight Segment 1 (Backing truck) The lead vehicle signals a left turn and starts the turn, but stops suddenly due to the presence of a pedestrian on the opposite side of the road. 8. Curve right with obstruction Most like: Curve 3 (Curve with stop sign ahead) Drivers negotiate a sweeping curve to the right and come upon a car that is parked partially in the road.
163
9. Straight movement through an intersection Most like: Intersection 1 (Truck blocking view of intersection) Drivers travel straight through the intersection while passing to the right of a turning vehicle.
5.1.4 ACT: The Theory Behind the Development of the Training Module -- Error Based Training: Mistakes, Mediation, and Mastery (3M) The same nine scenarios were used in this module as were used in the pre-test module (Table 47). In the training module, the participants learned through error based training (Ivancic and Hesketh, 2000). Error based training has been shown to be extremely effective in that when a person learns through error, they develop an inception. An inception-derived memory was associated with very high recall probabilities when compared to other training methods (Ivancic and Hesketh, 2000). Error based training is the most effective when it is embedded in a program which allows not only for errors (mistakes), but also for mediation and mastery. With such 3M training, the drivers are given the opportunity to make mistakes, are shown their mistakes and told why the mistakes lead to safety problems, and finally are allowed to master the skill being taught. The RAPT (Pradhan et al., 2009) and FOCAL (Pradhan et al., 2011) programs to which reference has been made are all 3M programs and have effects out to at least a year for novice drivers. Similar 3M programs have shown effects for up to two years with older drivers (Romoser and Fisher, 2009; Vlakveld et al., 2011).
164
Mediation as Rules. One question I asked was how best to implement the mediation portion of the 3M training program. Driver education programs have generally not used error training. This is not surprising. Letting students make errors, recording them, and then explaining the nature of the errors to novice drivers could be very dangerous on the open road where error training would have to occur. Instead, they emphasize mediation, often advocating a rule-based, “left-right-left” glance sequence before entering an intersection (Pennell, 2010) and turning to the left. The left-right-left rule-of-thumb assumes the most immediate hazard was in the nearest lane and to the left (in the U.S.). However, Klein (1997) indicated that when a subject was taught a rulebased method and a single cue deviates from a normal pattern, the ability to recognize a problem will suffer. In essence, if the most immediate hazard was anywhere other than to the left, driver performance suffers and crash probability will likely increase. Evidence that this is the case comes both from common sense and several studies. Horrey et al. (2006) examined responses of tactical military commanders in a simulated war zone after participants were trained to respond to the most threatening event. Horrey’s participants were encouraged to look toward one direction (a highlighted area); response times to threats in other areas were longer. Applied to drivers, we know that drivers’ abilities to respond to emergencies are negatively influenced by increases in visual eccentricity between where the driver is looking and where the hazard emerges (Summala, Lamble, and Laasko, 1998). Pritchett and Bisantz (2006) came to similar conclusions when examining aircraft radar detection tasks. Generalizing these findings to drivers suggests that when the most immediate hazard is anywhere other than to the left,
165
the “left-right-left” search sequence will likely result in longer response times, increased numbers of detection failures, and more crashes. Rather than uniformly suggesting drivers make a last look to the left when turning left, training should instead suggest that drivers “Neutralize the most immediate hazard first”. Neutralizing involves dealing with the most threatening unit or area first and doing it as early as possible. This is still a rule, but offers more flexibility while addressing the varying needs of different circumstances. For instance, if attempting a left turn at a location with a limited sight line to the right, the last look should be to the right, in order to reduce the likelihood that a vehicle will encroach on the driver or rider’s location without notice. When negotiating an intersection, the key is to limit the time before turning that the most immediate hazard is not monitored. As an example, if a driver glances to the shortest sight line last, he or she will reduce the possibility that a vehicle can arrive at their position undetected. Implementing Mediation. The next question is how to implement the training of rules and exceptions to rules. In her natural decision making research, Bisantz, et al. (2006) proposed a linear rule-based education by first training rules, and then exceptions to each rule. In the case of the left-right-left, drivers may be taught to consider the short sight line limitation as well as other limitations. However, a more effective strategy for training may be to reinforce why a driver must look in a particular direction at each point during a turn and under what situations the last glance may be in one direction, and in other situations in the other direction. This combination of rule-based training that
166
includes exceptions along with errors and explanations of the errors was what guided the development of the ACT training program. The specific rules that drivers were given in each scenario followed from an overarching rule. The overarching rule was that drivers should slow for a HRECCS (pronounced wreck). HRECCS stands for:
Hidden obstacle (or latent hazards as defined by RAPT) Roadside obstacles Escape routes are not available (do not pass between two vehicles if it can be avoided) Curves Closing on a lead vehicle when changing lanes is not an option Signals: be prepared to slow for changing traffic signals and heed warnings to slow whether than be from a traffic signal, a flagman, or some other specific warning.
These rules were reinforced and utilized in the context of the nine traffic scenarios. Trainees were taught to slow for a HRECC or to brake if two or more instances of a HRECC exist. After discussion of the concept of HRECCS, drivers were given direct training relative to where to brake, slow, and glance. The specific training was an extension of two concepts: 1. Slow for HRECCS (and brake if there are two or more of any of the HRECCS variables) and, 2. Maintain a safety bubble. The specific training related to the safety bubble concept involved teaching related to horn use and optimizing vehicle lane position. 5.1.5 The ACT Training Module The ACT training module consisted of three submodules: the safety bubble submodule, the HRECC submodule, and the feedback submodule, given in this order. The Safety Bubble Submodule.
167
Maintaining a safety bubble involved staying away from other obstacles and led the trainee into the concept of slowing for HRECCS. First, the user was taught to maintain a buffer between his or her vehicle and other obstacles. To do this, a driver was taught to position the vehicle within the lane so that sight line and buffer space were maximized. Next, drivers were taught that the horn was a collision avoidance device, not a greeting or chastising tool. When there was a situation where they believed others might not know they were there, they should sound the horn. Lastly, as a means of staying away from other vehicles, they should slow for HRECCS.
Figure 30. Safety bubble involves staying away from other obstacles by slowing or horn use (forward arrow), or moving left or right within the lane.
The HRECC Submodule. The need to brake, slow, change lane position or sound the horn when a HRECCS was present was explained through the use of both text and aerial and driver views. An example follows below. 168
Table 48: The four training slides for ACT Scenario 1, van blocking crosswalk scenario. Here, the drivers were taught about the concepts of hazard anticipation (glancing toward latent hazards), Slowing for HRECCS (Hidden hazards and roadside obstacles), maintaining a Safety Bubble by changing lane position (third slide), and using the horn when possible (when not steering) to keep others out of the Safety Bubble.
Glance Location
Slowing
Lane Position
Horn Use
The Feedback Submodule. In the feedback submodule, the nine scenarios in which novice drivers are most likely to crash, left-turn-across-path crashes, run-of-theroad crashes and rear end crashes, were trained one at a time. There were four components in the feedback module. (1) Aerial View Component. Referring to Table 48, the novice drivers were shown aerial views depicting the potential hazards that existed at each scenario. This training occurred after the drivers were taught how to use the program, and after the pre-test showing the drivers’ views of the same scenarios. With the aerial views, the possible mitigation tactics, slowing, 169
braking, horn use, lane positioning and anticipatory glance locations were discussed. Also, the limitations were discussed; for instance, horn use and emergency steering are contradictory actions. A driver should use the horn early in the event and not at a time where both hands on the wheel might be necessary. (2) General Mitigation Explanation Component. After being shown an aerial view of a scenario, three or more slides were presented. The first slide explained why a hazard was hidden in the scenario or, if visible, why it represented a threat. Animation was used to help enforce the concept. The second and subsequent slides explained one at a time the proper glance, lane position and slowing behaviors. For the van blocking the crosswalk (ACT Scenario 1), there was a forth training slide that addresses the possibility of horn use. (2) Exemplary Driver Component. To avoid an argumentative situation, drivers were not told that they were correct or incorrect, but instead were informed when their responses were consistent with, or inconsistent with the responses of exemplary drivers with no crash history. Results from Experiment 1 were compiled. The general findings displayed in Figure 28 were used as a guide to offer specific critiques. For example, experienced drivers were likely to glance toward the extent of the sightline or toward the potential hazard seven seconds or earlier before the incident area. The experienced drivers were also more likely to begin slowing when more than seven seconds from the incident, and braking when more than two seconds from the incident. Experienced drivers also selected optimal lane positions
170
which entailed being to the opposite side of the lane as the sight line obstruction and moving to the inside of the curve at the apex if safe to do so. (4) Individual Driver Feedback Component. Using the responses provided by an individual trainee during the pretest, the trainee’s responses were critiqued. The scenario was replayed one snapshot at a time. The experimenter offered an automated type critique, reading one text if the trainee gave the correct mitigation response at the appropriate time and another text if the trainee gave an incorrect mitigation response or a correct mitigation response at the wrong time. For example, if a driver failed to select a slowing response by the time the driver was shown the slide seven seconds from the curve, or if the driver failed to select a glance location toward an area of threat or latent hazard, feedback was offered by the experiment. 5.1.6 ACT: The Post-Test Module Once the training module was completed, all drivers participated in the ACT posttest and proceeded to the evaluation drive on the Realtime Technologies Driving Simulator. The same scenarios that were shown in the pre-test and training modules were included in the post-test module. The training was evaluated on similar scenarios in the Realtime Technologies Driving Simulator (Table 47). 5.1.7 ACT: Recording User Responses All responses in the posttest were shown at actual driving speeds and the results recorded. By the time the drivers were taking the posttest, they had responded to practice scenarios for each tactic (braking, glances, slowing, horn use, and lane position choice). Each participant who received ACT training had a practice driver, nine pre-tests, and 171
observed the nine scenarios for both training and feedback training. The slides were shown at one slide per second and at a speed that replicated a speed of 25 mph (36 feet per second).
5.2 Development of the Placebo Training Program The ACT and placebo versions of the program appear similar but the training components were much different. In the training component, the placebo version of the program had questions regarding signs and asked drivers to react to objects that appeared at traffic scenes, like a reaction time tester (see Appendix F).
5.3 Participants Thirty-six drivers participated in the study: 18 newly licensed (novice) drivers (16-17 years old) who received placebo training (control) and 18 novice drivers (16-17 years old) trained with the ACT program. The recruiting process was similar to Experiment 1; however most of the participants in Experiment 2 were recruited and scheduled by the Pioneer Driving School in Amherst, MA.
5.4 Experimental Design There were two groups of drivers. Specifically, one group of novice drivers was given placebo training. The other group completed ACT training. Half of the drivers were assigned to placebo training, the other half to ACT training. The procedures used to evaluate the effects of training were similar to the procedures used to evaluate the effects of experience that were described in Experiment 1. Drivers were exposed to nine different scenarios on a driving simulator. 172
Each of the scenarios that was used to evaluate the effects of training was described in a previous portion of this dissertation. The evaluation scenarios were again counterbalanced in the same way as in Experiment 1.
5.5 Procedure Upon arrival participants were asked to read and sign a consent form following University policy (see Appendix A). By signing the document, the participants indicated their understanding of the experiment and their willingness to continue. This document also informed the participants that compensation would be provided at the end of the experiment. Drivers from both the ACT trained and placebo trained groups received a preexperimental questionnaire. The pre-experimental questionnaire was given to the drivers to record whether they use corrective lenses and demographic information such as age and gender (see Appendix B). Both in the recruiting materials and on the questionnaire both groups’ members were asked their age and date they received a license. The remaining questions allowed for posttest analysis to determine whether individual differences, such as health and miles driven each year, had a significant impact on the responses by any particular participant. All instructions were written down and read aloud to each driver to assure that each participant received the same instructions, delivered in the same way. These instructions can be seen in Appendix C. A four minute practice drive was completed before testing. Based upon the research by McGehee et al. (2005), a practice drive of three minutes or more was 173
associated with the time necessary to reach a point that behavior was no longer influenced by unfamiliarity with simulator controls and virtual world. During the practice drive (and before the final instructions were read), the drivers were encouraged to “test” the vehicle’s steering, brakes and acceleration capabilities. The drivers were given as much time as they believed was necessary to become comfortable with the handling of the vehicle during the practice drive. There were hazards and traffic during the practice drive and the drivers learned that all vehicles stopped for stop signs and other traffic acted as was typical of most drivers. At the end of the practice drives the participant was asked to assume he was driving a friend’s vehicle that handled just as the simulator was handling and to drive accordingly. Participants were asked to travel at 40 mph through mixed road types that included four-lane suburban business arterials and two-lane rural and residential roads. Drivers were asked to drive as they normally would if it was necessary to get to an important meeting on time. Specifically, each driver was told that if he or she drove an average speed near that of the posted speed limit, the driver would arrive at the appointment on-time.
5.6 Data Collection and Analysis The experimental scenarios on the driving simulator were used to evaluate two main hypotheses about the effectiveness of the training, whether drivers looked at the potential risk (or area where a risk could emerge) and if they looked, did they diminish the potential risk in some way? Most importantly, the purpose of Experiment 2 was to determine if ACT was an appropriate tool to use to train drivers to mitigate crash risk by
174
slowing or moving within their lanes. The specific hypotheses being tested were addressed in Chapter 3 of this research. Data used to evaluate these hypotheses were collected using a head mounted eyetracker system (for eye behavior), and the simulator itself (for vehicle behavior). Drivers’ glance and slowing behaviors were recorded for at least ten seconds before the target zone (intersection or apex of a curve). There were a series of one-second launch zones, each treated like discrete bins. Anticipatory glances, slowing and lane positions were recorded as they were in Experiment 1. The ten seconds before a curve or intersection was also the area of the roadway in which a driver action (e.g., taking a foot off the accelerator) could potentially mitigate a threat. Again, the scoring was binary. The averages of three samples (each 0.05 seconds in length) were taken every one second from the simulator information for throttle position (on or off), brakes on or off, and vehicle speed. The results were rounded to an integer. For example, if there were two samples where the brakes were on and one where the brakes were off, a 1 was recorded. Yet if two samples recorded the brakes off, a 0 was recorded, regardless of the brake pressure.
5.7 Results 5.7.1 PC Posttest Performance of ACT Trained Versus Controls (Placebo Trained) Drivers were randomly assigned to one of two training programs. After the training, every driver received a posttest on a PC. The posttest involved the users viewing nine scenarios which included three curves, three intersections and three straight road segments. 175
If on the posttest a driver selected a glance location in an area of the next most threatening event, specifically, at the extent of the sight line at any time earlier than five seconds before the event, the glance was scored as a correct response. A correct foot pedal selection was to select slowing or braking no later than three seconds before the event, and braking not later than three seconds before the event. A correct lane position was recorded if the user selected a position within the lane that offered the greatest sightline and buffer space. The results are shown in the Figure 31. To repeat, there were 18 participants in each of the placebo trained and ACT trained groups. The matrices show the number of correct responses. Each row represents one of the nine training scenarios. The training scenarios are shown in Appendix E and are listed below. Each ACT training scenario is listed and the corresponding scenario that each driver faced in the driving simulator is listed as well. The hope was that each of the ACT training scenarios would be generalized by the drivers to the corresponding driving simulator scenarios.
176
Figure 31: Number of Correct responses for ACT trained drivers in the pretest and posttest compared to the correct responses by the placebo trained drivers in the posttest.
After comparing the ACT trained drivers’ performances in the pretest with those of the placebo trained drivers in the posttest, we can see there were no significant differences in glances, lane position choices, braking or slowing choices (Figure 31, bottom right). Yet, when we compare the ACT trained posttest performances with the placebo trained posttest performances, for every driver safety factor, glances, lane position, blowing and braking choices, each was significantly better for the ACT trained drivers (Table 30 bottom middle). In Figure 31, cells that are green suggest that many of the ACT trainees were correct in that response. Red indicates that very few responded correctly. Again, a correct response was someone who indicated a glance toward the most threatening area at least five seconds before the incident, selected a slowing or braking option seven seconds or earlier before the incident, selected a proper lane position three seconds before the incident, and selected the brake at least three seconds before the incident. 177
As an example, we can compare one scenario. The second row was training for a curve right with roadside pedestrians along the opposite side of the road. In that scenario, an oncoming vehicle could cross the center line to avoid the pedestrians as the driver was approaching the curve. In the pretest, 11 participants correctly identified the extent of the sightline to be the proper glance location. Only one participant indicated that he or she would move to the left in the lane to increase sightline (and eventually move to the inside of the curve). Two participants indicated that they would brake at least three or more seconds before the apex of the curve and ten of the eighteen participants indicated that they would begin to slow as least seven seconds before the curve. In the posttest, we can see that the same participants made 17 correct glance responses, 13 correct lane position responses, 15 correct brake decisions, and selected the slow option correctly 16 times. Also of note is that the placebo trained drivers responded similarly to the ACT trained drivers’ pre-test scores. Clearly, the ACT trained participants understood the training and scored significantly better in selecting glances (Mean = 15.0 vs. Mean = 9.0), lane position (Mean = 12.6 vs. Mean = 4.9), braking (Mean = 13.9 vs. Mean = 5.2), and slowing (Mean = 15.4 vs. Mean = 12.1). Each of these differences between the posttest responses of the ACT trained and placebo trained drivers was statistically significant. The next question that needs to be answered is whether the ACT training actually improved drivers’ performance on the simulator. The ACT training program addressed lane position, horn use, glances, and speed management. Results are reported only for the glances, speed choice and lane keeping performance of the ACT trained and the
178
placebo trained drivers on curves, intersections and straight segments just as in Experiment 1. 5.7.2 Geometry: Horizontal curves Recall that the target speed thresholds of 20 mph for the sharp curve right and 34 mph for the moderate curve left were utilized to compare the drivers’ speed choices. First, the aggregate glance, braking and slowing behaviors were compared. Second, these behaviors were compared conditional on an anticipatory glance being made (or not made). 5.7.2.1 Aggregate Glance, Braking and Slowing Behaviors, and Lane Position The ACT trained and placebo trained drivers were evaluated on whether anticipatory glances were made when approaching curves. Specifically, anticipatory behaviors of glancing to the far extent, slowing to target speed and moving within the lane were compiled. The behaviors of ACT trained drivers were compared to those who received placebo training. The open question here was whether the ACT trained drivers showed improved performance on each of the anticipatory measures. Glances. The prerequisite to anticipatory slowing was anticipatory glances. The percentage of ACT trained drivers that glanced at each second when approaching a sharp curve and a routine curve were compared. Significantly more ACT trained drivers made glances across the curve (to the far extent) five to seven seconds before arriving at the curve compared to the novice drivers. When five to nine seconds from the curve, nearly four times as many ACT trained drivers made a glance toward the far extent than had placebo trained drivers. Figure 32 and Table 49 also show that ACT trained drivers were
179
42 percentage points more likely to glance across the curve than placebo trained drivers when five seconds from the curve.
Figure 32: The cumulative percentage of ACT trained and placebo trained drivers that made a glance to the far extent (across the curve) as they approached a sharp curve to the right. (* The Asterisks - The difference between the ACT trained and placebo trained drivers’ glances were significant.)
The statistical significance of the difference between the glance proportions of the ACT trained and placebo trained drivers for each second at the sharp curve is shown in Table 49. At every one-second bin, ACT trained drivers were more likely to glance across the curve. When six and five seconds from the curve, 46% and 57%, respectively, of the ACT trained drivers had made a glance across the curve to the far extent. At the same time, fewer than 15% of the placebo trained drivers made an anticipatory glance toward the far extent when five seconds from the curve.
180
Table 49: Comparison of ACT trained and Placebo trained Drivers’ Glances within Each 1-second epoch before a curve. (Numbers in bold yellow indicate significant differences.)
181
At the moderate curve left, ACT trained drivers differed from placebo trained drivers only 14 percentage points in the likelihood that they glanced to the far extent, with most of that difference occurring in the last two seconds before the curve. (See Figure 33).
Figure 33: The cumulative percentage of ACT trained and placebo trained drivers that made a glance to the far extent (across the curve) as they approached a curve to the left.
The statistical significance between the cumulative percentage of glances by the ACT trained and placebo trained drivers for the moderate curve left during each second is shown in Table 50. Of interest, half of the ACT trained drivers made a glance toward the far extent when more than a second from the curve, while not more than half the placebo trained drivers made a far extent glance.
182
Table 50: Cumulative percentage of ACT trained and placebo trained drivers who glanced toward the far extent of the sight line in each second while approaching a sharp curve to the right.
Speed. The speed profiles of the ACT trained and placebo trained drivers show that both ACT trained drivers and Placebo trained drivers had slowed considerably from an initial speed near 40 mph on the preceding straight segment (the speed limit) to something close to 30 mph when nine seconds before the sharp curve (Figure 34). However, when nine seconds before the curve, the ACT trained drivers were traveling some 3 mph slower than the Placebo trained drivers. This early, slower speed selection worked in the ACT trained drivers’ favor. The average speed of both the experienced and placebo trained drivers began to decline sharply when four seconds from the curve. The fact that placebo trained drivers did not reduce speed as much early means that they collectively did not reach the desired target speed of 20 mph. 183
Figure 34: Average speed of ACT trained and placebo trained drivers in the ten seconds before reaching a sharp curve to the right.
ACT trained and placebo trained drivers both slowed considerably before entering the ten second window of this research when approaching the sharp curve. When approaching the curve left, both the ACT trained and placebo trained drivers were traveling above the 40 mph speed limit until as late as six seconds before the curve. While the ACT and placebo drivers started at similar speeds, the ACT trained drivers were slowing from as far back as nine seconds before the curve. The speed profile for the placebo trained drivers does not dip much until five seconds before the curve. Figure 35 shows three asterisks that refer to the times when the ACT trained drivers slowed to a target speed of 34 mph or less. Even though the speed is still fast at these times, the ACT
184
trained drivers were slowing at a rate that would allow them to slow to a speed of 34 mph or less by the time they reached the curve.
Figure 35: Average speed of ACT trained and placebo trained drivers in the ten seconds before reaching a moderate curve left. (*The Asterisks - refers to times when significantly more ACT trained drivers slowed to a target speed of 34 mph.)
If you recall, because the left curve did not have as small a radius as the sharp curve, less speed loss was necessary for safe operation according to the Bonneson model. According to the equation by Bonneson and Pratt (2009), we would expect an average curve speed to be 20 mph at the sharp curve and 34 mph at the moderate curve left. Also, you might recall that we are comparing target speeds as a means of attempting to understand the anticipatory behavior of the drivers. A target speed considers the current speed as well as the deceleration. 185
At the sharper curve, ACT trained drivers were more likely to reduce speed to a target speed below 20 mph than were the placebo trained drivers. When one second before the curve, 77% of the ACT trained drivers had slowed compared to only 48% of the placebo trained drivers. (See Table 51). Table 51: Percentages of ACT trained and placebo trained drivers that slowed to target speed of 20 mph when approaching a sharp curve to the right.
At the moderate curve left, ACT trained drivers were significantly more likely to slow to a target speed of 34 mph when six to eight seconds before the curve than were the placebo trained drivers. Even when within five seconds before the curve, the difference between the percentage of ACT and Placebo trained drivers who had slowed to the target speed of 34 mph ranged between 19 and 32 percentage points. (See Table 52).
186
Table 52: Percentages of ACT trained and placebo trained drivers that slowed to target speed of 34 mph when approaching a sharp curve to the right. (Numbers in bold yellow indicate significant differences.)
The groups were compared relative to the extent at which the drivers slowed to the target speed. For instance, if Driver A slowed to target speed five seconds from the curve and Driver B slowed to the target speed one second before the curve, the average for these two drivers would be 3.0 seconds. Using this same methodology, the average time bin at which the drivers in each group slowed to the target speed was compared. At the sharp curve right, ACT trained drivers on average slowed to target speed 2.7 s. (SD = 0.4 s.) before the curve. Placebo trained drivers did not reach target speed until 1.8 s. (SD = 1.7 s.) before the curve, a difference which was significant. [t (29) = 2.36, P = 0.03]. At the moderate curve left ACT trained drivers slowed to target speed
187
5.7 s. (SD = 4.3 s.) before the curve, compared to placebo trained drivers who did not reach target speed until 2.5 s. (SD = 3.7 s.) before the curve [t (17) = 1.98, P = 0.16]. Lane Positioning. Table 53 shows that ACT trained drivers started near the center of the lane (0.03 ft.) and moved to a position 1.81 ft. to the right of the center of the lane at the apex of the curve. This 1.84 ft. movement to the inside of the curve is consistent with the findings by Bonneson et al. (2009), though only about half of the size of the lateral movement made by experienced drivers. However, during the third and second seconds before the apex of the curve, the ACT trained drivers actually drifted left of the lane center by about a foot and moved quickly back to the right of lane center.
The
placebo trained drivers had no real plan of attack either. They started 0.34 ft. to the right of lane center and ended 2.01 ft. to the right of lane center, a lateral movement of 1.77 ft. The average placebo trained drivers maintained a position near the center of the lane throughout most of the curve and then, towards the end wandered to the left of center and again at the last second moved to the inside of the curve. The lane position results suggest that both ACT trained drivers and placebo trained drivers had difficulty negotiating the lane position in the curve.
188
Table 53: Average lane position of ACT trained and placebo trained drivers when approaching a sharp curve right. Distances refer to the position left (negative in parentheses) or right (positive) of the center of the lane.
The differences in the lane position offset of the ACT trained drivers and Placebo trained drivers were equally hard to distinguish in the moderate curve left (Table 54). The Placebo trained drivers started farther to the right than the ACT trained drivers at 9 s (1.54 ft. versus 0.90 ft.). But they also ended less far to the left at the apex of the curve (0.23 ft. versus 0.84 ft.). It is noteworthy that the ACT trained drivers moved to the right quite a bit at 8 s (1.62 ft.). Thus, the maximum lateral shift for the ACT trained drivers (1.85 ft.) was about what it was for the Placebo trained drivers (1.74 ft.). In short, both sets of drivers appeared to negotiate the curve about equally well.
189
Table 54: Average lane position of ACT trained and placebo trained drivers when approaching a moderate curve left. Distances refer to the position left (negative in parentheses) or right (positive) of the center of the lane.
5.7.2.2 Conditional Braking, Slowing and Crashes Conditional Glancing and Slowing to Target Speed. The conditional probability that a driver glanced, given that he or she had glanced when five to eight seconds before the sharp curve right, was computed. A comparison was then made between these conditional probabilities for ACT and Placebo trained drivers. Table 55 show the joint outcomes of the eighteen drivers in the ACT and Placebo groups the first time they drove through the sharp curve. Table 56 shows the conditional outcomes. Drivers from all four groups varied on subsequent drives through the sharp curve and when materialized. My goal here was to compare drivers’ abilities to recognize a latent or unfamiliar hazard. Clearly, there were improvements in the performances by all groups in subsequent drivers. Vlakveld et al. (2011) has indicated that crashing events might create error training events that cause drivers to learn. This research will not address this in that the
190
sharp curve was the only scenario where such differences were noted. Instead, for a fair comparison, I compared only the first drive through the sharp curve. Consider first the joint outcomes (Table 55). The ACT trained drivers were five times more likely to make a glance to the far extent (11 vs. 2) and 1.16 times more likely to slow to the target speed of 20 mph (14 vs. 12). Importantly, 9 ACT trained drivers both glanced and slowed whereas only 2 Placebo trained drivers both glanced and slowed. Table 55: Number of ACT trained and placebo trained drivers that slowed to target speed after glancing to the far extent when five to eight seconds before a sharp curve right.
If there was an anticipatory glance, both ACT trained and placebo trained drivers were much more likely to slow (ACT trained 9 out of 11, 89%; placebo trained 2 out of 2, 100%) (Table 56). If the placebo trained driver did not make an anticipatory glance, the probability reduced to 63% (10 out of 16) from 100%. Likewise, 71% (5 out of 7) of the
191
ACT trained drivers that failed to make a glance to the far extent slowed to the target speed compared to the 89% that slowed with a glance. Table 56: Percentage of ACT trained and placebo trained drivers that slowed if they had glanced to the far extent or had not glanced to the far extent. (Numbers in bold yellow indicate significant differences.)
Consider next the joint and conditional outcomes for the moderate curve left (Table 57 and Table 58). Unlike the very dangerous sharp curve scenario where many drivers crashed, or responded sharply, and learned from each exposure to the curve, at the curve left, there were no crashes. Thus, there were no noticeable differences in the behavior of the drivers on the first time through the curve or when the hazard was materialized. The ACT trained drivers were actually less likely to glance (12 versus 14) and only 1.19 times more likely to slow than the Placebo trained drivers (31 vs. 26). More importantly, 11 ACT trained drivers both glanced and slowed whereas only 8 Placebo trained drivers both glanced and slowed (Table 57).
192
Table 57: Number of ACT trained and placebo trained drivers that slowed to target speed after glancing to the far extent when five to eight seconds before a moderate curve left.
Consider next the conditional probabilities (Table 58). Of those who glanced to the far extent at a time earlier than five seconds before the curve, significantly more ACT trained drivers slowed (92%; 11 out of 12) than did placebo trained drivers (57%; 8 out of 14). When there was not a glance, ACT trained drivers and placebo trained drivers slowed in similar percentages (ACT trained 20 out of 24; Placebo trained 18 out of 22).
193
Table 58: Percentage of ACT trained and placebo trained drivers that slowed if they had glanced to the far extent or had not glanced to the far extent when approaching a curve left. (Numbers in bold yellow indicate significant differences.)
While glancing across the curve appears to be predictive of speed loss on the sharp curve right, such does not appear to be the case for the moderate curve left for the ACT or Placebo trained drivers. On the sharp curve right, fully 89% of the ACT trained drivers reached target speed given that they glanced and 71% of the ACT trained drivers reached target speed given that they did not glance (a difference of 18 percentage points). Of the Placebo trained drivers who glanced to the far extent on the sharp curve right, 100% slowed, while 63% slowed if they had not glanced (a difference of 37 percentage points). Conversely, on the moderate curve left, 92% of the ACT trained drivers reached target speed, given that they glanced, whereas 83% slowed given that they did not glance (a difference of only 9 percentage points). Of the Placebo trained drivers who glanced to the far extent on the moderate curve left, 57% slowed, while 82% slowed if they had not glanced (a difference in the direction opposite that predicted). These results show that ACT training is having an effect primarily on the number of novice drivers who glance to the far extent of the curve when that curve has as tight 194
radius. Five times as many ACT trained drivers glanced to the far extent in the tight curve right as did Placebo drivers. Given that a driver glanced in the sharp curve right, regardless of whether an ACT or Placebo trained driver, the driver was very likely to slow. On the moderate curve left, more Placebo trained drivers glanced to the far extent than did ACT trained drivers. The percentage of drivers that slowed in the ACT group, given that the drivers glanced, is almost twice as large as it is for the Placebo group (92% versus 52%). However, given that the drivers did not glance, the percentage that slowed is almost the same for the ACT and Placebo groups (83% versus 82%). This indicates that a glance across the far extent is not necessary for adequate slowing in the moderate curve left, perhaps because drivers in both groups can pick up the geometry well enough in their peripheral field of view. Conditional Outcomes in Crashes. While there were no crashes at the moderate curve to the left, there were six crashes at the sharp curve to the right. The placebo trained drivers were responsible for five of the six crashes. Additional information can be learned by examining how each occurred. Of the crashes, four of the five placebo trained crashes involved a driver completely leaving the road. The lone ACT trained driver that crashed drove off the road. Of the ACT and Placebo drivers that crashed, no driver made a glance to the far extent when five to eight seconds from the curve. The lone ACT trained driver that crashed slowed to a target speed of 19.7 mph, and did not reach target speed until reaching the curve, P(S|C) = 1.0 (Table 59,). None of the placebo trained drivers that crashed had slowed to target speed, and three never slowed to less than 28 mph, P (S|C) = 0. Of those who did not crash, ACT trained and placebo trained drivers slowed 93% (13 195
out of 14) and 92% (12 out of 13), respectively [Table 57, P (S|nC)]. Yet no novice driver that crashed had slowed (0 out of 5). Table 59: Percentage of ACT and Placebo trained drivers who crashed, slowed before crashing given that they crashes, slowed given that they did not crash, glanced and slowed given that they crashed and glanced and slowed given that they did not crash
I also compared the percentage that crashed of those who slowed versus those who did not slow. ACT trained drivers were much less likely to crash if they had slowed to the target speed, P (C|S) = 0.07 versus P (C|nS) = 0.25. But Placebo trained drivers were significantly less likely to crash if they slowed to target speed before the curve, P (C|S) = 0.0 versus P (C|nS) = 0.83 (Table 60). Table 60: Comparison of the percentage of ACT trained and placebo trained drivers that crashed based upon whether the driver slowed to the target speed or did not slow to the target speed. . (Numbers in bold yellow indicate insignificant differences.)
196
5.7.3 Geometry: Intersections Consider next intersections. In the above analysis of novice and experienced drivers behaviors at curves, the question was whether and when the two sets of drivers reached the target speed.
Recall that the target speed is the current speed minus the
speed loss in the previous second projected forward in time until the location of the intersection. In order to compute the target speed, one needs to know how much speed needs to be lost from the posted speed limit by driver when the driver enters a given arrangement. The speed that needs to be lost at a curve has been studied relatively extensively. Thus, the target speed can easily be computed. However, the speed that needs to be lost at intersections has been studied less extensively. Thus, it can be more difficult to know a priori what target to use as a threshold when comparing the two sets of drivers. The speed that needs to be lost from the posted speed before the intersection will depend on the geometry of the intersection, the traffic, and the environment. At Intersection 1, (Left turning truck), there is a large truck in the adjacent lane and a raised curbing to the right. According to the several authors (O’Leary, 2006; Lisle et al., 1980; Thompson et al., 1985), drivers reduced speeds by 1.0 to 2.0 mph when an obstacle was placed alongside the driver’s lane. These studies suggested a speed loss of approximately 1.5 mph without a sightline obstruction when entering an intersection. I assumed drivers would reduce speed by at least twice the amount. Three mph was selected by extrapolating from the literature as well as considering how speed is lost. If a driver coasts, average rolling resistance would result in a speed loss of approximately 0.3 mph to 2.2 mph in each second (Warner et al., 1983). Thus a speed loss of greater than 3 mph 197
would more likely be a result of a conscious choice by the driver. I computed predicted speeds at each one second interval before Intersection 1 based on the current speed and speed loss in the preceding second. Using these predicted speeds, I compared the percentage of the novice and experienced the drivers who would have reached a target speed of 37 mph when they entered the intersection (reduced their speed by 3.0 mph from the posted speed limit). The geometry and traffic is different at Intersection 3. When approaching a left turn at a busy intersection, drivers should be prepared to stop. While a stop might not be required, certainly, it could be necessary. Thus, drivers should be prepared to be ontarget to slow to a stop until they deem it is not necessary to do so. The percentage of placebo trained and ACT trained drivers who reach a target speed of zero mph when approaching a left turn at a busy intersection was compared at each second prior to entering the intersection. This intersection is referred to as a busy intersection in that there is traffic approaching on all three legs. At Intersection 1, the time when the target speed was attained, the percentage of glances to the near extent and the lane position in each second ten seconds before the intersection was compared for ACT and placebo trained drivers. At the busy intersection (Intersection 3), the time to target speed, the percentage of glances to the side roads and secondary glances, and the lane position in each second ten seconds before the intersection was compared for ACT trained and placebo trained drivers. Each group was analyzed when driving through the intersection once when there was not a materialized hazard and once when there was a materialized hazard. Recall that the materialized hazard at the left turning truck (Intersection 1) was a previously obstructed opposite 198
direction car that made a left turn into the path of the driver. The intruding vehicle was previously obstructed by a truck in the right lane. In the last two seconds, the left turning car could be seen by drivers after passing the stopped left-turning truck. At the busy intersection (Intersection 3), the materialized hazard was a car entering the intersection in the left lane on the side road to the left of the driver. 5.7.3.1 Aggregate Glance, Braking and Slowing Behaviors, and Lane Position Using the target speed of 37 mph for the intersection where a left turning truck obstructed the view and zero mph for the intersection where the drivers turned left, the speed loss, glance behaviors and steering behaviors of the ACT trained and placebo trained drivers were compared. First, the aggregate glance, braking and slowing, and steering behaviors are compared. Second, these behaviors were compared conditional on an anticipatory glance being made (or not made). Glances. When the drivers approached a curve, the glances toward the threatening area were followed by slowing. When approaching an obstructed intersection or left turn across path situation, slowing has to precede the measureable glance. Here, glancing was confirmation that the slowing behaviors related to anticipatory actions rather than random behaviors. Clearly, there must have been some anticipatory actions that occurred before slowing began, but such glance behaviors could not be recorded with the available equipment. Yet the glances that occurred later suggest that slowing was anticipatory of some hazard or unrelated to any particular hazard. Two intersection configurations were addressed. At a busy intersection the driver must slow for a left turn (Intersection 3), and at the intersection with a view obstruction due to a left turning truck the driver may slow for the reasons discussed above even 199
though he or she travels straight through (Intersection 1). Because the intersection movements were much different, glances had to be recorded differently as well. When approaching a left turning truck in the left lane, the sight line toward traffic within the intersection is obstructed (Intersection 1). When approaching the left turning truck, a driver cannot see in front of the truck until two to five seconds before the intersection. A glance to the right edge of the stopped truck is not strong evidence that the driver was anticipating traffic in front of the truck when as far back for six to ten seconds. However, if a driver made a glance toward the front of the truck as he or she was passing it, it suggests that the glance was anticipatory. Also, when examining the results, it was apparent that the ACT trained (and placebo trained) drivers made glances to the near extent during several one-second bins while the placebo drivers did not. Thus, when I compared cumulative glances, there was little difference between the groups. For these reasons, the percentages that glance in each second tell me much more about how the drivers differed when approaching these intersections. At the busy intersection, the potential hazard could emerge from any of the three remaining legs of the four-way intersection (Intersection 3). Imagine traveling toward the intersection. First drivers need to assure that traffic on the side roads has stopped or will not enter the intersection. With side road glances, I was interested in when they occurred as the driver approached the intersection. Therefore, side road glances were reported as the cumulative percentage of drivers who glanced to the side roads in each second. I use the plural “roads”, meaning that a driver must glance toward both the left and right legs of the intersection.
200
After assuring that the side road traffic is neutralized, a driver must then assure that the turn is safe. With an oncoming vehicle that may be traveling fast or slow, a driver should make a secondary glance. A secondary glance is a glance toward oncoming traffic at the moment of the turn or immediately after the turn starts. The secondary glance offers the driver a last chance to abort the turn. Secondary glances were reported as a percentage of drivers that made the glance. Lastly, the percentage of drivers that both made the necessary side road glances and the secondary glance was reported. At Intersection 1, ACT trained drivers were much more likely to make glances toward the near extent throughout the approach to the intersection except at two seconds from the intersection when glances were equal to placebo trained drivers. (See Figure 36).
201
Figure 36: Percentage of ACT trained and placebo trained drivers that glanced to the near extent left in each one-second periods when approaching an intersection with the left lane obstructed by a turning truck.
The statistical significance of the difference between the percentage of ACT trained and placebo trained drivers who glanced to the near extent for each second before the intersection is shown in Table 61. At every one-second bin except 1 s and 2 s, ACT trained drivers were more likely to glance toward the near extent of the sightline (recall that the near extent is at the right front of the stopped truck). However, only when nine seconds before the intersection were ACT trained drivers significantly more likely to make a glance to the near extent. The difference between ACT trained drivers and placebo trained drivers was at least 16 percentage points (in favor of the ACT trained drivers) when four, six and seven seconds from the intersection.
202
Table 61: Comparison of experienced and novice drivers’ Glances within Each 1-second epoch when approaching an intersection with its view obstructed by a left-turning truck. (Numbers in bold yellow indicate significant differences.)
At Intersection 3, when approaching a busy intersection, the differences in glances between the ACT trained and placebo trained drivers were not as radically different as at the obstructed view intersection. Half of the ACT trained drivers had glanced to both legs when six seconds from the intersection compared to only 28% for the placebo trained drivers. The difference between the percentage of ACT trained (89%) and placebo trained (78%) drivers who made glances to both side roads in the two seconds before turning was 11 percentage points.
203
Looking now at the far right of Figure 37, it can be seen that the difference between the percentage of ACT trained (61%) and placebo trained (50%) drivers who made secondary glances before turning was 11 percentage points. The difference between the percentage of ACT trained and placebo trained drivers who made both side road glances and a secondary glance toward the oncoming traffic before turning was 17 percentage points. All three legs refers to a driver who looked to both side roads and also made the secondary glance ahead before turning into the opposite lane.
Figure 37: Cumulative percentage of drivers that completed both glances to the left and right side road in the nine seconds before turning. Also, the percentage of drivers that made a secondary glance toward oncoming traffic before turning, and lastly, the percentage of drivers that both completed the side road glances and the secondary glance before turning.
204
Although the percentage of glances for the ACT and placebo trained drivers differ absolutely in each one second bin, sometimes by a considerable amount, they do not significantly differ from each other in any of the one second bins (see Table 62). Nor are there differences in the percentage of secondary glances or the percentage of glances towards all three legs. Table 62: Cumulative percentage of ACT trained and placebo trained drivers that completed glances all three glances to all three legs of the intersection. The cumulative percentage of ACT trained and placebo trained drivers that completed both side road glances in each second (0 – 9 s) before the curve. Also, the percentage of ACT trained and placebo trained drivers that made secondary glances and finally (bottom row), the percentage drivers that completed the necessary side road glances and a secondary glance before turning.
Speed Loss. The speed choice of the ACT trained and placebo trained drivers did not differ initially at the straight through intersection (Figure 38). Only as the two groups approached the intersection did the speed between the two groups begin to differ. 205
Significantly more ACT trained drivers slowed to target speed than did placebo trained drivers. In Figure 38, the asterisks refer to moments when significantly more ACT trained drivers had slowed to the target speed of 37 mph at Intersection 1 (also see Table 63). When two to four seconds before the intersection, more than half the ACT trained drivers had slowed to a target speed of less than 37 mph, yet only 29% of placebo trained drivers ever slowed to a target speed of 37 mph (Table 63).
Figure 38: Average speed of the ACT trained and placebo trained drivers during each second when approaching an intersection with an obstructed view due to a left turning truck in the adjacent left lane. * The Asterisks - suggests that significantly more ACT trained drivers slowed to the target speed of 37 mph in that second.
206
Table 63: Comparison of percentage of ACT trained and placebo trained drivers’ Glances within Each 1-second epoch when approaching an intersection with its view obstructed by a left-turning truck. (Numbers in bold yellow indicate significant differences.)
At the left turning truck scenario (Intersection 1), cognition, or anticipation was necessary by the drivers to recognize a need to slow. At a busy intersection (Intersection 3) where the goal was to turn left, the speed differences between the ACT trained and placebo trained drivers were not as noticeable. While the speed did not differ much, the target speed did. In Figure 39 the asterisks refer to times when the ACT trained drivers were on-target to slow to zero mph (also see Table 64). Again, target speed is the predicted speed at Intersection 3 based upon the current speed and the speed loss in the previous second. The dotted line in the figure shows the projected slowing if the drivers
207
continued to slow the rate they did between three and six seconds before the turn. The dotted line depicts a target speed. When within five seconds of a busy intersection, more than half the ACT trained drivers slowed to a target speed of 0 mph (See Table 64). Half of the placebo trained drivers did not slow to a target speed of 0 mph until two seconds from the intersection. When six to nine seconds before the intersection, significantly more ACT trained drivers slowed to target speed than did placebo trained drivers.
Figure 39: Average speed of the ACT trained and placebo trained drivers during each second when approaching a busy four-way intersection. (* The Asterisks - suggests that ACT trained drivers were significantly more likely to slow to the target speed of zero mph in that period.)
208
Table 64: Cumulative percentage of ACT trained and placebo trained drivers to slowed to a target speed of zero mph in the nine seconds before arriving at a busy intersection. (Numbers in bold yellow indicate significant differences.)
At the intersection where the truck obstructed the view (Intersection 1), ACT trained drivers slowed to target speed significantly earlier. ACT trained drivers slowed to the target speed of 37 mph and average of 5.7 s. (SD = 4.3 s.) before the intersection while placebo trained drivers did not slow to target speed until 2.5 s. (SD = 3.7 s.). [T (35) = 2.18, P = 0.04]. At the busy intersection (Intersection 3), ACT trained drivers slowed to target speed earlier, but not significantly so. ACT trained drivers slowed to a target speed of 0 mph when 4.3 s. (SD = 3.6 s.) from the intersection compared to 2.7 s. (SD = 3.5 s.) for the placebo trained drivers [t (35) = 1.00; P=0.33]. 209
Lane Position. Some drivers might choose to move within the lane and offer additional buffer space as a way to mitigate the need to slow as much. At Intersection 1, when eight and nine seconds from the intersection, ACT trained drivers began significantly left of center in the lane compared to placebo trained drivers. As the drivers passed the truck, ACT trained drivers moved to a position significantly right of the placebo trained drivers, but the amplitude of the difference appears to be only 0.1 m (about 4 inches). It would be difficult to claim that this difference changes the likelihood of a crash. (See Table 65). Table 65: Average lane position of ACT trained and placebo trained drivers when approaching an intersection with a left turning truck obstructing the view. (Distances refer to the position left (negative) or right (positive) of the center of the lane. Numbers in bold yellow indicate significant differences.)
At the busy intersection, there were notable differences in lane position. The ACT trained drivers, who received specific training relative to lane position when approaching an intersection stayed slightly more than ½ foot left of center of the lane, which is not remarkable, but the position of the placebo trained drivers was almost onefoot right of center. (See Table 28). The placebo trained drivers also began to turn
210
earlier than did the ACT trained drivers. The placebo trained drivers were more likely to make a longer arching turn that left them exposed to conflicting traffic longer. At the start of the intersection, the average placebo trained driver was over 4 feet left of center. Given a vehicle of average width, that places the vehicle over the center line of the road. The lane position results means that the ACT trained drivers remained approximately ½foot to 1-foot left of center until making the turn. The placebo trained drivers began the turn earlier than did the ACT trained drivers. Instead, the ACT trained drivers moved farther into the intersection before turning and apparently used that extra distance (and time) to make more glances. Table 66: Average lane position of ACT trained and placebo trained drivers when approaching a busy four-way intersection. Distances refer to the position left (negative) or right (positive) of the center of the lane.
Post test data snooping revealed that three novice drivers and three experienced drivers crossed the centerline before entering the intersection (The elongation of the closest edge of the side roads). Furthermore, of these drivers who started the turn early, only one experienced driver and one novice driver made glances to all three legs of the intersection before the turn. Comparing these data with the percentages that glanced to all three legs (50% vs. 33%, for the experienced and novice drivers respectively) suggests 211
that moving farther into the intersection before turning was associated with a marginal improvement in the likelihood of glancing to all three legs of the intersection. 5.7.3.2 Conditional Braking and Slowing and Crashes At Intersection 1, the left turning truck at the intersection, more ACT trained drivers slowed to target speed (67%) than placebo trained drivers (39%). ACT drivers were also more likely to glance toward the near extent. When comparing the conditional probabilities, the ACT trained drivers were much more likely to slow if they had glanced (71%) and less likely to slow if they did not make a glance to the near extent (50%). However, the placebo trained drivers were nearly equally underachievers relative to slowing whether they had glanced (40%) or did not glance (38%). (See Table 67 and Table 68). Table 67: Number of ACT trained and placebo trained drivers that slowed to target speed after glancing to the far extent when two to five seconds before an intersection with a left turning truck.
When comparing the conditional probabilities, the experienced drivers were much more likely to slow if they had glanced (71%) and less likely to slow if they did not make a glance to the near extent (50%). However, the novice drivers were nearly equally likely 212
to slow if they had glanced (40%) or not glanced (38%). These results might suggest that the glances by the novice drivers were not mindful or anticipatory. Given a glance, ACT trained drivers were 1.78 times more likely to slow than were the placebo trained drivers. This difference was nearly significant. Without a glance, only 12% more ACT trained drivers slowed than placebo trained drivers. Table 68: Percentage of ACT trained and placebo trained drivers that slowed if they had glanced to the near extent or had not glanced to the near extent when approaching an intersection that was obstructed by a left turning truck.
Consider next the busy intersection (Intersection 3). A comparison was made of the percentage of novice and experienced drivers who slowed, conditioned on their having glanced to both side roads and toward oncoming traffic before entering the opposite lane. The conditional outcomes of the eighteen drivers when confronted with the materialized hazard were compared. ACT trained drivers were more likely to slow (78% versus 62%) and more likely to glance (50% versus 33%) than placebo trained drivers. If the drivers glanced toward the near extent, 67% of ACT trained drivers slowed and 50% of Placebo trained drivers slowed. (See Table 69 and Table 70).
213
Table 69: Number of ACT trained and placebo trained drivers that slowed to target speed of 0-mph and glanced toward all three legs of the intersection before turning left.
Of interest was whether the experienced or novice drivers reduced speed more or less if they had glanced or not. ACT trained drivers were more likely to slow if they had made glances to all three legs of the intersection (Experienced 67%; Novices 50%) or if there was not a glance(89% vs. 67%) (See Table 70). Table 70: Percentage of ACT trained and placebo trained drivers that slowed if they had glanced to the far extent or had not glanced to the far extent.
214
Conditional Outcomes in Crashes. One of the reasons for this research was to investigate reasons for crashes among placebo trained and ACT trained drivers at high crash risk locations. No driver crashed at the busy intersection. However, several drivers crashed at the intersection where the truck was making a left turn. Recall that the materialized hazard here was a car turning left across the path of the driver. The view of the car had previously been obstructed by a large truck that was waiting to turn left. Placebo trained drivers were involved in thirteen crashes, out of the possible 18 drivers. Six ACT trained drivers also crashed. Of those who did not slow, ACT trained and placebo trained drivers crashed 83% and 100% of the time. Yet for those who slowed, ACT trained drivers crashed only 8% and placebo trained crashed 29%. Clearly, all drivers that both glanced and slowed were much less likely to crash than those who did not both glance to the extent and slow to target speed. (See Table 71). Table 71: Percentage of drivers who slowed before crashing, or did not slow before crashing at an intersection with the view obstructed by a left turning truck.
I also compared within the ACT trained and within the placebo trained groups the probability of a crash for those who slowed versus those who did not slow (Table 72).
215
Both ACT trained and placebo trained drivers were significantly less likely to crash if they had slowed to the target speed. Table 72: Comparison of the percentage of ACT trained and placebo trained drivers that crashed at an intersection with a left turning truck based upon whether the driver slowed to the target speed or did not slow to the target speed. (Numbers in bold yellow indicate significant differences.)
5.7.4 Geometry: Straight Segments The target speed threshold for the straight road segments was 23 mph. For an explanation of the reasoning behind this measure, please refer back to Experiment 1. Using these target speed thresholds as a benchmark, the speed loss and glance behaviors of the ACT trained and placebo trained drivers were compared. First, the aggregate glance, braking and slowing behaviors were compared. Second, these behaviors were compared conditional on an anticipatory glance being made (or not made). 5.7.4.1 Aggregate Glance, Braking and Slowing Behaviors Glances. The glancing results here for the bus blocking crosswalk scenario (Straight Segment 3) are similar to the left turning truck at the intersection scenario. Other than the similarity of a large vehicle in the adjacent lane, the glance patterns are also similar. Recall that at the intersection with the left-turning truck, ACT trained drivers had significantly more glances toward the near extent at three, seven and nine 216
seconds before the intersection. Considerably more ACT trained drivers made glances toward the near extent than did placebo trained drivers at the bus blocking crosswalk scenario. (See Figure 40).
Figure 40: Cumulative percentage of ACT trained and placebo trained drivers that made a glance toward the near extent in each second when approaching a bus stopped and partially blocking the view of a crosswalk. *The Asterisks - indicates that the ACT trained drivers made significantly more glances than did the placebo trained drivers in that second.
Significantly more ACT trained drivers made glances to the near extent when three seconds from the crosswalk. There were between a 17 and 40 percentage point difference between ACT trained and placebo trained drivers glancing toward the near extent at each of the one second epochs between one and eight seconds. (See Table 73).
217
Table 73: Percentage of ACT trained and placebo trained drivers that glanced toward the near extent when approaching a bus stopped in front of a crosswalk. (Numbers in bold yellow indicate significant differences.)
Next, consider Straight Segment 3. When approaching a roadside pedestrian on the left and perpendicular truck on the right, similar percentages of ACT trained and placebo trained drivers glanced both to the right and left before arriving near the truck, with ACT trained drivers being only one to 8% more likely to make a near extent glance than the placebo trained drivers. (See Table 74 and Figure 41).
218
Figure 41: Cumulative percentage of ACT trained and placebo trained drivers that glanced toward both a roadside pedestrian to the left and perpendicular truck to the right in each second when approaching the truck.
The difference in the likelihood of glances between the ACT trained and placebo trained drivers was not significant. When I compared the percentage to glance in each second for both groups, the difference between groups was not significant. (See Table 74).
219
Table 74: Cumulative percentage that glanced both toward a pedestrian along the roadside left, and a truck parked perpendicularly along the roadside right before passing the truck.
Speed Loss. In the straight segment scenarios, ACT trained and placebo trained drivers had already slowed to speeds well below the posted speed limit of 40 mph. A likely reason for this is that there were cues to slow when much farther back than ten seconds from the incident locations. In both scenarios, the drivers slowed to a minimum speed and then accelerated to a speed near 23 mph when passing the incident area. First, consider Straight Segment 1. When approaching the bus stopped at the crosswalk, the ACT trained drivers and placebo trained drivers had slowed to a speed of less than 35 mph by the time they were ten seconds from the crosswalk (Straight Segment 1). ACT trained drivers had slowed to an average speed near 21 mph when three seconds
220
from the crosswalk, which was below the target speed of 23 mph. Also at three seconds before the crosswalk, the placebo trained drivers were traveling 22 mph. (See Figure 42).
Figure 42: Average speed of ACT trained and placebo trained drivers when approaching a bus stopped in front of a mid-block crosswalk.
The bus stopped near a crosswalk scenario creates issues that are both obvious and latent. The presence of the lead vehicle and bus are obvious to an approaching driver. However, the possibility of a pedestrian emerging from in front of the bus is a latent hazard. At the bus-crosswalk scenario, there appears to be a difference in the speed choice between the ACT trained and placebo trained drivers although that relationship was not significant.
221
When approaching the two roadside obstacles, the pedestrian to the left and truck to the right, there were no latent hazards. The issue when approaching the roadside obstacles was whether the driver prepared for the possible intrusion of either obstacle or the sudden slowing of the lead vehicle. At the roadside obstacle situation, ACT trained drivers and placebo trained drivers slowed nearly the same way. The only difference of note is that the placebo trained drivers were slightly slower to reach base speed than were the ACT trained drivers. See Figure 43).
Figure 43: Average speed of ACT trained and placebo trained drivers when approaching a roadside pedestrian to the left and truck to the right.
When approaching the bus, drivers also responded to the presence of a lead vehicle. To differentiate when the slowing occurred, I analyzed the percentage of ACT trained and placebo trained drivers that slowed to the target speed of 23 mph in each 222
second while approaching the crosswalk. As the drivers were within two to six seconds before the crosswalk, ACT trained drivers were between 8% and 14% more likely to slow to the target speed of 23 mph than were the placebo trained drivers. When nine seconds from the crosswalk, significantly more ACT trained drivers (33%) slowed to the target speed of 23 mph, while no placebo trained driver had slowed to the target speed at that time (See Table 75). Table 75: Percentage ACT trained and placebo trained drivers that slowed to a target speed of 23 mph in each second when approaching a bus stopped at a crosswalk. (Numbers in bold yellow indicate significant differences.)
Next, consider Straight Segment 3. The bus stopped near a crosswalk creates issues that are both obvious and latent. The presence of the lead vehicle and bus are obvious to an approaching driver. However, the possibility of a pedestrian emerging 223
from in front of the bus is a latent hazard. At the bus-crosswalk scenario, there appears to be a difference in the speed choice between the experienced and novice drivers although that relationship was not significant. When approaching two roadside obstacles, the pedestrian to the left and truck to the right, there were no hidden (latent) hazards. The issue when approaching the roadside obstacles was whether the driver prepared for the possible intrusion of either obstacle or the sudden slowing of the lead vehicle. When approaching the roadside obstacles, typically 20% fewer placebo trained drivers slowed to target speed in each second from three to nine seconds from the crosswalk. The ACT trained drivers were significantly more likely to slow to the target speed than the placebo trained drivers when nine seconds from the crosswalk. (See Table 76). Table 76: Cumulative percentage to slow to a target speed of 23 mph when approaching a roadside pedestrian and truck.
224
When approaching the bus stopped in front of the crosswalk, ACT trained drivers slowed to target speed only slightly earlier but the difference was not significant (ACT trained 11.7 s. (SD = 2.2 s.); Placebo trained 10.8 s. (SD = 2.8 s.). [t (35) = 1.03, P = 0.31). When approaching the roadside obstacles, ACT trained drivers and placebo trained drivers slowed slightly earlier than did the placebo trained drivers. (ACT trained 9.6 s. (SD = 5.2 s.); Placebo trained 7.2 s. (SD = 5.9 s.). [t (35) = 1.31, P = 0.20] Lane Position. When approaching the stopped bus that was to the driver’s right, being farther left would offer greater buffer space and sightline for what might lie ahead of the bus. ACT trained drivers maintained a position slightly more left of the lane throughout the approach toward the bus, but both the ACT trained and placebo trained drivers drove left of center of the lane. The ACT trained drivers remained approximately one foot farther to the left than the placebo trained drivers when six to nine seconds before arriving at the truck. These differences were significant. When within five seconds of the truck, the placebo trained drivers started to move to the left of center as well. (See Table 77).
225
Table 77: Average lane position of ACT trained and placebo trained drivers when approaching a bus stopped at a crosswalk. Distances refer to the position left (negative in parentheses) or right (positive) of the center of the lane. (Numbers in bold yellow indicate significant differences.)
When approaching the roadside obstacles, recall that the truck protrudes into the road on the right and there is a roadside pedestrian left as well as an oncoming vehicle coming from the opposite direction. ACT trained drivers actually drove at a position that was right of the placebo trained drivers. In this instance, the ACT training did not prove to be effective in that placebo drivers were significantly farther left by approximately ½ foot in the last two seconds before passing the truck. (See Table 78).
226
Table 78: Average lane position relative to the center of the lane when approaching a roadside truck to the right and pedestrian to the roadside left. (Numbers in bold yellow indicate significant differences.)
5.7.4.2 Conditional Braking and Slowing and Crashes The conditional probability that a driver slowed given that he or she had glanced toward the near extent when passing the bus can tell us if there was anticipation associated with the drivers’ actions. When within two to four seconds of the crosswalk, a time when a driver was first able to look in front of the bus, 67% of the ACT trained drivers glanced toward the near extent compared to only 42% of the placebo trained drivers. Fewer placebo trained drivers slowed (52%) if they did not glance to the near extent. Conversely, more of the ACT trained drivers slowed if they did not glance (75%). (See Table 79 and Table 80).
227
Table 79: Number of ACT trained and placebo trained drivers that slowed to 23 mph based upon whether they glanced to the near extent two to four seconds before the crosswalk.
Table 80: Percent of ACT trained and placebo trained drivers that glanced to the near extent when two to four seconds before the crosswalk and slowed as they approached a bus parked in front of a crosswalk.
At the roadside pedestrian and truck, 78% of ACT trained drivers and 61% of placebo trained drivers slowed to a speed of 23 mph. Here, the hazards are not latent, but directly ahead, so anticipatory slowing is not surprising. However, there is still the need for drivers to process the roadside pedestrian and truck and 11% more ACT trained
228
drivers made glanced toward each before arriving at that location than did the placebo trained drivers. At the roadside obstacles, the ACT trained drivers and placebo trained drivers slowed more often when there was not a glance to both the pedestrian (left) and truck (right). ACT trained drivers were more likely to slow than placebo trained drivers whether there was a glance (73% to 53%) or not glance (84% compared to 78%). (See Table 81 and Table 82). Table 81: Number of ACT trained and placebo trained drivers that slowed to 23 mph based upon whether they glanced to the near extent two to four seconds before the roadside pedestrian and truck.
229
Table 82: Percent of ACT trained and placebo trained drivers that glanced and slowed as they approached a roadside pedestrian and truck
Conditional Probability related to Crashes. When approaching the roadside truck and pedestrian, no drivers crashed, but five drivers crashed at the location where the bus was stopped in front of the crosswalk. Of the five that crashed, there was one ACT trained driver and four placebo trained drivers. In each of these crash situations, speed choice played a large part in the cause of the crash even when the speed when passing the bus was not as great. Of those who crashed, two of the placebo trained drivers (50%) slowed to the target speed. The ACT trained driver that crashed did not slow to the target speed. Those who were not involved in a crash slowed 94% of the time if ACT trained and 89% of the time if placebo trained. Oddly, many of the drivers that crashed also glanced. The ACT trained driver that crashed also glanced as did three of the four placebo trained drivers. Yet only one placebo trained driver both glanced and slowed to the target speed. The ACT trained drivers that glanced to the near extent and slowed to the target speed of 23 mph were
230
significantly less likely to crash than the novice placebo trained drivers that also both glanced and slowed. (See Table 83). Table 83: Percentage of ACT trained and placebo trained drivers that slowed to a speed of 23 mph based upon whether they crashed or not. (Numbers in bold yellow indicate significant differences.)
I also compared the groups for the probability of a crash based upon whether the driver slowed or did not slow. If they slowed, the ACT trained drivers did not crash. The Placebo trained drivers crashed 20% of the time if they slowed and 25% if they did not slow. The ACT trained drivers crashed 17% of the time if they did not slow. (See Table 84). Table 84: Percentage of drivers that crashed based upon whether they slowed to a speed of 23 mph or not.
231
5.8 Pattern of Glances, Speed Loss, and Lane Position Finally, consider the six locations where drivers crashed in Experiment 2. They include: sharp curve right, left turning truck at intersection, and a bus at a crosswalk. It is of interest to note when the ACT trained drivers differed significantly from the placebo trained drivers in their glances, speed loss and choice of lane position. This is depicted in Figure 44. The differences speak to a tactical plan of action initiated by the ACT trained drivers where anticipatory glances either predict (slowing) or postdict (lane position) the mitigating behaviors that are not prevalent among the placebo trained drivers. The areas highlighted in yellow refer to times when ACT trained drivers were ½ a foot or more in a safer position in their lane than the placebo drivers at all six scenarios, and also the three crash-related scenarios. As an example, referring to Figure 44, where there is yellow in the 9s, 8s, 7 s, 6s, and 0 s columns, this means that when one compares the performances of the ACT and placebo trained drivers at all six scenarios, the experienced drivers maintained a position that was at least ½ foot better than did the novice drivers at those times. Red refers to times when the difference between the percentage of ACT and placebo trained drivers slowing to the target speed was at least 20 percentage points across all six scenarios. There was at least a 15 percentage point difference at each of the six scenarios when one to seven seconds before the incident. The light green refers only to the three scenarios where crashes occurred (sharp curve right, left turning truck at intersection, and a bus at a crosswalk). At each one-second time represented by the light green, there was a 20 percentage point or greater spread between ACT and placebo trained drivers in the percentage of glances to the far extent, the difference being in the favor of the ACT trained. 232
9 s.
8 s.
7 s.
6 s.
5 s.
4 s.
3 s.
2 s.
1 s.
0 s.
Glance Slow Lane_Pos. Figure 44: The highlighted regions represent the times where ACT trained drivers outperformed the placebo trained drivers at the three locations where drivers crashed.
Now, I want to interpret the pattern of lane position (yellow), glances (green) and slowing (red) differences in Figure 44. Imagine that the driver is traveling from left to right. The yellow bars tell us that the trained drivers were much more likely than placebo trained drivers to be moving to a more optimal location in the lane when six to nine seconds before the incident. This movement was based on information from the forward view that we could not measure with eye movements because it was straight ahead, an area where all drivers glanced 10 s before the end of a scenario. From seven to three seconds from the incident, more ACT trained drivers were gathering information by glancing toward the area that was to the extent of the sight line or to a threatening area. Finally, the ACT trained drivers, who were in the safer lane position and had gathered more relevant scene information slowed in response to what they saw in much greater numbers when one to four seconds from the incident. When comparing the ACT trained and placebo trained drivers’ performances at the three locations where no crashes occurred, there was not as great a difference, although ACT trained drivers continued to perform better than the placebo trained drivers 233
overall. At the three sites where no crashes occurred, ACT trained drivers were 20 percentage points or more likely to slow one and five seconds earlier and change position within the lane one to six seconds before the incident. However, the differences in glances to the extent were not significant.
234
CHAPTER 6 DISCUSSION William James, the first American psychologist, used a metaphor of a stream to explain a person’s consciousness (1890). Similarly, drivers are confronted with streaming information. Drivers with different levels of information make use of the information differently. This research explained how consciousness, specifically situational awareness, changed in various driving scenarios. Just as there are ebbs and flows in a stream, so to, there were ebbs and flows in the way drivers responded to the driving environments. The objective variables used to measure the ebbs and flows of situational awareness in this research were speed reduction, lane position and anticipatory glances. The current research was designed to examine the way drivers make anticipatory glances, changes in lane position and speed reductions at selected environments that are associated with greater and lesser crash risk. The goal was to compare novice drivers to experienced drivers in Experiment 1 and ACT trained drivers to placebo trained drivers in Experiment 2. Overall, those drivers with more a priori information, experienced and ACT trained drivers, were more likely to make anticipatory glances, slow more at locations associated with greatest crash risk, and position themselves more safely in the lane. When crash risk is reduced and when the hazard becomes more overt, those with less experience and training begin to respond as do those with more experienced and training.
235
At high risk locations, when hazard anticipation was most essential, experienced and ACT trained drivers exhibited behaviors that suggests there is a direct link between anticipatory glances and anticipatory risk mitigation (speed reduction and changes in lane placement). At the three locations where drivers crashed, those who glanced were cumulatively more were more likely to have slowed to target speed than those who did not glance --[P(S|G) > P(S|nG)] -- and those who slowed were much less likely to crash than those who did not slow --[P(C|S) < P(C|nS)] . Also, those who crashed were much less likely to have made anticipatory glances. Selection of lane position also appears to be associated with those who glanced or slowed to the target speed. ACT trained drivers moved to a significantly safer lane position than placebo trained drivers at Intersection 1 (The left turning truck) and the Straight Segment 1 (The Bus blocking the crosswalk). Both these scenarios were among the locations where crashes occurred in this research. At scenarios where no crashes occurred, only in the last two seconds as drivers traveled past the parked truck at Straight Segment 3 were ACT drivers significantly more likely to select a safer lane position. Overall, the results speak to a tie-in between anticipatory glances, slowing and lane position behaviors. In Experiment 1, experienced drivers were much more likely to make anticipatory glances followed by anticipatory slowing and lane positioning. Consider the findings by Glennon, et al., (1983). Glennon at al. argued that the three seconds before a curve was the critical region. The current research extends that region. These results speak to an eight second critical region, where anticipatory glances, lane positioning, and slowing occurs seven seconds or more before an incident and anticipatory slowing was evidenced from one to five seconds before the incident in 236
Experiment 2. Experiment 2 studied whether novice drivers, who received ACT hazard anticipation and mitigation training, could be taught to understand and implement the anticipatory glances, slowing and lane position tactics within the critical region established by the experienced drivers. Drives in the driving simulator were utilized to evaluate the effects of experience and of ACT training in the virtual world. While novice drivers in Experiment 1 crashed 23 times, they crashed only eight times in Experiment 2 after ACT training. Experienced drivers crashed nine times in Experiment 1 and placebo trained drivers crashed 24 times in Experiment 2. Therefore, each driver averaged approximately one crash per drive. Experienced and ACT trained drivers crashed 0.47 times each, while novice and placebo trained drivers crashed 1.31 times each.
6.1 Experiment 1 In Experiment 1, the hazard mitigation strategies of novice and experienced drivers were compared for the purpose of finding the differences between experienced and novice driver speed mitigation tactics at locations that are known to be over represented in crashes. Drivers were exposed to 27 scenarios which included three curves, three intersections, and three straight segments each driven three times. These 27 scenarios each encompassed approximately 12 seconds of the driver’s time. The scenarios were embedded within a 40 minute drive along with other scenarios of no particular consequence and were presented in a randomized order. In Experiment 1, information was gathered to identify drivers’ proper tactics for reducing crashes and improving safety. The three most common fatal crash arrangements 237
were utilized as locations to measure driver performance. The three most common crash types for novice drivers are single-vehicle off the road, left turn intersection crashes, and rear end crashes. Two curves and two intersections and two straight segments where a rear end crash materialized were selected for my analysis. The ultimate goal of Experiment 1 was to identify the differences in crash mitigation behaviors of experienced and novice drivers. Here, the crash mitigation behaviors were anticipatory glances, lane positioning, and speed reduction. The results were analyzed for the purpose of finding rules that can be taught to novice drivers so a training program could be developed. If the experienced drivers outperformed the novice drivers at a specific higher risk location, that information could be used to tailor a hazard mitigation (crash avoidance) training program, ACT. ACT was evaluated in Experiment 2. 6.1.1 Curves Drivers were evaluated on two of the three curves. All curves are associated with greater crash risk (Zegeer et al., 1990). However, the two curves that were selected represent a curve associated with lower relative crash risk and greater relative crash risk. In this research, as predicted by the literature, there were no crashes on the moderate curve left (lower crash risk curve) and eight crashes on the sharp curve right that were associated with elevated crash risk. At the moderate curve, experienced drivers selected lower speeds, started braking earlier and made earlier glances across the curve. The percentage of experienced drivers glancing to the far extent when three to eight seconds from the curve was at least 27
238
percentage points larger than the percentage of novice drivers doing such. Significantly more experienced drivers slowed to the 34 mph target speed when three to six seconds before the curve. Also, experienced drivers started at a position in the lane significantly to the right of the novice drivers at nine and two seconds before the curve, and moved to a position significantly farther inside the curve when one second from the apex. Between the times when two seconds before the curve to the apex of the curve, experienced drivers shifted 1.8 feet from the outside to the inside of the curve while novice drivers shifted 0.5 feet, but were still right of the lane center when entering the curve. Bonneson and Pratt (2009) indicated the average lane shift is 3.0 feet. At the sharp curve right experienced drivers selected lower speeds, started braking earlier and made earlier glances across the curve. At the sharp curve the experienced drivers’ results show that they better recognized the greater danger posed by this curve. The percentage of experienced drivers glancing to the far extent when three to eight seconds from the curve was at least 20 percentage points larger than the percentage of novice drivers doing such. Even at eight and nine seconds from the curve the experienced drivers were still at least four times more likely to glance to the far extent. At five and six seconds from the curve, experienced drivers were significantly more likely to glance to the far extent. Also, significantly more experienced drivers slowed to the 20 mph target speed when three to four seconds before the curve. Also, experienced drivers started at a position in the lane that was at least ½ foot left of the novice drivers when six, three and two seconds before the curve. Between the times when two seconds before the curve to the apex of the curve, experienced drivers shifted 2.7 feet from the outside to the inside of the curve while novice drivers shifted 2.4 feet. The lane shifts at 239
the sharp curve are more indicative of those reported by Bonneson and Pratt (2009). An example of how slowing to target speed and lane positioning changed relative to glances can be seen in Figure 45. When nearly 50 percent of the drivers from each group have made a glance to the far extent, we can see a noticeable increase in the percentage of drivers that slowed to target speed, and also an improvement in the lane position of drivers in each group.
Figure 45. Glancing, slowing, and average lane position of experienced and novice drivers when approaching a sharp curve to the right. [Red Lines: Proportion of drivers who glance to the far extent right. Black Lines: Proportion of drivers who slowed to the target speed of 32 km/h (20 mph). Blue Lines: Average lane position of drivers when approaching the sharp curve to the right (negative is left). Each time represents the time before the curve.]
240
When we consider the actions at the two curves, the experienced drivers moved from a position at the outside of the curve and moved to the inside of the curve. The novice drivers did so only at the sharp curve. Moving from the outside to the inside of the curve flattens out the curve and allows a driver to maintain better control of the vehicle once they enter the curve. At the outset, before entering the curve, the experienced drivers were already in a position to better control the vehicle, yet the experienced drivers did not stop there. If I collapse the results from the two curves together, the experienced drivers were 20 percentage points more likely to glance to the far extent when four to seven seconds from the curve. Also, experienced drivers were 23 percentage points more likely to glance to the far extent at the sharp curve when eight seconds from the curve. It is no surprise that the experienced drivers slowed to target speed much more than novice drivers at both curves as well. In the five seconds before the curve, the experienced drivers were much more likely to slow to target speed than the novice drivers at both curves. Regardless of the curve left or the curve right, there was no significant difference in the speed of the experienced and novice drivers when at the apex of the curve or when more than seven seconds from the curve. These results show that the drivers were initially driving similar speeds and that the speed loss that we see was due to the anticipation of the potential hazards. When near the apex of the curve, the novice drivers were more likely to brake hard in the last second. The last second hard braking by some of the novice drivers increased the likelihood of those drivers to slow to the target speed in the last second before the apex of the curve.
241
Six novice drivers crashed at the sharp curve compared to two experienced drivers. Of those who crashed, one slowed to the target speed of 20 mph and none of these drivers glanced to the far extent and slowed. Conversely, at the more moderate curve to the left, no drivers crashed. Based upon mathematical model that predicts crash risk at curves (Anderson and Krammes, 2000), these results are not surprising in that the curve to the right was associated with a crash risk that was 4.4 times that of the curve to the right. The curves in which no glance was made, but experienced drivers slowed corroborates the findings by Horrey et al. (2006) that some (experienced) drivers are able to process the environment ambiently. Newly licensed drivers clearly needed to glance across the curve before slowing. Nearly four of five drivers who glanced across the curve also slowed adequately. Until the time newly licensed are able to process the Gestalt of the scene, these drivers must be taught to make direct glances that lead to slowing. Comparison to Related Results in the Literature. At the sharp curve to the right, when ten seconds from the curve the average speed of drivers was 33 mph which was considerably less than the posted 40 mph speed. These findings are consistent with the equations that predict the distance before a curve that slowing has begun (Mikolajetz et al., 2009). Based on the Mikolajetz models we would expect some drivers to start slowing when as far back as 15 seconds before this curve. At the sharp curve right, drivers had already slowed 7 mph when more than nine seconds from the curve. At the moderate curve to the right, the Mikolajetz model predicts slowing to begin 4.9 seconds before the curve with a 95th percent confidence interval between 3.0 and 6.7 seconds. When six seconds from the curve, 50% of the experienced drivers slowed to the 242
target speed (were on target to slow to 20 mph). Only 35% of the novice drivers slowed to the target speed. According to the Mikolajetz model, the novice drivers performed worse than the 5th percentile performers and the experienced drivers were above average but responded within the predicted range. At the moderate curve to the left, novice drivers failed to slow to the baseline set by the Mikolajetz and Bonneson models. The equations by Bonneson and Pratt (2009) offered a reliable prediction of the slowing by experienced drivers in this study, but under predicted the slowing of novice drivers in this simulator study. At the curve right, the Bonneson model would suggest that near 50% of the drivers would slow to a speed of 20 mph at the curve. Again, the drivers in this research responded as would be expected with 78% of the experienced drivers slowing to the target speed and only 43% of the novice drivers slowing to the target speed. At the moderate curve left, 89% of the experienced drivers and 35% of the novice drivers slowed to the target speed. If we were to combine the results from both groups and both curves, we can see that the Bonneson model offers a relatively accurate estimate of the slowing of drivers if applied to these results from the driving simulator. We might also say that the drivers in this research slowed similarly as was the case in field research. The glances by the experienced drivers in this experiment were consistent with the findings by Land and Lee (1994) and Suh et al. (2006). Most of the experienced drivers had glances dwell five to ten degrees to the right when approaching the curve right or five to ten degrees to the left when approaching the curve left. Most of the glances by the novice drivers when approaching both these curves remained in the area of the inside fog line which is not as far across the curve as was found by either Land et al. or Suh et al. 243
Glennon et al. (1985) referred to the three seconds before the curve as the critical region. These findings are consistent with his findings relative to speed loss. However, anticipation must be considered as well. In this research, drivers who made anticipatory glances four to eight seconds before the curve were more likely to slow to target speed when within Glennon’s three second critical region than drivers who did not make such anticipatory glances. In the current research, these curves were sharper than the typical curve a driver might negotiate. At these sharper curves, drivers slowed earlier than predicted by Glennon. Glennon’s critical region did not consider some of the other factors that might cause the critical region to increase or decrease. Thus, the critical region should also be expanded to include the time for anticipating. 6.1.2 Intersections Intersections offered reinforcement to the findings at curves but with a twist. Rather than speed loss following far extent glances, at the intersection, glances to the near extent followed speed loss. Clearly, the glances to the near extent were not the glances that prompted the speed loss (which occurred earlier), but they are an indication that the prior slowing was not random but instead was based upon some earlier act of hazard anticipation. At the intersection where the left turning truck obstructed the intersection, experienced drivers made significantly more glances to the near extent when three, seven, and nine seconds before the intersection. At all times from three to nine seconds before the intersection, experienced drivers were much more likely to make glances to the near extent than were novice drivers.
244
On the basis of these results, glances and speed loss are reasonably linked in that experienced drivers slowed to target speed significantly more often when one to four seconds before the intersection. Experienced drivers also selected lane positions that were farther right when passing the left turning. Also, the experienced drivers crashed six times, but the novice drivers crashed twice as often, so much so that 67% of the novice drivers crashed at this intersection. At the busy intersection, experienced drivers remained left of center more so than the novice drivers when approaching the intersection. They also slowed slightly earlier and were more likely to make glances toward the traffic on each leg of the intersection before making a turn. Novice drivers were more likely to start the turn earlier rather than delayed apexing (Hough, 2008). Delayed apexing is when a driver moves farther forward before turning. Delayed apexing allows the driver more time to scan for other traffic as well as limits the time the driver is exposed to conflicting traffic when turning. Fewer than half the novice drivers made a glance toward all three legs of the intersection before turning. It was surprising that a novice driver would be willing to take such a gamble, particularly considering that there was moving traffic on every leg of the intersection at some time before the drivers turned. The average turning speed of all experienced and novice drivers (if turning without stopping) was 18 mph. As a comparison, I compared the drivers with the assumption that 18 mph was the target speed (rather than 0 mph). When doing so, novice drivers reached the target speed of 18 mph at nearly equal percentages as the experienced drivers. Therefore, it appears that the novice drivers set the anticipatory speed at a speed slow enough to turn safely without regard of a possible need to stop. 245
There was not a significant difference in speed choices of the novice and experienced drivers at the busy intersection for two primary reasons. First, the potential hazard of making a left turn and the intersection were laid out in front of the drivers. While latent hazards existed, the potential hazards that were available directly ahead created cues that caused the novice drivers to brake at a rate near that of the experienced drivers. Secondly, the busy intersection did not have the same crash risk that the left turning truck scenario had. It appears that there is a relationship between the differences in speed loss by experienced and novice drivers and the crash risk associated with the scenario. At scenarios where no driver crashed, the differences in the speed loss of novice and experienced drivers is relatively small. At scenarios where there were the most crashes, the speed differences between experienced and novice drivers was greatest. Implications. At the intersection where there was a left turning truck (Intersection 1), the difficulty here for many drivers was that because the traffic signal showed green they somehow were in a protected state. A green ball traffic signal is a permitted signal meaning that opposing traffic may turn. Also, most driving laws require drivers to yield to vehicles and pedestrians within the intersection, even with a green signal. Nothing here is absolving a driver traveling the opposite direction from making sure it is safe before turning, but driving into an intersection with an obstructed view at nearly full speed is an extremely unsafe act as shown by the stunning number of novice drivers that crashed in this scenario. At the busy intersection, a hazard could emerge from any of the other three legs. For truck drivers, there is the added danger of traffic approaching from the blind spot behind the truck when making a wide turn. Potentially, there are at least three and 246
sometimes four legs of the intersection that must be conceptually neutralized before attempting a turn. The experienced drivers were much more likely to make these glances than were the novice drivers. Furthermore, the additional glances take additional time, the additional time apparently translated in a later turn. The experienced drivers started the turn later in the event. Evidence of a later turn can be seen in the lane position results. At the entrance of the intersection (time = 0 s), the novice drivers were a full 1.75 feet left of the average position of the experienced drivers. While a driver should strive to move left within the lane before a left turn, the average position of the experienced drivers was 2.2 feet left, which would place them just right of the center line. However, the novice drivers were 4.0 feet left of center of the lane, which placed them partially into the opposite lane before entering the intersection. In addition to more glancing time, a later turn (delayed apexing) reduces the time the turning driver is in a conflict zone. There are two ways to reduce crash probability: first, crash avoidance tactics, and second, reduced exposure. The experienced drivers exhibited delayed apexing with additional glance time and by starting the turn later in the event; both actions would likely reduce crash risk (See Figure 46).
247
Figure 46. An example of delayed apexing versus a sweeping turn that starts much earlier and how each turn tactic remains within the zone where it is most likely to be in conflict with other vehicles or pedestrians. (Yellow highlight is the conflict area; green highlight is the safer region).
6.1.3 Straight Segments Slowing. In Fisher et al.’s 2002 paper, they too examined the speed loss of drivers at the vehicle blocking crosswalk scenario. Not unlike the findings by Fisher et al. (2002), here too there was no significance in speed choices between the experienced and novice drivers at the straight segment scenarios. In the current experiment drivers also had the additional cueing of the lead vehicle. Fisher et al. later reasoned that the difference was due to non-homogeneity and after removing those drivers that stopped and those that did not slow at all, a significant difference was found. A concern with this approach is that crashes might be caused by the anomalous responders. Even still, I examined this to determine if there was some additional explanation.
248
I eliminated six experienced and six novice drivers that slowed to a speed less than three mph. I also eliminated two novice drivers that did not slow at all, and one experienced driver that did not slow. Still, experienced drivers reduced speed more than novice drivers but there was no significant difference between the groups [t (55) = 1.88; P = 0.06]. Glances. The ability to use the output from the eye tracker to determine on which object exactly a driver is fixating begins to diminish when the driver is ten and more seconds from the object, yet drivers had already significantly reduced speed when within this time window. Thus the ability to discover a direct relationship between early glances and early speed loss was not feasible. However, here, a glance can be used as an indicator that the driver did not simply reduce speed ambiently, but through conscious effort. When at the roadside truck and pedestrian scenario, although the difference did not reach significance, experienced drivers were nearly 1.96 times more likely to glance toward both the truck and pedestrian in each second from two to seven seconds from the truck. At the bus blocking crosswalk scenario, experienced drivers made 1.51 times more glances toward the far or near extent than novice drivers when two to seven seconds from the crosswalk. It is in this region that a driver should begin preparing if the lead vehicle should have to stop suddenly for a passenger emerging from in front of the bus. It is also the time where drivers, if anticipating, would be glancing around the edges of the bus. Given such a large difference between groups, one might expect there to be a significant difference, but here, many experienced drivers made glances in several onesecond bins, while a few novice drivers made a single glance in a single time bin. There 249
does appear to be a factor relative to glance density or frequency that was not addressed in this paper. Not only did more experienced drivers make earlier near and far extent glances, they also made far more glances toward the forward roadway. It was not uncommon for a novice driver to have glanced toward a hazardous region, but do so only once or twice compared with the typical experienced driver who would be more likely to make several glances toward a dangerous area. Conditional Analysis. First, consider both the conditional probability that experienced and novice drivers glance to the near and far extent and that experienced and novice drivers slow. Experienced drivers were 1.56 times more likely to glance toward the extent of the sightline and slow to target speed in the bus crosswalk scenario and the roadside obstacles in the roadside truck and pedestrian scenario than were the novice drivers. Next, consider the conditional probabilities when the drivers glanced versus when there was not a glance. If the novice drivers glanced, the probability of slowing decreased and when they did not glance, the probability of slowing increased. When the hazard is ahead and within the sightline, there is not the need for focal attention. Indeed, these novice drivers were able to appreciate the need to slow without a direct glance when all the available information was laid out in front of them. Unfortunately, this type of situation is more common than the other scenarios in this research, which might lead these drivers to be conditioned to believe that ambient attention might suffice for all driving situations. Among the experienced and novice drivers, the probability of slowing stayed the same for the roadside truck and pedestrian scenario, and for the bus scenario. These 250
results might suggest that the experienced drivers slowed for the lead vehicle or bus (earlier in the event) and not for the possibility of a pedestrian (later in the event), thus had no more slowing than the novice drivers later in the event. However, Fisher et al. (2002) came to the same findings. Fisher et al. compared the speed loss of experienced and novice drivers at a truck blocking a crosswalk scenario and did not find a significant difference. The lack of a difference in slowing between the experienced and novice drivers can be traced to two likely causes. First, any anticipatory slowing for a potential pedestrian intrusion was likely masked by the slowing for all the other cues that were present. Secondly, all the initial cues, such as the slowing bus, and the lead vehicle changing lanes offered cues to an approaching driver that could be processed ambiently and did not require a great deal of situational awareness. Again, novice drivers likely face situations like this (where all hazards can be processed ambiently) which might lead them to believe that the forward driving field can be processed ambiently or intermittently. Instead, the significant decrease in speeds by both groups implies that all drivers, regardless of experience, recognized the situation as one in which slowing was necessary. A likely reason for this recognition is ambient vision and early detection. Most drivers recognized the need to slow well before the ten second window of this research and others processed the hazards ambiently (without a direct glance). Thus, there was no measurable relationship between glances and slowing. Instead, maintaining a safe following distance and being prepared for hazards was the best approach. Crashes. Finally, consider crashes at the straight road segments. There are several reasons a driver might crash. While novice drivers’ errors appear to be related to 251
lack of knowledge at curves and intersections, at the straight road scenarios, the errors (crashes) appear more related to slips. All of the drivers who crashed also glanced, but only 60% of the novice drivers that crashed had slowed to a target speed. In each of the straight segment crashes, the hazardous situation generally unfolded when directly ahead of the driver and the drivers glanced toward the hazardous areas. The straight segment crashes cannot be attributed to inadequate glances, but instead a failure to maintain a safe buffer zone from other vehicles. Here, the crash drivers saw the hazard but failed to respond appropriately. Accordingly, training drivers to keep a proper buffer space and safety zone might be of value. 6.1.4 Implications of All Findings in Experiment 1 Not surprisingly, when a driver fails to heed a potential hazard, the potential hazard might develop into an immediate hazard and crash situation. Such an instance occurred throughout this research. Surprisingly, of the 46 novice driver crashes only 30 were when the hazard was materialized. In 16 instances, the novice drivers crashed with no provocation. In plain language, the novice drivers crashed without the help of experimental manipulation. Simply mixing a known hazardous location with an inexperienced driver created a dangerous mix. Yet drivers who both glanced toward the most threatening areas and slowed to target speed were significantly less likely to crash. These results speak for an action-reaction (glance & slow) combination that must be taught to newly licensed drivers. When the results are examined, it appears that the only time when speed was significantly different was at locations where the hazard was latent or glances were not made to the extent of the sightline. For instance, where the hazard materialized directly 252
ahead at the straight segments or busy intersection, there were no significant differences in slowing to target speed. At these three scenarios, the only evidence of a difference in hazard anticipation was that experienced drivers selected significantly better lane positions when approaching a roadside truck and when entering the busy intersection. Experienced drivers also made significantly more glances toward the side roads when two and one second before the turn at the busy intersection. Overall, when the hazard could be seen ahead, the differences between the experienced and novice drivers were relatively subtle, but not so when more hazard anticipation was required. Yet when approaching either curve or the intersection that was obstructed by a left turning truck, a driver was required to anticipate a hazard to safely negotiate the area. At these latent hazard locations, the differences between the experienced and novice drivers were pronounced. Experienced drivers were significantly more likely to glance to the near or far extent at the intersection with the left turning truck (Intersection 1) and when approaching the sharp curve to the left than were novice drivers. Experienced drivers selected significantly safer lane positions at the curve left and the intersection with the left-turning truck, and significantly more experienced drivers slowed to the target speeds at all three of these locations. Still, when the novice drivers slowed to target speed they were three times less likely to crash when compared to when they did not slow to the target speed at the sharp curve right and intersection with the left-turning truck. Interestingly, of those experienced drivers that crashed, they were nearly eight times more likely to crash if they did not slow to the target speed at the sharp curve and intersection with the left-turning truck. These results speak to the need to anticipate a hazard, a skill not yet mastered by novice drivers, but also, the need for anticipatory 253
slowing and risk mitigation. While significantly more experienced drivers slowed to target speed than did novice drivers at these three scenarios associated with hazard anticipation, when a driver did not slow to target speed the probability of a crash increased significantly regardless of experience. The straight segment scenarios allow most of the information a driver must process to be relatively straight ahead. As evidence of this, we can see that even as early as ten seconds from the truck (straight segment 2) or the crosswalk (straight segment 1), the average speed was reduced to 26 mph or less. As Horrey et al., (2006) pointed out, drivers are able to control vehicle position, and presumably speed, with ambient vision. Ambient vision implies that a driver could control the lane position of the vehicle without a direct glance. However, recognition required focal (direct) vision. In the straight segment, the presence of the bus (straight segment 1), truck (straight segment 2) and lead vehicle (both straight segments) offered information to the drivers that could be processed ambiently and from great distances. As a result, we see that novice drivers almost match the experienced drivers in glances and speed reduction. Still, when focal vision tasks were necessary, such as anticipating a pedestrian emerging from in front of the bus and the ability to process the additional information and traffic movements, the novice drivers were three times more likely to crash. The simple findings here were that experienced drivers were much more likely to glance five to nine seconds before the event, slow one to five seconds before the event and move within the lane both eight and nine or one to three seconds before the incident. The combination of the glances, slowing and lane position movement speaks to
254
anticipatory tactics that experienced drivers have established that should be taught to novice drivers. Some experienced drivers that did not make a direct glance apparently devote some cognitive activity to the surrounding enough so that half still slowed adequately. The no-glance and slowing by some experienced drivers corroborates the findings by Horrey et al. that some drivers are able to process the environment ambiently. To account for ambient driving, drivers should also be taught to slow at specific high risk locations such as obstructed intersections, obstructed crosswalks, and curves. Unlike the experienced drivers who, at times, showed an ability to process the environment ambiently, novice drivers clearly needed to glance across the curve before slowing. At the three scenarios where crashes occurred, two of every three experienced and novice driver who glanced toward the near or far extent also slowed adequately. Until the time newly licensed drivers are able to process the Gestalt of the scene, these drivers must be taught to make direct glances that lead to slowing.
6.2 Experiment 2 Experiment 2 had two primary components. First, from the literature review and the results of Experiment 1, a PC-based computer program was written to teach novice drivers some of the tactics for safe driving at high risk scenarios that were exhibited by the exemplary drivers in Experiment 1. After the program was written and pilot tested, the training was evaluated using a pre-test/posttest comparison, as well as an evaluation driver in the same virtual worlds as were driven in Experiment 1. 6.2.1 ACT Training Program 255
Given that novice drivers do not mitigate hazards as well as experienced drivers, the next goal was to develop a training program that can be used to teach newly licensed drivers to better mitigate hazards. The results from Experiment 1 show that a prerequisite for hazard mitigation is hazard anticipation. Also, some drivers, who either do not appreciate the need to glance or have difficulty making the necessary anticipatory glances, should also be taught to slow at locations that are known to be associated with elevated crash risk. The ACT training taught simple rules. Specifically, slow for HRECCS: hidden obstacles, roadside obstacles, places where there is no escape route, situations where one is closing on a lead vehicle and passing is not a possibility, curves, and traffic signals. Drivers were taught to maintain a safety bubble and to glance toward potentially dangerous areas or to the extent of the sight line. The concept of a safety bubble teaches drivers to select safer lane positions. Participants were told to begin braking at least eight seconds before an incident. While this is earlier than some drivers, it was clear from Experiment 1 that many drivers began braking at least this early. Therefore, it is an effective goal for a driver. Drivers were also taught to utilize the horn to warn other traffic when they might not be visible. In addition to hazard mitigation tactics, and because hazard anticipation has been strongly associated with hazard mitigation, many of the training scenarios contained hazard anticipation training modules. 6.2.2 Simulator Evaluation of ACT
256
The results show that the ACT trained drivers were indeed less likely to crash, more likely to glance toward threatening areas and more likely to slow before a curve than the untrained or placebo trained drivers in, respectively, Experiments 1 and 2. Participants who completed the ACT training performed significantly better on the post test at identifying the proper lane position, slowing and braking locations and glance locations at nine potentially hazardous scenarios. Of the nine potentially hazardous scenarios, they encompass the eight most common crash configurations that 16 and 17 year old drivers are involved in and 60% of all crashes by that age group. Those who were randomly assigned to the placebo training did no better or worse on the posttest than those assigned to the ACT training pretest. Curves. At the sharp curve to the right, ACT trained drivers were significantly more likely to glance toward the far extent when five to seven seconds from the curve. At the moderate curve to the left, the differences were not significant, but the difference between the ACT and placebo trained drivers that glanced to the far extent was 14 percentage points. At the sharp curve to the right the difference between the ACT trained drivers and placebo trained drivers who slowed to the 20 mph target speed was between 14 and 29 percentage points. At the curve left, the ACT trained drivers were significantly more likely to slow to the target speed of 34 mph when six to eight seconds from the curve. Relative to lane position, ACT trained drivers started at a position farther to the outside of the curve (two seconds before the curve) and moved farther to the inside of the curve (at the apex) than the placebo trained drivers at both curves. This lateral shift
257
allows a driver to reduce the lateral friction and maintain better control of the vehicle. According to Bonneson and Pratt (2009) drivers exhibit this lateral shift in a curve to flatten out the radius of the turn, which allows the driver to have more control at higher speeds. At the two curves, ACT trained drivers were more likely to slow if they glanced to the far extent. However, the placebo trained drivers were less likely to slow if not glancing at the sharp curve right than if glancing. At the curve left, placebo trained drivers slowed more when they did not glance toward the extent than when they did. In either case, the placebo trained drivers’ probability of slowing was not better or worse than the ACT trained drivers. At the curve right, ACT trained drivers and drivers who did not receive the ACT training differed by more than ½ foot in average lane position when one and three seconds before the curve right, and when zero, eight and nine seconds before the curve left. At both curves, ACT trained drivers implemented a larger lateral lane position change which shows that they were utilizing tactics that would allow them to better control the vehicle once they entered the curve. When examining conditional probabilities, slowing was not always associated with glances. At the curve right 89% of the ACT trained drivers and 100% of the placebo trained drivers slowed to the target speed of 20 mph. Clearly, slowing was associated with an anticipatory glance. However, only two of the 18 placebo trained drivers glanced to the far extent. When there was not a glance to the far extent, the percentage of drivers that slowed dropped to 71% for the ACT drivers and 62% for the placebo trained drivers. The results for the ACT trained drivers are similar to those by the experienced drivers in 258
Experiment 1 and are a notable improvement on the performance of the novice drivers in Experiment 1. At the curve left, while similar numbers of ACT trained and placebo trained drivers made glances to the far extent, many more ACT trained drivers utilized the information gained and slowed. Of those who glanced to the far extent at the curve left, 92% of the Act trained drivers slowed to the target speed of 34 mph compared to only 57% of the placebo trained drivers. These results are better than those from the experienced drivers in Experiment 1. Intersections. Unlike a curve where slowing is necessary to maintain or conserve frictional capabilities to maintain control, drivers at an intersection with a green traffic signal have no issue with vehicle control. The issue here is an inability to adequately anticipate hazards due to the restricted sightline. Braking offers the driver more time to survey the intersections as well as reducing response time by eliminating the need for leg movement time. The ACT trained drivers reduced speed by an average of nearly 7 mph at the obstructed intersection when the average speed loss by the placebo trained drivers was less than 2 mph. The differences in speed loss for each second are quite extraordinary. This reinforces the saying that ignorant people are only ignorant if they don’t know what they are ignorant about. Here, the ACT trained drivers were taught how to respond to a situation like the left turning trucks at an intersection. The placebo drivers did not receive hazard mitigation training and the difference was nearly 6 mph. If a driver needs two seconds to respond to a crash, a vehicle will travel almost another 18 feet if traveling
259
6 mph faster. Also, the crash rate for the placebo drivers was three times greater than for the ACT trainees. Also of note is the similarity in the data between Experiment 1 and Experiment 2. When we compare those who slowed to the 37 mph target speed, ACT trained drivers in Experiment 2 slowed slightly more often (72% to 68%) than the experienced drivers at the intersection with the left-turning truck. At the busy intersection more ACT trained drivers slowed to the target speed of 0 mph than experienced drivers (78% compared with 59%). Conversely, the novice drivers who did not receive ACT training (placebo drivers) were much less likely slow to target speed with only 30% slowing adequately at the intersection with the left turning truck and 61% slowing to the target speed of 0 mph at the busy intersection. At the intersection with the left turning truck, ACT trained drivers positioned themselves in a safer position by at least ½ foot when compared to the placebo trained drivers when four to six seconds from the intersection. ACT trained drivers made more anticipatory glances, were more likely to slow, and even selected better lane positions than did the placebo trained drivers at both intersections. The conditional probabilities show that ACT trained drivers that glanced to the near extent as they approached the intersection were much more likely to slow to the 37 mph target speed than were drivers who did not make an anticipatory glance. ACT trained drivers were also more likely to slow to the target speed and make an anticipatory glance than the placebo drivers whether or not an anticipatory glance was made.
260
Straight Segments. At the bus blocking crosswalk scenario, the ACT trained drivers slowed only seven percentage points more than the placebo trained drivers. Yet ACT trained drivers were more likely to slow to the target speed of 23 mph, more likely to glance to the near extent, and more likely to select a safer lane position. At the location where a truck and pedestrian were at the roadside, similar numbers of ACT and placebo trained drivers glanced toward the truck and pedestrian, yet ACT trained drivers were 20 percentage points more likely to slow to the target speed of 23 mph than were the placebo trained drivers. Also, the ACT trained drivers selected safer lane positions when passing the truck. Repeatedly, the ACT trained drivers’ performances were better than those of the placebo trained drivers. 6.2.3 Training Implications of the Research The results from this research show that rule-based training, specifically, hazard anticipation tied with hazard mitigation and rules for slowing can be an effective method of training novice drivers to select better speed management strategies. The ACT program was such a tool. This research has focused on the ACT training relative to speed management and hazard anticipation. Future research might examine the program as a tool for teaching optimal horn use, and lane positioning. Also, in instances where the hazards are obvious, such as along the roadside, a failure to act might be more associated with a mistake rather than lack of knowledge. In such instances FOCAL (Attention maintenance training) might also be beneficial. The PC-based ACT training is portable and relatively easy to run. While it might not have the full fidelity of the driving simulator, it does offer similar concepts. The
261
users in this research showed evidence of understanding by the performances on the posttest. Clearly, users are shown traffic scenes when approaching known high risk locations. It is reasonable to believe that the ACT training program will be a cost effective way to teach eye tracking, speed management, lane management, and horn use without the need for the expensive eye tracking and driving simulator equipment. Finally, in a posttest questionnaire, drivers were asked to comment on the ACT training. Most drivers did not comment. One driver stated that the training was “a bit too long”. Four others stated that they thought the posttest was difficult and three made comments similar to saying the training was fun, or challenging. Two stated it reminded them of a video game (both explained that this was a positive comment). 6.2.4 Other Implications This research showed there was a strong link between drivers’ glances to the near and far extent and subsequent speed loss. At intersections the speed loss precedes the glance and at curves the glances precede the speed loss. In either case, both speed loss and glances to the extent of the sight line are behaviors that point to anticipatory thoughts by the driver. On the other hand, at locations where there was a reduced crash risk, the differences in glances and speed loss between experienced and novice drivers, and later between ACT trained versus placebo trained drivers, is diminished. The inclusion to specific speed management training was shown to be effective as well. At instances where slowing was not strongly associated with anticipatory glances, ACT trained drivers typically slowed more often than did those who did not receive training. Thus, speed and lane position management along with hazard anticipation
262
showed to be an effective way to train novice drivers to respond and avoid hazardous situations and crashes in a simulated environment.
6.3 Comparison of Experiment 1 to Experiment 2 The results as they stand are encouraging, but further, there appeared to be a heightened state of awareness by the placebo trained drivers. The ACT training posttest showed many similar situations as were faced in the driving simulator. Yet, some of these results explain the difference between hypersensitivity and actual anticipatory actions. An example of this can be seen in the moderate curve left scenario. In Experiment 1, 44% of the novice drivers slowed if they glanced to the far extent and 37% slowed if they did not glance. In Experiment 2, after taking the placebo training and ACT posttest (with three curves scenarios), 57% of the placebo trained drivers slowed to the target speed after a glance to the far extent (an insignificant improvement) and 82% slowed to the target speed without a glance. Slowing without an anticipatory glance suggests that these drivers, if hypersensitive to slowing, still did not appreciate where they should be looking or where to place the vehicle within the lane. The descriptive results in Experiment 2 strongly and repeatedly show that the ACT trained drivers performed better and safer than did the placebo trained drivers. However, when we consider all the results, the ACT trained drivers fell short of the glancing, slowing, and lane choice selections of the experienced drivers in Experiment 1. This finding was not surprising in that it is possible to train drivers to implement better driving tactics, but it is not possible to endow novice drivers with more experience.
263
6.4 Applications and Generalizability of This Research Future research should allow the participants to complete all three training programs: RAPT, FOCAL and ACT. Clearly, there is an association between glances and speed. Several studies have shown that drivers that are engaged in greater mental workload, such as cell phone tasks have a tendency to drive slower. Cut outs (driveways) are also associated with slower speeds. The prior research suggests that the more processing and glances, the lower the speed. Hazard anticipation training alone could clearly cause a reduction in speed. However, many of the findings here show that hazard anticipation alone did not account for many instances of speed loss, including the several instances where drivers slowed without glancing to the hazard or extent of the sight line. These results show that experienced drivers were more likely to glance toward the extent of the sightline or a hazard ahead when five to seven seconds before the incident. Experienced drivers were also more likely to slow when within three seconds of an incident and to move within the lane to optimize the sightline and minimize the lateral friction necessary. Essentially, in Experiment 1 I learned that experienced drivers are more likely to slow for HRECCS. In Experiment 2, novice drivers were taught these concepts with the use of the ACT PC-based computer program. The ACT trained drivers were taught to select slowing options when eight seconds from HRECCS and braking as an option when within three seconds or if two or more of the items in the acronym HRECCS were present. Drivers were also taught how to select a lane position that would allow for the
264
largest safety bubble and to glance into areas that are at the extent of the sight line or the next most hazardous threat. The training was evaluated on a driving simulator where nine scenarios were driven three times each. Among the nine scenarios were scenarios that are the most representative of the crashes in which novice drivers are involved. Of these scenarios, two curves, two intersections and two straight segments that represent a higher risk and lower risk scenario were selected as locations to be evaluated. ACT trained drivers repeatedly showed more safe speed choice, lane position choice and glancing behaviors than did placebo trained drivers in Experiment 1 or Novice drivers in Experiment 1. These results show that experienced and trained drivers are more likely to glance towards a potentially hazardous region when five to nine seconds from the event and that they move within the lane as far back from the event as nine seconds. Experienced drivers and ACT trained drivers are also likely to begin slowing when within two to five seconds of the incident. In many instances fewer than half the placebo trained drivers and novice drivers made anticipatory glances and slowing actions at potentially dangerous situations. The lack of anticipatory tactics was a major reason why these groups were overrepresented in crashes in this research and a likely reason teenagers are over represented in the crash scenarios studied in this research. There are two remaining issues to discuss. First, it is understood that a fixedbased driving simulator cannot replicate the feel that a driver has when negotiating a curve. However, this research addressed driver anticipation (glances) and pre-curve
265
responses (target speeds) before lateral g-forces come into play. There might also be factors associated with intersection and straight road segment scenarios that are dissimilar in a virtual world when compared to being in the field. Additional research in the field might be able to uncover additional differences between experienced and novice drivers. As an example, there might also be differences in the choice and amplitude of the maneuver after negotiating the apex of a curve (Maeda et al., 1977). Finally, one can ask whether training might be able to decrease run-off-road crashes. Given that failures of hazard anticipation were clearly implicated as a major cause of ROR crashes (at the sharp curve right), and left-turn-across-path movement crashes (at the intersection with the left-turning truck) for novice drivers and given that hazard anticipation training has proven to have effects both on a driving simulator (Pollatsek et al., 2006) and in the field (Pradhan et al., 2006), effects that can last up to a year on the open road (Taylor et al., 2011), it would appear that it should be possible to train novice drivers to anticipate the hazards these scenarios present. As seen in this study, it is also the case that novice drivers did not slow enough (mitigate the hazard appropriately) even when they anticipate that hazard (glance at the hazard). The results from this and prior research suggest that novice drivers can be trained to slow appropriately and select safer lane positions if they anticipate a given hazard (Fisher et al., 2002).
266
APPENDIX A INFORMED CONSENT
INFORMED CONSENT Participant Name: ______________________________________________________ Principal Investigator: Donald L. Fisher________________________________________ Sponsor: University of Massachusetts, Amherst____________________________ Project Title:
HARMLESS
WHAT IS THIS FORM? Welcome to the Human Performance Laboratory. This is an Informed Consent Form. It will give you information about the study so you can make an informed decision about participation. a) If you are under 18, then both you and one of your parents need to sign this form. b) If you are 18 years old or older you do not need to have a parent sign this form to participate in this study. Your signature on this form indicated that you are giving your permission to participate in this study. 2.
WHO IS ELEGIBLE TO PARTICIPATE?
You have been selected to participate in this study because you are a fully licensed driver who is either (1) under age 18 and licensed for less than 8 months, or (2) Between ages 25 and 60 without a crash, infraction or driving arrest within the past five years. 3.
WHO IS SPONSORING THIS STUDY?
The National Institute of Health is sponsoring this study. 4.
WHAT IS THE PURPOSE OF THE STUDY?
You will be participating in a research project designed to better understand how you drive in various environments. As a result of your participation in the experiment, it is hoped that researchers will have a better understanding of drivers’ behavior. Knowing more about your driving behavior can potentially lead to improved driver training curriculum.
267
5.
WHERE WILL THE STUDY TAKE PACE AND HOW LONG WILL IT
LAST? The experiment will take place in the Human Performance Laboratory at the University of Massachusetts, Amherst. It is expected that 48 subjects will be enrolled in this study. The entire study is expected to last a total of two months, but your participation in the study should exceed no more than two hours. 6.
WHAT WILL I BE ASKED TO DO?
You will be asked to fill out a questionnaire, both before and after the experiment. First, you will be asked to complete a pretest questionnaire. The questions are to determine if you qualify for the experiment and if so, whether your individuality influenced your performance. Second, you will participate in the driving experiment where you will be seated as the driver in an actual car (see figure to right). You will be asked to drive a practice driver, followed by four experimental drivers. In each scenario, your task will be to remain on Route 66. In the subsequent four drives you will be asked to look for signs that will provide you information about whether you need to remain on the current road or to turn. We will ask you to take appropriate action (i.e., turning or staying on the current road) as soon as you see these signs while obeying all traffic laws and posted speed limits to the best of your ability. If you fail to change lanes, the experimenter will ask you to do such after passing a sign. In addition, before you begin driving, you will be fitted with a head mounted optical device (see figure at right) and complete a calibration process. During the calibration process you will be asked to keep your head still while looking at specific points displayed in the screen in front of you. Once the calibration process is completed you can then look freely anywhere in a scenario as you try to negotiate the traffic. In order for you to get used to the eye tracker and the driving simulator, you will be given a practice drive. Finally, you will be given the test drives. 7. ARE THERE ANY RISKS OR BENEFITS ASSOCIATED WITH PARTICIPATION? You should understand there is an uncommon, but foreseeable risk to you if you agree to participate in this study. A small percentage of participants who drive the 268
simulator may experience feelings of nausea or actual nausea. You should inform the experimenter at any point you feel any discomfort. I can stop the simulation immediately, which should quickly reduce the discomfort. Moreover, if you already experience any motion sickness while in a real car, you should not participate in the experiment. 8. VEHICLE?
WHO WILL SEE THE RESULTS OF MY PERFORMANCE IN THE
The results of the research may be published but that your name will not be revealed. To maintain confidentiality of your records, the researchers will use subject codes, rather than names, to identify any data collected during your simulation drive or any verbal response that you give. The subject codes will be secured in the Human Performance Laboratory and will be accessible only by Dr. Donald Fisher and Jeffrey Muttart. It is possible that your research record, including sensitive information and/or identifying information, may be inspected and/or copied by the study sponsor (and/or its agent), or federal or state government agencies, in the course of carrying out their duties. If your record is inspected by the study sponsor (and/or its agents), or by any of these agencies, your confidentiality will be maintained to the extent permissible by law. 9. STUDY?
WILL I RECEIVE ANY PAYMENT FOR TAKING PART IN THE
You will be compensated in the amount of $40.00 for your participation in the experiment. This payment will be made regardless of whether you finish the experiment. Again, I want to emphasize that you should inform the experimenter at any time you are feeling any discomfort. The experiment will be stopped at that point and you will be fully compensated. 10.
WHAT IF I HAVE A QUESTION?
Should you have any questions about the experiment, your treatment or any other matter relative to your participation in this project, you may call: Jeffrey Muttart at 860912-2280. If you experience a research related injury at any time during this study, you may contact: Donald L. Fisher at 413-549-1734. If, during the study or later, you wish to discuss your participation or concerns regarding it with a person not directly involved in the research, you can talk with the Human Subjects Administrator at
[email protected]; (413) 545-3428. A copy of this consent form will be given to you, for your records.
269
APPENDIX B HUMAN PERFORMANCE LABORATORY POST-STUDY QUESTIONNAIRE
270
271
APPENDIX C
INSTRUCTIONS TO PARTICIPANTS: SIMULATOR DRIVES Welcome to the Human Performance Laboratory. You have agreed to participate in a research on driving behavior in a simulated virtual world. This document discusses the overall procedures as well as specific instructions on what you are required to do in each part of the experiment. Part 1: Pretest Questionnaire First, you will be asked to complete a pretest questionnaire, which consists of questions about your background and ability to participate in this research. Also, you will be asked to read and sign an informed consent form. Page one of the questionnaire will be completed before participation. The remaining portion of the questionnaire will be completed after your driving. Part 2: Driving Simulator Calibrating the eye tracker During the first part of this experiment you will be fitted with a head mounted optical device and complete a calibration process. During the calibration process you should keep your head still while looking at specific points displayed in the screen in front of you (you will be told when and where to look). When I ask you to look at the “A”, please look toward the top right corner of the small box under the letter “A”. Do the same for each of the other letters I ask you to look at. Getting used to the Vehicle and Eye Tracker Calibration: You have been fitted with safety glasses that record your eye movements and your voice. We have calibrated the eye tracking system to your eye movements. If the glasses should slip from the current position, or if you touch them or move them, please tell me and we can recalibrate. It is important that you tell me if the eye tracking glasses move in any way. Also, even though I asked you to keep your head still during the calibration process, please feel free to move your head normally when driving. Practice Drive The first drive is designed to make you comfortable with the simulated driving environment, the vehicle itself, and the experimental task. You will practice driving in a simple driving environment so you can learn to appreciate the small, but noticeable, differences between simulated and actual driving. Please make yourself comfortable and adjust your seat and steering wheel position as necessary. The practice drive lasts approximately three to four minutes. You will notice that traffic moves very much as expected in the real world. During this drive we ask that you pay particular interest to the steering and braking of this vehicle. Many of those who drive in this simulated environment report that the steering and braking is different than the car they normally drive. To account for this we ask that you drive the practice drive 272
as many times as is necessary to become accustomed to the steering and braking of this vehicle. If you tell me that you are ready to begin the experiment, I will assume that you have a mastery of this vehicle and that any good or bad performances on your behalf are a direct result of your driving abilities. Experimental Drives The Driving Simulator is an actual car in front of which the virtual world is projected. You will operate the controls of the Driving Simulator just as you would the controls of a normal vehicle, causing you to slow, turn or accelerate in the virtual world in a manner consistent with your control action. Each drive is designed to replicate real life driving. You should not experience anything during your drives here today that is something that would be completely unexpected if driving on real roads. Additional users will share the road with you. You will hear directions when turning is necessary. Otherwise travel straight. We would like you to imagine that you are running late for work. You must arrive to work on time. If you travel a speed near the posted 40 mph speed limit, you will arrive on time. However, all the dangers that you might experience in real life will also be present here, so we ask that you drive expeditiously but safely. You can adjust the seat, move your head and use the equipment on the vehicle as you would in real life. The next nine drives are similar to the practice drive. Each drive takes 2 ½ to 4 minutes depending upon your speed. At the end of each drive I will ask you to stop. When I do so, please stop and place the car into park. I will then load a new virtual world onto the computer. When a new world is loaded, you will see the road in front of you. When you see the road, you may put the car into drive and drive away. You do not need my permission to start once the road is in front of you. You will be allowed to take a short break before each of the nine drives. If I do not hear otherwise, I will assume that you would like to continue. Risks and Discomforts: A small number of participants who drive the simulator may experience discomfort related to the perceived movement when none actually exists. Only drivers who successfully complete the practice drive, free of simulator sickness, should continue with the experiment. Drivers who experience frequent motion sickness while in a car should discontinue their participation in the experiment, regardless of how you feel in the simulator. Those who withdraw or are uncomfortable with this experiment for any reason may withdraw and will be compensated fully. It can be helpful to push yourself lightly against the back of the seat when braking. Some participants have reported that this reduces discomfort. If you have any questions, please feel free to ask the researchers at this point. I will discuss the instructions one more time just before the first and second experimental drives. For control purposes no questions will be answered during the experimental sessions. Thanks for your participation 273
APPENDIX D
EYE GLANCE SCORING TRAINING
274
275
276
277
278
279
280
APPENDIX E
ACT TRAINING
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
REFERENCES AASHTO, (2001). A Policy on Geometric Design of Highways and Streets, Washington, DC. Abdel-Aty, M. A., Chen, C. L., Radwan, A. E. (1999). Using conditional probability to find driver age effect in crashes, Journal of Transportation Engineering, 125, 502-
509 Agran, P. F., Winn, D. G., & Anderson, C. L. (1994). Differences in child pedestrian injury events by location. Pediatrics, 93, 284-288. Agran, P. F., Winn, D. G., Anderson, C. L., Tran, C., & Del Valle, C. P., (1996). The role of the physical and traffic environment in child pedestrian injuries. Pediatrics, 98, 1096-1103. Anderson, I. B., Krammes, R. A. (2000). Speed reduction as a surrogate for accident experience at horizontal curves on rural two-lane roads, Transportation Research Record 1701, Washington, DC: Transportation Research Board. Bisantz, A. M., et al. Knowledge versus execution in dynamic judgment tasks. Ed. Alex Kirlik. Adaptive Perspectives on Human-Technology Interactions: Methods and Models for Cognitive Engineering and Human-Computer Interaction. New York: Oxford University Press, 2006. Blythe, W. Day, T., Grimes, W. (1998), 3-Dimensional simulation of vehicle response to tire blow outs. (Technical paper no. 980221). Warrendale, PSA: Society of Automotive Engineers.
308
Bonneson, J. A., Pratt, M. P. (2009). A model for predicting speed along horizontal curves on two-lane highways, Paper No. 09-1419, Washington, DC: Transportation Research Board. Boyle, L. N., Tippin, J., Paul, A., and Rizzo, M. (2010). Driver Performance in the Moments Surrounding a Microsleep, Transportation Research Part F: Traffic Psychology and Behavior. 11(2): 126–136. Braitman, K. A., Kirley, B. B., McCartt, A. T., and Chaudhary, N. K. (2008). Crashes of Novice Teenage Drivers: Characteristics and Contributing Factors. Journal of Safety Research, 39 (1), 47-54. California Department of Motor Vehicles (2008). Fatality Analysis Reporting. http://www.dmv.ca.gov/teenweb/more_btn6/traffic/traffic.htm# Accessed August 15, 2009. Chen, H. Y., Ivers, R. Q., Martiniuk, A. L. C., Boufous, S., Senserrick, T., Woodward, M., Stevenson, M., Williamson, A., Norton, R.
(2009). Risk and type of
crash among young drivers by rurality of residence: Findings from the DRIVE Study, Accident Analysis and Prevention 41, 676–682. Choueiri, E. M. and Lamm, R. (1987). Rural Roads Speed Inconsistencies Design Methods: Part I, Operating Speeds and Accident Rates on Two-Lane Rural Highway Curved Sections. Department of Civil and Environmental Engineering, Clarkson University, NY.
309
Chovan, J.D., Tijerina, L., Pierowicz, J.A., and Hendricks, D.L., (1994). Examination of Unsignalized Intersection Straight Crossing Path Crashes and Potential IVHS Countermeasures, Report DOT HS 808 152, DOT-VNTSCNHTSA- 94-2. Washington, D.C.: Volpe National Transportation Systems Center. Cullinane, B., Green, P. A. (May 2006). Visual Demand of Curves and FogLimited Sight Distance and Its Relationship to Brake Response Time, Safety Vehicles using adaptive Interface Technology, UMTRI 102436, Ann Arbor, MI: University of Michigan Transportation Research Institute Cynecki, M. J., Sparks, J. W., and Grote, J. L. (1993). Rumble strips and pedestrian safety, ITE Journal, 63(8), 18-24. Daily, J., Shigemura, N., Daily, J. (2006). Fundamentals of Traffic Crash Reconstruction (2nd Ed.). Jacksonville, FL: Institute of Police Technology and Management, University of North Florida. Emmerson, J. (1969). Speeds of cars on sharp horizontal curves, Traffic Engineering & Control, 135-137. Fatal Accident Reporting System [FARS]. (2012). (National Highway Traffic Safety Administration) Retrieved November 12, 2012, from http://www.nhtsa.gov/people/ncsa/fars.html Fisher, D. L. (2008). Evaluation of PC-Based Novice Driver Risk Awareness DOT HS 810 926. Washington, DC: National Highway Traffic Safety Administration.
310
Fisher, D. L., Collura, J., Romoser, M. R. E., Knodler, M., Gomez, R., Samuel, S., Roman, L., Abdul, M. (2011). Evaluating the Effect of Advance Yield Markings and Symbolic Signs on Vehicle-Pedestrian Conflicts at Marked Midblock Crosswalks across Multilane Roads. ISA # 58738. Massachusetts Department of Transportation Fisher, D. L., Laurie, N. E., Glaser, R., Connerney, K., Pollatsek, A., Duffy, S. A. and Brock, J. (2002). The use of an advanced driving simulator to evaluate the effects of training and experience on drivers’ behavior in risky traffic scenarios. Human Factors, 44, pp. 287-302. Fisher, D. L., Pollatsek, A. and Pradhan, A. (2006). Can novice drivers be trained to scan for information that will reduce their likelihood of a crash? Injury Prevention, Volume 12, Supplement 1, i25-i29. Fisher, D. L., Pradhan, A. K., Pollatsek, A. and Knodler, M. A. Jr. (January 2007). Empirical evaluation of hazard anticipation behaviors in the field and on a driving simulator using an eye tracker. Proceedings of the 86th Transportation Research Board Annual Meeting CD-ROM, TRB, National Research Council, Washington, D.C. Fisher, D. L., Thomas, F. D. III, Pradhan, A. K., Pollatsek, A., Blomberg, R. D., Reagan, I. (2010). Development and evaluation of a PC-based attention maintenance training program, (Technical paper no. DOT HS 811 252). Washington, DC: National Highway Traffic safety Administration. Fitzpatrick, K., Fambro, D. B., and Stoddard, A. M. (2000). Safety Effects of Limited Stopping Sight Distance on Crest Vertical Curves. Transportation Research Record (1701), 17-24. 311
Fitzpatrick, K., Krammes, R. A., and Fambro, D. B. (1997). Design speed, operating speed and posted speed relationship, ITE Journal, 67(12), 52-59. Garay-Vega, L., Fisher, D. L. (2005). Can novice drivers recognize foreshadowing risks as easily as experienced drivers?, Proceedings of the Third International Driving Symposium on Human Factors in Driver Assessment, Training and Vehicle Design, 471-477. Glennon, J. C., Neuman, T. R., Leisch, J. T. (1986). Safety and operational considerations for design of rural highway curves, (DOT – FH – 11 – 9575). McLean, VA: Federal Highway Administration. Hallmark, S. L., Veneziano, D., McDonald, T., Graham, J., Bauer, K. M., Patel, R., Council, F. M. (2006). Safety Impacts of Pavement Edge Drop-offs, Washington, DC: AAA Foundation for Traffic safety. Happer, AJ, Peck, MD, Hughes, MC, (2009). Analysis of Left-Turning Vehicles at a 4-way Medium-Sized Signalized Intersection, Warrendale, PA: Society of Automotive Engineers. Harrell, W.A. (1994). Effects of Pedestrians' Visibility and Signs on Motorists' Yielding, Perceptual and Motor Skills, 78, 355-362. Highway Capacity Manual (2010). Washington, DC: Transportation Research Board.
312
Horrey, W.J. and Wickens, C.D. (2004). Focal and ambient visual contributions and driver visual scanning in lane keeping and hazard detection. Proceedings of the Human Factors and Ergonomics Society 48th Annual Meeting (pp. 2325-2329). Santa Monica, CA: Human Factors and Ergonomics Society. Horrey, W.J., Wickens, C.D., and Consalus, K.P. (2006). Modeling drivers’ visual attention allocation while interacting with in-vehicle technologies. Journal of Experimental Psychology: Applied, 12(2), 67-78. Hough, D. L. (2008). Proficient Motorcycling: The Ultimate Guide to Riding Well, 2nd Ed. Irvine, CA: BowTie Press. Hulse, M.C., Dingus, T. A., Fisher, T., and Wierwille, W.W. (1989). The Influences of Roadway Parameters on Driver Perception of Attentional Demand, Advances in Industrial Ergonomics and Safety I, 451-456. Integrated Driver Response Research, (2012). New London, CT: Crash Safety Solutions. Ivancic, K., and Hesketh, B. (2000). Learning from errors in a driving simulation: Effects on driving skill and self-confidence. Ergonomics, 43, 1966-1984. Klein, G., C. Zsambok, An Overview of Naturalistic Decision making Applications. Naturalistic Decision Making. Mahwah, NJ: Lawrence Erlbaum Associates Publishers, 1997. Kohlberg, L. (1973). The Claim to Moral Adequacy of a Highest Stage of Moral Judgment. Journal of Philosophy, 70 (18), 630–646.
313
Lamm, R. and Choueiri, E. M. (1987). Recommendations for evaluating horizontal design consistency based on investigations in the State of New York. Geometric Design and Operational Effects -Transportation Research Record, 1122, 6878" Land, M. F., & Lee, D. N. (1994). Where we look when we steer, Nature, 369(6483), 742-744. Liebowitz, H. W. & Owens, D. A. (1991). Can normal outdoor activities be carried out in civil twilight? Applied Optics, 30, 3501-3503. Liebowitz, H. W., Owens, D. A., and Tyrrell, R. A. (1998). The Assured Clear Distance Ahead Rule: Implications for traffic safety and the law. Accident Analysis and Prevention, 30, 93-99. Lisle, F. N., and Hargroves, B. T. (1980). Evaluation of the performance of portable precast concrete traffic barriers. Transportation Research Record 769. Washington, DC: Transportation Research Board. Maeda, T., Irie, N., Hidaka, K., Nishimura, H. (1977). Performance of drivervehicle system in emergency avoidance, (Technical paper 770130). Warrendale, PA; Society of Automotive Engineers. Masten, S. V. and Foss, R. D. (2010). Long-term effect of the North Carolina graduated driver licensing system on licensed driver crash incidence: A 5-year survival analysis. Accident Analysis and Prevention, 42, 2010.
314
Mayhew, D. R.; Simpson, H. M. and Pak, A. (2003). Changes in collision rates among novice drivers during the first months of driving. Accident Analysis and Prevention, 3, 683-91. McCartt, A. T., Shabanova, V. I. and Leaf, W. A. (2003). Driving experience, crashes and traffic citations of teenage beginning drivers. Accident Analysis and Prevention, 35, 311-320. McLaughlin, S. B., Hankey, J. M., Klauer, S. G., Dingus, T. A. (2009). Contributing factors to run-off-the-road crashes and near crashes, DOT HS 811 079 Washington, DC; National Highway Traffic Safety Administration. McLean, J. (1981). Driver speed behaviour and rural road alignment design. Traffic Engineering & Control, 22, 208-211 Mikolajetz, A., Henning, M. J., Tenzer, A., Zobel, R., Krems, J. F., Petzoldt, T. (2009). Curve negotiation: identifying driver behavior around curves with the driver performance database, Proceedings of the fifth International Driving Symposium on Human Factors in Driver Assessment, Training and Vehicle Design, 391 - 397. Mourant, R. R., and Rockwell, T. H. (1972). Strategies of visual search by novice and experienced drivers. Human Factors, 14, 325–335. Muttart, J. W., Messerschmidt, W., & Gillen, L. (2005). Relationship between Relative Velocity Detection and Driver Response Times in Vehicle Following Situations (Technical paper No. 2005-01-0427). Warrendale, PA: Society of Automotive Engineers.
315
OECD Publications. (October 2006). Young Drivers the road to safety, Policy Brief. http://www.internationaltransportforum.org/jtrc/safety/YDpolicyBrief.pdf Accessed July 2009. O'Leary, P., Bettinardi, C., Daniel, T., Nielsen, E., & Curry, D. G. (2006). Driver Response to Roadside Obstacles. Proceedings of the Human Factors and Ergonomics Society 50th Annual Meeting (pp. 2013-2017). Human Factors and Ergonomics Society. Pennell, J. The Driving Hazard Book. [Online] 2010. http://www.shellstations.com/answerbook/driving/driving_6.htm. Accessed July 22, 2010 Pollatsek, A., Fisher, D. L. and Pradhan, A. K. (2006). Identifying and Remediating Failures of Selective Attention in Younger Drivers. Current Directions in Psychological Science, 15, 255-259. Polus, A., Fitzpatrick, K., and Fambro, D. B. (1999). Predicting operating speeds on straight sections of two-lane rural highways. Transportation Research Record, 1737, 50-57. Pradhan, A. K., Fisher, D. L. and Pollatsek, A. (2006). Risk Perception Training for Novice Drivers: Evaluating Duration of Effects on a Driving Simulator. Transportation Research Record, 1969, 58-64. Pradhan, A. K., Fisher, D. L., Pollatsek, A., Knodler, M., and Langone, M. (October, 2006). Field Evaluation of a Risk Awareness and Perception Training Program for Younger Drivers. Proceedings of the Human Factors and Ergonomics Society 50th Annual Meeting, San Francisco, 2388-2391.
316
Pradhan, A. K., Hammel, K. R., DeRamus, R., Pollatsek, A., Noyce, D. A. and Fisher, D. L (2005). The Use of Eye Movements to Evaluate the Effects of Driver Age on Risk Perception in an Advanced Driving Simulator. Human Factors, 47, 840-852. Pritchett, A. R., Bisantz, A. M. Assessing the fit between human judgment and automated alerts: A study of collision detections. Ed. Alex Kirlik. Adaptive Perspectives on Human-Technology Interactions: Methods and Models for Cognitive Engineering and Human-Computer Interaction. New York: Oxford University Press, 2006. Read, J. H. (1969). Traffic accidents involving children pedestrians: a program for their prevention. Pediatrics, 44 Suppl. 838-847. Resnick, R. A. (2004). Visual sensing without seeing, Psychological Science, 15, 27-32. Robinson, G. H., Erickson, D. J., Thurston, G. L., & Clark, R. L. (1972). Visual Search by Automobile Drivers, Human Factors, 14(4), 315-323. Romoser, M. R. E. & Fisher, D. L. (2009). The effect of active versus passive training strategies on improving older drivers’ scanning for hazards while negotiating intersections. Human Factors, 51(5), 652-668. Shinar, D., McDowell, E., and Rockwell, T. H. (1977) Eye movements in curve negotiation. HumanFactors, 19(1), 63-71. Simons, D. J. (1999). Gorillas in our midst; sustained inattentional blindness for dynamic events. Perception, 28, 1059-1074.
317
Smith, K., Thome, A., Blaberg, C., Bargman, J. (2009). An invariant may drive the decision to encroach at unsignalized intersections, Proceedings of the Fifth International Driving Symposium on Human Factors in Driver Assessment, Training and Vehicle Design. Snowden, R. J., Stimpson, N., Ruddle, R. A. (1998). Speed perception fogs as visibility drops, Nature, 392, 450. Summala, H., Lamble, D., and Laakso, M., (1998). Driving Experience and perception of the lead car's braking when looking at in-car targets. Accident Analysis and Prevention, 30(4), 401-407. Summala, H., Nieminen, T., Punto, M. (1996). Maintaining lane position with peripheral vision during in-vehicle tasks, Human factors, 38, 442-451. Tate, F., Turner, S. (2007). Road geometry and drivers’ speed choice, IPENZ Transportation Conference, Tauranga, NZ. Taylor, T., Masserang, K., Divekar, G., Samuel, S., Muttart, J., Pollatsek, A., Fisher, D. L., (2011). Long-Term Effects of Hazard Anticipation Training on Novice Drivers Measured on the Open Road, 6th International Driving Symposium on Human Factors in Driver Assessment, Training, and Vehicle Design, Olympic Valley, CA. Thompson, S. J., Fraser, E. J., & Howarth, C. I. (1985). Driver behavior in the presence of child and adult pedestrians, Ergonomics, 28, 1469-1474.
318
Trick, L. M., Lochner, M., Toxopeus, R., Wilson, D. (2009). Manipulating drive characteristics to study the effects of mental load on older and younger drivers, Proceedings of the Fifth International Driving Symposium on Human Factors in Driver Assessment, Training and Vehicle Design, 363-369. Triggs, T. J. (1980). The influence of oncoming vehicles on automobile lateral position, Human Factors, 22, 111-114 Van Winsum, W. (1996). Speed Choice and Steering Behavior in Curve Driving, Human Factors, 38(3), 434-441 Varhelyi, A. (1998). Drivers' Speed Behaviour at a Zebra Crossing: A Case Study. Accident Analysis and Prevention, 30 (6), 731-743. Vlakveld, W. P. (2005). Young, novice motorists, their crash rates, and measures to reduce them; a literature study (Technical Report R-2005-3). Leidschendam: Stichting Wetenschappelijk Onderzoek Verkeersveiligheid (SWOV). Vlakveld, W., Romoser, M. R. E., Mehranian, H. Diete, F., & Fisher, D. L. (2011). Does the experience of crashes and near crashes in a simulator-based training program enhance novice driver’s visual search for latent hazards?, Transportation Research Record No. 2265, Transportation Research Board of the National Academies, Washington, D.C., 153-160. Walsh, W. B., Betz, N. E. (2000). Tests and Assessments, 4th Ed., Upper Saddle River, NJ: Prentice Hall.
319
Winston, F. (2011). 677 teens who were involved in serious crashes between 2005-2006: Query from the National Motor Vehicle Crash Causation Survey (NMVVCS), 2008. DOT HS 811 051, Washington, DC: National Highway Traffic Safety Administration (email correspondence to D. L. Fisher, June 10, 2011). Wright, P.H. and Robertson, L.S. (1979). Amelioration of Roadside Obstacle Crashes, Journal of Transportation Engineering, 105, 609-622. Zegeer, C. V., Stewart, R., Huang, H. H., & Lagerwey, P. A. (2002). Safety Effects of Marked vs. Unmarked Crosswalks at Uncontrolled Locations: Executive Summary and Recommended Guidelines. US Department of Transportation, Federal Highway Association. Zegeer, C., Stewart, R., Reinfurt, D., Council, F.M., Neuman, T., Hamilton, E., Miller, T., and Hunter, W. (1990). Cost Effective Geometric Improvements for Safety Upgrading of Horizontal Curves. Publication No. FHWA-RD-90-021, University of North Carolina, Highway Safety Research Center, Chapel Hill, NC.
320
