Performance Criteria for Retroreflective Pavement Markers (2022)

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

55   Treatment Recognition, Visibility, and Driver Behavior Evaluations To assess the effectiveness of RPMs, the research team developed a three-pronged approach: a closed-course visibility and driver behavior study, an operational assessment of SHRP 2 data rela- tive to RPM use, and a safety analysis. This chapter is dedicated to the first approach and covers a closed-course recognition and visibility examination of pavement markings and RPMs and a closed-course evaluation of driver behavior through curves with different RPM treatments. The Texas A&M University System RELLIS Campus, henceforth referred to simply as the RELLIS Campus, is located at the site of the former Bryan Air Force Base. The series of runways and taxiways provided the test location on which to conduct the closed-course studies. For this study, data collection efforts for three analyses were simultaneously undertaken, focusing on the ability of study participants to recognize various delineation treatments, factors affecting the visibility of treatments, and driver behavior through curves. In order for on-road guidance treatments to effectively relay information to road users, the treat- ments must be visible from a reasonable distance. The treatment recognition study was designed to identify the distance at which study participants could recognize RPMs or pavement marking tape treatments that diverged to either the left or right from the centerline travel path straddled by the study vehicle. The participants identified the treatment when they could determine which direction the treatment deviated toward and relay that information to a Texas A&M Transporta- tion Institute (TTI) researcher operating the vehicle. Retroreflectivity is the most common metric used to quantify the visibility of TCDs, including pavement markings and RPMs. However, retroreflectivity is a physical property of a material and does not account for ambient lighting of the roadway, the geometric relationship between the device and the vehicle, the headlight output, or the driver’s visual capability. Therefore, it is necessary to quantify visibility level (VL) for assessing the visibility of RPMs, based on drivers’ visual demands, to determine their visual performances in different settings and by drivers with various visual capabilities. In wet road conditions, the visibility of pavement markings can be substantially reduced, especially when the markings and RPMs have degraded due to age and wear. This issue is a common complaint of motorists and is a particular issue at locations of alignment changes (i.e., curves). To assess the effectiveness of the RPMs at conveying alignment information to the study participants, several performance metrics on which to gauge driver performance were identi- fied: the number of drivers that encroached the centerline in the treatment area, the maximum deceleration that was observed on the test segment, the distance to the point of tangency of the treatment curve at which the maximum deceleration was observed, the speed loss in the curve, and the lane position variation in the curve. Driver performance in terms of speed and position can then be used to gauge how well the study participants were able to see each treatment and use that information to navigate the curve. C H A P T E R 4

56 Performance Criteria for Retroreflective Pavement Markers To simulate testing in wet conditions, the research team installed custom pavement markings in the treatment areas with the intent to simulate the low visibility of pavement markings in wet conditions. Using a thin application of waterborne paint with a low bead drop rate, the research team striped the treatment areas and then abraded the markings with a steel-reinforced sweeper brush. The results were very faded markings with retroreflectivity levels less than 25 mcd/m2/lux. The resulting treatment areas can be seen in the images included in Appendix A. 4.1 Data Collection Procedure TTI conducted the data collection for the closed-course studies at the RELLIS Campus in December 2017. Study participants came to the RELLIS Campus, arriving at a gated entrance to the property in pairs, with four people typically participating each night of the study. The first study pair for a given night would arrive shortly after dusk, with another pair arriving approxi- mately two hours later. Upon arriving at the gate, the study participants were met by a member of the research team and guided to a check-in building near the runway system. At the check-in building, demographic information for each of the study participants was collected. Following the collection of the demographic information, a vision test was administered to identify the visual acuity of each study participant as well as their ability to distinguish contrast. Next, each participant was briefed on the general procedure of the data collection process for both the recognition and curve studies. Once the participants had a cursory understanding of the procedure, one participant was taken to a Toyota Sienna to complete the first and second of four laps of the curve study, while the other participant was taken to a Ford Explorer to com- plete the first of two laps for the recognition study. The nuances of the data collection efforts are discussed in more detail in the following paragraphs. The locations of the treatments for each of the data collection efforts are indicated in Figure 6. The RPMs examined in this research were obtained new from Ennis-Flint, with an initial retro- reflectivity of 800–1,000 mcd/lux. Throughout, the recognition, visibility, and curve behavior study, RPMs of five retroreflectivity levels were used as treatments. RPMs were considered to have high retroreflectivity when they were examined in the new-from-the-manufacturer condi- tion. The medium-, low-, and heavy-wear RPMs were created from the new RPMs by degrad- ing the retroreflective surface using sandpaper. Medium-retroreflectivity RPMs were degraded to a level of approximately 167 mcd/lux, which is the ASTM standard for new RPMs (ASTM International 2015). Low-retroreflectivity RPMs were degraded to 65 mcd/lux, representing the Texas DOT (TxDOT) 12-month field test material specification for RPMs (TxDOT 2018). The heavy-wear RPMs were degraded to slightly less than one-half of the TxDOT standard to simulate RPMs nearing the end of their service life. Finally, obscured RPMs were created by applying masking tape to the retroreflective surface. These RPMs represented nonreflective RPMs. During the recognition and visibility studies, the medium-, low-, and heavy-wear RPMs were used as well as two types of tape markings. The tape markings consisted of a new-from-the- manufacturer material with a retroreflectivity of 500 mcd/m2/lux and a simulated-wear marking (made from the same material but degraded) that had a retroreflectivity of 100 mcd/m2/lux. In the curve study, high-, medium-, low-, and obscured-retroreflectivity RPMs were used. The treatments are summarized in Table 39. 4.1.1 Recognition Study The recognition and visibility studies included 37 participants. The demographic and visual data for these participants are presented in Table 40. Throughout the recognition study data collection process, a researcher drove the test vehicle, the 2016 Ford Explorer SUV shown in Figure 7, and subjects observed the road ahead from the vehicle’s passenger seat. The vehicle was

Treatment Recognition, Visibility, and Driver Behavior Evaluations 57   (a) (b) Figure 6. Locations of recognition (a) and curve (b) study treatments. Delineator Type Treatment Label Retroreflectivity RPM High-retro 800–1,000 mcd/lux RPM Medium-retro 167 mcd/lux RPM Low-retro 65 mcd/lux RPM Heavy-wear 30 mcd/lux RPM Obscured <10 mcd/lux Tape Marking New 500 mcd/m2/lux Tape Marking Simulated-wear 100 mcd/m2/lux Table 39. Treatment summary. equipped with a mobile GPS unit operated through a laptop and keypad allowing the researcher to mark GPS locations as the vehicle moved along the course. Along the RPM recognition course, the researcher drove the vehicle at approximately 15 mph while straddling the centerline of the course. The light-emitting diode (LED) low-beam headlights were aligned before the study and used at all times. All treatments were located on straight sections of the course, as previously indicated in Figure 6a. The RPM recognition study began with the researcher driving the participant to a test section of the course to practice the procedure of identifying deviations of RPMs or pavement markings away from the centerline of the course. This test section served to familiarize the participants

58 Performance Criteria for Retroreflective Pavement Markers with the procedure so that hesitation due to unfamiliarity would be less likely to occur during the data collection. Throughout the actual test course, study participants encountered RPMs of three different retroreflectivity levels, and tape markings of two different retroreflectivity levels were used as treatments. These treatments, all of which were yellow, are described in Table 41. The alphanumeric treatment designation refers to a specific part of the runway, as identified in Figure 6a. On part of the course along the north end of the treatment area on runway 35C, four overhead luminaires at 160-ft spacing with high-pressure sodium lamps provided lighting to simulate RPMs and pavement markings under a roadway lighting condition. Between the two halves of the study, treatments along 35C (denoted as A1–A4 and B1–B4) would be swapped. Whether participants viewed the A or B treatments first alternated between each group of participants in the study. During the data collection for the recognition study, the Ford Explorer was driven along the centerline, which consisted of low-retro markings/markers spaced at 80-ft intervals. The center- line markings/markers were not the treatment and only provided guidance regarding the course layout. The markings (broken line) or RPMs of each treatment started on the centerline path and deviated to the right or left in a straight line at a slight angle, as shown in Figure 8 (only three treatments are shown for demonstration). All deviations were of the same angle. The RPMs of Characteristic Number of Participants Characteristic Number of Participants Gender Age Female 18 28 or younger 7 Male 19 29–64 9 Miles driven per year 65 or older 21 <10,000 7 Visual acuity 10,000–15,000 9 20/13 5 >15,000 21 20/15 4 Typical driving location 20/20 12 City street 28 20/25 5 Rural roads 3 20/30 4 Freeways 6 20/40 7 Table 40. Recognition and visibility study participant details. Figure 7. 2016 Ford Explorer test vehicle.

Treatment Recognition, Visibility, and Driver Behavior Evaluations 59   each treatment deviating from the path were spaced at 40 ft longitudinally, with the first one on the centerline and the next four deviating to the right or left of the centerline path. The lateral offset of the RPMs increased by 2 ft as each subsequent marker moved away from the centerline. Markings were 10 ft long for all four segments and had a 30-ft gap longitudinally, and lateral offset increased by 2 ft for each subsequent marking. The deviation direction of each treatment is shown in Table 41. When a participant verbally declared a deviation, the location of the declara- tion would be marked using GPS, allowing an observation distance to be calculated. 4.1.2 Driver Behavior Through Curves For this study, 37 participants took four laps each through a test course consisting of six unique curves; however, only 33 participants were used in the analysis because four drove a different Treatment Treatment Type Deviation Direction Overhead Lights Runway A1 New marking tape Left Yes 35C A2 Heavy-wear RPM Left Yes 35C A3 Low-retro RPM Right No 35C A4 New marking tape Right No 35C B1 Low-retro RPM Right Yes 35C B2 Medium-retro RPM Right Yes 35C B3 Medium-retro RPM Left No 35C B4 Heavy-wear RPM Left No 35C C1 New marking tape Left No 35L C2 Medium-retro RPM Right No 35L C3 Heavy-wear RPM Right No 35L C4 Low-retro RPM Left No 35L C5 Simulated-wear marking tape Right No 35L Table 41. Recognition treatment details. Figure 8. Treatment settings in the recognition study.

60 Performance Criteria for Retroreflective Pavement Markers vehicle through the course due to mechanical issues and were subsequently omitted. Upon arrival at the RELLIS Campus, each study participant’s demographic information was veri- fied, and participants were given a visual performance test for visual acuity and contrast sensitivity. Details for the participants specifically examined in the curve study can be found in Table 42. After completion of the visual performance tests, the study participants would traverse the course in an instrumented Toyota Sienna, shown in Figure 9. The vehicle is a 2015 model with halogen projector low-beam lights. The equipment installed in the Sienna included the following: • Dewetron computer system connected to vehicle controller area network bus • GenSys Model ADMA-G-ECO automotive dynamic motion analyzer • NovAtel GPS-702-GG dual-frequency GPS and GLONASS antenna The Dewetron computer system recorded vehicle operational characteristics and positioning data at a frequency of 10 Hz as study participants drove the vehicle around the course. Fields of potential interest were as follows: • Percent accelerator depression • Percent brake depression • Steering wheel angle (degrees/percent) Characteristic Count Characteristic Count Gender Age Female 16 28 or younger 6 Male 17 29–64 8 Miles driven per year 65 or older 19 <10,000 11 Visual acuity 10,000–15,000 11 20/13 3 >15,000 11 20/15 4 Typical driving location 20/20 10 City street 25 20/25 5 Rural roads 3 20/30 4 Freeways 5 20/40 7 Table 42. Participant information (n = 33). Figure 9. Side and front views of the instrumented Sienna.

Treatment Recognition, Visibility, and Driver Behavior Evaluations 61   • Engine revolutions per minute • Vehicle speed (speedometer/GPS) • GPS status • X, Y, and Z accelerometers • Latitude and longitude • Time • Roll • Pitch • Heading • Elevation In addition to logging the data, the Dewetron computer also displayed information in real time, allowing the research team to verify that data being collected were valid and to trouble- shoot, as necessary. Figure 10 shows the data that were available to the study team in real time. During the study, a TTI employee was seated in the passenger seat to provide high-level directions to help the participant navigate the course. Exclusive of the curves being studied, the route through the course was delineated using RPMs covered by a mesh tape to produce a consis- tent retroreflectivity level of about 175 mcd/lux. Highly worn painted pavement markings were also present along the route; however, in most cases, they were barely visible, with retroreflectivity levels typically below 25 mcd/m2/lux. Each study participant navigated the course four times. Each curve was delineated using RPMs of varying retroreflectivity levels. The treatment sec- tions started 200 ft before the point of curvature (PC) and ended 200 ft after the point of tangent (PT). The treatments were spaced evenly along the length of the curve based on the desired spacing for each particular curve. The treatments were broken down into multiple installation scenarios for odd and even days in the study. For example, Curve 1 had low-retroreflectivity RPMs during Lap 1 for the first participant on Day 1 of the study and medium-level RPMs for Lap 1 for the first participant on Day 2 of the study. Further details of the order in which partici- pants encountered the various RPM retroreflectivity levels and images of the treatments in the curves can be found in Appendix A. In addition to the RPMs, several of the curves were treated without RPMs or with RPMs whose retroreflective material had been obscured via masking tape. The layout of the course can be seen in the aerial image of the RELLIS Campus runways shown in Figure 6. Table 43 provides details of the geometric properties of the curves. During Laps 1 and 3, drivers entered the course at the north end of Curve 1, proceeding through Curves 1, 2, and 3 before using a connecting roadway to circle back to the north end of Curve 1, passing through the curve again and then proceeding through Curves 4, 5, and 6. Figure 10. Dewetron real-time data display.

62 Performance Criteria for Retroreflective Pavement Markers During Laps 2 and 4, drivers started north of Curve 6, proceeding through Curves 6, 7, and 8, entering Curve 1 from the southern end, proceeding north to the connecting roadway, and entering Curve 3 from the north end before proceeding through Curves 2 and 1. Trips when the driver passed through Curve 1 immediately before or after driving through Curve 4 were excluded from all analyses since drivers had to brake and turn toward the different areas of the course. Table 43 indicates that Curve 5 had glare present. This glare was provided by the halogen head- lights of a Ford Fusion located to simulate oncoming headlights. Two vehicles were used such that one vehicle provided glare during Laps 1 and 3 while the other provided glare during Laps 2 and 4. The vehicles were stationary during the study and were located such that they would not interfere with the study participants’ travel paths. The retroreflectivity levels assigned to the RPMs were the result of a deliberate attempt at simulating the wear that occurs over the service life of an RPM. Using a RoadVista 1200F field retroreflectometer, new RPMs were degraded using sandpaper until they fit in one of the desired ranges of retroreflectivity. The retroreflectivity levels are described in Table 44. Images of each of the installed treatments can be found in Appendix A. Each participant completed two laps, then participated in the marking recognition study that was occurring simultaneously on a different part of the test area (described previously in this chapter), and then returned to complete the final two laps. Between the pairs of laps, treatments in the curves were changed to present the drivers with different scenarios. Due to the layout of the course, Laps 1 and 3 consisted of only right-curved treatments, while Laps 2 and 4 consisted of only left-curved treatments. While driving through the course, drivers were instructed to operate at speeds of 40 mph and maintain a comfortable speed in the curves. No speed limit, curve warning, or curve advisory speed signs were present on the test course. The test course was laid out on a former runway; subsequently, there was no superelevation in any of the curves. The AASHTO Green Book Curve Treatment Length (ft) Arc Length (ft) Curve Radius (ft) Central Angle (degrees) RPM Retro Levels RPM Spacing (ft) Glare Curve 1 755 358 525 39 Low, Medium 40 No Curve 2 754 356 203 101 None, Low, Medium 40 No Curve 3 688 358 510 40 None, High 40 No Curve 4 695 297 301 57 Low, Medium 20 No Curve 5 632 239 308 44 Obscured, Medium 40 Yes Curve 6 832 435 283 88 Low, Medium 40 No Table 43. Curve properties. Retroreflectivity Level Details Target Retroreflectivity RoadVista 1200F Retroreflectivity Range (measured) High New RPM New 800–1,000 mcd/lux Medium ASTM D4280 new RPM minimum 167 mcd/lux 157–177 mcd/lux Low TxDOT DMS-4200 in- service RPM minimum 65 mcd/lux 57–73 mcd/lux Obscured Retroreflective face covered with masking tape 0 mcd/lux 0 mcd/lux Table 44. Curve study RPM retroreflectivity levels.

Treatment Recognition, Visibility, and Driver Behavior Evaluations 63   (AASHTO 2011) specifies the following relationship between speed, superelevation, side friction, and curve radius: R V e f15 0.01 [3]min 2 max max( ) = + where R = minimum curve radius, V = the speed in miles per hour, emax = the maximum allowable superelevation, and fmax = the maximum available side friction. Using 40 mph for a design speed, 0% superelevation due to the flat nature of the runway, and fmax of 0.16 (the value specified in the Green Book for 40 mph), the minimum curve radius is 667 ft, which is larger than any of the study curves. Conversely, the equation can be solved for V to identify an appropriate speed. These values are presented in Table 45. Despite the fact that the study curves are not designed to be navigated at 40 mph (the speed at which the participants were instructed to drive through the tangent portion of the course), no advisory speed signage, chevrons, or other navigational signs were present on the course. Subsequently, all vehicle operation changes that occurred in the study were directly attributable to the participant’s ability to detect the curve geometry based on the RPMs on the course and the heavily worn pavement markings. The worn markings, which were used in lieu of wet markings, were solid, double-yellow markers on the centerline in the curves and solid white on the edge line. In nontreatment portions of the course (i.e., tangent sections), the markings varied in type. Images of the curves illustrating the pavement markings present can be found in Appendix A. 4.2 Statistical Methodology While analyzing the data collected for the recognition and curve studies, the distribution of the various performance metrics used to quantify study participant performance was found to be right-skewed. Subsequently, preliminary statistical analyses using ordinary least squares (OLS) linear regression conducted on the data revealed that the resulting distribution of errors was also right-skewed, as opposed to normally distributed, violating one of the underlying assumptions of OLS. In order to provide reliable parameter estimates and thus make reliable inferences regarding the data, generalized linear models (GLMs) with a log-link function and gamma error distribution were estimated. In this framework, the probability of observing a particular value of yi given parameters αi and θi is given by the probability density function of the gamma distribution: f y y e yi i i i yi i i for , and 0 [4]1 i i i( ) ( ) = θ Γ α α θ >( ) α α − − θ where α and θ are shape (overdispersion) and scale parameters, respectively, and the distribution has mean αi/θi and variance αi/θi2. Curve Curve 1 Curve 2 Curve 3 Curve 4 Curve 5 Curve 6 Speed (mph) 36 22 35 27 27 26 Table 45. Design speeds for study curves.

64 Performance Criteria for Retroreflective Pavement Markers Using the log-link function, the parameterization of the model is given by the following equation: ( )= βEXP [5]Y Xi i where Yi = the response variable characterized by the gamma distribution, Xi = a vector of covariates associated with observation i, and β = a vector of estimable parameters. Unlike the simple linear models estimated with OLS, maximum likelihood methods are used to estimate GLMs. To assess encroachment probability, another GLM, the binary logistic regression (logit) model, was used. Binary logit models describe the log odds of an event happening, in this case, a centerline encroachment. The log odds are a transformation of the probability of an event happening, which is then related to explanatory factors using the following equation: ( ) ( )−       = βln 1 [6]P y P y Xi i i where P(yi) = the probability that an encroachment occurred on trip i through a curve, β represents a vector of estimable parameters, and Xi = a vector containing the characteristics of the ith trip through a curve. The logit-link function used to relate the dependent and independent variables in the equation yields coefficients for binary variables that can be interpreted as having an effect of (eβ−1) p 100% change in the log odds of an encroachment occurring. Coefficients for the continuous variables can be interpreted such that a 1-unit change in X results in a (eβ−1) p 100% change in the log odds of an encroachment occurring. 4.3 Driver Recognition of Treatments 4.3.1 Preliminary Data Exploration Based on the information collected about study participant characteristics and the ability to recognize the pavement marking and marker patterns, a data set was developed where each observation represented the distance at which one participant identified one treatment. The observed distance was then matched with the corresponding demographic and visual perfor- mance data for that participant as well as detailed information regarding the specific treatment observed. Visual performance was based on two tests: visual acuity and contrast detection. According to the American Optometric Association (2018), visual acuity is a ratio based on the following principles: 20/20 vision is a term used to express normal visual acuity measured at 20 ft. A person with 20/100 vision needs to stand 20 ft from an object to see with equal clarity what a person with normal (20/20) vision sees at 100 ft. For this analysis, the expression of visual acuity can simply be treated as a number (e.g., 20/20 vision equals 1.00, 20/100 vision = 0.2), where larger numbers indicate better acuity. The combination of participants and treatments observed in this study resulted in a data set consisting of 666 observations. Descriptive statistics for the data set are provided in Table 46. In order to explore the data and gain insight into the relationships between participants and treatment recognition, a series of plots were developed. Two participant characteristics

Treatment Recognition, Visibility, and Driver Behavior Evaluations 65   that tended to be of interest were age, particularly whether the participant was 65 years of age or older, and visual acuity. Figure 11 is a box-and-whisker plot of the A2 treatment (heavy- wear RPM under the roadway lighting) aggregated by age group and visual acuity. The red and green on the plots indicate the interquartile range. The point where the colors meet is the median value. The plot indicates two clear trends: older study participants needed to be closer to the treatment to see it, and as the visual acuity denominator increased (i.e., vision worsened), study partici- pants needed to be closer to the treatment to be able to see it. This trend was observable over nearly every treatment. Figures for the other treatments have been included in Appendix B. Building further on the information presented in Figure 11, Figure 12 presents the average dis- tance at recognition for each treatment versus visual acuity. Parameter Average Std. Deviation Minimum Maximum Female 0.49 0.50 0.00 1.00 Age 56.11 23.03 19.00 87.00 65 or older 0.57 0.50 0.00 1.00 Less than 10,000 0.30 0.46 0.00 1.00 10,000 to 15,000 0.35 0.48 0.00 1.00 Greater than 15,000 0.35 0.48 0.00 1.00 City streets 0.76 0.43 0.00 1.00 Rural roads 0.08 0.27 0.00 1.00 Freeways 0.16 0.37 0.00 1.00 Visual acuity ratio 0.95 0.34 0.50 1.54 Heavy-wear RPM 0.22 0.42 0.00 1.00 Low-retroreflectivity RPM 0.22 0.42 0.00 1.00 Medium-retroreflectivity RPM 0.22 0.42 0.00 1.00 Simulated-wear tape 0.11 0.31 0.00 1.00 New tape 0.22 0.42 0.00 1.00 Right side 0.44 0.50 0.00 1.00 Overhead lights 0.22 0.42 0.00 1.00 Observation distance 525.58 329.08 32.34 1859.73 Table 46. Recognition study descriptive statistics (37 participants, 666 observations). 0 100 200 300 400 500 600 700 65 or over Less than 65 Re co gn iti on D is ta nc e (ft ) Age A2 By Age Group 0 100 200 300 400 500 600 700 20/13 20/15 20/20 20/25 20/30 20/40 Re co gn iti on D is ta nc e (ft ) Visual Acuity A2 By Visual Acuity Figure 11. A2: heavy-wear RPM recognition results.

66 Performance Criteria for Retroreflective Pavement Markers Figure 12 simultaneously demonstrates that participants needed to be much closer to the treat- ments to recognize them as visual acuity worsened, as retroreflectivity decreased, and in the presence of overhead lighting. Collectively, these observations provide valuable insight to expound upon via statistical analysis. Another way that the information can be considered is in terms of preview time. Table 47 presents preview times that were calculated for 45 mph, 55 mph, and 65 mph based on the median recognition distance for each treatment retroreflectivity level and runway location. The table has been color coded such that red cells represent the preview time being less than 1.8 seconds, orange cells are less than 2.2 seconds, yellow cells are less than 3 seconds, green cells are less than 3.65 seconds, and blue cells are greater than 3.65 seconds. The first two values are based on the European Cooperation in the Field of Scientific and Technical Research 331 report (COST 1999), while the 3.65-second threshold was identified by Zwahlen and Schnell (1998). 0 200 400 600 800 1000 1200 1400 20/13 20/15 20/20 20/25 20/30 20/40 A ve ra ge D is ta nc e at R ec og ni tio n (ft ) Visual Acuity Worn Marking Tape New Marking Tape Heavy Wear RPM Low Retro RPM Medium Retro RPM New Marking Tape Under Lights Heavy Wear RPM Under Lights Low Retro RPM Under Lights Medium Retro RPM Under Lights Figure 12. Average recognition distance versus visual acuity. Treatment Medium- Retro RPM Low- Retro RPM Heavy- Wear RPM New Tape Simulated- Wear Tape With Overhead Lighting— A and B Locations Median Distance 560.95 393.74 386.05 221.51 45 mph Preview Time 8.50 5.97 5.85 3.36 55 mph Preview Time 6.95 4.88 4.79 2.75 65 mph Preview Time 5.88 4.13 4.05 2.32 No Overhead Lighting— A and B Locations Median Distance 741.55 619.59 528.21 289.08 45 mph Preview Time 11.24 9.39 8.00 4.38 55 mph Preview Time 9.19 7.68 6.55 3.58 65 mph Preview Time 7.78 6.50 5.54 3.03 No Overhead Lighting— C Locations Median Distance 1063.73 661.92 632.92 264.44 140.95 45 mph Preview Time 16.12 10.03 9.59 4.01 2.14 55 mph Preview Time 13.19 8.21 7.85 3.28 1.75 65 mph Preview Time 11.16 6.94 6.64 2.77 1.48 Table 47. Preview times (in seconds) for various speeds based on median recognition distances (in feet).

Treatment Recognition, Visibility, and Driver Behavior Evaluations 67   The table illustrates that based on the median recognition distances for all participants, most of the treatment scenarios had adequate preview time. However, the simulated-wear tape performed poorly in terms of preview time, with unsatisfactory results for speeds in excess of 45 mph and barely meeting the comfortable minimum criteria at 45 mph. Similar tables have been developed for other percentiles and considering only subsets of the participants. These tables can be found in Appendix B. 4.3.2 Statistical Analysis A statistical analysis was conducted on the data set to simultaneously examine all relevant characteristics of the drivers as well as the physical characteristics of the pavement markings and markers. This study used a generalized linear model with a gamma distribution estimated with the GLM package using the statistical software R. The results of that analysis are shown in Table 48. Separate models were estimated considering Lap 1 and Lap 2, as well as a joint model considering both laps; the latter is the focus of discussion for the remainder of the paper. The parameter estimates for the Lap 1 and Lap 2 models are shown to illustrate the similarity between the three models. The log link used to estimate the regression equation yields coefficients for binary variables that can be interpreted as having an effect of (eβ−1) p 100% on the dependent variable and coefficients for the continuous variables that can be interpreted such that for a 1-unit change in x, a (eβ−1) p 100% change in y occurs. The ability of study participants to recognize the treatments was found to decrease as vision worsened and the visual acuity parameter increased. This observation is highly intuitive; the visual acuity ratio is designed such that decreasing values indicate worsening vision. Based on the results from this exercise, a person with 20/10 vision (visual acuity of 2.00 in the model) would recognize the treatments from 53% farther away than a person with 20/20 vision (visual acuity of 1.00 in the model). Older participants were 13% closer to the treatments at the time of recognition. Decreased visual acuity is generally expected in older people, and this was considered in the analysis pro- cess; however, in this data set, the two characteristics were not shown to be highly correlated based on the examination of variance inflation factors. Additionally, interaction terms between the older participant variable and age (as well as with each of the treatment variables) were Parameter Estimate Percent Change Std. Error p- value Lap 1 Estimate Lap 2 Estimate Intercept 5.359 0.067 <0.001 5.345 5.516 Visual Acuity 0.426 53 0.041 <0.001 0.459 0.392 65 or Older -0.144 -13 0.028 <0.001 -0.106 -0.180 Heavy-Wear RPM 0.657 93 0.034 <0.001 0.729 0.695 Low-Retro RPM 0.759 114 0.034 <0.001 0.865 0.830 Medium-Retro RPM 1.042 183 0.035 <0.001 1.109 1.128 Simulated-Wear Tape -0.719 -51 0.045 <0.001 -0.703 -0.666 Overhead Lights -0.445 -36 0.027 <0.001 -0.208 -0.207 Right Side -0.116 -11 0.025 <0.001 -0.167 -0.153 Lap 2 0.071 7 0.022 0.001 — — Dispersion 0.081 -0.274 -0.196 Intercept-Only Log- likelihood -4707.379 -0.380 -0.349 Final Log-likelihood -4188.736 0.086 0.070 Table 48. Recognition of treatment model results.

68 Performance Criteria for Retroreflective Pavement Markers not shown to improve model fit. Subsequently, the effects of these parameters were considered separately in the analysis. With respect to the various treatments, new pavement marking tape was considered to be the base case upon which all other treatments were compared. As the retroreflectivity of the RPMs increased, the distance at which the treatment was recognizable also increased by 93%, 114%, and 183% for heavy-wear, low-retro, and medium-retro RPMs, respectively. Conversely, participants needed to be 51% closer to the simulated-wear pavement marking tape to be able to detect it. Participants had a harder time recognizing the treatments that were located under overhead lighting, needing to be 36% closer to recognize them. This is likely due to the overhead lighting overpowering the light retroreflected from the treatments and effectively reducing the contrast between the treatment and the surface. Further research is needed to identify if other treatment types (e.g., contrast markings) may be more recognizable in well-lit areas. Learning behavior was observed in the study since participants were able to observe the treat- ments at 8% longer distances during Lap 2. The learning behavior may be due to participants becoming more familiar with the course itself or with the study procedure. The output of the statistical model represents an equation that can be used to estimate the distance at which a person would be able to recognize an RPM or pavement marking tape treat- ment. While the study was conducted at low speeds (∼15 mph), the observation distance of the treatments can easily be converted to preview distance, which is expressed in seconds and dependent on the speed at which the vehicle is traveling. Following the transformation from distance to time, the output of the model can be interpreted as available preview time, that is, the amount of preview time for a road user dependent on the parameters included in the model. Figure 13 is a graphical representation of the statistical model with the output converted to preview time versus speed in miles per hour assuming an older driver, for drivers with 20/20 and 20/40 vision. These values represent typical and worst-case scenarios in terms of study participant vision. The treatments that were evaluated under overhead lighting are indicated with dashed lines. Figure 13 demonstrates that the RPMs provided significantly greater preview distance, regardless of the retroreflectivity levels, in comparison to the pavement marking tape materials. For older drivers with 20/40 vision, all of the RPMs met the 3.65-second criteria when not in the presence of overhead lighting, while the medium-retroreflectivity RPMs, corresponding to the ASTM ini- tial minimum standard for new RPMs, met the criteria across all speeds and lighting conditions (for older drivers with 20/20 vision, even the RPMs in lit areas provided sufficient preview time until speeds approached 75 mph). The RPMs representing TxDOT’s 1-year in-service criteria and the heavy-wear RPMs were capable of providing adequate preview time under the lights for speeds below 60 mph. The new tape markings were only able to meet the 3.65-second criteria for the lower speeds in unlit conditions, whereas the simulated-wear marking tape was unable to meet the 3.65-second criteria in any of the conditions. Figure 13 illustrates that as the denominator of the visual acuity ratio increased, the preview time associated with each treatment decreased. The new or simulated-wear tape markings were unable to provide adequate preview time at 75 mph for all participants. The low-retro and heavy- wear RPMs under the lit conditions were unable to meet the 3.65-second preview time criteria for all participants except those with very good vision. 4.3.3 Driver Recognition Study Summary The driver recognition study included 37 participants and 666 total observations of various treat- ments. The treatments included three levels of RPMs and two levels of pavement markings

Treatment Recognition, Visibility, and Driver Behavior Evaluations 69   in dark conditions and conditions with overhead lighting. The collected data were used to determine visibility of the treatments and provide data for the visibility modeling portion of the research (described in Section 4.4). Key aspects of the results are outlined below: • Older study participants needed to be closer to the treatment to see it, and as the visual acuity denominator increased (i.e., vision worsened), study participants needed to be closer to the treatment to be able to see it. This trend was observable over nearly every treatment. • Participants needed to be much closer to the treatments to recognize them as visual acuity worsened, as retroreflectivity decreased, and in the presence of overhead lighting. • Based on the median recognition distances for all participants, most of the treatment scenarios provided adequate preview time. However, the simulated-wear tape performed poorly in Figure 13. Preview time versus speed for 20/20 and 20/40 vision.

70 Performance Criteria for Retroreflective Pavement Markers terms of preview time, with unsatisfactory results for speeds in excess of 45 mph and barely meeting the comfortable minimum criteria at 45 mph. • The ability of study participants to recognize the treatments was found to decrease as vision worsened and the visual acuity parameter increased. Older participants were 13% closer to the treatments at the time of recognition. Decreased visual acuity is generally expected in older people; however, in this data set, the two characteristics were not shown to be highly correlated based on examination of variance inflation factors. • As the retroreflectivity of the RPMs increased, the distance at which the treatment was rec- ognizable compared to the new pavement marking tape increased by 93%, 114%, and 183% for heavy-wear, low-retro, and medium-retro RPMs, respectively. Conversely, participants needed to be 51% closer to the simulated-wear pavement marking tape to be able to detect it. • Participants had a harder time recognizing the treatments that were located under overhead lighting, needing to be 36% closer to recognize them. This is likely due to the overhead light- ing overpowering the light retroreflected from the treatments and effectively reducing the contrast between the treatment and the surface. Further research is needed to identify if other treatment types (e.g., contrast markings) may be more recognizable in well-lit areas. These findings provide insight into the visibility differences between RPMs and pavement markings of different retroreflectivity levels. The influence of visual acuity, participant age, and overhead lighting all play a role in the visibility of the treatments. The data collected in this portion of the study are the basis for the visibility modeling using the VL model. 4.4 RPM Visibility Modeling The VL model allowed the research team to look at the data collected from the recognition study from a different perspective by using modeled visibility requirements in combination with the study test conditions. The VL model is predicated on the assumption that the luminance of RPMs provided by vehicle headlamps follows the inverse square law. Consequently, additional details concerning the properties of the RPMs (beyond those used in the recognition study) needed to be collected to ensure that the model was properly calibrated. The basis of the VL model, the procedure for validating it, and the application of the VL model using data collected during the recognition study are discussed in the following subsections. 4.4.1 VL Model Method In the 1980s, Adrian established a VL model for a circular target (Adrian 1989). In 1999, the COST 331 Management Committee (COST 1999) extended Adrian’s VL model to pavement markings. In a TTI study for the Alaska DOT (Ye, Carlson, and Miles 2013), the COST VL was modified to evaluate the VL for pavement markings in Alaska. In this study, the VL model was enhanced and further extended to RPMs. The VL model for RPMs can be summarized in three steps: (1) calculate the actual lumi- nance difference between RPMs and the adjacent pavement surface; (2) calculate the threshold luminance difference between RPMs and the adjacent pavement surface; and (3) calculate the VL of a set of RPMs. More details on these steps are provided below. 4.4.1.1 Step 1—Calculate the Actual Luminance Difference In the study of the VL of RPMs, a set of RPMs is treated as the target, and their adjacent pave- ment surface is treated as the background. Accordingly, the luminance difference between the target and the background is calculated by Equation 7.

Treatment Recognition, Visibility, and Driver Behavior Evaluations 71   L L Lt b [7]actual∆ = − where ΔLactual = the actual luminance difference between RPM and pavement surface, Lt = the luminance of the target (i.e., RPM), and Lb = the luminance of the background (i.e., pavement surface). 4.4.1.2 Step 2—Calculate the Threshold Luminance Difference The threshold luminance difference is defined as the target contrast required for a human to detect an object with respect to its background, which indicates a value at which a target of defined size becomes perceptible with near 100% probability under the observation conditions. There are three main steps for calculating the threshold luminance difference, as stated below. Step 2.1—Calculate the Target Size of RPMs. In the model, target size refers to an angular size of a circular target. However, a set of RPMs is more complicated than a circular target. First, a set of RPMs is composed of several discrete surface areas, which subsequently cannot be seen as a circular object. Second, the nighttime luminance of RPMs and the pavement sur- face are not uniform, unlike daytime or when overhead lighting is present. The light cast on the RPMs is supplied from the headlamps of the study vehicle, which is at a different distance from each RPM in a specific treatment. Therefore, the luminance of RPMs is not uniform but is based on the inverse square law (illuminance from a point lighting source is inversely proportional to the square of the distance from the source). Therefore, an equivalent target size is used for RPMs, which is defined as the size of a circular target of the same solid angle as obtained by luminance-weighted integration over the surface of the RPMs considered for evaluation. For RPMs, target size α, measured in minutes of arc, is obtained by Equation 8. The equation was derived from the relationship between the solid angle of a cone (Ω) with apex angle α (i.e., target size in this study). ( )α = × ×π − Ωπ180 60 2 arccos 1 2 [8] where Ω = the solid angle of the RPMs obtained by luminance-weighted integration, with unit in steradian (sr). According to the definition of the equivalent target size, solid angle of RPMs from a specific distance D0 is the luminance-weighted integration over the surface beyond the distance D0, as calculated by Equation 9. L L d [9] 0 ∫Ω = Ω where dΩ = the solid angle of a differential element on RPMs, L = the luminance at the location of the small differential element on RPMs, which can be related to L0 based on the inverse square law, and L0 = the luminance at the target location of the differential element on RPMs (i.e., from distance D0).

72 Performance Criteria for Retroreective Pavement Markers Based on the denition of the solid angle of a cone, which is the ratio of the area cut out on a spherical surface (with its center at the apex of that cone as shown in Figure 14) to the square of the radius of the sphere, dΩ can be given by Equation 10. d D dAcos [10] 2 Ω= θ where θ = the angle of view, measured from the direction of view to the normal of the pavement surface, cosθ = the cosine of θ, calculated by θ=cos 0H D (H0 is the height of a driver’s eyes, D is the distance between the driver and RPMs), and dA = the area of a dierential element on RPMs. By combining Equations 9 and 10, it is easy to obtain Equation 11 for the solid angle of RPMs. L L H D dA [11] 0 0 3∫Ω= × e calculation of Equation 11 is complicated due to dierent combinations of the following factors: type of light sources, such as headlight or diuse light (daylight or roadway lighting); geometry with respect to the location of RPMs and the driver; geometry of the road, such as the horizontal curvature, road width, lane width, and number of lanes; and target distance for analysis (i.e., D0). e details and further derivation of Equation 9 can be found in the Alaska study (Ye, Carlson, and Miles 2013). Step 2.2—Calculate the Adjusted Background Luminance. e presence of glare sources, such as oncoming headlights, impairs drivers’ vision, which results in a higher requirement for target contrast (i.e., threshold luminance) in order for targets to remain visible. e eects of disability glare on the VL of RPMs can be expressed by adding a veiling luminance to the luminance of the pavement surface (Lb). e calculations for veiling luminance are based on well-established visibility work (COST 1999). Step 2.3—Calculate the reshold Luminance Dierence. Based on the adjusted back- ground luminance calculated above, the standard threshold luminance dierence can be calcu- lated, subdividing the functions into three ranges of background luminance. Accordingly, the threshold luminance dierence (ΔLthreshold) is calculated by incorporating the contrast polarity factor, exposure time factor, and age factor, if there is any, as shown in Equation 12. L L F F AFCP t [12]threshold∆ = ∆ ∗ ∗ ∗ where ΔL = the standard threshold luminance dierence, FCP = the contrast polarity factor, Ft = the expose time factor, and AF = the age factor. e calculation of the contrast polarity factor, exposure time factor, and age factor follows the same method as used in the COST VL model. More details can be found in the COST (1999) report. Figure 14. Solid angle of a circular target.

Treatment Recognition, Visibility, and Driver Behavior Evaluations 73   4.4.1.3 Step 3—Calculate the VL As the final step, VL is obtained by the ratio of the actual luminance difference of RPM, calculated by Equation 7, to its threshold value, calculated by Equation 12. Although the VL model has been developed based on solid theories and long-term proven empirical studies, the model has not been tested for RPMs. Therefore, before applying the VL model to evaluate the visibility performances of RPMs, the model was validated using the field-measured data. 4.4.2 VL Model Validation The VL model was built based on a variety of assumptions and empirical equations from lab and field experiments. It is difficult to check each assumption and equation without conducting a study dedicated to the evaluation. For validation purposes in this study, the key assumption of inverse square law in RPM luminance was verified, which had not been done before. In addi- tion, the overall performance of the VL model was estimated by comparing the real VL with the calculated values in order to check whether there was a need to modify the model to produce data-supported outcomes. 4.4.2.1 Validation of Inverse Square Law in RPM Luminance As illustrated in Step 2.1, calculation of the equivalent target size in the VL model is based on luminance-weighted integration over the surface of RPMs. When vehicle headlamps are the only light source to RPMs in intermittent groups, their luminance is assumed to rely on the distance from headlamps by the inverse square law. This assumption was verified using the photometric information collected as discussed below. This assumption is a fundamental one in the VL model. In order to verify the assumption, a set of RPMs were placed in a straight line on a runway at the RELLIS Campus and their luminance was captured with a photometric camera mounted in a 2004 Toyota Highlander. Figure 15 provides an image of the photometric camera from the side. Yellow RPMs were placed to the left of the vehicle at 40-ft spacing and white RPMs to the right at 80-ft spacing (shown Figure 16a). The simulated driving lane was 12 ft wide. The vehicle’s low- beam headlamps were used. Researchers used ProMetric software to obtain luminance values of 3 pixels within each marker in the image, as shown in Figure 16b. The average of the 3 pixels’ luminance values was used to represent the luminance of that marker. Based on the measured luminance, the luminance ratio of two consecutive RPMs in the line was calculated and compared to the inverse ratio of their square distances to the vehicle, Figure 15. Charge-coupled device photometric camera.

74 Performance Criteria for Retroreflective Pavement Markers as shown in Figure 17 and Figure 18 for yellow and white RPMs, respectively. If the inverse square law assumption is valid, those two ratios should be equal in theory. According to both figures, the white RPMs have a larger discrepancy between the two ratios than the yellow markers due to larger spacing, which leads to a larger bias in the luminance ratio. Considering that markers are not uniformly degraded and get dirty unevenly on roads, which results in differences in luminance from one marker to another, the discrepancy between the two ratios for both white and yellow RPMs is acceptable. The finding indicates that the assumption of inverse square law in RPM luminance is appropriate for the test conditions. (a) (b) Figure 16. RPM luminance measurements. 0 0.5 1 1.5 2 2.5 3 3.5 240 280 320 360 400 440 480 520 560 600 640 680 720 760 800 Lu m in an ce R ati o Distance between Vehicle and the First RPM (feet) Luminance Ratio Inverse Ratio Figure 17. Luminance ratio versus inverse distance ratio square (yellow RPMs on the left).

Treatment Recognition, Visibility, and Driver Behavior Evaluations 75   4.4.2.2 Estimation of Overall Model Performance To estimate the overall performance of the VL model, the data collected from the recognition study described in Section 4.3 were used. To estimate the overall performance of the VL model, the marking/RPM treatments without road lighting were used and are summarized in Table 49. For the purposes of this analysis, the median detection distances for two age groups, older drivers (65 and older) and other drivers (i.e., younger drivers), were identified as relevant data points. The medians of the detection distances for these age groups are summarized in Table 50. At this point in the study, additional luminance measurements for the treatments (both RPMs and markings) and the adjacent concrete pavement were collected using a charge-coupled device photometer based on the median detection distances for each age group. The measured lumi- nance values for the first marking segment/RPM and its adjacent pavement surface are also listed in Table 50. The images in Figure 19 were captured using the photometric camera (see Figure 15) and have the colors adjusted to better show the luminance difference between the treatments and the background. The luminance values at the median detection distances for each age group were used as the inputs of the VL model to estimate the average visibility performances of each treatment from the perspective of younger drivers and older drivers. Meanwhile, the average ages of each group (31 years for the younger group and 75 for the older group) were used correspondingly in the VL calculation. The calculated VLs of each treatment by each age group are summarized in Figure 20. 0 0.5 1 1.5 2 2.5 3 3.5 240 320 400 480 560 640 720 800 Lu m in an ce R ati o Distance between Vehicle and the First RPM (feet) Luminance Ratio Inverse Ratio Figure 18. Luminance ratio versus inverse distance ratio square (white RPMs on the right). Treatment Type Retroreflectivity Level Deviation Marking A4 New tape 500 mcd/m2/lux Right C1 New tape 500 mcd/m2/lux Left C5 Simulated-wear tape 100 mcd/m2/lux Right RPM A3 Low-retro RPM 65 mcd/lux Right B3 Medium-retro RPM 167 mcd/lux Left B4 Heavy-wear RPM 30 mcd/lux Left C2 Medium-retro RPM 167 mcd/lux Right C3 Heavy-wear RPM 30 mcd/lux Right C4 Low-retro RPM 65 mcd/lux Left Table 49. Marking/RPM treatments in the recognition study.

76 Performance Criteria for Retroreflective Pavement Markers Treatment Median Detection Distance (age group) Luminance (cd/m2) Marking/RPM Pavement Marking A4 357 ft (younger) 0.885 0.051 249 ft (older) 1.958 0.071 C1 309 ft (younger) 0.692 0.038 242 ft (older) 1.085 0.042 C5 206 ft (younger) 0.503 0.081 123 ft (older) 1.175 0.180 Treatment Median Detection Distance (age group) Luminance (cd/m2) Marking/RPM Pavement RPM A3 620 ft (younger) 5.284 0.048 469 ft (older) 11.003 0.030 B3 819 ft (younger) 12.764 0.092 522 ft (older) 21.519 0.069 B4 611 ft (younger) 4.737 0.019 422 ft (older) 13.538 0.032 C2 1,254 ft (younger) 7.500 0.001 814 ft (older) 20.254 0.002 C3 762 ft (younger) 3.381 0.001 511 ft (older) 13.555 0.026 C4 785 ft (younger) 5.943 0.006 574 ft (older) 10.849 0.024 Table 50. Marking/marker luminance at median detection distances. Figure 19. Luminance measurement of markings (top image) and RPMs (bottom image). The detection distance is the distance between the participant and the target, where the partic- ipant initially detects the target, so the VL value at the detection distance should be the threshold of visibility. A VL value of 10 has been used for traffic situations (COST 1999; Adrian 1989). In Figure 20, the horizontal dashed line represents the VL threshold (i.e., VL = 10). Comparing the calculated VLs with the VL threshold shows that the VLs vary by the treatments and age groups but bounce up and down around 10. For markings, the mean values of the calculated VLs at median detection distances are 11.3 for younger drivers and 10.4 for older drivers, while the cor- responding values are 10.3 and 10.4 for RPMs. The finding indicates that the VL model overall provided a reasonable estimation of VL for both markings and RPMs in the studied scenarios. In addition, the fluctuation of the calculated VLs at drivers’ detection distances does not show any pattern across the retroreflectivity levels and drivers’ age groups for either markings or RPMs. It

Treatment Recognition, Visibility, and Driver Behavior Evaluations 77   indicates that the VL model has no obvious systematic deviation but some measurement errors or variations in drivers’ visibility performances. 4.4.3 VL Model Application Because the VL model provided a good estimation of visibility performances of studied markings and RPMs, it was used to expand the breadth of understanding of the effect different factors have on treatment visibility. These factors included retroreflectivity, spacing, number of consecutive RPMs, marker/marking in-service condition, and glare. Using the visibility model allowed researchers to study these factors in a more cost-effective way compared to additional human factor testing. Since visibility performances are distance-based, detection distances were first determined to study the VL of markings/RPMs. A 3.65-second preview time was chosen in the study for detection distances, making the driving task easier and allowing drivers to make comfortable adjustments compared to a shorter preview time. A preview time of 3.65 seconds has been used in other research regarding markings/RPM visibility at night, including the study on the pro- posed minimum required retroreflectivity for pavement markings (Zwahlen and Schnell 1998). With 3.65 seconds as the preview time, the corresponding detection distance is 97 m (318 ft) for a vehicle speed of 45 mph, 118 m (387 ft) for 55 mph, and 140 m (459 ft) for 65 mph. As inputs to the VL model, luminance values of marking/RPM and the adjacent pavement surface at a detection distance were measured and are summarized in Table 51. The markings/ RPMs were the same ones used in the visibility study for the VL model validation. The average luminance values of markings/RPMs of the same retroreflectivity level were used to represent the typical luminance value of that retroreflectivity level at a specific detection distance. The luminance of the pavement listed in Table 51 was also the average value of the measured luminance at the detection distances. Note that at the three detection distances in Table 51, there is no significant difference between the luminance values of the same markings/RPMs on the left and right sides of the vehicle. Rela- tive to the long longitudinal distances, the small change in the lateral offset of marking/RPM does not cause any big change in luminance. Therefore, the following analysis did not specify the lateral location of marking/RPM of a driving lane. VL=10 0 2 4 6 8 10 12 14 A4 C1 C5 A3 B3 B4 C2 C3 C4 Marking RPM Vi sib ili ty L ev el Treatment young older Note: A4 and C1 = new tape; C5 = simulated-wear tape; A3 = low-retro RPM; B3 = medium-retro RPM; B4 = heavy-wear RPM; C2 = medium-retro RPM; C3 = heavy-wear RPM; C4 = low-retro RPM. Figure 20. Calculated VL for marking/RPM by age group.

78 Performance Criteria for Retroreflective Pavement Markers Meanwhile, the driver’s age of 65 was used in the VL calculation since this research aimed to study the minimum performance criteria of RPMs that needs to meet older drivers’ visual demands. By assuming markings/RPMs on a tangent and drivers in a regular passenger car, researchers calculated the VLs for markings/RPMs at different retroreflectivity levels. The markings were assumed to be broken lines 4 in. wide, 10 ft long, and with 30-ft gaps in between. RPMs are regularly sized with their face size of 13⁄16 inch by 2 5⁄8 inch at a 30-degree angle up from the ground. In the VL calculation, different marker numbers were tried—three (at least three markers are needed to guide the roadway alignment) to 10— and four spacings were involved: 20 ft, 40 ft, 80 ft, and 120 ft. The results of the calculated VLs are provided in the following sections. 4.4.3.1 Marking versus RPM Visibility In general, the calculated VLs of RPMs were much higher than those of markings at all three detection distances since the markers’ luminance was much larger than that of the markings. As shown in Figure 21, the VLs of three RPMs along the road spaced at 120 ft (lowest VL for RPMs among the studied scenarios) were compared with a set of broken markings. With the minimum Treatment Target Detection Distance (driving speed) 97 m (45 mph) 118 m (55 mph) 140 m (65 mph) New Tape (500 mcd/m2/lux) Marking 2.166 1.230 0.792 Pavement 0.070 0.047 0.030 Simulated-Wear Tape (100 mcd/m2/lux) Marking 0.367 0.183 0.118 Pavement 0.057 0.038 0.030 Medium-Retro RPM (167 mcd/lux) RPM 83.472 71.777 42.726 Pavement 0.056 0.041 0.030 Low-Retro RPM (65 mcd/lux) RPM 51.101 33.788 29.004 Pavement 0.057 0.032 0.026 Heavy-Wear RPM (30 mcd/lux) RPM 18.624 16.186 12.159 Pavement 0.057 0.030 0.028 Table 51. Luminance of marking/RPM and adjacent pavement (cd/m2). Figure 21. Comparison of VLs between RPMs and markings.

Treatment Recognition, Visibility, and Driver Behavior Evaluations 79   marker numbers and largest spacing, markers of all three retroreflectivity levels still had better visibility performances than the markings. This finding is consistent with the common observa- tion that markers are much brighter than markings on roads. Meanwhile, RPM had VLs much larger than 10, which indicates that all markers including those of low retroreflectivity can meet older drivers’ visual needs on roads with speed limits of 45 mph, 55 mph, or 65 mph. Older drivers did not see markings as clearly as RPMs. Older drivers had trouble seeing the simulated-wear tape at all three speeds and the new higher retroreflec- tivity pavement marking tape at speeds of 65 mph, even on clear nights. Therefore, solely relying on markings to provide guidance on roads may not be sufficient for the given test conditions. The impact of wet weather was not considered in the data presented in Figure 21. 4.4.3.2 RPM Performance Criteria To estimate the performance criteria of RPMs, VLs of markers were calculated in terms of different combinations of spacing and the number of markers in view for three driving speeds (45 mph, 55 mph, and 65 mph). A 65-year-old was used in the calculation to represent older drivers’ visual demands. VL Calculation Based on Three Sets of Luminance. Since the study focused on the mini- mum performance of RPMs, luminance measurements from the heavy-wear RPMs—30 mcd/ lux (one-half of the TxDOT 12-month in-service retroreflectivity level)—were used in the VL calculation. However, the measured luminance of this type of marker might not be low enough to represent the typical visibility performance of in-service RPMs, which will have lower retro- reflectivity levels as they approach their end of service life. Consequently, three sets of marker luminance values were used: the first set was the measured luminance of markers of heavy- wear retroreflectivity, representing ideally low luminance of in-service markers; the second set reduced the measured luminance by half, representing more realistically low luminance (dis- counted luminance) of markers on roads; and the last set added the glare impact into the second set, representing adverse conditions and low-quality markers. Calculated VLs Using Measured Luminance of Heavy-Wear RPMs (Ideally Low Luminance). Using the luminance values of the heavy-wear RPMs (RI = 30 mcd/lux) at different detection distances (see Table 51) as the VL model input, VLs were calculated for different combinations of spacing and number of consecutive markers visible. The results are summarized in Figure 22. Note that the number of RPMs in the figure is only up to seven because the increase of luminance with more RPMs is negligible when the number of RPMs reaches seven. A general pattern can be found in the above figure showing that a larger spacing between markers reduces the visibility performance of RPMs. On the other hand, with the same spacing, adding markers in view can increase the VL, but the amount of increase slows down and eventu- ally stops with the increase of distance between marker and vehicle. This finding indicates that it is important to ensure drivers see a sufficient number of markers that provide clear guidance on the road at night. The calculated VLs were compared with the threshold VL value (i.e., VL = 10), as shown in Figure 22. For all three driving speeds, the markers were bright enough to meet older drivers’ visual demands at night regardless of spacing and number of markers. Calculated VLs Using Discounted Luminance (Realistically Low Luminance). RPMs have a service life of a few years, with the retroreflective surface providing some level of retro- reflectivity for up to 3 years; the survey results in Chapter 3 indicate that most states reinstall markers in a window between 2 and 5 years. The heavy-wear RPMs in the study represented RPMs in service for more than a year but might not represent the in-service markers near the end

80 Performance Criteria for Retroreflective Pavement Markers of their service life that were even less visible. The measured luminance values of the heavy-wear RPMs shown in Table 51 were reduced by 50% (termed discounted luminance) to analyze VLs of in-service markers under lower visibility conditions. The calculated VLs using the discounted luminance are summarized in Figure 23. The same patterns of VLs can be found in Figure 23. However, with lower luminance, the VLs dropped for all conditions. Compared with the VL threshold, the markers with discounted luminance still Figure 22. VLs based on measured luminance of heavy-wear RPMs (ideally low luminance). Figure 23. VLs based on discounted luminance (realistically low luminance).

Treatment Recognition, Visibility, and Driver Behavior Evaluations 81   met older drivers’ visual needs at driving speeds of 45 mph and 55 mph for all the combinations of spacing and number of markers. For 65 mph roads, markers could only be spaced at 20 ft or 40 ft to be seen clearly from the required detection distance. Calculated VLs Using Discounted Luminance with Glare (Adverse Condition Luminance). The glare from headlamps of opposing vehicles is not negligible for drivers’ visibility, especially on two-way, two-lane highways. Consequently, glare from headlamps of an oncoming vehicle in the adjacent lane was taken into account by adding 0.098 cd/m2 veiling luminance in the VL cal- culation (see VL Model Method Step 2.2 for more details), termed adverse condition luminance. The discounted RPM luminance values were used in the analysis. The results are summarized in Figure 24. For driving speeds of 45 mph, all the combinations of spacing and number of markers met older drivers’ visual demands with the existence of glare. For 55 mph, markers spaced at 20 ft or 40 ft met the needs, but larger spacing resulted in a VL below the threshold. For 65 mph, spacing needed to be set as small as 20 ft with at least six mark- ers in view for older drivers to detect markers in time. 4.4.3.3 Suggested RPM Performance Criteria Based on the above analysis for RPMs using the three sets of luminance and with various combinations of spacing and number of markers, the setting criteria and minimum required luminance and retroreflectivity were derived and proposed for the three driving speeds. RPM Setting Criteria. Based on Figure 22, Figure 23, and Figure 24, RPM setting criteria in terms of spacing and number of markers in view are summarized in Table 52 for the three driving speeds. “X” in the table means there is no requirement for the case because older drivers’ visibility needs were met in all tested scenarios. According to the table, the heavy-wear RPMs in the ideally low luminance condition are bright enough for older drivers to detect at night for all three driving speeds, with the largest Figure 24. VLs based on discounted luminance and glare (adverse condition luminance).

82 Performance Criteria for Retroreflective Pavement Markers spacing of 120 ft and a minimum number of three markers. More realistically for RPMs at the realistically low luminance condition level on roads of 65 mph, markers need to be spaced at 20 ft or 40 ft (with at least four RPMs in view). Considering the adverse luminance condition of glare and discounted RPM luminance level on roads of 55 mph, markers cannot be spaced at 80 or 120 ft, and for roads of 65 mph, markers need to be spaced at 20 ft with at least six markers in view. Minimum Required Luminance and Retroreflectivity. In order to find the minimum required luminance for RPMs at various settings for the three speeds, additional analyses were completed using varying luminance in the VL model. The breakpoint luminance was achieved when the calculated VL was around the threshold value (i.e., VL = 10), and the breakpoint luminance was used as the minimum required luminance. Table 53 summarizes the minimum required marker luminance for different combinations of spacing and number of markers without glare. The required values are listed in Table 54 when glare from the headlamps of an oncoming vehicle in the adjacent lane was included in the calculation. For a specific distance and geometry between a vehicle and a set of markers, the luminance of the markers is proportional to their retroreflectivity under the same headlamp illumination. Therefore, by setting the ratio between the minimum required luminance (in Table 53 and Table 54) and measured luminance of the heavy-wear RPM (in Table 51) equal to the ratio between their corresponding retroreflectivity values, the minimum required marker retro- reflectivity was derived and is summarized in Table 55 without glare and Table 56 with glare. Based on the VL model, the minimum maintained retroreflectivity of the RPMs should be 20 mcd/lux. Driving Speed (mph) Setting Criteria Luminance Condition Ideally Low Realistically Low Adverse 45 Spacing (ft) X X X Number of markers X X 4+ for 120-ft spacing 55 Spacing (ft) X X 20, 40 Number of markers X X X 65 Spacing (ft) X 20, 40 20 Number of markers X 4+ for 40-ftspacing 6+ Table 52. RPM setting criteria for different scenarios. Speed (mph) Spacing (ft) Number of Markers3 4 4+ 45 20 3.7 3.1 3.0 40 4.5 4.2 3.9 80 5.8 5.7 5.5 120 6.8 6.6 6.4 55 20 4.2 3.6 3.2 40 5.0 4.5 4.2 80 6.5 6.2 6.0 120 7.8 7.5 7.5 65 20 5.5 4.7 4.2 40 6.4 5.8 5.4 80 8.3 7.8 7.6 120 9.8 9.5 9.3 Table 53. Minimum required RPM luminance without glare (cd/m2).

Treatment Recognition, Visibility, and Driver Behavior Evaluations 83   Speed (mph) Spacing (ft) Number of Markers3 4 4+ 45 20 5.1 4.5 4.2 40 6.0 5.8 5.3 80 7.9 7.9 7.5 120 9.3 9.2 9.2 55 20 6.1 5.4 5.0 40 7.4 6.8 6.4 80 9.6 9.2 8.9 120 11.4 11.1 11.1 65 20 8.3 7.2 6.4 40 9.6 8.7 8.1 80 12.3 11.4 11.2 120 14.6 14.1 13.9 Table 54. Minimum required RPM luminance with glare (cd/m2). Speed (mph) Spacing (ft) Number of Markers3 4 4+ 45 20 6.0 5.0 4.8 40 7.2 6.8 6.3 80 9.3 9.2 8.9 120 11.0 10.6 10.3 55 20 7.8 6.7 5.9 40 9.3 8.3 7.8 80 12.0 11.5 11.1 120 14.5 13.9 13.9 65 20 13.6 11.6 10.4 40 15.8 14.3 13.3 80 20.5 19.2 18.8 120 24.2 23.4 22.9 Table 55. Minimum required RPM retroreflectivity without glare (mcd/lux). Speed (mph) Spacing (ft) Number of Markers 3 4 4+ 45 20 8.2 7.2 6.8 40 9.7 9.3 8.5 80 12.7 12.7 12.1 120 15.0 14.8 14.8 55 20 11.3 10.0 9.3 40 13.7 12.6 11.9 80 17.8 17.1 16.5 120 21.1 20.6 20.6 65 20 20.5 17.8 15.8 40 23.7 21.5 20.0 80 30.3 28.1 27.6 120 36.0 34.8 34.3 Table 56. Minimum required RPM retroreflectivity with glare (mcd/lux).

84 Performance Criteria for Retroreflective Pavement Markers This will provide adequate visibility when three or more RPMs are in view at 80-ft spacing and travel speeds are 65 mph or less. The minimum maintained retroreflectivity value may be lower if a shorter RPM spacing is used, a greater number of visible markings are maintained, or travel speeds are less. The minimum maintained retroreflectivity value may be higher if a longer RPM spacing is used, travel speeds are higher, or oncoming vehicle glare is a concern. 4.4.4 Limitations The suggested RPM performance criteria, including setting criteria and minimum luminance/ retroreflectivity levels presented in this study, are the product of the VL model developed for this study and simplifying assumptions based on previous research efforts. Therefore, these recom- mendations are subject to the following limitations. • The study relies on the measured luminance from yellow RPMs. The effect of color on markers’ visibility was not included in the study, with the assumption that the detection dis- tances to yellow and white markers will have little difference if their retroreflectivity is at the same level. • The calculated VLs were based on straight roadways and longitudinally intermittent RPMs. Curved roadway segments, especially those with small radii, change the geometry between vehicle and markers and place the markers in a different location in the projected headlamp beam pattern. Consequently, the visibility of RPMs on curves is different from those on tangents. However, whether they are more or less visible at curves is not certain without further study. • Drivers at the age of 65 years old were used in the study. If the percentage of drivers older than 65 is high, the proposed RPM performance criteria need to be adjusted accordingly. • The analysis was based on the data collected from RPMs. Snowplowable RPMs and recessed RPMs were assumed to have similar visibility performances to surface-applied raised RPMs. • Luminance data of the pavement surface were measured on concrete pavement. Asphalt is darker than concrete pavement, which leads to a larger contrast between markers and the pavement surface and might improve the visibility of markers. However, this study assumed the difference in visibility caused by types of pavement was negligible. • The study suggests RPM performance criteria under three conditions: ideally low (measured luminance of heavy-wear RPMs), realistically low (discounted luminance), and adverse (dis- counted luminance with oncoming glare) conditions. However, understanding how to adopt the criteria would rely on engineering judgment to evaluate whether or not a specific condi- tion warrants the use of markers with stricter requirements. Ideally, the RPMs would have been evaluated in a rainy nighttime condition, which is when RPMs will provide their greatest benefit. • The proposed minimum required RPM retroreflectivity values were based on the luminance measurements using the Ford Explorer’s LED headlights as light sources in the study. In reality, each vehicle has different headlamp characteristics, resulting in different illumination and RPM luminance values. The retroreflectivity values stated are what is required based on the analysis. • Since the VL model can only evaluate the visibility performance of RPMs or markings sepa- rately, the proposed RPM performance criteria did not consider the potential impacts of the coexistence of markings or other TCDs on the overall visibility. 4.4.5 Visibility Modeling Summary The results from the visibility modeling indicate that the VL model is a valid tool for evalu- ating the VL of RPMs and pavement markings. The model proved valid, and evidence of the

Treatment Recognition, Visibility, and Driver Behavior Evaluations 85   effectiveness of RPMs across a range of factors can be efficiently and economically evaluated. Ultimately, the findings provide additional insight into the visibility performance of RPMs, and how that performance compares to drivers’ needs. Key aspects of the findings are out- lined below. • The VL model was validated and proved to provide a good estimation of visibility performances of studied markings and RPMs. • The VL model was used to expand knowledge on the effect retroreflectivity, spacing, number of consecutive RPMs, marker/marking in-service condition, and glare have on visibility. Using the VL model allowed researchers to study these factors in a more cost-effective way compared to additional human factor testing. • By using the VL model with various combinations of spacing (20 ft, 40 ft, 80 ft, and 120 ft) and the number of markers (from 3 to 10), the calculated VLs of RPMs were much higher than those of markings for all scenarios at the distances evaluated. The finding is consistent with the common observation that markers are much brighter than markings on roads. • Older drivers did not see markings as clearly as RPMs. Older drivers had trouble seeing the simulated-wear tape at all three evaluated speeds and the new higher retroreflectivity pave- ment marking tape at speeds of 65 mph, even on clear nights. Therefore, solely relying on markings to provide guidance on roads may not be sufficient for the given test conditions. • RPMs of differing retroreflectivity levels will have different VLs. Different quantities of RPMs will be needed in view depending on the retroreflectivity level and speed of travel. The heavy- wear RPMs (ideally low luminance condition) are bright enough for older drivers to detect at night for all three evaluated driving speeds, with the largest spacing of 120 ft and a minimum number of three markers. More realistically for RPMs at the discounted luminance (realisti- cally low luminance condition) level on roads of 65 mph, markers need to be spaced at 20 ft or 40 ft. Considering the adverse condition of glare and discounted RPM luminance level on roads of 55 mph, markers cannot be spaced at 80 or 120 ft, and for roads of 65 mph, markers need to be spaced at 20 ft with at least six markers in view. • Specific minimum requirements for luminance and retroreflectivity were found based on a VL of 10 for different driving scenarios. 4.5 Driver Behavior Through Curves Analysis To begin to understand the driver behavior in the curves, speed profiles were developed using a moving average speed of all vehicles as they passed through a curve under a specific treatment scenario. The moving average considered the previous 10 and next 10 observations from each point relative to the location along the curve (and preceding roadway area). Perhaps the most interesting speed profile was developed for Curve 3, shown in Figure 25. The figure clearly captures the fact that vehicles tended to enter Curve 3 at higher speeds when no RPMs were present, particularly on the first lap through the course. This tended to result in a large reduction of speed occurring in the curve itself, along with generally poor lanekeeping behavior that is not captured in the figure. Subsequently, several of the analytical approaches used in this study were intended to capture changes in vehicle speed before, at, and in the curve. Speed profiles for the other curves can be found in Appendix A. A three-pronged approach was used to investigate the effectiveness of RPMs in curves involving (a) a table identifying the percentage of trips through each curve where the centerline was encroached, (b) a series of plots investigating the nonencroachment performance measures, and (c) a statistical analysis investigating several of the performance measures. To conduct these analyses, the data collected for this study were reduced such that each observation of the data set was representative of one pass through a specific curve by a specific driver.

86 Performance Criteria for Retroreective Pavement Markers e analysis unit for the statistical analyses was identied as being a single pass through a curve by a vehicle. With the analysis unit in mind, descriptive statistics for the data set on which the statistical analyses were conducted are shown in Table 57. 4.5.1 Centerline Encroachments Centerline encroachments were tabulated using ArcGIS to determine whether the GPS antenna, which was located in the center of the vehicle, crossed the centerline of the road- way between the PC and PT of the curve. Using a special query, instances where the center of the vehicle crossed the centerline of the roadway were identied and classied as severe encroachments. is methodology, which represents a conservative tally of the number of centerline encroachments, was used due to the precision (1 m) of the GPS equipment used to collect the data. Higher-precision GPS would have allowed for identifying when the driver’s side tire (or some other oset from the center of the vehicle) encroached the centerline. e percentages of trips resulting in centerline encroachments by curve, lap, and RPM retrore- ectivity level are shown in Table 58. Note that each curve was treated with multiple RPM retroreectivity levels. e table indicates a nding that is not surprising: more centerline encroachments were observed on curves treated with no RPMs or with RPMs with obscured retroreective material. Perhaps a more interesting observation can be made for Curve 3 specically, which suggests a signicant improvement in successful curve navigation by drivers in Laps 2, 3, and 4 versus the initial lap when no RPMs are present. is observation does not translate to Curve 2, which was the other curve to be treated with no RPMs, nor Curve 5, which was treated with the obscured RPMs. To assess the eect of RPMs on encroachment while controlling for participant Figure 25. Curve 3 speed prole.

Treatment Recognition, Visibility, and Driver Behavior Evaluations 87   Parameter Mean Std. Deviation Minimum Maximum Female 0.48 0.50 0.00 1.00 Age 56.70 23.35 19.00 87.00 Under 29 0.18 0.39 0.00 1.00 Over 64 0.58 0.49 0.00 1.00 Vehicle Speed at Maximum Brake Application 32.44 4.07 0.00 42.31 RPM Presence 0.83 0.37 0.00 1.00 20 ft RPM Spacing 0.17 0.37 0.00 1.00 No-RPM 0.17 0.37 0.00 1.00 Obscured RPM 0.08 0.28 0.00 1.00 Low-Retro RPM 0.29 0.45 0.00 1.00 Medium-Retro RPM 0.38 0.48 0.00 1.00 High-Retro RPM 0.08 0.28 0.00 1.00 Curve Arc Length (ft) 340.39 60.44 238.77 434.62 Curve Radius (ft) 354.99 120.11 202.61 525.39 Central Angle (degrees) 61.46 24.13 39.06 100.57 Maximum Brake Application (%) 33.24 13.41 0.90 74.77 Speed at PC 29.80 3.99 18.09 41.73 Oncoming Headlights 0.17 0.37 0.00 1.00 Treatment Start to PT (ft) 532.39 62.85 435.25 633.49 Encroached Centerline 0.04 0.20 0.00 1.00 Maximum Deceleration (g’s) 0.34 0.11 0.14 0.86 Distance to PT at Maximum Brake Application (ft) 440.21 157.66 0.72 951.12 Curve Speed Differential (mph) 2.89 3.67 0.00 36.03 Curve Lateral Position Change (ft) 1.31 2.76 0.11 44.34 Table 57. Descriptive statistics of analysis data set. RPM Level Curve/Lap High Medium Low Obscured None Curve 1 – 3.03% 0.00% – – Lap 1 – 5.56% 0.00% – – Lap 2 – 0.00% 0.00% – – Lap 3 – 6.67% 0.00% – – Lap 4 – 0.00% 0.00% – – Curve 2 – 0.00% 0.00% – 10.61% Lap 1 – 0.00% 0.00% – 0.00% Lap 2 – 0.00% 0.00% – 18.75% Lap 3 – 0.00% 0.00% – 17.65% Lap 4 – 0.00% 0.00% – 5.88% Curve 3 0.00% – – – 24.24% Lap 1 0.00% – – – 64.71% Lap 2 0.00% – – – 0.00% Lap 3 0.00% – – – 18.75% Lap 4 0.00% – – – 12.50% Curve 4 – 1.52% 0.00% – – Lap 1 – 0.00% 0.00% – – Lap 2 – 0.00% 0.00% – – Lap 3 – 0.00% 0.00% – – Lap 4 – 5.56% 0.00% – – Curve 5 – 1.52% – 4.55% – Lap 1 – 0.00% – 0.00% – Lap 2 – 0.00% – 10.53% – Lap 3 – 0.00% – 7.14% – Lap 4 – 7.14% – 0.00% – Curve 6 – 1.52% 0.00% – – Lap 1 – 5.26% 0.00% – – Lap 2 – 0.00% 0.00% – – Lap 3 – 0.00% 0.00% – – Lap 4 – 0.00% 0.00% – – Table 58. Centerline encroachments by curve, lap, and RPM level.

88 Performance Criteria for Retroreflective Pavement Markers characteristics, a binary logistic regression, or logit, model was estimated. The results of the statistical analysis are presented in Table 59. At first glance, the results of the logit model for severe encroachment (where the center of the car is over the centerline treatments) are intuitive. Two participant characteristics that were found to affect the probability of severe encroachment were gender and visual acuity ratio. Unsurprisingly, these characteristics were also found to play a significant role in lateral posi- tion change and are discussed again later. Females were found to have 60% lower log odds to encroach the centerline. Participants with greater visual acuity, and therefore larger visual acuity ratios, are associated with a lower probability of encroaching the centerline. The effect of no RPMs in Curves 2 and 3 is positively associated with a higher probability of encroachment. The observed effect is more pronounced for Curve 2, which had the smallest radius of any of the course curves. Obscured RPMs are also positively associated with a higher probability of encroachments. No statistically significant effect was observed for the three RPM treatments of different retroreflectivity levels. Last, a binary variable indicating the loop number is associated with a lower probability of encroachment. This could likely be due to participants achieving some familiarity and comfortability in driving along curves after the first loop. One drawback to the use of logit models is the interpretation of the model. The response variable, the log odds of an event occurring, is a transformation of the probability of the event occurring. Subsequently, the probability of an event occurring can be back-calculated from the model results. Figure 26 and Figure 27 show the probability of encroachment, expressed as a value ranging from 0 to 1, for various levels/ratios of visual acuity for the various treatments of the centerline [i.e., the presence of any retroreflectivity level RPM; the presence of obscured (taped over) RPMs; and individually for Curves 2 and 3, the lack of RPMs]. The three retro- reflectivity levels of RPMs are defined in Table 59. As can be observed by comparing the two plots, females exhibit a lower probability of encroaching the centerline, and this effect is present across all RPM VLs, as well as in the presence of obscured RPMs, and in the lack of RPMs for Curves 2 and 3, individually. Additionally, males exhibit a higher magnitude of change in probability of encroachment across visual acuity ratios. 4.5.2 Maximum Deceleration Maximum deceleration was selected as a performance measure in order to capture hard brak- ing due to differences in visibility between curve treatments. Figure 28 presents the maximum deceleration force observed in each curve. In order to assess deceleration related to RPMs, only the area 400 ft prior to the treatment start through the end of the curve (not the overall treatment area) was considered. These bounds were chosen to mitigate Parameter Estimate (β) Std. Error p-value Intercept -2.702 0.738 < 0.001 Female -0.916 0.438 0.036 Visual Acuity Ratio -1.654 0.733 0.024 Curve 3—No RPMs 2.710 0.612 < 0.001 Curve 2—No RPMs 3.796 0.550 < 0.001 Obscured RPMs 1.782 0.752 0.018 Laps 2 and 4 -0.932 0.427 0.029 Intercept-Only Log-Likelihood -130.843 Final Log-Likelihood -91.717 Table 59. Encroachment model results.

Treatment Recognition, Visibility, and Driver Behavior Evaluations 89   Figure 26. Probability of encroachment versus visual acuity ratio for male participants for varying RPM treatments. Figure 27. Probability of encroachment versus visual acuity ratio for female participants for varying RPM treatments.

90 Performance Criteria for Retroreflective Pavement Markers braking behavior due to overall confusion with the course layout as opposed to behavior attrib- utable to the RPMs. Figure 28 suggests that the maximum deceleration force was relatively consistent within each curve, with Curve 3 being the exception. For Curves 3, 5, and 6 during Lap 1 (and Curve 3 during Lap 3), the maximum deceleration force observed tended to increase as the level of retroreflectivity of the RPMs decreased, with the most pronounced difference occurring between the high- retroreflectivity RPM condition and no-RPM condition observed in Curve 3. This observation suggests that due to the inability of drivers to see the curve, harder braking is necessary to navigate the curve (or in some cases to stop and re-orient the vehicle on the course). Surprisingly, the no-RPM condition in Curve 2 did not seem to be associated with any change in deceleration force. NOTE: The horizontal bar represents the mean. Figure 28. Maximum deceleration.

Treatment Recognition, Visibility, and Driver Behavior Evaluations 91   4.5.3 Distance from the PC at Maximum Brake Application The second performance measure considered in this study was the distance from the PC when the brake pedal was applied (depressed) the most (which is not necessarily where the highest deceleration force occurred). This performance metric was intended to capture the point at which study participants reacted most strongly to curve conditions. Positive values indicate that the maximum brake depression occurred prior to the driver passing by the PC, while nega- tive values indicate that the driver was already in the curve when maximum brake application occurred. These distances were aggregated by treatment and are summarized in Figure 29. Figure 29 indicates that the maximum brake depression occurred closer to the PC when no RPMs were present as opposed to when any RPMs were present. Surprisingly, maximum brake depression typically occurred farthest from the PC when the reflective material of the RPM was obstructed with masking tape, particularly during Laps 2 and 4. Exclusive of that notable excep- tion, the distance from the PC at which maximum brake depression occurred tended to decrease as the retroreflectivity of the marker increased. Figure 30 presents the same data for each treat- ment type in each curve individually. Figure 30 illustrates that in Curve 3, maximum brake application generally occurred farther from the PC when high-retroreflectivity RPMs were present. This figure is likely capturing drivers not being able to see Curve 3 when the RPMs were not present and subsequently braking hard at a later point to compensate and navigate the curve. This action does not necessarily appear to be a trend with other curves. In Curve 5 particularly, where a vehicle was positioned in the oncoming lane to replicate headlight glare from oncoming traffic, the obscured RPMs tended to be associated with brake depression occurring farther from the PC. One possible explanation is that the presence of oncoming headlight glare in the absence of visual cues on the roadway caused drivers to behave more cautiously and brake earlier than they otherwise would have. Given the possibility that brake depression may vary by more parameters than can be accounted for via graphical analysis, GLMs were estimated to predict the distance from the end of the curve at which maximum brake depression occurred as a function of curve geometry, RPM retro- reflectivity level, and vehicle operational characteristics. Driver characteristics were also examined but not found to play a significant role with respect to this particular performance metric. This included investigating potential interactions between participants of various ages with RPMs of specific retroreflectivity levels. In order to capture possible differences attributable to the drivers learning the course and modifying their behavior, two models were estimated: one considering data from all laps and one considering data from only the first lap. The results of the regression analysis can be seen in Table 60. The models were parametrized such that the dependent variable is the distance from the end of the curve (PT), as opposed to the PC, due to a requirement of gamma-distributed data taking only values greater than 0. To account for this, an offset term was included in the model—the natural log of the distance between the start of the treatment and the end of the curve—effectively shifting the model to be based on the start of the curve. The log link used to estimate the regression equation yields coefficients for binary variables that can be interpreted as having an effect of (eβ−1) p 100% on the dependent variable and co efficients for the continuous variables that can be interpreted such that for a 1-unit change in x, a (eβ−1) p 100% change in y occurs. The results of the regression models were relatively consistent between the full data set and the Lap-1-only model. An exception was the central angle (deflection angle) of the curve not being a significant parameter in the Lap-1-only model, and the indicator for Laps 2 and 4, which obviously could not be included in the Lap-1-only model.

92 Performance Criteria for Retroreflective Pavement Markers Figure 29. Distance to PC at maximum brake depression by RPM type for Curve 3.

Treatment Recognition, Visibility, and Driver Behavior Evaluations 93   Figure 30. Distance to PC at maximum brake depression by curve. The effects of the RPMs of varying retroreflectivity levels were compared to the effects of the high-retroreflectivity RPMs. When RPMs were not present, maximum brake depression tended to occur closer to the PT than when they were present for RPMs of high-, medium-, or low- retroreflectivity levels. The absence of RPMs was found to result in maximum brake application occurring 22.5% closer to the PT than when RPMs were on the roadway. Glare due to the presence of oncoming headlights when the RPMs were obscured was found to result in a 37.5% decrease in the distance between the PT and the location of maximum brake application. When glare was present at the same time as RPMs with medium-level retroreflectivity, a 25.3% decrease in distance was found. Curve 5 was the only location with obscured RPMs, which were intended to mimic the effect of no RPMs being present or RPMs with zero retro- reflectivity. Curve 5 was also the only curve with headlight glare from an opposing vehicle. It is

94 Performance Criteria for Retroreflective Pavement Markers reasonable to conclude that RPMs help mitigate the effect of glare from oncoming headlights; however, it is possible that the nonretroreflective RPMs played some role in the drivers’ braking behavior. For each mile per hour that the vehicle was traveling at the time of maximum brake depres- sion, the distance to the PT increased by 1.2%. This indicates that drivers in the study attempted to brake farther from the curve when traveling at higher speeds, thus giving themselves a greater distance to slow down prior to entering the curve. As the central angle of the curve increased, the distance from the PT at which maximum brake depression occurred decreased. The coefficient associated with the central angle indicates that a 1-degree increase in the central angle will result in maximum brake application occurring 0.2% closer to the PT. This observation suggests that the drivers in the study may have had difficulty in assessing the sharpness of the curve, causing them to brake harder as they approached and/or traveled through the curve. Curve radius was also examined but was not found to be a significant predictor of this performance measure. Finally, the natural log of the treatment length (from the first retroreflective treatment RPM to the PT) was used in the models as an offset variable (forcing the coefficient to be 1), resulting in a 1% change in the treatment length corresponding with a 1% change in the distance from the PT at maximum brake application. This offset term effectively accounts for the maximum brake application occurring farther from the PT in cases where the curve is longer. Figure 31 presents the relationship between vehicle speed at maximum brake application and the distance from the PC at maximum brake application for each of the centerline delineation scenarios that were examined while holding the other explanatory factors at their average value. While the statistical model itself was estimated in terms of the distance from the PT, the average curve length was subtracted from the model output to express the value in terms of distance from the PC, which is a more intuitive value. 4.5.4 Speed Differential in the Curve The speed differential in the curve was calculated by subtracting the minimum speed in the curve (between the PC and PT) from the speed observed at the PC. This performance metric was Parameter Full Model Lap 1 Model Estimate (β) Std. Error p-value Estimate (β) Std. Error p-value Intercept -0.297 0.098 < 0.001 -0.405 0.209 0.054 Central Angle -0.002 0.000 0.056 – – – No RPMs -0.255 0.044 < 0.001 -0.467 0.101 < 0.001 Glare and Medium RPM -0.292 0.044 < 0.001 -0.309 0.088 0.001 Glare and Obscured RPM -0.469 0.052 < 0.001 -0.651 0.118 < 0.001 Low-Retro RPMs -0.125 0.042 < 0.001 -0.266 0.095 0.005 Medium-Retro RPMs -0.114 0.042 < 0.001 -0.213 0.092 0.022 Speed (mph) 0.012 0.003 < 0.001 0.016 0.006 0.006 20 ft RPM Spacing -0.191 0.031 < 0.001 -0.161 0.066 0.016 Laps 2 and 4 0.039 0.021 0.053 – – – Ln (Treatment Length) 1.000 NA NA 1.00 NA NA Dispersion Parameter 0.085 – – 0.103 – – Null Log-Likelihood -5149.363 – – -1279.639 – – Full Log-Likelihood -5009.836 – – -1223.762 – – NOTE: – indicates no estimate for a parameter in a specific model. Table 60. Distance from PT at maximum brake depression.

Treatment Recognition, Visibility, and Driver Behavior Evaluations 95   selected using the thought process that larger speed discrepancies in the curve indicate difficulty for the driver to see the curve and therefore determine an appropriate speed at which to enter. The speed differentials observed during this study are documented in Figure 32. Figure 32 suggests that as retroreflectivity decreases, minor changes in driver speed differen- tial can be observed. The most pronounced differences were in Curve 3; however, the y-axis was truncated to make the minor changes in other curves more perceptible. To further investigate this point, as well as other variables that could be affecting driver performance, two linear regres- sion models were estimated: a model considering all data and a model considering only Lap 1. The model results are shown in Table 61. In order to use a log-link function when estimating a GLM, all observations of the response variable must be positive. Subsequently, 0.001 was added to every value for speed decrease such that the minimum value was 0.001. The log link used to estimate the regression equation yields coefficients for binary variables that can be interpreted as having an effect of (eβ−1) p 100% on the dependent variable and coefficients for the continuous variables that can be interpreted such that for a 1-unit change in x, a (eβ−1) p 100% change in y occurs. As vehicle speed at the PC increased, the decrease in speed through the curve tended to increase. This result is relatively intuitive since it should be expected that vehicles entering a curve at higher speeds may not have clearly seen the curve or been able to properly assess the degree of curvature. Vehicle speed at the PC was found to have an impact on the speed differen- tial in both the full data and Lap-1-only models. With regard to curve geometry, larger curves tended to be associated with lower speed changes. Again, this finding is intuitive since the forces acting on the vehicle while navigating a curve become less substantial at a given speed as the curve radius increases, leading drivers to feel more comfortable and subsequently not reduce speed in the curve itself. Somewhat surprisingly, curve radius was not found to play a significant role in the speed differential in the model that focused solely on Lap 1. The effects of the RPMs of varying retroreflectivity levels were compared to the effects of the high-retroreflectivity RPMs. The two models suggest a step-wise progression of increasing speed differential as the retroreflectivity of the RPM decreased, with the most pronounced speed dif- ferential occurring when no RPMs were present, specifically in the first lap model. This detail Figure 31. Distance from PC at maximum brake application versus vehicle speed at maximum brake application.

96 Performance Criteria for Retroreflective Pavement Markers Figure 32. Speed differential in curve. again suggests learning behavior among drivers as they gain familiarity with the course. This observation agrees with what was illustrated in Figure 32. The spacing of the RPMs was found to play a role in speed differential in the overall model, but not in the Lap-1-only model. When the RPMs were spaced every 20 ft in Curve 4 (as opposed to every 40 ft in the other curves), the speed differential in the curve tended to be larger by nearly 1 mph. This may seem counterintuitive since drivers should be able to see this curve more clearly. The closeness of the RPMs may be indicating to drivers that the curve is sharper than they would otherwise think, causing them to decelerate more (and farther into the curve) than they would if the RPMs were spaced every 40 ft. This finding should be investigated to a greater extent in further research using a larger variety of RPM spacings. Driver age was shown to play a role in speed differential through the curve, with younger drivers (28 or younger) tending to decelerate less in the curve. Initial modeling efforts looked at older drivers as well; however, the inclusion of visual acuity in the model (drivers with better visual

Treatment Recognition, Visibility, and Driver Behavior Evaluations 97   acuity tended to brake less in the curve, indicating that they likely saw the curve from farther away) was strongly correlated and a better predictor. Finally, the learning behavior of drivers through multiple laps in the course can be seen by using indicator variables for each lap. The coefficient for Laps 2 through 4 is negative, indicating that drivers did not change speed as much in the curves after the initial lap. The coefficient for Lap 3 is smaller than Lap 2, indicating that the period between Laps 2 and 3 where the drivers went to participate in the recognition study may have caused the drivers to forget some details of the course. The relationship between curve radius and speed loss in the curve is illustrated in Figure 33. The plot was made assuming a driver over the age of 28 with average values for all other continuous variables. 4.5.5 Lateral Position Change The final performance metric considered in the evaluation of RPM impact on driver behavior in curves was the change in vehicle lateral position in the curve, specifically between the PC and PT. To determine this value, the offset from the route centerline at the PC was compared to the offset throughout the remainder of the curve until the maximum difference was identified. The changes in lateral position are illustrated in Figure 34. The y-axis (reflecting lateral position change) has been truncated in Figure 34 due to the generally limited change. In Curve 3, however, there was a huge change in lateral position asso- ciated with no RPMs being present. This again suggests that without RPMs, drivers are unable to properly see the curve. One counterintuitive observation based on the figure is that in Curve 2 during the first lap, the lane position change decreased with retroreflectivity level, including when no RPMs were present. This could be due to other visual cues present in Curve 2 that were not present in Curve 3. Parameter Full Model Lap 1 Model Estimate (β) Std. Error p-value Estimate (β) Std. Error p-value Intercept 4.085 0.853 < .001 4.557 0.947 < .001 Speed at PC (mph) 0.066 0.011 < .001 0.070 0.014 < .001 Curve Radius (ft) -0.879 0.138 < .001 -0.928 0.170 < .001 No RPMs 1.223 0.185 < .001 1.393 0.152 < .001 Obscured RPMs 0.550 0.211 0.009 0.552 0.212 0.010 Low-Retro RPMs 0.437 0.173 0.012 – – – Medium-Retro RPMs 0.443 0.169 0.009 – – – 20 ft RPM Spacing 0.237 0.117 0.043 0.341 0.148 0.022 Driver Age 28 or Younger -0.259 0.113 0.022 -0.506 0.144 0.001 Visual Acuity Ratio -0.265 0.132 0.045 -0.338 0.170 0.048 Lap 2 -0.425 0.116 < .001 – – – Lap 3 -0.293 0.116 0.012 – – – Lap 4 -0.639 0.116 < .001 – – – Dispersion 1.323 0.545 Intercept-Only Log- Likelihood -1610.362 -481.95 Fully Specified Log- Likelihood -1494.024 -423.45 NOTE: – indicates parameter not included in Lap 1 Model. Table 61. Speed differential model results.

98 Performance Criteria for Retroreflective Pavement Markers Once again, two GLMs were estimated to examine a variety of factors that could be affect- ing driver performance relative to the presence and condition of RPMs. One model uses all of the available data, and one model considers only Laps 2, 3, and 4. The results of this regression analysis are shown in Table 62. Attempts were initially made to estimate a stand-alone model for Lap 1 as well; however, the model was largely uninformative. The log link used to estimate the regression equation yields coefficients for binary variables that can be interpreted as having an effect of (eβ−1) p 100% on the dependent variable and coef- ficients for the continuous variables that can be interpreted such that for a 1-unit change in x, a (eβ−1) p 100% change in y occurs. Both the full data model and the Laps 2, 3, and 4 models indi- cate that the magnitude of lateral position change tended to decrease with curve radius (which was transformed via natural log for the purpose of model estimation). This is likely attributable to decreased centrifugal force acting on the vehicle as it maneuvers through curves of higher radii (assuming that speed remains constant). In the full data model, due to the magnitude of change of lateral position associated with Curve 3 when no RPMs were present, the no-RPM condition in Curve 3 was captured sepa- rately from the no-RPM condition in Curve 2 (the only other curve where no RPMs was a treatment). In Curve 2, a much less pronounced variation in lateral position was found. The relatively large p-value associated with this parameter indicates that there is weak evidence to suggest that the no-RPM condition played much of a role in lateral position change in Curve 2. However, when considering the strong evidence that lateral position change increased due to this condition in Curve 3 in conjunction with the observation that the obscured RPM condi- tion also tended to increase the magnitude of the lateral position change when obscured RPMs were present, it is likely that other visual cues present in Curve 2 offset the impact. Conversely, the presence of high-retroreflectivity RPMs was shown to decrease the magnitude of the lateral position change that occurred as a driver traveled through a curve. In addition to the retroreflectivity of the RPMs, the spacing was found to play a role in lateral position change. When spaced at 20 ft, the magnitude of the change in lateral position decreased Figure 33. Speed loss in curve versus curve radius.

Treatment Recognition, Visibility, and Driver Behavior Evaluations 99   by 27%. This indicates that drivers maintain a more consistent path when the RPMs are spaced closer together. The presence of glare from oncoming headlights resulted in improved lateral position mainte- nance. While this may seem counterintuitive at first, the headlights provided study participants with an idea of the general location of the study roadway. This observation needs to be con- sidered with caution because in a real-world driving scenario, both vehicles would be moving (instead of just the study vehicle) and both drivers would likely be basing their position in the lane relative to each other. Two participant characteristics appear to impact the amount of change in lane position through curves: visual acuity and gender. Drivers with better visual acuity tended to maintain more consistent paths through the curves, as did female drivers. It is expected that drivers with Figure 34. Lateral position change in curve.

100 Performance Criteria for Retroreflective Pavement Markers better vision would be better able to see, and thus navigate, the curve more consistently. One possible explanation for the better performance of females is that they are generally more risk- averse than male drivers and may have driven the course more cautiously. Overall, lane position variation tended to be less in Laps 2 through 4. While this result poten- tially represents some learning behavior of drivers, it can also be seen that the parameter estimates for Laps 2 and 4 are approximately equal (and were subsequently combined into one parameter for the Laps 2, 3, and 4 model). This indicates that lane variation behavior is less varied when drivers navigate left turns as opposed to right turns. Figure 35 and Figure 36 present the model output graphically controlling for curve radius. Each line represents one of the centerline delineation scenarios that were examined in the study. Figure 34 explicitly looks at the model assuming 20/20 vision, while Figure 36 assumes 20/40 vision. Each plot was created assuming a female driver; the values for a male driver would be slightly higher by comparison. Ultimately, these figures reinforce the observation that driv- ers had more difficulty navigating Curve 3 when no RPMs were present compared to the other delineation scenarios. 4.5.6 Curve Study Summary Collectively, the results from the study of the effects of RPMs on driver behavior through curves indicate that RPMs provide significant guidance to road users, particularly in conditions where other road markings and visual cues may be obscured, such as when the roadway is wet or when pavement markings are of poor quality or condition. Evidence of the effectiveness of RPMs is consistent across several performance metrics. Key aspects of the analysis of those per- formance metrics are outlined below. • Curves with no RPMs present were more likely to experience centerline encroachments. Nearly 65% of drivers encroached the centerline in Curve 3 on their first time through the course when no RPMs were present compared to 0% of drivers encroaching the centerline Parameter Full Model Laps 2, 3, and 4 Model Std. Error p-value Estimate (β) Std. Error p-value Intercept 3.170 0.658 < 0.001 2.663 0.525 < 0.001 Ln (Curve Radius) -0.416 0.112 < 0.001 -0.384 0.089 < 0.001 Curve 3—No RPMs 1.274 0.113 < 0.001 – – – Curve 2—No RPMs 0.155 0.118 0.190 – – – No RPMs – – – 0.241 0.079 0.002 Obscured RPMs 0.254 0.133 0.058 0.228 0.134 0.090 High-Retro RPMs -0.197 0.113 0.080 -0.180 0.109 0.101 20 ft RPM Spacing -0.304 0.080 < 0.001 -0.268 0.080 0.001 Headlight Glare -0.374 0.104 < 0.001 -0.357 0.106 0.001 Visual Acuity Ratio -0.376 0.085 < 0.001 -0.232 0.084 0.006 Female -0.111 0.055 0.043 – – – Lap 2 -0.404 0.077 < 0.001 – – – Lap 3 -0.201 0.077 0.009 – – – Lap 4 -0.432 0.077 < 0.001 – – – Laps 2 and 4 – – – -0.252 0.058 < 0.001 Dispersion 0.586 – – 0.443 – – Intercept-Only Log- likelihood -963.717 -453.731 Final Log-likelihood -686.889 -371.268 (–) indicates parameter not estimated in the model. Estimate (β) Table 62. Lateral position change model results.

Treatment Recognition, Visibility, and Driver Behavior Evaluations 101   Figure 35. Absolute lateral position change versus curve radius for females with 20/20 vision. Figure 36. Absolute lateral position change versus curve radius for females with 20/40 vision.

102 Performance Criteria for Retroreflective Pavement Markers in Curve 3 on their first time through the course when the high-retroreflectivity RPMs were present. • Visual analysis indicates that the maximum braking force on Curve 3 was substantially higher when no RPMs were present in contrast to high-retroreflectivity RPMs due to hard braking by drivers unable to successfully navigate the curve. • When no RPMs were present, maximum brake application occurred 23% to 37% closer to the PT of the curve than when high-retroreflectivity RPMs were present. This indicates maximum brake application occurred later when no RPMs were present compared to when the high- retroreflectivity RPMs were present. • The distance between the PT and the vehicle at the point of maximum brake application decreased by 37% in the full data model and 48% in the Lap 1 model when glare from oncom- ing headlights was present with obscured RPMs. These changes were reduced to 25% and 27% when medium-level RPMs were present. This indicates that RPMs still provide some level of guidance to motorists in conditions that typically reduce the visibility of pavement markings and other visual navigation cues. • Closely spaced (20 ft) RPMs were found to decrease the distance between the vehicle and the PT at the maximum brake application by 15%–17%. This is potentially due to tight spacing conveying the sense of a sharper curve, regardless of whether that is actually the case. • The difference between the speed at the PC of the curve and the minimum speed of the vehicle in the curve increased as the retroreflectivity of the RPMs in the curve decreased, indicating that higher retroreflectivity RPMs improve a driver’s ability to maintain consistent speed in a curve. • Lower visual acuity tended to be associated with larger speed decreases. • Drivers tended to have more variable lane position as the retroreflectivity of the RPMs decreased, with the most appreciable increase in variability at locations with no RPMs. • Lateral position tended to increase as visual acuity decreased. • Coefficients denoting each lap through the course for the speed differential and lateral posi- tion models indicate that drivers altered their behavior after the first lap, subsequently main- taining more consistent speed and lateral position during the later laps. Additionally, driver performance tended to differ slightly between making right turns (Laps 1 and 3) and left turns (Laps 2 and 4), with drivers operating the vehicle more consistently when making left turns. Ultimately, these findings provide valuable insight into the effects of RPMs in curves when other visual cues are limited. A key gap that still needs to be bridged is relating the retroreflectiv- ity of the RPMs to their age, thereby allowing for the development of a maintenance schedule. Ideally, this endeavor would consider a more varied spectrum of RPM retroreflectivity to more clearly outline the relationship between retroreflectivity and driver performance. 4.6 Closed-Course Limitations This portion of the study has several limitations that should be discussed in order to properly contextualize the results. The first important aspect of the closed-course study that bears discus- sion is the fact that a runway was used to simulate highway pavement. There are several differ- ences between highway pavement and runway pavement that could influence the outcomes of this portion of the study: • The runway is designed to be flat; therefore, the curves do not have superelevation. • The simulated roadways are not demarcated by a definite edge of pavement, shoulder, curb, or other roadside treatment/environment. • The pavement seams present on the runway are not always consistent with typical roadway seams. In addition to those differences between a road and the runway, several other limitations of the study exist. First, the faded pavement markings on the runway at the RELLIS facility are

Treatment Recognition, Visibility, and Driver Behavior Evaluations 103   intended to simulate the relative visibility of wet markings; however, no water was used to mask the markings. The visibility of the markings was as poor as would be the case in wet conditions, but the road surface was dry, and rain was not falling on the windshield. Second, the study involved a relatively small sample of drivers in a controlled environment. Perhaps the most chal- lenging issue in analyzing these data was due to variation in the geometric layout of the curves, which varied in radius from 200 ft to 500 ft and in central angle from 40 degrees to 90 degrees. Additionally, some treatments were only used in one specific curve, which leaves the possibility that conditions at a given curve that were unaccounted for in the data could potentially be biasing the supposed effectiveness of the RPMs. The treatment issue is particularly problematic when considering Curve 3, which was treated with high-retroreflectivity markers as well as no markers but also had the fewest visual cues aside from the pavement markings of any of the treatment locations. From the participants’ perspective, while Curve 3 appeared to be a relatively wide- open concrete area, Curve 2 had buildings in the horizon and vegetation delineating the edge of the pavement surface, which may have elicited more consistent driver behavior.