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.

12 C H A P T E R 2 2.1 Safety-Related Studies Although RPMs have been researched regularly for decades, genuine safety evaluations of the potential countermeasures are relatively scarce, especially considering that most modern guide- lines seem primarily based upon NCHRP Report 518 (Bahar et al. 2004). That report collected literature regarding past studies, commented upon the quality of those studies, and presented a more conclusive evaluation of RPMs. Since the publication of NCHRP Report 518, few studies have presented novel evaluations of the safety benefits of the various forms of RPMs. Instead, most extant literature presents research on the durability and retroreflectivity of different markers, the capabilities of RPMs to improve delineation and affect driver speed and behavior, or the applications of RPMs within pedestrian accommodations. The literature review within this report presents the results of those earlier studies with other historic evaluations as well as post- NCHRP Report 518 evaluations in order to construct a more comprehensive sense of how RPMs affect safety, especially at horizontal curves on roadways. This safety-related literature review is divided into two subsections. First, unique evaluations of the safety effectiveness of different types of RPMs are presented. Although NCHRP Report 518 cataloged the most prominent safety evaluations of RPMs within the literature, a handful of other contemporary studies have also been conducted and are reviewed. The second section reviews research projects that looked at RPM evaluations that used surrogate measures of safety. It is important to keep in mind that the methodologies used to design a study can greatly impact the applicability of the results. Isolating the effects of the treatment, in this case RPMs, is often difficult. A key limitation of before-and-after studies is that multiple treatments may be applied at once. RPMs are typically applied when a road is both repaved and restriped, resulting in a combined effect that can mask the specific safety benefits of RPMs. 2.1.1 Safety Evaluations of RPMs RPMs are traditionally considered to improve roadway safety by improving the preview dis- tance of roadway features and by improving visibility in bad weather and low light (Bahar et al. 2004). To that end, RPMs are typically installed on the centerline and left edge of pavement, par- ticularly around horizontal curves and in addition to other measures like retroreflective mark- ings (Zhao et al. 2015). NCHRP Report 518 (Bahar et al. 2004) is among the most frequently referenced literature regarding the effectiveness of RPMs. Guidance from that report is included in the current edition of the Highway Safety Manual and has influenced the installation stan- dards of some state departments of transportation (DOTs). The results of NCHRP Report 518 are not without controversy, though. Some agencies feel that the limitations of the study negatively influence the results. Literature Review

Literature Review 13   NCHRP Report 518 summarizes the results of seven noteworthy evaluations of the safety effectiveness of RPMs; those seven studies include the following: • Evaluation of the effect of RPMs on nighttime crashes in Georgia (Wright et al. 1982) • Before-and-after analysis of daytime and nighttime crashes in Texas (Kugle, Pendleton, and Von Tress 1984) • Reexamination of the data used by Kugle, Pendleton, and Von Tress (1984) through the use of cross-product and logit severity analysis (Mak, Chira-Chavala, and Griffin 1987) • Reexamination of the data used by Kugle, Pendleton, and Von Tress (1984) through the use of weighted log odds ratios for night crashes (Griffin 1990) • Empirical Bayes analysis of nighttime crashes in Michigan (Pendleton 1996) • Two before-and-after studies of nighttime and wet-weather crashes in New York (NYSDOT 1989; NYSDOT 1997) • Evaluation of nighttime crashes in Pennsylvania using the odds ratio methodology from Griffin 1990 (Orth-Rodgers and Associates Inc. 1998) These studies were long considered a source of confusion regarding the efficacy of RPMs for crash reduction due to their often conflicting results. Wright et al. (1982) found reductions in nighttime crashes for sites modified in 1976 and 1977 but an increase in nighttime crashes for sites modified in 1978. Kugle, Pendleton, and Von Tress (1984) reported a significant increase in nighttime crashes after the application of RPMs but a nonsignificant decrease in wet-weather crashes, noting that sites experiencing an increase in crashes were in the minority; however, the crash increases that occurred were large. Mak, Chira-Chavala, and Griffin (1987) also found conflicting results, with four sites showing a significant reduction in nighttime crashes and nine sites showing a significant increase compared to daytime crashes. The previous two studies used the cross-product approach to compare nighttime to daytime crashes. Under the cross-product analysis framework, if the number of nighttime crashes decreases at a lower rate than the daytime crashes, the nighttime crashes are said to be increasing; this may undervalue night- time crash reductions, especially in the case of RPMs, which may have some daytime benefits due to their audible/tactile properties when driven over. Analyzing the same data set as Mak, Chira-Chavala, and Griffin (1987)—which was based on the data from Kugle, Pendleton, and Von Tress (1984)—Griffin (1990) estimated a 16.8% increase in nighttime crashes following the instal- lation of RPMs. Pendleton’s (1996) analysis found an increase in nighttime crashes on undivided roadways and a decrease in nighttime crashes on divided roadways after installation of RPMs, but the use of daytime crashes at treated sites as a comparison group cast doubt on the results. NYSDOT (1989) analysis, through a simple before-and-after study of 20 unlit suburban and rural sites where RPMs had been applied, found a significant decrease in nighttime crashes and night- time wet-weather crashes with nonsignificant decreases in total crashes. NYSDOT (1997) study, a before-and-after study of RPMs on 50 long-highway sections, found a significant decrease in total crashes, a nonsignificant decrease in nighttime crashes, and a nonsignificant increase in night- time wet-weather crashes. Last, Orth-Rodgers and Associates Inc. (1998) found slight decreases in nighttime crashes for sites with raised RPMs and significant increases for sites with recessed RPMs. To account for these conflicting results, Bahar et al. (2004) examined the research method- ology of these studies closely and found several conflicting factors and potential sources of errors that may have informed these discrepancies. The results for at least some of these studies may have been confounded by the following factors: • Changes in traffic volumes—Most of these studies assumed constant traffic volumes before and after the treatment was enacted—except perhaps for Pendleton’s (1996) Empirical Bayes analysis—and, perhaps as problematically, assumed those volumes would remain in constant ratio between the day and nighttime crashes (Wright et al. 1982; Kugle, Pendleton, and Von Tress 1984; Mak, Chira-Chavala, and Griffin 1987; Griffin 1990; Orth-Rodgers and Associ- ates Inc. 1998). Crashes have a strong relationship to traffic volume, though not necessarily

14 Performance Criteria for Retroreflective Pavement Markers a linear one, and it is erroneous to assume that volume will be constant before and after the installation of some treatment. • Time trends—In addition to traffic volumes, safety is also linked to other socioeconomic factors, such as technology and demographics. These socioeconomic factors may change over time, so care should be taken to avoid any external influences on the effects of RPMs. However, it is unclear whether any of these studies accounted for a wide range of potential confounding factors, and even using daytime crashes as a comparison could be problematic if some factor causes daytime crash frequencies to change. • Regression to the mean—Because crashes, though linked to traffic volume and other socio- economic factors, are random events, sites may experience uncharacteristically high or low frequencies in specific years; if sites are selected for treatment based on these odd peaks, the magnitude of a treatment’s effect on safety may be miscalculated due to the normalizing nature of safety over time. Only Pendleton’s (1996) study accounted for regression to the mean through the use of an Empirical Bayes methodology. Therefore, the true magnitude of the effects of RPM treatments in the other studies cannot be accurately described. • Multiple treatments—None of the reported studies accounted for other treatments that may have been installed in the study periods, so it is difficult to conclude with accuracy whether any crash reductions or increases were caused solely by RPMs. Often, RPMs are installed or replaced at the same time as resurfacing, and not accounting for this can lead to biased results, especially because resurfacing itself has been found to have a significant impact on crashes (Hauer, Terry, and Griffith 1994). Even NCHRP Report 518 (Bahar et al. 2004) does not expli c itly discuss if other changes (e.g., resurfacing) occurred at the same time as the installation of RPMs. The ability to isolate the effects of RPMs on safety from other factors has limited most previous safety studies. • Selection of comparison group—Most of the studies used daytime crashes as the comparison group for nighttime crashes. This selection, however, is flawed because it assumes that RPMs will not also affect daytime crashes. Orth-Rogers and Associates Inc. (1998) identified that the selection of daytime crashes for comparison was problematic and that any treatment like RPMs has the potential to migrate crashes to different locations, perhaps because the increased delineation encourages changes in speed or driver behavior at comparison sites near the treat- ment sites. Other studies, such as that conducted by Jiang (2006), reported changes in daytime operations when RPMs were present. If comparison entities and crash types are not specially selected, the true effect of countermeasures cannot be isolated (Bahar et al. 2004). Bahar et al. (2004) summarized an evaluation of the seven aforementioned studies in a table, reproduced as Table 1. Note that the site types and installation locations varied between studies. Wright et al. (1982) investigated centerline RPMs on horizontal curves on two-lane highways. Kugle, Pendleton, and Von Tress (1984) examined RPMs on two-lane and multilane roadways in different environments, as did Mak, Chira-Chavala, and Griffin (1987) and Griffin (1990). Pendleton (1996) examined divided and undivided arterials, focusing primarily on centerlines and lane lines. NYSDOT (NYSDOT 1989; NYSDOT 1997) examined suburban and rural road- ways, and Orth-Rogers and Associates (1998) examined rural, interstate highways. To account for the myriad problems outlined above and to offer sound guidance on when and where to apply RPMs to reduce crashes, Bahar et al. (2004) conducted their own evaluation study for NCHRP Report 518. The research team surveyed 29 different states to gather information about applications of RPMs, and the following types of data were collected: • Inventory of sites where RPMs had been applied • Geographic locations of both the reference group and comparison group • Historical crash data • Roadway attributes • Traffic volumes • Other delineation and guidance treatments

Table 1. Summary of RPM study information from Bahar et al. (2004). Study Reference and Location Site Type Installation Location Time Period Sample Sizes for Treatment and Comparison Groups Dependent Variable Independent Variables Analyzed Comparison Group Other Notes Estimated Effects Wright et al. (1982) Georgia Horizontal curves on two- lane highways in excess of 6 degrees of curvature. Centerline. I—1976–1978 B—1 to 3 years A—2 to 4 years Treatment—662 locations. Comparison—same as treatment group. Total nighttime crashes. Average daily traffic (ADT), degree of curvature. Total daytime crashes. Both raised and recessed reflective markers were used; at some locations warning signs, chevron markers, or other guidance devices were installed. 22% reduction in nighttime crashes; single-vehicle crashes reduced 12% more than other nighttime crashes; reductions independent of ADT or horizontal curvature for curves with degree of curve greater than 6. Kugle, Pendleton, and Von Tress (1984) Texas Two-, three-, four-, five-, and six- lane roadways. Does not specify. I—1977–1979 B—2 years A—2 years Treatment—452 locations. Comparison—same as treatment group. Total nighttime crashes, some analysis by crash and severity. ADT, number of lanes, number of wet-weather days. Total daytime crashes. None 15% to 31% increase in nighttime crashes; no significant effect on wet-weather crashes. Mak, Chira- Chavala, and Griffin (1987) Texas Two-, three-, four-, five-, and six- lane roadways. Does not specify. I—1977–1979 B—2 years A—2 years Treatment—87 locations. Comparison—same as treatment group. Total nighttime crashes, some analysis by crash and severity types. Intersection type, within/outside city, horizontal curvature, grade, structures, number of lanes, divided/ undivided. Total daytime crashes. Used a subset of the data from Kugle, Pendleton, and Von Tress (1984). 4.6% of locations showed significant reductions; 10.3% showed significant increases; 85.1% showed nonsignificant effects. Griffin (1990) Texas Two-, three-, four-, five-, and six- lane roadways. Does not specify. I—1977–1979 B—2 years A—2 years Treatment—86 locations. Comparison—same as treatment group. Total nighttime crashes. None. Total daytime crashes. Used a subset of the data from Kugle, Pendleton, and Von Tress (1984). 16.8% increase in nighttime crashes, with the 95% confidence interval between a 6.4% and 28.3% increase. (continued on next page)

I = installation period; B = before-period length; A = after-period length. Study Reference and Location Site Type Installation Location Time Period Sample Sizes for Treatment and Comparison Groups Dependent Variable Independent Variables Analyzed Comparison Group Other Notes Estimated Effects Pendleton (1996) Michigan Divided and undivided arterials. Centerline on undivided arterials, lane lines on divided arterials. I—1989 B—2 years A—2 years Treatment—17 locations totaling 56.11 mi (90.3 km). Comparison—42 locations totaling 146.28 mi (235 km). Total nighttime crashes. Divided/ undivided and vehicle miles traveled used in Empirical Bayes analysis. Total daytime crashes, total nighttime crashes at comparison sites. None. No significant effect; direction of effect positive or negative dependent on method used and access control. NYSDOT (1989, 1997) New York Suburban and rural roadways. Does not specify. I—unknown B—unknown A—unknown Selective Installation: Treatment—20 locations totaling 26 mi (41.84 km). Comparison—none used. Nonselective Installation: Treatment—50 locations. Comparison—none used. Total crashes, total nighttime crashes. None. None. Regression to the mean is cited as being a factor. 26% decrease in nighttime crashes when placed selectively; no significant effect when installed nonselectively. Orth-Rodgers and Associates Inc. (1998) Pennsylvania Interstate highways in rural nonillumi- nated areas. Does not specify. I—1992–1995 B—1–3 years A—1–3 years Treatment—3,376 locations depending on crash type studied. Comparison—same as treatment group. Total nighttime crashes, nighttime wet road, nighttime wet road sideswipe, or fixed object. None. Total daytime crashes, daytime wet road, daytime wet road sideswipe, or fixed object. Both raised and recessed reflective markers were used. 18.1% overall increase in nighttime crashes; nighttime wet condition crashes increased from 30% to 47%; nighttime wet road sideswipe or fixed object increased by 56.2%. Table 1. (Continued).

Literature Review 17   Ultimately, six states were used for the evaluation: Illinois, Missouri, New Jersey, New York, Pennsylvania, and Wisconsin. The data from Illinois, New Jersey, New York, and Pennsylvania pertained to two-lane roadways, while data from Missouri, New York, Pennsylvania, and Wisconsin pertained to four-lane freeways. Additionally, the Pennsylvania and Wisconsin data sets also contained information for four-lane expressways. Only data from Illinois, New Jersey, New York, and Pennsylvania were used to develop crash modification factors (CMFs) for two-lane roadways, and only data from Missouri, New York, and Pennsylvania were used to develop CMFs for four-lane freeways. Crash data from at least 2 years before implementation and 1 year after implementation were used for the evaluation. Ultimately, Bahar et al. (2004) conducted two sets of analyses to investigate the impacts of RPMs on traffic safety. The first analysis was a composite before-and-after procedure that used Empirical Bayes to overcome the regression-to-the-mean nature of crash modifications. The second analysis used the results of the before-after evaluation to determine the specific safety effects associated with site characteristics such as roadway type, average annual daily traffic (AADT), and horizontal curvature. The second analysis essentially estimated crash modifica- tion functions, with CMFs being dependent variables and site characteristics being independent variables. The CMFs for nighttime application of RPMs for different combinations of roadway type, AADT, and horizontal curvature, are shown in Table 2 and Table 3. Bahar et al. (2004) provided many tables showing the sample sizes (e.g., number of sites and miles by roadway type) for the four different states that were part of the evaluation. However, specific information about the number of miles/sites or expected crashes for the disaggregate results in Table 2 was not available. Bahar et al. did not document the number of miles/sites for the two-lane roadways with AADT between 0 and 5,000 and degree of curvature larger than 3.5. Bahar et al. did try to account for other treatments by including factors for rumble strips, illu- mination, chevrons, and post-mounted delineators (PMDs). Factors such as age/condition of pavement surface, pavement marking presence/quality, and retroreflectivity quality of the RPMs were not considered. With these limitations in mind, the results in both the tables (Table 2 and Table 3) show a consistent trend: RPMs are more effective at night on roads with higher traffic volumes. At lower volumes, installation of RPMs may actually increase crash frequencies. Simi- larly, RPMs were less effective in larger degrees of curvature. The relationship to volume may indicate a reduction in the head-on crashes that naturally increase in probability as traffic vol- umes increase (Bahar et al. 2004). Although NCHRP Report 518 addressed several previous studies, there were several older studies not addressed in the report. In 1982, Zador, Wright, and Karpf conducted a before- and-after analysis of the effects of RPMs on nighttime crashes in Georgia. A statistical test, the Mantel-Haenszel procedure, was used to find whether there was a statistical difference between nighttime crashes and daytime crashes after application of RPMs on the centerline of a hori- zontal curve with a curvature larger than 6 degrees. The researchers concluded that nighttime crashes were reduced for sites modified in 1976 and 1977 but increased for sites modified in 1978. This study mirrors that of Wright et al. (1982) and shares the same limitations as it reports on the project. Facility Type AADT (veh/day) CMF when Degree of Curvature ≤ 3.5 CMF when Degree of Curvature ≥ 3.5 Two-Lane Roadways 0–5,000 1.16 1.43 5,001–15,000 0.99 1.26 15,001–20,000 0.76 1.03 Table 2. Nighttime CMFs for RPMs on two-lane roadways (Bahar et al. 2004).

18 Performance Criteria for Retroreflective Pavement Markers In 1986, Agent and Creasey developed crash reduction factors (CRFs) for various delineation treatments on horizontal curves in Kentucky. Before-and-after crash data were gathered for 1 year before and 1 year after for horizontal curves in Jefferson County and Louisville. Nighttime crashes and daytime crashes were compared, and the researchers concluded that RPMs caused a slight decrease in total crashes, a slight increase in wet surface crashes, and a slight decrease in nighttime crashes after implementation of RPMs. This study did not account for the numerous limitations of this type of analysis, such as regression to the mean and poor comparison site selection, listed previously. Additionally, the sample size for RPM installation was limited to four test sites. In 1996, Agent, Stamatiadis, and Jones produced several CRFs related to RPM installation by surveying different state agencies and reviewing existing literature on RPMs. A total of 43 states were surveyed, 37 of whom use CRFs related to RPMs; these CRFs were synthesized with those found in the literature to develop general recommendations and expected crash reductions. Table 4 is adapted from that report and shows expected crash reductions based on the state survey and the literature review. As is evident in Table 4, there is a lack of consistency between the state surveys and the litera- ture regarding the CRF potential of RPMs. Moreover, these factors are provided by facility type, and it is unclear what methodologies were used to determine these percent reductions. Agent, Stamatiadis, and Jones (1996) further synthesized these results into recommended CRFs for the three listed crash types in Table 4: 10% for total crashes, 20% for nighttime crashes, and 25% for wet-weather nighttime crashes. Corben et al. (1996) conducted before-and-after evaluations of different countermeasures in Victoria, Australia. Two of the sample sites selected were outfitted with RPMs. The before period consisted of 5 years. The duration of the after period was not provided, but the study report men- tions that “after-periods comprised the maximum number of full years of crash data available.” Fatal run-off-the-road collisions with fixed objects were compared on sites without treatment in Facility Type AADT (veh/day) CMF Four-Lane Freeways ≤ 20,000 1.13 20,001–60,000 0.94 > 60,000 0.67 Table 3. Nighttime CMFs for RPMs on four-lane freeways (Bahar et al. 2004). Crash Type Percent Reduction Found in State Survey Percent Reduction Found in Review of Literature Number of References Range of Reduction Average Reduction Number of References Range of Reduction Average Reduction Total 15 4–50 13 7 4–15 6 Wet- Weather Nighttime 7 20–25 21 3 20–46 29 Nighttime 8 10–26 17 4 10–26 18 Table 4. Percent reduction in crash types found by Agent, Stamatiadis, and Jones (1996), adapted.

Literature Review 19   the before-and-after period to sites with the treatment. The study found an 11.6% reduction in crashes after installation of RPMs, but this result was not statistically significant. The reduction from RPMs applied in addition to curve delineation was 6.2%, but this result was not statistically significant either. This study may have been subject to a regression-to-the-mean bias because data were drawn from a black-spot identification program. Ogden (1997) conducted a before-and-after evaluation of the effect of paving shoulders in rural Victoria. The analysis examined two-lane, two-way roads and found that there was a statistically significant reduction in fatal crashes equal to approximately 41% after paved shoulders were installed on two-lane roadways. Although RPMs were not examined explicitly in this study, Ogden noted that repaving efforts on many of these roadways likely included installation of RPMs along the outside pavement markings of these roadways, so the RPMs may have contrib- uted to the reduction of fatal crashes. The magnitude of that effect, however, is unknown. Just 1 year before the publication of NCHRP Report 518, a different NCHRP report detailed two studies that showed benefits of reducing run-off-the-road crashes through the implementa- tion of RPMs. Volume 6 of the NCHRP Report 500 series, A Guide for Addressing Run-Off-Road Collisions (Neuman et al. 2003), includes information regarding RPM evaluations in New Jersey and Ohio. In New Jersey, before-and-after analyses were conducted for several two-lane rural highways using 2 years of before and 1 year of after-crash data; the study showed a statistically significant reduction in nighttime crashes. An Ohio study that included horizontal curves, narrow bridges, stop approaches, and interchanges showed reductions in both total crashes and injuries after a 1-year before-and-after crash evaluation of RPMs. However, the NCHRP report notes that both New Jersey and Ohio studies [and the cited New York analyses (NYSDOT 1989; NYSDOT 1997)] likely did not control for regression to the mean. Volume 6 of NCHRP Report 500 concluded that at the time, the effectiveness of RPMs was questionable and reliable CRFs could not be reported (Neuman et al. 2003). Several other studies have been published since NCHRP Report 518. These studies are detailed below. Persaud and Lyon (2007) published a meta-analysis of before-and-after studies conducted for traffic safety. The authors discussed many of the issues addressed in NCHRP Report 518, particularly the poor selection of comparison groups. Persaud and Lyon concluded that previ- ously reported evaluations of RPMs in Missouri, Pennsylvania, Illinois, and New Jersey may be unreliable due to a scarcity of sound data from untreated sites in the after period. The Delaware Valley Regional Planning Commission (DVRPC 2007) published the results of a road safety audit (RSA) conducted in the state on state highway PA 663. This RSA exam- ined 125 crashes on the highway between 2003 and 2006 using data provided by the Pennsyl- vania Department of Transportation (PennDOT). These crashes were clustered and mapped along the highway corridor, and the audit team discussed potential strategies, based on crash data and other roadway characteristic data, for reducing some of the problems identified. The road safety audit team concluded, based on the results of the RSA, that RPMs should be used to improve delineation throughout the corridor, particularly on the centerline; DVRPC indicated that RPMs could produce a high-potential safety benefit. However, RSAs lack the rigor of statistical analysis methods, so the results may be unreliable; no crash reduction fac- tor was reported. The 2009 edition of the Handbook of Road Safety Measures (Elvik et al. 2009) contains guid- ance related to RPMs. Although the authors reported that there was no significant effect of RPMs used as a sole countermeasure, they indicated that RPMs considered in conjunction with delineator posts might provide crash reduction benefits. Table 5 shows the reported CRFs for

20 Performance Criteria for Retroreflective Pavement Markers RPMs and delineators for different crash types and severities as different CMFs. The CMFs reported in Table 5 were developed through a log odds method of meta-analysis; therefore, the CMFs are subject to the limitations of the literature used in the meta-analysis. These studies are mostly older analyses subject to the limitations reported in NCHRP Report 518 and include a 1988 study by Creasey, Ullman, and Dudek, the 1990 Griffin study, NCHRP Report 518 (Bahar et al. 2004), and possibly other, unspecified studies categorized as “various types of road mark- ings” (Elvik et al. 2009). Carlson, Park, and Andersen’s (2009) meta-analysis of the safety effects of pavement mark- ings included some comments related to RPMs, but these results were difficult to distinguish from more general findings for pavement markings. The authors included regression coeffi- cients for different pavement marking and roadway variables from a negative binomial regres- sion model of Illinois data for rural two-lane roads. These coefficients for edge line width may include effects from RPMs, but the specific effect of RPMs cannot be explained. Similarly, the authors noted that in New Zealand, RPM and other delineation treatments are applied progres- sively on roadways as traffic volume increases, and the effects of both edge lines and centerlines are difficult to parse from the effects of RPMs. The phenomenon of RPMs overpowering other pavement markings has been observed in the United States as well. Based on this meta-analysis, the true effects of RPMs may be difficult to isolate from other treatments. This finding can apply to most of the research that has evaluated RPMs. Pant and Panta (2009) published the results of an evaluation of treatments used to reduce rear-end crashes in Ohio that included a simple before-and-after comparison of the effect of RPMs on freeways. The results were conflicting, showing an increase in crashes after installation of RPMs at one site and a decrease in crashes following resurfacing with new RPMs and thermo- plastic markings at another. These conflicting results may be explained by the small sample size, simplicity of methodology, and lack of accounting for conflicting factors, so no conclusions can be drawn regarding the effectiveness of RPMs. Agent and Green’s 2009 report on the evaluation of the use of snowplowable, raised pavement markers for the Kentucky Transportation Cabinet contained a safety evaluation. The researchers compared crash data on rural two-lane two-way roads that were and were not part of the snow- plowable RPM system roads and had an ADT above 2,500. Researchers found lower crash rates on the snowplowable RPM system compared to other roads. Researchers also found crashes occurring during nighttime and wet nighttime conditions were lower on the snowplowable RPM system roads. Researchers did note that roadways on the snowplowable RPM system typically have better geometrics than roads that were not part of the snowplowable RPM system. Researchers recommended continued use of the snowplowable RPMs if the castings are properly installed and the pavement is maintained. In 2013, FHWA recognized a noteworthy application of RPMs in Alabama (FHWA 2013). The Alabama DOT (ALDOT) worked with Mobile County to identify rural roadways with the Crash Type CMF Crash Severity Best Estimate 95% Confidence Interval All Crashes Unspecified 0.99 (0.97, 1.01) All Crashes Injury 0.97 (0.97, 1.01) Crashes in Darkness Unspecified 0.98 (0.93, 1.04) Crashes in Darkness Injury 0.99 (0.75, 1.29) Daylight Crashes Unspecified 1.01 (0.96, 1.06) Head-on Collisions Unspecified 0.99 (0.67, 1.46) Loss-of-Control Crashes Unspecified 0.97 (0.88, 1.07) Table 5. CMFs associated with RPM treatments (Elvik et al. 2009).

Literature Review 21   highest total of run-off-the-road crashes. Mobile County and ALDOT installed RPMs along the edge line of horizontal curves on the top 10 sites in order to improve delineation and visibility. A simple before-and-after study showed the average number of crashes on these roadways decreasing by 85.3%, but the rigor of the statistical evaluation was not reported. One of the most noteworthy evaluations of RPMs took place in Louisiana on behalf of the Louisiana Department of Transportation and Development (DOTD) in 2013. In the report, Sun and Das (2013) first reviewed the existing RPM literature discussed by Bahar et al. (2004) before presenting their own CMFs for RPM applications developed through multiple statistical analysis methods. In Louisiana, DOTD installs RPMs on all freeways, but evaluations before this study were subjective and lacked conclusive evidence of efficacy. Due to a lack of comprehensive data regarding installation dates, alternate methods were used to derive the CMFs for RPMs. These methods used 9 years of crash data, compared as crash rates, and engineer-designated annual ratings of pavement striping quality. First, Sun and Das compared crash rates on different free- ways by reported quality, determining from statistical t-tests that higher-quality RPMs corre- sponded to lower crash rates. The researchers then performed a “with and without” analysis of RPMs and reported quality, considering “with” roadways to be those with good quality ratings and “without” roadways to be those with poor quality ratings. This methodology showed a crash rate reduction attributable to RPMs for crashes occurring during a 24-hour period and specifically for crashes occurring at night by comparing markers that were subjectively rated as good in comparison to those rated as poor. Based on these reported reductions, Sun and Das reported the CMFs presented in Table 6 for RPM applications on rural freeways under different volume considerations. Based on their analysis, Sun and Das concluded that RPMs reduce crashes on rural freeways under all volume conditions, but the treatment has no benefit for urban freeways. The authors also noted various limitations, including the fact that the statistical methods could not account for other countermeasures that may have been installed during the study years and potentially other differences between roads with and without RPMs. As previously noted, the limitation described by Sun and Das applies to many of the research projects that have evalu- ated RPMs. In the 2016 edition of Local Roadway Safety: A Manual for California’s Local Road Owners, the California DOT (Caltrans) recommends the installation of RPMs in both centerline and edge line markings to reduce total crashes, head-on crashes, and run-off-the-road crashes. The reported CRF of edge line and centerline installation is 25% (with a range of 0%–44%), and an expected life of 10 years (Caltrans 2016). These recommendations were derived from the FHWA CMF Clearinghouse (FHWA 2017), NCHRP reports, the Roadway Departure Safety manual (Golembiewski and Chandler 2011), and the Desktop Reference for Crash Reduction Factors (Bahar et al. 2008). A recent study authored by Park, Abdel-Aty, and Wang (2017) explored the safety effects of pavement resurfacing through a time series analysis. Specifically, a before- and-after approach was applied to crash data on urban arterials in Florida to develop CMFs for pavement resurfacing. The authors noted that the sites where resurfacing was implemented also had RPMs installed, but the two treatments were not analyzed separately, so the effects of RPMs may have confounded the effects of resurfacing. This may be an issue in many before-and-after studies since RPMs are generally installed when pavements are resurfaced. The authors noted that the CMFs increased over time as the pavement resurfacing degraded, but the additional RPMs installed likely improved visibility and safety, particularly in low-light conditions. As the above literature review indicates, many modern studies still feature many of the same methodological problems identified by Bahar et al. (2004), such as not accounting for possible bias due to regression to the mean or traffic volume or lacking generalizable results due to con- founding treatment installations that limit isolated safety effects of the RPMs. These limitations may explain the fact that the efficacy of RPMs is still debated.

22 Performance Criteria for Retroreflective Pavement Markers Recommendations for the use of RPMs may also be drawn from Bahar et al. 2006 study of general retroreflectivity as part of NCHRP Project 17-28. In this study, Bahar et al. conducted both a simulation study using data from the National Transportation Product Evaluation Pro- gram and a time series analysis of 5,000 mi of road segments in California, investigating the safety benefits of different retroreflective treatments. This two-pronged analysis was undertaken to assess the reliability of the time series methodology, particularly when it comes to modeling treatments that may change over time. In fact, one of the key limitations of before-and-after studies is that multiple treatments may be applied at once when a major road rehabilitation is undertaken. In the case of RPMs, this limitation may occur when a road is repaved, restriped, and treated with RPMs, resulting in a combined effect that may mask the true safety benefits of RPMs. The authors determined that the combined study methodology was reliable. They ultimately concluded that retroreflectivity may not be significantly important for delineation, with there being no significant difference in safety between high retroreflectivity and low retro- reflectivity when markers and markings are maintained to California standards. Instead, the authors noted that all that is important for curve detection and safety is that some sort of pave- ment markings are present and visible to drivers, regardless of the quality of retroreflectivity. The authors suggest that the reason for this may be psychological; drivers likely slow down Highway Type Feature Crash Hour Rating N Mean CMF AADT ≤ 20,000 Rural RPM Night Good 291 0.139 0.81 Poor 200 0.172 Rural RPM 24 hr Good 291 0.635 0.91 Poor 200 0.700 Rural RPM+Striping Night Good 225 0.138 0.69 Poor 86 0.201 Rural RPM+Striping 24 hr Good 225 0.644 0.75 Poor 86 0.856 20,000 ≤ AADT≤ 60,000 Rural RPM Night Good 436 0.141 0.79 Poor 382 0.179 Rural RPM 24 hr Good 436 0.596 0.81 Poor 382 0.738 Rural RPM+Striping Night Good 329 0.148 0.76 Poor 165 0.195 Rural RPM+Striping 24 hr Good 329 0.602 0.78 Poor 165 0.770 AADT > 60,000 Rural RPM Night Good 745 0.153 0.86 Poor 596 0.178 Rural RPM 24 hr Good 745 0.655 0.87 Poor 596 0.757 Rural RPM+Striping Night Good 606 0.155 0.78 Poor 285 0.200 Rural RPM+Striping 24 hr Good 606 0.655 0.78 Poor 285 0.841 N = number of sites; Mean = average number of crashes at those sites. Table 6. RPM CMFs developed for Louisiana freeways (Sun and Das 2013).

Literature Review 23   when visibility is low and speed up when visibility is high. This finding echoes a study by Molino et al. (2004) that claimed that yellow centerlines could deteriorate by more than 99% and not experience any reduction in curve detection distance if new RPMs were installed. However, the Bahar et al. 2006 study did have a few limitations. The time series analysis generated an average seasonal effect that did not divide results by climate region, so while cyclic weather may have been captured in the model, specific road conditions may not have been. The specific locations of RPMs on the roadway were not linked well to geometric features, such as curves, and the sample of RPMs itself was limited. Although the study did capture the cyclic effects of changing treat- ments, the authors acknowledged that the study did not focus on the safety effects of RPMs, and the interaction between RPMs and pavement markings is not clearly captured in the model. The nature of the data set resulted in treatments that were maintained to California standards. This resulted in few sections where the visibility of the treatments was very poor, limiting the ability of the analysis to determine a minimum maintained visibility level. These findings, considered in conjunction with those of the other reported studies, indicate that RPMs may indeed provide some safety benefits in conjunction with lane markings, regardless of the level of retroreflectivity, but the true effect is still unknown. 2.1.2 Other Types of RPM Evaluations Although the number of safety evaluations of RPMs is limited, several other studies have been conducted to evaluate other properties of RPMs. These properties include RPMs’ ability to provide good vibratory information and feedback (Kao 1969), RPMs’ use as a delineation device (Blaauw 1985; Zador et al. 1987; Shankar, Mannering, and Barfield 1995; Smiley 2008), and RPMs’ instal- lation requirements (Grant and Bloomfield 1998). Finally, instead of gauging RPM effective- ness in terms of crashes, several studies relied on surrogate measures of safety (i.e., centerline encroachments, lateral position variability, and speed) to assess effectiveness. Several of these studies are detailed in the following paragraphs. Krammes and Tyer (1991) investigated several measures of vehicle operation that they identi- fied in the existing literature at the time as being correlated with crashes in curves on rural, two- lane roads: speed and lateral placement. The data they collected indicated that midpoint speeds in curves after the installation of new RPMs were 1 to 3 mph higher than with the previously installed PMDs, that lateral placement was 1 to 2 ft farther from the centerline, and that fewer vehicles crossed the centerline. Furthermore, the authors reexamined some of the selected sites after 11 weeks and observed similar driver behavior to new RPMs, while after 11 months, speeds were reduced back to PMD levels and lateral position was similar to new RPMs. There are several possible explanations for these findings: the distance at which RPMs were visible decreased with age, resulting in the speeds decreasing, or drivers simply became used to the presence of the RPMs and subsequently reverted to old driving habits. A number of studies have utilized speed and lateral position as surrogate measures of safety to assess roadway countermeasures. Zhao et al. (2015) evaluated the impact of chevrons on driver behavior (in China) with a driving simulator. The interaction between chevrons and curve radius was evaluated using average speed, the standard deviation of speed, and the standard deviation of lane position. Rasanen (2005) collected encroachment and lane position using video surveil- lance to assess the impact of rumble strip barrier lines that were implemented along a two-lane rural curve in Finland. Ben-Bassat and Shinar (2011) assessed effectiveness of shoulder width, presence of guardrail, and curve radius by comparing mean speed and changes in lateral posi- tion using 22 drivers in a driving simulator. McKnight, McKnight, and Tippets (1998) assessed the combined effect of lane edge marking width and marking contrast on lane keeping using 124 participants in a driving simulator. They used heading error, position error, and lane excursions as surrogate measures. Bella (2013) evaluated different roadway characteristics and

24 Performance Criteria for Retroreflective Pavement Markers presence of different types of guardrails using a driving simulator in Italy. Mean speed and lane position were used as surrogate measures. Gates et al. (2012) used video surveillance to assess passing maneuvers, lateral lane position, and centerline/edge line encroachments before and after the installation of centerline and shoulder rumble strips on rural two-lane roads in Michigan. A number of studies have also used speed metrics, encroachments, or lateral positions to eval- uate the impact of RPMs on driver behavior. Mullowney (1982) evaluated driver perfor m ance with RPMs on two rural two-lane roads. Encroachments and speed were collected using pneumatic road tubes. A 6% reduction in edge line encroachments and a 4% and 12% reduction in center- line encroachments were found after installation. They also found the speed profile through the curve was smoother in both dry and rain conditions, as evidenced by smaller changes in speed between data collection points after the RPMs were installed. Zador et al. (1987) evaluated the impact of several different delineation treatments includ- ing RPMs on rural two-lane curves in Georgia and New Mexico. The RPM installations were at 80-ft spacing before the curve and 40-ft spacing within the curve. They found a minor increase in mean speed (0.7 mph) at night at locations with RPMs. They also found vehicles shifted away from the centerlines with RPMs at night, but the change was less than 0.1 ft. Hammond and Wegmann (2001) evaluated the effects of RPMs on the number of encroach- ments, encroachment distance, and average speed at two horizontal curves. The researchers tested RPMs spaced at 20 ft and 40 ft apart. The researchers found that high degrees of lane encroachments decreased by 7.5%, moderate degrees of lane encroachments decreased by 7%, and low degrees of lane encroachments decreased by 14.5% with the 40-ft spacing. Likewise, the researchers found similar results for the 20-ft spacing. However, the researchers did not find any conclusive results for changes in average speed. Nissner (1984) evaluated speeds before and after installation of RPMs with 40-ft spacing at three horizontal curves. No differences in daytime or nighttime speeds were noted, but fewer centerline or edge line encroachments were noted. Agent and Creasey (1986) evaluated two S-curves and one regular curve where centerline RPMs were installed. They reported a 2 mph and 7 mph reduction in daytime speed and a 2 mph and 3 mph reduction for nighttime speeds. Encroachments decreased at one site by 3%, while the other site experienced an increase of 4%. Freedman et al. (1988) used simulated curves to evaluate driver behavior in the presence of RPMs, finding smoother operation and improved lane tracking when RPMs were present. The study also included a field observation component that corroborated the simulation. The research team made recommendations of minimum luminance contrast ratios of 0.5 in ideal conditions and 1.0 in glare on dry pavement while stating that contrast levels 2.0 or 3.0 were necessary to achieve 3 seconds of preview distance for younger and older divers on wet road- ways. Similar sentiment was expressed in the Pezoldt et al. (1990) study, with the research team stating that no one specific intensity value could be specified without consideration of other roadway geometric and operational characteristics. While it may seem obvious to some that improved delineation will result in improved driver behavior, the idea has been called into question. In a nighttime driving field study, Donnell et al. (2007) found little evidence of improved speed and lateral position consistency based on analysis that was conducted at the curve level (comparing multiple treatments at individual curves), using observations taken at specific points along the curve. Other applications of RPMs that have been evaluated include crash reductions at intersections, where the CRF is estimated to be 6%–13% for total crashes, 20%–30% for nighttime crashes, and 20%–46% for wet nighttime crashes (Institute of Transportation Engineers 2004); traffic calming at pedestrian crossings (Davis and Hallenbeck 2008); and speed reduction in conjunction with other traffic calming devices (Galante et al. 2010). Considering these other applications, the

Literature Review 25   general consensus is that RPMs provide safety benefits, though the magnitudes of those effects vary by application and location. 2.2 Application-Related Studies In addition to the in-depth scan of research literature to identify potential safety benefits of RPMs, the research team examined journal articles and reports to determine existing or sug- gested guidelines for the applications of RPMs. The researchers conducted this search to answer several questions: • What volume criteria are appropriate for RPM use? • Where in the roadway should RPMs be applied? • What functional classifications and lane configurations are best suited to RPM usage? • How frequently should RPMs be spaced? • What form of RPMs should be used? • How frequently should RPMs be maintained? • Should RPMs be used in conjunction with other delineation devices? The following sections are subdivided into two larger topics: the form and function of RPMs and the application of RPMs. Where applicable, justifications for particular standards or guide- lines are provided. 2.2.1 Form and Function of RPMs 2.2.1.1 Types of RPMs State agencies report the implementation of a wide variety of RPMs, often making it difficult to compare the functional characteristics and operational effectiveness of different types. Albin et al. (2016) reported the use of both reflective and nonreflective, recessed, and snowplowable RPMs. Even within these different categories, there are a number of shapes and heights (Bahar et al. 2004), and FHWA provides guidance for construction materials based on the type of RPMs in its 1998 guidance (Grant and Bloomfield 1998). Multiple states have reported concerns over the use of snowplowable RPMs because these markers may dislodge and damage windows or break plows (Grant and Bloomfield 1998). These concerns are discussed further in Chapter 3, “Survey and Review of State Practices.” There are also concerns that the retroreflectivity of recessed RPMs, which are designed to not conflict with plowing operations, may be degraded over time by residue and water (Bahar et al. 2006). Zaidel, Hakkert, and Pistiner (1992) also raised concerns that RPMs may degrade smoothness and serviceability for road users, although they reported that RPMs do not cause significant noise issues for communities. Chapter 3 of this report details more broadly the types of RPMs used throughout the United States. 2.2.1.2 Maintenance Maintenance practices for RPMs vary from jurisdiction to jurisdiction and may depend on the form and function of the markers. Confounding these factors is a lack of clarity in report- ing regarding the scope of maintenance; because snowplowable RPMs consist of both a casting and a retroreflective marker, some maintenance operations may replace only the casting or may replace the entire marker itself. Despite the lack of uniformity and consistency in applications, some guidance exists on installation, maintenance, and rehabilitation procedures. Researchers and agencies have commonly indicated that RPMs are either installed or replaced when pave- ment is resurfaced; for example, Park, Abdel-Aty, and Wang (2017) reported that in Florida, both new RPMs and replacement reflectors are installed during repaving projects. Chapter 3 of this report contains various maintenance cycles corresponding to repaving. In Guidelines for the

26 Performance Criteria for Retroreflective Pavement Markers Use of Raised Pavement Markers, FHWA recommends that RPMs should not be installed for an entire year after the use of a rejuvenating agent on pavement (Grant and Bloomfield 1998). In the MUTCD, FHWA (2012) recommends that any marker that is no longer in use should be obliterated rather than left on the roadway to be a hazard to drivers. Factors that may correspond to a degradation in effectiveness and a need for more frequent replacement may include place- ment on horizontal curves, higher percentages of heavy vehicles in daily traffic, and Portland cement concrete surfaces (Guo, Lu, and Yu, 2015). For RPMs used on asphalt concrete surfaces, Fontaine and Diefenderfer (2011) reported more frequent failure and increased need for main- tenance on stone mastic asphalt than hot mix asphalt (HMA). Fontaine and Diefenderfer also reported the traffic loading thresholds for RPM replacement shown in Table 7. The replacement cycles are based on the number of missing RPMs considered acceptable to still provide adequate delineation. Based on these results and the other reported research, agencies have a variety of replacement and maintenance cycle options to maintain safe delineation depending on RPM type and facility. 2.2.2 Application of RPMs 2.2.2.1 Roadway Type Researchers and state officials most consistently report that RPMs are applicable on all free- way types. This guidance seems to extend from FHWA’s 1998 Guidelines for the Use of Raised Pavement Markers, which recommends the use of snowplowable RPMs on all freeways and interstate highways (Grant and Bloomfield 1998). This FHWA recommendation is based on a 1993 technical report based on best practices for highway delineation (Pennell 1993); RPMs enhance delineation at high speeds and volume conditions of freeways and are therefore sug- gested for use on this facility type. The FHWA recommendations have since been supported by other researchers. Researchers also reported a few other noteworthy applications of RPMs. The older FHWA guidelines also indicate the use of RPMs on rural two-lane roads, especially on curves (to enhance delineation; Grant and Bloomfield 1998), and Persaud and Lyon (2007) indicated that some states apply RPMs more selectively on two-lane roadways. Last, Indiana DOT (INDOT) recommends the use of RPMs on other multilane highways but not at locations with sufficient illumination based on a sample of best practices reported by individual districts throughout the state (Jiang 2006). 2.2.2.2 Roadway Location Researchers and agencies typically recommend specific roadway locations for RPM use in conjunction with roadway-type suggestions. In 1998, FHWA recommended RPMs should be used on both centerlines and left edge lines to promote proper delineation. However, FHWA does not recommend the use of RPMs on right edge lines unless there is a specific safety prob- lem; there is concern that right edge line delineation may encourage inappropriate speed and Percentage of Snowplowable RPMs Missing or Damaged Traffic Threshold (millions of vehicles) Estimated Years to Achieve Traffic Threshold Given Current AADT 20,000 veh/day 40,000 veh/day 60,000 veh/day 80,000 veh/day 3.0 74.5 12 6 4 3 4.6 150.8 >25 12 8 6 6.1 212.6 >25 17 11 8 Table 7. Possible inspection and replacement cycles for snowplowable RPM castings from Fontaine and Diefenderfer (2011).

Literature Review 27   result in departure crashes (Pennell 1993; Grant and Bloomfield 1998). The current edition of the MUTCD also does not support use of RPMs on right edge lines (FHWA 2012). The rationale for this standard is that edge line RPMs may confuse drivers and cause them to mis- interpret the markers as lane lines (Albin et al. 2016). In an earlier study, Zador et al. (1987) found that RPMs, when used on double-yellow centerlines, increased delineation for drivers but may also inadvertently increase vehicle speeds. They therefore encouraged the use of one-way RPMs when extra delineation is necessary. An internal document from the Oregon Department of Transportation also recommends RPMs on centerlines and lane lines to assist delineation (Leaming 2016). In comparison, Caltrans (2016) also recommends the use of RPMs on edge lines or centerlines to improve driver delineation. In fact, Caltrans engineers suggest that RPMs be used in conjunction with edge lines to assist drivers in understanding the limits of the roadway to reduce crashes on local roads. In summary, most researchers and agencies recommend RPMs for use on left edge lines and centerlines at the least because studies have shown that the treatment can improve roadway delineation. Right edge line applications may also be appropriate when it is likely that drivers may not understand roadway geometry or when objects are near the edge of the road. Although their design and exact mechanism differ, audio-tactile lane markings (ATLMs) also accomplish similar delineation goals and function- ality; Hatfield et al. (2009) recommended similar applications of ATLMs as RPMs, especially as a delineator on the centerline. The authors differ, however, from some researchers in that they also recommend ATLMs as a delineation device on both edge lines when applied to rural freeways (Hatfield et al. 2009). In addition to edge line and centerline applications, some researchers and agencies also rec- ommend that RPMs be applied specifically to horizontal curves. The federal guidance distributed by FHWA in 1998 reported that RPMs can be used on rural two-lane roads to assist drivers navi- gating horizontal curves (Grant and Bloomfield 1998). An earlier study by Agent and Creasey (1986) informed this recommendation; based on a roadway sample from Kentucky, the authors concluded that RPMs improve delineation around horizontal curves. Studies conducted after the publication of the FHWA guidance also confirm these results. Jiang (2006) reported applica- tions of RPMs on rural roadway curves throughout Indiana; in contrast to the recommendations of Bahar et al. (2004) in NCHRP Report 518, Jiang even recommended that RPMs should be considered on especially winding rural roads despite potential safety degradation. Jiang recom- mended the use of RPMs even if the degree of curvature exceeds 3.5 degrees, due to the positive impact of RPMs on delineation. Two studies published in 2015 also support the use of RPMs on horizontal curves. In one, Zhao et al. (2015) drew from a 2000 simulation study of speed limit- ers (Comte and Jamson 2000) and a 2010 simulation study of treatments that highlight curve radius (Jamson, Lai, and Jamson 2010) to suggest that RPMs may be an adequate alternative to chevrons, large horizontal arrows, or advance warning signs on horizontal curves. It should be noted that neither of the studies referenced by Zhao et al. explicitly mention RPMs or present real-world evaluations. In the other 2015 study, Guo, Lu, and Yu found positive safety effects of RPMs for delineation on horizontal curves and even suggested that RPMs should be placed such that they face the tire direction to minimize the impact angle. Last, an internal Oregon Department of Transportation document (Leaming 2016) supported the findings in NCHRP Report 518 and recommended that RPMs should be discontinued 5 seconds before a curve when the degree of curvature is greater than 3.5, and the speed limit is greater than or equal to 50 mph. The consensus, then, is that RPMs should be used on horizontal curves to improve delineation, especially on rural roads. However, care should be taken regarding application to curves with high degrees of curvature; RPMs may be applicable in some cases when especially winding roads are present, but engineers should be cautious when the degree of curvature exceeds 3.5 degrees. Bahar et al. (2004) also cautioned against the use of both edge lines and centerlines on sharp curves because the combination may confuse drivers.

28 Performance Criteria for Retroreflective Pavement Markers 2.2.2.3 Spacing Agencies typically install RPMs at three different spacing lengths dependent on lane configu- ration, striping, and roadway location. These spacing lengths are 80 ft, 40 ft, and 20 ft. In the 1998 guidance, FHWA recommends a maximum spacing of 80 ft for all RPM applications and more specifically under certain operations such as centerlines (except on yellow double-solid lines on multilane roads); broken lane lines; and horizontal curves when the degree of curvature is less than 3 degrees (Grant and Bloomfield 1998). Before the release of the FHWA guidance, Zador et al. (1987) identified that states often applied RPMs at both 80-ft and 40-ft spacings with no clear rationale. In NCHRP Report 518, Bahar et al. (2004) reported widespread state compliance with the Roadway Delineation Practices Handbook recommendations for RPM spacing (Migletz, Fish, and Graham 1994); the standard application of RPMs on tangents is 80 ft (Bahar et al. 2004). Bahar et al. (2004) noted that this spacing should provide a sufficient preview time for drivers equal to 2–3 seconds. More recently, Jiang (2006) reported a standard practice among individual districts in Indiana consistent with the 80-ft spacing widely reported elsewhere. In some cases, a 40-ft spacing may be more appropriate for RPMs. As noted, Zador et al. (1987) reported some usage of 40-ft spacing in different states without a clear rationale. Further clarification was offered in FHWA’s 1998 guidance to recommend a tighter, 40-ft spacing on edge lines or on horizontal curves when the degree of curvature is between 3 and 15 degrees; when the degree of curvature is 8–9 degrees, a 40-ft spacing is the maximum allowed (Grant and Bloomfield 1998). Bahar et al. (2004) also reported this recommendation, as did Jiang (2006). The reason for this spacing is to promote continual visibility of multiple RPMs while traveling through a curve (Bahar et al. 2004). A spacing of 20 ft may be applicable in specific cases. The FHWA guidelines recommend a 20-ft spacing to ensure clear delineation on horizontal curves with a degree of curvature greater than 15 degrees; for hazardous curves where there is a risk of a roadway departure or collision with a fixed object, the recommended right edge line spacing for RPMs to ensure proper delinea- tion is 20 ft (Grant and Bloomfield 1998). However, a 2001 study by Hammond and Wegmann testing driver encroachment during the daytime following an RPM spacing reduction from 40 ft to 20 ft found no significant difference between the two spacing options. Jiang (2006) reported finding no significant encroachment difference relative to 20-ft spacing to the INDOT. Recently, FHWA (2013) noted a positive safety benefit to 20-ft spacing of RPMs in curves in Alabama. Engineers from Mobile County and ALDOT implemented a systemic installation of RPMs on all roadways in the county, with a uniform spacing of 20 ft on all curves, and reported a net reduction in crashes. ALDOT engineers concluded that the widespread application, including the 20-ft spacing on curves, was critical for giving guidance to drivers during dark and rainy conditions. In addition to these more widely applied spacing standards, both individual researchers and FHWA have provided alternative guidance for RPM application. In the MUTCD, the recom- mended spacing for RPMs is 2N (where N is equal to the length of one segment and one gap, typically 40 ft). The MUTCD notes that RPMs may be spaced at 3N on freeways or N at other desired locations (FHWA 2012). In contrast to the MUTCD requirements, Chrysler, Carlson, and Williams (2005) reported that a fixed spacing for RPMs equal to twice the MUTCD rec- ommended length should be sufficient for tangent sections before and after a horizontal curve. Jiang (2006) reported that INDOT allows another large spacing equal to 100 ft on toll roads, seemingly due to the higher capacity and design speed of this facility type. Recently, in Low- Cost Treatments for Horizontal Curve Safety 2016, Albin et al. (2016) did not specify any strict spacing requirements, instead noting that spacing on curves and winding segments should be reduced enough to adequately show the alignment. Based on these references, it can be con- cluded that spacing should be adequate to allow drivers the ability to see enough RPMs to detect

Literature Review 29   roadway geometry. The spacing may vary under different roadway conditions and operational requirements. 2.2.2.4 Volume Criteria NCHRP Report 518 (Bahar et al. 2004) concluded that RPMs should only be applied under certain operational conditions. According to the report, when the traffic volume is below 20,000 veh/day on four-lane freeways, the CMF for RPMs is 1.13. More specifically, the researchers found a 43% increase in crashes when the curvature was greater than 3.5 degrees, and the AADT was less than 5,001 veh/day on rural two-lane roads. However, other studies reported conflicted operational requirements. Drawing from RPM placement practices used in Illinois (Matthias 1988), FHWA produced the example look-up table for delineation by RPMs shown in Table 8. Note that these recommendations are aged and cover few operational situations. As discussed in the safety evaluation literature review, Sun and Das (2013) studied the safety effects of RPM implementation on freeways throughout Louisiana, ultimately concluding that RPMs provided safety benefits in all operational conditions. As shown previously in Table 6, the CMFs developed for all three volume categories (AADT ≤ 20,000 veh/day; 20,000 veh/day ≤ AADT ≤ 60,000 veh/day; AADT ≥ 60,000 veh/day) on rural roads demonstrated significant crash reduction potential. Sun and Das found, however, that the largest reduction was for night- time crashes on rural freeways with an AADT ≤ 20,000 veh/day where RPMs were used in con- junction with quality striping. However, as was discussed earlier, the study by Sun and Das had some significant limitations. Other researchers identified other volume criteria for applications of RPMs. In a study of interstate segments with AADT > 15,000 veh/day in Virginia, Fontaine and Diefenderfer (2011) found that total traffic volume is a failure criterion for snowplowable RPMs. They identified a threshold of 111 million vehicles for failure of snowplowable RPMs, at which point the snow- plowable RPM castings likely break or dislodge and should be replaced. This finding does not directly reveal potential safety effects for RPMs, but it does indicate a potential cycle for main- tenance to ensure that RPMs do not deteriorate sufficiently to adversely affect safety. Finally, internal recommendations for the Oregon Department of Transportation in Leaming (2016) included a base AADT greater than 20,000 veh/day for the safety benefits of RPMs to be realized. Leaming also noted that RPMs may be implemented on nonfreeways when justified by safety concerns in the regional plan. In conclusion, research seems to indicate that the most significant benefits of RPMs may be realized on facilities designed to carry larger volumes or when specific safety problems can be solved by enhanced delineation on curves. 2.2.2.5 Delineation and Other Usage RPMs may be used in a variety of contexts to provide or improve roadway delineation for drivers. Zador et al. (1987) recommended RPMs as a supplemental delineation system based on a measurement of mid-curve speed and direction on rural two-lane highways. The researchers found that RPMs provided some displacement in the curve but may also increase mean approach speed when compared to other delineation devices. Gawron and Ranney (1990) used Table 8. Example RPM operational look-up table from Matthias (1988). By Average Daily Traffic (ADT) Rural Highway (two-way) Divided Highway Urban Street (two-way) Urban Street (one-way) One-lane Two-lane ADT 15,000 Multilane ADT 2,500 Not specified ADT 7,500

30 Performance Criteria for Retroreflective Pavement Markers a simulation study to prove that RPMs may be used effectively in conjunction with chevrons to move vehicles away from centerlines. Persaud, Retting, and Lyon (2004) reported that some jurisdictions simply use RPMs as alternatives to rumble strips; considering the Bahar et al. (2006) retroreflectivity study, this application may still improve delineation even if reflectivity is not the goal. Carlson, Park, and Anderson (2009) reported that wider pavement markings, which in many cases are applied in tandem with RPMs, may improve delineation in curves by decreasing drivers’ need for central focus. More visibility-related studies are discussed in the next section. 2.3 Visibility-Related Studies 2.3.1 Vision and Driving RPMs have been recommended for use as delineation, specifically for isolated horizontal curves, particularly because they can provide guidance both downstream from and immediately at the vehicle (Stimpson et al. 1977) in dry and wet-weather nighttime driving conditions (Liptak 1980). Drivers use visual cues provided by RPMs, pavement markings, and the surrounding environment to navigate roads. Past human factor research described the driver as a three-part processor. The three parts of the processor are the compensatory subsystem (which adjusts vehicle position based on errors from the other two systems), the pursuit subsystem (which responds to stimuli further down the road and allows the driver to anticipate a desired path), and the precognitive subsystem (which applies previously learned movements and skills) (McRuer et al. 1977). In the 1970s, research led by Shinar found that drivers tended to fixate on objects farther from their vehicle on tangent roadway sections, while on curves, drivers intermittently performed saccadic scans (characterized by rapid, simultaneous eye movement) closer to their vehicle and down the road. Drivers performed scans while generally attempting to maintain a preview distance of 2.5–3.5 seconds (Shinar, McDowell, and Rockwell 1977). The Shinar study is one of several studies cited in FHWA guidelines that collectively present a range of 2 to 5 sec- onds as appropriate preview distances (Grant and Bloomfield 1998). When drivers are able to maintain an adequate preview distance, they are able to use the pursuit processing subsystem. The pursuit subsystem is responsible for planning routes through the road, thereby allowing the driver to maintain more consistent vehicle operation (McRuer et al. 1977). A study led by Carlson compared the preview time provided by one type of new RPM to a variety of new paints, tapes, thermoplastics, and other pavement marking materials (Carlson et al. 2007). The study demonstrated that RPMs provided preview time in excess of 4 seconds at speeds ranging from 30 mph to 65 mph during both wet and dry nighttime conditions, while the markings, some of which were designed for wet-weather performance, provided substantially less, particularly during wet-night, high-speed conditions. Markings and markers were considered in terms of cost per foot of visibility, finding that RPMs provided the most visibility per dollar spent. A report from New Zealand came to similar conclusions regarding the visibility and cost-effectiveness of RPMs through a series of field evaluations of curve delineation (Thomas et al. 2017). A study con- ducted for the Ohio DOT evaluated the visibility and durability of snowplowable RPMs and wet-reflective pavement markings in dry, wet, and rainy conditions (Abbas and Sarker 2012). The Ohio DOT study found the pavement markings provided high initial dry and wet visibility, but performance was significantly compromised during the first and second winter seasons. The study found the snowplowable RPMs provided consistently higher wet-night visibility dis- tances compared to the pavement markings. The researchers concluded that the snowplowable RPMs were the most cost-effective wet-night delineation for the conditions evaluated (Abbas and Sarker 2012). The SHRP 2 naturalistic driving data have provided an avenue to study driver behavior in the presence of a wide variety of on-road delineation scenarios. A study by researchers at Iowa

Literature Review 31   State University and the University of Iowa used the SHRP 2 data to investigate driver behavior through curves. One finding indicated in the study was that RPMs (as well as chevrons) may provide improved advanced information for drivers based on when drivers reacted to seeing the curve (Hallmark et al. 2015). 2.3.2 Degradation of Visual Properties The time relative to installation at which some of the previously discussed studies evaluated the driver performance relative to the condition of RPMs may have contributed to the findings of each study. Based on observations of 6-month-old RPMs, Zador et al. (1987) found no differ- ence in speed or lateral position with respect to RPM age. The authors of a Texas study found that the specific intensity (SI) of RPMs deteriorated quickly (loss of 70%) within the first 6 months of installation and plateaued over the remainder of a 12-month-period (Pezoldt et al. 1990), while Ullman (1994) noted that RPMs experience significant losses in reflectivity over short periods of time. However, it is not necessarily clear how much deterioration of material properties must occur before the visibility of the RPM is affected. Bullough and Liu (2018) recently conducted a laboratory study to investigate the relationship between RPM visibility and luminance based on relative visual performance (RVP), ultimately finding that used RPMs with luminance values 20%–30% lower than new RPMs had similar visibility characteristics to new devices. 2.3.3 Challenges of Visibility Data The apparent lack of difference in RVP relative to changes in luminance is not necessarily surprising, given that Zhang et al. (2009) noted that laboratory tests (such as those used to quantify luminance) do not necessarily correlate with the field performance of RPMs. Field measure- ments can be problematic as well, with Pike (2017) noting variation in portable retroreflecto- meter measurements of up to 40% when measuring RPMs, as well as discrepancies between portable retroreflectometer and vehicle-mounted retroreflectometer measured values (Pike and Peretin 2018). 2.4 Summary The existing body of literature with regard to RPMs is quite large. Through an extensive examination of that literature, several observations that serve to form the basis of the remainder of this document are clear. First, the safety literature has conflicting results as to the benefit of RPMs. Much of the safety literature is handicapped by the difficulty of isolating the impact of RPMs on safety. This difficulty is attributable to the quality of data available regarding RPM installation, more specifically, data indicating other crash countermeasures and site modifications that are installed concurrently with RPMs. Three studies highlight a multitude of the challenges of assessing RPM perfor- mance: Kugle, Pendleton, and Von Tress (1984); Mak, Chira-Chavala, and Griffin (1987); and Griffin (1990). These studies make use of multiple versions of the same data to draw similar conclusions; however, these conclusions are drawn by comparing nighttime crashes to day- time crashes and focusing on high crash locations. Ultimately, a retrospective safety analysis regarding RPMs that overcame the aforementioned shortcomings was not feasible as a part of this study. Second, in lieu of crash-based safety assessments, a substantial part of the extant RPM litera- ture uses surrogate safety measures such as mean speed, lateral position, and encroachments to examine RPM effectiveness. Generally speaking, these studies indicate that RPMs are capable of

32 Performance Criteria for Retroreflective Pavement Markers influencing driver behavior in ways that should impact safety in a positive manner. However, some of the effects of RPMs on the surrogate safety measures, such as increased speed, may actu- ally be indicative of overdriving, which could reduce safety. While the existing literature has shown that RPMs provide greater preview time than mark- ings, there have been significant challenges in establishing a clear relationship between RPM age (and subsequently, observable visibility characteristics) and driver behavior. This summary of the literature review serves to illustrate why the three data-driven compo- nents of this study are important. A closed-course recognition and visibility examination, a closed-course evaluation of driver behavior through curves, and a cross-sectional investigation of naturalistic driving data through curves and tangents with and without RPMs are described later in this document to help address gaps in the existing body of literature.