In this paper, visibility evaluations include research directed toward the identification of correlation rates between pavement marking visibility (measured in terms of retroreflectivity) and crashes, and research evaluating the detection of pavement markings with various characteristics and properties. Detection distances are thought to be a surrogate for crash data in that longer detection distances have a positive effect on vehicle-control measures and, consequently, crashes. Because of this, incremental visibility improvements provided by pavement markings have been viewed as a proxy for improved roadway safety, although no direct link has been established.

Studies of Visibility in Terms of Retroreflectivity and Crashes
Recently, there have been several attempts to statistically link pavement marking retroreflectivity levels to crash rates. A significant challenge has been that pavement marking retroreflectivity levels are dynamic in that they continuously change. Attempts to model pavement marking retroreflectivity degradation curves have not been widely successful (39, 40). While there is some predictability to pavement marking retroreflectivity (it is generally accepted that ADT is a significant predictor variable), it can change unpredictably and substantially as a function of frequency and intensity of rain (to clean markings), quality of installation, or even the condition of the pavement. Therefore, it is difficult to know the retroreflectivity levels of the pavement markings at the exact time and location of each crash. While crash data are available, researchers have had to make assumptions regarding the retroreflectivity levels for their analyses. Some researchers model retroreflectivity using various sources of measured data, while others make assumptions about the retroreflectivity without measurements. This section of the paper describes attempts to statistically link pavement marking retroreflectivity levels to crash rates.

In 2006, researchers in New Zealand studied the safety impacts of brighter pavement markings and concluded that there was a not a conclusive improvement in safety (41). In 1997, New Zealand implemented a minimum maintained retroreflectivity policy of 70 mcd/m²/lx for their state system. Using a before-after approach, the authors compared the crash rates before the change in policy. They assumed that markings were brighter during the after period. It should also be pointed out that, in New Zealand, all state roadways are delineated as a function of traffic volume. As volumes increase, they progressively apply the following treatments: delineators, centerlines, edge lines, and then RRPMs. Therefore, roadways with centerlines had delineators too. Previous research in the United States has shown that supplemental delineation treatments, such as delineators or RRPMs, overpower the potential effect of pavement markings (42).

Also in 2006, the results of a National Cooperative Highway Research Program (NCHRP) study were published with the following conclusions: “…the difference in safety between new markings and old markings during non-daylight conditions on non-intersection locations is approximately zero” (43). While the study incorporated large amounts of crash data and utilized the latest statistical techniques, there were significant limitations to the study. For instance, the research only included crashes from California and modeled retroreflectivity (no measurements were made). While the study included efforts to overcome the possible limitations in modeling retroreflectivity, these efforts presuppose that markings in California never reach a value where there is an adverse impact on safety. The pavement marking maintenance policy of California is such that they restripe their higher-volume highways up to three times a year with paint, or every two years with thermoplastic markings. As a result, there is only the occasional roadway with retroreflectivity levels below 100 mcd/m²/lx.

Overlooking the concerns regarding the modeled retroreflectivity levels, perhaps even more concerning is the binning of the modeled retroreflectivity levels. The binning thresholds were derived linearly, which by itself is a limitation since the performance of retroreflectivity has been repeatedly shown to be best modeled logarithmically rather than linearly (47, 48). In addition, the authors binned the modeled retroreflectivity data such that the lowest bins for the edge lines included retroreflectivity levels from 21 to 183 mcd/m²/lx, thus including both inadequate levels and near-desired levels in the same bin (see TABLE 2). Eight additional bins included retroreflectivity levels up to 413 mcd/m²/lx. Therefore, all binning used in the analyses included levels deemed to be acceptable or at least above minimum retroreflectivity levels. Combined, these limitations and concerns seriously challenge the quoted concluding remarks shown above.

In 2007, researchers reported results from an effort to develop a statistical association between measured pavement marking retroreflectivity and traffic crash frequency (44). The results suggest that increased levels of the average pavement marking retroreflectivity on multi-lane highways may be associated with lower expected target crash frequencies; however, the association is small in magnitude and not statistically significant. On two-lane highways, the association between pavement marking retroreflectivity and crash frequency is larger in magnitude and marginally significant. While this study used measured retroreflectivity levels (recorded once per year), it should be noted that all the retroreflectivity data were well above what might be considered minimum levels, and even near what might be considered desired levels (all data were above 100 mcd/m²/lx with an overall average of 240 mcd/m²/lx). These researchers are continuing to evaluate their data using innovative techniques such as modeling retroreflectivity using neural networking techniques.

In 2008, a similar effort was reported that included 3 years of measured retroreflectivity (measured once per period) in Iowa (48). These data were analyzed along with crash records from the same year. The distributions and models of the entire database, and a subset including only two-lane highways, did not show that pavement marking retroreflectivity correlated to crash probability. When truncating the data to only records with retroreflectivity values ≤ 200 mcd/m²/lx, a statistically significant relationship was determined. However, the correlation was small. This research is also being continued using retroreflectivity thresholds near the generally accepted minimum levels of 100 mcd/m²/lx.

Studies of Visibility in Terms of Detection Distances

Pavement marking detection distances are usually measured with two different techniques. In a static setup, the driver counts the number of skip lines visible. In a dynamic setup, where the research participant is driving the vehicle, the driver is tasked with detecting either the beginning or end of a long line, an isolated skip line (52), or a discontinuity such as a taper (46). The results are reported in maximum nighttime detection distances. These studies are usually conducted with pavement markings of various retroreflectivity levels (measured dry or wet), of different widths, and from different vehicles (in consideration of their size and headlamp type). The studies have repeatedly shown that pavement marking detection distances are correlated to retroreflectivity in a logarithmic fashion (47, 48).

While the use of wider pavement markings continues to grow across the United States (49) and research results are looking favorable in terms of the impact on safety, the results of studies based on maximum detection distances of pavement marking widths are inconclusive. On one hand, a number of research efforts show increased visibility for wider lines (22, 50, 51), while on the other hand, research findings also show that there are no consistent statistical or practical differences (23, 52). An empirical study has shown that theoretical calculations of marking detection distance as a function of marking width are invalid, and more work is needed to develop mathematical relationships between marking width and detection distances (53).