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.
C H A P T E R 2 Identifying Final Lane Departure Research Questions and Relevant FactorsOne of the goals of this project was to develop a set of research questions that could be explored using existing NDS data. The intent was to then determine which research questions could adequately be addressed given the data and the likely limitations of the SHRP 2 full-scale NDS (hereafter referred to as the full-scale NDS or full-scale study). In order to identify lane departure research questions, the research team first identified which driver, vehicle, roadway, and environmental factors were likely to contribute to the occurrence and severity of lane departure crashes, based on an in-depth literature review described in the next section and on the teamâs expertise in lane departure issues. The team then reviewed data from existing NDSs, as well as information available about the full-scale NDS. Research questions that could not feasibly be answered because the necessary data would not be available or could not be extracted were identi- fied. This chapter identifies relevant factors, Appendices A and B address the feasibility of extracting the data elements from the UMTRI and VTTI databases, and Chapter 4 com- ments on which data factors are expected to be available with the full-scale study. The team also identified which research questions could be addressed in the present research. Although the information to select final research questions is based on the information provided in the following chapters, the infor- mation is also provided in this chapter to simplify report organization. Relevant Data Elements Identified in Existing Literature In order to formulate research questions, it was necessary to determine which data elements are the most relevant. This section addresses factors necessary or desirable to evaluate lane departure crashes. The data elements were selected through a review of currently available literature regarding which road- way, environmental, vehicle, and driver variables are correlated to road departure crash occurrence. Roadway data elements10and crash data information were also selected based on the teamâs experience in working with road departure crashes and its understanding of what roadway variables are likely to be available or collected. Roadway Factors Horizontal and Vertical Curves Horizontal and vertical curvature, as well as grade, have been correlated with crash occurrence in a number of studies. Torbic et al. (2004) report that the crash rate for horizontal curves is approximately three times that of tangent sections. The authors also indicate that approximately 76% of curve-related fatal crashes are single-vehicle run-off-road (ROR) crashes and 11% are head-on with an oncoming vehicle. A review of the Iowa DOT crash data indicates that in Iowa (2001â2005), 12% of all fatal crashes and 15% of all major injury crashes occurred on curves; 14% of all urban fatal crashes and 11% of all urban major injury crashes occurred on curves; and 11% of all rural fatal crashes and 19% of all rural major injury crashes occurred on curves. Miaou and Lum (1993) studied heavy truck crashes using 1985â1989 Utah Highway System Information System (HSIS) data and evaluated horizontal curvature, vertical grade, and width of paved shoulder. They found that as vertical grade increased, truck accident involvement also increased. They also found that truck crash involvement increased as horizon- tal curvature increased, depending on the length of the curve. Hauer et al. (2004) used a negative multinomial model using Washington HSIS crash data to predict the nonintersec- tion accident frequency of urban, four-lane, undivided roads (1993â1996). They found no significant correlation between crashes and vertical grade. Lamm et al. (1988) and Council (1998) found that crash rates increased as degree of curve increased, even when traf- fic warning devices were used to warn drivers of the curve.
11Mohamedshah et al. (1993) found a nonintuitive negative correlation between crashes and degree of curve for two-lane roadways. Council (1998) also found that the presence of spirals on hor- izontal curves reduced crash probability on level terrain, but did not find the same effect for hilly or mountainous terrain. Vogt and Bared (1998) evaluated two-lane rural road seg- ments in Minnesota and Washington State using HSIS data and found a positive correlation between injury crashes and degree of horizontal curve. Shankar et al. (1998) evaluated divided state highways with- out median barriers in Washington State and found a relation- ship between the number of horizontal curves per kilometer and median crossover crashes. Zegeer et al. (1992) evaluated 10,900 horizontal curves on two-lane roads in Washington State using a weighted linear regression model. They found that crash likelihood increased as the degree and length of curve increased. Alternatively, Deng et al. (2006) evaluated head-on crashes on two-lane roads in Connecticut for 720 segments using an ordered probit model. They included geometric characteristics in the analysis, but did not find that the presence of horizontal or vertical curves was significant. The vehicle speed reduction required for traversing a curve has an impact on the frequency and severity of crashes on curves. Abrupt changes in operating speed resulting from changes in horizontal alignment have been suggested to be a major cause of crashes on rural, two-lane roadways (Lamm et al., 1988). Higher crash rates were experienced on horizontal curves that required greater speed reductions (Anderson et al., 1999). This finding was also supported by Fink and Krammes (1995), who indicated that curves requiring no speed reduction had no significantly different mean crash rates from their pre- ceding roadway tangents. The roadway tangent length also influences driver behavior. The effect of a long tangent preced- ing a curve becomes more of a factor on sharp curves. Roadway tangent lengths also impact crash rates on steep downgrade curves. Crash rates on curves with long tangent lengths are more pronounced when the curve is located on a downgrade of 5% or more, with tangent lengths of more than 200 m. Preston (2009) found that crash rate increases as radius decreases below 2,000 ft and that around 90% of fatal and 75% of injury crashes occurred on curves with radii less than 1,500 ft. McLaughlin et al. (2009) evaluated ROR events using the VTTI 100-car NDS data. In that study, ROR crashes and events included those where one or more tires contacted a curb or left the roadway before returning to the roadway, where the vehicle departed the road and came to a stop, where the vehi- cle collided with a lane delineation object (e.g., curb, con- struction barrels), or where the driver braked hard and swerved to avoid a crash. The authors found a total of 122 ROR events,which included 28 crashes and 94 near crashes. They reported that 30% of the ROR events occurred on curves, 56% occurred on tangent sections, and 14% occurred at intersections. Roadway Cross Section Lane width, shoulder type, shoulder width, median type, and median width have all been associated with crash experience. A summary of some of the available literature follows. Table 2.1 also summarizes the information.Miaou and Lum (1993) studied heavy truck crashes using 1985â1989 Utah HSIS data and evaluated horizontal curva- ture, vertical grade, and width of paved shoulders. The authors found that as the width of the inside paved shoulder increased, truck involvement decreased. Mohamedshah et al. (1993) used Utah HSIS data to model truck crash involvement on two-lane rural roads. The authors found a negative relationship between two-lane truck crashes and increased shoulder width. Vogt and Bared (1998) developed an accident model for two-lane rural segments and intersections using Minnesota and Washington State HSIS data (1985â1989). The authors found a negative correlation between lane width and shoulder width and injury crashes. Hauer et al. (2004) used a negative multinomial model using Washington State HSIS crash data to predict the non- intersection accident frequency of urban, four-lane, undivided roads (1993â1996). The authors found no correlation between crashes and lane widths. The range of lane widths modeled was 10â12 ft. Moreover, the authors found that roadway seg- ments with two-way left-turn lanes (TWLTL) had fewer off- road crashes. Garber and Ehrhart (2000) considered crash factors for two-lane roadways in Virginia with speed limits of 55 mph. The authors used deterministic models to relate crash rate with mean speed, flow per lane, lane width, and shoulder width. They found that the effect of mean speed, shoulder width, and lane width was negligible. Deng et al. (2006) used an ordered probit model to analyze head-on crashes for 720 two-lane road segments in Connecti- cut (1996â2001). The authors found a positive relationship between narrow roadways and narrow road segments. Zhang and Ivan (2005) evaluated the effect of geometric characteristics on head-on crash incidents for two-lane roads in Connecticut. The authors used negative binomial general- ized linear models to evaluate the effects of roadway geomet- ric features on incidents of head-on crashes for 655 segments using 1996â2001 crash data. They found a correlation between horizontal and vertical curvature but indicated that lane and shoulder width were not significant. Zegeer et al. (1992) evaluated the safety effects of geometric improvements for 10,900 horizontal curves on two-lane roads
12Authors Data Set Shoulder Lane Other Miaou and Lum, 1993 Mohamedshah et al., 1993 Vogt and Bared, 1998 Hauer et al., 2004 Garber and Ehrhart, 2000 Deng et al., 2006 Zhang and Ivan, 2005 Zegeer et al., 1992 Heimbach et al., 1974 Sosslau et al., 1978 Zegeer et al., 1981 Abboud and Bowman, 2001 Souleyrette, 2001 Table 2.1. Summary of Literature for Roadway Cross Section UT HSIS data: Heavy truck crashes UT HSIS data: Heavy truck crashes MN and WA HSIS data: Rural two-lane WA HSIS data: Urban four- lane undivided VA data: Two-lane roads CT: Head-on crashes on two- lane roads CT: Head-on crashes on two- lane roads WA: Horizontal curves on two-lane roads Two-lane highways KY: State primary, secondary, and two-lane roads AL: 2- and 4-foot paved shoul- ders on two-lane roads IA: Rural two-lane, four-lane, expressways Negative correlation: Inside paved shoulder width and truck crashes Negative correlation: Width and truck crashes Negative correlation: Width and crashes No correlation: Width and crash rate No correlation: Width and crashes Lower crash rate for paved than for unpaved section Negative correlation: Width and crash rate Decrease in ROR, head-on, and opposite direction side- swipe crashes for gravel or paved shoulder width increase from 0 to 9 ft No correlation: Paved shoulder and crashes Could not detect relationship Negative correlation: Width and crashes No correlation: Width and crashes No correlation: Width and crash rate Positive correlation: Narrow roadway and crashes No correlation: Lane width and crashes Correlation: Superelevation deficiencies and crashes in Washington State. The authors found a statistical relation- ship between crash occurrence for sharper curves, narrower curve widths, locations with lack of spiral transitions, and increased super-elevation deficiencies. Heimbach et al. (1974) found that rural, two-lane high- ways with paved shoulders had a significantly lower crash rate than those with unstable shoulders. Sosslau et al. (1978) found that paved shoulders exhibit safety benefits. This NCHRP report concluded that roads with paved shoulders have lower crash rates than unpaved shoul- ders of the same width. The report also concluded that shoul- der widths, paved or unpaved, have a greater effect on crash rates than lane widths. A linear model was developed to pre- dict crash rates for roadways with varying lane and paved shoulder widths. The model was generally able to represent predicted relationships, but there were some inconsistencies.In general, crash rates decreased as shoulder widths increased. This rule applied for sections of roadways with three degrees or less of horizontal curvature, but the opposite result was true for roadways with an average daily traffic (ADT) of fewer than 1,000 vehicles per day (VPD) or more than 5,000 VPD. Zegeer et al. (1981) conducted a comparative analysis study of state primary, state secondary, and rural, two-lane roads in Kentucky. The sections were selected so that they did not include any major intersections. A database of 15,944 miles of road was compiled from computer tape, and eight classifi- cations of roads based on ADT were used. Because about 70% of the total sample had no shoulder, shoulders were defined as paved or densely graded. Grass and soil were not considered shoulders because they are not suitable for driving. Zegeer et al. found that ROR, head-on, and opposite-direction side- swipe crash rates decreased as shoulder width increased from
130 to 9 ft, but the crash rates increased slightly for shoulders from 10 to 12 ft wide. Crash severity, however, did not decrease with wider shoulders. Zegeer et al.âs results indicated that it is economically beneficial for roadways with lane widths greater than 10 ft to widen the shoulders if there are at least five ROR or opposite-direction crashes in one year. For roads without shoulders, the optimal shoulder width to install was found to be 5 ft. Not all studies have concluded that paved shoulders offer a significant benefit, however. Abboud and Bowman (2001) evaluated 2- and 4-ft paved shoulders on two-lane highways in Alabama and compared them against county statistics for the expected number of crashes on the treated segments. Crash records were not kept on specific routes with similar charac- teristics; therefore, total county crashes in the before and after periods were used as a control. Crash frequency by type and severity was analyzed, but no statistically significant differ- ences were found at the 0.05 alpha confidence level. Similarly, a study conducted by Souleyrette (2001) was unable to present significant results. Souleyretteâs study focused specifically on rural, two-lane and rural, four-lane, divided, noninterstate freeways in Iowa. Only targeted crashes were considered for this study. Intersection, median, and roadway crashes were excluded because they were assumed to be non- shoulder related. Limited data availability because of conser- vative shoulder construction practices in Iowa prevented statistical significance from being obtained with any of the results. Trends of reduced crash rates were noted but could not be verified with confidence. Shankar et al. (2004) used a zero-inflated negative binomial model to consider the interaction among design, traffic, and weather on roadside crashes using 318 segments. The authors found that weather and design factors play a statistically sig- nificant role in roadside crash occurrence. The authors also found that shoulder width and presence of a divided median were related to crash occurrence. In another study, Shankar et al. (1998) analyzed 275 sections of divided state highways and found that median width was a statistically significant factor in crash history. Driveway Density Vogt and Bared (1998) developed an accident model for two- lane rural segments and intersections using Minnesota and Washington State HSIS data (1985â1989). The authors found a positive correlation between driveway density and injury crashes. Deng et al. (2006) used an ordered probit model to analyze head-on crashes for 720 two-lane road segments in Connecti- cut (1996â2001). Among other factors, the authors found that nighttime crashes and density of access points were sig- nificantly related to more severe crashes.Roadway Lighting A number of studies have demonstrated that nighttime crash rates are significantly higher than daytime crash rates and that lighting can play a positive role in reducing nighttime crashes. Deng et al. (2006) used an ordered probit model to analyze head-on crashes for 720 two-lane road segments in Connecti- cut (1996â2001). Among other factors, the authors found that nighttime crashes and density of access points were signifi- cantly related to more severe crashes. A before-and-after study of lighting along a five-lane road- way in Chicago from 1952 to 1958 was reported by Lewin et al. (2003), who found a reduction of 48% in fatal night crashes. Billion and Parson (1962) compared crashes on 6 miles of unlighted and 6 miles of lighted major routes with mountable medians. The night/day crash rate per million miles was 1.5 times higher for unlighted sections than for lighted. Another study in Illinois compared the night crash rate before and after a major traffic route was lighted. A night crash reduction of 36% was recorded (Box, 1989). A New York study com- pared lighted and unlighted major and collector streets. The study reported that streets with little or no illumination had substantially higher nightâday crash ratios (Box, 1972). Elvik (1995) conducted a meta-analysis of 37 published studies, reported from 1948 to 1989 in 11 countries, that eval- uated the safety effects of lighting. Analysis of the studies indi- cated roughly a 65% reduction in nighttime fatal accidents, a 30% reduction in injury accidents, and a 15% reduction in property-damage-only (PDO) accidents for both intersections and roadway segments on rural, urban, and freeway facilities when lighting was installed. The effect of installing lighting was greater at intersections than at nonintersections; similar results were found for rural, urban, and freeway environments. A comparative study in the Netherlands reported that the ratio of night/day crashes for unlighted rural freeway routes was 28% greater than for lighted routes (International Commission on Illumination, 1992). A recent Iowa State University/Center for Transportation Research and Education (ISU/CTRE) study evaluated rural expressway safety. The researchers did not eval- uate lighting per se but evaluated other safety aspects of rural expressways, such as variation in medians and older driver issues, which may be of interest in evaluating potential safety benefits of lighting (Maze and Burchett, 2004). Two studies were found that evaluated lighting on rural pri- mary routes. Sabey and Johnson (1973) evaluated 43 sites on trunk highways before and after lighting. The authors found a statistically significant reduction (50%) in crashes for 19 of the roads that were high-speed (70+ mph) segments. They found no statistically significant reduction for lower-speed segments. Another study by Cornel and MacKay found no statistical dif- ference in night and serious crash frequencies before and after lighting was installed on rural highways (FHWA, 1982).
14Rumble Strips Rumble strips have also been found to reduce lane departure crashes. Table 2.2 summarizes the results of the literature review.Authors Data Set Shoulder/Edgeline Rumble Strips Centerline Rumble Strips Garder and Davies, 2006 Smith and Ivan, 2005 Corkle et al., 2001 Perrillo, 1998 Hickey, 1997 Miles et al., 2005 Persaud et al., 2004 Russell and Rys, 2005 Kohinoor and Weeks, 2009 Outcalt, 2001 Table 2.2. Summary of Literature for Rumble Strips MN CT: State highways Summarized 8 studies NY: State highways PA turnpike TX: Two-lane roadway 7 states: Rural two-lane Summarized other studies AZ: Arterials, minor arterials, and collectors CO: Two-lane Crash reduction: Overall, 27%, sleep- related ROR, 58%; dry-road ROR, 43% Crash reduction: SV, fixed object, 33%; ROR, 48.5% Crash reduction: ROR, 20 to 72% Crash reduction: Overall, 65% to 70% Crash reduction: ROR, 70%; drift-off- road, 60% Reduced shoulder encroachments, 46.7% Crash reduction: All injury, 14%; front and opposing-direction sideswipe injury crashes, 25% Crash reduction: Injury, 15%; head-on and opposing-direction injury, 25% Crash reduction: Fatal and serious injury crashes, 61% Crash reduction: Head-on, 34%; opposite sideswipe crashes, 36.5%Hanley et al. (2000) evaluated four crash-reduction factors currently used by the California Department of Transporta- tion (Caltrans), including rumble strip installation, defined as any construction for which a laterally positioned rumble strip had been installed. In most cases, the researchers indicated that some shoulder widening occurred as well. They found statistically significant accident-reduction factors with rumble strip installation. Garder and Davies (2006) evaluated the effectiveness of con- tinuous shoulder rumble strips (CSRS) on reducing crashes on rural interstates in Maine. The authors found that the pres- ence of CSRS reduced crashes overall by 27%, reduced sleep- related ROR crashes by about 58%, and reduced dry-road ROR crashes by about 43%. They also found that fatal crashes were reduced more than other crashes. Smith and Ivan (2005) evaluated the crash reduction result- ing from milled-in shoulder rumble strips on limited-access highways within a 3-year period before and after installation on sections of 20 freeways, including some sections without rumble strips. The authors found that shoulder rumble strips overall reduced single-vehicle, fixed-object crashes by 33%. They indicated that crashes were reduced by as much as48.5% within interchange areas and by as little as 12.8% on sections where the speed limit was less than 65 mph. They also found that crashes increased in areas where rumble strips were not installed. Corkle et al. (2001) summarized eight research studies on edgeline rumble strips (ERS) and found that ROR crashes were reduced by 20% to 72%. The New York State Department of Transportation (NYS- DOT) began installing continuous shoulder rumble strips in 1993. It began to include continuous shoulder rumble strips with its regular construction and as site-specific projects on existing roadways. The New York State Thruway Authority (NYSTA), which owns and operates private toll roads, installed continuous shoulder rumble strips between 1992 and 1996. The advantage of the data drawn from the NYSTA installations was uniformity, because the data were recorded by a dedicated troop of the state police force and there were a limited number of miles from which to collect data. Both agencies had a lim- ited amount of before-and-after data, so statistical significance was not tested, but both agencies found a reduction in crashes of 65% to 70%. It should be noted, however, that some obser- vations were made during years that included construction of a â[non-]significant percentageâ of continuous shoulder rum- ble strips (Perrillo, 1998). Rumble strips were installed on 80% of the Pennsylvania Turnpike between 1989 and 1994. Early results after the first five projects were completed found a 70% reduction in ROR
15crashes. After speculation of regression to the mean and other factors affecting the results, a follow-up study was conducted. The study included all reportable accidents from 1990 to 1995 and found a slightly more modest result of a 60% reduction in drift-off-road (DOR) crashes (Hickey, 1997). These results, however, were not tested for statistical significance. A preliminary study (Miles et al., 2005) was conducted to determine the extent of the benefits received by ERS. The study was conducted on a two-lane road in Texas with an 11-ft travel lane in each direction separated by a 4-ft-wide center segment with centerline pavement markings. Before-and-after data were collected along this 5-mi segment of road between September 10 and 22, 2004, and between November 5 and 17, 2004, respec- tively (Miles et al., 2005). The geometry of the roadway lim- its the applicability of the findings to a typical two-lane rural road and the brief study time period limits the conclusiveness of the results, but the study still provides an interesting insight into the operational effects of ERS. The study by Miles et al. (2005) used rumble strips that were 12 in. wide; 4 in. was on the marked edge line and 8 in. was on shoulder pavement. Pneumatic road tubes were used to collect volume, speed, and lateral position data. Video footage was also collected to classify the shoulder encroach- ment maneuvers and determine if the ERS caused any erratic maneuvers by drivers. A total of 2,985 shoulder encroach- ments were observed during the 13 days of before-installation footage and the 13 days of after-installation footage. No erratic maneuvers were observed in the video data. Statistical t-tests were performed on the data to determine significance at the 95% confidence level for any changes in driver behavior (Miles et al., 2005). The data from Miles et al. (2005) revealed an overall reduc- tion in shoulder encroachments during the after period of 46.7%. When broken down by encroachment type, the âotherâ case experienced the greatest proportional decrease in shoulder encroachments. The âotherâ case included âinadvertent contact with the edge line because of natural lane shifting, driver inat- tention or fatigue, swaying motions of trailers, or large load width.â Encroachments classified as âotherâ are categorized as one of four types ranging from âright tires hit,â where only the right tires contact the rumble strips, to âaround,â where both sets of tires completely cross over the rumble strips (Miles et al., 2005). While the number of encroachments decreased, the lat- eral position of vehicles increased in distance beyond the edge line. This was not statistically significant, however, and standard deviations were large. Persaud et al. (2004), using empirical Bayes, analyzed about 98 treatment sites (210 mi) on rural, two-lane roadways in seven states before and after installation of centerline rumble strips. The authors found a 14% reduction for all injury crashes combined (at a 95% confidence level), a 25% reduction for front- and opposing-direction sideswipe injury crashes (at a95% confidence level after installation), and an overall reduc- tion in crashes of 12% (at a 95% level of significance). Russell and Rys (2005) summarized the results of several studies and suggested that the use of centerline rumble strips reduced overall injury crashes by 15% and reduced head-on and opposing-direction crashes involving injury by 25%. Roadway Delineation and Signing Sun et al. (2007) investigated the distribution of vehicle lateral position before and after implementation of edgeline mark- ings on seven tangent and three curve sections of two-lane roads with less than 22-ft pavement widths in Louisiana. The authors found that after implementation of the edge lines, vehicles were more likely to move away from the pavement edge. They also found that centerline crossings increased at several sites during the daytime but decreased at night. Donnell and Mason (2006) evaluated the operation effect of wider edge lines along curves (8 in. vs. 4 in.) in Pennsylva- nia. The authors compared differences in several operational metrics, including change in mean speed, lateral placement, encroachment frequency, and vehicle position in the travel lane. Results indicated that wider edge lines did not change the encroachment proportion, mean speed, or lateral position along curves. Tsyganov et al. (2006) compared crash statistics for rural, two-lane highways in Texas with and without edge lines on roadways with 9-, 10-, and 11-ft travel lanes with shoulder widths less than 4 ft using crashes from 1998â2001. On sections with two or more accidents, highways without edge lines had an 8% higher mean accident ratio than similar sections with edge lines. The authors also found an increase in crash fre- quency with lane-width reductions on sections without edge lines, but not on roadways with edge lines. On curved seg- ments, highways without edge lines had a 25.8% higher crash frequency than those with edge lines. Pavement Edge Drop-off Evaluating fatal crashes in Georgia in 1997, Dixon (2005) ran- domly selected 150 two-lane rural fatal crashes on state and nonstate system roads. She estimated that in 38 of the 69 (55%) nonstate system fatal crashes, edge rutting or edge drop-off was present. The author also determined that edge drop-off appeared to be one of the crash causal factors for 21 of the 38 (55%) sites where there was drop-off. The study indicated that drop-off was from 2.5 to 5.0 in. on the rural highway edges. The Federal Highway Administration (FHWA) estimated that approximately 160 fatalities and 11,000 injuries result from crashes related to edge drop-off each year in the United States (FHWA, 2004). Although a quantitative relationship between pavement edge drop-off and safety has not been
16derived, the U.S. Department of Transportation (USDOT) has suggested that a drop-off of 3 in. or more of vertical differ- ential is considered unsafe (FHWA, 2004). AASHTO (1996) suggested that no vertical differential greater than 2 in. should occur between lanes. A study by Humphreys and Parham (1994) found that vertical drop-offs of 4 in. or more between the roadway surface and adjacent shoulder were unsafe. Zimmer and Ivey (1982) also showed that safety was related to pavement edge shape. Hallmark et al. (2006) reviewed crash reports for Iowa and Missouri to determine whether crashes were related to pave- ment edge drop-off. The authors found that approximately 18% of rural ROR crashes in Iowa were potentially edge drop- off related. They also found that crashes that were potentially edge drop-off related were more likely to result in a fatal or major injury crash than other rural ROR crashes. In Missouri, the authors found that approximately 23% of rural ROR crashes were potentially pavement edge drop-off related. They also found a relationship between crashes and an edge drop- off of 2.5 in. or more. McLaughlin et al. (2009) evaluated 122 ROR events using the VTTI 100-car NDS data. The authors reported that change in lane boundaries was involved in 22% of the events. This factor included start of median, narrowing of lane, lane drop, or unusual roadway geometry. Environmental Factors Maze et al. (2006) used the 2005 FARS data to evaluate the impact of weather. They found that pavement condition is listed as rain, snow, or ice for only 12% of fatal crashes. How- ever, as noted, this value does not represent the scope of the problem, because rain-, snow-, or ice-related incidents are only present during a small amount of driving time. For instance, in Iowa approximately 21% of crashes are winter-weather related. The amount of time snow or ice is present is significantly less than 21%. Additionally, fatal crash frequency during the win- ter in rural Iowa when pavement conditions are snowy or icy is about twice the fatal crash frequency when alcohol is a con- tributing factor. Deng et al. (2006) used an ordered probit model to analyze head-on crashes for 720 two-lane road segments in Connecti- cut (1996â2001). Among other factors, the authors found a positive relationship between wet roadway surface and crashes. Shankar et al. (2004) used a zero-inflated negative bino- mial model to consider the interaction between design, traf- fic, and weather on roadside crashes using 318 segments. The authors found that weather plays a statistically significant role in roadside crash occurrence and contributes to 19.3% of the likelihood of crash occurrence, while the weather and design interactions contribute around 6% to the likelihood of crash occurrence. Their results indicated that the presence of pre- cipitation in the fall was positively correlated, and the presenceof precipitation in the spring was negatively correlated with crash occurrence. The authors also indicated that average monthly snowfall exceeding 4 in. and the interaction between snow depth and horizontal curves had a statistically significant effect on roadside crash frequency. McLaughlin et al. (2009) evaluated 122 ROR events using the VTTI 100-car NDS data. They reported the following among their findings: ⢠A ROR event is 2.5 times more likely to occur on dark unlighted roads than during daylight conditions; ⢠It is 1.8 times more likely on wet roads than dry; ⢠It is 7 times more likely on roads with snow or ice than on dry roads; and ⢠It is 2.5 times more likely to occur during the presence of precipitation (fog, mist, rain) than during clear conditions. Vehicle Variables Vehicle type is relevant because rollover incidents may result in more serious outcomes for a ROR crash. Pickup trucks and sport-utility vehicles have a higher center of gravity, which may result in a different outcome for the same initial sequence of events during a road departure. Little information was found about which specific vehicle factors are related to ROR crashes. It is generally accepted that sport-utility vehicles and pickup trucks are more prone to rollover. However, little information was found that describes specific vehicle charac- teristics in relation to lane departure risk. For naturalistic driving studies, most vehicle variables can be collected up front when the instrumentation packages are installed. It is important to provide representative distribu- tion of vehicle types for the full-scale NDS. Driver Factors General Spainhour et al. (2005) evaluated fatal crashes in Florida involv- ing heavy trucks. The authors found that human factors were the primary contributing factor for 94% of the crashes, with the most common factors being alcohol/drug use, inattention, and decision errors. Dissanayake (2003) used logistic regression to identify influ- ential factors in young-driver (16 to 25 years old), single- vehicle ROR crashes. The author used crash data from 1997 and 1998 from police-reported crashes in Florida. Influence of alcohol or drugs, existence of a curve or grade, and vehicle speed significantly increased the probability of having a more severe ROR crash. McGinnis et al. (2001) analyzed FARS and National Per- sonal Transportation Studies data for ROR fatal crashes from
171975 to 1997. The authors evaluated how trends changed over time and found that young drivers, male drivers, drivers over the age of 70, drivers in utility vehicles, and drivers using alco- hol had higher involvements in fatal ROR crashes. Khattak and Hummer (1998) analyzed crashes on two-lane rural roadways from two counties in North Carolina. The authors indicated that consumption of alcohol, roadway surface condition, and horizontal alignment appeared to influence the occurrence of ROR crashes. McLaughlin et al. (2009) evaluated 122 ROR events using the VTTI 100-car NDS data. The authors reported that in 40% of the events, the most common contributing factor was distraction/inattention. The most common distractor (90%) was secondary task distraction, which included use of a cell phone or dialing a cell phone, talking to or looking at passen- gers, or devoting attention to in-vehicle devices. Younger Drivers Ulmer et al. (1997) examined the National Highway Traffic Safety Administration (NHTSA) General Estimate System (1993) for 16-year-old drivers and reported that 16-year- old drivers were more likely than other drivers to be involved in single-vehicle crashes and in crashes from 6:00 p.m. to 12:00 a.m. These teen drivers were also more likely to be accompanied by other teen passengers than were 17-, 18-, or 19-year-olds. Williams et al. (1997) evaluated fatal crash involvement among 15-year-old drivers in states that required a learnerâs permit for 15-year-olds and found that crashes involving 15- year-old drivers were usually single-vehicle crashes, occurred late at night (between 12:00 a.m. and 6:00 a.m.), and had a number of passengers in the car. Driving factors that con- tributed to 15-year-old-driver fatal crashes included speeding and failure to drive in the proper lane. A University of North Carolina study (Highway Safety Research Center, 2000) found that 80% of 16-year-old-driver nighttime crashes occurred between the hours of 9:00 p.m. and 12:00 a.m. and 73% of 17-year-old driver nighttime crashes occurred from 9:00 p.m. to 12:00 a.m. The crash risk for 16- and 17-year-old drivers was nearly three times greater between 10:00 p.m. and 12:00 a.m. than during the daylight hours. Based on the study, the risk per mile driven is even greater after midnight because most of the nighttime vehi- cle miles traveled (VMT) by 16- and 17-year-olds occurred before midnight. Ulmer et al. (1997) examined NHTSAâs General Estimates System for 16-year-old drivers and found that 16-year-olds were more likely than other drivers to be involved in crashes from 6:00 p.m. to 12:00 a.m. Williams et al. (1997) evaluated fatal crash involvement for 15- and 16-year-olds and found that fatal crashes for 15-year-olds were more likely to occur between 12:00 a.m. and 6:00 a.m.Rice et al. (2004) evaluated how nighttime driving affected injury crash rates for young drivers in California before implementation of a graduated driverâs license (GDL) in 1998 and found that crash risk increased after 10:00 p.m. Adolescent impulsiveness is a natural behavior, but it results in poor driving judgment and participation in high-risk behav- iors, such as speeding, inattention, drinking and driving, and not using a seat belt. Peer pressure also often encourages risk taking (Chein et al., 2011). According to NHTSA, risk tak- ing among adolescents appears to be a critical factor in explain- ing the high number of crashes. For example, younger drivers tend to accept narrower gaps when pulling out into traffic. They also have been observed to have shorter following dis- tances and to drive faster (Ferguson, 2003). Williams (2001) reported on a study that indicated that for teenage drivers the presence of one passenger nearly doubles the fatal crash risk compared with driving alone. In another study, the fatal crash risk with two or more passengers was found to be five times as high as driving alone. There was excess risk for young drivers with passengers during both day and night hours (Williams, 2001). Another study indicated that the crash risk when three or more passengers were present was about four times greater than when driving alone (NHTSA, 2005b). The increased crash risk existed for both daytime and nighttime crashes, although overall crash risk was much higher at night. In one study, death rates from 10:00 p.m. to 6:00 a.m. were 1.74 times higher with passengers than without pas- sengers. During the daytime, rates were 1.77 times higher (Williams, 2003). More teen fatal crashes occurred when pas- sengers, usually other teenagers, were in the car than when no passengers were in the car. Two out of three teens who die as passengers are in vehicles driven by other teenagers (Williams, 2003). Summarized List of Factors Table 2.3 summarizes roadway, environmental, vehicle, and driver factors that have been identified in the literature or through team expertise as being relevant to the occur- rence and severity of lane departure crashes. Other factors that may be necessary to analyze lane departures using NDS data, such as factors to position a vehicle in respect to the roadway, factors to identify potential lane departures (triggers), or factors relevant to crash surrogates, are not identified.Research Questions One of the goals of this project was to develop a set of research questions that could be explored using existing NDS data. The intent was to then determine which research questions could adequately be addressed given the data and the limitations of
18Roadway Factors Horizontal curves Vertical curves Roadway cross section Signing Speed limit Roadway delineation Roadway defects Other Clear zone Countermeasures not included in other roadway factors (e.g., paved shoulders) Environmental Factors Pavement surface condition Ambient conditions Vehicle Factors Vehicle characteristics Kinematic Driver Factors General Condition Teen-specific factors Substance use Distractions Table 2.3. Factors Contributing to Occurrence and Severity of Lane Departure Crashes Length Spirals Relationship to other curves Length Relationship to other curves Lane width Cross slope Median type and width Presence and type Posted speed limit Presence and quality of pavement markings Pavement edge drop-off Surface friction Driveway density Type and location of objects within clear zone Presence of object delineators Edgeline and center rumble strips Additional delineation, such as channelizers, raised pavement markings Presence of snow, ice, rain, debris Time of day Precipitation Size Width Advanced technologies (e.g., lane departure warning system, OnStar) Speed Age Driving experience Reaction time Fatigue Emotional state Presence of passengers Alcohol Prescription drugs Type of distraction Level of engagement in distraction Radius or degree of curve Superelevation Grade Surface type Shoulder type and width Advisory speed limits Presence and type of overhead street lighting Surface irregularities Road debris Sight distance Slope beyond edge of shoulder Guardrail, barriers Speed feedback signs Cable median barrier Amount of snow, ice, rain Temperature Visibility (precipitation, fog, smog, dust) Type (e.g., SUV, van) Center of gravity Braking capabilities Acceleration Gender Aggressiveness Medical condition Driver licensing requirements Illegal drugs Duration of distraction
19the SHRP 2 full-scale NDS. Three sets of questions are pre- sented in the following sections: 1. The first set of questions includes those that were addressed in this report. These research questions reflect a need for information that would set the stage to answer research questions in the full-scale study. These questions could also be explored further in the full-scale NDS. 2. The second set of questions includes those identified through this research as being feasible for the full-scale study. Examples are also provided of more specific ques- tions within those categories that the team feels can be realistically addressed, given what was learned during this research and what is known about the data expected from the full-scale NDS. 3. The third set of questions includes specific research ques- tions that the team has determined cannot be realistically addressed given the data expected to be available from the full-scale study and given the review of existing NDS data. The information that supports the research teamâs best estimate about what can or cannot be answered for the full- scale study is based on information provided in the following chapters. However, the research questions are placed in this section because they provide an overview of what follows in the rest of the report. In order to identify lane departure research questions, the team first identified which driver, vehicle, roadway, and envi- ronmental factors were likely to contribute to the occurrence and severity of lane departure crashes, based on an in-depth literature review, as described in the section âRelevant Data Elements Identified in Existing Literatureâ (p. 10), and on the teamâs expertise in lane departure issues. The team then reviewed data from existing NDS from VTTI and UMTRI and evaluated the feasibility of extracting from these various vehicle, roadway, driver, and environmental factors. This information is provided in Appendices A and B. Chapter 4 summarizes data elements that are expected to be necessary to answer the research questions, reviews the roadway data elements identified by SHRP 2 Safety Project S03, Roadway Measurement System Evaluation, and reviews what is expected to be available from the instrumented vehicle study based on a review of information from Safety Project S05, Design of the In-Vehicle Driving Behavior and Crash Risk Study. The abil- ity to extract data from existing NDS was also explored and summarized, as this ability relates to the full-scale NDS. Lane Departure Research Questions Addressed in Scope of Research The first set of research questions includes those necessary to set the stage for answering research questions in the full-scale NDS.These questions were explored in this research, and the results are presented in the following sections. These questions may also be further addressed using data from the full-scale NDS. Research Question A-1: What driver, vehicle, roadway, and environmental factors are necessary to answer a range of research questions related to lane departures using NDS and roadway data? Identifying data needs is an important step in determining which lane departure research questions can feasibly be answered. Roadway, driver, environmental, and vehicle fac- tors expected to influence the occurrence and severity of lane departures was summarized using a literature review (see sec- tion âRelevant Data Elements Identified in Existing Litera- ture,â p. 10). Sources for the various data elements were identified based on the most current available information for SHRP 2 Safety Project S07, In-Vehicle Driving Behavior Field Study, and Safety Project S04B, Mobile Data Collection. Existing NDS from UMTRI and VTTI were examined, the data elements necessary to answer lane departure research questions were extracted, and the feasibility of obtaining the data was determined. This information is summarized in Chapter 4. Research Question A-2: What kinematic variables can be used to identify lane departure incidents (e.g., lateral drift, lane departure, near crash)? For instance, a side acceleration of X ft or a roll rate of Y might define a lane departure. This question addresses the need to identify vehicle kine- matic variables that can be used to flag lane departure inci- dents in the full-scale study. A significant amount of data will result, and it will be necessary to determine some method to flag potential incidents in an automated process. The team conducted an exploratory analysis of kinematic variables for normal driving, as well as for left- and right-side lane depar- tures, using the UMTRI and VTTI data sets, as described in Chapter 5. Research Question A-3: What environmental, roadway, driver, or vehicle factors influence whether a vehicle departs its lane? This research question addresses understanding environ- mental, roadway, driver, and vehicle variables that influence the occurrence of lane departures. Lane departures from the UMTRI data set were identified, and factors were extracted from the various corresponding data sets. Several different analyses were conducted using the UMTRI data that demon- strated approaches to answering this research question. The approaches are described in Chapter 6.
20Relevant Lane Departure Research Questions for Full-Scale NDS The section provides general categories of lane departure ques- tions that can be answered using the full-scale study. Exam- ples are also provided of more specific questions within those categories that can be realistically addressed given what was learned during this research and based on what is known about the data expected to result from the full-scale NDS. This task was based on a review of information available as of September 2009, when the first draft of this report was sub- mitted. The team is not aware of any additional information that alters its original assessment as of January 2010, which is the date for the final submission of this report. This set of questions can be addressed by researchers for SHRP 2 Safety Project S08, Analysis of the SHRP 2 Naturalis- tic Driving Study Data. To address these questions, researchers will need data from instrumented vehicles (Safety Project S07), as well as data that will be gathered or collected during the roadway data collection effort (Safety Project S04A, Roadway Information Database Developer, Technical Coordination, and Quality Assurance for Mobile Data Collection, and Safety Project S04B). The determination of what can be addressed is also based on reviewing and extracting variables from the UMTRI and VTTI NDS data. Research Question B-1: What environmental, roadway, driver, or vehicle factors influence whether a vehicle departs its lane? This research question addresses understanding driver, road- way, environmental, and vehicle variables that influence the occurrence of lane departures. More specific research ques- tions that could be answered under this general topic include the following: ⢠How does roadway surface condition affect lane departure frequency? ⢠Are lane departures less likely when pavement markings are highly visible? ⢠Does signing have any impact on frequency of road depar- tures (e.g., large chevrons may make a driver alert to an adverse horizontal curve)? ⢠How does roadway lighting affect driver scanning patterns at night and what is the impact on lane departures? ⢠What curve characteristics influence the likelihood of a lane departure? ⢠What role does distraction play in lane departure frequency? ⢠What is the relationship between speed and lane departures on curves? ⢠How does alcohol consumption influence driver response to changes in roadway geometry, and what is its impact on lane departures?⢠Are drivers of sport-utility vehicles and pickup trucks more likely to engage in aggressive driving behaviors (e.g., speed- ing, overtaking), and what is the impact of such behaviors on likelihood of lane departures? Research Question B-2: What environmental, roadway, driver, or vehicle factors influence lane departure outcome? This research question involves understanding driver, road- way, environmental, and vehicle variables that influence the outcome of a lane departure when one occurs. More specific research questions that could be answered under this general topic include the following: ⢠How do weather conditions affect lane departure outcome? ⢠Are drivers who leave the roadway more likely to recover and safely return to their lane on paved shoulders than on gravel or earth shoulders? How much of an impact does shoulder width have on outcome? ⢠What is the relationship between speed and lane departure outcome? ⢠Are drivers in vehicles without automatic braking systems (ABS) more likely to overcorrect when encountering snow/ice or loose shoulder material than drivers in vehicles with ABS? ⢠How does level of driver forward scanning before a lane departure influence the likelihood of recovery? ⢠How do drivers react when encountering various types of slope beyond the shoulder edge, and how do these reactions affect lane departure outcome? ⢠What factors lead to driver overcorrection, and what is its impact on lane departure outcome? Research Question B-3: What is the impact of lane departure countermeasures on lane departure frequency and outcome? This research question addresses how drivers interact with countermeasures and addresses why countermeasures are or are not effective. More specific research questions that could be answered under this general topic include the following: ⢠Are drivers more likely to lane keep on roadways with edge- line rumble strips? ⢠How likely are drivers to overcorrect or counter-steer away from edgeline rumble strips and potentially encroach into an adjacent lane rather than experience a road departure? ⢠Are drivers who leave the roadway more likely to recover and safely return to their lane on paved shoulders than on gravel or earth shoulders? How much of an impact does shoulder width have on outcome? ⢠Do edgeline or centerline rumble strips have the same impact on distracted drivers as on nondistracted drivers?
21⢠Are drivers more likely to travel at unsafe speeds during win- ter storm events when median cable barriers are present? ⢠Does additional delineation affect driver forward attention on curves? What is the impact of delineation on frequency and outcome of lane departures? Research Question B-4: What is the relationship between lane departure crash surrogates and crashes? One of the main advantages of the full-scale NDS is that it will provide a unique opportunity to develop relationships between crash surrogates and crashes. Agencies frequently are unable to conduct crash analyses to compare the impact of a treatment for a number of years after the treatment is installed, or they have few sites for comparison. As a result, it is often difficult to conduct crash analyses. Understanding the relationships between lane departure crash surrogates and lane departure crashes would provide agencies with opportunities to conduct evaluations sooner. Answering this research question can provide some infor- mation on the potential effectiveness of countermeasures such as edgeline or centerline rumble strips or treatments that reduce speeds on curves. Specific research questions devel- oped under this category may also indicate what factors pos- itively affect the outcome of a lane departure. Relevant Research Questions That Cannot Feasibly Be Addressed in the Full-Scale NDS Several factors expected to be correlated to lane departure crash frequency and severity will not be collected in any data sets available during the full-scale NDS. Other factors may be available, but extraction may be infeasible. The team has deter- mined that certain research questions cannot be addressed or cannot be realistically addressed given the data expected to be available during the full-scale study and based on a review of existing NDS data.Highly relevant research questions related to the occur- rence, frequency, and severity of lane departure crashes that cannot be answered include those concerning the following: ⢠Occurrence or level of alcohol use by the driver (i.e., what is the relationship between blood alcohol level and fre- quency of lane departures?): The instrumented vehicleâs data acquisition system (DAS) is expected to have an alco- hol sensor that will indicate the presence (but not amount) of alcohol in the vehicle and will not be able to isolate the user. Driver alcohol use may be inferred if the driver is the sole occupant. ⢠Occurrence or level of drug use by the driver (i.e., does the driverâs illegal drug use similarly affect the frequency of lane departures?): No sensors are available that will pick up drug use or identify drugs in the driverâs system. ⢠Pavement friction (i.e., what is the relationship between lane departures on curves and pavement surface friction?): Pavement surface friction is unlikely to be collected using the mobile mapping vans. Even if collected, surface friction will change over the course of the full-scale study, depend- ing on such factors as wear and winter maintenance. ⢠Impact of pavement edge drop-off on lane departure out- come: Data on pavement edge drop-off is not likely to be collected by the mobile mapping van because drop-off can change over short periods of time. As a result, recording the presence and amount of drop-off at one time period may not reflect conditions at a future time period. For instance, there may be several inches of drop-off during the time the mobile mapping van collects data, but shoulder maintenance could occur several days later and thus change conditions drastically. ⢠Quantitative measure of rain, snow, and ice on road: The presence of rain, snow, or ice can be determined from the instrumented vehicle forward video or from environmental records, but the amount of precipitation on a given stretch of roadway cannot be measured.