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.
54 A p p e n d i x A The ability to predict the occurrence of crashes on roadways has been a challenge to the transportation profession since the early days of motorized transportation, and a large number of studied have been conducted to relate highway safety to highway design, environmental and driver characteristics. The literature on these studies is extensive. A sample of studies relevant to the current research is briefly reviewed. Highway design Factors There are three primary geometric elements in highway design and, thus, also in highway characteristics which have been studied in relation to crashes. These are cross section, horizontal alignment, and vertical alignment. Note that there is extensive literature relating intersection character- istics and crashes. However, because this study is concerned with lane departure crashes that occur on segments, the review does not include studies concerned with intersection crashes. Cross Section The major elements of cross section include the number and width of lanes, presence and type of median, type and width of shoulders, and roadside features (e.g., side slope, clear zone, placement and types of roadside obstacles). The effects of cross-sectional elements on crash occurrence have been examined in many empirical studies. A classic early study by Schoppert (1957) examined crash occurrence on two-lane rural roads in Oregon and developed descriptive and predictive models using regression analysis. Schoppert found that vehicle crashes were directly related to traffic volume and certain fea- tures of the roadway, including lane and shoulder widths. He found that the crash rate increased with reduced cross-section width but reported that lane and shoulder widths did not serve as good predictors of the number of crashes. Versace (1960), in analyzing Schoppertâs data, recognized that roadway features were correlated with each other; that is, good cross- sectional elements usually go together, and furthermore, good cross-sectional elements were usually found together with good alignment. These, he noted, are the result of road design and construction practices. Versace identified shoulder and lane widths as factors affecting crash occurrence but noted that their effects were not as important as that of traffic vol- ume. Increased crash rates with decreased lane and shoulder widths were also reported by Dart and Mann (1970) and Roy Jorgensen and Associates, Inc. (1978). Kihlberg and Tharp (1968) investigated the relation- ship between motor vehicle crashes and highway design elements using data from five states by analyzing crash counts on homogenous road segments of two- and multi- lane roadways. They found that number of lanes and median affect crash rates. The effect of the median, however, was not very marked and was found in only some of the states examined. Cleveland, Kostyniuk, and Ting (1984, 1985) examined crash data for two-lane rural highway segments from 14 states by using statistical categorical techniques. They confirmed Versaceâs observation that good (or bad) geometric features were usually found together, and grouped cross-sectional fea- tures (lane width, shoulder width, side slope, ditch condition) that were usually found together in roads into a set of âgeo- metric bundlesâ that varied from excellent to poor. An effect of the geometric bundles on crashes was found, but it was not as important as the effects of traffic volume and access density and the interactive effect of the geometric bundles and access density. A study by Zegeer and Deacon (1987) quantified the effects of lane width, shoulder width, and shoulder type on high- way crash experience based on the analysis of data for nearly 5,000 mi of two-lane highway from seven states. The study controlled for many roadway and traffic features, includ- ing roadside hazard, terrain, and traffic volume. Lane width Literature Review: Crash Rates and Highway, Environmental, and Driver Factors
55 Horizontal and Vertical Alignment Elements of horizontal alignment include degree and length of horizontal curve, presence of spiral or other transition curve, and the superelevation. Elements of vertical alignment include vertical lines or grades and vertical curves (sags and crests). In a study of two-lane and multilane rural roads from 15 states, Raff (1953) found that crash rates increased with the degree of curvature. On two-lane roads, the crash rate increased with curve frequency. Crash rates also increased with sight distance restrictions, which are primarily due to crest vertical curves. The study also found that grade alone did not have an effect on crash rates on tangent (straight) sections of road, but there was an increase in crash rates when both grade and horizontal curvature were present. Increased crash rates on combinations of grades and horizontal curves have also been reported by Bitzel (1957). Bitzelâs data was from 25,500 crashes on 1,300 mi of German highways. However, Bitzel found that crash rates increased as grades increased. Kihlberg and Tharp (1968) found effects on crash rates of horizontal curves only for curves of 4 degrees or more and for grades of 4% or more. The effect of horizontal curves on crashes was also investi - gated by Glennon, Neuman, and Leisch (1986). A database that included crash, geometric, and traffic data for two-lane rural highway segments from four states was developed for this study. There were more than 3,000 segments with hori- zontal curves and about 350 control tangent sections. Care was taken to select sites with uniform lane and shoulder conditions and to avoid influences of bridges, intersections, curbs and other nearby horizontal curves. Analysis of covari- ance methods were used to develop a model that related the number of crashes on curves to the traffic volume, degree of curve, and length of curve. Matthews and Barnes (1988) analyzed curve crashes on 2,000 km of highways in New Zealand. They identified prior curvature (total number of curvature in the two km preced- ing the curve where a crash occurred) as having the largest effect on curve crash rates, followed by grades, and radius of curve. They also reported that crash risk was particularly high on short radius curves located at the end of long tan- gents and on steep down grades. Safety effects of horizontal curves on two-lane rural roads were studied by Zegeer et al. (1990), using 5 years of crash data from Washington State and detailed information on horizontal alignment [degree of curve, length of curve, curve direction, central angle, pres- ence of spiral transition, roadside (recovery distance, road- side hazard rating), cross section (lane width, width and type of shoulder), and traffic volume]. In all, there were more than 12,000 crashes with an average of 0.22 crashes per curve. Sta- tistical analysis revealed significantly higher curve crashes for sharper curves, narrower lane width on curves, lack of spiral transitions, and increased superelevation deficiency. and shoulder type and width were found to be related to crash rates and could also be related to crash type. A crash prediction model was developed and used to determine the expected effects of lane and shoulder widening improve- ments on related crashes. A large effort at the Federal Highway Administration (FHWA 1982; Cirillo, Dietz, and Beatty 1969; Cirillo 1970) investigated the effects of geometric and traffic parameters on crashes on the Interstate system. By using data from 24 states, regression models were developed for 19 model categories for various segments of the Interstate freeway system, including interchanges of the mainline roadway. The basic finding of these analyses concerning geometric elements, which included cross section, was that because the geometrics on Interstate roads are generally very good, their variations, when they occur, have little influence on crashes. The Interactive Highway Safety Design Module (IHSDM) is a suite of software analysis tools developed by FHWA to evaluate safety and operational effects of geometric design decisions during the design process (FHWA 2003). The IHSDM consists of modules, among which is the Crash Prediction Module, which estimates the number and severity of crashes that could be expected on specified road segments based on its geometric design and traffic characteristics. The IHSDM includes an algorithm for predicting the crashes on two-lane rural roads (Harwood et al. 2000). The base model provides an estimate of the safety performance on a roadway or inter- section for a set of assumed nominal conditions. The modi- fication factors adjust the base model predictions to account for the effects on safety for roadway segments of various geometric and operational features. For cross-sectional ele- ments, the base conditions are two 12-ft lanes, paved 6-ft shoulders, and a roadside characterized as marginally recov- erable. The adjustment factors vary by traffic volume condi- tions. According to the IHSDM crash prediction algorithm, the effects of lane width and shoulder type and width at low traffic volumes are very limited, but become larger at traffic volumes of more than 2,000 vehicles per day. The IHSDM algorithm also includes the effects of passing lanes on crash rates on two-lane rural roads. Based on the work of Harwood and St. John (1985), the algorithm predicts a reduction of crash rates with the installation of short four- lane sections that allow passing. It is worth noting that the IHSDM algorithm is based on historical accident data, before-and-after studies, as well as expert judgment as the basis for the factor analysis, and is an excellent example of a classical analysis that provides a contrast to the work con- tained in the present project. Perhaps a future extension of the IHSDM algorithm will include further data from natu- ralistic driving, site-based trajectory analysis and analysis of spatially referenced data.
56 mile increased with higher AADT, higher number of lanes, greater access density, higher proportion of time the road surface is wet, and higher traffic speed variation but lower speed and lower skid number (a measure of the surface fric- tion). Shankar, Mannering, and Barfield (1995) found that crash frequency on an Interstate in the Snoqualmie Pass area of Washington State increased with a higher number of hori- zontal curves, higher maximum grades, higher frequency of rainy days, higher maximum daily snowfall in a month, inter- actions of maximum snowfall with grade, and with curves. Overall, the literature shows that crash frequencies are higher in adverse weather conditions because of reduced visibility and reduced road friction. Traffic Volume Relationships between crash occurrence and geometric and operational characteristics of roadways often use a mea- sure of traffic volume as the exposure measure of crash occurrence. However, there is strong empirical evidence of relationships between crash rates and traffic volume, con- ditional on roadway characteristics (e.g., Schoppert 1957; Versace 1960; Cleveland, Kostyniuk, and Ting 1984, 1985; Hall and Pendleton 1989; Stokes and Mutabazi 1996; Garber and Gadiraju 1990). Schoppertâs (1957) study found that crash rates increased with increases in vehicle volume. He also reported that crashes on low-volume roads did not appear to be related to any roadway feature. Versace (1960) found average daily traffic (ADT), which is a measure of traffic vol- ume, to be the variable most highly related to crash occur- rence. Cleveland, Kostyniuk, and Ting (1984, 1985) found the relationship between crashes on road segments and ADT to be nonlinear and the best predictor of crashes on two- lane rural roads. They also found the interactive effects of access point density with ADT to be very important in predicting crashes. In roadways built for high-design speeds, such as freeways, traffic volume appears to be the most important predictor of crashes. Other studies that FHWA studied of crashes on the Interstate system (Cirillo, Dietz, and Beatty, 1969; Cirillo 1970) concluded that the traffic volumes and commercial traffic volumes were the main con- tributors to the explanation of the crashes on the Interstate system of roads. ADT was also found to be the most impor- tant variable in the relationship between traffic crashes and highway geometric design elements and traffic volumes on interchange ramps and speed-change lanes (Bauer and Harwood 1997). The interactive effect of traffic volume on crash occurrence is built into the IHSDM crash prediction algorithm for two- lane rural roads and intersections (Harwood et al. 2000). The effects of each of the design or operational features are given for different levels of ADT. All else being equal, higher traffic volume and longer curves were associated with significantly higher frequency of curve crashes. Federal Highway Administration studies (FHWA 1982; Cirillo, Dietz, and Beatty 1969; Cirillo 1970) of the effects of geometric and traffic parameters on crashes on the Interstate system did not find a significant contribution of horizontal or vertical alignment on crash rates on freeways. However, a study by Dunlap et al. (1978) that examined the effects of horizontal and vertical curves on crash rates on the Pennsylvania and Ohio Turnpikes found no significant relationship between crash rates and grades and horizontal curves in Ohio, but there were increases in crash rates with increasing curvature of horizontal curves in Pennsylvania. The IHSDM crash prediction algorithm for two-lane rural roads (Harwood et al. 2000) includes the effects of horizontal and vertical alignment. The base model provides an estimate of the safety performance on a tangent and flat road segment. The crash rate for long flat curves is only slightly higher than for tangent roadways. However, the crash rate increases with the sharpness and shortness of the curve. Spiral transitions to the curve mitigate the crash rates as does adequate super- elevation. The algorithm also predicts an increase in crash rate for increases in grades at steeper grades. environmental Factors For road departure crashes the main factors of additional interest are weather and traffic volumes. Of course other fac- tors such as the presence of construction zones, access den- sity, and intersection design all have a strong influence on crash rates and are of great general interest. Because the pres- ent focus is on road departure crashes, those factors will not be reviewed here. Weather Weather constitutes a set of environmental factors that can influence crash occurrence by increasing crash risk. Empiri- cal evidence suggests that a wet road surface increases crash frequency (Jones, Janssen, and Mannering 1991) and that truck-involved freeway collisions increase on wet and icy road surfaces (Golob and Recker 1987). Many studies have investigated the impacts of adverse weather and road geom- etry on crashes (Khattak, Kantor, and Council 1998; Ivey et al. 1981; Jovanis and Delleur 1981; Snyder 1974; Brodsky and Hakkert 1988; Shankar, Mannering, and Barfield 1995). Satterwaitte (1976) analyzing California data, found a ratio of the number of crashes during 24 h when almost all crashes occurred in wet conditions to the number of crashes occur- ring in dry conditions to be 2.23 times. A study on Texas road- ways (Ivey et al. 1981) found that wet crash frequency per
57 and Moskowitz 2001; Charlton et al. 2004; Walsh et al. 2004). A study of driving errors by 533 older drivers (mean age 76.2 years), whose competence was in question (Di Stefano and Macdonald 2003) found that their most frequent errors were failure to turn their heads to look behind, nonuse of mir- rors, problems maintaining lane position and lane keeping, and keeping up with traffic. Driving performance problems and elevated crash risk among young drivers have been related to inexperience, lack of knowledge about factors such as driving situations, vehicle handling, problem solving, and decision-making strategies (Eby and Molnar 1998). Young drivers also tend to perceive less risk in driving than do older drivers and are poorer at identifying hazards when driving (Eby and Molnar 1998; Jonah 1997). Sensation seeking among young drivers has also been linked to unsafe driving behaviors and purposeful viola- tion of traffic laws and regulations, crashes, and violations (Eby and Molnar 1998; Jonah 1997). High levels of sensation seek- ing [as measured on the sensation-seeking scale (Zuckerman 1994)] have been associated with drinking and driving in young driver populations. The relationship was much weaker for female drivers than male drivers, and appears to decline with age (Jonah 1997). There are also other gender effects on the driving task and crash risk. Men and women differ in cognitive abilities (Halpern 1992), risk taking (Jonah 1997), and automobile crash rates (NHTSA 2006b). These differences suggest that men and women may also differ in driving performance in a given situ- ation. Studies that have examined visual aspects of attention demand find that female drivers require more time viewing than do male drivers for the same situation (Courage, Milgram, and Smiley 2000; Tsimhoni and Green 1999; Tsimhoni and Green 2001; Tsimhoni, Yoo, and Green 1999). infrastructure and Highway Features Horizontal curves and lane width are two roadway features associated with road departure crash risk. As such, human factors research includes many studies of visual demand (the amount of time drivers look at road) associated with these road features. Results of these studies indicate that curved roads require more visual demand than straight sections. Visual demand increased with decreasing lane width (e.g., Courage, Milgram, and Smiley 2000; Senders et al. 1967; Van der Horst and Godthelp 1989). Age Whether considered on a per mile basis or a per population basis, crash rates vary as a function of age (NHTSA 2006a). There is clear evidence of age effects in selective, divided, driver Factors The driving task is also affected by the driverâs characteristics and behaviors. The SHRP 2 Safety program has a strong focus on the safety impacts of human behavior, especially driver error and the interaction with relevant highway and other factors. Human factors research focuses on the specific ways in which the human characteristics and behaviors influence the driving process. This is useful in understanding the basic mechanisms of how, for example, highway factors interact with human factors in determining crash risk in any particular situation. To help understand how human behaviors affect crash risk, Wierwille et al. (2002) proposed the following taxonomy of four major factors degrading driving performance: 1. Inadequate knowledge, training, and skill: A lack of under- standing or misunderstanding of traffic laws, vehicle kinematics, driving techniques, or driver capabilities and limitations. 2. Impairment: Fatigue and drowsiness; use of alcohol and illegal drugs; illness; lack of or incorrect use of medication; disability or uncorrected disability. 3. Willful inappropriate behavior: Purposeful violation of traffic laws, regulation; aggressive driving; use of vehicle for intimidation or as a weapon. 4. Infrastructure, environmental problems: Traffic control device related; roadway related; alignment, sight distance, delineation; weather, visibility related. State and national crash databases include some information on drivers including age, sex, and some contain information on driver impairment and distraction. For example, the National Automotive Sampling System (NASS) Crashworthiness Data System (CDS) contains variables on police-reported alcohol presence and distraction. Analysis of these data shows that in 2001, police noted alcohol presence for about 5% of all crash- involved drivers in the United States; 25% of crash-involved drivers were distracted in some way, including 5% who were sleepy or asleep. Among drivers involved in single-vehicle road departure crashes, alcohol presence was noted for 18%; 40% were distracted in some way, including 21% who were sleepy or asleep (Eby and Kostyniuk 2004b). Crash risk varies with age. Whether considered on a per mile basis or a per population basis, crash rates vary as a function of age (NHTSA 2006a). Safe and efficient driving requires the adequate functioning of a range of abilities including vision, perception, cognitive functioning and physical abilities, which tend to decline with age; the loss of efficiency in any of these functions can reduce performance and increase risk on the road (Janke 1994; Marottoli et al. 1998; Stutts et al. 1998; Wood 2002; Oxley et al. 2006). Chronic disease and medica- tions also affect driving abilities and crash risk (e.g., Wilkinson
58 and sustained attention (Comalli, Wapner, and Werner 1962; Parasuraman and Greenwood 1998; Sexton and Geffen 1979). Tsimhoni, Yoo, and Green (1999) and Tsimhoni and Green (2001) assessed the differential effects of young drivers (aged 21 to 28 years) and older drivers (aged 66 to 73 years) on visual demand (the proportion of time that a driver views the roadway over a segment of interest). Using the visual occlu- sion method in a driving simulator, they found that older drivers had significantly higher visual demand for straight roadway sections and three curves of different radii. In a simi- lar study, Tsimhoni and Green (1999) investigated differences in visual demand among three age groups (18 to 24; 35 to 54; 55 and older). Again, subjects drove both straight and curved sections of roadway. These researchers found increased visual demand by age group for all curve radii studied. On straight roadways, however, visual demand was significantly different only for the oldest age group. Thus, these studies show that there is a clear interaction among age, roadway type, and visual demand, and that for a given driving situation, older drivers will experience a greater visual demand than younger drivers. Gender Men and women differ in cognitive abilities (Halpern 1992), risk taking (Jonah 1997), and automobile crash rates (NHTSA 2006b). These differences suggest that men and women may also differ in driving performance in a given situation. Several studies have addressed the visual aspect of this issue (Cour- age, Milgram, and Smiley 2000; Tsimhoni and Green 1999; Tsimhoni and Green 2001; Tsimhoni, Yoo, and Green 1999). Generally these studies find that female drivers require more time viewing the road in a given situation than male drivers. For example, Courage, Milgram, and Smiley (2000) had sub- jects drive straight and curved roadways that varied in width. Over all conditions, they found that women required 8% more time viewing the road than did men. This significant difference was of the same magnitude as the effect of lane width found in the same study. driving Speed Driving speed is not a human factor, but rather a human response to the situation urgency of the trip, to conditions on the road, and to conditions within the vehicle. It is a trade-off between reduced journey time and risk of crash or conviction (or other social impact), as well as a result of decisions made on comfort and workload issues. Speed is a result of driver choice and has a profound effect on crash risk. Numerous investigations have shown that as velocity increases, the percent of time viewing the forward scene also increases (Courage, Milgram, and Smiley 2000; Godthelp, Milgram, and Blaauw 1984; Mourant and Ge 1997; Senders et al. 1967). Senders et al. studied velocities ranging from 5 to 75 mph in an on-road occlusion study. They found a monotonic relationship between velocity and percent of time viewing the roadway. Mourant and Ge (1997) considered two velocities (20 and 60 mph). They found a 9-percentage-point increase in visual demand as velocity increased from 30 to 60 mph. In a study using similar speeds Courage, Milgram, and Smiley (2000) found slightly greater increases in visual demand. Thus, there is a clear relationship between the speed at which a driver is traveling and the visual demand of the driving situation. Visual demands Associated with Highway Features Human factors research includes many studies of visual demand associated with road features, focusing on lane width and hori- zontal curves. Studies using the visual occlusion method have found that visual demand increased with decreasing lane width (Courage, Milgram, and Smiley 2000; Senders et al. 1967; Van der Horst and Godthelp 1989). For example, Courage, Milgram, and Smiley (2000) varied lane width in a medium- fidelity driving simulator. They found that as width varied from 3.7 to 2.7 m (12 to 9 ft), visual demand increased by 6%. Thus, the effect of lane width is significant but not strong. The visual demand of driving horizontal curves has also been studied extensively with the visual occlusion method (Courage, Milgram, and Smiley 2000; Godthelp 1986; Mourant and Ge 1997; Senders et al. 1967; Shafer, Brackett, and Krammes 1995; Tsimhoni and Green 1999; Tsimhoni, Yoo, and Green 1999; Wooldridge et al. 1999; Wooldridge et al. 2000). Generally, these studies show that drivers need more visual input for curves than for straight sections of roadway, indicating that curves require greater visual demand. Those studies that have systematically varied the features of curves (e.g., Shafer, Brackett, and Krammes 1995; Tsimhoni and Green 1999; Tsimhoni, Yoo, and Green 1999; Wooldridge et al. 2000) have found that visual demand (a) is inversely related to the radius of curvature, (b) does not vary much with deflection angle, (c) begins to rise at the end of the approach tangent and peaks at the beginning of the curve followed by a decline throughout the curve, and (d) was higher for S-curves than for broken-back curves, but the effect was weakened with a large separation between the curves. Note that a broken-back curve has two curves in the same direction, whereas an S-curve has two curves in opposite directions. Those findings held for both on-the-road and simulator studies. Vertical curves and combinations of vertical and horizon- tal curves have not been studied in human factors research. The most likely reason for this is that driving simulators do not simulate vertical curves adequately and test courses are usually flat, and hence techniques such as visual occlusion cannot be easily applied.
