Accident Analysis and Prevention 104 (2017) 65–73

Contents lists available at ScienceDirect

Accident Analysis and Prevention

journal homepage: www.elsevier.com/locate/aap

The effects of lane width, shoulder width, and road cross-sectional MARK
reallocation on drivers’ behavioral adaptations
⁎
Sami Mecheria, Florence Roseyb, Régis Lobjoisa,
a
Laboratory for Road Operations, Perception, Simulators and Simulation, IFSTTAR, Paris-Est University, France
b
DITM-GESM, CEREMA Normandie-Centre, France

A R T I C L E I N F O A B S T R A C T

Keywords: Previous research has shown that lane-width reduction makes drivers operate vehicles closer to the center of the
Lane width road whereas hard-shoulder widening induces a position farther away from the road’s center. The goal of the
Road reallocation present driving-simulator study was twofold. First, it was aimed at further investigating the respective effects of
Traffic condition lane and shoulder width on in-lane positioning strategies, by examining vehicle distance from the center of the
Driver behavior
lane. The second aim was to assess the impact on safety of three possible cross-sectional reallocations of the
Lateral position
width of the road (i.e., three lane-width reductions with concomitant shoulder widening at a fixed cross-
sectional width) as compared to a control road. The results confirmed that lane-width reduction made
participants drive closer to the road’s center. However, in-lane position was affected differently by lane
narrowing, depending on the traffic situation. In the absence of oncoming traffic, lane narrowing gave rise to
significant shifts in the car’s distance from the lane’s center toward the edge line, whereas this distance remained
similar across lane widths during traffic periods. When the shoulders were at least 0.50 m wide, participants
drove farther away from both the road center and the lane center. Road reallocation operations resulted in
vehicles positioned farther away from the edge of the road and less swerving behavior, without generating
higher driving speeds. Finally, it is argued that road-space reallocation may serve as a good low-cost tool for
providing a recovery area for steering errors, without impairing drivers’ behavior.

1. Introduction 1.1. Lane width

Humans are highly skilled at controlling locomotion. However, Despite the lane width is a basic factor in accident analysis research
while individuals commonly experience efficient locomotor adaptations (e.g., Hadi et al., 1995; Karlaftis and Golias, 2002), a few studies,
in complex and dynamic environments (Warren and Fajen, 2004), lane whether on a driving simulator (Dijksterhuis et al., 2011; Godley et al.,
keeping during driving remains a critical safety task. Almost one third 2004; Green et al., 1994; Lewis-Evans and Charlton, 2006) or in real-
of all road-accident fatalities (31.4%) in the European Union during the world conditions (De Waard et al., 1995), have investigated its effects
decade 2004–2013 occurred in single vehicle accidents (ERSO, 2015), on steering control and lateral position. The findings have consistently
most of which were related to lane departure (see Najm et al., 2007). indicated that lane narrowing results in less variability of the vehicle’s
Furthermore, only 33% of these single-vehicle fatalities occurred in the lateral position (De Waard et al., 1995; Dijksterhuis et al., 2011; Godley
dark, whereas 51% occurred in daylight or twilight (ERSO, 2015). The et al., 2004; Green et al., 1994) and speed reduction (De Waard et al.,
large number of crashes involving run-off-road collisions without third- 1995; Godley et al., 2004; Lewis-Evans and Charlton, 2006). As an
party involvement or severe perceptual impediment indicates that explanation of why lateral-position variability decreases with lane-
drivers are often confronted to inadequate lateral control issues. width reduction, evidence indicates that driving in a narrow lane
Keeping a proper distance from the lane boundaries along a road thus requires more steering effort on the part of the driver. Larger steering-
appears to merit further research. The present work was aimed at wheel angle deviations were found for narrow lanes as compared to
investigating the relationship between drivers’ lateral positioning and a wider ones in Godley et al. (2004), while subjectively experienced
road’s cross-sectional dimensions, in particular lane width and shoulder mental workload was increased by exposure to narrow lanes in
width. Dijksterhuis et al. (2011). The lateral-position variability has also been

⁎
Corresponding author at: Laboratory for Road Operations, Perception, Simulators and Simulation, French Institute of Sciences and Technology for Transport, Development and
Networks, Paris-Est University, 14-20 Boulevard Newton, Cité Descartes, Champs sur Marne, F-77447 Marne la Vallée Cedex 2, France.
E-mail addresses: [email protected] (S. Mecheri), fl[email protected] (F. Rosey), [email protected] (R. Lobjois).

http://dx.doi.org/10.1016/j.aap.2017.04.019
Received 16 December 2016; Received in revised form 21 March 2017; Accepted 25 April 2017
Available online 06 May 2017
0001-4575/ © 2017 Elsevier Ltd. All rights reserved.
S. Mecheri et al. Accident Analysis and Prevention 104 (2017) 65–73

shown to be negatively correlated with workload ratings (Green et al., the results obtained by Bella (2013), who showed that the presence of a
1994). Concerning travel speed, an influential hypothesis for speed 0.50 m shoulder produced a lateral position that was 0.20 m farther
choice in narrow lanes is based on models of steering control. As away from the road’s center than without a shoulder, regardless of the
supported by a number of experimental results, drivers use time-to-line- roadside configuration. Thus, shoulder widening gave rise to effects
crossing – i.e., the time needed by the vehicle to reach a lane boundary, opposite to lane-width reduction, since it produces higher speeds and
assuming the steering wheel angle remains the same – as a critical cue lateral positions closer to the edge line.
to controlling the lateral safety margin (Van Winsum and Godthelp,
1996). In this view, narrowing lane width causes drivers to reduce their 1.3. Road cross-sectional reallocation
speed so that time-to-line-crossing remains constant and safety margin
is not jeopardized (Godthelp, 1988). For nearly a decade now, a low-cost safety plan consisting of
The lane width was also shown to influence drivers’ lateral positions reconfiguring the combination of lane and shoulder widths by moving
(De Waard et al., 1995; Dijksterhuis et al., 2011; Green et al., 1994; the location of the edge line toward the center of the road has been in
Lewis-Evans and Charlton, 2006). By defining a vehicle’s lateral effect in France. In this system, each lane (defined by road markings) is
position as the distance between the car and the center of the road, it reduced in width and the width of the shoulder is increased without
has been shown that narrow lanes produce lateral positions that are changing the cross-sectional dimension of the road. This provides a
significantly closer to the road center than do wider lanes, both in larger recovery area to drivers. This reallocation of the total paved
simulation (Green et al., 1994) and in real-world studies (De Waard width thus combines the effects of two road operations (lane narrowing
et al., 1995). However, according to Green et al. (1994), this observa- and shoulder widening) on drivers’ lateral positioning and speed,
tion may not necessarily reflect different in-lane positioning if drivers known to work in opposite directions when taken separately.
systematically keep their offset from lane center constant, regardless of Examples in the literature include a simulator study conducted on a
its width. When a vehicle’s lateral position is computed as its distance road with a 7 m cross-section and two lanes 3.50 m wide reduced to
from the lane’s center, empirical evidence suggests, on the contrary, create 0.50 m shoulders, and a real-road study conducted on a road
that drivers use different in-lane positioning strategies that depend on with a 6.60 m cross-section and two 3.30 m lanes reduced to create
lane width (Dijksterhuis et al., 2011; Lewis-Evans and Charlton, 2006). 0.30 m shoulders (Rosey and Auberlet, 2012; Rosey et al., 2009). The
Lewis-Evans and Charlton (2006) showed that participants drove closer results of these two studies showed that lane narrowing by relocating
to the road’s axis in a narrow (3.00 m) lane (i.e., with a 0.37 m shift to the edge line caused participants to drive closer to the road’s center, but
the left of the lane’s center, when driving on the right-hand side of the it increased lateral-position variability and had no impact on driving
road), while a wide lane (4.60 m) caused vehicles to drive closer to the speeds (Rosey and Auberlet, 2012; Rosey et al., 2009). While the lane-
edge line (i.e. with a 0.26 m shift to the right of the lane’s center). The reduction effect on lateral position generated by relocating the edge
findings for different in-lane positions when lane width was varied were lines is in line with what was observed when the entire road (or
marginally replicated by Dijksterhuis et al. (2011), who reported a physical lane) was reduced, the results for lateral-position variability
significant effect of lane width on lateral position but no significant and speed were clearly at variance with those found in the physical
pairwise comparisons in a-posteriori analyses. This study also revealed lane-reduction literature (De Waard et al., 1995; Dijksterhuis et al.,
a significant interaction between lane width and traffic density. While 2011; Godley et al., 2004; Green et al., 1994; Lewis-Evans and Charlton,
an increase in oncoming traffic from low to high density led to a lateral 2006). Although one could assume that lane-width reduction has a
displacement of the vehicle toward the right side of the lane in the four greater influence on drivers’ behavior than shoulder-width enlarge-
lane–width conditions (3.00, 2.75, 2.50 and 2.25 m), the largest effect ment, the systematic covariation of lane and shoulder widths due to the
of traffic condition was found in the 2.75 m lane (Dijksterhuis et al., fixed roadway width found in existing reports (Rosey and Auberlet,
2011). Lastly, lane-width effects on drivers’ positioning do not appear 2012; Rosey et al., 2009) precludes drawing any clear conclusions
to result from conscious steering control – most participants did not about the respective effects of these two cross-sectional characteristics
notice changes in lane width – but rather from implicit processing of the when road space is reallocated.
road environment (Lewis-Evans and Charlton, 2006; see also Coutton-
Jean et al., 2009). Narrowing the driving lane can therefore be regarded 1.4. The present study
as a powerful way to influence road-user behavior, since the effect
occurs at an implicit perceptual level that lowers driving speed and The present study was carried out on a fixed-base driving simulator
lateral-position variability while producing a lateral position closer to to find out how speed and lateral position are affected by different lane-
the center of the road. width and shoulder-width combinations when the road’s cross-sectional
width was reallocated. While the primary research goal was to assess
1.2. Shoulder width the effectiveness of lane–shoulder combinations given a fixed road
width, lane and shoulder widths were manipulated separately to
Another cross-sectional feature of the road is the shoulder adjacent combat the above-mentioned covariation issues. Driving behavior was
to the driving lanes which may serve multiple purposes, including examined at fixed and variable pavement width on rural two-lane
providing a recovery area for driver errors (Hall et al., 1998). As for roads, which account for around two-thirds of the single vehicle
lane width, the effects of shoulder width on crash rate have been accident fatalities in France (68%) and Europe (62%) (ERSO, 2015).
investigated (e.g., Hadi et al., 1995; Wang et al., 1998; Zegeer and Based on the earlier results, it was hypothesized that lane-width
Council, 1995) but studies aimed at understanding its effects on driver reduction would induce shorter vehicle distances from the road center,
behavior are scarce. Recently, two driving-simulator studies revealed lower lateral-position variability, and slower driving speeds than would
significant shoulder-width effects on speed and lateral position (Bella, wider lanes. This study was also aimed at documenting the effects of
2013; Ben-Bassat and Shinar, 2011). The results showed that drivers lane width on in-lane positioning strategies (i.e., the vehicle’s distance
drove at higher speeds when shoulders were present (Bella, 2013) or from the center of the lane), since available findings have not made it
became wider (Ben-Bassat and Shinar, 2011). Regarding lateral posi- entirely clear what behavioral results can be expected on this issue. A
tion, Ben-Bassat and Shinar (2011) highlighted that in the presence of second hypothesis was that shoulder widening would have effects
guardrails, participants drove in the left side of the lane when the opposite to lane-width reduction and would make participants drive
shoulder was narrow (0.50 m), but shifted toward the middle of the faster and closer to the edge line. Regarding cross-sectional realloca-
lane and toward the right side of the lane when the shoulder was tion, it was expected that lane-narrowing effects at a fixed pavement
enlarged (1.20 and 3.00 m, respectively). This outcome is supported by width would be mitigated by an increase in the shoulder width. Lastly,

66
S. Mecheri et al. Accident Analysis and Prevention 104 (2017) 65–73

the impact of oncoming traffic on driver speed and lateral position was Table 1
also analyzed. A stream of traffic travelling in the opposite direction Mean and standard deviation of the vehicle distance from the road center (DRC) and the
lane center (DLC), and lateral-position variability (SDLP), for each lane-shoulder width
was included in some parts of the simulated road environment. Drivers
combination and traffic condition (Mean ± SD). The lane and shoulder widths do not
were expected to decrease their speed and deviate from the lane’s include the edge-line width.
center in the direction of the edge line during traffic periods, in order to
safely maneuver the vehicle. Lane Shoulder Cross- Traffic DRC (cm) DLC (cm) SDLP (cm)
width width (m) section
(m) (m)
2. Method
3.50 0 7.30 No 183 ± 27 8 ± 27 23 ± 8
2.1. Participants Yes 218 ± 28 43 ± 28 20 ± 8
0.25 7.80 No 186 ± 26 11 ± 26 22 ± 7
Yes 226 ± 25 51 ± 25 21 ± 7
A total of 34 drivers were recruited through web announcements
0.50 8.30 No 188 ± 26 13 ± 26 23 ± 9
and were paid for their participation. Four participants, who did not Yes 227 ± 27 52 ± 27 20 ± 6
complete the experiment due to simulator sickness (n = 3) or inability 0.75 8.80 No 199 ± 28 24 ± 28 24 ± 10
to carry out the driving task (n = 1), were excluded from later analysis, Yes 231 ± 27 56 ± 27 20 ± 7
leaving 30 participants (21 males and 9 females). The age range was 3.25 0 6.80 No 180 ± 24 18 ± 24 21 ± 8
20–52 (mean = 29.7 ± 10.1 years). All participants had had their Yes 209 ± 26 46 ± 26 17 ± 6
driver’s license for a minimum of two years and had at least 10,000 km 0.25 7.30 No 175 ± 20 13 ± 20 21 ± 7
Yes 214 ± 21 52 ± 21 19 ± 7
of driving experience. The self-reported total mileage ranged from 0.50 7.80 No 183 ± 26 21 ± 26 22 ± 9
10,000 to 580,000 km (mean = 109,233 km) and the annual mileage Yes 217 ± 25 54 ± 25 19 ± 7
ranged from 5000 to 30,000 km (mean = 9459 km). All had normal or 0.75 8.30 No 184 ± 23 21 ± 23 21 ± 7
corrected-to-normal vision and were unaware of the hypotheses under Yes 217 ± 26 54 ± 26 19 ± 6
investigation. Before taking part in the study, the participants were told 3.00 0 6.30 No 166 ± 21 16 ± 21 19 ± 8
that they were free to withdraw at any time, for any reason. All gave Yes 192 ± 21 42 ± 21 16 ± 6
0.25 6.80 No 169 ± 17 19 ± 17 18 ± 6
their consent. This study was approved by the local ethics committee.
Yes 197 ± 18 47 ± 18 17 ± 5
0.50 7.30 No 171 ± 24 21 ± 24 18 ± 6
2.2. Experimental set-up and stimuli Yes 202 ± 22 52 ± 22 18 ± 6
0.75 7.80 No 171 ± 21 21 ± 21 19 ± 6
The present experiment was conducted in a fixed-base driving Yes 202 ± 22 52 ± 22 17 ± 5

simulator equipped with real vehicle controls. The comparison of real 2.75 0 5.80 No 155 ± 20 18 ± 20 18 ± 8
and simulated road-reallocation data in the study by Rosey et al. (2009) Yes 182 ± 18 45 ± 18 13 ± 4
0.25 6.30 No 155 ± 20 18 ± 20 17 ± 7
indicated similar reallocation effects on lateral positioning and driving
Yes 185 ± 18 48 ± 18 15 ± 6
speed. Driving simulation can therefore be regarded as a reliable tool 0.50 6.80 No 159 ± 21 21 ± 21 17 ± 6
for providing a picture of driving-behavior adaptation as a function of Yes 189 ± 21 52 ± 21 15 ± 6
roadway design modifications. The simulator included a gearbox, a 0.75 7.30 No 163 ± 21 26 ± 21 17 ± 7
force-feedback steering wheel, three pedals, a dashboard, and a model Yes 192 ± 22 54 ± 22 16 ± 5

of a combustion engine. The roadway scene was displayed through

three video-projectors (refresh rate of 60 Hz, with a resolution of
and passing a specific point at an interval of 5 s. All vehicles were
1280 × 1024 pixels) on three screens each 1.86 m wide and 2.55 m
identical (width = 1.60 m) and all were positioned in such a way that
high. The distance between the participants’ head and the center of
the distance between the side of the cars (excluding the mirror) and the
each screen was 2.26 m. The visual angle subtended by the three
road center was 0.20 m. The second traffic condition was a control
screens was 135° horizontally and 56° vertically. The images were
condition where no oncoming vehicles were present. No vehicles were
calculated and projected at eye level so that the viewing angle was
in the participant’s lane throughout the experiment.
aimed at the simulated scene’s vanishing point. Realistic engine sounds
and environmental noises were generated by four speakers to enhance
the reality of the driver’s experience. The data collected at a frequency 2.3. Experimental design
of 60 Hz included information on the driver’s behavior (steering wheel
and pedals), the simulated car (position, speed), and other vehicles in Participants drove on a total of 16 experimental rural roads (see
the scene (identity, position, and speed). Table 1) created by manipulating four lane widths (3.50, 3.25, 3.00,
The simulated route was a two-lane rural road, with one lane in and 2.75 m) and four hard-shoulder widths (0, 0.25, 0.50, and 0.75 m).
each direction, stretching 6.2 km between two roundabouts. The route Except for the lane and shoulder widths, the 16 experimental roads
consisted of two uninterrupted straight sections of 3 km separated by were identical in terms of their geometry. This selection of lane widths
one curve. The curve had an arc length of 200 m and a radius of 240 m, was based on the typical design of lane width in France, 3.50 m (see
as measured from the center of the road. The road surface was textured Hall et al., 1998). The shoulder-width range was determined by the
and marked on the edges and in the center by a discontinuous line magnitude of lane reduction across the different lane widths. The edge
0.15 m wide. The surrounding terrain was a flat rural design containing line was not included in the lane or shoulder width, as shown in
trees placed with a minimum offset of 40 m from the road edge. Tree Table 1. Although the present experimental design focused on the most
spacing was random but replicated every 100 m in order to control the prominent cross-sectional width of 7.30 m in France, the selected
number of discontinuities that passed by a fixed point in the driver’s parameter combinations are also representative of road widths in other
visual field (Denton, 1980). The participants drove the simulated car on countries, and may thus provide insight into reallocation effects at an
the right side of the road. The driving conditions were daylight with international level.
good visibility. The speed limit was 90 km/h (i.e., the legal speed limit For each change in road width, the trees were moved to follow the
for two-lane rural roads in France). new road edge. For the purposes of data analysis, the data recorded in
Two traffic conditions were scripted to alternate with each straight the first and last 250 m of each 3 km straight section were excluded
section of the experimental road (i.e., after the curve). The first was a from the analyses because of pre-positioning for making a turn or post-
continuous traffic generated by private vehicles travelling at 90 km/h positioning after a turn. This resulted in 2.5 km straight sections for use

67
S. Mecheri et al. Accident Analysis and Prevention 104 (2017) 65–73

Table 2
Main and interaction effects for the vehicle distance from the road center (DRC) and the lane center (DLC), and lateral-position variability (SDLP).

Variables DRC DLC SDLP

F p ηp2 F p ηp2 F p ηp2

Lane 213.85 < 0.001 0.88 1.48 0.22 0.05 50.21 < 0.001 0.63
Shoulder 16.08 < 0.001 0.36 16.08 < 0.001 0.36 1.16 0.33 0.04
Traffic 153.31 < 0.001 0.84 153.31 < 0.001 0.84 47.71 < 0.001 0.62
Lane × Shoulder 1.08 0.38 0.04 1.07 0.38 0.04 0.32 0.97 0.01
Lane × Traffic 7.27 < 0.001 0.20 7.27 < 0.001 0.20 1.06 0.37 0.04
Shoulder × Traffic 3.78 < 0.05 0.12 3.78 < 0.05 0.12 2.48 0.07 0.08
Lane × Shoulder × Traffic 0.91 0.52 0.03 0.91 0.52 0.03 0.98 0.46 0.03
Reallocation – – – 10.14 < 0.001 0.26 10.29 < 0.001 0.26
Traffic – – – 153.90 < 0.001 0.84 13.69 < 0.001 0.32
Reallocation × Traffic – – – 3.58 < 0.05 0.11 0.97 0.41 0.03

in the statistical analyses. shoulder width, and traffic on all dependent variables were assessed
using 4 (lane width: 3.50, 3.25, 3.00, 2.75 m) × 4 (shoulder width: 0,
2.4. Procedure 0.25, 0.50, 0.75 m) × 2 (type of traffic: no traffic, traffic) repeated-
measures analyses of variance (ANOVAs). Given that the aim of the
The participants were initially informed about the nature and study was also to investigate the effects of concomitant lane and
requirements of the task. More specifically, they were told that they shoulder width variations on drivers’ behavior at a fixed total pavement
would have to drive as if they were driving in a real car on a real two- width (i.e., French standard cross-sectional width of 7.30 m), all
lane rural road, while obeying the speed limit of 90 km/h. After filling dependent variables were analyzed using 4 (road reallocation: [3.50,
out biographical and consent forms, the participants were seated in the 0], [3.25, 0.25], [3.00, 0.50], [2.75, 0.75]) × 2 (type of traffic: no
simulator and encouraged to adjust their seat position to feel comfor- traffic, traffic) repeated-measures ANOVAs.
table. They were then given the opportunity to get used to the simulator The data were analyzed using the General Linear Model for
by driving around on a winding rural road (length: 6.2 km) without any Repeated Measures analysis in STATISTICA (version 12). For every
other traffic. Each participant then drove on the 16 experimental roads, dependent variable, the main and interaction effects were analyzed by
making the distance driven approximately 100 km for the entire using the univariate approach for repeated measures. All statistical tests
experiment. For each experimental road, the two traffic conditions were done with p set at 0.05. Mauchly's test of sphericity was conducted
were presented in succession, with one condition per straight section. to determine whether the sphericity assumption was violated. In cases
Participants were informed that the start location of each trial would be of violation, the Greenhouse-Geisser (GG) correction was used to adjust
right in the middle of a roundabout and that they would drive until they the degrees of freedom. Post-hoc comparisons were conducted to follow
reached a second roundabout, on which they would stop. The order of up on significant effects using Tukey's honestly significant-difference
the experimental roads, and the order of the type of traffic on each (HSD) procedure. For each effect, partial eta-squared (ηp2) was
road, were randomized to avoid order effects. A 30-s break between calculated to determine the proportion of total variability accounting
each road was used to give the experimenter time to enter the scenario for the effect.
for the next trial. The participants took a 5-min break every four roads,
during which they left the simulator. The entire study session was 3. Results
completed in about one hour and 45 min.
Table 1 summarizes the lateral position of the vehicle and the
2.5. Data and statistical analysis variability of this position as a function of lane-shoulder width
combination and traffic condition. Table 2 displays the ANOVA results.
2.5.1. Distance from road center
This variable was defined as the distance in centimeters between the 3.1. Effects of lane and shoulder width for various pavement widths
center of the participant’s vehicle and the center of the road. A value of
0 indicates that the vehicle’s center was positioned exactly above the 3.1.1. Distance from road center
road’s center. Positive values correspond to a deviation toward the edge The ANOVA revealed significant main effects of lane width
line. (F3,87 = 213.85, p < 0.001, ηp2 = 0.88), shoulder width
(F3,87 = 16.08, p < 0.001, ηp = 0.36), and traffic (F1,29 = 153.31,
2

2.5.2. Distance from lane center p < 0.001, ηp2 = 0.84) on the distance from the road’s center. The
This variable was defined as the distance in centimeters between the distance from the road’s center decreased significantly for every degree
center of the participant’s vehicle and the middle of his or her driving of lane narrowing (3.50 m: 207 ± 33 cm; 3.25 m: 197 ± 29 cm;
lane. Positive values correspond to a deviation toward the edge line and 3.00 m: 184 ± 25 cm; 2.75 m: 172 ± 25 cm). Concerning shoulder
negative values correspond to a deviation toward the center of the road. width, pairwise comparisons revealed that the distance from the road’s
center increased with the width of the shoulder: the distance in the no-
2.5.3. Lateral-position variability shoulder condition (186 ± 30 cm) was significantly shorter than in the
The sampled values of the distance from the road center were used medium (i.e., 0.50 m; 192 ± 32 cm) and large (i.e., 0.75 m;
to calculate this variable, which corresponded to the standard deviation 195 ± 32 cm) shoulder conditions. A significant difference between
of the car’s lateral position. the small (i.e., 0.25 m; 189 ± 30 cm) and large-shoulder width con-
ditions was also observed. Concerning the main effect of traffic,
2.5.4. Speed participants drove significantly farther away from the center of the
Participant-determined travel speed of the vehicle in km/h. Driving road in the traffic condition (206 ± 27 cm) than in the no-traffic
behavior was assessed for each straight section of 2.5 km along a given condition (174 ± 26 cm).
experimental road, as detailed above. The possible effects of lane width, The interactions between lane width and traffic (F3,87 = 7.27,

68
S. Mecheri et al. Accident Analysis and Prevention 104 (2017) 65–73

Fig. 1. Distance from the center of the road, as a function of lane width and traffic condition (left panel), and shoulder width and traffic condition (right panel). The error bars represent
the standard error of the mean.

p < 0.001, ηp2 = 0.20) and between shoulder width and traffic from the lane’s center was affected differently by the shoulder width,
(F3,87 = 3.78, p = 0.013, ηp2 = 0.12) were also found significant. depending on traffic (see Fig. 2). In the no-traffic condition, the
Regarding the lane width × traffic interaction (Fig. 1), a-posteriori distance from the lane’s center increased significantly with shoulder
tests indicated that all the pairwise comparisons yielded significant width starting at a 0.50 m width. In the traffic condition, this distance
differences except the comparison between the 2.75 m lane in the increased significantly with shoulder width, except for the shoulder
traffic condition (187 ± 20 cm) and the 3.50 m lane in the no-traffic widening from 0.25 to 0.50 m and 0.50 to 0.75 m. Again, participants
condition (189 ± 27 cm). The shoulder width × traffic interaction were found to drive closer to the edge line in the traffic condition than
revealed that the shoulder-width effects on the distance from the road’s in the no-traffic condition at each shoulder width.
center differed across traffic conditions. In the no-traffic condition, the
distance from the road’s center increased significantly with the shoulder 3.1.3. Lateral-position variability
width starting at a 0.50 m wide. In the traffic condition, the distance Lane width (F3,87 = 50.21, p < 0.001, ηp2 = 0.63) and traffic
from the road’s center increased significantly with the shoulder width, (F1,29 = 47.71, p < 0.001, ηp2 = 0.62) had significant main effects
except for the widening from 0.25 to 0.50 m and 0.50 to 0.75 m (see on the lateral-position variability of the vehicle. Lateral-position
Fig. 1). Note that participants also drove significantly farther away from variability was not related to shoulder width (F3,87 = 1.16,
the road’s center in the traffic condition than in the no-traffic condition p = 0.33). No significant interactions were found. Lateral-position
at each shoulder width. variability in the 3.50 (22 ± 8 cm), 3.25 (20 ± 7 cm), 3.00
(18 ± 6 cm), and 2.75 m (16 ± 6 cm) lane-width conditions all
differed significantly. Concerning the traffic condition, significantly
3.1.2. Distance from lane center
more lateral-position variability was found in the no-traffic condition
The ANOVA revealed significant main effects of shoulder width
(20 ± 8 cm) than in the traffic condition (18 ± 7 cm).
(F3,87 = 16.08, p < 0.001, ηp2 = 0.36) and traffic (F1,29 = 153.31,
p < 0.001, ηp2 = 0.84) on the distance from the lane’s center.
Concerning shoulder width, a-posteriori tests revealed a significantly 3.1.4. Speed
shorter distance from the lane’s center in the no-shoulder condition The participants’ driving speed was not related to lane width
(30 ± 27 cm) than in the medium (36 ± 29 cm) and large (F3,87 = 0.99, p = 0.40), shoulder width (F3,87 = 0.19, p = 0.90), or
(39 ± 28 cm) shoulder conditions. A significant difference between traffic (F1,29 = 2.03, p = 0.165). No significant interactions were
the small (32 ± 27 cm) and large shoulder-width conditions was also observed. Participants averaged a driving speed of 99 ± 6 km/h
observed. Concerning the main effect of traffic, participants drove throughout the experiment.
significantly farther away from the lane’s center in the traffic
(50 ± 23 cm) than in the no-traffic (18 ± 23 cm) condition. 3.2. Effects of road reallocation
The ANOVA also revealed the following interactions: lane width × -
traffic (F3,87 = 7.27, p < 0.001, ηp2 = 0.20) and shoulder width × - Since the effects of lane and shoulder width on the distance from the
traffic (F3,87 = 3.78, p = 0.013, ηp2 = 0.12). The lane width × traffic road’s center obtained in the present study are in line with robust
interaction revealed that the lane-width effects on the distance from the findings in the literature, the results for the distance from the road’s
lane’s center differed across traffic conditions. In the no-traffic condi- center are not reported in the following analyses. This will allow us to
tion, lane narrowing from 3.50 (14 ± 27 cm) to 3.00 m (19 ± 21 cm) better focus on the effects of road reallocation on in-lane positioning
and from 3.50 to 2.75 m (21 ± 21 cm) significantly increased the strategies, steering stability, and speed control. The analyses pertain to
deviation from the lane’s center toward the edge line. In the traffic the most prominent road cross-section in France: i.e. 7.30 m.
condition, the distance from the lane’s center was not affected by lane
width, and the participants averaged an offset from the lane center of 3.2.1. Distance from lane center
50 ± 23 cm (see Fig. 2). This interaction also revealed that the The 4 (road reallocation) × 2 (traffic) ANOVA revealed significant
deviation from the lane center (in the direction of the edge line) was main effects of road reallocation (F3,87 = 10.14, p < 0.001,
higher in the traffic than in the no-traffic condition at each lane width. ηp2 = 0.26) and traffic (F1,29 = 153.90, p < 0.001, ηp2 = 0.84) on
The shoulder width × traffic interaction also showed that the distance the distance from the lane’s center. The mean distance was significantly

69
S. Mecheri et al. Accident Analysis and Prevention 104 (2017) 65–73

Fig. 2. Distance from the center of the lane, as a function of lane width and traffic condition (left panel), and shoulder width and traffic condition (right panel). All positive values mean
that participants drove consistently on the right-hand side of the lane. The error bars represent the standard error of the mean.

shorter in the [3.50, 0] condition (26 ± 32 cm) than in the [3.00, reallocation, depending on the traffic (see Fig. 3). In the no-traffic
0.50] (37 ± 28 cm) and [2.75, 0.75] (40 ± 26 cm) conditions. A condition, the distance from the lane’s center increased significantly
significant difference between the [3.25, 0.25] (32 ± 28 cm) and with concomitant lane narrowing and shoulder widening, except for the
[2.75, 0.75] reallocation conditions was also observed. Regarding the reallocation from [3.50, 0] to [3.25, 0.25] and from [3.00, 0.50] to
traffic effect, participants drove significantly farther away from the [2.75, 0.75]. In the traffic condition, reallocation of road space from
center of the lane in the traffic condition (50 ± 24 cm) than in the no- [3.50, 0] (43 ± 28 cm) to [3.25, 0.25] (52 ± 21 cm) significantly
traffic condition (17 ± 24 cm). increased the distance from the lane’s center by producing shifts toward
The ANOVA also revealed a significant road reallocation × traffic the edge line; however, changes in road-space reallocation from the
interaction (F3,87 = 3.58, p = 0.017, ηp2 = 0.11), which indicated that [3.25, 0.25] condition to the [3.00, 0.50] and [2.75, 0.75] conditions
the distance from the lane’s center was affected differently by road did not produce any changes in the distance from the lane’s center. Here

Fig. 3. Lateral control measures and speed on the control road and the three reallocated roads, in the absence (top row) and presence (bottom row) of traffic. The simulated car (width:
160 cm) is depicted in proportion to lane width. The distance from the road’s center, edge of the road and the edge line are also provided as a guide (values are in centimeters). The edge
of the road and the edge line overlapped on the control road.

70
S. Mecheri et al. Accident Analysis and Prevention 104 (2017) 65–73

also, participants were found to drive significantly farther away from lane narrowing literature (De Waard et al., 1995; Dijksterhuis et al.,
the lane center in traffic than in no traffic, for each reallocation 2011; Godley et al., 2004; Green et al., 1994). This indicates that the
condition (see Fig. 3). participants coped with different lateral demands by decreasing the
zone in which they drove by around 2 cm between each lane-width
3.2.2. Lateral-position variability change. The results also showed that lateral-position variability was
Road-space reallocation (F3,87 = 10.29, p < 0.001, ηp2 = 0.26) greater in the absence of oncoming traffic than in its presence. This
and traffic (F1,29 = 13.69, p < 0.001, ηp2 = 0.32) had significant suggests that drivers paid closer attention to path tracking during
main effects on the vehicle’s lateral-position variability. No interaction periods of traffic. This decrease in lateral-position variability may in
effect was observed. Concerning road reallocation, lateral-position fact demonstrate that the drivers were less prone to letting the vehicle
variability was significantly higher in the [3.50, 0] condition deviate more within the current driving lane. This probably occurred
(21 ± 8 cm) than in the [3.00, 0.50] (18 ± 6 cm) and [2.75, 0.75] because the presence of continuous traffic resulted in a significant
(17 ± 6 cm) conditions. A significant difference between the [3.25, deviation (about 30 cm) from the lane’s center in the direction of the
0.25] (20 ± 7 cm) and the [2.75, 0.75] reallocation conditions was shoulder that placed drivers substantially closer to the edge line.
also observed. Concerning the traffic variable, significantly more Because the speed at which the participants were driving was controlled
lateral-position variability was found in the no-traffic condition by the simulator in Dijksterhuis et al. (2011), the data collected in this
(20 ± 7 cm) than in the traffic condition (18 ± 7 cm). experiment points out the drivers’ ability to provide the extra steering
effort needed to drive in a narrower zone in the presence of oncoming
3.2.3. Speed traffic, without using a potential compensatory reaction consisting of
The participants’ driving speed was not related to road-space slowing down.
reallocation (F3,87 = 0.45, p = 0.715) or traffic (F1,29 = 1.50,
p = 0.231). No interaction effect was observed. Participants averaged 4.2. Lateral position and shoulder width
a driving speed of 100 ± 6 km/h.
Shoulder width had a major impact on drivers’ lateral positioning.
4. Discussion Indeed, the distances from both the road center and the lane center
significantly increased with the shoulder width, irrespective of oncom-
4.1. Lateral position and lane width ing traffic. Contrary to what was observed for lane width, shoulders at
least 0.50 m wide modified how drivers positioned themselves with
Analysis of the vehicle-position data indicated significant differ- respect to the center of their driving lane in both the absence and the
ences in the distance from the road’s center across of all the lane-width presence of oncoming traffic, by moving toward the edge line. In this
conditions, whereas lane-width variations did not produce significant respect, the presence of large shoulders seems to tell drivers that they
in-lane position differences. This is due to the fact that participants have a recovery area that is taken into account in the driving task.
positioned their vehicle on the right-hand side of the lane throughout These results are in line with what was found in two studies examining
the experiment, with an average offset from the lane center of 34 cm the effects of shoulder width on driving behavior (Bella, 2013; Ben-
toward the edge line for all four lane widths (for similar lateral Bassat and Shinar, 2011). The reason for driving farther away from the
positions, see Dijksterhuis et al., 2011; and Green et al., 1994). This middle of the lane in the presence of a medium-size or large shoulder is
position most likely occurred because while there was a concrete probably that drivers sense that there is more space and feel safer, as
penalty for driving too far to the left on the experimental roads (risk demonstrated by the subjective measures found in the Ben-Bassat and
of running into oncoming traffic), the penalty for driving too far to the Shinar, 2011Ben-Bassat and Shinar study (2011), when shoulders were
right was only implicit (lane departure). combined with guardrails. Importantly, while the drivers systematically
While the distance from the road’s center was expected to decrease used the lane space available to maneuver – even when the conse-
with lane width, which is a well-documented finding (De Waard et al., quence was an increase in lateral-position variability – this variability
1995; Green et al., 1994; Rosey et al., 2009), the literature has (influenced by lane width) was unaffected when the shoulder was large.
produced somewhat contradictory findings on in-lane positioning as a This lack of a difference in lateral-position variability between the
function of lane width (Dijksterhuis et al., 2011; Lewis-Evans and different shoulder widths strongly suggests that drivers may not see the
Charlon, 2006). The results showing that in-lane position was affected shoulder width (or the whole road width) as a driveable portion of the
by lane-width variations only when no oncoming vehicles were present road, but only as a recovery area.
offer a new picture of lane narrowing effects and can help in
documenting this issue. In the absence of oncoming traffic, lane 4.3. Driving speed
narrowing from 3.50 m to 3.00 and 2.75 m significantly increased the
distance from the lane’s center as the vehicle moved closer to the edge The results showed that the participants’ driving speed was not
line. Insofar as the distance from the road’s center decreased with lane affected by any of the manipulated cross-sectional features. Although
width, drivers may have attempted to compensate for this decrease by this result is consistent with what was reported in a previous study that
shifting to the right and thus maintaining their lateral clearance from implemented lane narrowing by relocating road markings (Rosey et al.,
the opposing lane. This trajectory in response to lane narrowing led to a 2009), it conflicts not only with the evidence showing that wider
large decrease in the distance from the edge line, which probably helps shoulders lead to higher speeds (Bella, 2013; Ben-Bassat and Shinar,
account for the smaller lateral-position variability elicited in the two 2011), but also with the findings reported in the physical lane
narrowest lanes. When oncoming traffic was present, however, the narrowing literature (De Waard et al., 1995; Godley et al., 2004;
distance from the lane’s center was not influenced by lane narrowing, Lewis-Evans and Charlton, 2006). A potential explanation for the
most certainly because the drivers began to shift substantially toward similar speeds used on different lane-shoulder width combinations in
the edge line starting in the largest lane. Based on this result, lane the present experiment may be related to speed controls on in-vehicle
narrowing (irrespective of shoulder width) should not be employed as a displays. This is illustrated by the lane-width effect on driving speed
tool for influencing drivers’ in-lane positioning strategies, since ex- found in Godley et al. (2004), where the participants were denied
posure to oncoming traffic is highly frequent on two-lane rural roads in access to the speedometer. But while this factor needs to be considered,
real driving conditions. it cannot provide overwhelming support, since Lewis-Evans and
Concerning lateral-position variability, drivers swerved less as lane Charlton (2006) reported speed differences due to lane width while
width decreased, which is in line with a robust finding in the physical displaying the speedometer on the screen. However, these authors

71
S. Mecheri et al. Accident Analysis and Prevention 104 (2017) 65–73

observed an asymmetric pattern of speed adaptation brought about by constitutes a relevant positioning strategy. More generally, based on the
road changes: the narrow road decreased speed but the wide road did strict effect of road reallocation on lateral position (i.e., in the absence
not significantly increase speed over that of the control width. of traffic), these findings highlight a strategy on the part of the
Similarly, while mean speed was significantly lower for the narrow participants consisting of reversing the road surface distribution from
lane width than for the medium lane width in Godley et al. (2004), a large space on the left of their vehicle on the control road, to a large
speed on the wide lane (control road) did not significantly differ from space on the right of their vehicle on the [2.75, 0.75] road. Note that
the other two lanes. Additionally, in a real road study by Lum (1984), the drivers adapted their position in such a way that the road width was
narrowing the perceptual lane width (within the same physical lane) roughly equivalent on the two sides of the vehicle in the [3.25, 0.25]
did not result in slower driving speeds. These findings indicate an (95 and 95 cm) and [3.00, 0.50] (91 and 99 cm) reallocated roads.
absence of systematic speed changes as a function of lane width, Proposing recommendations for road-space reallocation thus requires
notably because several factors that potentially influence speed beha- choosing between the likelihood of head-on and run-off-road accidents.
vior differed across the studies in question (e.g., lane and shoulder Based on a general description of accidents, it seems that a vehicle’s
width, surrounding terrain, or oncoming traffic). The literature on the lateral position relative to the edge line is more significant in terms of
relationship between speed and lane width thus remains unclear and safety than its position relative to the centerline (see Mäkinen et al.,
thereby precludes drawing any clear conclusions about the effects of 1999). From this angle, it would be highly worthwhile for road
cross-sectional road features on driving speed. reallocation programs to make drivers operate their vehicle farther
Examination of data also revealed that the participants did not away from the edge of the pavement; this is likely to lower the risk of
strictly adhere to the legal speed limit throughout the experiment but run-off-road accidents.
drove slightly over the limit (99 km/h). This can probably be explained
by the long sight distance and the high proportion of straight-line 5. Conclusion
stretches on the simulated roads, which no doubt made the driving task
somewhat monotonous due to its low control demands. The present study provided insight that allows making some useful
suggestions for improving road safety on two-lane rural roads via low-
4.4. Road space reallocation with a fixed pavement width cost road-space reallocation programs. The results showed that the
road-space reallocation with the greatest potential for increasing road
In road reallocation programs, lane and shoulder widths are safety on a control road 7.30 m wide was the 3.00 m lane with a 0.50 m
interrelated by total cross-sectional width. In this section, the empirical adjacent shoulder. This lane-shoulder combination seems particularly
before-and-after method usually employed for safety evaluation was effective since it produced significant road user behavior changes in
applied. It consists of comparing a control road site (two 3.50 m lanes lane positioning and steering stability. First, the [3.00, 0.50] reallo-
bounded by 0.15 m edge lines each), with three possible reallocation cated road elicited lateral-position strategies that significantly in-
operations on this control road. The results showed that the road-space creased the distance between the vehicle and the edge of the road, as
reallocations clearly affected the drivers’ behavior in terms of in-lane compared to the control road. Second, from the strict point of view of
positioning strategies and lateral-position variability. In contrast to the road-reallocation effect (i.e., when drivers were not confronted with
Rosey and Auberlet (2012), lane narrowing via the relocation of road oncoming traffic), the drivers adapted their in-lane positioning for this
markings led to significantly less lateral-position variability in the lane-shoulder combination in such a way that the road surface was
[3.00, 0.50] and [2.75, 0.75] conditions than on the control road. equally distributed on either side of the vehicle. Third, when partici-
Regarding in-lane positioning, the distance from the lane’s center pants were driving on the [3.00, 0.50] road, their steering trajectory
significantly increased between the control road and the [3.00, 0.50] resulted in less lateral-position variability than on the control road.
and [2.75, 0.75] lane-shoulder combinations, with mean shifts of 11 Additionally, because the data of this study strongly suggest that drivers
and 14 cm, respectively. This in-lane positioning strategy, however, did not consider the adjacent shoulder as an extra lane for travelling,
made drivers gradually end up with very little space on their right, even during traffic periods, a 3.00 m wide lane seems to be better than a
while leaving a large maneuvering space on their left. Indeed, on the 2.75 m reallocated lane on rural roads where trucks are expected.
control road with no traffic, drivers had 87 cm of maneuvering space on Importantly, not all of these behavioral adaptations were accompanied
the right side of their vehicle, while this value fell to 32 cm in the [2.75, by deleterious effects that counteracted any overall increase in safety,
0,75] condition (the remaining space on the vehicle’s left dropped from such as higher driving speeds or greater lateral-position variability with
103 to 83 cm; see Fig. 3). During traffic periods, only 3 cm separated shoulder widening. Therefore, the [3.00, 0.50] lane-shoulder width
the side of the car from the edge line in the [2.75, 0.75] condition, combination can be recommended for 90 km/h two-lane rural-road
whereas there were 52 cm on the right of the vehicle on the control reallocation with this particular cross-section.
road. Interestingly, however, while participants significantly increased In making these suggestions, we must add a note of caution,
their distance from the lane center between the control road and the however. The use of only one volume of oncoming traffic in this study
[3.25, 0.25] reallocated road during traffic periods, they did not deviate does not accurately reflect actual demands on the lane keeping task
differently from the lane center across the three different reallocated which emerge from various flow densities. Further, other road geo-
roads, most likely in order to keep within the bounds of the lane. By metric features such as access density or curves with varying radii were
doing so, lane narrowing across reallocations did not lead participants not part of the simulated environment whereas it is regularly encoun-
to drive dangerously by travelling on the shoulders. tered on real roads. Another limitation could be that elderly drivers
At first sight, it is striking to find that participants considered it were not included in the sample, as they were found to compensate for
acceptable to drive at just a few centimeters from the edge line even decreased motor skill by remaining closer to the middle of the lane
though substantial lane space was available on their left. But when when negotiating curves (Raw et al., 2012). Therefore, the influence of
taking a closer look at the data, this in-lane positioning also implied other traffic volumes, road geometric features, and age on drivers’
more and more road surface on the right-hand side due to shoulder behavioral adaptations should be analyzed as an extension of this work
widening. Indeed, although only 3 cm separated the side of the car from to strengthen the results.
the lane markings in the most restrictive situation ([2.75, 0.75] in the
presence of traffic), the drivers were positioned at 78 cm away from the Acknowledgments
edge of the road. In the light of these observations, it can be assumed
that the participants considered that attempting to maximize their This research is part of the French national multidisciplinary project
distance from oncoming traffic without going over the lane boundary PROFIL. This project was supported by funding from the Fondation

72
S. Mecheri et al. Accident Analysis and Prevention 104 (2017) 65–73

Sécurité Routière. The authors would like to thank Isabelle Aillerie and Practices, vol. 12. Bostonpp. 1–12.
Karlaftis, M.G., Golias, I., 2002. Effects of road geometry and traffic volumes on rural
Fabrice Vienne for their help in designing the simulated road and roadway accident rates. Accid. Anal. Prev. 34 (3), 357–365. http://dx.doi.org/10.
programming the driving scenarios. They are also grateful to Vivian 1016/S0001-4575(01)00033-1.
Waltz for English-language editing. Lewis-Evans, B., Charlton, S.G., 2006. Explicit and implicit processes in behavioural
adaptation to road width. Accid. Anal. Prev. 38 (3), 610–617. http://dx.doi.org/10.
1016/j.aap.2005.12.005.
References Lum, H.S., 1984. The use of road markings to narrow lanes for controlling speed in
residential areas. Inst. Transp. Eng. 54 (6), 50–53.
Mäkinen, T., Lundkvist, S., Beilinson, L., Kallio, M., Kulmala, P., Mikkola, J., Unhola, T.,
Bella, F., 2013. Driver perception of roadside configurations on two-lane rural roads:
1999. Task 400 – Effects of Road Markings on Driver Behaviour (No. COST 331).
effects on speed and lateral placement. Accid. Anal. Prev. 50, 251–262. http://dx.doi.
Brussels: European Commission, Directorate General Transport.
org/10.1016/j.aap.2012.04.015.
Najm, W.G., Smith, J.D., Yanagisawa, M., 2007. Pre-crash scenario typology for crash
Ben-Bassat, T., Shinar, D., 2011. Effect of shoulder width, guardrail and roadway
avoidance research. DOT-HS-810. National Highway Transportation Safety
geometry on driver perception and behavior. Accid. Anal. Prev. 43 (6), 2142–2152.
Administrationpp. p. 767.
http://dx.doi.org/10.1016/j.aap.2011.06.004.
Raw, R.K., Kountouriotis, G.K., Mon-Williams, M., Wilkie, R.M., 2012. Movement control
Coutton-Jean, C., Mestre, D.R., Goulon, C., Bootsma, R.J., 2009. The role of edge lines in
in older adults: does old age mean middle of the road? J. Exp. Psychol. Hum. Percept.
curve driving. Transp. Res. Part F 12 (6), 483–493. http://dx.doi.org/10.1016/j.trf.
Perform. 38 (3), 735–745. http://dx.doi.org/10.1037/a0026568.
2009.04.006.
Rosey, F., Auberlet, J.-M., 2012. Trajectory variability: road geometry difficulty indicator.
Denton, G.G., 1980. The influence of visual pattern on perceived speed. Perception 9 (4),
Saf. Sci. 50 (9), 1818–1828. http://dx.doi.org/10.1016/j.ssci.2012.04.003.
393–402. http://dx.doi.org/10.1068/p090393.
Rosey, F., Auberlet, J.-M., Moisan, O., Dupré, G., 2009. Impact of narrower lane width:
Dijksterhuis, C., Brookhuis, K.A., De Waard, D., 2011. Effects of steering demand on lane
comparison between fixed-base simulator and real data. Transp. Res. Rec. 2138,
keeping behaviour, self-reports, and physiology. A simulator study. Accid. Anal. Prev.
112–119. http://dx.doi.org/10.3141/2138-15.
43 (3), 1074–1081. http://dx.doi.org/10.1016/j.aap.2010.12.014.
Van Winsum, W., Godthelp, H., 1996. Speed choice and steering behavior in curve
ERSO (European Road Safety Observatory), 2015. Traffic Safety Basic Facts on Single
driving. Hum. Factors: J. Hum. Factors Ergon. Soc. 38 (3), 434–441. http://dx.doi.
Vehicle Accidents. European Commission, Directorate General for Transport June.
org/10.1518/001872096778701926.
Godley, S.T., Triggs, T.J., Fildes, B.N., 2004. Perceptual lane width, wide perceptual road
De Waard, D., Jessurun, M., Steyvers, F.J.J.M., Reggatt, P.T.F., Brookhuis, K.A., 1995.
centre markings and driving speeds. Ergonomics 47 (3), 237–256. http://dx.doi.org/
Effect of road layout and road environment on driving performance, drivers’
10.1080/00140130310001629711.
physiology and road appreciation. Ergonomics 38 (7), 1395–1407. http://dx.doi.org/
Godthelp, H., 1988. The limits of path error-neglecting in straight lane driving.
10.1080/00140139508925197.
Ergonomics 31 (4), 609–619. http://dx.doi.org/10.1080/00140138808966703.
Wang, J., Hughes, W.E., Stewart, R., 1998. Safety effects of cross-section design on rural
Green, P., Lin, B., Bagian, T., Driver workload as a function of road geometry: a pilot
multi-lane highways. International Symposium on Highway Geometric Design
experiment. Technical Report UMTRI-93-39/GLCTTR 22-91/01, 1994, Ann Arbor,
Practices, vol. 18, 1–14 Boston.
MI: University of Michigan Transportation Research Institute, 1001(48), 109.
Warren, W.H., Fajen, B.R., 2004. From optic flow to laws of control. In: Vaina, L.M.,
Hadi, M.A., Aruldhas, J., Chow, L.-F., Wattleworth, J.A., 1995. Estimating safety effects of
Beardsley, S.A., Rushton, S. (Eds.), Optic Flow and Beyond. Springer, Kluwer, pp.
cross-section design for various highway types using negative binomial regression.
307–337.
Transp. Res. Rec. 1500, 169–177.
Zegeer, C.V., Council, F.M., 1995. Safety relationships associated with cross-sectional
Hall, L.E., Powers, R.D., Turner, D.S., Brilon, W., Hall, J.W., 1998. Overview of cross
roadway elements. Transp. Res. Rec. 1515, 29–35.
section design elements. International Symposium on Highway Geometric Design

73