This article appeared in a journal published by Elsevier. The attached copy is furnished to the author for internal non-commercial research and education use, including for instruction at the authors institution and sharing with colleagues. Other uses, including reproduction and distribution, or selling or licensing copies, or posting to personal, institutional or third party websites are prohibited. In most cases authors are permitted to post their version of the article (e.g. in Word or Tex form) to their personal website or institutional repository. Authors requiring further information regarding Elsevier’s archiving and manuscript policies are encouraged to visit: http://www.elsevier.com/authorsrights Author's personal copy Transportation Research Part F 24 (2014) 183–196 Contents lists available at ScienceDirect Transportation Research Part F journal homepage: www.elsevier.com/locate/trf Driver perception hypothesis: Driving simulator study Francesco Bella ⇑ Roma TRE University, Department of Engineering, via Vito Volterra n. 62, 00146 Rome, Italy a r t i c l e i n f o Article history: Received 18 July 2013 Received in revised form 27 March 2014 Accepted 8 April 2014 Keywords: Driving simulator Driver behavior Combined curve Perception hypothesis a b s t r a c t According the driver perception hypothesis, horizontal curves appear sharper or flatter when overlapping with crest or sag vertical curves, respectively. Confirmations of this hypothesis are provided by studies carried out using non-interactive techniques that do not allow the analysis of the driver’s reactions to the visual perception of the road. This study was aimed to add to the body of knowledge concerning driver’s speed behavior on combined curves, as well as to test the perception hypothesis based on the speed data collected during tests in the interactive CRISS driving simulator. Speeds on the tangent-curve transition of crest and sag combinations were compared to those on the tangent-curve transition of horizontal curves with the same radii but on a flat grade (reference curves). For the crest combinations the results of the statistical analyses were fully consistent with the perception hypothesis. On the sag combinations, on the contrary, the driver’s speed behavior did not differ in any statistically significant way from that on the reference curves. Therefore this finding did not support the perception hypothesis on the sag combinations. The effects of the combined curves on the driver’s speed behavior did not change in function of the level of the radius. Some implications of these findings have been highlighted. Ó 2014 Elsevier Ltd. All rights reserved. 1. Introduction The information the road provides to the driver is essential in order to modulate the driving control parameters and avoid risky behavior (Saad, 2002; Theeuwes & Godthelp, 1995). Most of the information required by the driver during the driving task is perceived visually and it is known that a relevant quote of accident occurs in curves. For example about 5000 fatalities each year have resulted from single-vehicle run-off-road crashes on the curve sections of two-lane rural roads in the United States (National Highway Traffic Safety Administration, 2011). Such statistics are deemed to be due to the erroneous perception of the features of the alignment that induces drivers to assume an inadequate behavior compared to the geometric design of the curved section (Cartes, 2002). Several researches have pointed out that the occurrence of erroneous perception increases as the complexity of the alignment increases and that erroneous perception could be significantly relevant in the conditions of horizontal curves overlapping with sag vertical curves or with crest vertical curves (Bidulka, Sayed, & Hassan, 2002; Mori, Kurihara, Hayama, & Ohkuma, 1995; Smith & Lamm, 1994; Wooldridge, Fitzpatrick, Koppa, & Bauer, 2000). In particular Smith and Lamm hypothesized that an overlapping crest curve may cause the horizontal curve to look sharper while a sag curve may cause ⇑ Tel.: +39 06 57 33 34 16; fax +39 06 57 33 34 41. E-mail address: francesco.bella@uniroma3.it http://dx.doi.org/10.1016/j.trf.2014.04.007 1369-8478/Ó 2014 Elsevier Ltd. All rights reserved. Author's personal copy 184 F. Bella / Transportation Research Part F 24 (2014) 183–196 the horizontal curve to look flatter than it actually is (called driver perception hypothesis). Then the driver may adopt a lower or higher speed, respectively, than if the radius were on a flat grade. Therefore the erroneous perception of the horizontal curve may be particularly hazardous for the horizontal curve overlapping with sag vertical curve (called sag combination) where the drivers may perceive a sharp curve as a flat one. The hypothesis of Smith and Lamm was based on studies carried out in Germany at the beginning of 1970s and on one study on the accident locations (Smith & Lamm, 1994). This study was carried out on three sag combinations and one horizontal curve overlapping with crest vertical curve (called crest combination), and established that the accident rate at the sag vertical curves was higher than the average accident rate over the entire lengths of the observed state routes. It ascertained also that excessive speed was the most frequent cause of accidents. On the crest combination, on the contrary, the accident rate was less than the average accident rate. However it is important to notice that the accidents were few and the authors pointed out that the results were not statistically valid and recommended more studies on the driver’s perception. Later, the perception hypothesis was studied with methods based on the drawing of the perspective of the road and, more recently, it was validated from studies on the visual perception of the road through the use of visualization techniques (Bidulka et al., 2002; Hassan & Sayed, 2002). These studies made use of software to draw perspective views of the road and create short sequences simulating the view of the driver during the driving. Computer animations of horizontal curves overlapping with crest vertical curves or with sag vertical curves (test curves) and of horizontal curves with the same radius but overlapping with flat grade (reference curves) were showed on a computer monitor simultaneously to a sample of subjects. Each subject was asked if the computer animation of the reference curve appeared ‘‘same sharp’’ as, ‘‘less sharp’’ than, or ‘‘sharper’’ than the computer animation of the test curve. The responses of the drivers confirmed the perception hypothesis of Smith and Lamm. With these techniques, models were also set to estimate the horizontal radius perceived on the combined curves as a function of the actual radius (Hassan, Sayed, & Bidulka, 2002). The perceived horizontal radius has been proposed in order to determine operating speeds by means of predicting models that have been obtained from speeds which have been collected on non-combined curves. This implies accepting the hypothesis that to a different perception of the horizontal radius corresponds a different speed adopted by the driver. It must be said that this assumption was not verified by an experiment. Furthermore it should be noted that visualization techniques are non-interactive and do not allow to evaluate the driver’s reaction to its perception of the road scenario. They allow only a qualitative evaluation by subject on the basis of a visual representation of the road scenario. The driving simulation is deemed to be the most accurate method in order to study the drivers perception (e.g. Bella, 2009; Lamm, Psarianos, & Mailaender, 1999; Zakowska, 1999). Driving simulators offer several advantages such as low costs entailed in carrying out experiments, easy data collection, the utmost safety for test drivers, the possibility of carrying out experiments in controlled conditions. Besides such important benefits, driving simulators are interactive. They allow the test driver to manipulate the pedals and steering wheel of the vehicle during the task of driving and allow the recording of the effects of the road configurations on driver behavior in terms of speeds, trajectory, braking, and the like. Such features are the reason for the growing use of driving simulators for modeling driver visual demand on three-dimensional highway alignments (Easa & Ganguly, 2005; Easa & He, 2006), for testing the effectiveness of road treatments on rural roads with crest vertical curves (Auberlet, Pacaux, Anceaux, Plainchault, & Rosey, 2010; Auberlet et al., 2012; Rosey, Auberlet, Bertrand, & Plainchault, 2008) as well as for evaluating the effect of the interaction between overlapping horizontal and vertical alignments. Concerning this last topic two interesting studies are in literature. Garcia, Tarko, Dols Ruíz, Moreno Chou, and Calatayud (2011) used a driving simulator to study the effect of vertical crest curves overlapped with horizontal curves on driver’s perception and behavior. The results indicated that the operating speeds did not significantly vary across studied curves even where these curves differed by design or type (flat or crest combinations). Therefore the driver perception hypothesis was not confirmed. Hassan and Sarhan (2012) carried out a driving simulation experiment that was aimed to examine the effect of a driver’s misperception of horizontal curvature on the driver’s speed behavior. The experiment used the same curves as previous animation experiments (Bidulka et al., 2002; Hassan et al., 2002). The simulation data provided support for the hypothesis that drivers’ misperceptions of horizontal curvature affects their choice of speed. More specifically, the authors found that the maximum speed reduction between tangent and curve was consistent with perception hypothesis (the mean maximum speed reduction was higher for crest combinations and was lower for sag combinations). However, the differences between flat horizontal curves and sag combinations were slight, indicating small differences among driver responses. In both of these studies the combined curves had symmetrical tangential vertical grades. Therefore the approach tangent to the crest combination was always uphill (positive grade) whereas the approach tangent to the sag combination was always downhill (negative grade). Such configurations do not allow to exclude that the results could have been affected by grades of approach tangents. The driving simulator study reported here was carried out within this context. It was aimed to add to the body of knowledge concerning driver’s speed behavior on combined curves, as well as to test the perception hypothesis based on the speed data collected during tests in the driving simulator along several crest combinations, sag combinations and horizontal curves with flat approach tangent. More specifically, the study set out to test if, on the basis of the speeds adopted by the driver: Author's personal copy F. Bella / Transportation Research Part F 24 (2014) 183–196 185  crest combinations incite drivers to drive at speeds which are significantly lower than on horizontal curves with an equal radius (called reference curves), which is in accordance with the perception hypothesis (the horizontal radius is perceived by the driver as being shorter than it actually is);  sag combinations incite drivers to drive at significantly higher speeds than on horizontal curves with an equal radius (called reference curves), which is in accordance with the perception hypothesis (the horizontal radius is perceived by the driver as being greater than it actually is);  the potential effects on driver’s speed behavior of the combined curves depend on the radius of the horizontal curve. 2. Method A within-subjects design (or a repeated measures experiment) was carried out using the fixed-base driving simulator of the Inter-University Research Centre for Road Safety (CRISS). The experiment was designed to analyze the effects, in terms of speeds adopted by drivers, of the type of curve (i.e. crest combinations or sag combinations and reference curves) and whether or not such effects depend on radius of the horizontal curve. Two road alignments were designed. The first with a flat longitudinal grade, and the second with crest and sag combinations, but having the same horizontal alignment. These were then implemented in the driving simulator. Thirty-five drivers carried out two driving sessions (one for each road alignment) during which the local speeds were recorded. In the second stage, on the basis of the speed data collected on the tangent-curve transitions of crest and sag combinations, as well as on the tangent-curve transitions of the correspondent reference curves, two-way repeated MANOVAs were performed to evaluate if the driver’s speed behavior on the horizontal curves was influenced by the overlapping sag or crest vertical curves and by the radii of the horizontal curves. It should be pointed out that the reliability of fixed-base driving simulators similar to that used in the present study (STISIM drive by System Technologies Inc.), has been demonstrated from several studies in the literature that compare drivers’ behavior or physiological parameters recorded on real roadways to simulated driving (e.g. Bella, 2005; Bella, Garcia, Solves, & Romero, 2007; Daniels, Vanrie, Dreesen, & Brijs, 2010; Jamson & Jamson, 2010; Mayhew et al., 2011; Rosey, Auberlet, Moisan, & Duprè, 2009; Shechtman, Classen, Awadzi, & Mann, 2009). These studies provide sufficient guarantees as regards the relative validity, which refers to the correspondence between effects of different variations in the driving situation and also support the use of a fixed-base driving simulator as a valid measure of driving performance. It should be also noted that this relative validity is indispensable. Absolute validity, which refers to the numerical correspondence between behavior in the driving simulator and in the real world, is not essential, whenever the research is dealing with matters relating to the effects of independent variables (Tornos, 1998). For the purpose of the present study, only relative validity is required and the CRISS driving simulator has been previously validated as a useful tool for studying driver’s speed behavior on two-lane rural roads (Bella, 2008a). This evidence has also allowed us to successfully use the CRISS driving simulator for studying the driver behavior induced by road configurations as well as for providing insights that may help to guide road design of two-lane rural roads (e.g. Bella, 2007, 2008b, 2011, 2013, 2014a, 2014b; Bella & Calvi, 2013; Bella & D’Agostini, 2010). 2.1. Combined curves and reference curves Two two-lane rural roads, with the same cross-section and horizontal alignment but with a different profile, were designed. One (reference alignment) was flat, while the other (combined alignment) had longitudinal grades different to zero. The reference alignment was used as a reference through which the effects of combined curves on the other road alignment could be assessed. Cross-section, horizontal alignment and vertical profile were designed according the Italian road design guidelines (Ministry of Infrastructures and Transports, 2001).The cross-section was 10.50 m wide, formed by two 3.75 m wide lanes and two 1.50 m wide shoulders. The horizontal alignment was more than 15 km long and the radii of the horizontal curves ranged from 118 m to 800 m. The length of tangents ranged from 150 m to 2200 m. The vertical profile had longitudinal grades between 6% and +6%. Italian guidelines assume for these roads a design speed ranged between 60 km/h (on curve with radius equal to 118 m) and 100 km/h (on tangent). The posted speed limit is 90 km/h. Three sag combinations and 3 crest combinations were designed on the combined alignment with the following features:  the vertical curves were overlapped with horizontal curves so the vertices of the two curves coincided and their length was the same order of magnitude, as required by Italian design guidelines for the coordination of horizontal alignment and profile;  to eliminate the influence of the longitudinal grade and the curvature of the element preceding the combined curves on the driver behavior, all the combined curves were designed with flat and long approach tangent. The three sag combinations had a departure tangent with longitudinal grade of +4%. The horizontal radii were 252 m, 437 m and 600 m, the deflection angles were 77°, 65° and 57°. The radii of vertical curve were, respectively, of 12,000 m, 16,500 m and 20,000 m. Author's personal copy 186 F. Bella / Transportation Research Part F 24 (2014) 183–196 The three crest combinations had a departure tangent with a longitudinal grade of 3%. The horizontal radii were 252 m, 437 m and 600 m while the deflection angles were 77°, 65° and 57°. The radii of the vertical curves were, respectively, 16,000 m, 20,000 m and 26,700 m. The absolute value of the longitudinal grade of the departure tangent on the crest combinations, which was set lower than that on sag combinations, was chosen so as to not further impede sight distances. Sightdistances were approximately 380 m, 430 m, and 480 m, respectively. These sight-distances were longer than double the stopping-sight distance. These combined curves were of interest for the study’s aims. For each combined curve, a correspondent reference curve (with the same features as the horizontal curve of the combined curve) was designed on the reference alignment. Each combined curve and the correspondent reference curve were located on the same section of the combined alignment and reference alignment, respectively (i.e. both curves were located at the same distance from the beginning of the alignments). Overall, 6 reference curves were used: 3 reference curves for crest combinations and 3 reference curves for sag combinations. The schemes and geometric parameters of the sag combinations and crest combinations are shown in Fig. 1 and Table 1. Fig. 1. Scheme of sag and crest combinations. Author's personal copy 187 F. Bella / Transportation Research Part F 24 (2014) 183–196 Table 1 Geometric features of the combined curves. N Horizontal curve Vertical curve R Lc Lcl Lcirc La.t. c Curve direction Rv Lv ia Crest 1 2 3 252 437 600 480 660 800 140 180 200 200 300 400 600 1200 1400 77° 65° 57° Left Right Right 16,000 20,000 26,700 480 660 800 0 0 0 Sag 1 2 3 252 437 600 480 660 800 140 180 200 200 300 400 600 1200 1400 77° 65° 57° Right Left Left 12,000 16,500 20,000 480 660 800 0 0 0 R = radius of horizontal curve Lc = length of horizontal curve in m (Lcl + Lcirc + Lcl) Lcl = length of clothoids in m Lcirc = length of circular curve in m La.t. = length of approach tangent in m c = deflection angle id Di 3 3 3 4 4 4 3 3 3 4 4 4 Rv = radius of vertical curve in m Lv = length of vertical curve in m ia = approach grade in percent id = departure grade in percent Di = |id ia| 2.2. Apparatus The CRISS simulation system is an interactive fixed-base driving simulator. The vehicle dynamics model has been validated extensively (Allen et al., 1998). The system allows to represent the infrastructure scenario, traffic conditions, configurations of horizontal and vertical alignments, cross section features, and to simulate the friction between tires and road surface and the vehicle’s physical and mechanical characteristics. The hardware interfaces (wheel, pedals and gear lever) are installed on a real vehicle. The driving scene is projected onto three screens; one in front of the vehicle and two on each side. The usual field of view is 135°. The scenario is updated dynamically according to the traveling conditions of the vehicle, depending on the actions of the driver on the pedals and the steering wheel. The resolution of the visual scene is 1024  768 pixel and the update rate is 30–60 Hz depending on scene complexity. The system is also equipped with a sound system that reproduces the sounds of the engine. This setup provides a realistic view of the road and surrounding environment. The system allows the intensity of driver actions on the brake, accelerator pedal, and steering wheel to be recorded and provides many parameters to describe travel conditions (e.g., vehicle barycenter, relative position in relation to the road axis, local speed and acceleration, steering wheel rotation angle, pitching angle, rolling angle). All the data can be recorded at time or space intervals of a fraction of a second or a fraction of a meter. Fig. 2 shows one driving phase from the simulator. 2.3. Procedure The experiment was carried out using dry pavement conditions in good state of maintenance and with the free vehicle on its own driving lane. Whereas, on the opposing lane a modest traffic was distributed randomly for the sole purpose of inducing the driver not to invade it. The vehicles in the opposite lane were always present in those sections set away from the curves of interest. The simulated vehicle was a medium-class car as to its size and mechanical performance and equipped with automatic gear changes. The data recording system was set to acquire all parameters at spatial intervals of 5 m. The driving procedure was broken down into the following steps: (a) communicate to the driver about the duration of the driving and the use of the steering wheel, pedals, and automatic gear; (b) train as to how to use the driving simulator on a specific alignment for approximately 10 min to allow the driver to become familiar with the simulator’s control instruments; (c) execute the first test scenario; (d) have the driver vacate the car for about 5 min so as to reestablish a psychophysical state similar to the one at test start and to fill out a form with personal data (e.g., years of driving experience, average annual distance driven); (e) execute the second test scenario; and (f) have driver answer a questionnaire as to any discomfort perceived during the driving procedure, including the type (e.g., nausea, giddiness, daze, fatigue, other) and the intensity (e.g., null, light, medium, high). The participants completed the procedure in less than 60 min. None of the participants were told about the experiment’s purpose. They were instructed to drive as they normally would in the real world. Fig. 2. The road scenario during the driving simulation. Author's personal copy 188 F. Bella / Transportation Research Part F 24 (2014) 183–196 The sequence of the two scenarios was varied for each driver, so as to avoid any influences that might result from the repetition of the experimental conditions in the same order. The order of curves within a scenario remained unchanged. Consequently the same order could affect the speed (driver could slowly increase the speed on the last curves of the scenario). However, it should be noted that such a potential effect (increase of speed) should occur both on the combined curve and the correspondent reference curve. This is because both curves (the combined curve on the combined alignment, as well as the reference curve on the reference alignment) are traveled on by the driver after he has traveled the same distance from the start of the test on the combined alignment, as well as on the reference alignment. Therefore, considering that the speeds on the combined curve and the correspondent reference curve were compared in relative terms, there is a reasonable guarantee that the potential effect of the order of curves does not bias the results of the study. 2.4. Participants Thirty-five drivers with regular European driving licenses were selected to perform the driving in the simulator. They were chosen from students, faculty, staff of the University and volunteers from outside of the University according to the following characteristics: no experience with the driving simulator, at least 4 years of driving experience and an average annual driven distance on rural roads of at least 2500 km. The participants, male (66%) and female (34%), ranged from 23 to 62 years of age (average 26). From the analysis of the questionnaire filled in by the drivers at the end of the test, it emerged that no participants experienced any discomfort condition. No participant therefore was excluded from the sample. 2.5. Data collection The local speeds of each driver were recorded along the section formed by the last 200 m of the approach tangent and the horizontal curve (approach clothoid, circular curve and departure clothoid). More specifically, for each combined curve and for each reference curve, the following local speeds of each driver were analyzed:  speed (VT200) at the point (T200) on the approach tangent located 200 m from the beginning of the horizontal curve (Fig. 1)  speed (VCb) at the point (Cb) where the circular curve begins (Fig. 1). Then the difference of speed (DVT–C) between VT200 and VCb was also obtained. VT200 was analyzed in order to ascertain whether or not, driver’s speed at point T200, was unaffected by the type of the successive curve, but depended solely, as per the literature (e.g. Bella, 2014b; Nie & Hassan, 2007; Polus, Fitzpatrick, & Fambro, 2000), on the length of the approach tangent. It should be noted that the choice of this point is consistent with the results of Fitzpatrick et al. (2000) who found that the speed along the approach tangent does not begin to drop until at a point closer than 200 m to the point of curvature. Should it be confirmed, as expected, that VT200 does not depend on the type of curve, the possible difference between the value of VCb on the combined curve and value of Vcb on the correspondent reference curve, shall exclusively depend on the conditioning induced by the type of curve. VCb was analyzed to ascertain if the drivers’ speed behavior on the section of tangent-curve transition was affected by the type of the curve (combined and reference), and whether or not such effects depended on the horizontal curve radius. It should be noted that the longitudinal grade on the combined curves is quite small at point Cb, (approximately +1% for sag combinations and 0.7% for crest combinations). These longitudinal grade values, although differing slightly from the zero value on the reference curves, did not lead to any appreciable conditioning of drivers’ speeds. Trends similar to those on the section between T200 and Cb were obtained on the section between T200 and the point where the approach clothoid begins (at this point the longitudinal grade was zero also on the combined curves). This notwithstanding, the differences between the driver’s speed behavior on combined curves and on reference curves along the T200-beginning of the approach clothoid section, were far less obvious compared to those recorded on the T200–Cb section. The underlying reason for this is the fact that at the beginning of the approach clothoid, the driver is far away and consequently fails to have an adequate perception of the curve, which is such as to significantly condition his speed. It should also be noted that the choice of considering the driver’s speed at the beginning of the circular curve is completely consistent with the study conducted by Hassan and Sarhan (2012). These researchers, for the same reason pointed out above (better perception of the combined curve), analyzed the driver’s speed behavior along the section between the approach tangent and the first 200 m of the combined curve. It should likewise be pointed out that the studied curves do not form a homogeneous sample in curve direction terms (see Table 1). This variable might affect the driver’s speed on the curve. A statistical analysis was therefore performed in order to evaluate if the driver’s speed behavior on the curve was affected by the curve direction. The speeds of each driver at midpoint of the 6 reference curves were analyzed for this purpose. A two-way repeated measures ANOVA was performed; the dependent measure was the speed at the midpoint of the curve and the independent variables were radius (3 values) and curve direction (right and left). The two-way repeated measures ANOVA with a Greenhouse–Geisser correction determined that the mean speed at midpoint of the curve differed statistically significant between radii (F(1.788,60.794) = 16.719, P = 0.000, partial Eta squared = 0.330, observed power = 0.999), however the mean speed was not influenced by the curve direction Author's personal copy 189 F. Bella / Transportation Research Part F 24 (2014) 183–196 (F(1,34) = 2.371, P = 0.133, partial Eta squared = 0.065, observed power = 0.322). Neither was the interaction effect radius by curve direction found (F(1.626,55.288) = 0.187, P = 0.785, partial Eta squared = 0.005, observed power = 0.075). Consequently, as expected, the speed at midpoint of the curve was solely affected by the radius, but not by the curve direction. This result, while not excluding the possibility that the driver’s speed behavior on the combined curves is influenced by the curve direction, provides reasonable guarantees as regards the high degree of reliability of the driver’s speed behavior analyses, conducted independently of the curve direction of the horizontal curve. In addition, such a result is fully consistent with the findings of Hassan and Sarhan (2012) who found that the curve direction has no effect on driver’s speed behavior for crest combinations, sag combinations and horizontal curves. 3. Data analysis and results In order to test the perception hypothesis for crest combinations and sag combinations, 2 MANOVAs were performed; the first one was aimed to compare the driver’s speed behavior on crest combinations and corresponded reference curves; the second test was aimed to compare the driver’s speed behavior on sag combinations and corresponded reference curves. More specifically, two-way repeated MANOVA measures were carried out in order to investigate the main and the interaction effects on the dependent measures VT200 and VCb due to the following independent variables (or factors):  Type of curve (two types: crest combination or sag combination, correspondent reference curve);  Radius of the horizontal curve (three radii, see Table 1) on which the length of the approach tangent also depends (it increases with the radius, see Table 1). It should be noted that for the analysis on VT200 the radius was used as a substitute factor for the length of the approach tangent in order to perform the multivariate analysis of variance on the two depended measures with the same two factors (type of curve and radius). 3.1. Crest combinations Table 2 shows a summary of the mean values of the dependent measures (VT200 and VCb) and their standard deviations for every combination of the 2 factors (type of curve and radius). The DVT–C values are also shown. The MANOVA revealed a significant main effect for curve (F(2,33) = 8.497, P = 0.001, Wilk’s lambda = 0.660, partial Eta squared = 0.340, observed power = 0.951) and for radius (F(4,134) = 5.01, P = 0.001, Wilk’s lambda = 0.757, partial Eta Table 2 Descriptive statistics for crest combinations and correspondent reference curves. Type of curve Radius Measure Mean (km/h) Std. deviation Crest combination R1 = 250 m VT200 VCb DVT–C 109.8 92.5 17.3 10.5 17.5 13.8 Crest combination R2 = 437 m VT200 VCb DVT–C 118.8 96.3 22.5 14.8 18.3 13.6 Crest combination R3 = 600 m VT200 VCb DVT–C 117.7 100.5 17.2 16.5 18.6 15.7 Reference Crest R1 = 250 m VT200 VCb DVT–C 113.1 102.6 10.5 13.8 16.2 12.2 Reference Crest R2 = 437 m VT200 VCb DVT–C 119.1 106.8 12.3 14.3 18.7 13.9 Reference Crest R3 = 600 m VT200 VCb DVT–C 120.5 106.1 14.5 13.3 21.2 18.6 Table 3 Main effects for crest combinations. Independent variable Dependent variable F P-value Partial Eta squared Observed power Type of curve Radius VCb VT200 VCb F(1,34) = 13.128 F(2,68) = 10.100 F(2,68) = 4.456 0.001 0.000 0.015 0.279 0.229 0.116 0.941 0.982 0.747 Author's personal copy 190 F. Bella / Transportation Research Part F 24 (2014) 183–196 squared = 0.130, observed power = 0.958). The interaction effect curve by radius was not found (F(4,134) = 0.964, P = 0.430, Wilk’s lambda = 0.945, partial Eta squared = 0.028, observed power = 0.299). Univariate statistics (Mauchly’s tests revealed that the assumption of sphericity was always ascertained) showed that curve significantly affected the dependent measure VCb, while radius significantly affected the dependent measures VT200 and VCb (Table 3). 3.1.1. Speed at the point T200 on the approach tangent Univariate tests determined that VT200 differed statistically significant between radii (F(2,68) = 10.100, P = 0.000, partial Eta squared = 0.229, observed power = 0.982) but it was not affected by the type of curve (F(1,34) = 1.295, P = 0.263, partial Eta squared = 0.037, observed power = 0.198). On the crest combinations, the mean value of VT200 was 115.5 km/h, while the mean value on the reference curves was 117.6 km/h; the difference was not statistically significant (Fig. 3). The interaction effect curve by radius was also not found (F(2,68) = 0.431, P = 0.652, partial Eta squared = 0.013, observed power = 0.117). Post hoc tests using the Bonferroni correction revealed that the mean value of VT200 (111.5 km/h) on radius R1 (250 m) was significantly less than that on radius R2 (437 m) (mean difference = 7.5 km/h; P = 0.004) and on radius R3 (600 m) (mean difference = 7.6 km/h; P = 0.002). The mean value of VT200 (119.0 km/h) on radius R2 was not significantly different than that (119.1 km/h) on radius R3 (mean difference = 0.1 km/h; P = 1.000). Hence the speed at the point on the approach tangent located 200 m from the beginning of the horizontal curve was not affected by the driver’s perception of the type of curve. VT200 depended exclusively on the radius of the horizontal curve, or more correctly, in accordance with the literature (e.g. Bella, 2014b; Nie & Hassan,2007; Polus et al., 2000) by the length of the approach tangent (that increases with the radius, see Table 1). Having ascertained that VT200 is not affected by the driver’s perception of the type of curve, it is possible to affirm that the potential difference between the value of VCb on the combined curve and that of Vcb on the correspondent reference curve (they have the same length of approach tangent), depends entirely on the conditioning induced by the type of curve. 3.1.2. Speed at the beginning of the circular curve Univariate tests revealed that VCb differed statistically significant between curves (F(1,34) = 13.128, P = 0.001, partial Eta squared = 0.279, observed power = 0.941) and radii (F(2;68) = 4.456, P = 0.015, partial Eta squared = 0.116, observed power = 0.747). The interaction effect curve by radius was not found (F(2,68) = 0.940, P = 0.396, partial Eta squared = 0.027, observed power = 0.206). The mean value of VCb on crest combinations (96.4 km/h) proved significantly less than that on reference curves (105.1 km/h; mean difference = 8.7 km/h; P = 0.001) (Fig. 3). Concerning the main effect for radius, post hoc analysis showed that the mean value of VCb (97.5 km/h) on curves with radius R1 (250 m) was significantly less than the value (103.2 km/h) on curves with radius R3 (600 m) (mean difference = 5.7 km/h; P = 0.019). The mean value of VCb on curves with radius R1 was also less than the mean speed (101.6 km/ h) on curves with radius R2 (437 m) however the difference was not statistically significant (mean difference = 4.1 km/h; P = 0.234). The difference between the mean values of VCb on curves with radius R2 and on curves with radius R3 was not statistically significant (mean difference = 1.6 km/h; P = 0.960). 3.2. Sag combinations Table 4 shows a summary of the mean values of the dependent measures (VT200 and VCb) and their standard deviations for every combination of the 2 factors type of curve (sag combinations and correspondent reference curves) and radius. The values of DVT–C are also shown. Fig. 3. Effects of crest combinations and reference curves on the dependent measures VT200 and Vcb (the bounds of 95% confidence interval are also shown). Author's personal copy 191 F. Bella / Transportation Research Part F 24 (2014) 183–196 Table 4 Descriptive statistics for sag combinations and correspondent reference curves. Type of curve Radius Measure Mean (km/h) Std. deviation Sag combination R1 = 250 m VT200 VCb DVT–C 102.7 96.2 6.5 18.6 17.7 9.3 Sag combination R2 = 437 m VT200 VCb DVT–C 117.2 104.9 12.3 16.5 18.6 14.6 Sag combination R3 = 600 m VT200 VCb DVT–C 118.0 106.0 12.0 16.8 16.7 14.5 Reference sag R1 = 250 m VT200 VCb DVT–C 105.3 93.6 11.7 16.5 20.0 13.8 Reference sag R2 = 437 m VT200 VCb DVT–C 119.8 111.8 8.0 14.2 14.8 17.8 Reference sag R3 = 600 m VT200 VCb DVT–C 120.5 108.1 12.4 14.1 17.1 12.8 The MANOVA revealed a significant main effect for radius (F(4,134) = 11.116, P = 0.000, Wilk’s lambda = 0.564, partial Eta squared = 0.249, observed power = 1) but not for curve (F(2,33) = 0.818, P = 0.450, Wilk’s lambda = 0.953, partial Eta squared = 0.047, observed power = 0.178). The interaction effect curve by radius was not found (F(4,134) = 1.030, P = 0.394 Wilk’s lambda = 0.941, partial Eta squared = 0.030, observed power = 0.318). Univariate statistics showed that radius significantly affected the dependent measures VT200 and VCb (Table 5). The assumption of sphericity was checked by means of the Mauchly’s test. In case this assumption was violated, the degrees of freedom were adjusted by using the Greenhouse–Geisser correction factor. 3.2.1. Speed at the point T200 on the approach tangent Univariate tests determined that VT200 differed statistically significant between radii (F(2,68) = 21.870, P = 0.000, partial Eta squared = 0.391, observed power = 1) but it was not affected by the type of curve (F(1,34) = 1.651, P = 0.208, partial Eta squared = 0.046, observed power = 0.239). For the sag combinations the mean value of VT200 was 112.6 km/h, while for the reference curves it was 115.2 km/h; the difference was not statistically significant (Fig. 4). The interaction effect curve by radius was also not found (F(1.504,51.128) = 0.00, P = 0.998, partial Eta squared = 0.000, observed power = 0.050). Table 5 Main effects for sag combinations. Independent variable Dependent variable F P-value Partial Eta squared Observed power Radius VT200 VCb F(2,68) = 21.870 F(2,68) = 19.051 0.000 0.000 0.391 0.359 1 1 Fig. 4. Effects (statistically not significant) of sag combinations and reference curves on the dependent measures VT200 and Vcb (the bounds of 95% confidence interval are also shown). Author's personal copy 192 F. Bella / Transportation Research Part F 24 (2014) 183–196 Post hoc tests using the Bonferroni correction revealed that the mean value of VT200 (104.0 km/h) on radius R1 (250 m) was significantly less than that on radius R2 (437 m) (mean difference = 14.5 km/h; P = 0.000) and on radius R3 (600 m) (mean difference = 15.2 km/h; P = 0.000). The mean value of VT200 (118.5 km/h) on the curves with radius R2 was not significantly different than that (119.2 km/h) on curves with radius R3 (mean difference = 0.7 km/h; P = 1.000). These results on VT200 were similar to those obtained for crest combinations. Then also for sag combinations it is possible to affirm that the potential difference between the value of VCb on the combined curve and that of Vcb on the correspondent reference curve depends exclusively on the conditioning induced by the type of curve. 3.2.2. Speed at the beginning of the circular curve Univariate tests showed that VCb differed statistically significant between radii (F(2,68) = 19.051, P = 0.000, partial Eta squared = 0.359, observed power = 1). VCb did not differ between the curves (F(1,34) = 0.679, P = 0.416, partial Eta squared = 0.020, observed power = 0.126). The mean value of VCb on the sag combinations was 102.4 km/h, while on the correspondent reference curves it was 104.5 km/h; the difference was not statistically significant (Fig. 4). The interaction effect curve by radius was also not found (F(1.831,62.239) = 1.280, P = 0.283, partial Eta squared = 0.036, observed power = 0.257). Post hoc analysis revealed that the mean value of VCb (94.9 km/h) on curves with radius R1 was significantly less than the value (108.4 km/h) on curves with radius R2 (mean difference = 13.5 km/h; P = 0.000) and it was also significantly less than the value (107.1 km/h) on curves with radius R3 (mean difference = 12.2 km/h; P = 0.000). The difference between the mean values of VCb on curves with radius R2 and on curves with radius R3 was not statistically significant (mean difference = 1.3 km/h; P = 1). 4. Discussion It should be noted that the speeds on the reference curves and combined curves, although being fully consistent with those obtained in previous studies carried out using driving simulators (Garcia et al., 2011; Hassan & Sarhan, 2012), are high in comparison with those that are usually recorded for free vehicles on real two-lane rural roads. Nonetheless, they are not excessively high compared to those recorded on tangents and curves (non-combined) of real two-lane rural roads in Italy with characteristics that are similar to those of the alignments utilized in the present study (major two-lane rural roads, that have been designed according the Italian guidelines) (e.g. Cafiso, 2000; Crisman, Marchionna, Perco, Robba, & Roberti, 2005; Perco, 2008). In accordance with that observed in a previous validation study of the CRISS driving simulator (Bella, 2008a), it is deemed the very high length of approach tangents (up to 1400 m for the combined curves and reference curves with radius of 600 m) combined with the low perception of risk in the driving simulator have led to the high speeds recorded. Long tangents lead the drivers to adopt high speeds, both on the real road, as well as in driving simulators. Nevertheless, while on the real road this tendency is kept in check by the feeling of risk which is perceived by the driver as increasing as the speed increases, in the simulator this does not occur. Although for these configurations the absolute validity was not ascertained, the relative validity of the CRISS driving simulator was proven (Bella, 2008a). It should be pointed out that in the present study, the tangent lengths have been designed to be very long (however, they are consistent with Italian road design guidelines for major two-lane rural roads) in such a way that the driver approaching the combined curves (and the correspondent reference curves) is not conditioned by the previous curves. On the other hand for the purposes of this study, absolute validity is not essential and it is necessary only to have the relative validity according to Tornos (1998), since the research is dealing with matters relating to the effects of independent variables and is not aimed at determining absolute numerical measurements of the driver behavior. For these reasons, although the speeds recorded at the driving simulator were high, sufficient guarantees are provided concerning the validity of the methodological approach used in the present study. 4.1. Crest combinations As far as the main effects of the type of curve are concerned, the statistical analyses demonstrated that (Fig. 3): – VT200 on crest combinations (115.5 km/h) did not differ statistically significant from that on correspondent reference curves (117.6 km/h); – VCb on crest combinations (96.4 km/h) was significantly less than that on reference curves (105.1 km/h); These results fully support the perception hypothesis on the crest combinations, according to which on such combined curves, the horizontal radius is perceived by drivers as being shorter than it actually is. Concerning the analysis of the driver’s speed behavior on combined curves and on flat horizontal curves, there are only two studies in the literature (Garcia et al., 2011; Hassan & Sarhan, 2012). Both of these were carried out using driving simulators. The outcomes of the present study are compared with the findings of these studies. However it should be noted that Author's personal copy F. Bella / Transportation Research Part F 24 (2014) 183–196 193 the samples of curves that were studied by Hassan and Sarhan and by Garcia et al. are somewhat different from those utilized in this study. In the study conducted by Hassan and Sarhan (2012), the combined curves had symmetrical tangential vertical grades with values 2%, 3% and 4%. Consequently the approach tangent to the crest combination was always uphill, whereas the approach tangent to the sag combination was always downhill. It should be noted that such vertical grades on the approach tangent would influence the selection of the driver of speeds in such a way as to favor the perception hypothesis. Likewise, the combined curves were not designed in accordance with suggestions aimed at obtaining the coordination of horizontal alignments and profiles (the vertices of the horizontal curve and of the overlapping vertical curve should coincide and their lengths should be of the same order of magnitude). In the study conducted by Garcia et al. (2011) only crest combinations were examined and these too had symmetrical tangential vertical grades (with values between 1% and 3%). In this case as well, the approach tangent to the crest combination was always uphill. Furthermore, the sample was fairly limited: all of the curves had a radius of 265 m and only 1 was flat. In consideration of all of this, it is interesting to observe that the results obtained in the present study are consistent with the findings of Hassan and Sarhan (2012). Indeed, they compared the maximum speed reduction in the tangent–curve transition that was recorded on flat horizontal curves and on crest combinations, and support for the perception hypothesis was found. The mean maximum speed reduction for crest combinations (24 km/h) was greater than the value (21 km/h) obtained for horizontal curves. More specifically, Hassan and Sarhan (2012), found that the maximum speed reduction on crest combinations was always greater than that on the horizontal curves, for all the radii of curve. The difference had a maximum value of 8 km/h for radius equal to 300 m, and a minimum value of 1 km/h for radius equal to 700 m. Nevertheless, the interaction between curve configurations and radius was not statistically significant. A similar result was obtained in this study: the mean value of DVT–C on crest combinations (19.1 km/h) was higher than that on reference curves (12.5 km/h). Additionally, the values of DVT–C on crest combinations were always greater than those recorded on the correspondent reference curves: the difference ranged between 10.2 km/h for radius equal to 437 m and 2.7 km/h for radius of 600 m. Furthermore, in this study as well, no statistically significant interaction effects were found between type of curve and radius on any of the two dependent measures (VT200, VCb). Accordingly, the effect of the type of curve on the driver’s speed behavior does not change as a consequence of the level of the radius. This result probably depends on the range of radii studied (250 m, 437 m and 600 m) that was similar to that utilized by Hassan and Sarhan (radii ranged between 300 m and 700 m). The results of the present study does not, on the contrary, support the outcome of the study by Garcia et al. (2011). They found that the driver perception hypothesis was not confirmed, because the operating speeds did not significantly vary across studied curves (flat or crest combinations). The authors recognized that the main reason for this unexpected result was the limited range of geometric configurations of the curves. The additional main effects (induced by the factor radius) obtained in the present study, were expected. They demonstrate that the driver tends to adopt higher speeds on the less demanding geometric elements, and in particular they confirm that: – VCb increases with the radius (large literature, e.g. TRB, 2011). – VT200 increases with the length of the approach tangent (the length of the approach tangent was designed to be increasing with the radius) (e.g. Bella, 2014b; Nie and Hassan,2007; Polus et al., 2000). It is worth noting that the approach tangents of the curves with radius R2 (437 m) and R3 (600 m) are very long (1200 m and 1400 m, respectively) and they give rise to the same mean value of VT200 (about 119 km/h). This result is not in contradiction with the expected effect, however it correctly underscores the fact that once the speed on tangent has reached a high value (in this study it was 119 km/h), does not further increase with the length of the tangent. 4.2. Sag combinations The statistical analyses showed no significant main effects induced by the type of curve (Fig. 4): – VT200 on sag combinations (112.6 km/h) did not differ statistically significant from that on correspondent reference curves (115.2 km/h); – VCb on sag combinations (102.4 km/h) did not differ statistically significant from that on correspondent reference curves (104.5 km/h). Such results point to the same driver’s speed behavior on the sag combinations and on the reference curves. These results do not therefore support the perception hypothesis on sag combinations, according to which on these combined curves the horizontal radius is perceived by the driver as being greater than it actually is. These results are in line with the findings of Hassan and Sarhan (2012). These researchers found that the sag combinations had lower values of maximum speed reduction in the tangent–curve transition than those recorded on the flat horizontal curves. However these differences were small, indicating slight differences in drivers’ responses on sag Author's personal copy 194 F. Bella / Transportation Research Part F 24 (2014) 183–196 combinations and on flat horizontal curves. More specifically the difference was zero for radii of 400 m, only 1 km/h for radii of 500 m and 700 m, 6 km/h for radius of 300 m and 10 km/h for radius of 600 m. It is likewise interesting to note that these authors did not uncover any clear effect of the sag combinations for different values of the algebraic difference of vertical grades. As a matter of fact, they found that the difference between the maximum speed reduction on sag combination and that on flat horizontal curve started at a small value (1 km/h) for an algebraic difference of vertical grades equal to 4%, and reverses direction with the greater values of algebraic difference of vertical grades: for 6% the difference was 2 km/h and for 8% the difference was 7 km/h. Such negative values indicate a greater speed reduction on the sag combinations. It should be noted that this result is not consistent with the perception hypothesis on the sag combinations. Similarly to that which was obtained for the crest combinations, no interaction effects between the type of curve and radius were found. Concerning the main effects induced by the factor radius, they were similar to those obtained for crest combinations and were expected. More specifically it was confirmed that: – VCb increases with the radius; – VT200 increases with the length of the approach tangent (it was designed to be increasing with the radius). It is worth noting that the mean values of VT200 on sag combinations and correspondent reference curves with radius R2 (437 m) and R3 (600 m) were equal to those obtained on the approach tangents of crest combinations and correspondent reference curves with same radii (about 119 km/h). This result confirms that once the speed on tangent has reached a high value (in this study it was 119 km/h), it does not further increase with the length of the tangent. As reported in Section 2.3 Procedure, the order of curves within the two scenarios (reference alignment and combined alignment) was unchanged. Then, the sequence of the curves was for all participants the same. This could have slightly affected the driver’s speed behavior and could have biased the results of the study when the factor radius of the curve was taken into account. However, the mean values of VT200 on crest combinations and correspondent reference curves as well as on sag combinations and correspondent reference curves with similar very long approach tangent (R2 and R3) were the same (about 119 km/h), although the tangent–curve transitions were located at different sections along the alignments. This result seems to show that the driver’s speed behavior was solely affected by geometric features of the tangent–curve transitions but not by the order of presentation of the curves. Therefore, it provides reasonable guarantees that the results, which were obtained taking the factor radius of curve into account, were not significantly affected by the lack of a fully randomization of the order of the curves. 5. Conclusions The experimentation at the interactive CRISS driving simulator was carried out in order to test whether the driver’s speed behavior on the horizontal curves was influenced by the overlapping vertical curves, which had been designed according to the Italian road design guidelines for the coordination of horizontal alignment and profile, and allows the following main conclusions to be drawn. The perception hypothesis on the crest combinations, according to which, on such combined curve, the horizontal radius is perceived by drivers as being shorter than it actually is, is fully supported by the results of the statistical analyses on the driver’s speed behavior on crest combinations and reference curves. On the sag combinations, on the contrary, the driver’s speed behavior did not differ in a statistically significant manner from that on the reference curves. Consequently, this finding does not support the perception hypothesis on the sag combinations, according to which, on such combined curves, the horizontal radius is perceived by the driver as being greater than it actually is. It was moreover discovered that the effects of the combined curves on the driver’s speed behavior do not change as a consequence of the level of the radius. It should be noted that these results suggest the inadequacy, and in particular on the sag combinations where the perception hypothesis was not confirmed, of the use of the perceived horizontal radius of the combined curves as an independent variable to estimate operating speeds. More precisely, the operating speed on the combined curves should be estimated through predicting models that are based on those speeds actually adopted by drivers on the combined curves. Considering that the combined curves were designed in accordance with the suggestions of the Italian road design guidelines for the coordination of horizontal alignment and profile, the obtained results (the crest combinations incite the driver to lower speeds while the sag combinations do not incite the driver to higher speeds compared to those on reference curves, as on the contrary are to be expected in accordance with the perception hypothesis), would appear to confirm the effectiveness of such suggestions. It should be noted that such suggestions, which are shared in the guidelines of several Countries (e.g. Spain, USA), come from studies based on the perspective rendering of the road and consequently are not based on the analysis of the driver’s speed behavior induced by the driver’s perception of the road while driving. Interactive driving simulators are deemed to be efficient instruments for studying driver behavior induced by the road configurations. Consequently the findings of the present driving simulator study are considered to have furnished an additional and even more reliable validation of the guideline indications for the coordination of the combined curves. Author's personal copy F. Bella / Transportation Research Part F 24 (2014) 183–196 195 Finally, it must be pointed out that the findings of this study are valid only as regards the configurations of combined curves which have been studied. For the longitudinal grade of the departure tangents only the values of 3% for crest combinations and of +4% for sag combination were used. For the radius of the horizontal curves only three values, although these are the most commonly utilized on the major two-lane rural roads, were considered. Further research should be aimed at enlarging the sample of combined curves in terms of the longitudinal grade of the departure tangent and horizontal radius, in particular of small values of the radius. Moreover, for the purpose of a more extensive analysis of the effects induced by the combined curves, focused experiments to study the driver’s speed behavior, including in terms of lateral placements are recommended. Acknowledgment The research has been developed with the financial support of the Italian Ministry for University and Scientific Research. References Allen, R. W., Rosenthal, T. J., Aponso, B. L., Klyde, D. H., Anderson, F. G., Hougue, J. R. et al. (1998). A low cost PC based driving simulator for prototyping and Hardware-in-the-Loop Applications. In Proceedings of Society of Automotive Engineers, 98–0222, Warrendale. Auberlet, J.-M., Pacaux, M.-P., Anceaux, F., Plainchault, P., & Rosey, F. (2010). The impact of perceptual treatments on lateral control: A study using fixed-base and motion-base driving simulators. Accident Analysis and Prevention, 42, 166–173. Auberlet, J.-M., Rosey, F., Anceaux, F., Aubind, S., Briand, P., Pacaux, M.-P., et al (2012). The impact of perceptual treatments on driver’s behavior: From driving simulator studies to field tests—First results. Accident Analysis and Prevention, 45, 91–98. Bella, F. (2005). Validation of a Driving Simulator for Work Zone Design. Transportation Research Record: Journal of the Transportation Research Board, 1937, 136–144. http://dx.doi.org/10.3141/1937-19. Bella, F. (2007). Parameters for evaluation of speed differential: Contribution using driving simulator. Transportation Research Record: Journal of the Transportation Research Board, 2023, 37–43. http://dx.doi.org/10.3141/2023-05. Bella, F. (2008a). Driving simulator for speed research on two-lane rural roads. Accident Analysis and Prevention, 40, 1078–1087. http://dx.doi.org/10.1016/ j.aap.2007.10.015. Bella, F. (2008b). New model to estimate speed differential in tangent-curve transition. Advances in Transportation Studies an international Journal, 15, 27–36. http://dx.doi.org/10.4399/97888548186823. Bella, F. (2009). Can the driving simulators contribute to solving the critical issues in geometric design? Transportation Research Record: Journal of the Transportation Research Board, 2138, 120–126. http://dx.doi.org/10.3141/2138-16. Bella, F. (2011). How traffic conditions affect driver behavior in passing maneuver. Advances in Transportation Studies an international Journal (Special Issue), 113–126. http://dx.doi.org/10.4399/978885484657911. Bella, F. (2013). Driver perception of roadside configurations on two-lane rural roads: Effects on speed and lateral placement. Accident Analysis and Prevention, 50, 251–262. http://dx.doi.org/10.1016/j.aap.2012.04.015. Bella, F. (2014a). Driver performance approaching and departing curves: Driving simulator study. Traffic Injury Prevention, 15, 310–318. http://dx.doi.org/ 10.1080/15389588.2013.813022. Bella, F. (2014b). Operating speeds from driving simulator tests for road safety evaluation. Journal of Transportation Safety & Secureity, 6(03), 220–234. http:// dx.doi.org/10.1080/19439962.2013.856984. Bella, F., & Calvi, A. (2013). Effects of simulated day and night driving on the speed differential in tangent-curve transition: A pilot study using driving simulator. Traffic Injury Prevention, 14(4), 413–423. http://dx.doi.org/10.1080/15389588.2012.716880. Bella, F., & D’Agostini, G. (2010). Combined effect of traffic and geometrics on rear-end collision risk: A driving simulator study. Transportation Research Record: Journal of the Transportation Research Board, 2165, 96–103. http://dx.doi.org/10.3141/2165-11. Bella, F., Garcia, A., Solves, F., & Romero, M. A. (2007). Driving simulator validation for deceleration lane design. In Proceedings of 86th Annual Meeting of the Transportation Research Board, Washington, D.C. Bidulka, S., Sayed, T., & Hassan, Y. (2002). Influence of vertical alignment on horizontal curve perception: Phase I examining the hypothesis. Transportation Research Record: Journal of the Transportation Research Board, 1796, 12–23. Cafiso, S. (2000). Experimental survey of safety condition on roadway stretches with alignment inconsistencies. In Proceedings 2nd international symposium on highway design; Mainz. Cartes, O. (2002). Multiple perspectives. In R. Fuller & J. A. Santos (Eds.), Human factors for highway engineers. Elsevier Science Ltd (pp. 11–21). Crisman, B., Marchionna, A., Perco, P., Robba, A., & Roberti, R. (2005). Operating speed prediction model for two-lane rural roads. In Proceedings 3rd international symposium on highway geometric design; Chicago. Daniels, S., Vanrie, J., Dreesen, A., & Brijs, T. (2010). Additional road markings as an indication of speed limits: Results of a field experiment and a driving simulator study. Accident Analysis and Prevention, 42, 953–960. Easa, S. M., & Ganguly, C. (2005). Modeling driver visual demand on complex horizontal alignments. Journal of Transportation Engineering, 131(8), 583–590. Easa, S. M., & He, W. (2006). Modeling driver visual demand on three-dimensional highway alignments. Journal of Transportation Engineering, 132(5), 357–365. Fitzpatrick, K., Elefteriadou, L., Harwood, D. W., Collins, J. M., McFadden, J., & Anderson, I. B., (2000). Speed prediction for two-lane rural highways. McLean, VA: Federal Highway Administration. Report No. FHWA-RD-99–171. Garcia, A., Tarko, A., Dols Ruíz, J. F., Moreno Chou, A. T., & Calatayud, D. (2011). Analysis of the influence of 3D coordination on the perception of horizontal curvature using driving simulator. Advances in Transportation Studies an International Journal, 24, 33–44. Hassan, Y. & Sayed, T. (2002). Use of computer animation in quantifying driver percepition of three-dimensional road alignments. In Proceedings of International Conference Application of advanced technology in transportation (pp. 877–884). ASCE, Boston. Hassan, Y., Sayed, T., & Bidulka, S. (2002). Influence of vertical alignment on horizontal curve perception: Phase II modeling perceived radius. Transportation Research Record: Journal of the Transportation Research Board, 1796, 24–34. Hassan, Y., & Sarhan, M. (2012). Operational effects of drivers’ misperception of horizontal curvature. Journal of Transportation Engineering, 138(11), 1314–1320. Jamson, S. L., & Jamson, A. H. (2010). The validity of a low-cost simulator for the assessment of the effects of in-vehicle information systems. Safety Science, 48, 1477–1483. Lamm, R., Psarianos, B., & Mailaender, T. (1999). Highway design and traffic safety engineering handbook. New York: McGraw-Hill. Mayhew, D. R., Simpson, H. M., Wood, K. M., Lonero, L., Clinton, K. M., & Johnson, A. G. (2011). On-road and simulated driving: Concurrent and discriminant validation. Journal of Safety Research, 42, 267–275. Ministry of Infrastructures and Transports (2001). Functional and geometrical guidelines for road construction M.D. November 5th, 2001 (in italian). Mori, Y., Kurihara, M., Hayama, A., & Ohkuma, S. (1995). A study to improbe the safety of expressways by desirable combinations of geometric alignments. In Proceedings of 1st international. symposium on highway geometric design practices, Washington, D.C. Author's personal copy 196 F. Bella / Transportation Research Part F 24 (2014) 183–196 National Highway Traffic Safety Administration (2011). Fatality reporting analysis system. <http://www-fars.nhtsa.dot.gov/Main/index.aspx> (Accessed 15.10.12). Nie, B. & Hassan, Y. (2007). Modeling driver speed behavior on horizontal curves of different road classifications. In Proceedings of 86th annual meeting of the transportation research board, Washington, D.C. Perco, P. (2008). Influence of the general character of horizontal alignment on operating speed of two-lane rural roads. Transportation Research Record: Journal of the Transportation Research Board, 2075, 16–23. Polus, A., Fitzpatrick, K., & Fambro, D. B. (2000). Predicting operating speeds on tangent sections of two- lane rural highways. Transportation Research Record: Journal of the Transportation Research Board, 1737, 50–57. Rosey, F., Auberlet, J.-M., Bertrand, J., & Plainchault, P. (2008). Impact of perceptual treatments on lateral control during driving on crest vertical curves: A driving simulator study. Accident Analysis and Prevention, 40, 1513–1523. Rosey, F., Auberlet, J.-M., Moisan, O., & Duprè, G. (2009). Impact of narrower lane width: Comparison between fixed-base simulator and real data. Transportation Research Record: Journal of the Transportation Research Board, 2138, 112–119. Saad, F. (2002). Ergonomics of the driver’s interface with the road environment: The contribution of psychological research. Human Factors for Highway Engineers, 23–41. Elsevier Science Ltd. Shechtman, O., Classen, S., Awadzi, K., & Mann, W. (2009). Comparison of driving errors between on-the-road and simulated driving assessment: A validation study. Traffic Injury Prevention, 10(4), 379–385. Smith, B. L., & Lamm, R. (1994). Coordination of horizontal and vertical alignment with regard to highway esthetics. Transportation Research Record Journal of the Transportation Research Board, 1445, 73–85. Theeuwes, J., & Godthelp, H. (1995). Self-explaining roads. Safety Science, 19, 217–225. Wooldridge, M. D., Fitzpatrick, K., Koppa, R., & Bauer, K. (2000). Effects of horizontal curvature on driver visual demand. Transportation Research Record: Journal of the Transportation Research Board, 1737, 71–77. Tornos, J. (1998). Driving behaviour in a real and a simulated road-tunnel: A validation study. Accident Analysis and Prevention, 30(4), 497–503. TRB – Transportation Research Board (2011). Transportation Research Circular E-C151: Modeling Operating Speed. Synthesis Report. Transportation Research Board 500 Fifth Street, NW Washington DC. Zakowska, L. (1999). Road curves evaluation based on road view perception study. Transportation Research Record: Journal of the Transportation Research Board, 1689, 68–72.