Drivers’ speed behaviour in real and simulated urban roads – A validation study

Transportation Research Part F 49 (2017) 1–17

Contents lists available at ScienceDirect

Transportation Research Part F journal homepage: www.elsevier.com/locate/trf

Drivers’ speed behaviour in real and simulated urban roads – A validation study Valentina Branzi ⇑,1, Lorenzo Domenichini 1, Francesca La Torre 1 Department of Civil and Environmental Engineering, University of Florence, Via Santa Marta 3, 50139 Firenze, Italy

a r t i c l e

i n f o

Article history: Received 10 June 2016 Received in revised form 26 May 2017 Accepted 3 June 2017

Keywords: Urban road Driving simulator Behavioural validity Field study Speed monitoring

a b s t r a c t Traffic accidents and injuries constitute a growing problem for Europe’s urban transportation system. Driving simulators can be active support tools in the design process of urban driving environments, provided that they are appropriately validated. The aim of this study was to establish the behavioural validity, in relative and absolute terms, of the motion-base driving simulator of the Road Safety and Accident Reconstruction Laboratory (LaSIS) of the University of Florence in order to use it to evaluate the effectiveness of the urban road safety treatment design process. The research was conducted by comparing the driving speed collected in field with those recorded in the 3D virtual reality experiments through conventional and integrative statistical methods. Speeds were recorded at twenty-one measurement sites located in homogeneous road sections characterized by a less or more demanding driving environment; thirty-four participants drove the virtual scenario which reproduced the real situation. The results of the comparative and conventional statistical analysis established the relative validity and also revealed that absolute validity was obtained in the measurement sites in which physical constraints are imposed by engineering treatments. The integrative regression analysis confirms these outcomes and showed that the simulation system is a reliable predictor of the real speed data also in absolute terms. It was also found that the driving experience levels significantly affect the speed adopted by the motorists in the virtual environment. These findings support the use of the driving simulator as a powerful approach to predict of the safety effectiveness of design solutions in urban areas, thus allowing important cost savings. Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction Urban areas are very complex driving environments in which different types of road users compete continually. The constant interaction between motorized traffic and vulnerable road users creates serious implication for traffic safety. In the last few years the problem of road crashes in urban areas has become increasingly relevant. In 2013, 26,090 people were killed in road accident throughout the European Union (EU),2 9923 of whom were killed in accidents on urban roads in ⇑ Corresponding author. Fax: +39 0552758800. E-mail addresses: [email protected] (V. Branzi), [email protected] (L. Domenichini), [email protected] (F. La Torre). Fax: +39 0552758800. 2 The European Union (EU) currently includes 28 countries and it has a population of approximately 740 million inhabitants. About 300 million passenger cars are present in the EU. The EU has almost one car for every citizens and the average annual mileage per European is 13,000 km. 1

http://dx.doi.org/10.1016/j.trf.2017.06.001 1369-8478/Ó 2017 Elsevier Ltd. All rights reserved.

2

V. Branzi et al. / Transportation Research Part F 49 (2017) 1–17

the EU. This corresponds to 38% of the road fatalities in 2013. Although the total number of fatalities within urban areas decreased since 2005, the proportion has slightly increased (from 35% to 38%) (Traffic Safety Basic Facts, 2016). In Italy the situation is even more serious. According to Italian data published by ACI-ISTAT in 2014 (ACI-ISTAT, 2015), the majority of fatal and injury accidents occurred on urban roads: 133,598 accidents, corresponding 75.5% of the total accidents with injured and/or fatalities occurred in the whole Italian road network. These accidents caused the death of 1505 people, corresponding 44.5% of the total number of victims of road traffic accidents on the Italian road network (ACI-ISTAT, 2015). Speeding has been identified as a key risk factor in traffic accidents on urban roads, influencing both the frequency and the severity of the injuries resulting from crashes. Many safety measures have been proposed to tackle this problem whose effectiveness depends on the characteristics of each specific urban environment and their interaction with the driver’s capabilities and expectations. The use of an interdisciplinary approach based on driving simulations is a promising method to study these interactions, to check the safety impact of the proposed interventions and to identify the most promising design alternatives. The use of driving simulator provides important advantages in terms of experimental control, capabilities effectiveness, cost, ease of data collection and safety of the study implementation. However, it presents also limitations mainly related to the reliability of the data acquired. For this reason, any driving simulator study should be preceded by questioning whether the simulator is sufficiently valid for the task or ability to be investigated (Kaptein, Theeuwes, & Van Der Horst, 1996). A wide range of driving parameters, environments and behaviours have been examined in driving simulator’s validation studies and generally a good correspondence between performance in the real world and during driving simulator experiments was obtained. Speed is the most commonly examined parameter in safety studies concerning road tunnels (Cao, Wang, & Luo, 2015; Törnros, 1998), rural roads (Bella, 2008; Bittner Jr., Simsek, Levison, & Campbell, 2002; Godley, Triggs, & Fildes, 2002; Santos, Merat, Mouta, Brookhuis, & De Waard, 2005), highway work zones (Bella, 2005; Bham, Leu, Vallati, & Mathur, 2014; McAvoy, Schattler, & Datta, 2007), signalized intersections (Yan, Abdel-Aty, Radwan, Wang, & Chilakapati, 2008) and speed management measures (Godley et al., 2002; Riemersma, van der Horst, Hoekstra, Alink, & Otten, 1990). Few studies considered other indicators such as vehicle’s lateral position (Blana & Golias, 2002; Wade & Hammond, 1998), driver’s reaction time (Brown, Dow, Marshall, & Allen, 2007; Engen, 2008; Hoffman, Brown, Lee, & McGehee, 2002), time gap (Abe & Richardson, 2006; Boer, Ward, Manser, Yamamura, & Kuge, 2005) and driving errors (Meuleners & Fraser, 2015). The validity of the simulation tools for assessing driving behaviour of specific driver groups, such as young drivers (Brown et al., 2007; De Winter et al., 2009; Hoffman et al., 2002; Mayhew et al., 2011; Shechtman, Classen, Awadzi, & Mann, 2009), older drivers (Hekamies-Blomqvist, Östlund, Henriksson, & Heikkinen, 2001; Lee, Cameron, & Lee, 2003; Lee, Lee, & Cameron, 2003) and subjects with temporary or permanent disabilities (Lee et al., 2007; Lew et al., 2005) has been investigated also. Few studies investigated the possible responsibility the drivers’ characteristics in differentiating the results of the real world and the driving simulator tests. The driving experience was found to be the most crucial driver characteristic (Gemou & Bekiaris, 2014) and validation study evaluation of driving errors showed that experienced drivers had the fewest errors compared to the beginner drivers (Mayhew et al., 2011). Gender and age also affect the behavioural validity of the driving simulator. Reed and Green (1999) found that the performance difference in terms of speed variability between young and old participants was significantly larger in simulation experiments than in the real world. Moreover, they found that the behavioural validity of lane keeping measures depends on both age and gender. In fixed-base simulators, participants older than 60 years produced a significantly larger standard deviation of the lateral position, compared to the 20–30 year-old participants, whereas in the real study no gender or age effects were observed. Additionally, they found a significant interaction between age and gender in the simulator, older females performing significantly poorer than other groups. The validation study carried out by Yan et al. (2008) revealed that driver age and gender significantly influence the operating speed in simulator experiments. They found that the mean speed registered with male participants was slightly higher than that with females and that, after the 20–24 age group, a decreasing trend in speed with increasing age was observed. Klüver, Herrigel, Heinrich, Schöner, and Hecht (2016) found no or only marginal age and gender differences between the real study and moving-base simulators. However, they found a significant interaction between age and gender in fixed-base simulators. Specifically, older participants showed a considerably higher headway and lane keeping variability, whereas the performance of younger participants was roughly the same as in the real world.

2. Framework for driving simulator validation Defining the validity of a driving simulator is a complicated and multidisciplinary task. The issue of transferability of results acquired on driving simulator experiments has been a concern for at least 35 years. The quality of a driving simulator is normally defined in terms of physical and behavioural validity (Blaauw, 1982). The first, often referred to as simulator fidelity, measures the degree to which the simulator reproduces the sensory stimuli present in a real driving situation, and depends on the simulator equipment (visual displays, simulator programme and dynamics – motion system). The second, also known as predictive validity, analyses the correspondence between the driver behaviour in the simulator and in the real world. In general, the physical fidelity is a prerequisite for ensuring behavioural fidelity. A recent research project carried out by Lee et al. (2013) revealed that simulators with high physical fidelity demonstrate high behavioural fidelity and likely

V. Branzi et al. / Transportation Research Part F 49 (2017) 1–17

3

provide good estimates of mean speeds in typical engineering applications such as roundabouts and roadway treatments designed to moderate drivers’ speed. Similar results were found by Klüver et al. (2016), revealing the usefulness of all simulators on a relative and partially on an absolute level, with moving-base simulators being preferable to fixed-base ones. The driving performance and behaviour is regarded as the more important validation technique. Generally, the behavioural validity is accomplished by comparing the simulation data with those obtained from on-road tests. Basically, two procedures may be adopted to collect traffic data in real traffic situations: with roadside measurements (laser, radar, video cameras and similar equipments) and instrumented vehicles. Both have their advantages and disadvantages. The first method is especially suited for temporal logging because the equipment is easily transported and mounted. Roadside data collection allows to record a greater number of passages, but the data collection is conducted at fixed points; it is also difficult to control all the effects contributing the possible changes in data. By using instrumented vehicles, instead, driving measurements along longer road sections are possible but the number of probe vehicles is usually limited. Furthermore, the study is conducted on drivers that who are aware of taking part in an experimental activity and thus likely influenced by the situation. Behavioural validity has been described in terms of relative and absolute validity (Blaauw, 1982). The relative validity is obtained when the driving tasks in response to an experimental manipulation have similar effects (e.g., a similar direction of change) and similar magnitude in the simulator and in real driving studies. Instead, the absolute validity is archived when the driver behaviour indicators in the simulated environment attain similar numerical values as those obtainable when driving a real road. Törnros (1998) indicated that, to consider a driving simulator as an useful research tool, relative validity is necessary but absolute validity is not essential. This is especially true when the research aims to compare changes in driving patterns under different treatment conditions (Mullen, Charlton, Devlin, & Bédard, 2011). In these cases the driving behaviour in simulators approximates (relative validity), but does not exactly replicate (absolute validity) on-road driving behaviour (Mullen et al., 2011). Mullen et al. (2011), when examining the behavioural validity, encouraged researchers to consider another approach based on statistical regression. It enables the computation of the effective ‘‘distance” between the real and simulated environments, pointing out systematic errors and system limitations. This procedure has proven to be effective for validating a fixed-based driving simulator (Losa, Frendo, Cofrancesco, & Bartolozzi, 2013). A recent and extensive literature review on the topic of driving simulator validity, specifically focusing on speed related measures (Knapper, Christoph, Hagenzieker, & Brookhuis, 2015), supported the use of simulator, finding that the majority of the studies reported indications for relative validity, whereas only few studies obtained absolute validity. These latter revealed that the driving speed in simulator might be influenced by the type of the geometrical features of the tested road layout. When the road layout does not include features requiring complex driving tasks (less demanding configurations, such as longer straights, high radius bends), the recorded speed during the simulation experiments tends to be higher than in real driving conditions (Bella, 2008; Bittner et al., 2002; Boer, Girshik, Yamamura, & Kuge, 2000; Simsek, Bittner, Levison, & Garness, 2000). The opposite occurs in more demanding road sections such as work zone area (Bella, 2005; Bham et al., 2014) along which the speeds in the simulator tend to be lower than real ones. This evidence can be related to the drivers’ perception of the field of view in the driving simulators, which likely differs from that of the real world, as suggest by Törnros (1998) in a behavioural validation study for road tunnels. In fact, the peripheral flow of information is the most important factor allowing the drivers to perceive the speed they are driving. Other factors may contribute, such as the lack of vestibular and other motion cues in combination with the absence of simulation of longitudinal forces in fixed-base driving simulators. Knapper et al. (2015) highlighted that moving based simulators may provide a slight advantage over fixed base simulators, when dealing with absolute validity of the speed registered during driving simulator studies. 3. Objective and methodology The paper is referred to a preliminary activity of the SCUP (Speed Control in Urban Projects) research project conducted at the Department of Civil and Environmental Engineering (DICeA) of the University of Florence, aiming at assessing the safety impact of road design solutions on urban roads by means of driving simulator experiments. In urban driving environments, speed reduction measures are often the most promising solution, especially on the roads encompassing both the mobility and the access road functions, provided that such measures are also designed in compliance with the driver’s capabilities and expectations (Domenichini, La Torre, Tartaglia, Branzi, & Fanfani, 2014). The purpose of the study was to establish the behavioural validity of the motion-base driving simulator of the Road Safety and Accident Reconstruction Laboratory (LaSIS) of the DICeA, in order to use it to evaluate the effectiveness of urban road safety treatments during the design process. Although physical validity (such as simulation dynamics and movements) could give the motorist a better feeling of driving a real car, the simulator behavioural validity, for speed management treatments based on human factors related criteria, is of paramount importance. In fact, beside road geometric features, perceptual treatments placed along the road layout are supposed to influence the driving speed (Domenichini et al., 2014). Therefore, the driving speed was considered as the surrogate safety indicator to evaluate the possible differences in the users’ driving behaviour among the real and simulated conditions.

4

V. Branzi et al. / Transportation Research Part F 49 (2017) 1–17

To accomplish this objective a campaign of speed measurements was conducted along an urban road, near Florence that afterwards has been virtually reconstructed and tested at LaSIS Driving Simulator. The simulator validation study was conducted assuming that driving was conditioned exclusively by the geometric features and the engineering treatments present along the road layout examined. The comparison between the data measured in field and those registered at the driving simulator was carried out by means of comparative and conventional statistical methods and regression techniques in order to evaluate the relative and absolute validity of the results provided by the simulator. The relative validation should be necessary to compare the safety effectiveness of the possible different design alternatives, while the absolute validation should allow verifying whether the proposal solutions are able to influence the drivers’ speeding behaviours along this type of roads. Finally, assuming that the behavioural validity may be affected also by the driver characteristics, the effect of some of these variables (driving experience, annual driven distance, specific driving experience on the urban street examined and gender) on the simulation speed data has been examined. 4. Field study 4.1. Urban road characteristics The speed monitoring activities were performed along Via Pistoiese, near Florence (Italy), which is two-lanes urban penetration collector linking the suburban areas located in the west boundaries of the Firenze Municipality to the city centre. Along both sides of Via Pistoiese two crowded residential and commercial districts are present, burdening the road with many intersections, pedestrian crossings and driveways. 3 intersections are signalized and n. 7 intersections are unsignalized. A total number of 10 pedestrian crossings are present, all identified by zebra crossings except three which have a raised platform. The road axis is composed of two long straights connected by a bend (R = 250 m) and has a total length of about 4.5 km (Fig. 1). The road cross-section has a total width 18.5 m and includes two 5.50 m wide traffic lanes, two 2.00 m wide lateral parking lanes and two 1.50 m wide sidewalks. The posted speed limit is 50 km/h. The external 2.5 km long section, starting from the roundabout under the A1 motorway to Via del Pesciolino (see Fig. 1) was selected to form the test bed for the driving simulation study. The driving environment along the test bed includes different infrastructural constraints (named ‘‘engineering treatments”) possibly inducing changes in the driven speed: the initial roundabout, the intermediate bend, the raised pedestrian crossings (Fig. 2), the presence of a continuous or discontinuous median (Fig. 3), the presence of 3 or 4 legs intersections allowing left turns and of signalized intersections or pedestrian crossings have been considered as speed influencing engineering treatments. The entire route (2.5 km long section) was subdivided into 11 homogeneous sections (named from A to M, as shows in Fig. 4) whose main characteristics are summarized in Table 1 and in Table 2. Inside each section one or more speed data collection sites were selected, for a total number of 21 sites, located as shown in Fig. 4 and characterized as described in Table 3. 4.2. Data collection and processing The speed data were recorded with a laser speed meter (KV Laser, supplied by Sodi Scientifica) installed on the windscreen of a private vehicle; the vehicle was placed in the parking spaces available along Via Pistoiese (Fig. 5) so as not to interfere with the traffic flow and to avoid biased behaviour of drivers. The laser speed meter is a stand-alone unit which does not require external detectors, thus eliminating the need for disruptive road works normally associated with the installation of other types of traffic monitoring systems. The KV Laser works in a time domain: the device calculates the speed and length of a vehicle from the time taken to break the two beams and how long the beams are disturbed for. These data are stored with time and date information. For each site, the data acquired were: vehicle’s speed and length, time (hour, minute and seconds) and driving direction. The driving direction towards the city of Florence was exclusively examined in this study.

Fig. 1. Overview of the road axis of Via Pistoiese.

V. Branzi et al. / Transportation Research Part F 49 (2017) 1–17

5

Fig. 2. Raised pedestrian crossing.

Fig. 3. Road segment with discontinuous median.

Fig. 4. Position and features of measurement sites on Via Pistoiese.

The speed measurements were carried out in good weather and visibility conditions and on a dry road surface, during October and November 2015. All recordings were performed during daytime off-peak periods for about 60 min (between 10:00 AM and 11:00 AM). The measured raw-data (approximately 8.800 passages collected) were filtered to capture speed data coherent with the specific experimental conditions considered in the driving simulator experiments: free flow traffic conditions, only

6

V. Branzi et al. / Transportation Research Part F 49 (2017) 1–17

Table 1 Characteristics of the homogeneous sections included in the test bed (from section A to section F). Section

Length (m)

Cross section

A

90

Two undivided lanes

B

240

Two undivided lanes

C

185

D

235

E

550

Two lanes divided by a surmountable and continuous median Two lanes divided by a surmountable and continuous median Two undivided lanes

F

150

Two undivided lanes

Engineering treatments

Classification as driving environment

Ending Points (EP)

Intermediate (I)

Upstream: roundabout Downstream: signalized pedestrian crossing Upstream: signalized pedestrian crossing Downstream: signalized intersection Upstream: signalized intersection Downstream: median opening

/

More demanding

n. 1 pedestrian crossing

Less demanding

n.1 intersection with local road marked with a yield sign

More demanding

n. 1 signalized pedestrian crossingn. 1 bend (R = 250 m)

More demanding

/

Less demanding

n.2 intersection with local road marked with a yield sign

Less demanding

Upstream: median opening Downstream: signalized pedestrian crossing Upstream: signalized pedestrian crossing Downstream: signalized intersection Upstream: signalized intersection Downstream: pedestrian crossing

Note. EP: at the ends of the section; I: within the section.

Table 2 Characteristics of the homogeneous sections included in the test bed (from section G to section M). Section

Length (m)

Cross section

G

320

H

80

Two undivided lanes + Left turn lane Two undivided lanes

I

160

L

75

M

165

Two lanes divided by a 1.20 m wide median identified by marking Two lanes divided by a surmountable and discontinuous median Lanes divided by a median surmountable and discontinuous

Engineering treatments

Classification as driving environment

Ending Points (EP)

Intermediate (I)

Upstream: pedestrian crossing Downstream: signalized intersection Upstream: signalized intersection Downstream: raised pedestrian crossing Upstream: raised pedestrian crossing Downstream: raised pedestrian crossing

/

Less demanding

/

More demanding

/

More demanding

Upstream: raised pedestrian crossing Downstream: median opening

/

More demanding

Upstream: median opening Downstream: raised pedestrian crossing

/

More demanding

Note. EP: at the ends of the section; I: within the section.

passenger cars (vehicles with a length between 3 m and 6 m), isolated vehicles (headway time equal to or >5 s) and operative speed >25 km/h (in the absence of traffic disturbances, speed values lower than 25 km/h were considered abnormal). In order to obtain speed data referred to free flow conditions only, two observers were placed 50 m upstream and downstream of each measurement site. The observers took note of traffic situations which created disturbance to the traffic flow (e.g. pedestrians crossing the road, car exiting from a parking space or a driveway, left turning vehicles etc.) in order to be able, during the post processing activity to delete the data concerning the vehicles passed during the disturbed traffic periods. Where the test site was nearby a signalized intersection, speed data was limited to green phases only. More specifically, after the green phase started and vehicles in queue were cleared, only the operating speeds of those which were not in a platoon were selected for data collection. 5. Driving simulator study 5.1. LaSIS driving simulator The LaSIS Driving Simulator is a full scale, dynamic simulator, with a complete Lancia Ypsilon cabin installed on a 6 axes Stewart Stewart’s platform, capable to reproduce all the sensorial stimuli typical of driving. It is the model AS 1200, supplied by AutoSim (Norway), and it is running on the software SimWorld version 2.8.2. The vehicle interior is identical to the commercial version and includes all commands normally available in such kind of cars, with steering wheel with force feed-back. The cabin is surrounded by a cylindrical screen about 200 degrees wide on which 4 projectors (with resolution of 1920
1200 pixels) reproduce the driving environment; rear mirrors are replaced by 6.500 LCD monitors. Sounds and noise are generated by a multichannel audio system, capable to reproduce both vehicle and environmental noise. All functions are

V. Branzi et al. / Transportation Research Part F 49 (2017) 1–17

7

Table 3 Description of speed data measurement sites. Site

A1 A2 B1 C1 C2 D1 D2 D3 E1 E2 F1 F2 G1 G2 G3 H1 I1 L1 L2 M1 M2

Speed influencing engineering treatments Upstream

Downstream

Roundabout (10 m far from) Roundabout (60 m far from) Signalized pedestrian crossing (50 m far from) Signalized intersection (55 m far from) / / / Tangent of the curve (50 m far from) / / / / Pedestrian crossing (35 m far from) Pedestrian crossing (70 m far from) Point in which changes cross-section (15 m far from) Signalized intersection (60 m far from) Raised pedestrian crossing (85 m far from) Raised pedestrian crossing (13 m far from) Raised pedestrian crossing (35 m far from) / /

Signalized pedestrian crossing (70 m far from) Signalized pedestrian crossing (10 m far from) Pedestrian crossing (50 m far from) / / Tangent of the curve (70 m far from) Tangent of the curve (30 m far from) Signalized intersection (75 m far from) / / Pedestrian crossing (50 m far from) Pedestrian crossing (10 m far from) Point in which changes cross-section (70 m far from) Point in which changes cross-section (35 m far from) Intersection marked with a yield sign (30 m far from) Raised pedestrian crossing (40 m far from) Raised pedestrian crossing (75 m far from) / / / /

Note: the distance of the measurement sites from the engineering applications are shown in brackets.

Fig. 5. Laser speed meter installed on a private vehicle.

supervised by a network of 5 computers, including the operator’s station from which the simulation is managed. The data acquisition frequency of the apparatus is 20 Hz. 5.2. Participants Participants were selected after a preliminary screening based on a questionnaire to gather information related to gender, age, education, job position, driving license release date, average annual driving travelled distances, specific driving experience on Via Pistoiese and health conditions. Thirty-nine subjects (16 women and 23 men) were recruited on a voluntary basis among the staff and students of the University of Florence, according to the following criteria: possession of an Italian valid driver’s license, with at least five years of driving experience, an annual driving distance >5000 km and low susceptibility to motion sickness. The selected participants are all resident in Florence and had normal or corrected-to-normal vision. Three drivers exhibited simulator sickness during the experiment implementation and two selected drivers withdrew afterwards from the study. Thus thirty-four subjects (13 women and 21 men) completed the driving simulator experiment.

8

V. Branzi et al. / Transportation Research Part F 49 (2017) 1–17

Age varied between 24 years and 65 years (mean = 43.15 years and standard deviation = 12.85 years). Their driving experience (measured in terms of years of driving license possession) varied between 5 years and 47 years (mean = 24.47 years and standard deviation = 12.96 years). Furthermore, 38% of participants annually covers a distance of between 5000 and 10,000 km, 34% between 10,000 and 20,000 km, while the remaining 28% exceeds 20,000 km. Finally, 41.2 % of subjects travelled through this urban street daily, 26.5% travelled there one a week, and 32.4% rarely travelled through the Via Pistoiese. The driving experience of the subjects, the annual driving distance travelled by them, the specific driving experience on Via Pistoiese and the gender represented the independent variables (factors) for the considered experimental study. For each factor, subjects were divided into three groups as described in Table 4. The characteristic of each group are shown in Table 5.

Table 4 Description of the factors considered. Factor

Levels

Description

Driving experience

Beginner driver Intermediate driver Experienced driver Reduced distances Medium distances Great distances Familiar driver Weekly driver Unfamiliar driver

<15 years of experiences 15–30 years of experiences >30 years of experiences 5000–10,000 km 10,000–20,000 km >20,000 km Travels through this street daily Travels through this street once a week Travels through this street rarely (less than once a month)

Average annual driving travelled distances

Specific driving experience on Via Pistoiese

Table 5 Participant characterization according to the factors considered. Factor 1: driving experience

Beginner driver

Intermediate driver

Experienced driver

Sample consistency Gender (females/males) Age Years of driving license possession Annual driven distance

5000–10,000 km 10,000–20,000 km >20,000 km Usual driver Weekly driver Casual driver

11 4/7 28.0 (3.16) 9.18 (3.06) 36.4% 18.2% 45.4% 23.7% 18.2% 54.5%

10 3/7 42.7 (5.77) 24.0 (5.46) 30% 40% 30% 60% 30% 10%

13 6/7 56.3 (4.73) 37.8 (5.04) 23.1% 53.8% 23.1% 38.4% 30.8% 30.8%

Factor 2: annual driven distance

Reduced distances

Medium distance

Great distance

Sample consistency Gender (females/males) Age Years of driving license possession Specific driving experience on Via Pistoiese

Usual driver Weekly driver Casual driver

10 4/6 42.10 (14.72) 23.60 (14.83) 40% 50% 10%

13 5/8 47.54 (11.21) 28.77 (10.94) 53.8% 7.7% 38.5

11 4/7 38.91 (12.36) 20.18 (12.93) 27.3% 27.3% 45.4%

Factor 3: Specific driving experience on Via Pistoiese

Familiar driver

Weekly driver

Unfamiliar driver

Sample consistency Gender (females/males) Age Years of driving license possession Annual driven distance

14 4/10 44.57 (11.01) 25.79 (10.91) 28.6% 50% 21.4% Males

9 4/5 44.78 (13.53) 26.33 (13.85) 55.6% 11.1% 33.3%

11 5/6 40.00 (14.99) 21.27 (15.13) 9.2% 45.4% 45.4%

Specific driving experience on Via Pistoiese

5000–10,000 km 10,000–20,000 km >20,000 km

Factor 4: Gender Sample consistency Age Years of driving license possession Annual driven distance

Specific driving experience on Via Pistoiese

5000–10,000 km 10,000–20,000 km >20,000 km Usual driver Weekly driver Casual driver

21 42.52 (13.44) 23.71 (13.72) 28.6% 38.1% 33.3% 47.6% 23.8% 28.6%

Females 13 44.15 (12.28) 25.69 (12.05) 30.8% 38.4% 30.8% 30.8% 30.8% 38.4%

Note: frequencies are given for the gender distribution; means and standard deviations in brackets, for the other parameters.

V. Branzi et al. / Transportation Research Part F 49 (2017) 1–17

9

5.3. Virtual reconstruction of via pistoiese Via Pistoiese was replicated in virtual reality reconstructing its cross section and its horizontal and vertical alignment. The existing road markings, signs, vertical billboard, roadside elements, signalized intersections, pedestrian crossings, etc. were incorporated into the virtual scenario in order to obtain driver’s perceptions similar to the real world. Furthermore, to achieve a field of view as much similar to the real world as possible, the background images (including trees, residential and commercial buildings) were composed by photos of the real environment. Fig. 6 shows the virtual environment as seen by the drivers during the experimental trials. The experimental scenario was composed of a 9.5 km long continuous route, composed of two successive sections: a preliminary 7 km long section of rural road, allowing the driver to become familiar with the specific situation, and a subsequent 2.5 km long section of urban road, representing the experimental Via Pistoiese section. The transition from the rural to the urban environment occurred at the existing roundabout located at the A1 Motorway overpass. The virtual scenario was characterized by autonomous traffic, made with 10 different vehicles, organized as ‘‘swarm” around the interactive vehicle. In the first section of rural road the autonomous traffic was organized in both directions, whereas in the examined Via Pistoiese section it was only on the direction opposite to the interactive vehicle in order to avoid influencing the drivers’ speed. The scenario didn’t require any emergency braking response unless the driver made an error. All the traffic signals present along the simulated scenario were turned to green during the driving experiments. 5.4. Experimental procedures Upon their arrival at the laboratory, each participant was briefed on the experiment procedure and was asked to read and sign an informed consent declaration. The participants were advised to drive and behave as in real life situations. The subjects were also warned about the simulator sickness and informed they could stop the test in any moment. They were not briefed about the objectives of the research. The drivers performed a preliminary 10-min training phase in order to familiarize with the interactive vehicle and its control instruments. At the end of the training phase, the subject was asked to get down from the cabin and fill in a post-training questionnaire. After a 5 min rest in order to re-establish psycho-physical conditions similar to those at the beginning of the test, the drivers started the experimental session. The participants were not paid for their involvement, which lasted 30 min. 5.5. Background of statistical analysis A statistical analysis was carried out to verify in absolute terms the behavioural validity of the LaSIS driving simulator. Conventional statistical approaches for group comparisons (such as parametric tests for unpaired samples: independent t-test, ANOVA test, etc.) are typically used to examine the difference of driving performance measures recorded at the driving simulator and on the road. To select the statistical tests to adopt, preliminarily checks of the normality distribution and the

Fig. 6. Driving simulator scenario.

10

V. Branzi et al. / Transportation Research Part F 49 (2017) 1–17

homoscedasticity assumptions of speed data in each measurement site both in the field and in the driving simulator were performed. The Shapiro-Wilk’s test was conducted to verify the goodness of fit of data to the normal distribution. It is appropriate for small sample sizes (<50 samples) as the ones available. The null-hypothesis of Shapiro-Wilk’s test is that the population is normally distributed. Thus, if the p-value was equal or less than to chosen value, then the null hypothesis is rejected and this means that the tested data are not from a normally distributed population. The Fischer’s test used methods to test whether the variances of two samples examined are equal (null hypothesis); the null hypothesis was rejected when Fratio > Fcritical. If a violation of the Shapiro-Wilks’s test and/or the Fisher’s test occurs, a nonparametric equivalent test is more appropriate. The Mann-Whitney U test (M-W test) and the Kolmogorov-Smirnov’s test (K-S test) are common nonparametric alternatives to the independent-samples t-test. The first one is the most common non-parametric tests and was used to verify the equality of two population means (null hypothesis); the second one to determine whether the samples are drawn from populations with the same distribution (null hypothesis). Although the K-S test is less powerful than the M-W test, it is considered more comprehensive because it is also sensitive to differences in the general shapes of the distributions in the two samples (i.e., to differences in dispersion, skewness, etc.). Therefore both tests were used in the validation study. 6. Results and discussion The speed values measured in field and those registered during the simulation experiments have been compared and statistically analyzed to ascertain the behavioural validity of the driving simulation results:  in relative terms (comparative analysis);  in absolute terms (statistical analysis). The impact of the drivers’ characteristics on the behavioural validity of the simulation results was also analyzed. 6.1. Comparative analysis To verify whether the driver behaviour in the driving simulator was similar to that observed in the real world, the speed values obtained from field measurements and simulation experiments at the selected data collection sites were compared. The mean speed, the standard deviation and the variance of the acquired data are summarized in Table 6. The continuous mean speed profile resulting from driving simulation experiments and the mean speed values recorded on site (MSr) or during the driving simulation experiments (MSs) are plotted in Fig. 7. A similar trend of the speed values in the real world and in the virtual reality can be observed: exiting from the roundabout the drivers keep accelerating then decelerate in section D to negotiate the bend; then re-accelerate up to constant

Table 6 Number of measurements, mean speed, standard deviation and variance in real world and in driving simulator. Site

A1 A2 B1 C1 C2 D1 D2 D3 E1 E2 F1 F2 G1 G2 G3 H1 I1 L1 L2 M1 M2

Variance (km2/h2)

Number of measurements

Mean speed (km/h)

Standard Deviation (km/h)

Real

Simulation

Real MSr

Simulation MSs

MSr-MSs (km/h)

Real SDr

Simulation SDs

SDr-SDs (km/h)

Real Vr

Simulation Vs

Vr-Vs (km2/h2)

189 148 153 70 92 136 77 60 94 117 95 112 139 115 81 46 142 79 159 115 39

34 34 34 34 34 34 34 34 34 34 34 34 34 34 34 34 34 34 34 34 34

38.07 43.01 47.95 53.21 54.87 58.24 58.14 52.18 53.33 54.01 54.00 51.03 52.73 49.46 49.14 46.28 46.64 39.47 44.76 48.78 51.18

34.98 43.09 47.68 52.92 54.06 54.99 54.94 52.14 59.53 61.21 61.14 60.34 56.60 55.50 56.31 44.33 46.60 37.78 42.18 47.66 49.46

3.09 0.08 0.27 0.29 0.78 3.25 3.20 0.04 6.20 7.20 7.14 9.31 3.87 6.04 7.17 1.95 0.04 1.69 2.58 1.12 1.72

5.14 6.42 6.38 6.74 8.96 8.98 8.86 9.00 7.38 7.57 8.85 9.00 9.06 9.12 7.56 9.49 6.70 5.80 8.40 9.93 8.36

5.19 4.84 6.09 5.54 5.55 6.01 6.62 5.19 6.76 7.64 9.42 10.00 10.55 9.81 7.77 7.02 6.05 7.62 7.13 7.17 7.65

0.05 1.58 0.29 1.20 3.41 2.97 2.24 3.81 0.62 0.07 0.57 1.00 1.49 0.69 0.21 2.47 0.65 1.82 1.27 2.76 0.71

26.42 41.22 40.70 45.43 80.28 80.64 78.50 81.00 54.46 57.30 78.32 81.00 82.08 83.17 57.15 90.06 44.89 33.64 70.56 98.60 69.89

26.94 23.43 37.09 30.69 30.80 36.12 43.82 26.94 45.70 58.37 88.74 100.00 111.30 96.24 60.37 49.28 36.60 58.06 50.84 51.41 58.52

0.52 17.79 3.62 14.74 49.48 44.52 34.68 54.06 8.77 1.06 10.41 19.00 29.22 13.06 3.22 40.78 8.29 24.42 19.72 47.20 11.37

V. Branzi et al. / Transportation Research Part F 49 (2017) 1–17

11

Fig. 7. Comparison of the mean speed from real world (MSr) and simulator (MSs).

speed along E and F sections. At the beginning of section G the drivers slightly reduce the speed and in the last sections, from G to M, variable motion conditions occur due to the presence of a series of raised pedestrian crossings. This finding allows to conclude that, in qualitative terms, the driving simulator provides speed trends that are comparable to the reality (relative validation). In terms of absolute values, the comparison brings to subdivide the test bed in three successive sections:  In the 1st (from section A to section D) and 3rd (from section H to section M) portions the speed measured values fit well with the simulator’s speed profile, with similar absolute values. The measured values are in general somewhat higher than the ones registered at the driving simulator;  In the 2nd portion (from section E to section G) the speed values resulting from the driving simulator are always higher than the measured ones, with a mean difference ranging from 9.3 km/h (at Site F2) to 3.9 km/h (at Site G1). Considering the homogeneous section classification shown in Table 1 it can be noted that the 1st and 3rd road portions coincide with the more demanding driving environments and that the 2nd portion coincides with the less demanding ones. This confirms the literature findings (Bella, 2005, 2008; Bham et al., 2014; Bittner et al., 2002) suggesting that driving conditions with a reduced number of stimuli (less demanding driving conditions) bring to adopt at the driving simulator higher speeds (perhaps due to a reduced risk perception) and wider speed variances than in reality. This effect disappears and even the opposite slightly occurs when the driver has to negotiate more demanding maneuvers. 6.2. Statistical analysis 6.2.1. Independent-samples test The two data sets (the one containing the results measured in each data collection site in the field and the other containing the simulation results registered in the same sites) have been statistically analyzed to understand if they belong to the same population (absolute validation). To select the statistical tests to adopt, the assumption concerning the normality and the homoscedasticity of both the data sets have been verified by means of the Shapiro-Wilk and the Fisher tests. Both the tests were performed at a significance level of 5%. Table 7 shows the results obtained in each measurement site in both the real world and in the virtual reality. In the Shapiro-Wilk test the data measured and registered in each measurement site were considered as fitting the normal distribution (H0 = null hypothesis) if the resulting p-value was equal to or higher than 0.05. In the Fischer’s test, instead, the null hypothesis (the variance of two data sets in each data collection site are equal) was rejected if Fratio > Fcritical.

12

V. Branzi et al. / Transportation Research Part F 49 (2017) 1–17

Table 7 Shapiro-Wilk’s test and F-test for field data and simulation data. Measurement sites

A1 A2 B1 C1 C2 D1 D2 D3 E1 E2 F1 F2 G1 G2 G3 H1 I1 L1 L2 M1 M2

Field study Simulator study Field study Simulator study Field study Simulator study Field study Simulator study Field study Simulator study Field study Simulator study Field study Simulator study Field study Simulator study Field study Simulator study Field study Simulator study Field study Simulator study Field study Simulator study Field study Simulator study Field study Simulator study Field study Simulator study Field study Simulator study Field study Simulator study Field study Simulator study Field study Simulator study Field study Simulator study Field study Simulator study

Shapiro-Wilk’s test

Fisher’s test

p-value

Test result

Fratio

Fcritical

Test result

0.013 0.230 0.142 0.510 0.093 0.592 0.038 0.077 0.814 0.745 0.199 0.846 0.163 0.179 0.948 0.768 0.003 0.219 0.012 0.117 0.008 0.208 0.037 0.420 0.006 0.913 0.010 0.228 0.010 0.817 0.768 0.068 0.012 0.674 0.028 0.298 0.289 0.051 <0.001 0.214 0.267 0.226

H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0

0.887

1.666

H0 not rejected

0.137

1.668

H0 not rejected

0.796

1.685

H0 not rejected

0.242

1.802

H0 not rejected

0.005

1.750

H0 not rejected

0.015

1.697

H0 not rejected

0.085

1.782

H0 not rejected

0.002

1.839

H0 not rejected

0.613

1.746

H0 not rejected

0.906

1.715

H0 not rejected

0.639

1.745

H0 not rejected

0.441

1.720

H0 not rejected

0.258

1.694

H0 not rejected

0.950

1.717

H0 not rejected

0.823

1.772

H0 not rejected

0.094

1.919

H0 not rejected

0.538

1.692

H0 not rejected

0.063

1.777

H0 not rejected

0.310

1.681

H0 not rejected

0.049

1.713

H0 not rejected

0.633

1.982

H0 not rejected

rejected not rejected not rejected not rejected not rejected not rejected rejected not rejected not rejected not rejected not rejected not rejected not rejected not rejected not rejected not rejected rejected not rejected rejected not rejected rejected not rejected rejected not rejected rejected not rejected rejected not rejected rejected not rejected not rejected not rejected rejected not rejected rejected not rejected not rejected not rejected rejected not rejected not rejected not rejected

According to the Fisher’s test results, the homoscedasticity assumption is satisfied in all the data collection sites at the 5% level of significance, whereas the Shapiro-Wilk test results indicated that the speed data measured in most of the data collection sites cannot be considered as fitting a normal distribution (p < 0.05). Therefore, a non-parametric test for unpaired samples of data was used to check the absolute validity of the simulation results. Both the Mann-Whitney’s test (M-W test) and Kolmogorov-Smirnov’s test (K-S test) were considered. The first one is the most common non-parametric tests and was used to verify the equality of two population means (null hypothesis). The second one was performed to determine whether the samples were drawn from the populations with the same distribution (null hypothesis). Although the K-S test is less powerful than the M-W test, it is considered more comprehensive because it is also sensitive to differences in the general shapes of the distributions in the two samples (i.e., to differences in dispersion, skewness, etc.). The outcomes of the statistical analysis (Table 8) showed that both tests lead to the same results, except for the sites G1 and L2. The absolute validity was achieved in 13 out of 21 measurement sites (from Site A2 to Site D3 and from Site H1 to Site M2). Therefore, the statistical analysis confirms the findings of the comparative analysis: the LaSIS driving simulator yields the same speed values as those recorded in the real situation in correspondence of the measurement sites in which physical constraints are imposed by engineering treatments (raised pedestrian crossing, curve, surmountable and continuous or discontinuous median, etc.). It should be noted that at the first portion of the test bed, the statistical analysis showed that the speed data belong the same population, except in the site A1, which is only 10 m far from the roundabout. An explanation might be that although motion cues are present in LaSIS driving simulator, the manoeuvre in correspondence of the roundabout is much more difficult than in real conditions.

13

V. Branzi et al. / Transportation Research Part F 49 (2017) 1–17 Table 8 Results of statistical tests for comparison of mean speed data. Measurement sites

A1 A2 B1 C1 C2 D1 D2 D3 E1 E2 F1 F2 G1 G2 G3 H1 I1 L1 L2 M1 M2

Mean Speed (km/h)

Mann-Whitney’s test

Kolmogorov-Smirnov’s test

Real

Simulation

p-value

Test result

p-value

Test result

38.07 43.01 47.95 53.21 54.87 58.24 58.14 52.18 53.33 54.01 54.00 51.03 52.73 49.46 49.14 46.28 46.64 39.47 44.76 48.78 51.18

34.98 43.09 47.68 52.92 54.06 54.99 54.94 52.14 59.53 61.21 61.14 60.34 56.60 55.50 56.31 44.33 46.60 37.78 42.18 47.66 49.46

0.004 0.871 0.859 0.799 0.644 0.069 0.077 0.930 <0.001 <0.001 0.001 <0.001 0.049 0.002 <0.001 0.414 1.000 0.133 0.071 0.970 0.389

H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0

0.020 0.563 0.814 0.598 0.347 0.098 0.114 0.336 <0.001 <0.001 0.007 0.001 0.098 0.032 <0.001 0.353 0.971 0.118 0.025 0.882 0.750

H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0 H0

rejected not rejected not rejected not rejected not rejected not rejected not rejected not rejected rejected rejected rejected rejected rejected rejected rejected not rejected not rejected not rejected not rejected not rejected not rejected

rejected not rejected not rejected not rejected not rejected not rejected not rejected not rejected rejected rejected rejected rejected not rejected rejected rejected not rejected not rejected not rejected rejected not rejected not rejected

6.2.2. Regression analysis According to Mullen et al. (2011), establishing whether a simulator exactly replicates on-road performances may be less important than determining how well a simulator predicts the on-road performance. Therefore, the regression technique has been proposed as a valid analysis tool allowing to determine how well the driving simulator predicts the on-road performance (Losa et al., 2013). A regression analysis between the results of the field measurements and the simulation experiments in each data collection site was performed and the findings are plotted in Fig. 8 together with the dotted equality line. A good correlation among the speed values collected with both systems (R2 = 0.719) has been obtained. This means that the mean speed registered during the simulation experiments can be considered as a statistically significative prediction of the mean speed in the real world, F(1, 19) = 48.61, p < 0.001. The error (deviation from the equality line) is less than or equal to 5% inside the simulated speed range of 43–54 km/h in which all the data observed in the data collection sites characterized by the presence of engineering applications are contained (the points represented with green and red colours in Fig. 8). According to Bham et al. (2014), an error less than or

Fig. 8. Linear regression of speed data.

14

V. Branzi et al. / Transportation Research Part F 49 (2017) 1–17

equal to 5% does not invalidate the behavioural analysis of drivers. Therefore an excellent correspondence between the two data sets was achieved in proximity of road sections where engineering applications are present. If reference to an error equal to 10% of the field speed (Bella, 2008) is considered, a wider mean speed range (39–64 km/h) can be considered for accepting the driving simulation tool as a good predictor of the real world speed. Only the speed data registered in the A1 collection site (at the roundabout exit) exceed this range. Therefore, it can be concluded that the speeds registered in the urban scenario tested with LaSIS driving simulator are a reliable prediction of the real world speeds also in absolute terms, mainly where an engineering treatment is present. 6.3. Driver characteristics The aim of this analysis was to test the effect of some driver characteristics on the speed adopted by them in the virtual environment. Four factors (driving experience, gender, annual driven distance and specific experience on Via Pistoiese) were considered and the average driving speed for each factor was evaluated in all the measurement sites (Table 9). The results showed that the mean speed adopted by the drivers with less driving experience was higher than one adopted by others and that increasing speeding behaviours occurred in this driver category in less demanding driving conditions (from Site E1 to Site G3). The greatest difference was recorded between beginner drivers and experienced drivers. Specifically, the differences between mean speeds ranged from 0.48 km/h (at Site A1) to 10.95 km/h (at Site F2). It is worthy to be noted that in each measurement sites (except that in the sites located in the last two homogeneous sections) lower

Table 9 Descriptive statistics for all levels of each factor examined in the virtual environment. Site

A1 A2 B1 C1 C2 D1 D2 D3 E1 E2 F1 F2 G1 G2 G3 H1 I1 L1 L2 M1 M2

Driving Experience

Gender

Annual driven distance

Specific experience on Via Pistoiese

Beginner driver

Intermediate driver

Experienced driver

Males

Female

Reduced distances

Medium distances

Great distances

Familiar driver

Weekly driver

Unfamiliar driver

35.73 (6.28) 43.97 (5.35) 49.09 (4.88) 55.25 (5.34) 56.03 (5.30) 57.91 (5.27) 58.69 (5.57) 53.99 (3.14) 62.65 (5.95) 65.03 (8.04) 65.98 (9.41) 65.98 (8.78) 62.17 (10.15) 61.56 (9.35) 61.95 (5.18) 46.51 (4.99) 45.93 (5.09) 36.39 (7.38) 41.08 (7.24) 47.70 (5.80) 50.08 (5.53)

33.79 (4.63)

35.44 (4.11)

42.83 (5.83)

42.43 (4.19)

48.19 (8.61)

45.73 (4.71)

53.19 (5.06)

50.36 (3.87)

54.72 (5.72)

51.46 (3.38)

55.98 (6.09)

51.06 (4.55)

55.43 (7.21)

50.66 (3.42)

53.19 (7.27)

49.19 (3.28)

59.99 (8.22)

55.87 (3.65)

62.12 (9.41)

56.48 (6.14)

60.97 (11.15)

56.46 (6.17)

59.97 (12.32)

54.99 (6.70)

55.46 (12.19)

52.11 (7.10)

53.31 (11.03)

51.64 (5.80)

54.29 (10.29)

52.62 (4.37)

45.46 (9.67)

41.03 (6.20)

46.86 (7.35)

46.99 (5.57)

37.98 (9.71)

38.97 (4.90)

42.29 (9.05)

43.16 (4.36)

48.15 (8.05)

47.06 (7.33)

52.17 (9.58)

46.11 (6.92)

36.97 (5.01) 43.79 (4.64) 47.97 (6.52) 52.82 (5.85) 54.21 (5.69) 55.92 (6.37) 55.99 (7.29) 53.16 (5.76) 61.01 (6.80) 62.15 (9.15) 62.94 (9.87) 62.23 (10.77) 58.87 (11.05) 57.72 (9.98) 58.25 (7.40) 45.57 (8.16) 46.64 (5.67) 38.98 (8.02) 43.78 (7.23) 49.52 (6.79) 50.62 (7.96)

32.96 (4.10) 42.39 (5.51) 47.36 (5.71) 53.02 (3.46) 53.90 (4.13) 54.04 (4.83) 53.89 (4.24) 51.09 (3.44) 57.99 (5.85) 60.26 (7.05) 59.32 (8.85) 58.38 (8.89) 54.39 (9.12) 53.27 (8.78) 54.32 (8.14) 42.99 (5.71) 46.53 (6.31) 36.57 (5.64) 40.49 (5.65) 45.77 (6.73) 48.23 (7.27)

33.66 (3.03) 42.53 (2.84) 47.21 (5.05) 53.89 (5.08) 55.03 (5.10) 55.99 (6.39) 54.59 (7.34) 52.01 (6.62) 60.88 (7.21) 62.12 (9.37) 61.31 (11.43) 61.49 (12.05) 56.57 (11.99) 57.07 (10.76) 58.02 (9.73) 45.99 (7.98) 46.88 (5.47) 38.01 (5.39) 40.35 (5.43) 45.93 (5.67) 47.98 (7.54)

34.35 (5.45) 41.44 (5.76) 46.54 (7.37) 51.73 (3.72) 53.31 (4.36) 54.97 (4.71) 55.33 (4.45) 52.33 (3.99) 60.69 (6.57) 60.75 (9.84) 61.99 (9.27) 61.64 (9.90) 55.16 (10.18) 54.48 (9.16) 55.48 (7.12) 43.97 (8.09) 46.99 (6.19) 39.04 (8.88) 43.77 (7.91) 49.19 (6.55) 51.25 (7.32)

36.91 (5.54) 45.31 (5.02) 49.25 (5.58) 53.08 (6.34) 53.87 (6.10) 53.95 (6.82) 54.93 (7.72) 52.11 (5.08) 57.07 (5.72) 57.69 (4.87) 57.06 (8.16) 57.93 (9.04) 55.01 (10.21) 54.94 (9.86) 55.49 (7.08) 42.93 (5.57) 45.82 (6.18) 36.31 (6.79) 42.32 (6.57) 47.79 (8.15) 49.18 (8.52)

38.82 (2.01) 44.46 (3.15) 48.38 (6.03) 53.34 (3.78) 54.97 (4.03) 55.12 (5.98) 54.79 (7.15) 52.27 (5.22) 60.51 (6.32) 61.66 (6.97) 61.64 (9.87) 61.08 (11.11) 58.08 (11.13) 55.63 (9.68) 56.78 (6.26) 44.73 (7.77) 46.97 (5.55) 39.14 (7.98) 43.69 (6.22) 47.21 (7.09) 47.88 (7.87)

33.02 (2.05) 42.79 (5.70) 47.49 (6.79) 51.59 (5.87) 51.71 (6.01) 53.09 (6.56) 53.45 (7.03) 51.13 (6.66) 57.49 (7.62) 57.25 (9.91) 57.46 (10.09) 56.59 (10.37) 53.71 (11.06) 53.99 (10.59) 54.85 (9.60) 44.14 (9.03) 46.78 (7.91) 36.51 (6.22) 40.49 (5.69) 46.73 (6.06) 47.24 (8.67)

33.09 (7.08) 41.92 (6.09) 47.15 (6.28) 53.86 (5.85) 55.52 (5.17) 56.74 (4.94) 56.57 (4.51) 52.95 (3.48) 60.55 (6.11) 64.65 (7.93) 64.26 (8.33) 63.33 (8.32) 58.09 (9.41) 56.85 (9.30) 57.27 (8.61) 44.09 (5.62) 46.16 (4.64) 37.64 (7.32) 42.33 (8.34) 49.02 (7.75) 53.21 (5.52)

Note: means and standard deviations in parentheses are given for all levels of each factor.

15

V. Branzi et al. / Transportation Research Part F 49 (2017) 1–17

standard deviations were recorded for drivers belonging to the experienced cluster. It was also found that the mean speed for males was slightly higher than females: the maximum difference was +4.38 km/h (at Site G1). A greater variability of results was recorded for the remaining two factors (annual driver distance and familiarity with Via Pistoiese) thus demonstrating their reduced influence on the drivers’ behaviour during the performed simulation experiments. A one-way ANOVA test was conducted for each factor considered, in order to verify if the speed adopted by motorists in virtual reality can be significantly influenced by some of the driver’s characteristics. Residual analysis was performed preliminarily to test for the assumptions of the statistical test. There were no outliers, as assessed by box-plot; data was normally distributed for each group, as assessed by Shapiro-Wilk’s test (p > 0.05); and there was homogeneity of variances, as assessed by Levene’s test of homogeneity of variances (p > 0.05). Table 10 Results of the one-way ANOVA test for each factor. Site

A1 A2 B1 C1 C2 D1 D2 D3 E1 E2 F1 F2 G1 G2 G3 H1 I1 L1 L2 M1 M2

Driving experience

Annual driving distance

Specific driving experience on Via Pistoiese

Gender

F (2, 31)

p-value

Partial g2

F (2, 31)

p-value

Partial g2

F (2, 31)

p-value

Partial g2

F (2, 31)

p-value

Partial g2

0.417 0.288 0.959 3.038 3.065 4.701 6.789 3.123 5.674 2.770 3.410 4.161 3.226 4.436 6.066 2.186 0.064 0.309 0.243 0.121 2.161

0.663 0.752 0.654 0.062 0.061 0.016 0.004 0.058 0.013 0.078 0.046 0.025 0.053 0.020 0.006 0.129 0.938 0.736 0.786 0.887 0.132

0.026 0.018 0.058 0.164 0.165 0.233 0.305 0.168 0.202 0.152 0.180 0.212 0.172 0.223 0.281 0.124 0.004 0.020 0.015 0.008 0.122

1.359 2.032 0.661 0.592 0.344 0.306 0.020 0.026 1.084 0.780 0.711 0.437 0.059 0.157 0.448 0.947 0.159 0.432 0.691 1.156 0.526

0.272 0.148 0.523 0.560 0.712 0.738 0.981 0.974 0.351 0.467 0.499 0.650 0.943 0.856 0.643 0.399 0.853 0.653 0.509 0.328 0.596

0.081 0.116 0.041 0.037 0.022 0.019 0.001 0.002 0.065 0.048 0.044 0.027 0.004 0.010 0.028 0.058 0.010 0.027 0.043 0.069 0.033

5.892 1.173 0.089 0.551 1.734 0.977 0.696 0.276 0.535 1.909 1.287 1.156 0.533 0.375 0.228 0.155 0.061 0.372 0.597 0.310 2.113

0.007 0.323 0.915 0.582 0.193 0.388 0.506 0.760 0.591 0.165 0.290 0.328 0.592 0.691 0.797 0.857 0.941 0.692 0.557 0.735 0.138

0.275 0.070 0.915 0.034 0.101 0.059 0.043 0.018 0.033 0.110 0.077 0.069 0.033 0.024 0.015 0.010 0.004 0.023 0.037 0.020 0.120

4.907 0.942 0.054 0.128 0.029 0.835 1.046 1.562 1.691 0.403 1.291 1.230 1.438 1.484 2.163 0.895 0.029 0.917 1.957 2.430 0.746

0.034 0.339 0.818 0.723 0.867 0.368 0.314 0.220 0.203 0.530 0.264 0.276 0.239 0.232 0.151 0.351 0.925 0.345 0.171 0.129 0.394

0.133 0.029 0.002 0.004 0.001 0.025 0.032 0.047 0.050 0.012 0.039 0.037 0.043 0.044 0.063 0.027 <0.001 0.028 0.058 0.071 0.023

Note: boldface indicates statistically significant values with 5% level of significance; Italic indicates statistically significant values with 10% level of significance.

Table 11 Results of Turkey post hoc test (p-value). Site

A1 A2 B1 C1 C2 D1 D2 D3 E1 E2 F1 F2 G1 G2 G3 H1 I1 L1 L2 M1 M2

Pair-wise comparison Beginner/Intermediate

Beginner/Experienced

Intermediate/Experienced

0.657 0.865 0.930 0.699 0.866 0.629 0.250 0.826 0.672 0.676 0.414 0.244 0.255 0.136 0.040 0.960 0.968 0.859 0.937 0.988 0.810

0.970 0.740 0.387 0.055 0.064 0.015 0.003 0.058 0.027 0.067 0.036 0.019 0.045 0.017 0.006 0.149 0.935 0.724 0.767 0.942 0.356

0.774 0.981 0.630 0.299 0.202 0.142 0.159 0.216 0195 0.364 0.457 0.523 0.726 0.707 0.831 0.264 0.996 0.977 0.945 0.884 0.129

Note: boldface indicates statistically significant values with 5% level of significance; Italic indicates statistically significant values with 10% level of significance.

16

V. Branzi et al. / Transportation Research Part F 49 (2017) 1–17

The findings of the statistical test (Table 10) revealed that there was a statistically significant difference between groups having different driving experience, while the other three factors did not significantly affect the speed adopted by drivers in virtual reality. The outcomes of the statistical analysis revealed that there are significant differences between the driving experience groups as a whole. In order to investigate about which specific groups are significantly different from the others, a Turkey post hoc test was performed. The findings of this analysis (Table 11) showed that there is a statistically significant difference only between beginner and experienced drivers, but no other group differences were statistically significant. It is important to note that the statistically significant differences were recorded in the sites in which the absolute validity was not obtained. These results are in agreement with previous research findings which suggest that the faster drivers tend to be younger rather than older (McGwin & Brown, 1999) and that young drivers are more likely to underestimate hazards, while experienced drivers are more likely to show anticipatory avoidance of hazard by changing speed, direction, level of vigilance, focus of attention and information transmitted to other road users (Fuller, 2005). This attitude could be increased in virtual environment, in which the perception of risk is less compared to the real world. The differences recorded could be responsible of the differentiation among the real world and virtual environment registered in the less demanding driving conditions (see Section 6.2.1). This behaviour was only recorded in the virtual environment. Due to the method and the equipment used during the field study, the collected indicators refer to average sample values and cannot be characterized in terms of demographic factors (age, gender, driving experiences, etc.). Naturalistic driving tests should conduct to check the findings achieved from simulation. 7. Conclusion A validation study has been performed to assess the behavioural validity of the LaSIS Driving Simulator when used to evaluate the speed related safety measures in urban roads. The relative and the absolute behavioural fidelity of the driving simulator were considered in the study. The comparison of speed data measured in field with those registered at the simulator in the same sites allowed to find out a very good correspondence between the performance in the real world and in the virtual scenario. The drivers increased or decreased the vehicles speed with the same trend measured in the field, thus allowing to confirm the simulator validity in relative terms. The conventional statistical analysis of the two sets of data (those measured on site and those registered at driving simulator) allowed to confirm the validity of the research tool also in absolute terms in all the situations in which the road infrastructure is able to influence the driving behaviour by means of a more demanding driving environment. This means that the speed data registered during the simulation experiment closely predict those adopted in the real world when engineering treatments (bend, bumps, lane narrowing, raised pedestrian crossings, etc.) are present on the road. On the other hand, when a less demanding environment occurs, the speed adopted by the drivers at the simulator were always higher than those measured on site, likely due to the lower perception of risk occurring when driving a simulator rather than in real world. The obtained absolute validity of the simulator allows to consider the speed measured on the simulator as a good predictor of the driver’s behaviour in real world when road engineering measures, such as those considered in all traffic calming solutions, are present. Further research is necessary to understand the underlying mechanics of these effects, to prove the considered hypothesis and find the relationship between the differences of the speeds in virtual environment and on real road. The integrative regression method that has been used for the comparison analysis of the simulator and on-road measurements confirmed the simulation system reliability in terms of absolute validation. In proximity of road sections where the engineering treatments are present, the difference between the mean speeds obtained from the field and from the simulation was less or equal to 5%. The integrative procedure has revealed to be efficient and flexible; it can be applied for other types of driving simulator, different virtual driving scenarios and types of traffic environments, driver groups, including different thresholds, depending on the specific experimental need. These findings demonstrated that the driving simulator can be a powerful support tool in the urban road design process, clearly indicating the influence on speed of the engineering treatments that could be introduced to improve safety. The value of this approach consists in a better understanding of the driver behaviour in relation to road features price to implementation of the latter, thus offering important savings. The drivers characteristics that mainly influence the speed choice in virtual simulation experiments are the gained driving experience and, even if at a reduced extent, the age. According to the obtained results, the familiarity with the driven road, the gender and the annual driven distances play a marginal role. Naturalistic driving tests should be conducted in order to confirm the latter findings. References Abe, G., & Richardson, J. (2006). Alarm timing, trust and driver expectation for forward collision warning systems. Applied Ergonomics, 37(5), 577–586. http://dx.doi.org/10.1016/j.apergo.2005.11.001. ACI-ISTAT (2015). Incidenti stradali in Italia anno 2014. Bella, F. (2005). Validation of a driving simulator for work zone design. Transportation Research Record: Journal of the Transportation Research Record, issue 1937, Washington, D.C., 136–144, http://dx.doi.org/10.3141/1937-19.

V. Branzi et al. / Transportation Research Part F 49 (2017) 1–17

17

Bella, F. (2008). Driving simulator for speed research on two-lane rural roads. Accident Analysis and Prevention, 40(3), 1078–1087. http://dx.doi.org/10.1016/ j.aap.2007.10.015. Bham, G. H., Leu, M. C., Vallati, M., & Mathur, D. R. (2014). Driving simulator validation of driver behavior with limited safe vantage points for data collection in work zones. Journal of Safety Research, 49(6), 53–60. http://dx.doi.org/10.1016/j.jsr.2014.02.012. Bittner Jr., A. C., Simsek, O., Levison, W. H., & Campbell, J. L. (2002). On-road versus simulator data in driver model development driver performance model experience. Transportation Research Record: Journal of the Transportation Research Record, issue 1803, Washington, D.C., pp. 38–44. Blaauw, G. (1982). Driving experience and task demands in simulator and instrumented car: A validation study. Human Factors: The Journal of the Human Factors and Ergonomics Society, 24(4), 473–496. http://dx.doi.org/10.1177/001872088202400408. Blana, E., & Golias, J. (2002). Difference between vehicle lateral displacement on the road and in a fixed-simulator. Human Factors, 44, 303–313. Boer, E. R., Girshik, A. R., Yamamura, T., Kuge, N. (2000). Experiencing the same road twice: A driver comparison between simulation and reality. In Proceedings of Driving Simulation Conference DSC 2000, Paris. Boer, E. R., Ward, N. J., Manser, M. P., Yamamura, T., & Kuge, N. (2005). Driver performance assessment with a car following model. In 3rd international driving symposium on human factors in driver assessment, training and vehicle design. Brown, T., Dow, B., Marshall, D., & Allen, S. (2007). Validation of stopping and turning behavior for novice drivers in the National Advanced Driving Simulator. Iowa City, Iowa, USA. Cao, C., Wang, J., & Luo, Y. (2015). Driving simulator validation for research on driving behavior at entrance of urban underground road. In Proceedings of the 3rd international conference on transportation information and safety, Wuhan, China (pp. 147–150), http://dx.doi.org/10.1109/ICTIS.2015.7232083. De Winter, J., De Groot, S., Mulder, M., Wieringa, P., Dankelman, J., & Mulder, J. (2009). Relationships between driving simulator performance and driving test results. Ergonomics, 52(2), 137–153. http://dx.doi.org/10.1080/00140130802277521. Domenichini, L., La Torre, F., Tartaglia, V., Branzi, V., & Fanfani, F. (2014). Safety improvements in Urban Areas based on Human Factors principles. A case study: Via Pistoiese. In BASt - Ageing and Safe Mobility Conference 27–28 November 2014, Bergisch - Gladbach, Germany. Engen, T., 2008. Use and validation of driving simulators. Norwegian University of Science and Technology, PhD Thesis 2008:2010, Trondheim. Fuller, R. (2005). Towards a general theory of driver behaviour. Accident Analysis & Prevention, 37(3), 461–472. http://dx.doi.org/10.1016/j.aap.2004.11.003. Gemou, M. & Bekiaris, E. (2014). Simulator vs. real traffic conditions: how different parameters influence the transferability of results. In Proceedings of the 4th Transport Research Arena Conference, Paris, France. Godley, S. T., Triggs, T. J., & Fildes, B. N. (2002). Driving simulator validation for speed research. Accident Analysis and Prevention, 34(5), 589–600. http://dx. doi.org/10.1016/S0001-4575(01)00056-2. Hekamies-Blomqvist, L., Östlund, J., Henriksson, P., & Heikkinen, S. (2001). Elderly car drivers in a simulator – A validation study. VTI Rapport, Issue 464, 2000, Swedish National Road and Transport Research Istitute (VTI). Hoffman, J. D., Brown, T. L., Lee, J. D., & McGehee, D. V. (2002). Comparisons of braking in a high fidelity simulator to braking on a test track. Transportation Research Record, 1803, 59–65. Kaptein, N., Theeuwes, J., & Van Der Horst, R. (1996). Driving simulator validity: Some considerations. Transportation Research Record, 1550(1), 30–36. http:// dx.doi.org/10.3141/1550-05. Klüver, M., Herrigel, C., Heinrich, C., Schöner, H., & Hecht, H. (2016). The behavioral validity of dual-task driving performance in fixed and moving base driving simulators. Transportation Research Part F: Traffic Psychology and Behaviour, 37, 78–96. http://dx.doi.org/10.1016/j.trf.2015.12.005. Knapper, A., Christoph, M., Hagenzieker, M., & Brookhuis, K. (2015). Comparing a driving simulator to the real road regarding distracted driving speed. European Journal of Transport and Infrastructure Research, 15(2), 205–225. Lee, H. C., Cameron, D., & Lee, A. H. (2003). Assessing the driving performance of older adult drivers: On-road versus simulated driving. Accident Analysis and Prevention, 35(5), 797–803. http://dx.doi.org/10.1016/s0001-4575(02)00083-0. DOI: 10.1016/S0001-4575(02)00083-0. Lee, J. D., Ward, N., Boer, E., Brown, T. L., Balk, S. A., & Ahmad, O. (2013). Exploratory Advanced Research: Making driving simulators more useful for behavioural research – Simulator characteristics comparison and model-based transformation. Report N. FHWA-HRT-13-073 U.S. Department of Transportation, Washington, DC. October, 2013. Lee, H. C., Falkmer, T., Rosenwax, L., Cordell, R., Granger, A., Vieira, B., & Lee, A. (2007). Validity of driving simulator in assessing drivers with Parkinson’s disease. Advances in Transportation Studies and International Journal, 81–90. Special Issue. Lee, H. C., Lee, A. H., & Cameron, D. (2003). Validation of a driving simulator by measuring the visual attention skill of older adult drivers. American Journal of Occupational Therapy, 59, 97–100. Lew, H. L., Poole, J. H., Lee, E. H., Jaffe, D. L., Huang, H. C., & Brodd, E. (2005). Predictive validity of driving simulator assessments following traumatic brain injury: A preliminary study. Brain Injury, 19(3), 177–188. http://dx.doi.org/10.1080/02699050400017171. Losa, M., Frendo, F., Cofrancesco, A., & Bartolozzi, R. (2013). Procedure for validating fixed-base driving simulators. Transport, 28(2), 420–430. http://dx.doi. org/10.3846/16484142.2013.867281. 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(4), 267–275. http://dx.doi.org/10.1016/j.jsr.2011.06.004. McAvoy, D. S., Schattler, K. L. & Datta, T. K. (2007). Driving simulator validation for nighttime construction work zone devices. Paper presented at 86th Annual Meeting of the Transportation Research Board, Washington, DC. McGwin, G., & Brown, D. B. (1999). Characteristics of traffic crashes among young, middle-aged, and older drivers. Accident Analysis and Prevention, 31(3), 181–198. Meuleners, L., & Fraser, M. (2015). A validation study of driving errors using a driving simulator. Transportation Research Part F: Traffic Psychology and Behaviour, 29(2), 14–21. http://dx.doi.org/10.1016/j.trf.2014.11.009. Mullen, N., Charlton, J., Devlin, A., & Bédard, M. (2011). Simulator validity: Behavior observed on the simulator and on the road. In Handbook of driving simulation for engineering, medicine, and psychology (pp. 13.1–13.18). Reed, M. P., & Green, P. A. (1999). Comparison of driving performance on-road and in a low-cost simulator using a concurrent telephone dialling task. Ergonomics, 42(8), 1015–1037. http://dx.doi.org/10.1080/001401399185117. Riemersma, J. B. J., van der Horst, A. R. A., Hoekstra, W., Alink, G. M. M., & Otten, N. (1990). The validity of a driving simulator in evaluating speed-reducing measures. Traffic Engineering and Control, 31, 416–420. Santos, J., Merat, N., Mouta, S., Brookhuis, K., & De Waard, D. (2005). The interaction between driving and in-vehicle information systems: Comparison of results from laboratory, simulator and real-world studies. Transportation Research Part F: Traffic Psychology and Behaviour, 8(2), 135–146. http://dx.doi. org/10.1016/j.trf.2005.04.001. 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. http://dx.doi.org/10.1080/15389580902894989. Simsek, O., Bittner, A. C., Jr., Levison, W. H., & Garness, S. (2000). Development of prototype driver model for highway design. Task B: Speed-decision simulator experiment. Seattle: Battelle Human Factors Transportation Center. Törnros, J. (1998). Driving behaviour in a real and a simulated road tunnel - A validation study. Accident Analysis & Prevention, 30(4), 497–503. Traffic Safety Basic Facts 2015: Urban Areas (2015). Available from Internet: . Wade, M. G. & Hammond, C. (1998). Simulator validation: evaluating driver performance in simulation and the real world. Report N°28, Minneapolis, MN: Minnesota Department of Transportation. Yan, X., Abdel-Aty, M., Radwan, E., Wang, X., & Chilakapati, P. (2008). Validating a driving simulator using surrogate safety measures. Accident Analysis and Prevention, 40(1), 274–288. http://dx.doi.org/10.1016/j.aap.2007.06.007. DOI: 10.1016/j.aap.2007.06.007.