Simulator-based driving test prescreening as a complement to driver testing – Toward safer and more risk-aware drivers

Young people represent a high-risk group of drivers and the prevalence of road traffic crashes among young drivers is high. Thus, to increase traffic safety, it is essential to ensure that new drivers are both sufficiently educated in and assessed for risk awareness. The aim of this study was to examine the possibility and potential benefit of using a driving simulator screening test as a complement to the existing on-road driving test. The main idea is to detect drivers who are not ready to proceed to the driving test. A comparative study was performed with participants who passed and failed a simulator test and an on-road driving test, respectively. A comparison between subjective and objective measures of performance and risk was also included. A driving simulator was placed at a traffic school and customers were recruited as participants. In total, 70 participants took part in the study and the simulated drive consisted of rural roads, urban traffic, and motorways with 16 different scenarios, constructed from the second level of the GDE matrix, to examine driving behavior, attention, and risk perception. The results show that with a screening test in a driving simulator, it is possible to detect drivers who consider themselves ready to take a driving test, but who have not yet reached the level of risk awareness required to be a safe driver. Test scenarios should be suited to detect deficiencies in risk awareness, test different levels of the GDE matrix and, to complement the driving test, be difficult to assess in an on-road driving test. Deficiencies in self-evaluation that are well-known among young drivers are again confirmed. To practice self-evaluation, the driving simulator is suggested as a pedagogical tool, linked to the GDE matrix.


Introduction
In Sweden, young drivers (18-24 years) account for more than 30 % of single-vehicle crashes resulting in death or injury (The Swedish Transport Agency, 2017).Similar figures exist for Europe, where 16 % of drivers who died in traffic in the years 2015-2019 were younger than 25 (European Commission, 2021).The high prevalence of road traffic crashes among young drivers is not specific to Sweden or Europe (Foss et al., 2011;Siskind et al., 2019;Walshe et al., 2017).Neither is the overrepresentation of young drivers' withdrawn driving licenses country-specific, and the reasons are commonly offenses like drunk driving and speeding (Faulks, 2022;The Swedish Transport Agency, 2022).
It is well-recognized that young people represent a high-risk group of drivers (Banz et al., 2019;Elvik, 2010;Walshe et al., 2017), which can partly be explained by a lack of knowledge and risk awareness (Fisher et al., 2006;McKnight & McKnight, 2003).There are also additional underlying reasons, including developmental, social, and cultural factors (Arnett, 2000;Bates et al., 2019;Walshe et al., 2017).Contributing factors associated with crashes involving young drivers, such as risky driving behaviors, inadequate hazard perception, and distracted driving have been suggested to indicate poor executive function (Walshe et al., 2017).Executive function is not fully developed before early adulthood with the maturation of the frontal lobe (Diamond, 2013) and is related to impulsivity, sensation seeking, and other risk-taking behaviors in young adults (Romer et al., 2017).Thus, to increase traffic safety, it is essential to ensure that new drivers are both sufficiently educated in and assessed for risk awareness.
The Swedish Transport Administration and the Swedish Transport Agency, the authorities involved in driver licensing in Sweden, have reported a decline in the pass rate of the driving tests.Difficulties in assessing risk perception and self-awareness during the on-road driving test have been highlighted (The Swedish Transport Administration & The Swedish Transport Agency, 2019).Furthermore, the possibility of using driving simulators for certain elements of driver training and the driving test has been discussed, since driving simulators are wellaccepted tools for training in different types of transport and for various professional groups of drivers (Bates et al., 2018;Casutt et al., 2014;Olsson et al., 2021;Pollatsek et al., 2011).
Prior research has explored the advantages of using driving simulators for training, including the opportunity to train in critical situations ( Ābele et al., 2019;de Winter et al., 2012;Olsson et al., 2021), gain better self-awareness ( Ābele et al., 2018;Vlakveld et al., 2011), and learn from mistakes ( Ābele et al., 2019;Allen et al., 2007;de Winter et al., 2012), as well as time and cost benefits ( Ābele et al., 2019;Olsson et al., 2021;Pollatsek et al., 2011).Simulators have also been suggested suitable for training and testing not only hazard perception but also decision-making skills (Jackson et al., 2009).Undoubtedly, driving simulators have the potential to improve traffic safety in various ways.The focus of this paper is the potential of driving simulators to complement the on-road driving test, specifically regarding decision making in rare risk critical scenarios.

Aim and research questions
The aim of this study was to examine the possibility and potential benefit of using a driving simulator screening test as a complement to the existing on-road driving test.The main idea was to detect drivers who are not ready to proceed to the driving test, which, due to the natural limits of complete assessment, they might pass.Drivers that either show risky driving behavior or insufficient self-awareness could be flagged as not yet ready to proceed to the on-road driving test.The following research questions have been formulated to address the aim of the study.
• How can risk-prone drivers who are not yet ready for a driving test be detected with a simulator screening test?o What kind of traffic scenarios should be included in the test?o How can self-awareness be assessed in the same test?

Background
In this section, the current routine for driving license training in Sweden is presented, followed by a discussion of cognitive aspects involved in safe driving and a brief overview of the use of driving simulators in Europe today.

Drivers' education in Sweden
In Sweden, driver's license training can begin at age 16 and be completed through private supervision by, for example, parents or a driving instructor (Gregersen et al., 2000).Either way, the student needs to apply for a learner's permit from the Swedish Transport Agency.For private practice driving, an application to become a supervisor is also required.However, attending risk training is mandatory.The first part of the risk training consists of information about risky driving behaviors involving factors such as alcohol, drugs, and fatigue.The second part covers speed, safety, and driving in different road and weather conditions, as well as practical strategies for driving on slippery roads.
The curriculum is based on the Goals for Driver Education (GDE) matrix (Hatakka et al., 2002), which emphasizes motivational aspects of driver education and replaces reevaluated pedagogical methods with, for example, active learning methods and the use of self-reflection to achieve the goals.The GDE matrix defines the competencies that a driver must have to be a safe driver, and it shows how driving is a process that involves an interaction between the driver's own abilities and motives.The GDE matrix consists of four hierarchical levels (vehicle maneuvering, mastering traffic situations, goals and context of driving, and goals for life and skills for living) with three categories on each level (knowledge and skills, risk increasing factors, and self-evaluation).The first three levels of the GDE matrix encompass and expands on the operational, tactical, and strategic levels in the driver behavior model proposed by Michon (1986).The curriculum specifies goals for selfevaluation.However, it has been suggested that there is room for improvement (Mynttinen et al., 2009), and the authorities are aiming to incorporate self-assessment as a natural element throughout the training (The Swedish Transport Administration and The Swedish Transport Agency, 2019).

Cognitive aspects of safe driving
During the last decades, the terminology has slowly been shifting from traffic accidents to traffic crashes.The argument is that an accident is an act of fate and is therefore unpredictable, while that is not true in the case of traffic crashes (Evans, 2015;Rumar, 1988;Sabey, 1991).Most traffic crashes have been attributed to human causes, such as recognition and decision-making errors, as the "immediate reason for the critical event" (Singh, 2015).This implies the need for considering risk awareness and the ability for self-reflection as part of drivers' education, in addition to the traditional focus on procedural skills and vehicle control, such as prescribed by the GDE matrix (Hatakka et al., 2002;Näätänen & Summala, 1974;Reason et al., 1990).
Driving is a complex everyday task that involves a range of cognitive processes, including perception, attention, learning, memory, decisionmaking, and action control (Groeger, 2000).Driving simulators have the potential to engage drivers in the types of cognitive activities involved in real-world driving, that is, to attain cognitive fidelity (Tsang and Vidulich, 2003).Among the cognitive activities that are important to safe driving are speed management, attention to the forward roadway, and hazard perception, i.e., the ability of drivers to anticipate potential hazards as they unfold over time (Chan et al., 2010;Groeger, 2000).
Hazard perception, a form of situational awareness (SA), has shown to predict crash risk (Horswill, 2016).Endsley describes a model of SA in three levels: perception (1); comprehension (2); and projection (3).A driver needs to be able to perceive the hazard, understand that it is a hazard and project other road users' future behavior and their role in that situation.Still, it is possible to make inaccurate decisions with accurate SA, making SA and hazard perception a necessary but not sufficient skill for safe driving (Endsley and Garland, 2000).Hazard perception can also be defined as having both tactical and strategical aspects as to perceive and react to explicit hazards or anticipate future hazards (Michon, 1986;Krishnan et al., 2019).
These cognitive abilities are self-paced, i.e., the driver can make it easier or harder for themselves by for example decrease their speed and thus gain time to act or by planning their vehicle control in a way that it doesn't distract from their visual search in a potential hazardous situation (Näätänen & Summala, 1974;Fuller, 2005).Simulator studies have shown that novice drivers are less likely to appropriately adapt to speed limit traffic signs (Fisher et al., 2006) and to scan for and anticipate hidden or latent hazards (Borowsky et al., 2010;Pradhan et al., 2005).Importantly, these results have been found to be highly consistent with results from on-road studies (Chan et al., 2010;Shinar, 2008).In addition, there is substantial evidence that hazard perception can be improved using simulator-based training (Horswill et al., 2017;Lu et al., 2022).Taken together, these findings suggest that driving simulators could potentially be used to preventively detect or screen for safetycritical cognitive deficiencies in individuals undergoing driver training.

Simulators for driver education
In several European countries, driving simulators are used as a complement to driver training in real traffic.The motives vary, however, and are related to country-specific challenges.Finland started using driving simulators primarily for night driving, since real-life night training is difficult in certain parts of the year (Mikkonen, 2007).Practicing night driving as well as long distances to reach an urban environment are motives expressed in Norway (Robertsen et al., 2016;Saetren et al., 2018), where, as in Sweden and Denmark, simulators are not yet widespread ( Ābele et al., 2019;Selander & Thorslund, 2021).More densely populated countries, like France and Germany, have a lack of driving instructors or financial motives for replacing some of the driver training with driving simulators (Fretay, 2022;Reindl et al., 2016).In sum, the two main motives for using driving simulators in European driver education are improving efficiency and completeness.To the authors' knowledge, no European countries use driving simulators for passenger car driving exams assessment.

Method
A comparative study was performed with participants who passed and failed a simulator test and an on-road driving test, respectively.A comparison between subjective and objective measures of performance and risk was also included.

Participants
A driving simulator was placed at a traffic school and customers were recruited as participants.Some tests were canceled due to technical problems (n = 7) or simulator sickness (n = 1).In total, 70 participants completed the study, 46 women and 24 men.The mean age was 20.7 years (SD = 4.3) across all participants, 20.8 years (SD = 1.66) for the women, and 20.5 years (SD = 4.1) for the men.
The 30-minute simulated drive consisted of rural roads, urban traffic, and motorways with a series of scenarios to examine driving behavior, attention, and risk perception.The drive was divided into three parts.After each part, the participant was asked to answer questions displayed on a side screen next to the simulator.The questions covered their selfassessed driving performance and perceived risk in each scenario.

Screening test
A reference group with experts from the driving test unit at the Swedish Transport Administration, the Swedish Transport Agency, Sweden's national association of traffic schools, and additional traffic educators has been actively involved in the design of the simulator screening test.The primary goal of the design was to include scenarios that are rare and thus difficult to test in an on-road driving test, but very important to test to ensure that the driver is sufficiently aware of their risks.All scenarios were carefully chosen in close cooperation with the reference group and with the GDE matrix as a guiding framework.With the aim of complementing the on-road driving test, the scenarios were chosen primarily from the second level of the GDE matrix.
In total, the screening test included 16 different scenarios in urban traffic, on rural roads, and on motorways (see Table 1).Among other things, the scenarios featured slow-moving vehicles, children playing, vulnerable road users at pedestrian crossings, road work, turning on country roads, and merging on motorways.The speed limit varied between 30 and 110 km/h (about 20-70 mph).Each scenario was designed with the aim of being critical enough for a failure to justify a failed screening test.Hence, passing the test required passing all scenarios.The criteria for passing or failing were set by the reference group, focusing mainly on tactical hazard perception and decision making.

Table 1
Pictures, descriptions, and fail criteria for the screening test scenarios (S1-S16).

Self-assessment
The participants were prompted to stop the simulator on three occasions to self-assess their performance and the perceived risk.On each occasion, images of 5 to 6 of the 16 scenarios were displayed separately.After answering the questions for each scenario respectively, the participants clicked forward to the next scenario.This is in line with the recommendations to administer self-evaluation measures in close connection to the event that is subject to evaluation (Sundström, 2009).Driving performance was captured by the question: "How would you rate your driving performance in the situation illustrated below?"Participants chose one of the alternatives: 1 (excellent), 2 (good), 3 (moderate), 4 (not so good), 5 (poor).Perceived risk was assessed by the question: "Did you experience any risk in the situation illustrated below?"Participants chose one of the alternatives: 1 (no risk at all), 2 (fairly small risk), 3 (moderate risk), 4 (quite a risk), 5 (very high risk).

Equipment
A fixed-base driving simulator without a motion platform was used for the screening test.This was equipped with gas and brake pedals (automatic transmission), a driver's seat, a steering wheel, and an instrument panel.Three screens provided a broad overview of the traffic environment.Three rear-view mirrors were also integrated into the screens.A small screen to the right next to the steering wheel was used to answer questions about risk awareness and performance.See Fig. 1.

Procedure
The screening test was carried out a few days before the on-road driving test.Initially, participants received instructions about the study's purpose, both orally and in writing, and signed the informed consent.
Staff at the traffic school started the simulator, assigned a participant ID, showed the controls and pedals, and checked if the driver's seat needed to be adjusted.A 5-minute training session was included to allow for acclimation to both the simulator and the simulated environment and thereby reduce the risk of simulator sickness.After this, the participant was left alone, and the screening test began.
Neither the student, the driving school nor the driving test examiner received information about the results of the simulator screening test.This was saved only for the researchers' analysis.After completion of the on-road driving test, the driving school received information about the result (pass or fail).For each participant, this information together with age and gender was saved by the driving school in a file and linked to the driving test number assigned in the simulator test.This file was sent to the research group.

Statistical analyses
After classifying all scenarios as pass or fail, a file was compiled for data analysis.The file contained gender and age, outcomes from the simulator test scenarios according to the criteria in Table 1, selfassessment for each event in the simulator test, and results from the on-road driving test.SPSS statistics were used for all statistical analyses, including ANOVAs, frequency analysis, and correlations.

Results
Four groups were created according to the possible outcomes after the two tests (see Table 2).ANOVA tests revealed no effect of age or gender on group distribution, or on the outcome of the screening test or the on-road driving test.

Detection of risk-prone drivers
Forty-four participants (63 %) failed the screening test, due to the criteria of failing at least one scenario.Among these participants, a majority (66 %) passed the on-road driving test and were categorized in group B. In most cases (n = 32, 73 %), participants failed one scenario.10 participants failed 2 or 3 scenarios, and 2 participants failed 4 or 5 scenarios.Among participants who failed 1-2 scenarios, less than 60 % passed the on-road driving test, while both participants who failed 4 or 5 scenarios passed the on-road test.See Table 3 for the distribution of participants according to the number of failed scenarios and the proportion who passed the on-road test.

Screening test scenarios
The fail rates for each scenario in the screening test are displayed in Fig. 2. The scenario in which participants most frequently failed was S9 (truck approaches from behind on a rural road).60 % of the participants did not stop on the right to let the truck pass before taking a turn to the left.On closer inspection, this scenario was deemed more difficult than intended, with the driver having too little foresight to be able to act correctly.S9 was therefore excluded from the analysis, including the distributions presented in Tables 2 and 3.
ANOVA revealed a significant effect of gender in the motorway entrance scenario (S12), such that 30 % of the women and 4 % of the men failed, F(1, 68) = 6.92, p =.011.The most common reason for failing this scenario was approaching at a speed that was too low (11 cases) and the second most common was too short a distance to the vehicle in front (8 cases).
In one scenario, ANOVA showed a significant difference between the two groups that failed the screening test (B and D).With the cyclist crossing the road in S3, there were 3 participants in group D and none in group B who failed to yield the right of way, F(1, 42) = 6.92, p =.012.

Self-awareness assessment
Mean values of self-rated performance show that drivers consider their performance to be moderate or better (from 1 = excellent to 5 = poor) in all scenarios.The scenario with the lowest-rated performance is the one with the girl appearing behind the bus (S2).In this scenario 39 % of the participants failed; however, the mean self-rated performance is moderate (M = 3.01, SD = 1.25).The ambulance overtaking scenario has the second-worst performance ratings (S10, M = 2.84, SD = 1.07), and in this scenario, no participant failed.
Correlation analysis for each of the scenarios showed statistically significant relationships (p <.05) between outcomes and self-rated performance for five scenarios (S2, S3, S5, S12, S13), such that failure is associated with lower-rated performance.
The scenario with the motorway entrance was the only one with a statistically significant correlation between perceived risk and objective performance, such that participants who failed the scenario rated the risk higher.In Table 4 the scenarios with any significant correlations discovered between self-rated performance, perceived risk, and outcome are presented.

Discussion
The aim of this study was to examine the possibility and potential benefits of using a driving simulator-based screening test as a complement to the existing on-road driving test in Sweden.The motive is that although the curriculum of the driver education is based on the GDE matrix and the intention is to assess driving skills and risk awareness at the on-road driving test, it is hard to control the content of the on-road driving test.The purpose is not to replace the on-road driving test, but to complement it with a screening test to enable a more comprehensive assessment.While a major advantage of the screening test is the possibility of controlling scenarios, the main advantage of the on-road driving test is the realism.The main research question concerns how to detect risk-prone drivers with a simulator screening test, with sub-questions of what kind of traffic scenarios to include and how to assess selfawareness.In this section, the results are discussed for each research question respectively, followed by suggestions for future work.

Detection of risk-prone drivers
The fact that 43 participants failed the screening test indicates that a simulated screening test can detect risky drivers, since the scenarios are  designed to be critical and have no tolerance for risky behavior.Of the 41 participants who passed the on-road driving test, 71 % had failed the screening test and were, according to the stipulations set by this project, too risky to have a driver's license.These drivers, categorized in group B, are of specific interest, since in the screening test they drove in a way that would have posed a large risk both to themselves and to their fellow road users if they drove that way in the real world, but they nevertheless passed the on-road driving test and received their driving licenses.Essentially, they showed major deficiencies in the simulator in awareness of vulnerable road users, when they must give way, and how to safely merge on the motorway.These skills are possible test points in the onroad driving test, but it can be difficult to create similar testing opportunities in real-world driving.By complementing the driving test with a simulated screening test, the chances of detecting risk-prone drivers increase.
Of the 26 participants who passed the screening test, 14 (54 %) failed the on-road driving test and were thus categorized in group C. Like with group B above, this result demonstrates that the two tests measure different things.The screening test guarantees exposure of situations demanding hazard perception and decision making in a way that the onroad driving test cannot.The on-road driving test on the other hand assesses level one of the GDE matrix, which is not included in the screening.One possible reason for failing the driving test after having passed the screening test may also be insufficient scanning, which has been stated as important examination criteria (Lidestam et al., 2010) and important for hazard perception ( Ābele et al., 2018;Lee et al., 2008;Underwood, 2007).Real-world situations, compared to a simulator environment, facilitate assessment of both scanning behavior and the current driving performance.Furthermore, participants in group C may have been more worried while taking the on-road driving test compared to the screening test, and therefore performed worse (Fairclough, Tattersall, & Houston, 2006).

Screening test scenarios
The purpose of the screening test was to include scenarios that are safety-critical and appropriate for detecting deficiencies in risk awareness, but difficult to assess in an on-road driving test.Each scenario should have clear pass or fail criteria, as the idea is that any future mandatory screening test should be able to be carried out in a simulator, be fair, and not require any evaluation or interpretation of the results.Some scenarios did not turn out as safety-critical as planned.However, with some adjustments in implementation, the scenarios can be included in a screening test.The same applies to the scenario that was more difficult than intended.All scenarios in which participants failed reveal deficiencies in risk awareness and inadequate hazard perception, making it appropriate to include variants of these in a screening test supplementary to the on-road driving test.
With driving being a self-paced task mediated by executive function (Walshe et al., 2017), the scenarios test participants' ability to detect hazards in the simulated traffic environment.This engages the participant in higher-order cognitive processes, e.g.decision-making and planning (cf.Section 2.2), that are affected by adjustments to their own driving behavior in response to potential hazards.This differs from traditional hazard perception tests (e.g.Borowsky et al., 2010;Lee et al., 2008), in which participants are limited to only detecting developing hazards instead of potential hazards and risky situations.
Importantly, different types of scenarios involve different types of cognitive challenges, mainly tactical.Consider the scenario where a child runs out on the roadway from the right-hand side of the vehicle (S15).Here, the participant is driving straight and is likely to experience low cognitive load from vehicle control.The child enters the participant's field of view in the periphery.This scenario engages the participant in a somewhat different (but overlapping) set of cognitive processes than the scenario where a pedestrian is walking out from the left-hand side of the road (S3).Here, the pedestrian appears relatively far from the middle of the participant's field of view, and detecting the pedestrian requires a visual search by shifting their gaze, which is a behavior governed by executive function (Walshe et al., 2017).Similarly, detecting the child running out behind a bus (S2) requires the participant to anticipate the hazard and respond to it proactively, which also depends on executive function.
These cognitive challenges are described in the different levels of the GDE matrix (Hatakka et al., 2002).For example, passing S15 is facilitated by the driver having automated vehicle control so they can direct gaze toward the roadway (Vehicle control).By contrast, S3 and S2 relate more directly to the second and, somewhat, to the third level (Driving in traffic situations and Goals and context of driving), respectively.Introducing distractors related to vehicle control, e.g., turning, shifting, and accelerating, will increase the workload and task difficulty unless the participant strategically plans around this, which is a part of the third level (Goals and context of driving).

Self-awareness assessment
Self-rated performance show that drivers considered themselves to have performed moderately or better than moderately in all scenarios.The scenario with the lowest rated performance is the one in which a girl runs out from behind a bus, in which 39 % of the participants failed.However, the average of the self-ratings indicates that the drivers considered themselves to have performed moderately.In the scenarios that received the second-worst self-ratings, no one or only a few failed.The small variation of self-rated performance between scenarios may indicate an insufficiently fine-grained scale, but the results are in line with previous research showing deficiencies in self-evaluation ability (McKnight & McKnight, 2003).This deficiency may be partly due to a not fully developed frontal lobe (Diamond, 2013) and possibly also affected by the drivers not being used to performing self-assessment.In the GDE framework self-evaluation is suggested to be a critical ability for a driver to be able to adapt their driving to be within their own level of skill and the demands that are put upon them by the traffic environment (Hatakka et al., 2002).
The results indicate that the participants were slightly better at assessing risk than self-performance, although many participants indicated low risk even in scenarios where they exposed themselves or others to direct danger and thus failed.Participants who failed the motorway entrance rated the risk significantly higher, and commonly failed due to insufficient speed.Since this is the only scenario in the screening test where insufficient speed may be dangerous and is thus a criterion for failing, it may be suggested that these participants may be lacking in risk awareness, due to a failure of understanding what is risky and why (e.g.driving slower when a higher speed is safer).The GDE framework highlights the importance to be aware of the risks and correctly evaluate the own ability to perform the appropriate actions to correctly pace the driving.This is also the only scenario where a gender difference appeared, such that female drivers failed more often.A more cautious driving behavior among female drivers (Lajunen et al., 2022) may in this case be a disadvantage.
It is reasonable to believe that those who failed the screening test generally had a lower risk awareness, but this only applies to a few scenarios.In the scenario with roadwork and oncoming traffic (S14), participants who failed the screening test rated the scenario as lower risk, which may indicate deficiencies in risk awareness or that they incorrectly perceive a high level of control over the situation.The scenario is complex, and a series of events are required for it to become dangerous.The GDE framework highlights the importance to be aware of these complex risky scenarios beforehand, because of the limitations to correctly evaluate the risk when it is novel to them.Furthermore, risk awareness may involve the connection between understanding when there is a risk and understanding one's own ability to handle the situation.A driver may be risk-aware, yet not understand their own limitations, especially as a young novice driver (Fisher et al., 2006).

Future work
This project contributes to an increased understanding of how simulators can be used to complement driver training and on-road driving tests.A driving simulator enables testing not only the drivers' hazard perception, but also their risk awareness in case of detected hazards and decision-making.Evaluation of adapted behavior to decrease the demands below their own limitations is also possible (i.e., if they drive in a self-paced way).This makes driving simulators a valuable training tool and may give the student more influence and thus making their selfevaluation matter more in the assessment.
The GDE matrix could be used as a framework to systematically evaluate the possible uses a driving simulator could have in the Swedish driving license system.Both for knowing what skills to evaluate and devising ways of evaluating them for potential uses as a screening test or as a teaching tool.In future studies, it would also be useful to include questions in the background questionnaire to cover driver type and sensation seeking.
The next steps are scaling up with multiple versions of the same test scenarios, developing and validating many different and unpredictable screening scenarios, and reviewing regulations.However, the aim is not to replace the on-road driving test with simulator testing.These are two different tests that have different purposes and measure different abilities, and therefore they should complement each other.

Conclusion
The results of this study show that with a screening test in a driving simulator, it is possible to detect drivers who consider themselves ready to take a driving test, but who have not yet reached the level of risk awareness required to be a safe driver.
The GDE matrix can be used to create scenarios that consistently, and fairly, test the driverś risk awareness, hazard perception, situational awareness, decision-making, and vehicle control.These scenarios should be designed to complement the on-road driving test and clearly and systematically expose the participant to rare situations with significant risk.The GDE matrix can also be used when creating clear pass or fail criteria for the scenarios considering the knowledge and skills on the different levels of the matrix.
The high self-ratings of performance, even when failing, confirm the well-known lack of self-evaluation skills among young drivers, and although perceived risk was rated higher than moderate in most of the scenarios, major deficiencies in self-evaluation appear.To practice selfevaluation, the simulator can be used as a pedagogical tool, linked to the GDE matrix, and with the opportunity to match perceived hazards to perceived abilities to cope with these hazards.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Fig. 1 .
Fig. 1.Driving simulator used for the screening test.

Table 3
Number of participants who failed a certain number of scenarios and the number of these participants who subsequently passed their on-road driving test.
Fig. 2. Number of participants (n = 70) who failed, presented for each scenario in the screening test.

Table 4
Correlation values from test scenarios with at least one significant correlation between self-rated performance, self-rated risk, or outcome (pass or fail).* p <.05, ** p <.01.