CHAPTER 4
Conclusions and Suggested Research

Conclusions

This research highlights several practical implications for practitioners and policymakers in the field of teen driver training and licensing. First, the review of existing driving training and licensure practices among U.S. states and OECD countries demonstrates a general lack of scientific justification for many current testing procedures. The experimental research findings are a first step toward addressing the lack of existing evidence-based interventions by identifying a hazard perception training intervention that can significantly improve hazard perception test performance. Further, the research found a link between test performance and driving risk. These results underscore the potential of integrating hazard perception training and testing into driver education curricula to better prepare young drivers for real-world challenges. Additionally, these results provide a data-driven basis for state driver licensing agencies to consider mandating or incentivizing hazard perception training as part of the licensing process.

This study was the first to prospectively enroll a large cohort of teenage drivers at the start of their driver education. By leveraging multiple data streams—including driver education milestones, hazard perception test results, and telematics data—the research team identified new associations that deepen understanding of teen driver risk. Notably, the finding that the intervention group outperformed the control group on the hazard perception test suggests an initial benchmark of 63 percent as a potential passing threshold for the post-training assessment. This benchmark could also serve as a screening tool to identify drivers needing further education or training before obtaining their licenses. Additionally, the experience of piloting the hazard perception training and testing paradigm allowed identification of key barriers that may affect large-scale implementation. Recognizing these known challenges enables implementing agencies to proactively address them, facilitating a smoother adoption of hazard perception training and testing going forward.

Suggested Research

To advance research based on the findings of this report, further studies are needed to explore the relationships between key predictors and long-term outcomes that occur further upstream in the driving experience. This includes examining how hazard perception performance influences driving risk following licensure, as well as crash risk during the first year of independent driving. The findings relating hazard perception training and behind-the-wheel exam performance with driving risk based on telematics data represent an important step forward for the field, with the use of objective data sources to understand how driver education and testing can be used to help novice drivers become safe drivers. Given that this is first-of-its-kind research and research that may be hindered by the potential for measurement error with use of a smartphone-based

telematics app, it is important to conduct confirmatory research to see if the findings can be replicated with other populations and at other time points. Assessing associations with additional outcome data, such as crash and citation data, could be important to add further weight to the driving risk classification. Additionally, longitudinal research could help identify developmental trajectories of hazard perception skills and determine the most critical intervention points to enhance driver safety from the outset.

For drivers who score poorly on hazard perception scenarios, it is necessary to identify the most effective intervention strategies. Options include providing additional computer-based skills training, offering more behind-the-wheel practice, or a combination of both. Furthermore, identifying the prospective association between performance on specific hazard perception scenarios and safety outcomes is needed.

To improve the implementation of the recommendations outlined in this report, further investigation is necessary to assess the scalability of hazard perception training and testing across different contexts. Research should focus on identifying the most effective and feasible methods for integrating hazard perception training into existing driver education programs. Consideration should be given to tools needed to support driving instructors and ways to support the accessibility of the training modules across various learner populations, such as including subtitles for non-native English speakers. Additionally, the content could be further expanded to provide additional scenarios. In the current iteration, only the rear-end section was predictive of driving risk. However, with additional modifications, researchers could continually assess how changes to training and testing relate to outcomes and can work toward a paradigm with testing that has greater predictive value.

This study further highlights the need to develop evidence-based tools for states to include in the driver training and licensing system. A national clearinghouse of evidence-based tools could be created to simplify the adoption of evidence-based driving examination processes. This could be achieved through the dissemination of tools to states to update their existing driving exams. To realize this vision, what is required is a commitment from states to the continual improvement of their driving examination processes.

Limitations

Several limitations should be considered when interpreting the findings of this study.

First, adherence to the requirement to use the telematics driving app was not universal. Approximately two-thirds of participants installed the app, and among those who did, data collection was subject to certain constraints. To function correctly, the app required sufficient phone battery and enabling of permissions. Additionally, the trips included in the analysis relied on the algorithm developed by the telematics provider to correctly infer transportation mode and driver status. This introduced the possibility of misclassification, whereby some driving trips may have been omitted and some nondriving trips mistakenly included, potentially affecting the performance of driving behavior assessments.

A further limitation to reliance on telematics data is the inability of telematics data to account for the context of driving. Although hard-braking events are considered key indicators for crash risk, there may be instances where hard braking is the safer option, such as when a driver needs to quickly stop for a traffic light that changed to yellow. Such contextual considerations are not able to be incorporated into telematics data at this time.

Second, the evaluation of the interventionʼs effect on hazard perception performance and driving behavior followed an intention-to-treat approach. This means that all students assigned

to the intervention were included in the analysis, even if they did not complete the training. Moreover, among those who did complete the training, engagement levels are likely to be varied. These factors may have led to an underestimation of the interventionʼs true effect. Furthermore, if hazard perception training and testing are mandated by a state then the effects may be increased. However, the models incorporating post-training assessment performance in relation to driving risk (via logistic and ordinal regression) provide a more direct measure of potential intervention effects.

Additionally, there are limitations in the hazard perception training and assessment in the current format. Specifically, the hazard perception exam did not have any time constraints, which enabled participants to take as long as they needed to try to identify the hazard present in a given scenario. Future iterations of the hazard perception training and testing program could incorporate in the scoring process information on how long participants take on each question. A further limitation is that the form of the test was identical between pre- and post-tests. For those who completed the post-test in quick succession, this could have resulted in some learning the test rather than the skills. Future iterations of the hazard perception test could incorporate additional driving scenarios to avoid repetition.

Finally, analysis of the association between the behind-the-wheel test data and risky driving behaviors measured by telematics was limited by several factors. The predictive power of the model was modest. These results suggest that while the model successfully captures and predicts some elements of driving risk, additional data is needed to characterize risk profiles more fully. To overcome this limitation, exploratory factor analysis was employed to identify those groupings of driving test maneuvers that predicted risky driving, showing that drivers who were penalized for lane control and intersection speed errors and parallel parking and hill parking failures during the behind-the-wheel exam were roughly 30 percent more likely to fall into a higher-risk driving category. Here, again, further research is recommended to support or validate these findings.

Despite these limitations, this study provides valuable insights into the effectiveness of hazard perception training and its potential role in teen driver education and testing. Future research should address these challenges by exploring ways to improve data completeness, engagement with training materials, and integration of hazard perception assessments into driver education curricula and driver testing processes.

Driving Test Recommendations

The skills exam represents a critical junction where safe driving skills can be verified before licensure and where practicing safe driving skills can be encouraged for novice drivers preparing for the exam. The current form of the skills exam centers around having novice drivers drive a predefined route that contains certain features (e.g., at least two traffic lights) and complete a series of maneuvers (e.g., parallel parking). Failure to complete a maneuver or completing the maneuver while making an error, such as striking the curb, results in point deductions. There are also specified dangerous actions that result in automatic failure, such as dodging out of the way by a pedestrian.

The exam in its current form has some association with driving risk, as measured through a telematics device, although the strength of the association is limited. This could in part be because novice drivers adjust their driving to be safer while in the presence of the examiner, such as by avoiding speeding and phone use during the exam. It could also be due to measurement error, such as with the telematics device improperly classifying a trip as a passenger as a driving trip. Despite limitations with using telematics data as the source of driving risk classifications,

important insights can be drawn from the review of skills exam data in relation to driving risk, which suggests several recommendations to strengthen the skills exam. In the longer term, the research team recommends transitioning to a simplified competency-based scoring sheet. However, as an initial step in this direction, the team recommends incorporating an independent drive component to the driving exam that follows a competency-based rubric. These and other recommendations are described in more detail here:

1. Incorporate an independent drive component into the skills exam. The review of driving exam components adopted in other countries found that some countries—including the Netherlands, United Kingdom, and Chile—incorporate an independent drive component when testing drivers before licensure. These components provide the driver an opportunity to drive without instructor direction and can provide a critical window into general driving performance in an unstructured format. They allow instructors space to observe the driver and their approach to engaging in traffic scenarios and evaluate whether the independent drive is completed in a safe manner. The components can also provide a space to assess safe driving skills such as situational awareness.
2. Incorporate a hazard perception exam after the knowledge exam and before the skills exam. These research findings suggest the current driving exam may fail to fully capture aspects of unsafe driving, while the hazard perception test can offer some added predictive value for unsafe driving behaviors. Therefore, building on the hazard perception test and integrating it into the licensure exam protocol could offer a way to strengthen the utility of the licensure process in ensuring safe driving. The hazard perception exam used in this pilot study offers an evidence-based set of 18 questions that can form the basis of a testing platform. The platform could further encompass complex scenarios that mirror real-world driving challenges, in which multiple high-predictive skills must work together. As further evidence of efficacy of other hazard perception testing becomes available, new test scenarios can be incorporated in addition to or as a substitute for the RAPT scenarios used in this pilot study. Incorporating this test requirement and offering free training could be one mechanism for incentivizing novice drivers to focus on developing this key driving skill. Over the longer term, this testing platform could integrate other components, such as attention maintenance or adaptive testing, if such items are supported by evidence to improve safe driving.
3. Emphasize items identified by the modeling that predict risky driving, by adding additional weight to these categories in the scoring protocol. This analysis of individual driving exam items identified two key groups of items that were predictive of high-risk driving. One category identified drivers with a lack of vehicle control who were unable to successfully parallel park and who struggled to complete the task of parking on a hill. A second category is related to judgment in traffic situations, including items related to lane control and intersection behavior. Such items could be given greater weight in the scoring protocol.
4. Eliminate items that fail to differentiate between low- and high-risk drivers. To accommodate new elements to the driving exam, the research team recommends eliminating those behind-the-wheel test items that have limited predictive value for driving risk and are rarely scored as errors on the test. Specifically, analysis suggests that the following maneuvers could be potential candidates for elimination: following, passing, and response to traffic control devices. Where appropriate, foundational items related to these maneuvers could still be incorporated into a set of critical errors that lead to penalties. Aspects could also be integrated into other areas, such as having “following too closely” as an item leading to a poorer result on the independent driving section.
5. Transition to a performance-based driving exam scoring system. Findings from the pilot study indicate that a driverʼs ability to perceive hazards is a stronger predictor of driving risk than individual driving errors recorded during an exam. This supports a shift from an error-based scoring system to one focused on evaluating real-world driving performance and the

1. core competencies required for safe driving. To this end, the research team recommends requiring a hazard perception assessment before licensure and incorporating an independent driving component into the skills exam. If these additions demonstrate a positive impact on the licensing process, the next phase should involve redefining the entire exam around essential driving competencies. This would include building consensus on the core skills required for safe driving and developing a standardized rubric to evaluate drivers on those skills. The AAMVA Model Driverʼs License Manual (2022) offers a potential roadmap of key driving maneuvers (under Section Seven) such as turning, parking, and changing lanes. It also offers a set of key aspects of safe driving (under Section Eight) such as visual search, speed management, space management, and communication.