Despite the tremendous efforts made in recent years towards improving overall health status of adolescents, road traffic crashes remain a global problem worldwide among teen drivers. It is well established that the first few months of independent driving are the most dangerous. Indeed, crash risk among adolescent drivers is particularly high during the early months of independent driving, after which it starts to rapidly decrease over a period of over a period of years. Hypotheses for this decline have focused on Some researchers have hypothesized accumulation of driving experience, maturation, and increasing self-regulation. However, the mechanisms by which they interact to decrease teen crash risk in few months are not well understood. Additionally, safety researchers are engaged in a longstanding quest to fundamentally improve teen driving. To that end, increasing number of studies have been striving for solutions. Understanding the processes underlying patterns in teen crash risk and catalyze effective teen driving interventions can benefit from techniques for modelling complexity. The goal of this project was to develop a model that provides initial insights into the mechanisms underlying adolescent risky driving patterns over time. The purpose of the modeling is to investigate how much faster the early improvement of teen risky driving could be with interventions.
This study utilized naturalistic driving data derived from a clinical trial study. A sample of newly-licensed teen drivers and at least one of their parents was recruited from high schools in Iowa and randomly assigned to one out of three groups: control group, feedback group, and feedback plus parent communication group. Each participant's vehicle was equipped with an event triggered video recording system to gather data on near-crashes and crashes as well as their proxies denoted risky driving events. The video recording system was installed in the vehicles of the control group only for data collection purposes. For the feedback intervention group, teen drivers received an immediate feedback via blinking of LED lights on the in-vehicle video system when a driving error occurred. In addition, each teen and their parent in this feedback group received a weekly report card that summarized the types of driving errors made by the teen and provided video clips of those errors. The feedback plus parent communication group was exposed to the feedback intervention described above plus communication strategies for discussing safe driving with teens. The video recording system was also used to collect data on mileage, driver behaviors (eg. traffic violations, cell phone use), and traffic conditions (eg. snow, rain).
The first aim of this study thoroughly investigated heterogeneity in driving outcomes within the population of teen drivers. Results showed two distinct risky driving trajectories, including one inverted U-shaped pattern (initial increase in risky driving followed by a steady decrease) and one relatively constant pattern over time. Risk-taking behavior trajectories were found to follow the same patterns as risky driving. The study also identified two groups of teens with respect to amount of driving: one group has a linear increase in the amount of driving and the second group has an upward U-shaped pattern. Teens classified in the high risk-taking behavior group are more likely to be in the high risky driving group whereas the teens classified in the low risk-taking behavior are more likely to be in the low risky driving group. Results showed that males are more likely to be in high risky driving and high risk-taking behavior groups compared to females.
The second aim of this project was to develop a dynamic model of teen risky driving and use this framework as a guide to leverage an understanding of the dynamic process underlying patterns in teen risky driving over time. The analysis suggests that the natural risky driving behavior (absent intervention) is slow improvement followed by faster improvement, and finally a plateau: that is, S-shaped decline in errors. The results showed that a model that includes cumulative miles driven and recent risky driving events as stock variables and their feedback is capable of explaining the dynamics of teen risky driving over time. The analysis suggests the existence of a reinforcing loop and two negative feedbacks. The reinforcing loop arises from a decline in recent events leading to a faster increase in driving; this leads to a faster accumulation of driving and thus a greater decrease in driving error rate; the decrease in driving error rate leads to a further decline in recent events via a slow replenishing of the stock “recent events”. The first negative feedback is from recent events to amount of driving. By this feedback mechanism, more recent events (or memories of events) lead to less driving, and thus slow accumulation of driving experience (cumulative miles driven). The second feedback in the model is from recent events to event rate. A greater number of recent events (or memories of events) leads to a decrease in event rate perhaps via corrective actions taken by the teen driver. Thus, more recent events (or memories of events) lead to a decrease in event rate, but slow accumulation of experience via less driving. The results highlight that variations of individualized trends in driving event rate and monthly driving are more likely due to significant variations in the stock “cumulative miles driven” and the stock “recent events”. Variations in these stocks are influenced by initial event rates and driving need. The methodological approach provides an explanation for the peak in crash rates during the latter months post-licensure rather than the first month, which was not fully understood.
The third, and final, aim of this teen driving dynamic model project sought to simulate driver feedback intervention and conduct its cost effectiveness analysis. To examine the impact of driver feedback intervention and its tradeoffs, a previous version of the model was extended to create a model that allows the simulation of the intervention and the comparison between its expected costs and benefits. The analysis suggested that the simulated intervention data are comparable to data from actual feedback intervention group. The simulation results indicate significant differences in the period over which the intervention is needed. While the intervention is economically beneficial for some drivers, it is worthless for others. The model also suggests the need of combining several interventions for some drivers for a faster improvement in risky driving.
This research offered initial insights for understanding risky driving patterns, risk-taking behavior, and amount of driving among adolescent drivers and can be helpful when designing teen driving interventions, as the different trajectories may represent unique strata of crash risk level. The dynamic model developed can be used to design and evaluate teen driving interventions in order to identify key leverage points to guide policy and direct the optimum combination of prevention strategies.