Computational Cognitive Modeling of Intervention Effects on Adolescent Drivers’ Crash Risk

The etiology of risky driving behavior among adolescents is of great interest to transportation scientists, yet existing explanatory frameworks of within- person changes in crash risk are underdeveloped. Population-level crash rates incrementally decrease following licensure, in a power-law pattern, which has led to speculation that recently licensed teen drivers’ crash risk also decreases incrementally as they accrue experience. However, it cannot be assumed that individual-level changes in crash risk mirror the population-level changes in crash rates. In statistics, this is known as an ecological fallacy and in formal logic it is known as the fallacy of division, a type of category error. Further, Myung et al., 2000 demonstrated that a power-law “artifact” can occur when data from non-linear models are arithmetically averaged in the presence of individual difference factors. Previously the authors demonstrated that aggregating individual-level abrupt decreases in crash risk accurately fits population-level crash rate data from over 1 million adolescents, thus demonstrating that the power-law artifact can be present in teen drivers’ crash data. To address the lack of explanatory frameworks that specify why population-level crash rates change so dramatically during the initial months of licensure and to contribute to the growing body of literature relating to the power-law artifact as it is observed in the context of insight, learning, and skilled performance the authors used a computational cognitive modeling approach to define a phase transition framework specific to the teen driver. The central principles of the computational model were: (1) transitioning from a riskier novice driver to a safer (less novice) driver can be accomplished through a series of phase transitions due to adopting new strategies, with abrupt transitions happening at different times for different adolescents and (2) interventions can cause an immediate phase transition as well as increase the likelihood of a future transition. In this report the authors demonstrate how this phase transition model accounts for effects of two interventions found to reduce police-reported MVCs (ODA, Mirman et al., 2018 and RAPT, Thomas et al., 2016). These effects are inconsistent with an incremental model of crash risk reduction predicated on the accrual of post-license experience. The authors used computational cognitive modeling to develop an individual-level account of the data from two US-based intervention trials administered to teens prior to the accumulation of any post-licensure driving experience and that used police- reported crashes as the main outcome. The phase transition model was defined mathematically using a sigmoid function, which produces a value between 0 (lowest possible risk state) and 1 (highest possible risk state), and a relatively rapid transition from the high-risk state to the low-risk state. Critically, the timing of that transition was assumed to differ across individuals and to be expedited or induced by the interventions of interest. Crash events were simulated based on binomial sampling from the individual crash risk curves. Results of the RAPT simulation illustrate the phase transition model’s strong fit to the data (Fig1). A modest (10%) increase in phase transition probability can translate into a substantial reduction in crash rate, consistent with the observed trial data. The ODA simulation results were highly similar. This model reflected a possible key difference between how the ODA and RAPT may have affected phase transition timing: inducing immediate phase transitions (RAPT) and increasing the probability of a phase transition (ODA). Results suggest that researchers should consider cognitive change processes that beget translational transformations in crash risk trajectories (i.e., shifting crash risk curves) instead of thinking primarily about learning rates and the relationship between accruing experience and crash reduction. Strategy acquisition is one such process. More detail about how to use the computational cognitive modeling approach to evaluate interventions and theories for teen drivers will be provided in our presentation.

  • Supplemental Notes:
    • This paper was sponsored by TRB committee ANB30 Standing Committee on Operator Education and Regulation.
  • Corporate Authors:

    Transportation Research Board

  • Authors:
    • Mirman, Jessica H
    • Curry, Allison E
    • Mirman, Daniel
  • Conference:
  • Date: 2019


  • English

Media Info

  • Media Type: Digital/other
  • Features: Figures; References;
  • Pagination: 5p

Subject/Index Terms

Filing Info

  • Accession Number: 01697703
  • Record Type: Publication
  • Report/Paper Numbers: 19-01938
  • Files: TRIS, TRB, ATRI
  • Created Date: Mar 1 2019 3:51PM