Evaluating and Comparing Spatial Variations in the Effectiveness of Graduated Driver Licensing (GDL) Programs in USA


This study aims to (i) identify areas where focused promotional and educational efforts are needed to achieve greater GDL’s effectiveness, (ii) detect variables that are associated with the fastest and slowest growth in effectiveness, and (iii) indicate the point at effectiveness plateaus within areas. We have focused on developing a general modeling framework in that we considered the number of fatal teenage-driver car crashes in counties in Michigan.

Injury resulting from motor vehicle crashes is the leading cause of death to teens in the US. Few programs or policies have been found that are effective in reducing the crash risk of young novice drivers. For example, driver education, which is required for driver licensure under age 18 in most states, has no safety effect. GDL is an effective policy that has been implemented in most US states and Canadian provinces. The effectiveness of policies can be tested as both temporal and spatial effects. Temporal effects are in terms of onset and variation over time. Spatial effects are seen as variation across regions, and can be correlated with regional characteristics to understand policy adoption. Information about the spatial distribution of policy effects can be used to achieve more uniform policy adoption and better overall effectiveness. Published evaluations of GDL effectiveness has focused on the temporal domain analysis using pre-post study designs, and no studies we are aware of have evaluated GDL effectiveness from a spatial perspective. This study used the Fatality Analysis Reporting System (FARS) and other public databases from US Census Bureau and US Bureau of Labor Statistics to provide a spatial evaluation of the GDL effectiveness. This study aims to (i) identify areas where focused promotional and educational efforts are needed to achieve greater GDL’s effectiveness, (ii) detect variables that are associated with the fastest and slowest growth in effectiveness, and (iii) indicate the point at effectiveness plateaus within areas. We have focused on developing a general modeling framework in that we considered the number of fatal teenage-driver car crashes in counties in Michigan. To account for the spatial dependence we introduced spatial correlation among counts relative to adjacent counties through spatial random effects. Because these spatial random effects a Conditionally AutoRegressive (CAR) model, our approach improves statistical power by borrowing information from neighboring counties. More importantly, our proposed approach enabled us to study and understand how the spatial variation in the effectiveness of GDL may be associated with some socio-economic factors. Our findings would provide some data evidence useful for policy-making and intervention pertaining to teenage-driver safety. Our analysis confirms the previous finding that the GDL in Michigan is an effective policy to significantly reduce the risk of vital car accident. In addition, our analysis unveils that rurality index is important contextual variable to explain the spatial difference of GDL effectiveness in the state of Michigan, so is white American percent. This project has attracted two new searchers, PhD student Yu Chen and Dr. Veronica Berrocal, into the field of injury research. The development of a general spatial modeling framework and computing methods based on winBUGS software is exemplary to the analysis of car crash data. Our findings from a spatial modeling perspective about important contextual variables associated with the spatial variations of the GDL effectiveness in Michigan are of interest, and will be submitted for publication.