Development of a Crash Prediction Model for Older Drivers
Dr. Michael A. Knodler Jr., Principal Investigator
Assistant Professor, Department of Civil and Environmental Engineering
University of Massachusetts/Amherst
Sponsor: New England University Transportation Center, Region 1
Older drivers are more likely to crash and to be fatally injured when involved in a crash. (IIHS, 2001; Li, 2003) As the driving population ages, jurisdictions are struggling to find ways to re-assess older driver competency in an equitable, cost effective manner. Because it will allow drivers to be identified as “high risk” based on objective criteria – recent driving performance – crash prediction modeling is a logical mechanism for identifying high-risk older drivers. In addition, the factors contributing the greatest weight to the prediction model will likely identify potential areas for focused retraining. We propose to use historical data from Massachusetts’ statewide crash, driver licensing, and citation datasets to derive and validate a crash prediction model that will identify a subgroup of older drivers at high risk for a near term injury causing crash. First, a crash prediction model will be derived using information from Massachusetts’ crash, citation, and driver history datasets. These data will be deterministically linked using unique identifiers at the driver level for each of the more than 670,000 licensed drivers aged over 64 in 2004. To be included in the derivation set, drivers will have licensing data available for the previous 10 years. The primary outcome of interest will be driver participation in an injury crash in 2004. Secondary analysis will evaluate for significant differences in the model if moderate or serious injury/fatal crashes are the primary outcome. We intend to use Poisson or negative binomial regression modeling, depending on the characteristics of the data.