Identifying Factors Related to the Estimation of Near-Crash Events of Elderly Drivers

Abstract

This study attempted to identify factors associated with driving behavior of elderly drivers to assess their safety and estimate their risk during naturalistic driving. We performed binomial logistic regression using self-reported past crash involvement as a response variable to identify critical factors and provided an estimation model has 18 variables. However, applying driver category based on crash and near-crash events (CNCs) collected from naturalistic driving study employed on-dash cam instead of self-reported crash involvement to the previous model showed lower predictive performance (0.63 for sensitivity and 0.51 for specificity). This implies that the model based on self-reported crash experiences was difficult to detect for drivers with CNC during naturalistic driving. Then, we performed binomial logistic regression based on CNC involvement and indicated another model, where the predictive performance was improved, with 0.81 for sensitivity and 0.70 for specificity. To predict the number of CNCs as drivers’ risk, this study adopted Poisson regression analysis using nine variables selected from the second model. The analyses showed a plausible model and significant variables for the estimation of CNCs. Mini-Mental State Examination (MMSE) was one of the better predictor putting in this model, and showed the probability that lower performance associated with higher number of CNCs. This model for CNC estimation would be helpful for the development of safety programs for elderly drivers with possible incidents.