Unobserved heterogeneity in ramp crashes: Insights from nonlinear random parameter and random parameter with heterogeneity in means and variance approaches

The dissertation aims to investigate the heterogenous effects of factors on crash frequency & severity on ramps while considering unobserved heterogeneity through advanced models. The dissertation comprises of two parts. The first section investigated the heterogeneous effects of ramp type, configuration, spatial footprints, traffic, and geometric characteristics on ramp crash frequency, while the second part showed a new method to capture nonlinear relationship of ramp crash severity and contributing factors involved in crash considering unobserved heterogeneity. The first part of the dissertation presents a negative binomial random parameter model with heterogeneity in means and variance to capture the effect of heterogeneous effect of ramp type, alignment, truck volume and interchange geometry and on freeway ramp crash frequency. Two years (2018–2019) of crash data on freeway ramps in Washington State were analyzed. Model estimation results show ramp type (directional, semi-directional and loop), alignment, and traffic characteristics significantly impact ramp crash frequency. The northwest loop ramp indicator has a random parameter. The minimum horizontal curve radius and the total number of vertical curves on the ramp appear to be statistically significant sources of heterogeneity in the mean of this parameter. Heterogeneity in the mean of the random effect is influenced by single truck percentage and the low AADT indicator (<=1,340 vehicles per day). Heterogeneity in the variance of the northwest loop ramp random parameter appears to be associated with the southwest loop ramp indicator indicating unobserved effects due to same-side loop geometries. Directional ramp indicators (on- and off-ramps) and interactions involving speed limit, AADT and horizontal curve radius are statistically significant (as fixed parameters) in their impact on ramp crash frequency. Total centerline mile footprint of all ramps at the interchange is a continuous fixed parameter effect. Ramp-specific lengths (longer than 0.335 miles) also appear to be statistically significant. The findings in this study suggest that ramp and interchange design need to account for a holistic integration of spatial footprint, type of ramp and alignment factors, in addition to traffic flow variables. The second part of the dissertation explores the true nonlinear relationship between contributing factors such as driver behavior and vehicle-related factors and injury severity on different ramp segments while considering unobserved heterogeneity. Nonlinear and linear (standard) random parameter multinomial logit models with heterogeneity in means were estimated using three years of crash data (2018-2020) on freeway ramps in Washington State for the gore-to-gore, merge-diverge, and combined ramp segments. Model estimation results show the nonlinear models are superior to the linear models of all ramp segments based on log-likelihood, AIC, BIC, and model-predicted versus observed crashes by injury severity. Nonlinearity and heterogeneous effects are captured in the random parameter of the total number of vehicles involved in a crash under the possible injury utility. The sources of heterogeneity in the mean of random parameter estimates differ by ramp segment. Heterogeneity in the mean of the random effect is influenced by driver age of 70 or higher, sideswipe collision type, and distracted driver in the gore-to-gore ramp segment, by following too closely in the merge-diverge segment, and by speeding vehicles in the combined ramp segments. In addition, the significant variables with fixed parameter effects vary in the different ramp segments. Indicator variables of not granting a right of way, the interaction of foggy weather and wet road conditions, and the interaction of vertical and horizontal curves were only found statistically significant in the merge-diverge ramp segments due to additional workloads on drivers such as left and right turns, and interactions with traffic in other direction. While indicators of speeding vehicles, and same-direction vehicle movement prior to the crash, which reflects wrong-way driving in ramps, were only found statistically significant in the gore-to-gore ramp segments. These signify the necessity of investigating the traffic safety of ramps by segments. The integrated approach of nonlinearity and unobserved heterogeneity consideration provides detailed insight into the complex interaction of the total number of vehicles. This finding contributes to the knowledge of multiple-vehicle crashes, which are recorded to be substantially higher than single-vehicle crashes in the United States.