- This dissertation provides a comparison of statistical and econometric frameworks, using a previously unused freight data source, to study crash frequency by crash type and heavy-vehicle hard braking in Oregon. Hard braking can serve as a proxy for several factors, one of which is safety. Therefore, with the hard braking literature being limited to regenerative braking of electric vehicles, behavior modeling, and stopping performance, this dissertation uniquely fills the gap in literature as it pertains to hard braking of heavy-vehicles explicitly in a safety context. Hence, based on the data, the four most occurring crash types at areas prone to heavy-vehicle hard braking are analyzed: rear-end crashes, turning movement crashes, fixed-object crashes, and sideswipe (overtaking) crashes.
The new norm in transportation safety analyses is to account for the unobserved heterogeneity (unobservables) in the crash data. However, specific events may be linked spatially, which can result in spatial correlation. With this in mind, this dissertation seeks to analyze each crash type by fitting a model that accounts for unobserved heterogeneity and a model that accounts for spatial correlation. This is accomplished through a detailed data process and spatial analysis. To compare these analytic methods, overall model fit (log-likelihood values) and the rate of correctly predicted crash frequencies are assessed. With crash frequencies being non-negative integer count values, specific count-data models are applied: Poisson regression if the data is not over- or under-dispersed, Negative Binomial regression if the data is over- or under-dispersed, and Spatial Lag of X variants of these models if there is significant spatial correlation. Results show that three of the four crash frequency models have a better fit when accounting for spatial autocorrelation. But, the rate of correctly predicted crash frequencies for each crash type is substantially higher when accounting for unobserved heterogeneity.
As stated previously, through the comparison of the crash frequency analysis frameworks, the predictability power of the “heterogeneity” models outperformed the predictability of the spatial models. In addition to identifying a preferred method to model crash frequency, this dissertation shows the viability of a new freight data source for transportation research. More, the importance of studies regarding hard braking and crash frequency have been presented. This dissertation uniquely fills this gap in literature.
Finally, an analytical foundation and recommendations have been provided. With regard to the analytical method, several Departments of Transportation (DOT) use traditional crash frequency analysis methods; but, they typically do not account for unobserved heterogeneity. This work has shown that the Oregon DOT (location of the current study) can generate a higher rate of prediction by accounting for these unobservables in crash data. For recommendations, this dissertation identifies methods to monitor hard braking (both for heavy-vehicles and other classes of vehicles). In addition, the locations of heavy-vehicle hard braking hot spots, along with the significant crash frequency contributing factors, can assist the Oregon DOT in identifying specific countermeasures to mitigate hard braking events and crash frequency by crash type.