Exploration of Advances in Statistical Methodologies for Crash Count and Severity Prediction Models

This report first describes the use of different copula based models to simultaneously estimate the two crash indicators: injury severity and vehicle damage. The Gaussian copula model outperforms the other copula based model specifications (i.e. Gaussian, Farlie-Gumbel-Morgenstern (FGM), Frank, Clayton, Joe and Gumbel copula models), and the results indicate that injury severity and vehicle damage are highly correlated, and the correlations between injury severity and vehicle damage varied with different crash characteristics including manners of collision and collision types. This study indicates that the copula-based model can be considered to get a more accurate model structure when simultaneously estimating injury severity and vehicle damage in crash severity analyses. The second part of this report describes estimation of cluster based SPFs for local road intersections and segments in Connecticut using socio-economic and network topological data instead of traffic counts as exposure. The number of intersections and the total local roadway length were appropriate to be used as exposure in the intersection and segment SPFs, respectively. Models including total population, retail and non-retail employment and average household income are found to be the best both on the basis of model fit and out of sample prediction. The third part of this report describes estimation of crashes by both crash type and crash severity on rural two-lane highways, using the Multivariate Poisson Lognormal (MVPLN) model. The crash type and crash severity counts are significantly correlated; the standard errors of covariates in the MVPLN model are slightly lower than the other two univariate crash prediction models (i.e. Negative Binomial model and Univariate Poisson Lognormal model) when the covariates are statistically significant; and the MVPLN model outperforms the UPLN and NB models in crash count prediction accuracy. This study indicates that when simultaneously predicting crash counts by crash type and crash severity for rural two-lane highways, the MVPLN model should be considered to avoid estimation error and to account for the potential correlations among crash type counts and crash severity counts