An econometric framework for integrating aggregate and disaggregate level crash analysis

Traditionally, aggregate crash frequency by severity and disaggregate severity analysis have been conducted independently in the safety literature. The current research effort contributes to the safety literature by bridging the gap between these two different streams of research by using both aggregate and disaggregate level crash data simultaneously. To be specific, the study proposes a framework that integrates aggregate and disaggregate level models. The proposed framework allows for the influence of independent variables at the crash record level to be incorporated within the aggregate level propensity estimation. The empirical analysis is based on the crash data drawn from the city of Orlando, Florida for the year 2019. The disaggregate level analysis uses 20,204 crash records that contain crash specific variables, temporal characteristics, roadway, vehicle and driver factors, road environmental and weather information for each record. For aggregate level model analysis, the study aggregated the crash records by severity class over 300 traffic analysis zones. An exhaustive set of independent variables including roadway and traffic factors, land-use attributes, built environment, and sociodemographic characteristics are considered in this analysis. The empirical analysis is further augmented by employing several goodness of fit and predictive measures. A validation exercise is also performed using a holdout sample to highlight the superior performance of the proposed integrated model relative to the non-integrated crash count by severity model. The proposed model can also accommodate common unobserved spatial correlation among crash records within the same zone. The model results illustrate the benefits of developing an integrated model system for crash frequency and severity.

Automated summary

This study bridges the gap between aggregate and disaggregate crash data analysis by proposing an integrated model framework. The proposed model incorporates both aggregate and disaggregate data to improve the prediction accuracy of crash frequency and severity. The empirical analysis on Orlando's crash data demonstrates the superior performance of the integrated model.