A macro-level analysis of safety data using geospatial techniques and spatial econometric methods and models
Motor vehicle accidents are a source of many preventable injuries and deaths, worldwide. Several statistical and econometric models have been developed to predict and explain crash events. Research indicate that 93% of traffic accidents are due to human error. The objective of this research is twofold – first, to develop a macro level safety planning framework by identifying socioeconomic factors that influence crash frequencies and second, to characterize traffic congestion attributed to a crash events. To this effect, a Geographically Weighted Poisson Regression (GWPR) model, a suite of Spatial Econometric models and a Mixed Logit model were estimated. Data used included crash records from 2009 to 2013 in Alabama comprising 647,477 crash events. These included 4,814 crashes on Interstate 65 and 21,818 crashes related to Driving Under the Influence (DUI). Other data comprised socioeconomic data from US census, weather data, traffic data, spatial data from ESRI and crowd sourced speed data. Results indicate that DUI crash rates and frequencies at postal code level are predominantly influenced by rate of employment, income, population density, level of education, household size and housing characteristics. In addition, level of congestion attributed to a crash depends on factors including traffic volume, speed, weather, time of the event, severity of the crash, presence of physical barrier separating opposing traffic lanes, work zone, percent of heavy trucks and whether the crash occurred in an urban area or rural area. These results are unequivocal regarding the importance of geographic variation and heterogeneity in driver behavior and the general road safety.