Multinomial Logistic Regression Model for Single-Vehicle and Multivehicle Collisions on Urban U.S. Highways in Arkansas

Abstract

Multivariate analysis can be used to identify the effects of several factors on the causes of a crash compared with univariate analysis. This paper uses a multivariate-analysis technique, the multinomial logistic regression (MLR) model, to examine the differences in crash-contributing factors for six collision types for both divided- and undivided-highway nonjunctions, given that a crash has occurred. Multinomial logistic regression was used to investigate (1) single-vehicle and (2) multivehicle collisions, which included (1) angular, (2) head-on, (3) rear-end, (4) sideswipe-same-direction, and (5) sideswipe-opposite-direction collisions. The risks associated with different collision types were found to be significantly influenced by various vehicle actions. The risk of sideswipe-same-direction collisions was higher while changing lanes and merging on undivided and divided highways. Similarly, while merging, drivers were prone to angular collisions, and when slowing down to rear-end collisions on undivided and divided highways. On weekdays, there was higher risk of multivehicle collisions, whereas on weekends, single-vehicle collisions were found to be statistically significant. The risk of single-vehicle collisions attributable to drivers negotiating a curve, driving on a wet road surface, during nighttime, when vision was obscured, avoiding objects on the roadway, and driving under the influence of alcohol was higher compared with that of other collision types. On the basis of the results of the analysis, it was found that the risk of single-vehicle collisions was higher on divided and undivided highways compared with other collision types. Further, binary logistic regression model was used to identify the factors that contribute to crash-injury severity, given that a crash has occurred. Drivers and passengers who did not wear lap and shoulder belts, and drove under the influence of alcohol were involved in serious crash injuries. Drivers involved in a crash on horizontal and vertical curves were prone to severe crash injuries compared with crashes on straight and level roadways. Head-on and single-vehicle collisions were found to be at a higher risk for severe injuries compared with other collision types. Additionally, collision types were strongly related to driver behavior (decision making) parameters, such as merging, changing lanes, and slowing/stopping, compared with parameters such as roadway geometry, atmospheric conditions, and surface conditions. From these results, the importance of different statistical techniques is evident, as the significant variables varied for crash severity and different collision types.