Analyzing the Maximum Abbreviated Injury Scale in Vehicle Crashes by Using a Logistic Normal Model

This paper presents a model centered on the estimation of the maximum abbreviated injury scale–based injury predicting models by using a logistic normal model with random effects. The study identified energy equivalent speed, collision type, crash year, location, driver's age, and gender for the analysis of the gravity in vehicle crashes. The study showed that the collision type variable was modeled better by random effects than by fixed effects and that energy equivalent speed, crash year, location, driver's age, and gender were contributing factors with fixed effects to the injury severity. Therefore, the authors took advantage of the mixed logit model's ability to account for unobserved effects that are difficult to quantify and may affect the model estimation. Crashes from the database of detailed studies of personal accidents occurring from 1991 to 2010 in France were used. The estimation of the parameters (fixed and random effects) was performed by several approximation methods. Energy equivalent speed was found to be, by far, the most significant variable. If drivers age 75 and older were considered as a reference, drivers in the 25-to-34 age bracket were significant contributors and were followed by those in the 35-to-44 age bracket and then by those in the 55-to-65 age bracket. In addition, vehicles that crashed between 1995 and 1999 were more significant contributors than were those in the reference period, 1991 to 1994. For assessment of the performance of the model proposed, a binary logit model was compared with the mixed logit by means of cross validation. The obtained results revealed that the logistic normal mixed model is preferred because it detects more vehicle crashes with serious injuries than does the classic logistic regression model.