Modeling accident occurrence at signalized tee intersections with special emphasis on excess zeros

In implementing effective remedial treatments at hazardous intersections, it often is necessary to identify the geometric and traffic factors that lead to accident occurrence. However, one particular problem frequently encountered in accident studies is how to distinguish virtually safe intersections with little likelihood of accident occurrence from those that have happened to have no accident due to the random process. To deal with this problem, the "excess" records of zero accident, the zero-inflated negative binomial was used to assign the probability to the accident outcome. Accident data at 104 signalized tee intersections in Singapore over a period of 9 years were employed for model development. The model indicates that uncontrolled left-turn slip road, permissive right-turn phase, existence of a horizontal curve, short sight distances, large number of signal phases, total approach volume, and left-turn volume may increase accident occurrence. On the other hand, right-turn channelization, acceleration section on the left-turn lane, median railings, and more than 5% approach gradient may reduce accident occurrence. Moreover, there is a trend of reducing accidents over the years.