Microscopic accident prediction models for signalized intersections

• Main Author: Quintero Toscano, Mario Alberto
• Language: English
• Published: 2009
• Online Access: http://hdl.handle.net/2429/10333

The main objective of this thesis is to develop microscopic accident prediction models for estimating the safety potential of 4-leg signalised intersections in the City of Vancouver, B.C. and describes the applications of these models in traffic safety analysis. The aim, therefore, is to examine the traffic variables that appear to underlie the occurrence of accidents of these intersections and explain, in a statistical sense, the generation of accidents as a function of these variables. Generalised linear regression was employed to develop the models because of its superiority over conventional linear regression in modelling accident occurrence. The statistical software package GLIM4 was used to accomplish this task. The study made use of a sample of 8466 accidents that occurred at 170 4-leg signalised intersections during the years of 1994-1996. The data on accident frequencies and traffic volumes were obtained from the City of Vancouver. Several models that have different applications in the field of traffic safety were developed in this study for the 4-leg signalised intersections of the City of Vancouver. Different error structures that can be utilised to model the relationship between accidents and traffic flows are reviewed. Microscopic models for different accident types were developed in conjunction with macroscopic models for Total, Severe and Property Damage Only accidents. The microscopic models are presented in conjunction with the three macroscopic models which all resulted in statistical significance. Several model applications are discussed. Examples of how to obtain location-specific safety estimates, how to identify accident prone locations, how to rank the accident prone locations, and how to conduct a before and after safety evaluation are presented. Microscopic and macroscopic models are used simultaneously to determine which intersections should be regarded as accident prone locations according to specific accident patterns that can be effectively treated by engineering countermeasures.