Highway Research Record

V. TRAFFIC AND TRANSPORTATION
2. SAFETY AND ENVIRONMENT
NEW PROJECTS
Road Accident Modelling for Highway Development and Management in Developing Countries

Date of Start:December 2004
(i) Indian Institute of Technology, Madras (R,C)
(ii) Transport Research Laboratory, United Kingdom (S)

Scope and objectives
(i) To improve transport safety and to reduce impact of accidents on people in rural and urban areas.
(ii) To provide reliable predictors of road accidents for highway development models used in the planning of new and upgraded      roads.
(iii) To provide a new road safety component in the HDM-4 model.

Methodology

The project aims at modeling road accidents as a function of exposure and roadway characteristics. 500km road length has been selected for the study, which includes National Highways, State Highways, Major and Other District Roads and Hill Roads in urban and rural locations. Kilometer wise accident data on the selected road stretches, for the past three years, has been collected from police records. Other data including traffic volume, width of pavement/shoulder, condition of pavement/shoulder, condition of signs/markings, sight distance at curves, roadside friction due to pedestrian activity, type and proximity of off-shoulder hazards and number public/private accesses have been collected through field surveys. As the data includes variables of both quantitative and subjective nature, accident frequency is being modeled using Generalised Linear Modelling (GLM) technique.

Conclusions
The following interim conclusions are drawn from an initial analysis on part of the data
(i) Road accident frequency on two-lane, undivided rural highways was modeled as a Log-linear function of classified traffic     volume, number of public/private accesses, condition of carriageway/shoulder and off-shoulder hazard rating. It was     assumed that accident frequency follows Poisson distribution.
(ii) The observed accident frequency showed over-dispersion due to preponderance of zero-crash observations. This suggests      use of distributions other than Poisson for further modelling.

Significance/Utilisation Potential
The final model will provide a new road safety component in the HDM-4 model.

Reports/Publications
Lt. Col. Balachandran, G. (2005), ‘Road Accident Modelling for Two Lane Undivided Rural Highways Under Mixed Traffic Flow Conditions’, M. Tech. Theis report (unpublished), Indian Institute of Technology, Madras.