Methodology to Derive National Estimates of Injuries and Fatalities in Road Traffic Crashes in India

The Road Accident Sampling System - India (RASSI) accident database being developed by an international consortium of manufacturers and safety researchers is currently India’s only source of in-depth crash data. The database includes information on accident, vehicle, and driver factors associated with each crash, which is collected through on-scene crash investigations conducted by trained crash investigators, from four key sample regions (Coimbatore, Pune, Ahmedabad, and Kolkata). As the RASSI database continues to grow, the next step is to ensure that the sample data can be reliably extrapolated to the whole of India. This paper is an initial attempt to develop national estimates by crash type based on a few sampling locations currently being investigated by the RASSI teams in India. RASSI data was treated as a stratified sample of Indian accidents, and the locations, where the crash data is being collected, were considered as primary sampling units. The “mark and recapture” statistical procedurefor population estimation was used to derive sampling weights by accident type and injury severity. Sampling weights were derived by comparing RASSI data with the police reported data from the sampling units for the same period. The weights were based on several factors, including crash types (single-/multiple-vehicle), injury severity, crash location (urban/rural) and type of road user (pedestrian/motorized two-wheeler/car). Data from police logs and RASSI were matched by selected strata (injury type/accident type), and the estimate of total population for that stratum was calculated using well-established statistical methods. Then, national estimates of the various single-vehicle accident types (collisions with fixed objects, rollover, pedestrian, motorcycle) and multiple-vehicle accident types (head-on, rear, side impact, and sideswipe) were derived. Driver contributing factors and consequences were also estimated. The derived estimates at an aggregate level were compared with published sources including MoRTH data to determine and improve adequacy and validity of the weights.