Survival Analysis Of Accident Risks Of Car Drivers In Northern Region, Ghana

Abstract

The accident risks of drivers and the factors affecting the risks were analyzed using survival models. The data consisted detailed records of fatal accidents in the years 2007 - 2009, generated by the Motor Traffic and Transport Unit (MTTU) of the Ghana Police Service, Northern region. The data on fatal accidents contained 398 drivers. The objectives of the study included both analysis and explanation of driver accident risk factors and investigation of how survival modeling method can be used in traffic accident analysis. The research questions that were addressed include: can driver involvement in accidents and exposure be examined with survival models? Are age and sex major risk factors? Do vehicle characteristics contribute to accident risk? Does the use of unworn out tyres reduce accident risks? The analysis of the data using the SAS package confirmed that the most significant variables to the risks of accident were driver characteristics (age, driver behaviour, kilometerage driven, speed and use of safety belt), drivers’ state (driving under influence of alcohol) and vehicle characteristics (vehicle age and weight, and condition of tyres). Young and inexperienced on one hand, or old and experience drivers on the other hand, had the highest fatal accident risks. Drivers using worn out tyres had a somewhat greater accident risk than drivers using unworn out tyres, but the difference was not statistically significant. Survival modeling was used to analyze the data, and it was concluded that survival modeling promises to be a useful tool for road safety analysis.

Summary

The study examined the effects of human factors, kilometerage driven, vehicle characteristics and roadway factors on amount of accidents and risks of accident associated with driving. The study was based on accident data from the MTTU regional office of Ghana Police Service. This data included information on 398 drivers who had been involved in fatal accidents during the period under consideration.

Drivers’ accident risks and their dependence on various factors were examined using survival accident models. Survival modeling is commonly applied in medicine to the study of serious diseases and treatment methods. The present study tries to assess the development needs of survival models in the area of traffic accident analysis.

Several factors have bearing on driver’s accident risks. This can be explained by driver’s age and sex, annual kilometerage driven, age of vehicle, driving under influence of alcohol, the use of safety belt, speed at the site of the accident, the weight of the vehicle and the tyre treated depth. More specifically, the analysis confirmed the following answers to the main questions of the study;

The kilometerage driven was negatively related to hazard of accident. The more kilometers are driven, the lower the probability of being involved in an accident. This particular outcome needs further probing. 

Driver age proved to be a statistically significant accident risk factor, however there was no clear sex-related differences between accident risks of male and female drivers. 

Some vehicles characteristics can be separated into own individual risk factors but they are strongly interrelated to many other risk factors. 

Drivers of worn out tyres had a somewhat larger accident risk than the users of unworn out tyres, but the obtained differences were not statistically significant. The effects of driver’s motives and attitudes were not directly modeled in this study, but some of their influence can be assumed to underlay the effects of driver age, sex or experience, as well as determine their choosing speed, drive under influence of alcohol, or not use of safety belt.

The survival modeling analysis included model formulation, parameter estimation and validation. The main survival model type used was Cox Proportional model. It however revealed that some explanatory power is lost when survival models are based on random time period. The results proved that better exposure data e.g. time spent in traffic used in the analysis could improve the models. 

Generally, the application of survival models to the accident data appears to be a promising approach but still needs further development.