Regression analysis for estimation of the influencing factors on road accident injuries in Oman Poisson Regression Model and Poisson Alternatives

Road safety programs use statistical models to predict the occurrence of accidents and casualties and to identify the influencing factors that affect their occurrence. They are also used to identify the causes of an accident and the hazardous locations where more accidents happen (the hot spots or black spots). Causal factors could depend on human behaviour, road geometries, traffic volumes, weather, or the interactions among these. For decision makers, it is very important to understand road patterns and behaviours to apply road safety improvements and road maintenance activities effciently. Statistical modelling of road safety is conducted by taking the data of past accidents and the attributes of many sites and using them to produce the best prediction models. The objective is to discover the relationship between a function of the dependent variable (e.g., expected number of accidents at a certain point), E(Yi) = λi, in relation to number of covariates, Xi1, Xi2, Xi3 ,....Xik that are assumed to have an effect on the dependent variable Yi. It is a standard practice in road safety research to model accident counts Yi as Poisson distributed random variables that Yi ~ Pois (λi) corresponds to a random distribution of the accidents over time and space. Accident data have often been shown to exhibit overdispersion, which make it essential to use alternatives of Poisson to model such data. In this research, we apply the Poisson regression model and its alternatives in addition to the binary and ordered probit logistic regression model.