Generalized Linear Modeling of Crashes on Urban Road Links

As crash data have distinctive behavior like over-dispersion, researchers used statistical methods to deal with this unique of specifically. This study employed generalized linear modeling techniques develop the model. It was assumed that accident counts followed negative-binomial distribution, and link function chosen be log function. Negative-binomial technique over Poisson distribution because it is most by many as may encounter over-dispersion. The set showed greater variability between its variance mean. frequency shown in highly skewed, a very high number road segments registering zero accidents. Negative binomial after comparing Akaike’s Information Criterion (AIC) Bayesian Criteria (BIC). method widely applied count data. Twenty-two parameters were estimated Since p < 0.05 omnibus test, null hypothesis rejected, which indicates model reasonably fit. strongest variables witnessed length links, lanes, average daily traffic, bus lane, buses trolleys, HGVs.