2017 Volume 4 Issue 4 Pages 653-662
This study tries to contribute to the understanding of road safety in Indonesia, by utilizing a Negative Binomial Regression (NBR) and Zero-inflated Negative Binomial Regression (ZINBR). The goal is to build an accident estimation model and have better understanding of how the varying road elements and traffic conditions contribute to accidents. This study also includes the International Road Assessment Program (iRAP) star rating methodology in the crash estimation model. The study produced three main conclusions. First, star ratings have negative correlation with accident occurrence and are also poorly correlated with accident occurrence, which indicates that the risk factors and the widely used iRAP methodology needs to be adjusted. Second, several land use, roadway type and traffic compositions were found to be significant explanatory variables. Finally, based on AIC, BIC, Vuong and the Prediction test, ZINBR is a better model compared to NBR.