A Horvitz-type estimation on incomplete traffic accident data analyzed via a zero-inflated Poisson model

Abstract

To improve the road safety, policy makers relay on data analysis to enact new traffic policies. Accordingly, statistical modeling has been linked in various studies of road crash counts with excess zeros. On top of this excess zero problem, missing data are also likely to occur in the road traffic accident data. Unless the missing data are resulted randomly, the popular naive estimation may not provide reliable results for policy making. In contrast, the implementation of the Horvitz method, which inversely weights the observed data by a weight that are obtained parametrically or nonparametrically, results in reliable estimators. We received satisfactory results on the performance of our approach handling the missing data problems in both a Monte Carlo simulation and a real traffic accident data exploration.