Collision Prediction Models with Longitudinal Data: an Analysis of Contributing Factors in Collision Frequency in Road Segments in Portugal

In spite of the strategic importance of the national Portuguese road network, there are no recent studies concerned with either the identification of contributory factors to road collisions or collision prediction models (CPMs) for this type of roadway. This study presents an initial contribution to this problem by focusing on the national roads NR-14, NR-101 and NR-206, which are located in Portugal’s northern region. This study analyzed the collisions frequencies, average annual daily traffic (AADT) and geometric characteristics of 88 two-lane road segments through the analysis of the impact of different database structures in time and space. The selected segments were 200-m-long and did not cross through urbanized areas. Data regarding the annual traffic collision frequency and the AADT were available from 1999 to 2010. The GEE procedure was applied to ten distinctive databases formed by grouping the original data in time and space. The results show that the different observations within each road segment present mostly an exchangeable correlation structure type. This paper also analyses the impact of the sample size on the model’s capability of identifying the contributing factors to collision frequencies, therefore must work with segments homogeneous greatest possible. The major contributing factors identified for the twolane highways studied were the traffic volume (AADT), lane width, horizontal sinuosity, vertical sinuosity, density of access points, and density of pedestrian crossings. Acceptable CPM was identified for the highways considered, which estimated the total number of collisions for 400-m-long segments for a cumulative period of six years.