Accident prediction models with and without trend: Application of the Generalized Estimating Equations (GEE) procedure

Accident prediction models (APMs) are very useful tools for estimating the expected number of accidents on entities such as intersections and road sections. These estimates are typically used in the identification of sites for possible safety treatment and in the evaluation of such treatments. An APM is, in essence, a mathematical equation that expresses the average accident frequency of a site as a function of traffic flow and other site characteristics. The reliability of an APM estimate is enhanced if the APM is based on data for as many years as possible especially if data for those same years are used in the safety analysis of a site. With many years of data, however, it is necessary to account for the year-to-year variation, or trend, in accident counts because of the influence of factors that change every year. To capture this variation, the count for each year is treated as a separate observation. Unfortunately, the disaggregation of the data in this manner creates a temporal correlation that presents difficulties for traditional model calibration procedures. The objective of this paper is to present an application of a generalized estimating equations (GEE) procedure to develop an APM that incorporates trend in accident data. Data for the application pertains to a sample of 4–legged signalized intersections in Toronto, Canada for the years 1990 to 1995. The GEE model incorporating the time trend is shown to be superior to models that do not accommodate trend and/or the temporal correlation in accident data. Key words: accident prediction models, temporal correlation, GEE, GLM, signalized intersection