Procedures for Estimating Accident Reductions on Two‐Lane Highways

Abstract

Accident prediction models or the before‐and‐after study approach is commonly used to estimate the reduction in number of accidents resulting from highway improvements. The highway improvement evaluation results will depend on the approach used, if the two approaches do not yield the same number of accidents reduced. This study computed the accident reductions on two‐lane rural highways in Illinois using: (1) Mathematical models for accident frequency prediction; and (2) a before‐and‐after study with control sites. Accident data for two years before and two years after the improvements were collected. One linear and three multiplicative form models were used to fit the accident data collected from 51 projects with a total length of 349 mi. The linear models were found to be as good as, and in some cases even better than, the multiplicative models in explaining the variation of accident frequency in both before and after conditions. The linear models indicated a stronger correlation between accident frequency, traffic volume, and roadway geometry than the multiplicative models for an ADT range of 500–5,000. This finding is contrary to that of Boyce et al. (1988) and Zegeer et al. (1987a), which recommended the multiplicative models over the linear models. The before‐and‐after study with control sites estimated accident reductions 1.40–2.06 times higher than those of the linear model.