| description abstract | Ramp metering (RM) is a widely used active traffic management (ATM) system on freeways to enhance mobility. However, research on RM systems has predominantly focused on operational efficiency, with limited attention to their impact on traffic safety. In this paper, we develop short-term safety performance functions (SPFs) for RM at two peak periods, using microscopic traffic detector data and detailed RM operation data from three US states—Florida, California, and Wisconsin—to predict the total crashes for both freeway merge and on-ramp segments. The proposed short-term crash prediction has the potential to enhance accuracy and flexibility, providing deeper insights into the fluctuation of safety evaluation over time. Two Bayesian statistical methods, namely Poisson-lognormal (PLN) and negative binomial–Lindley (NB-L), are employed to develop the short-term SPFs. Independent variables such as traffic characteristics and specific geometric data (e.g., number of lanes, ramp configuration, and presence of weaving segments), along with the RM control strategy, are considered in the analysis. The proposed model shows that RM-implemented freeway segments had fewer crashes than non-RM segments. The traffic exposure and other variables for ramp and merge segments were significant. The significant factors vary between peak periods. Further, RM could potentially affect safety in weaving segment locations compared to other segment types. These results would aid practitioners, policymakers, and operators in identifying critical crash factors and assessing the safety effectiveness of ramp metering techniques. Ultimately, this research could pave the way for implementing appropriate safety interventions and advance safety evaluation research related to RM. | |
