A multivariate random parameter crash frequency model for exploring the associations between shoulder rumble strips on different crash types

Abstract

A systemic approach to safety management has potential to be more cost effective than a traditional hot-spot approach in reducing crash frequency as it aims to identify sites with promise (SWiPs) by targeting certain crash types and select countermeasures accordingly, and therefore, effectively connects target crashes to their corresponding countermeasures. Although it seems to be promising, there are many challenges to evaluate the effects of countermeasures under the systemic approach, i.e. evaluate the effects of countermeasures by crash types. These challenges mainly include (1) low sample mean, (2) correlation among crash types, and (3) correlation among the effects of countermeasures on different crash types. A random parameter multivariate Poison lognormal (random parameter MVPLN) model is proposed. It was shown that the random parameter MVPLN model not only mitigates the three challenges, but also provides useful information than traditional crash frequency models. Shoulder rumble strips were found to be effective in reducing fix-object crashes, the primary crash type targeted by this countermeasure. The effectiveness of shoulder rumble strips on fixed-object crashes was found to be positively correlated with opposite-direction sideswipe and head-on crashes, meaning that the effects of shoulder rumble strips on fixed-object crashes would spillover to opposite-direction sideswipe and head-on crashes.