Application of Random Effects Nonlinear Model for Analyzing Motorized and Nonmotorized Traffic Safety Performance.

In this paper, we investigated the influence of the cross-sectional characteristics of bicycle lanes on motorized and nonmotorized traffic safety. The safety effects of bicycle lanes with different widths were assessed through the estimation of safety performance functions (SPFs) and crash modification factors (CMFs). Four different types of crash prediction models [i.e., generalized linear model (GLM), random effects GLM, generalized nonlinear model (GNM), and random effects GNM] were examined to develop more reliable SPFs. The results indicated that the random effects GNMs provided the most reliable estimates. The goodness of fit was higher for the GNMs than for the GLMs because of the nonlinear relationship between the width of the bicycle lanes and the crash rates. In addition, the random effects models indicated better performance than the GLMs and GNMs. The results indicated that the installation of bicycle lanes is an effective safety measure to reduce four different types of crashes in general. The results of the estimated CMFs using the random effects GNMs indicated that the safety effects of bicycle lanes had nonlinear variations based on different widths. In general, the installation of bicycle lanes with widths of 1.5-1.8 m (5-6 ft) was most effective for reducing motorized crashes, whereas bicycle lanes with widths of 1.8-2.1 m (6-7 ft) were more appropriate for reducing nonmotorized crashes. According to the findings of this study, specific guidance on minimum bicycle lane widths for various roadway characteristics, traffic flows, and roadway types based on empirical evidence regarding traffic safety can be provided. [ABSTRACT FROM AUTHOR]