Safety Evaluation of Continuous Green T Intersections

EXECUTIVE SUMMARY

This study used a propensity scores-potential outcomes framework to evaluate the safety performance of the continuous green T (CGT) intersection relative to a conventional signalized T intersection. Data from 30 CGT (treated) and 38 conventional signalized (untreated) intersections from Florida were used in the evaluation, as were 16 treated and 21 untreated sites from South Carolina. In the propensity scores-potential outcomes framework, a propensity scores model was estimated using a binary logistic regression model, where the dependent variable was codified as a binary variable based on the presence of the CGT or the conventional T signalized intersection form. The independent variables in the propensity scores model included safety-influencing features present at the intersections, including the average annual daily traffic on the major and minor street approaches, the posted speed limit, cross-sectional widths, and the type of intersection channelization. The propensity scores were then used to match treated (CGT) to untreated (conventional signalized) intersections, mimicking a randomized experiment. After matching, the potential outcomes were estimated using mixed effects negative binomial or Poisson count regression models (where possible) and weighted using negative binomial regression with robust standard errors otherwise. The expected total, fatal and injury, and target crash frequencies were used as the dependent variables in the count models, while the intersection safety-influencing variables were used as independent variables. In addition, an indicator variable was used in the potential outcomes model to assess the safety performance of the CGT relative to a conventional T signalized intersection.

The results showed that there was a small but statistically insignificant benefit associated with the CGT intersection relative to the conventional signalized T intersection. The crash modification factors (CMFs) associated with total crashes, fatal and injury crashes, and target crashes were 0.958 (p-value = 0.699, 95-percent confidence interval (CI) = 0.772–1.189), 0.846 (p-value = 0.211, 95-percent CI = 0.651–1.099), and 0.920 (p-value = 0.519, 95-percent CI = 0.714–1.185), respectively. Because the propensity scores-potential outcomes framework involves matching, some treated and untreated intersections in the database were not included in the analysis sample. For purposes of comparison, cross-sectional regression models using all available data were estimated, and the results were similar to the propensity scores-potential outcomes results. In these models, the CMFs associated with total, fatal and injury, and target crashes were 0.886 (p-value = 0.389), 0.844 (p-value = 0.230), and 0.808 (p-value = 0.187), respectively. The benefit-cost analysis confirmed that the CGT is a cost-effective intersection design alternative to the conventional T signalized intersection.