Safety Effects for Widening Shoulder Width on Rural Multi-lane Roadways

1 This study assesses the safety effectiveness of widening shoulder widths by developing crash 2 modification factors (CMFs) and crash modification functions (CMFunctions). This study first 3 evaluates the CMFs for widening shoulder widths on rural multilane roadways using the 4 observational before-after with the empirical Bayes (EB) method to check the overall safety effects. 5 Secondly, the CMFs were calculated for each aggregated break points based on different roadway 6 characteristics such as the original shoulder widths of treated sites in the before period and the 7 actual widened widths. Lastly, the CMFunctions were developed using multiple linear regression 8 and multivariate adaptive regression splines (MARS) models to determine the variation of CMFs. 9 The MARS is one of the promising data mining techniques due to its ability to consider the 10 interaction impact of more than one variable and nonlinearity of predictors simultaneously. Crash 11 and roadway characteristics data were collected for roadways in Florida, and the Florida-specific 12 full SPFs for different crash types and severity levels were developed, and used for the 13 observational before-after with EB estimation. 14 The results indicated that widening shoulder widths resulted in a crash reduction of 12 percent for 15 total crashes (all severities), 18 percent for injury crashes, and 21 percent for severe crashes on 16 rural multilane roadways for all types of crashes. For run-off roadway (ROR) crashes, the overall 17 crash reduction of the treatment was 25 percent for total crashes, 28 percent for injury crashes, and 18 31 percent for severe crashes. Moreover, the results showed that the safety effects vary across the 19 sites with different roadway characteristics. In particular, the original shoulder widths of treated 20 sites in the before period and actual widened width are significant parameters that affect the 21 variation of CMFs. Moreover, it was found that the MARS models show more reliable estimates 22 than normal regression, if the variation of CMFs with specific parameters has nonlinear 23 relationships and interaction effects. The proposed CMFunctions provide general insights into 24 roadway design and selection of sites for widening shoulder width for reducing crashes. 25