**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

**Suggested Citation:**"Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data." National Academies of Sciences, Engineering, and Medicine. 2023.

*Pedestrian and Bicycle Safety Performance Functions*. Washington, DC: The National Academies Press. doi: 10.17226/27294.

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

100 This section of the report describes the development of pedestrian and bicycle SPFs and incor- porates pedestrian and bicycle exposure data into the modeling approach for potential incorpo- ration in HSM2. SPFs were developed to predict pedestrian and bicycle crashes for urban and suburban roadway segments and intersections. The SPFs were developed for potential consider- ation in the HSM predictive methods chapter for urban and suburban arterials (i.e., HSM Chap- terÂ 12) and/or for potential use in the PartÂ B network screening chapter. Urban and suburban roadway segments and intersections were the focus because: â¢ Most pedestrian and bicycle activity occurs within urban and suburban environments. â¢ The majority of fatal crashes involving pedestrians and bicyclists occur in urban and suburban areas. â¢ Pedestrian and bicyclist exposure data are primarily collected in urban areas. â¢ Urban and suburban roadway segments and intersections were deemed the highest priority for model development based on the survey results. SectionÂ 3.1 describes the site selection and data collection process for developing pedes- trian and bicycle SPFs, incorporating pedestrian and bicyclist exposure data into the modeling approach. SectionÂ 3.2 presents several descriptive statistics of the databases used for model development. SectionÂ 3.3 presents the statistical approach to SPF development. SectionÂ 3.4 presents the analysis results. SectionÂ 3.5 addresses the need for calibration, and SectionÂ 3.6 presents a summary and recommendations for incorporating the new SPFs into HSM2. 3.1 Site Selection and Data Collection During Phase I of the research, information was gathered on potential data sources to develop pedestrian and bicycle SPFs incorporating pedestrian and bicyclist exposure into the analysis approach. The types of data needed for model development included: â¢ Site characteristic information including pedestrian and bicycle infrastructure data. â¢ Motor vehicle volume data. â¢ Pedestrian and bicycle exposure data. â¢ Crash data. Of particular interest were potential data sources of pedestrian and bicycle exposure as the other types of data are often more readily available. Information was gathered from around the country from several metropolitan areas considered to have the most comprehensive pedes- trian and bicycle count programs. Information on other nontraditional data sources that can be used to estimate pedestrian and bicycle exposure was also evaluated (e.g., Strava data). S E C T I O N 3 Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 101Â Â Based on the results of the literature review, survey of practice, and discussions with selected agencies, it was decided that the best approach for incorporating pedestrian and bicycle exposure in the SPF development was to obtain count data for pedestrians and bicyclists and then develop direct demand models to estimate pedestrian and bicycle volumes for the study sites. Priority was given to searching for potential count data from cities, MPOs, and local transportation agencies compared to potential data sources from state transportation agencies as counts collected by state transportation agencies were considered too spread out geographically for efficient data collection purposes. The following cities were considered to have the most comprehensive pedestrian and/or bicycle count programs in the United States and were considered as potential locations to collect data for model development: â¢ Minneapolis, Minnesota, â¢ Philadelphia, Pennsylvania, â¢ Seattle, Washington, â¢ St. Paul, Minnesota, and â¢ Boulder, Colorado. This assessment was based on elements such as: â¢ The number of sites where counts are conducted. â¢ The duration of counts (e.g., short-term versus long-term counts). â¢ Availability of scaling factors or a valid approach for scaling short-term counts to 24-hour counts. â¢ Counts available for both pedestrians and bicyclists. â¢ Maturity of the pedestrian and bicycle count program (i.e., the number of years counts have been collected). In addition to the five cities listed above, the cities of Charlotte, North Carolina; San Diego, California; and Tucson, Arizona were considered potential locations to collect data for model development. After gathering information on each of the pedestrian and bicycle count programs for these respective cities, Minneapolis and Philadelphia were selected as the two urban/suburban locations to collect detailed data for model development. Having selected Minneapolis and Philadelphia as the locations for data collection, electronic files of roadway inventory data (i.e., site characteristics such as the number of lanes, geometric design features, and traffic control elements), traffic volume data, crash data, and pedestrian and bicyclist exposure data were requested from the local transportation agencies (i.e., the Minneapolis Public Works Department and the DVRPC as the federally designated MPOs for the nine-county region near Philadelphia). Information about installation dates of bike lanes, buffered bike lanes, etc., was also requested for consideration during data collection as well as model development to define the appropriate time period of a given site for analysis purposes. In addition, census data, socioeconomic data, and demographic data were either gathered from online sources or obtained directly from the local agency. Based on the availability of roadway inventory data, traffic volume data, crash data, and the ability to estimate pedestrian and bicycle volumes for a given site, specific roadway segments and intersections in Minneapolis and Philadelphia were selected as potential sites for data collection. Google Earth was used to collect detailed site characteristics for each of the potential study locations. To reduce potential errors during data collection and to streamline data entry, a data collection tool was created using Visual Basic for Applications. The items in the data collection tool dynamically changed as information was input by the data collector. For example, if the user indicated the presence of a median, a field for entering median width appeared.

102 Pedestrian and Bicycle Safety Performance Functions Additionally, information on the number of bus/transit stops, schools, and alcohol sales estab- lishments located in proximity to the study locations was gathered electronically. FigureÂ 8 is a screen capture of the data collection tool for roadway segments and illustrates the data elements that were manually collected for roadway segments. FigureÂ 9 illustrates, conceptually, the cross-sectional elements that were collected in each direction of travel. TableÂ 32 sum- marizes the site characteristic data elements that were collected for roadway segments, either manually or electronically, for use in model development. FigureÂ 10 and FigureÂ 11 are screen captures of the corresponding data collection tool for intersections. For intersections, data were collected for the intersection as a whole and by leg. For each intersection leg, data were gathered in both the inbound and outbound directions. TableÂ 33 summarizes the site characteristic data elements collected for intersections, either manually or electronically, for use in model development. FigureÂ 8. Illustration of data collection tool to manually collect roadway segment inventory data.

Shared Path Shared Path Buffer Sidewalk Sidewalk Buffer Outside Shoulder Travel Lanes Ln1.. Ln2.. Ln3.. Ln4..Ln5 Median Inside Shoulder Left Side of Road Inside Shoulder Travel Lanes Ln5.. Ln4.. Ln3.. Ln 2.. Ln1 Outside Shoulder Sidewalk Buffer Sidewalk Shared Path Buffer Shared Path Right Side of Road Bike Lane Buffer Bike Lane Parking Lane Bike Lane Buffer Bike Lane Parking Lane Traveled Way Roadway Separation from Traveled Way to Sidewalk (Left) Separation from Traveled Way to Sidewalk (Right) FigureÂ 9. Illustration of cross-sectional elements collected along roadway segments.

104 Pedestrian and Bicycle Safety Performance Functions Variable/Parameter Definition/Description Range or Permitted Values General Roadway Attributes Route identifier Unique identification number for roadway segment Character value Two-way versus one-way operation Indicates if the roadway operates with two-way traffic or one-way traffic One-way, two-way Presence of lighting Indicates if overhead lighting is present along the roadway Yes, no Presence of traffic calming Indicates if any traffic calming features (e.g., speed humps, chicanes) are present along the roadway Yes, no Median width Measured from outside of adjacent motor vehicle traffic lane in the one direction to the outside of adjacent motor vehicle lane in the opposing direction Values in feet Median type Type of median separating opposing directions of travel None, painted, depressed, raised, two-way left-turn lane (TWLTL) Presence of a midblock crossing Indicates the presence of a marked midblock crossing Yes, no Presence of advance yield/stop lines at the midblock crossing Indicates the presence of advance yield/stop lines at the midblock crossing Yes, no Type of traffic control at the midblock crossing Indicates the type of traffic control at the midblock crossing None, flashing beacon, signalized Length of midblock crossing Measured curb-to-curb minus the width of the refuge island (if present) Values in feet Bus/transit stops Indicates the number of bus/transit stopslocated along the roadway segment Integer value (0,1,..) Bus/transit stops (collected electronically) Indicates the number of bus/transit stops measured 1,000 ft from the center of the roadway segment Integer value (0,1,..) Schools (collected electronically) Indicates the number of schools measured 1,000 ft from the center of the roadway segment Integer value (0,1,..) Alcohol sales establishments (collected electronically) Indicates the number of alcohol sales establishments measured 1,000 ft from the center of the roadway segment Integer value (0,1,..) Begin year First year when the site is considered appropriate for use in the analysis 2006â2018 End year Last year when the site is considered appropriate for use in the analysis 2006â2018 Attributes in Each Direction of Travel Presence of a shared-use path Indicates the presence of a shared-use path adjacent to the direction of travel Yes, no Width of shared-use path Measured width of the shared-use path Values in feet Width of shared-use path buffer Measured width from the edge of shared- use path closest to the roadway to the curb (or outside shoulder) Values in feet Presence of a sidewalk Indicates the presence of a sidewalk adjacent to the direction of travel Yes, no Width of sidewalk Measured width of the sidewalk Values in feet Width of sidewalk buffer Measured width from the edge of sidewalk closest to the roadway to the curb (or outside shoulder) Values in feet Presence of sidewalk protection Indicates the presence of sidewalk protection (e.g., trees, barriers) separating the sidewalk from the roadway Yes, no Width of outside shoulder Measured width of outside shoulder Values in feet Presence of marked parking lane Indicates the presence of a marked parking lane in the direction of travel Yes, no Width of marked parking lane Width of marked parking lane Values in feet Type of bicycle facility Indicates the presence of a marked bike lane, buffered bike lane, protected bike lane, or travel lane with shared-lane marking in the direction of travel None, bike lane, buffered bike lane, protected bike lane, travel lane with shared-lane marking Width of marked bike lane Width of marked bike lane Values in feet Width of buffer Width of buffer associated with a buffered bike lane Values in feet TableÂ 32. Site characteristic variables collected for roadway segments.

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 105Â Â Variable/Parameter Definition/Description Range or Permitted Values One-way versus two operation along the bike lane Indicates whether bicycles travel in one direction or two directions along the bike lane One-way, two-way Presence of colored pavement along the bicycle facility Indicates the presence of colored pavement (e.g., green) in the marked bike lane (or buffered bike lane) Yes, no Number of travel lanes Indicates the number of motor vehicle travel lanes in the direction of travel 0, 1, 2, 3 Width of travel lanes Width of each motor vehicle travel lane in the direction of travel Values in feet Width of inside shoulder Measured width of inside shoulder Values in feet TableÂ 32. (Continued). FigureÂ 10. Illustration of data collection tool to manually collect intersection inventory data (general intersection elements).

106 Pedestrian and Bicycle Safety Performance Functions FigureÂ 11. Illustration of data collection tool to manually collect intersection inventory data (approach leg). For model development, priority was given to collecting data for roadway segment and inter- section types included in the urban and suburban arterials chapter of the first edition of the HSM ChapterÂ 12 and a subset of those planned for inclusion in the urban and suburban arterials chapter in the second edition of the HSM, namely: â¢ Roadway segments â Two-lane undivided arterials (2U). â Three-lane arterials including a center two-way left-turn lane (TWLTL) (3T). â Four-lane undivided arterials (4U). â Four-lane divided arterials (4D). â Five-lane arterials including a center TWLTL (5T). â Six-lane undivided arterials (6U). â Six-lane divided arterials (6D). â Seven-lane arterials including a center TWLTL (7T). â Eight-lane divided arterials (8D). â Two-lane one-way arterials (2OW). â Three-lane one-way arterials (3OW). â Four-lane one-way arterials (4OW).

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 107Â Â Variable/Parameter Definition/Description Range or Permitted Values General Intersection Attributes Intersection configuration (i.e., number of legs and type of traffic control) Indicates the number of legs and type of traffic control 3ST, 4ST, 3SG, 4SG Two-way versus one-way operation Indicates whether it is an intersection of two two-way streets or an intersection of a one- way street and a two-way street (only applicable for 4SG) 4SG 2x2 4SG 1x2 Presence of flashing beacons Indicates if overhead flashing beacons are present at the intersection proper Yes, no Presence of intersection lighting Indicates if overhead lighting is present at the intersection Yes, no Bus/transit stops (collected electronically) Indicates the number of bus/transit stops within 1,000 ft of the center of the intersection Integer value (0,1,..) Schools (collected electronically) Indicates the number of schools within 1,000 ft of the center of the intersection Integer value (0,1,..) Alcohol sales establishments (collected electronically) Indicates the number of alcohol sales establishments within 1,000 ft of the center of the intersection Integer value (0,1,..) Begin year First year when the site is considered appropriate for use in the analysis 2006â2018 End year Last year when the site is considered appropriate for use in the analysis 2006â2018 Approach Specific Attributes Route identifier Unique identification number for approach Character value Location at intersection Side/quadrant of the intersection on which the approach is located N, S, E, W, NE, NW, SE, SW Number of through lanes Includes dedicated through lanes and any lanes with shared movements. On the minor approach of a 3-leg intersection, if there is only one lane, then it should be classified as a through lane 0, 1, 2, 3 Width of through lanes (inbound/outbound) Width of dedicated through lanes and any lanes with shared movements Values in feet Presence/number of left-turn lanes The number of lanes in which only a left-turn movement can be made 0, 1, 2, 3 Width of left-turn lane Width of dedicated left-turn lane (i.e., lane in which only a left-turn movement can be made) Values in feet Left-turn operation Option that best describes the operation of the left turn (âDoes not applyâ is used where there is no left-turn movement) Does not apply, free flow, signal- permitted only, signal-protected only, signal-protected-permitted, stop control, yield control Presence/number of right-turn lanes The number of lanes in which only a right-turn movement can be made 0, 1, 2, 3 Width of right-turn lane Width of dedicated right-turn lane (i.e., lane in which only a right-turn movement can be made) Values in feet Presence of right-turn channelization Presence of right-turn channelization on the intersection approach Yes, no Right-turn operation Option that best describes the operation of the right turn (âDoes not applyâ is used where there is no right-turn movement) Does not apply, free flow, signal- right-turn-on-red (RTOR) permitted, signal-RTOR prohibited, stop control, yield control Width of inside/outside shoulder Measured shoulder width Values in feet Median width Measured from outside of adjacent motor vehicle traffic lane in the approaching direction to the outside of adjacent motor vehicle lane in opposing direction Values in feet Median type Type of median separating opposing directions of travel None, painted, raised Presence of inbound/outbound bicycle facility Indicates the presence of a marked bike lane, buffered bike lane, or protected bike lane parallel to the intersection approach or the presence of shared-lane marking None, bike lane, buffered bike lane, protected bike lane, travel lane with shared-lane marking Width of marked bike lane (inbound/outbound) Width of marked bike lane Values in feet TableÂ 33. Site characteristic variables collected for intersections. (continued on next page)

108 Pedestrian and Bicycle Safety Performance Functions Variable/Parameter Definition/Description Range or Permitted Values Width of buffer (inbound/outbound) Width of buffer associated with a buffered bike lane or protected bike lane Values in feet Presence of colored pavement in inbound/outbound bicycle facility Indicates the presence of colored pavement (e.g., green) in the marked bike lane, buffered bike lane, or protected bike lane Yes, no Width of marked parking lanes (inbound/outbound) Width of marked parking lane Values in feet Presence of crosswalk Indicates the presence of a crosswalk perpendicular to the intersection approach Yes, no Crosswalk control type Indicates the control type present at the crosswalk No crosswalk present, no control present, pedestrian crossing signal, pedestrian countdown timer Presence of shared-use path crossing Indicates the presence of a shared-use path crossing Yes, no Presence of advance yield/stop lines Indicates the presence of advance yield/stop lines Yes, no Crossing distance Indicates crossing distance measured curb-to-curb Values in feet Crossing distance (accounting for median/refuge island and right-turn channelizing island) Indicates crossing distance measured curb- to-curb (minus the width of the median/refuge island and/or right-turn channelizing island) Values in feet NOTE: 3ST = three-leg stop control intersection. 4ST = four-leg stop control intersection. 3SG = three-leg signal control intersection. 4SG = four-leg signal control intersection. 4SG 2Ã2 = four-leg signal control intersections with two-way/two-way operations. 4SG 1Ã2 = four-leg signal control intersections with one-way/two-way operations combined. TableÂ 33. (Continued). â¢ Intersections â Unsignalized three-leg intersections [stop control on minor-road approaches with two-way/ two-way (2Ã2), one-way/two-way (1Ã2), and one-way/one-way (1Ã1) operations] (3ST). â Unsignalized three-leg intersections on high-speed arterials (stop control on minor-road approaches) (3ST-HS). â Unsignalized three-leg intersections (all-way stop control) (3aST). â Signalized three-leg intersections [with two-way/two-way (2Ã2), one-way/two-way (1Ã2), and one-way/one-way (1Ã1) operations] (3SG). â Signalized three-leg intersections on high-speed arterials (3SG-HS). â Unsignalized four-leg intersections (all-way stop control) (4aST). â Unsignalized four-leg intersections [stop control on minor-road approaches with two-way/ two-way (2Ã2), one-way/two-way (1Ã2), and one-way/one-way (1Ã1) operations] (4ST). â Unsignalized four-leg intersections on high-speed arterials (stop control on minor-road approaches) (4ST-HS). â Signalized four-leg intersections [with two-way/two-way (2Ã2), one-way/two-way (1Ã2), and one-way/one-way (1Ã1) operations] (4SG). â Signalized four-leg intersections on high-speed arterials (4SG-HS). During detailed data collection, to the extent possible, the research team reviewed historical aerial images to determine whether a site had recently been reconstructed or improved and/or to validate the accuracy of the information provided to the research team regarding installation dates of bicycle facilities (e.g., bike lanes and buffered bike lanes) to determine which years of data should be used in model development. While assembling the final databases for model development, the research team performed quality control checks of the data. This included inputting missing years of AADT by interpolation or extrapolation where necessary; checking

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 109Â Â for outliers on all variables collected; cross-referencing selected variables for consistency of terminology/definition of terms; checking the data for completeness; and other checks as deemed necessary for each specific dataset. 3.2 Descriptive Statistics of Final Databases Data for approximately 253Â miles of urban and suburban roads and 271 urban and suburban intersections were available for the development of pedestrian and bicycle SPFs, incorpo- rating pedestrian and bicyclist exposure data into the modeling approach. SectionÂ 3.2.1 presents several descriptive statistics of the final databases for roadway segments, and SectionÂ 3.2.2 presents several descriptive statistics of the final databases for intersections. 3.2.1 Final Databases: Roadway Segments TableÂ 34 presents the number of sites, miles, and total mile-years by roadway segment type for which site characteristics, motor vehicle volumes, pedestrian volumes, bicycle volumes, and crash data were available for model development. Only those road types for which a sufficient amount of data were collected for potential model development are provided in the table. For Minneapolis, the dataset included up to 13Â years of data from 2006 through 2018. For Philadelphia, the dataset included up to 6Â years of data from 2013 through 2018. 3.2.1.1 Roadway Segments: Site Characteristics This section provides several descriptive statistics of the site characteristics and inventory data collected for roadway segments. TableÂ 35 provides information on the types of bicycle facilities present on the roads, by road type. Bicycle facility types were categorized as follows: â¢ Bike lane (i.e., conventional bike lane). â¢ Buffered bike lane (i.e., a conventional bike lane with a designated buffer space separating the bike lane from the adjacent motor vehicle lane). â¢ Travel lane with shared-lane marking (i.e., a motor vehicle travel lane with pavement markings designating a portion of the roadway or lane for preferential use by bicyclists). â¢ Protected bike lane (i.e., an exclusive bike facility physically separated from motor vehicle travel lanes, parking lanes, and sidewalks). Road type Minneapolis Philadelphia Total Number of Sites Miles Total Mile- Years Number of Sites Miles Total Mile- Years Number of Sites Miles Total Mile- Years 2U 651 65.32 568.87 706 48.76 278.54 1,357 114.08 847.41 4U 290 23.91 268.44 60 3.91 23.48 350 27.82 291.92 4D 44 4.98 56.27 48 6.19 34.33 92 11.17 90.60 1-lane, one-way 48 4.33 44.29 167 7.20 43.11 215 11.53 87.40 2-lanes, one-way 115 10.74 78.25 137 9.48 44.47 252 20.22 122.72 3-lanes, one-way 151 11.93 106.15 32 2.04 11.95 183 13.97 118.10 NOTE: 2U = two-lane undivided; 4U = four-lane undivided; 4D = four-lane divided. TableÂ 34. Number of sites, miles, and total mile-years by road type.

110 Pedestrian and Bicycle Safety Performance Functions â¢ Shared-use path (i.e., a bikeway outside the traveled way and physically separated from motorized vehicular traffic by an open space or barrier and either within the highway right-of-way or within an independent alignment. Shared-use paths may also be used by pedestrians.). TableÂ 35 indicates whether the bicycle facility was located on one side (i.e., one direction of travel) or both sides (i.e., both directions of travel) of the roadway. In a few instances, the bicycle facility type on one side of the road differed from the bicycle facility type on the opposite side of the road. TableÂ 36 indicates the number of roadway segments with lighting by road type. With the roadway segments being in urban and suburban areas, the majority of the sites had lighting. Road Type Bicycle Facility Type Number of Sites Miles Total Mile- Years 2U Bike lane (one side) 14 1.30 7.97 Bike lane (both sides) 550 40.83 236.38 Bike lane (one side) â buffered bike lane (one side) 7 0.52 1.72 Bike lane (one side) â roadway with shared-lane marking (one side) 4 0.31 1.37 Buffered bike lane (one side) 2 0.16 0.63 Buffered bike lane (both sides) 24 2.20 9.51 Protected bike lane (both sides) 1 0.80 0.16 Shared-use path (one side) 110 16.47 141.49 Shared-use path (both sides) 1 0.19 2.48 Roadway with shared-lane markings (one side) 1 0.13 1.00 Roadway with shared-lane markings (both sides) 91 6.11 26.65 No bicycle facility 552 45.78 418.04 4U Bike lane (both sides) 12 1.11 8.40 Bike lane (one side) â buffered bike lane (one side) 2 0.22 1.09 Bike lane (one side) â roadway with shared-lane marking (one side) 2 0.20 1.22 Buffered bike lane (both sides) 8 0.62 2.00 Shared-use path (one side) 1 0.12 0.70 Roadway with shared-lane marking (one side) 3 0.19 1.14 Roadway with shared-lane marking (both sides) 13 0.83 6.87 No bicycle facility 309 24.54 270.50 4D Bike lane (one side) 2 0.12 0.74 Bike lane (both sides) 21 2.83 17.73 Buffered bike lane (both sides) 2 0.69 2.08 Shared-use path (one side) 5 0.65 5.89 Travel lane with shared-lane marking (both sides) 6 0.61 4.27 No bicycle facility 56 6.26 59.89 1-lane one-way Bike lane (one side) 36 1.51 8.98 Travel lane with shared-lane marking 6 0.27 1.61 Shared-use path (one side) 14 1.71 15.35 No bicycle facility 159 8.04 61.47 2-lanes one-way Bike lane (one side) 29 1.81 11.80 Buffered bike lane (one side) 93 9.43 42.83 Protected bike lane 5 0.58 1.15 Travel lane with shared-lane marking 25 1.47 9.40 No bicycle facility 100 6.92 57.55 3-lanes one-way Bike lane (one side) 92 7.35 59.15 Buffered bike lane (one side) 21 1.64 7.29 No bicycle facility 70 4.98 51.67 NOTE: 2U = two-lane undivided; 4U = four-lane undivided; 4D = four-lane divided. TableÂ 35. Bicycle facility types by road type.

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 111Â Â TableÂ 37 provides information on the posted speed limit for sites by road type. The average speed limit for all roadway types was in the range of 25 to 30Â mph. TableÂ 38 provides information on median type and width, by road type. For four-lane divided roads, the majority of the study sites had raised medians. TableÂ 39 provides information on the presence of driveways along the roadway segments by road type. For the analysis, the number of driveways present along the roadway was repre- sented on a per-mile basis (i.e., number of driveways/m). For all roadway types considered, there were at least one or more study sites with no driveways present. TableÂ 40 provides information on the presence of bus stops along the roadway segments by road type. This information was collected manually while reviewing the sites. For the analysis, Road Type Total Number of Sites Presence of Lighting Yes No 2U 1,357 1,266 (93.3%) 91 (6.7%) 4U 350 335 (95.7%) 15 (4.3%) 4D 92 88 (95.7%) 4 (4.3%) 1-lane one-way 215 211 (98.1%) 4 (1.9%) 2-lanes one-way 252 244 (96.8%) 8 (3.2%) 3-lanes one-way 183 177 (96.7%) 6 (3.3%) NOTE: 2U = two-lane undivided; 4U = four-lane undivided; 4D = four-lane divided. TableÂ 36. Presence of lighting by road type. Road Type Total Number of Sites Speed Limit (mph) Mean Min. Max. Median Std. Dev. 2U 1,357 27 25 35 25 2.61 4U 350 30 25 40 30 2.34 4D 92 29 25 45 30 4.00 1-lane one-way 215 25 15 25 25 0.68 2-lanes one-way 252 29 25 45 30 3.91 3-lanes one-way 183 29 25 35 30 2.36 NOTE: 2U = two-lane undivided; 4U = four-lane undivided; 4D = four-lane divided. TableÂ 37. Posted speed limit by road type. Road Type Total Number of Sites Median Type Median Width (ft) None Depressed Raised Painted Mean Min. Max. Median Std.Dev. 2U 1,357 1,357 0 0 0 0 0 0 0 0 4U 350 350 0 0 0 0 0 0 0 0 4D 92 0 1 82 9 13.03 4.00 49.32 13.00 8.10 1-lane one-way 215 215 0 0 0 0 0 0 0 0 2-lanes one-way 252 252 0 0 0 0 0 0 0 0 3-lanes one-way 183 183 0 0 0 0 0 0 0 0 NOTE: 2U = two-lane undivided; 4U = four-lane undivided; 4D = four-lane divided. TableÂ 38. Median type and width by road type.

112 Pedestrian and Bicycle Safety Performance Functions the number of bus stops present along the roadway was represented on a per-mile basis (i.e., number of bus stops/mi). For all roadway types considered, there were at least one or more study sites with no bus stops present. TableÂ 41 provides information on the number of bus/transit stops within 1,000Â ft of the center of the roadway segment. This information was gathered from electronic data available through online portals. TableÂ 42 provides information on the number of schools within 1,000Â ft of the center of the roadway segment. This information was gathered from electronic data available through online portals. Road Type Total Number of Sites Driveway Density (number of driveways/mi) Mean Min. Max. Median Std. Dev. 2U 1,357 27.30 0.00 272.72 19.24 31.66 4U 350 26.32 0.00 277.35 25.30 26.64 4D 92 23.81 0.00 204.11 14.06 35.81 1-lane one-way 215 13.02 0.00 73.86 0.00 19.22 2-lanes one-way 252 25.01 0.00 201.07 17.49 29.89 3-lanes one-way 183 26.53 0.00 103.06 23.45 23.80 NOTE: 2U = two-lane undivided; 4U = four-lane undivided; 4D = four-lane divided. TableÂ 39. Driveway density by road type. Road Type Total Number of Sites Bus Stop Density (number of bus stops/mi) Mean Min. Max. Median Std. Dev. 2U 1,357 5.36 0.00 160.93 0.00 10.81 4U 350 2.94 0.00 48.92 0.00 7.84 4D 92 5.02 0.00 31.98 0.00 9.13 1-lane one-way 215 2.99 0.00 41.05 0.00 8.39 2-lanes one-way 252 3.53 0.00 34.16 0.00 7.13 3-lanes one-way 183 1.24 0.00 30.59 0.00 4.53 NOTE: 2U = two-lane undivided; 4U = four-lane undivided; 4D = four-lane divided. TableÂ 40. Bus stop density by road type. Road Type Total Number of Sites Number of Bus/Transit Stops Within 1,000 ft of the Roadway Segment 0 1 or 2 3 or 4 5 to 10 11 or more 2U 1,357 45 23 56 495 738 4U 350 0 3 14 104 229 4D 92 3 3 10 32 44 1-lane one-way 215 9 6 13 70 117 2-lanes one-way 252 0 0 3 46 203 3-lanes one-way 183 5 1 3 17 157 NOTE: 2U = two-lane undivided; 4U = four-lane undivided; 4D = four-lane divided. TableÂ 41. Number of bus/transit stops within 1,000Â ft of the center of the roadway segment by road type.

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 113Â Â TableÂ 43 provides information on the number of alcohol sales establishments within 1,000Â ft of the center of the roadway segment. This information was gathered from electronic data available through online portals. TableÂ 44 provides information on travel lane widths for the roadway segments by road type. Lane 1 is the right-most lane in the direction of travel. Lane 2 is to the left of Lane 1, and Lane 3 is to the left of Lane 2. TableÂ 44 provides lane width information for the individual lanes and all lanes combined. TableÂ 45 provides information on total travel lane width by road type. The total travel lane width does not include shoulder widths, parking lane widths, or bike lane widths. For two- directional roadways, the data reflect combined lane widths for both directions of travel but do not include the width of the median. TableÂ 46 provides information on conventional bike lanes, including placement information and lane widths by road type. For two-lane, two-way roads, four-lane undivided roads, and four-lane divided roads, if a bike lane was present at the study site, in most cases a bike lane was present in both directions of travel. For one-way roads, if a bike lane was present, the bike lane was located only on one side of the road in the direction of travel (i.e., no contraflow bike lanes were included in the analysis). TableÂ 47 provides information on buffered bike lanes, including placement information, the type of buffer or separation, buffer width, and bike lane width by road type. For two-lane, two-way roads, four-lane undivided roads, and four-lane divided roads, if a buffered bike lane was present Road Type Total Number of Sites Number of Schools Within 1,000 ft of the Roadway Segment 0 1 2 or more 2U 1,357 725 366 266 4U 350 172 83 95 4D 92 66 10 16 1-lane one-way 215 118 73 24 2-lanes one-way 252 108 65 79 3-lanes one-way 183 62 43 78 NOTE: 2U = two-lane undivided; 4U = four-lane undivided; 4D = four-lane divided. TableÂ 42. Number of schools within 1,000Â ft of the center of the roadway segment by road type. Road Type Total Number of Sites Number of Alcohol Sales Establishments Within 1,000 ft of the Roadway Segment 0 1 to 9 10 or more 2U 1,357 481 810 66 4U 350 71 205 74 4D 92 29 61 2 1-lane one-way 215 50 152 13 2-lanes one-way 252 58 125 69 3-lanes one-way 183 25 81 77 NOTE: 2U = two-lane undivided; 4U = four-lane undivided; 4D = four-lane divided. TableÂ 43. Number of alcohol sales establishments within 1,000Â ft of the center of the roadway segment by road type.

114 Pedestrian and Bicycle Safety Performance Functions Road Type Total Number of Sites Lane Width (ft) Mean Min. Max. Median Std. Dev. Lane 1 2U 1,357 13.36 7.83 29.70 11.41 4.49 4U 350 12.55 7.39 23.02 11.86 2.69 4D 92 11.92 9.08 25.19 11.59 1.97 1-lane one-way 215 15.13 9.04 27.00 12.03 5.87 2-lanes one-way 252 11.94 8.10 25.35 11.07 3.37 3-lanes one-way 183 13.04 9.00 24.14 11.30 3.46 Lane 2 2U N/A N/A N/A N/A N/A N/A 4U 350 11.13 8.96 19.63 11.16 1.10 4D 92 11.13 9.00 16.97 11.00 1.31 1-lane one-way N/A N/A N/A N/A N/A N/A 2-lanes one-way 252 11.77 8.00 26.80 11.01 3.27 3-lanes one-way 183 11.35 9.80 12.90 11.30 0.71 Lane 3 2U N/A N/A N/A N/A N/A N/A 4U N/A N/A N/A N/A N/A N/A 4D N/A N/A N/A N/A N/A N/A 1-lane one-way N/A N/A N/A N/A N/A N/A 2-lanes one-way N/A N/A N/A N/A N/A N/A 3-lanes one-way 183 14.26 7.00 24.42 12.00 3.95 All Lanes Combined 2U 1,357 13.36 7.83 29.70 11.41 4.49 4U 350 11.84 7.39 23.02 11.45 2.18 4D 92 11.52 9.00 25.19 11.26 1.72 1-lane one-way 215 15.13 9.04 27.00 12.03 5.87 2-lanes one-way 252 11.85 8.00 26.80 11.01 3.32 3-lanes one-way 183 12.88 7.00 24.42 11.49 3.28 NOTE: 2U = two-lane undivided; 4U = four-lane undivided; 4D = four-lane divided. TableÂ 44. Lane width by road type. Road Type Total Number of Segments Total Travel Lane Width (all lanes considered) (ft) Mean Min. Max. Median Std. Dev. 2U 1,357 26.53 16.77 57.43 22.97 8.66 4U 350 47.26 25.77 63.45 46.29 5.88 4D 92 45.98 37.92 65.56 45.67 5.43 1-lane one-way 215 15.13 9.04 27.00 12.03 5.87 2-lanes one-way 252 23.30 16.14 52.15 22.50 6.07 3-lanes one-way 183 38.03 25.93 56.31 38.53 6.27 NOTE: 2U = two-lane undivided; 4U = four-lane undivided; 4D = four-lane divided. TableÂ 45. Total travel lane width by road type.

Road Type Total Number of Segments Conventional Bike Lane Placement of Bike Lane Bike Lane Width (ft) None One Dir Both Dir Mixed Facilitya Other Bike Facilityb Mean Min. Max. Median Std. Dev. 2U 1,357 552 14 550 11 230 4.73 3.00 9.28 4.78 0.92 4U 350 309 0 12 4 25 5.81 3.00 11.00 5.39 1.62 4D 92 56 2 21 0 13 4.65 4.00 6.18 5.00 0.70 1-lane one-way 215 159 36 0 0 20 4.46 4.00 6.00 4.00 0.56 2-lanes one-way 252 100 29 0 0 123 5.22 4.00 7.16 5.00 1.05 3-lanes one-way 183 70 92 0 0 21 5.42 3.89 8.50 5.38 0.82 NOTE: 2U = two-lane undivided; 4U = four-lane undivided; 4D = four-lane divided; Dir = direction(s). aMixed Facility: A bike lane is present on the roadway in one direction of travel, and a different type of bicycle facility (i.e., buffered bike lane, travel lane with shared-lane marking, protected bike lane, or shared-use path) is present in the opposite direction. bOther Bike Facility: A bicycle facility other than a conventional bike lane (i.e., buffered bike lane, travel lane with shared-lane marking, protected bike lane, or shared-use path) is present on the roadway in one or both directions of travel. TableÂ 46. Conventional bike lanes by road type. Road Type Total Number of Segments Buffered Bike Lane Placement of Bike Lane Type of Buffer/Separation Buffer Width (ft) Bike Lane Width (ft) None One Dir Both Dir Mixed Facilitya Other Bike Facilityb Painted Raised Mean Min. Max. Median Std. Dev. Mean Min. Max. Median Std. Dev. 2U 1,357 552 2 24 7 772 57 0 2.65 1.10 7.00 2.38 0.97 5.91 3.90 7.24 6.03 0.70 4U 350 309 0 8 2 31 18 0 2.22 1.49 3.01 2.30 0.46 5.89 3.41 7.50 6.03 0.99 4D 92 56 0 2 0 34 4 0 4.25 4.00 5.00 4.00 0.50 4.75 4.00 6.00 4.50 0.96 215 159 0 0 0 56 N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A 252 100 93 0 0 59 93 0 3.84 2.00 5.88 4.14 1.02 5.47 4.00 7.81 5.56 1.03 1-lane one-way 2-lanes one-way 3-lanes one-way 183 70 21 0 0 92 21 0 2.34 1.15 4.37 2.03 1.15 4.81 4.26 6.99 4.39 0.69 NOTE: 2U = two-lane undivided; 4U = four-lane undivided; 4D = four-lane divided; Dir = direction(s). aMixed Facility: A bike lane is present on the roadway in one direction of travel, and a different type of bicycle facility (i.e., buffered bike lane, travel lane with shared-lane marking, protected bike lane, or shared-use path) is present in the opposite direction. bOther Bike Facility: A bicycle facility other than a conventional bike lane (i.e., buffered bike lane, travel lane with shared-lane marking, protected bike lane, or shared-use path) is present on the roadway in one or both directions of travel. TableÂ 47. Buffered bike lanes by road type.

116 Pedestrian and Bicycle Safety Performance Functions at the study site, in most cases a buffered bike lane was present in both directions of travel. For one-way roads, if a buffered bike lane was present at the study site, the buffered bike lane was located only on one side of the road in the direction of travel (i.e., no contraflow buffered bike lanes were included in the analysis). In all cases, the buffer area between the bike lane and the adjacent travel lane was painted or marked with thermoplastic. The separation between the bike lane and the adjacent travel lane was never provided using a raised feature or element. TableÂ 48 provides information on the presence of a travel lane with shared-lane markings and the width of the travel lanes with the markings by road type. For two-lane, two-way roads; four-lane undivided roads; and four-lane divided roads, if a travel lane with a shared-lane marking was present at the study site, in most cases, this type of bicycle facility type was present in both directions of travel. For one-way roads, no travel lanes with shared-lane markings were present on three-lane, one-way roads. TableÂ 49 provides information on the presence of shared-use paths by road type. For two- lane, two-way roads; four-lane undivided roads; four-lane divided roads; and one-lane, one-way roads, if a shared-use path was present at the study site, in most cases, a shared-use path was present in only one direction of travel. TableÂ 50 provides information by road type on the presence of sidewalks and separation distance (i.e., buffer width) between either the outside edge of the travel lane, bike lane, or parking lane and the inside edge of the sidewalk, or if a curb was present, the separation distance from the curb to the inside edge of the sidewalk. The majority of study sites had sidewalks present on one or both sides of the road. There were only a few study sites with no sidewalks present on either side of the road. The average buffer width by road type ranged between 1.42 and 3.44Â ft, but many sites had no separation distance (i.e., no buffer) between the curb and sidewalk. TableÂ 51 provides information on the presence of marked midblock crossings by road type. Across all road types, only nine study sites had a marked midblock crossing. TableÂ 52 provides information on the presence and width of parking lanes by road type. Across all road types, approximately 62Â percent of the study sites had parking lanes present on one or both sides of the road. 3.2.1.2 Roadway Segments: Traffic Volumes Traffic volume data were available for up to 13Â years from 2006 to 2018 for study sites in Minneapolis and for up to 6Â years from 2013 to 2018 for study sites in Philadelphia. TableÂ 53 shows the breakdown of traffic volume by road type. Study period (date range), number of sites and site-years, and basic traffic volume statistics are shown by city and combined across cities for each road type. 3.2.1.3 Roadway Segments: Pedestrian and Bicycle Volumes Direct demand models were developed to estimate pedestrian and bicycle volumes on individual roadway segments as a function of demographic variables, segment characteris- tics, and metrics associated with the built environment. Separate models were developed to estimate pedestrian and bicycle volumes, respectively, for each study location in Minneapolis and Philadelphia. 3.2.1.3.1 Data and Data Preparation The dependent variable in the pedestrian and bicycle exposure models was estimated annual average daily pedestrian and bicycle volumes, respectively, obtained from a subset of unique roadway segments within each of the Minneapolis and Philadelphia street networks.

Road Type Total Number of Segments Travel Lane with Shared-Lane Markings Placement of Markings Lane Width (ft) None One Dir Both Dir Mixed Facilitya Other Bike Facilityb Mean Min. Max. Median Std. Dev. 2U 1,357 552 1 83 4 717 13.58 8.21 22.15 11.48 4.72 4U 350 309 3 13 2 23 14.29 11.69 17.00 13.86 1.72 4D 92 56 0 6 0 30 13.48 10.42 16.95 13.11 1.71 1-lane one-way 215 159 6 0 0 50 16.03 9.49 26.00 12.21 7.80 2-lanes one-way 252 100 25 0 0 127 11.57 8.67 20.24 10.24 3.50 3-lanes one-way 183 70 0 0 0 113 N/A N/A N/A N/A N/A NOTE: 2U = two-lane undivided; 4U = four-lane undivided; 4D = four-lane divided; N/A = Not applicable. aMixed Facility: A bike lane is present on the roadway in one direction of travel, and a different type of bicycle facility (i.e., buffered bike lane, travel lane with shared-lane marking, protected bike lane, or shared-use path) is present in the opposite direction. bOther Bike Facility: A bicycle facility other than a conventional bike lane (i.e., buffered bike lane, travel lane with shared-lane marking, protected bike lane, or shared-use path) is present on the roadway in one or both directions of travel. TableÂ 48. Travel lane with shared-lane markings by road type. Road Type Total Number of Segments Shared-Use Path Placement of Path Buffer Width (ft) Path Width (ft) None One Side Both Sides Other Bike Facilitya Mean Min. Max. Median Std. Dev. Mean Min. Max. Median Std. Dev. 2U 1,357 552 110 1 694 20.75 0.00 115.00 14.10 23.14 10.40 5.46 14.57 10.00 1.55 4U 350 309 1 0 40 8.98 8.98 8.98 8.98 N/A 9.96 9.96 9.96 9.96 N/A 4D 92 56 5 0 31 13.57 0.50 45.00 13.24 14.53 9.83 8.00 11.31 10.00 0.98 215 159 14 0 42 20.19 0.00 100.00 5.50 29.82 9.36 8.00 10.00 9.73 0.76 252 100 0 0 152 N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A 1-lane one- way 2-lanes one- way 3-lanes one- way 183 70 0 0 113 N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A NOTE: 2U = two-lane undivided; 4U = four-lane undivided; 4D = four-lane divided; N/A = Not applicable. aOther Bike Facility: A bicycle facility other than a conventional bike lane (i.e., buffered bike lane, travel lane with shared-lane marking, protected bike lane, or shared-use path) is present on the roadway in one or both directions of travel. TableÂ 49. Shared-use paths by road type.

Road Type Total Number of Segments Sidewalk Placement of Sidewalk Buffer Width (ft) Sidewalk Width (ft) None One Side Both Sides Mean Min. Max. Median Std. Dev. Mean Min. Max. Median Std. Dev. 2U 1,357 71 111 1175 3.22 0.00 300 0.00 11.28 8.74 2.60 41.16 7.59 3.78 4U 350 4 11 335 2.07 0.00 19.00 0.00 3.18 8.62 2.38 23.16 7.70 3.86 4D 92 14 13 65 2.82 0.00 11.08 2.60 3.01 7.83 3.13 23.00 6.88 3.87 1-lane one-way 215 0 47 168 1.84 0.00 80.00 0.00 6.59 10.15 4.02 18.88 10.69 2.52 2-lanes one- way 252 2 4 246 3.44 0.00 10.91 3.76 3.10 8.84 3.00 24.91 7.85 3.62 3-lanes one- way 183 2 2 179 1.42 0.00 10.89 0.00 2.54 9.83 3.54 30.53 9.56 3.82 NOTE: 2U = two-lane undivided; 4U = four-lane undivided; 4D = four-lane divided. TableÂ 50. Sidewalks by road type. Road Type Total Number of Segments Midblock Crossings Type of Crossing Presence of Stop/Yield Lines None No Control Flashing Beacon Signalized No Yes 2U 1,357 1352 4 1 0 4 1 4U 350 349 0 1 0 0 1 4D 92 91 1 0 0 0 0 1-lane one-way 215 215 0 0 0 N/A N/A 2-lanes one-way 252 251 1 0 0 0 1 3-lanes one-way 183 182 0 0 1 0 1 NOTE: 2U = two-lane undivided; 4U = four-lane undivided; 4D = four-lane divided; N/A = Not applicable. TableÂ 51. Midblock crossings by road type.

Road Type Total Number of Segments Parking Lane Location of Parking Lane Parking Lane Width (ft) None One Dir Both Dir Mean Min. Max. Median Std. Dev. 2U 1,357 419 164 774 7.58 4.00 12.31 7.27 1.17 4U 350 247 13 90 7.27 4.81 12.00 7.53 1.11 4D 92 48 11 33 7.87 5.92 12.01 7.71 1.31 1-lane one-way 215 49 57 109 7.87 6.01 11.86 7.77 0.98 2-lanes one-way 252 74 60 118 8.77 6.06 13.00 8.08 1.90 3-lanes one-way 183 91 60 32 8.17 6.00 12.40 8.10 0.91 NOTE: 2U = two-lane undivided; 4U = four-lane undivided; 4D = four-lane divided. TableÂ 52. Parking lane by road type.

120 Pedestrian and Bicycle Safety Performance Functions In Minneapolis, annual average daily pedestrian and bicycle volume estimates were both available at 535 unique roadway segments. These estimates were provided by the City of Minneapolis, which applied expansion factors obtained from long-term counting stations to short-term counts at the individual sites. Short-term counts for these sites were performed between 2007 and 2017, and only a single count was performed at each site to estimate its annual average pedestrian or bicycle volume. At each count location, an imaginary screen line was drawn across the street and included any sidewalks or paths. All pedestrians and bicyclists crossing that line were counted so the pedestrian and bicycle volumes are for both directions of travel along the road. More details on the counting program are available from Lindsey etÂ al. (2013). In Philadelphia, there were some overlaps between locations with pedestrian and bicycle exposure estimates; however, in general, the counts were provided separately for each mode. Annual average daily pedestrian traffic data were available at 286 unique locations, while annual average daily bicycle traffic data were available at 393 unique locations. Counts were performed between 2010 and 2018, and multiple counts were often performed at each unique location during this period. Only the latest pedestrian or bicycle exposure estimate was used at each location for modeling purposes. Furthermore, pedestrian counts were available for one side of the road (i.e., one sidewalk) in some counts and available separately for both sides of the road (i.e., both sidewalks) in others. To account for this, counts on different sidewalks on the same roadway segment were combined to provide the total pedestrian exposure along that segment. Indicator variables were included in the dataset to represent if the count represented one sidewalk or two sidewalks to account for this difference in counting procedures. Independent variables considered for inclusion in the exposure model for both Minneapolis and Philadelphia consisted of the following characteristics associated with each roadway Two-Lane Undivided Roads Minneapolis 2006â2018 651 3,758 580 20,700 7,357 7,139 Philadelphia 2013â2018 706 4,044 1,239 16,429 7,642 6,900 Combined 2006â2018 1,357 7,802 580 20,700 7,505 7,088 Four-Lane Undivided Roads Minneapolis 2006â2018 290 3,238 3,390 29,969 14,653 13,346 Philadelphia 2013â2018 60 360 5,578 21,718 13,572 10,942 Combined 2006â2018 350 3,598 3,390 29,969 14,478 13,258 Four-Lane Divided Roads Minneapolis 2006â2018 44 501 6,339 23,550 13,341 12,892 Philadelphia 2013â2018 48 280 5,349 31,435 13,622 13,062 Combined 2006â2018 92 781 5,349 31,435 13,485 13,062 One-Lane, One-Way Roads Minneapolis 2006â2018 48 511 1,022 12,700 6,683 3,350 Philadelphia 2013â2018 167 1,000 1,074 12,695 4,827 3,857 Combined 2006â2018 215 1,511 1,022 12,700 5,242 3,857 Two-Lane, One-Way Road Minneapolis 2006â2018 115 877 2,450 18,896 8,634 8,785 Philadelphia 2013â2018 137 688 3,039 17,896 8,967 7,887 Combined 2006â2018 252 1,565 2,450 18,896 8,814 7,887 Three-Lane, One-Way Road Minneapolis 2006â2018 151 1,334 2,846 29,000 10,732 10,931 Philadelphia 2013â2018 32 187 6,141 17,161 13,055 12,624 Combined 2006â2018 183 1,521 2,846 29,000 11,139 12,001 TableÂ 53. Roadway-segment motor vehicle traffic volumes by road type.

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 121Â Â segment: employment and land use, infrastructure, transit, demographics, and street network connectivity. A summary of each of these variables, including its description, the scale at which it was collected, and the data source used to obtain it, is provided in TableÂ 54. Variables collected at the point scale refer to characteristics obtained for the specific roadway segment. Variables collected at the buffer scale refer to demographic or street network metrics that were calcu- lated for a circular area surrounding each roadway segment. Multiple radii (0.1Â mi, 0.25Â mi, and 0.5Â mi) were considered to determine the scale that yielded the best relationship between the respective variable and pedestrian or bicycle exposure. In all cases, the center of the circular buffer was placed at the geographic midpoint of the street segments. In some cases, the segments extend beyond the buffer; however, this was considered acceptable since it happened a few times and to keep data elements consistent across observations. Most of the variables in TableÂ 54 are self-explanatory, except for âEmployment square footage of foot traffic land usesâ and âLand-use entropy.â âEmployment square footage of foot traffic land usesâ aims to capture the scale of businesses within the buffer that may generate walking trips by attracting customers. This was estimated using ESRI Business Analyst data and filtered busi- nesses using existing North American Industry Classification System (NAICS) codes that may generate foot traffic. The specific codes considered here include: â¢ 44-45: Retail trade. â¢ 522: Banks. â¢ 54: Professional, scientific, and technical services. â¢ 62: Health care and social assistance. â¢ 71: Arts, entertainment, and recreation. â¢ 72: Accommodation and food services. â¢ 812: Personal and laundry services. â¢ 813: Religion, grantmaking, civil, professional, and similar organizations. Within this dataset, each business is categorized within one of four ranges, based on size: â¢ A: 1â2,499Â sq. ft. â¢ B: 2,500â9,999Â sq. ft. â¢ C: 10,000â39,999Â sq. ft. â¢ D: 40,000Â sq. ft and above. For categories A, B and C, the midpoint of each range was used as the estimate of business size. For category D, the physical amount of space within the buffer was used. âLand-use entropyâ describes the mixture of land-use types within the buffer area and was calculated using the following equation: (3-1) ln ln Land Use Entropy k p pi i i k ) = - ` ` j j: D / where pi is the proportion of land area in land use i, and k is the number of land uses (Frank, Andresen, and Schmid 2004). Following Hankey etÂ al. (2012), k = 4 was selected for modeling purposes. For Minneapolis, land uses were assigned using land-use codes provided by the City of Minneapolis as follows: â¢ Residential â 111, 112, 113, 114, 115, 116, 141. â¢ Commercial and office â 120, 130, 143. â¢ Social and institutional â 160. â¢ Parks and recreational â 170, 173.

122 Pedestrian and Bicycle Safety Performance Functions Variable Name Description Scale Minneapolis Data Source Philadelphia Data Source Employment/Land Use Emp Number of employees Buffer ESRI Business Analyst EmpSQFT Employment square footage of foot traffic land uses Buffer ESRI Business Analyst LUEntropy Land-use entropy Buffer General land use 2016 DVRPC 2015 land use Infrastructure Arterial Count is on a principal arterial. Point Minneapolis street centerline Delaware County road centerline, Montgomery County centerlines, Philadelphia street centerlines Minorart Count is on a minor arterial. Point Minneapolis street centerline N/A Collector Count is on a collector street. Point Minneapolis street centerline Delaware County Road Centerline, Montgomery County Centerlines, Philadelphia Street Centerlines Signal Intersection has a signal. Point Minneapolis street centerline Philadelphia intersection controls, and Google StreetView SPEED_LIM Speed limit on segment Point Minneapolis street centerline DVRPC AADT AADT for segment Point Minneapolis AADT DVRPC Transit TransitStops Transit stops within buffer Buffer Transit stops N/A BusStops Bus stops within buffer Buffer N/A SEPTA bus stops RailStops Rail stops within buffer Buffer N/A DVRPC passenger rail stations Demographics Pop Population Buffer U.S. Census ACS HseHld Number of households Buffer U.S. Census ACS WalkCom Number of walk commuters Buffer U.S. Census ACS TransCom Number of transit commuters Buffer U.S. Census ACS NoVeh Number of households with no vehicle Buffer U.S. Census ACS Degree Number of college degree holders Buffer U.S. Census ACS WhitePct Percent of population of White people alone Buffer U.S. Census ACS WalkComPct Walk commute mode share Buffer U.S. Census ACS TransComPct Transit commute mode share Buffer U.S. Census ACS NoVehPct Percent of households with no vehicle Buffer U.S. Census ACS DegreePct Percent of population with a college degree Buffer U.S. Census ACS Network Connectivity StSeg Number of street segments Buffer Minneapolis street centerline 2019 Census TIGER/Line NOTE: N/A = Not applicable; ACS = American Community Survey; TIGER = File/Topologically Integrated Geographic Encoding and Referencing (a U.S. Census database). TableÂ 54. Description of independent variables considered in pedestrian and bicycle exposure models.

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 123Â Â For Philadelphia, land uses with similar broad categories were considered as follows: â¢ Residential. â¢ Commercial. â¢ Institutional. â¢ Recreational. The set of available roadway segments was then sampled to ensure that the distribution of functional classifications for the roadway segments used to estimate the exposure models better matched that of the roadway segments on which the exposure models would be applied. Matching this distribution helped minimize unobserved impacts that might be associated with functional classification on pedestrian and bicycle exposure. A summary of these distributions is provided in TableÂ 55 for Minneapolis. As shown, of the roadway segments on which the expo- sure models were eventually applied, 2.9Â percent were classified as principal arterials, 33.3Â per- cent were classified as major collectors, 18.3Â percent were classified as A â minor augmentors, 20.5Â percent were classified as A â minor relievers, 16.4Â percent were classified as B â minor roadways, and 8.7Â percent had no available information on functional classification (N/A). However, this distribution is skewed more heavily toward segments without functional clas- sifications (N/A) in the set of roadway segments that have pedestrian and bicycle volumes. To alleviate this, 35 roadway segments with functional classification N/A and 153 major collector roadway segments were randomly sampled and included in the exposure modeling database, along with the remaining roadway segments for the other functional classifications. This provided a distribution that was more in line with that observed for the set of roadway segments for which the exposure model was applied. The final modeling database consisted of 418 roadway segments for both the pedestrian and bicycle exposure models. Fewer functional classification categories were available in Philadelphia (see TableÂ 56). As shown, of the roadway segments on which the exposure models were eventually applied, 55.6Â percent were classified as arterials, 29.5Â percent were classified as collectors, and the remaining 14.9Â percent (N/A) did not have a classification and were assumed to represent local roads. However, this distribution was skewed more heavily toward segments on arterials in sets of roadway segments that have pedestrian or bicycle volumes. To alleviate this, 107 roadway segments classified as arterials were randomly sampled for use in the pedestrian exposure modeling database, and only 141 segments classified as arterials were randomly sampled for use in the bicycle exposure modeling database. This provided a distribution that was more in line with that observed for the set of roadway segments where the exposure model was eventually applied. The final pedestrian modeling database consisted of 192 roadway segments, and the final bicycle exposure modeling database consisted of 253 roadway segments. Functional Classification Segments with Pedestrian/Bicycle Volumes Segments Included in Exposure Modeling Database Segments on Which Exposure Models Were Applied Count Percent Count Percent Count Percent Principal arterial 6 1.1% 6 1.4% 72 2.9% Major collector 154 28.8% 153 36.6% 835 33.3% A - Minor augmentor 64 12.0% 64 15.3% 458 18.3% A - Minor reliever 81 15.1% 81 19.4% 514 20.5% B - Minor 79 14.8% 79 18.9% 410 16.4% Not applicable 151 28.2% 35 8.4% 217 8.7% Total 535 100.0% 418 100.0% 2,506 100.0% TableÂ 55. Description of functional classification of roadway segments available for exposure models included in exposure model database and on which models were applied (Minneapolis).

124 Pedestrian and Bicycle Safety Performance Functions Select summary statistics for the roadway segments included in the Minneapolis analysis database are provided in TableÂ 57. This includes summary statistics for the pedestrian and bicycle volumes, the year during which counts were obtained, and the independent variables included in the final exposure models. Summary statistics for other variables were omitted for brevity. As shown, roadway segments included in the database had a wide range of pedestrian and bicycle volumes. Most counts were obtained from 2014 and after. Summary statistics for roadway segments included in the Philadelphia databases used for the development of the pedestrian and bicycle exposure models are provided in TableÂ 58 and TableÂ 59, respectively. This includes summary statistics for the pedestrian and bicycle volumes, the year during which counts were obtained, and the independent variables included in the final exposure models. Summary statistics for other variables were omitted for brevity. As shown, roadway segments included in the database had a wide range of pedestrian and bicycle volumes. 3.2.1.3.2 Model Development Two model types were considered for developing the exposure models in this work: linear regression and count regression models. For the linear regression models, log-linear models were explicitly considered in which the dependent variable was the natural logarithm of the pedestrian or bicycle volumes to ensure that only positive values could be obtained from the models. For the count regression models, the negative binomial model form was selected due to its ability to accommodate overdispersion in the dataset in which the variance of the observed data exceeds its mean. The two model forms were compared, and the negative binomial regression model form was selected since it provided a better fit to the observed data. This result is consis- tent with previous research in this area, which suggests that the negative binomial model form provides a better estimate of pedestrian or bicycle volumes compared to the log-linear model forms (Medury etÂ al. 2019). Negative binomial regression models take the following functional form. (3-2)ln Xi i im b f= + where: Î»i = expected annual average daily pedestrians or bicyclists for observation i, X = vector of independent variables, Î² = vector of estimated coefficients, and exp(Îµi) = error term following a gamma distribution. Functional Classification Segments with Pedestrian Volumes Segments Included in Pedestrian Modeling Database Segments with Bicycle Volumes Segments Included in Bicycle Modeling Database Segments on Which Exposure Models Were Applied Count (percent) Count (percent) Count (percent) Count (percent) Count (percent) Arterial 202 (70.4%) 107 (55.7%) 281 (71.5%) 141 (55.7%) 248 (55.6%) Collector 56 (19.5%) 56 (29.2%) 75 (19.1%) 75 (29.6%) 131 (29.5%) N/A 29 (10.1%) 29 (15.1%) 37 (9.4%) 37 (14.6%) 66 (14.9%) Total 287 (100.0%) 192 (100.0%) 393 (100.0%) 253 (100.0%) 445 (100.0%) TableÂ 56. Description of functional classification of roadway segments available for exposure models included in exposure model database and on which models were applied (Philadelphia).

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 125Â Â Continuous Variable Mean Std. Dev. Min. Max. Pedestrian volume (ped/day) 1,097.9 2,241.9 10 20,350 Bicycle volume (bike/day) 372.8 587.4 10 7,370 Population within 0.1 mi (people) 310.5 196.4 26 1,942 Population within 0.5 mi (people) 7,302.0 3,302.4 996 15,722 Walk commute mode share within 0.25 mi (proportion) 0.1108 0.1333 0 0.5657 Walk commute mode share within 0.5 mi (proportion) 0.1113 0.1194 0.0011 0.4654 Percent of population with a college degree within 0.5 mi (proportion) 0.2901 0.0926 0.1004 0.5032 Percent of population of White people within 0.5 mi (proportion) 0.6444 0.1987 0.1542 0.9377 Land-use entropy within 0.1 mi 0.4698 0.2436 0.0000 0.9592 Land-use entropy within 0.25 mi 0.5767 0.2047 0.0510 0.9820 Number of transit stops within 0.1 mi 5.9880 5.3958 0.0000 34 Categorical Variable Category Percent Count year 2007 2.4% 2008 1.4% 2009 4.3% 2010 1.0% 2011 1.4% 2012 1.0% 2013 6.0% 2014 17.0% 2015 15.3% 2016 20.8% 2017 29.4% Arterial roadway Yes 36.1% No 63.9% Speed limit Not available 0.2% 10 0.5% 25 6.7% 30 87.3% 35 3.6% 40 1.2% 45 0.2% 50 0.2% Type of bicycle facility present None 56.0% Physically separated bicycle facility 13.9% Nonphysically separated bicycle facility 30.1% TableÂ 57. Summary statistics for variables included in the exposure models (Minneapolis). Continuous Variable Mean Std. Dev. Min. Max. Pedestrian volume (ped/day) 2,457.5 3,404.6 7 18,051 Population within 0.5 mi (people) 14,776.8 8,030.8 333.3 31,186.3 Walk commute mode share within 0.25 mi (proportion) 0.5837 0.2555 0.0114 0.903 Employment square footage of foot traffic land uses within 0.1 mi (sq. ft) 1,414,404 3,554,560 0 25,286,500 Land-use entropy within 0.25 mi 0.6626 0.1686 0.0489 0.9615 Categorical Variable Category Percent Count year 2010 10.4% 2011 28.6% 2012 10.9% 2013 3.1% 2014 0.5% 2015 2.6% 2016 34.9% 2017 3.6% 2018 5.2% Arterial roadway Yes 55.7% No 45.3% TableÂ 58. Summary statistics for variables included in pedestrian exposure model (Philadelphia).

126 Pedestrian and Bicycle Safety Performance Functions The corresponding mean-variance relationship is described in the following equation. (3-3)E yVar y E y ii i 2 a= +` ` `j j j where: yi = number of crashes for observation i and Î± = overdispersion parameter. Under such assumptions, the probability distribution function and likelihood function of negative binomial models can be written as shown in Equation 3-4 and Equation 3-5 (Shankar, Mannering, and Barfield 1995). (3-4) ! P y y y i i i i i i i i i i m i i m m C C = + + + i yJ L K K J L K K` ` ` N P O O N P O Oj j j where: Î¸ = 1/Î± and Î = gamma function. (3-5) ! L y y i i i i N i i i 1 i m i i i m i i m m C C = + + + i = yJ L K K J L K K` ` ` N P O O N P O Oj j j% where: N = total number of observations in the sample. The model is estimated using the maximum likelihood estimation method. In this method, the model coefficients are selected such that they minimize the likelihood function provided in Equation 3-4. Since this number is usually very small, the natural logarithm of Equation 3-5 is typically used, which is more commonly known as the log-likelihood. Continuous Variable Mean Std. Dev. Min. Max. Bicycle volume (bike/day) 320.7 503.0 0 3574 Population within 0.5 mi (people) 14679.1 8849.6 204.7 37189.5 Walk commute mode share within 0.25 mi (proportion) 0.204 0.162 0.000 0.617 Percent of population that is White within 0.25 mi (proportion) 0.609 0.234 0.009 0.990 Land-use entropy within 0.25 mi 0.656 0.180 0.000 0.976 Categorical Variable Category Percent Count year 2010 9.5% 2011 22.2% 2012 9.5% 2013 1.6% 2015 7.9% 2016 21.8% 2017 18.3% 2018 9.1% Arterial roadway Yes 55.6% No 45.4% Type of bicycle facility present Physically separated bicycle facility 9.5% Other 90.5% TableÂ 59. Summary statistics for variables included in bicycle exposure model (Philadelphia).

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 127Â Â A forward-selection process was used to estimate the pedestrian and bicycle exposure models. In this method, each potential independent variable shown in TableÂ 54 was tested to examine its impact on the dependent variable and the most impactful variable selected. Non- linear transformations were considered for the continuous independent variables when such transformation improved the overall model fit and to identify the best functional form for each of the potential independent variables. Then, all other variables were sequentially added to determine the best pair. This process was repeated until the addition of no other variables could significantly improve the overall model fit. Note that Spearman correlation coefficients between all independent variables were examined to remove redundant effects from the models. Categorical variables were combined to ensure that a sufficient number of observa- tions were included within each group in the models. Finally, indicator variables for the year in which the count was taken were incorporated for the years 2013 to 2017, inclusive. These variables helped to account for unobserved annual factors that might have influenced pedes- trian or bicycle volumes, such as weather or active commuting campaigns. The resulting models for Minneapolis are provided in TableÂ 60 and TableÂ 61 for pedestrian and bicycle exposure (in both directions), respectively. These tables provide the set of variables included in the final models, the spatial level of each variable (where applicable), any transfor- mations that were used, the model coefficient, and associated p-value. Note that all variables were statistically significant at the 95Â percent confidence level, except for the yearly indicators. However, these variables were left in the final models since the models will be used for predic- tive (as opposed to explanatory) purposes, and they improved the overall prediction accuracy of the models. The pedestrian exposure model revealed that pedestrian volumes on a roadway segment increase with the population within 0.1Â mi of the roadway segment, the fraction of the popu- lation that identifies as walking commuters within 0.25Â mi, the fraction of population within 0.5Â mi with a college degree, land-use entropy (the measure of the mixture of land users near the roadway segment) within 0.1Â mi, and the number of transit stops within 0.1Â mi. Pedestrian volumes are also higher on arterial segments as opposed to nonarterial segments and on roadway segments with posted speed limits of 25Â mph or less. All these relationships are consistent with engineering expectations. The bicycle exposure model revealed that bicycle volumes on a roadway segment increase with the population within 0.5Â mi of the roadway segment, the fraction of the population that identifies as walking commuters within 0.5Â mi, the fraction of population within 0.5Â mi that Variable Spatial Level Transformation Coefficient p-Value Population (people) 0.10 mi Log 0.247 < 0.001 Walk commute mode share (proportion) 0.25 mi Square 10.668 < 0.001 Percent of population with a college degree (proportion) 0.50 mi None 2.130 < 0.001 Land-use entropy 0.10 mi None 1.242 < 0.001 Number of transit stops 0.10 mi None 0.032 0.001 Arterial roadway â Indicator 0.527 < 0.001 Speed limit 25 mph or less â Indicator 1.129 < 0.001 Year is 2013 â Indicator â0.226 0.271 Year is 2014 â Indicator â0.064 0.678 Year is 2015 â Indicator 0.130 0.409 Year is 2016 â Indicator 0.076 0.610 Year is 2017 â Indicator â0.039 0.777 Constant â â 3.085 < 0.001 Alpha â â 0.639 â NOTE: 2 x log-likelihood = â6135.141; root-mean-square error (RMSE) = 1548.6; â = Not applicable. TableÂ 60. Pedestrian exposure model (Minneapolis).

128 Pedestrian and Bicycle Safety Performance Functions is White, and land-use entropy within 0.25Â mi. Bicycle volumes decrease with the number of transit stops within 0.1Â mi. Bicycle volumes are also higher on arterial segments as opposed to nonarterial segments and on roadway segments with posted speed limits of 25Â mph or less. Finally, bicycle volumes are higher on roadways with bicycle facilities. Segments with physically separated bicycle facilities (e.g., protected bike lanes) generally have higher bicycle volumes than those with nonphysically separated bicycle facilities (e.g., conventional or buffered bike lane). All these relationships are consistent with engineering expectations. The models for Philadelphia are provided in TableÂ 62 and TableÂ 63 for pedestrian and bicycle exposure (in both directions), respectively. These tables provide the set of variables included in the final models, the spatial level of each variable (where applicable), any transformations that were used, the model coefficient, and associated p-value. Not all variables were statistically signifi- cant at the 95Â percent confidence level. However, these insignificant explanatory variables were kept because they were found to improve the overall model fit and had predictive coefficients Variable Spatial Level Transformation Coefficient p-Value Population (people) 0.50 mi None 0.00002 0.079 Walk commute mode share (proportion) 0.50 mi None 4.18900 < 0.001 Percent of population of White people (proportion) 0.50 mi None 1.10000 < 0.001 Land-use entropy 0.25 mi None 0.77600 < 0.001 Number of transit stops 0.10 mi None â0.02800 < 0.001 Arterial roadway â Indicator 0.29500 < 0.001 Speed limit 25 mph or less â Indicator 1.05000 < 0.001 Nonphysically separated bike facility (i.e., bike lane, buffered bike lane, bike boulevard, roadway with shared-lane marking) â Indicator 0.61000 < 0.001 Physically separated bike facility (trail, shared- use path, protected bike lane, ped/bike bridge) â Indicator 0.76700 < 0.001 Year is 2013 â Indicator â0.42200 0.020 Year is 2014 â Indicator â0.44700 0.001 Year is 2015 â Indicator â0.15100 0.292 Year is 2016 â Indicator â0.38500 0.005 Year is 2017 â Indicator â0.39300 0.002 Constant â â 3.83100 < 0.001 Alpha â â 0.49300 â NOTE: 2 x log-likelihood = â5358.449; RMSE = 474.7; â = Not applicable. TableÂ 61. Bicycle exposure model (Minneapolis). Variable Spatial Level Transformation Coefficient p-Value Population (people) 0.50 mi Log 0.229 0.043 Walk commute mode share (proportion) 0.25 mi None 2.804 < 0.0010 Employment square footage of foot traffic land uses (sq. ft) 0.10 mi None 4.250E-08 0.0860 Land-use entropy 0.25 mi None 2.944 < 0.0010 Arterial roadway â Indicator 0.644 < 0.0010 Year is 2011 â Indicator 0.060 0.8410 Year is 2012 â Indicator 0.319 0.3850 Year is 2013 or 2014 â Indicator 0.784 0.0750 Year is 2015 or 2016 â Indicator 0.215 0.4520 Year is 2017 or 2018 â Indicator 0.286 0.4430 Constant â â 1.832 0.0725 Alpha â â 0.974 â NOTE: 2 x log-likelihood = â3218.034; RMSE = 2309.2; â = Not applicable. The number of sidewalks was included as an offset variable. TableÂ 62. Pedestrian exposure model (Philadelphia).

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 129Â Â that were in line with expectations. Like the Minneapolis models, the yearly indicators were also included to improve prediction accuracy. The pedestrian exposure model revealed that pedestrian volumes on a roadway segment increase with the population within 0.5Â mi of the roadway segment, the fraction of the population that identifies as walking commuters within 0.25Â mi, the amount of employment square footage of foot traffic land uses within 0.1Â mi, and land-use entropy (the measure of the mixture of land users near the roadway segment) within 0.25Â mi. Pedestrian volumes are also higher on arterial segments as opposed to nonarterial segments. Since the number of sidewalks was included as an offset variable, the result of the exposure estimation provides the average pedes- trian exposure on one sidewalk. On roadway segments with two sidewalks, the estimated value would be doubled to estimate the pedestrian exposure on both sides of the road. All these relationships are consistent with engineering expectations. The bicycle exposure model revealed that bicycle volumes on a roadway segment increase with the population within 0.5Â mi of the roadway segment, the fraction of the population that identifies as walking commuters within 0.25Â mi, and the fraction of the population of White people within 0.25Â mi. Bicycle volumes are also higher on arterial segments as opposed to non- arterials. Finally, bicycle volumes are higher on roadways with protected bike lanes and buffered bike lanes. All these relationships are consistent with engineering expectations. 3.2.1.3.3 Exposure Estimates for Study Locations Using the direct demand models to estimate pedestrian and bicycle volumes for the study locations, TableÂ 64 summarizes the annual average daily pedestrian and bicycle (AADP and AADB, respectively) volumes for use in model development by city and road type. These are estimates of daily pedestrian and bicycle volumes along the road in both directions of travel. 3.2.1.4 Roadway Segments: Crash Data Crash data were available for up to 13Â years (from 2006 to 2018) for study sites in Minneapolis and for up to 6Â years (from 2013 to 2018) for study sites in Philadelphia. TableÂ 65 shows the breakdown of pedestrian and bicycle crash data by city and combined across both cites by severity and by road type. Pedestrian and bicycle crashes were assigned to roadway segments if the crash was located within the limits/boundaries of the site and the crash was designated as a nonjunction or driveway-related crash. Regarding severity, crash frequencies are provided for all severity levels combined (Total) and for fatal and suspected serious (FS) injury crashes. Variable Spatial Level Transformation Coefficient p-Value Population (people) 0.50 mi None 0.00004 < 0.001 Walk commute mode share (proportion) 0.25 mi None 2.80600 < 0.001 Percent of population of White people (proportion) 0.25 mi None 0.59000 0.047 Arterial roadway â Indicator 0.28100 0.045 Presence of protected bike lane or buffered bike lane â Indicator 0.37400 0.125 Year is 2012 or 2013 â Indicator 0.85800 < 0.001 Year is 2015 â Indicator â0.04700 0.860 Year is 2016 â Indicator â0.49300 0.010 Year is 2017 â Indicator 0.57500 0.004 Year is 2018 â Indicator 0.14400 0.568 Constant â â 3.66900 < 0.001 Alpha â â 1.08600 â NOTE: 2 x log-likelihood = â3255.406; RMSE = 465.3; â = Not applicable. There were no observations from 2014 to include in the model. TableÂ 63. Bicycle exposure model (Philadelphia).

City Date Range Number of Sites Number of Site-Years Annual Average Daily Pedestrian Volume (AADP) (ped/day) Average Annual Daily Bicycle Volume (AADB) (bike/day) Min. Max. Mean Median Min. Max. Mean Median Two-Lane Undivided Roads Minneapolis 2006â2018 651 3,758 68 11,185 328 201 105 2,565 440 201 Philadelphia 2013â2018 706 4,044 47 2,003 327 294 87 1,289 238 180 Combined 2006â2018 1,357 7,802 47 11,185 328 250 87 2,565 335 251 Four-Lane Undivided Roads Minneapolis 2006â2018 290 3,238 73 10,679 787 268 140 1,809 525 460 Philadelphia 2013â2018 60 360 158 489 280 255 106 315 187 183 Combined 2006â2018 350 3,598 73 10,679 705 266 106 1,809 471 420 Four-Lane Divided RoadS Minneapolis 2006â2018 44 501 80 452 211 210 146 1,131 451 469 Philadelphia 2013â2018 48 280 69 1,395 340 311 101 802 226 200 Combined 2006â2018 92 781 69 1,395 277 261 101 1,131 336 279 One-Lane One-Way Roads Minneapolis 2006â2018 48 511 354 646 473 468 618 810 707 702 Philadelphia 2013â2018 167 1,000 62 5,075 536 274 88 2,011 435 307 Combined 2006â2018 215 1,511 62 5,075 522 357 88 2,011 496 435 Two-Lane One-Way Roads Minneapolis 2006â2018 115 877 81 1,610 240 158 245 1,495 488 328 Philadelphia 2013â2018 137 688 246 7,341 3,021 2,194 195 1,992 875 832 Combined 2006â2018 252 1,565 81 7,341 1,747 523 195 1,992 698 609 Three-Lane One-Way Roads (3OW) Minneapolis 2006â2018 151 1,334 142 13,323 1,649 944 286 1,997 991 1,001 Philadelphia 2013â2018 32 187 248 6,871 1,757 472 196 1,130 441 234 Combined 2006â2018 183 1,521 142 13,323 1,668 859 196 1,997 895 925 TableÂ 64. Pedestrian and bicycle volumes by road type.

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 131Â Â Some of the crashes in the crash dataset were classified as property-damage-only crashes. By definition, it is difficult to understand how a pedestrian or bicycle crash could be classified as property damage only, but recognizing that some pedestrian and bicycle crashes are classified as such, these crashes were included in the count of total crashes (i.e., all severity levels combined). 3.2.2 Final Databases: Intersections TableÂ 66 presents the number of sites and site-years by intersection traffic control type and con- figuration for which inventory, exposure, and crash data were available for model development for City Date Range Number of Sites Miles Total Mile-Years Pedestrian Crashes Bicycle Crashes Total FS Total FS Two-Lane Undivided Roads Minneapolis 2006â2018 651 65.32 568.87 87 13 60 2 Philadelphia 2013â2018 706 48.76 278.54 163 10 50 1 Combined 2006â2018 1,357 114.08 847.41 250 23 110 3 Four-Lane Undivided Roads Minneapolis 2006â2018 290 23.91 268.44 89 32 91 10 Philadelphia 2013â2018 60 3.91 23.48 7 1 5 0 Combined 2006â2018 350 27.82 291.92 96 33 96 10 Four-Lane Divided Roads Minneapolis 2006â2018 44 4.98 56.27 4 3 3 0 Philadelphia 2013â2018 48 6.19 34.33 3 1 5 0 Combined 2006â2018 92 11.17 90.60 7 4 8 0 One-Lane, One-Way Roads Minneapolis 2006â2018 48 4.33 44.29 0 0 0 0 Philadelphia 2013â2018 167 7.20 43.11 15 2 12 1 Combined 2006â2018 215 11.53 87.40 15 2 12 1 Two-Lane, One-Way Roads Minneapolis 2006â2018 115 10.74 78.25 6 0 7 1 Philadelphia 2013â2018 137 9.48 44.47 25 0 8 1 Combined 2006â2018 252 20.22 122.72 31 0 15 2 Three-Lane, One-Way Roads Minneapolis 2006â2018 151 11.93 106.15 23 2 25 0 Philadelphia 2013â2018 32 2.04 11.95 10 1 4 0 Combined 2006â2018 183 13.97 118.10 33 3 29 0 TableÂ 65. Pedestrian and bicycle crash frequency by city, severity, and road type. Intersection Traffic Control Type and Configuration Minneapolis Philadelphia Total Number of Intersections Site-Years Number of Intersections Site-Years Number of Intersections Site-Years 3ST (2Ã2) 14 73 15 85 29 158 3SG (2Ã2) 15 86 20 117 35 203 4ST (2Ã2) 10 42 1 3 11 45 4SG (2Ã2) 84 441 43 204 127 645 4SG (1Ã2) 58 259 11 54 69 313 NOTE: 3ST 2Ã2= three-leg stop control intersections with two-way/two-way operations; 3SG 2Ã2= three-leg signal control intersections with two-way/two-way operations; 4ST 2Ã2 = four-leg stop control intersections with two-way/two-way operations; 4SG 2Ã2 = four-leg signal control intersections with two-way/two-way operations; 4SG 1Ã2 = four-leg signal control intersections with one-way/two-way operations combined. TableÂ 66. Number of sites and site-years, by intersection configuration.

132 Pedestrian and Bicycle Safety Performance Functions sites in Minneapolis and Philadelphia. Only those intersection configurations for which a sufficient amount of data was collected for potential model development are provided in the table. 3.2.2.1 Intersections: Site Characteristics This section provides several descriptive statistics of the site characteristics and inventory data collected for intersections. TableÂ 67 indicates the number of intersections where intersection lighting is present or not present. Approximately 91Â percent of the intersections had lighting. TableÂ 68 provides information on lane widths by major and minor roads. The lane widths are for lanes designated for motor vehicle traffic entering (inbound) and exiting (outbound) the intersection and include through lanes, lanes with shared movements, and dedicated turn lanes. TableÂ 69 provides information on the number of approaches with left-turn lanes present. No distinction is made on whether the left-turn lane is present on the major- or minor-road approach. TableÂ 70 provides information on the number of approaches with right-turn lanes present. No distinction is made on whether the right-turn lane is present on the major- or minor-road approach. Intersection Traffic Control Type and Configuration Total Number of Intersections Presence of Lighting 3ST (2Ã2) 29 3SG (2Ã2) 35 4ST (2Ã2) 11 4SG (2Ã2) 127 4SG (1Ã2) 69 No 10 4 2 3 5 Yes 19 31 9 124 64 NOTE: 3ST 2Ã2= three-leg stop control intersections with two-way/two-way operations; 3SG 2Ã2= three-leg signal control intersections with two-way/two-way operations; 4ST 2Ã2 = four-leg stop control intersections with two-way/two-way operations; 4SG 2Ã2 = four-leg signal control intersections with two-way/two-way operations; 4SG 1Ã2 = four-leg signal control intersections with one-way/two-way operations combined. TableÂ 67. Presence of lighting by intersection configuration. Intersection Traffic Control Type and Configuration Total Number of Intersections Approach Lane Width (ft) Mean Min. Max. Median Std. Dev. 3ST (2Ã2) 29 Major 11.34 8.00 21.00 11.00 2.40 Minor 14.24 8.80 25.00 12.00 4.83 3SG (2Ã2) 35 Major 11.58 6.75 21.50 11.00 2.59 Minor 13.42 8.40 38.30 11.25 4.85 4ST (2Ã2) 11 Major 13.03 7.75 19.50 12.50 2.96 Minor 16.89 6.90 42.25 17.00 4.36 4SG (2Ã2) 127 Major 12.39 6.25 25.50 11.50 3.45 Minor 13.72 5.60 30.00 12.10 3.93 4SG (1Ã2) 69 Major 12.80 6.70 27.40 12.00 3.66 Minor 13.93 6.00 30.00 12.00 4.34 NOTE: 3ST 2Ã2= three-leg stop control intersections with two-way/two-way operations; 3SG 2Ã2= three-leg signal control intersections with two-way/two-way operations; 4ST 2Ã2 = four-leg stop control intersections with two-way/two-way operations; 4SG 2Ã2 = four-leg signal control intersections with two-way/two-way operations; 4SG 1Ã2 = four-leg signal control intersections with one-way/two-way operations combined. TableÂ 68. Lane width by major and minor road by intersection configuration.

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 133Â Â TableÂ 71 provides information on the number of intersection legs where a marked parking lane is present in the inbound direction, outbound direction, or both. No distinction is made on whether the parking lane is present on the major or minor roads. TableÂ 72 provides information on the number of intersection legs with medians present. No distinction is made on whether the median is present on the major or minor roads. TableÂ 73 provides information on median widths by major and minor roads. TableÂ 74 provides information on the number of approaches with bike lanes or buffered bike lanes present. No distinction is made on whether the bike lane or buffered bike lane is present on a major- or minor-road approach. TableÂ 75 provides information on crossing widths, curb-to-curb. The crossing width includes widths of through lanes, left-turn lanes, right-turn lanes, parking lanes, shoulders, bike lanes, and medians. TableÂ 76 provides information on the crossing distance measured along the crosswalk. If a median was present, the median width was subtracted from the crossing distance. TableÂ 77 provides information on the number of lanes that a pedestrian must cross at an inter- section by major and minor roads. Both through lanes and turning lanes that are crossed by a pedestrian along a crossing path are considered. Intersection Traffic Control Type and Configuration Total Number of Intersections Number of Approaches with a Left-Turn Lane Present 0 1 2 3 4 3ST (2Ã2) 29 23 6 0 0 N/A 3SG (2Ã2) 35 16 15 4 0 N/A 4ST (2Ã2) 11 11 0 0 0 0 4SG (2Ã2) 127 64 13 37 6 7 4SG (1Ã2) 69 33 32 4 0 0 NOTE: 3ST 2Ã2= three-leg stop control intersections with two-way/two-way operations; 3SG 2Ã2= three-leg signal control intersections with two-way/two-way operations; 4ST 2Ã2 = four-leg stop control intersections with two-way/two-way operations; 4SG 2Ã2 = four-leg signal control intersections with two-way/two-way operations; 4SG 1Ã2 = four-leg signal control intersections with one-way/two-way operations combined; N/A = Not applicable. TableÂ 69. Number of approaches with left-turn lanes present by intersection configuration. Intersection Traffic Control Type and Configuration Total Number of Intersections Number of Approaches with a Right-Turn Lane Present 0 1 2 3 4 3ST (2Ã2) 29 27 2 0 0 N/A 3SG (2Ã2) 35 22 10 3 0 N/A 4ST (2Ã2) 11 10 1 0 0 0 4SG (2Ã2) 127 91 29 2 1 0 4SG (1Ã2) 69 50 17 2 0 0 NOTE: 3ST 2Ã2= three-leg stop control intersections with two-way/two-way operations; 3SG 2Ã2= three-leg signal control intersections with two-way/two-way operations; 4ST 2Ã2 = four-leg stop control intersections with two-way/two-way operations; 4SG 2Ã2 = four-leg signal control intersections with two-way/two-way operations; 4SG 1Ã2 = four-leg signal control intersections with one-way/two-way operations combined; N/A = Not applicable. TableÂ 70. Number of approaches with right-turn lanes present by intersection configuration.

134 Pedestrian and Bicycle Safety Performance Functions Intersection Traffic Control Type and Configuration Total Number of Intersections Number of Intersection Legs with a Parking Lane Present 0 1 2 3 4 3ST (2Ã2) 29 25 2 2 0 N/A 3SG (2Ã2) 35 19 1 9 6 N/A 4ST (2Ã2) 11 7 1 3 0 0 4SG (2Ã2) 127 37 17 33 10 30 4SG (1Ã2) 69 21 10 23 10 5 NOTE: 3ST 2Ã2= three-leg stop control intersections with two-way/two-way operations; 3SG 2Ã2= three-leg signal control intersections with two-way/two-way operations; 4ST 2Ã2 = four-leg stop control intersections with two-way/two-way operations; 4SG 2Ã2 = four-leg signal control intersections with two-way/two-way operations; N/A = Not applicable. 4SG 1Ã2 = four-leg signal control intersections with one-way/two-way operations combined; TableÂ 71. Number of intersection legs with marked parking lanes present by intersection configuration. Intersection Traffic Control Type and Configuration Total Number of Intersections Number of Intersection Legs with a Median Present 0 1 2 3 4 3ST (2Ã2) 29 22 4 3 0 N/A 3SG (2Ã2) 35 21 11 3 0 N/A 4ST (2Ã2) 11 11 0 0 0 0 4SG (2Ã2) 127 108 6 8 3 2 4SG (1Ã2) 69 45 17 3 4 0 NOTE: 3ST 2Ã2= three-leg stop control intersections with two-way/two-way operations; 3SG 2Ã2= three-leg signal control intersections with two-way/two-way operations; 4ST 2Ã2 = four-leg stop control intersections with two-way/two-way operations; 4SG 2Ã2 = four-leg signal control intersections with two-way/two-way operations; N/A = Not applicable. 4SG 1Ã2 = four-leg signal control intersections with one-way/two-way operations combined; TableÂ 72. Number of intersection legs with a median present by intersection configuration. Intersection Traffic Control Type and Configuration Total Number of Intersections Approach Median Width (ft) Meana Min. Max. Median Std. Dev. 3ST (2Ã2) 29 Major 13.06 7.20 35.00 11.00 7.86 Minor N/A N/A N/A N/A N/A 3SG (2Ã2) 35 Major 9.44 4.30 12.50 9.00 2.58 Minor 4.65 3.30 6.00 4.65 1.91 4ST (2Ã2) 11 Major N/A N/A N/A N/A N/A Minor N/A N/A N/A N/A N/A 4SG (2Ã2) 127 Major 9.96 3.00 42.00 6.25 8.93 Minor 10.00 5.25 18.00 10.38 4.09 4SG (1Ã2) 69 Major 9.21 5.90 12.00 9.50 2.29 Minor 15.13 5.98 42.25 12.00 9.78 NOTE: 3ST 2Ã2= three-leg stop control intersections with two-way/two-way operations; 3SG 2Ã2= three-leg signal control intersections with two-way/two-way operations; 4ST 2Ã2 = four-leg stop control intersections with two-way/two-way operations; 4SG 2Ã2 = four-leg signal control intersections with two-way/two-way operations; N/A = Not applicable. aValues of 0 ft median width are not included in the calculation of the mean. 4SG 1Ã2 = four-leg signal control intersections with one-way/two-way operations combined; TableÂ 73. Median width by major and minor roads by intersection configuration.

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 135Â Â Intersection Traffic Control Type and Configuration Total Number of Intersections Number of Approaches with a Bike Lane or Buffered Bike Lane Entering the Intersection 0 1 2 3 4 3ST (2Ã2) 29 23 2 4 0 N/A 3SG (2Ã2) 35 23 3 7 2 N/A 4ST (2Ã2) 11 5 0 6 0 0 4SG (2Ã2) 127 70 8 40 6 3 4SG (1Ã2) 69 19 28 12 9 1 NOTE: 3ST 2Ã2= three-leg stop control intersections with two-way/two-way operations; 3SG 2Ã2= three-leg signal control intersections with two-way/two-way operations; 4ST 2Ã2 = four-leg stop control intersections with two-way/two-way operations; 4SG 2Ã2 = four-leg signal control intersections with two-way/two-way operations; N/A = Not applicable. 4SG 1Ã2 = four-leg signal control intersections with one-way/two-way operations combined; TableÂ 74. Number of approaches with a bike lane or buffered bike lane entering the intersection by intersection configuration. Intersection Traffic Control Type and Configuration Total Number of Intersections Approach Crossing Width Curb-to-Curb (ft) Mean Min. Max. Median Std. Dev. 3ST (2Ã2) 29 Major 41.82 17.90 97.00 40.25 17.32 Minor 34.59 18.10 60.00 32.30 12.97 3SG (2Ã2) 35 Major 45.39 25.00 88.00 43.75 13.70 Minor 40.15 19.40 61.50 42.60 11.88 4ST (2Ã2) 11 Major 44.10 27.00 63.00 46.00 9.63 Minor 36.65 22.00 60.00 38.00 8.66 4SG (2Ã2) 127 Major 50.03 24.60 133.25 45.00 15.00 Minor 43.42 22.00 122.00 41.00 12.31 4SG (1Ã2) 69 Major 52.07 17.80 116.25 52.75 12.27 Minor 49.59 22.00 115.25 48.00 15.01 NOTE: 3ST 2Ã2= three-leg stop control intersections with two-way/two-way operations; 3SG 2Ã2= three-leg signal control intersections with two-way/two-way operations; 4ST 2Ã2 = four-leg stop control intersections with two-way/two-way operations; 4SG 2Ã2 = four-leg signal control intersections with two-way/two-way operations; 4SG 1Ã2 = four-leg signal control intersections with one-way/two-way operations combined. TableÂ 75. Crossing width curb-to-curb by major and minor roads by intersection configuration. Intersection Traffic Control Type and Configuration Total Number of Intersections Approach Crossing Distance Excluding Median Width (ft) Mean Min. Max. Median Std. Dev. 3ST (2Ã2) 29 Major 45.23 22.00 75.00 47.00 14.77 Minor 47.30 29.00 90.00 50.00 13.21 3SG (2Ã2) 35 Major 51.12 28.50 98.00 50.15 14.45 Minor 52.85 25.40 119.20 52.60 17.89 4ST (2Ã2) 11 Major 45.95 32.00 56.00 45.50 6.93 Minor 48.33 37.00 70.00 45.00 8.62 4SG (2Ã2) 127 Major 54.33 29.00 120.00 51.00 14.90 Minor 48.58 23.00 92.48 46.13 11.27 4SG (1Ã2) 69 Major 55.38 26.00 122.00 55.38 11.84 Minor 51.98 25.80 92.00 50.51 12.65 NOTE: 3ST 2Ã2= three-leg stop control intersections with two-way/two-way operations; 3SG 2Ã2= three-leg signal control intersections with two-way/two-way operations; 4ST 2Ã2 = four-leg stop control intersections with two-way/two-way operations; 4SG 2Ã2 = four-leg signal control intersections with two-way/two-way operations; 4SG 1Ã2 = four-leg signal control intersections with one-way/two-way operations combined. TableÂ 76. Crossing distance by major and minor roads by intersection configuration.

136 Pedestrian and Bicycle Safety Performance Functions TableÂ 78 provides information on the number of lanes that a pedestrian must cross at an intersection by major- and minor-road approaches, taking into consideration the presence of a median refuge. Both through lanes and turning lanes that are crossed by a pedestrian along a crossing path are considered. If the crossing path is broken by a median that provides a suitable refuge for the pedestrian so that the crossing may be accomplished in stages, then the number of lanes crossed in each stage is considered separately. TableÂ 79 provides information on the number of approaches with protected or protected/ permissive left-turn signal phasing. TableÂ 80 provides information on right-turn-on-red prohibitions provided along each approach by major and minor roads. Intersection Traffic Control Type and Configuration Total Number of Intersections Approach Number of Lanes Crossed Mean Min. Max. Median Std. Dev. 3ST (2Ã2) 29 Major 2.88 2.00 6.00 2.00 1.06 Minor 2.20 2.00 4.00 2.00 0.48 3SG (2Ã2) 35 Major 2.97 2.00 6.00 3.00 1.12 Minor 2.60 2.00 5.00 2.00 0.81 4ST (2Ã2) 11 Major 2.59 2.00 4.00 2.00 0.85 Minor 2.05 2.00 3.00 2.00 0.22 4SG (2Ã2) 127 Major 3.10 2.00 7.00 3.00 1.17 Minor 2.53 2.00 8.00 2.00 0.94 4SG (1Ã2) 69 Major 3.19 1.00 8.00 3.00 1.04 Minor 2.82 1.00 6.00 3.00 1.10 NOTE: 3ST 2Ã2= three-leg stop control intersections with two-way/two-way operations; 3SG 2Ã2= three-leg signal control intersections with two-way/two-way operations; 4ST 2Ã2 = four-leg stop control intersections with two-way/two-way operations; 4SG 2Ã2 = four-leg signal control intersections with two-way/two-way operations; 4SG 1Ã2 = four-leg signal control intersections with one-way/two-way operations combined. TableÂ 77. Number of lanes crossed by a pedestrian on major and minor roads by intersection configuration. Intersection Traffic Control Type and Configuration Total Number of Intersections Approach Number of Lanes Crossed Considering Median Refuge Mean Min. Max. Median Std. Dev. 3ST (2Ã2) 29 Major 2.60 1.00 4.00 2.00 0.86 Minor 2.13 1.00 4.00 2.00 0.51 3SG (2Ã2) 35 Major 2.74 1.00 5.00 2.00 1.10 Minor 2.51 2.00 5.00 2.00 0.82 4ST (2Ã2) 11 Major 2.59 2.00 4.00 2.00 0.85 Minor 2.00 1.00 3.00 2.00 0.32 4SG (2Ã2) 127 Major 2.89 1.00 5.00 3.00 0.97 Minor 2.44 1.00 5.00 2.00 0.75 4SG (1Ã2) 69 Major 3.05 1.00 6.00 3.00 1.01 Minor 2.64 1.00 5.00 2.50 0.97 NOTE: 3ST 2Ã2= three-leg stop control intersections with two-way/two-way operations; 3SG 2Ã2= three-leg signal control intersections with two-way/two-way operations; 4ST 2Ã2 = four-leg stop control intersections with two-way/two-way operations; 4SG 2Ã2 = four-leg signal control intersections with two-way/two-way operations; 4SG 1Ã2 = four-leg signal control intersections with one-way/two-way operations combined. TableÂ 78. Number of lanes crossed considering a median refuge by major and minor roads by intersection configuration.

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 137Â Â 3.2.2.2 Intersections: Traffic Volumes Traffic volume data were available for up to 13Â years (from 2006 to 2018) for study sites in Minneapolis and up to 6Â years (from 2013 to 2018) for study sites in Philadelphia. TableÂ 81 shows the breakdown of major-road traffic volume (AADTmaj), the minor-road traffic volume (AADTmin), and the sum of major- and minor-road traffic volumes (AADTtotal = AADTmaj + AADTmin) by intersection type. The study period (date range), number of sites and site-years, and basic traffic volume statistics are shown by city and combined across cities for each inter- section type. 3.2.2.3 Intersections: Pedestrian and Bicyclist Volumes For intersections, pedestrian and bicycle volumes were estimated based on the direct demand models developed for roadway segments (see SectionÂ 3.2.1.3). Using the direct demand models, pedestrian volumes (AADP) and bicycle volumes (AADB) were estimated for each leg of the intersection. From these estimates, daily pedestrian and bicycle volumes along the major roads (i.e., AADPmaj and AADBmaj) and minor roads (i.e., AADPmin and AADBmin) were estimated as well as the total number of pedestrians and bicyclists crossing each leg of the intersection (i.e., AADPcrossing = AADPmaj + AADPmin and AADBcrossing = AADBmaj + AADBmin). TableÂ 82 shows the breakdown of pedestrian volumes by major and minor roads and the total number of pedestrians crossing each leg of the intersections. TableÂ 83 shows the corresponding bicycle volumes. Intersection Traffic Control Type and Configuration Total Number of Intersections Number of Approaches with Protected or Protected/Permissive Left-Turn Signal Phasing 0 1 2 3 4 3ST (2Ã2) 29 N/A N/A N/A N/A N/A 3SG (2Ã2) 35 22 12 1 0 N/A 4ST (2Ã2) 11 N/A N/A N/A N/A N/A 4SG (2Ã2) 127 84 5 23 2 13 4SG (1Ã2) 69 43 24 2 0 0 NOTE: 3ST 2Ã2= three-leg stop control intersections with two-way/two-way operations; 3SG 2Ã2= three-leg signal control intersections with two-way/two-way operations; 4ST 2Ã2 = four-leg stop control intersections with two-way/two-way operations; 4SG 2Ã2 = four-leg signal control intersections with two-way/two-way operations; N/A = Not applicable. 4SG 1Ã2 = four-leg signal control intersections with one-way/two-way operations combined; TableÂ 79. Left-turn signal phasing by intersection configuration. Intersection Traffic Control Type and Configuration Total Number of Intersections Number of Approaches with Right-Turn-on-Red Prohibition 0 1 2 3 4 3ST (2Ã2) 29 N/A N/A N/A N/A N/A 3SG (2Ã2) 35 27 3 5 0 N/A 4ST (2Ã2) 11 N/A N/A N/A N/A N/A 4SG (2Ã2) 127 115 0 1 2 9 4SG (1Ã2) 69 60 5 4 0 0 NOTE: 3ST 2Ã2= three-leg stop control intersections with two-way/two-way operations; 3SG 2Ã2= three-leg signal control intersections with two-way/two-way operations; 4ST 2Ã2 = four-leg stop control intersections with two-way/two-way operations; 4SG 2Ã2 = four-leg signal control intersections with two-way/two-way operations; N/A = Not applicable. 4SG 1Ã2 = four-leg signal control intersections with one-way/two-way operations combined; TableÂ 80. Right-turn-on-red prohibitions by intersection configuration.

AADTmaj (veh/day) AADTmin (veh/day) AADTtotal (veh/day) City Date Range Number of Sites Number of Site-Years Min. Max. Mean Median Min. Max. Mean Median Min. Max. Mean Median Three-Leg Stop Control Intersections with Two-Way/Two-Way Operations Minneapolis 2006â2018 14 73 4,150 12,550 8,552 8,933 866 3,859 2,055 1,452 5,750 15,075 10,166 10,071 Philadelphia 2013â2018 15 85 1,246 20,945 8,707 8,147 1,246 20,945 7,733 4,972 2,491 41,889 16,441 12,576 Combined 2006â2018 29 158 1,246 20,945 8,632 8,700 866 20,945 5,331 2,525 2,491 41,889 13,412 10,700 Three-Leg Signal Control Intersections with Two-Way/Two-Way Operations Minneapolis 2006â2018 15 86 3,977 18,700 12,610 12,200 500 9,079 4,175 4,000 10,019 26,836 16,785 15,500 Philadelphia 2013â2018 20 117 3,057 29,521 10,572 9,077 2,774 29,521 7,947 4,778 6,114 59,043 18,519 14,848 Combined 2006â2018 35 203 3,057 29,521 11,446 9,629 500 29,521 6,330 4,500 6,114 59,043 17,776 14,905 Four-Leg Stop Control Intersections with Two-Way/Two-Way Operations Minneapolis 2006â2018 10 42 2,525 12,240 8,280 9,200 495 9,200 2,133 1,027 3,675 18,400 10,414 10,068 Philadelphia 2013â2018 1 3 2,132 21,32 2,132 2,132 2,132 2,132 2,132 2,132 4,263 4,263 4,263 4,263 Combined 2006â2018 11 45 2,132 12,240 7,721 9,200 495 9,200 2,133 1,100 3,675 18,400 9,855 9,920 Four-Leg Signal Control Intersections with Two-Way/Two-Way Operations Minneapolis 2006â2018 84 441 3,200 32,000 13,435 13,000 1,000 18,700 7,037 7,100 5,600 45,700 20,472 20,344 Philadelphia 2013â2018 43 204 3,552 19,356 9,988 10,053 1,703 14,817 5,890 5,633 6,486 33,039 15,878 14,600 Combined 2006â2018 127 645 3,200 32,000 12,268 11,400 1,000 18,700 6,648 6,214 5,600 45,700 18,916 17,500 Four-Leg Signal Control Intersections with One-Way/Two-Way Operations Minneapolis 2006â2018 58 259 5,600 33,500 14,663 12,738 2,000 21,300 8,573 7,900 8,150 44,000 23,236 22,206 Philadelphia 2013â2018 11 54 4,698 14,053 9,982 9,635 3,538 12,784 6,212 4,556 8,469 26,161 16,194 15,855 Combined 2006â2018 69 313 4,698 33,500 13,916 12,680 2,000 21,300 8,197 7,444 8,150 44,000 22,113 21,100 TableÂ 81. Major- and minor-road AADTs and total AADT by intersection configuration.

AADPmaj (ped/day) AADPmin (ped/day) AADPcrossing (ped/day) City Date Range Number of Sites Number of Site-Years Min. Max. Mean Median Min. Max. Mean Median Min. Max. Mean Median Three-Leg Stop Control Intersections with Two-Way/Two-Way Operations Minneapolis 2006â2018 14 73 116 549 231 171 79 523 182 130 201 1,036 414 298 Philadelphia 2013â2018 15 85 41 447 120 89 38 206 86 64 79 652 206 147 Combined 2006â2018 29 158 41 549 174 135 38 523 132 107 79 1,036 306 241 Three-Leg Signal Control Intersections with Two-Way/Two-Way Operations Minneapolis 2006â2018 15 86 154 3,319 799 337 127 2,615 684 278 306 5,305 1,483 588 Philadelphia 2013â2018 20 117 64 566 286 289 56 477 222 183 136 1,042 507 445 Combined 2006â2018 35 203 64 3,319 506 322 56 2,615 420 208 136 5,305 926 539 Four-Leg Stop Control Intersections with Two-Way/Two-Way Operations Minneapolis 2006â2018 10 42 103 441 209 205 91 405 170 113 209 847 379 315 Philadelphia 2013â2018 1 3 146 146 146 146 141 141 141 141 287 287 287 287 Combined 2006â2018 11 45 103 441 203 196 91 405 167 114 209 847 370 310 Four-Leg Signal Control Intersections with Two-Way/Two-Way Operations Minneapolis 2006â2018 84 441 72 3,752 405 263 69 3,766 334 201 143 7,518 739 473 Philadelphia 2013â2018 43 204 144 710 314 279 88 718 281 251 236 1,428 596 540 Combined 2006â2018 127 645 72 3,752 374 270 69 3,766 316 208 143 7,518 691 487 Four-Leg Signal Control Intersections with One-Way/Two-Way Operations Minneapolis 2006â2018 58 259 82 8,641 1,691 623 106 11,981 1,790 730 188 17,253 3,482 1,269 Philadelphia 2013â2018 11 54 250 8,640 1,966 500 130 11,981 2,491 382 380 20,621 4,457 875 Combined 2006â2018 69 313 82 8,641 1,735 571 106 11,981 1,902 492 188 20,621 3,637 1,013 TableÂ 82. Major- and minor-road AADPs and total number of pedestrians crossing at the intersection (AADPcrossing) by intersection configuration.

AADBmaj (bike/day) AADBmin (bike/day) AADBcrossing (bike/day) City Date Range Number of Sites Number of Site-Years Min. Max. Mean Median Min. Max. Mean Median Min. Max. Mean Median Three-Leg Stop Control Intersections with Two-Way/Two-Way Operations Minneapolis 2006â2018 14 73 80 550 247 165 78 430 184 164 159 980 431 345 Philadelphia 2013â2018 15 85 49 189 111 98 69 119 95 95 143 277 206 197 Combined 2006â2018 29 158 49 550 177 130 69 430 138 100 143 980 315 248 Three-Leg Signal Control Intersections with Two-Way/Two-Way Operations Minneapolis 2006â2018 15 86 118 865 437 399 115 478 246 236 232 1,343 682 665 Philadelphia 2013â2018 20 117 104 566 212 153 75 161 121 121 203 708 333 279 Combined 2006â2018 35 203 104 865 308 207 75 478 174 140 203 1,343 483 347 Four-Leg Stop Control Intersections with Two-Way/Two-Way Operations Minneapolis 2006â2018 10 42 91 929 301 263 90 485 205 135 181 1,413 506 398 Philadelphia 2013â2018 1 3 146 146 146 146 96 96 96 96 242 242 242 242 Combined 2006â2018 11 45 91 929 287 262 90 485 195 130 181 1,413 482 390 Four-Leg Signal Control Intersections with Two-Way/Two-Way Operations Minneapolis 2006â2018 84 441 53 693 228 182 58 793 214 155 124 1,480 441 339 Philadelphia 2013â2018 43 204 119 718 269 231 83 659 167 146 231 1,123 436 402 Combined 2006â2018 127 645 53 718 242 189 58 793 198 151 124 1,480 440 357 Four-Leg Signal Control Intersections with One-Way/Two-Way Operations Minneapolis 2006â2018 58 259 94 1,053 418 370 87 1,253 415 360 181 1,965 833 823 Philadelphia 2013â2018 11 54 182 1,142 534 500 182 890 408 328 364 2,032 941 812 Combined 2006â2018 69 313 94 1,142 437 397 87 1,253 413 347 181 2,032 850 812 TableÂ 83. Major- and minor-road AADBs and total number of bicycles crossing at the intersection (AADBcrossing) by intersection configuration.

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 141Â Â 3.2.2.4 Intersections: Crash Data Crash data were available for up to 13Â years (from 2006 to 2018) for study sites in Minneapolis and up to 6Â years (from 2013 to 2018) for study sites in Philadelphia. TableÂ 84 shows the breakdown of pedestrian and bicycle crash data by city and combined across both cities by severity and intersection type. Pedestrian and bicycle crashes were assigned to intersections if the crash was located within 75Â ft of the center of the intersection, and the crash was designated as intersection or intersection-related. This 75-ft intersection influence area for pedestrian and bicycle crashes is more consistent with the approach used in the usRAP methodology. In the usRAP methodology, pedestrian and bicycle crashes must occur at the intersection to be assigned to the intersection. Otherwise, the pedestrian and bicycle crashes are assigned to the roadway segment. A 75-ft intersection influence zone for pedestrian and bicycle crashes was based on inter- section data from Minneapolis. The average crossing distance was about 50Â ft across all inter- section types where data were collected in Minneapolis, but some of the larger intersections had crossing distances of around 100Â ft. By selecting a 75-ft intersection influence zone, pedestrian and bicycle crashes extending approximately 25Â ft beyond the boundary of the intersection proper where crosswalks would likely be located would be assigned to the intersection (see FigureÂ 12). Regarding severity, crash frequencies are provided for all severity levels combined (Total) and for fatal and suspected serious (FS) injury crashes. As indicated previously, some of the crashes in the crash dataset were classified as property-damage-only crashes. By definition, it is difficult to understand how a pedestrian or bicycle crash could be classified as property damage only, but recognizing that some pedestrian and bicycle crashes are classified as such, these crashes were included in the count of total crashes (i.e., all severity levels combined). 3.3 Safety Performance Functions: Model Development Negative binomial regression was used to develop the roadway segment and intersection SPFs in this study to be consistent with the models developed in the first edition of the HSM. The negative binomial model estimates relationships between the predicted number of crashes City Date Range Number of Sites Total Site-Years Pedestrian Crashes Bicycle Crashes Total FS Total FS Three-Leg Stop Control Intersections with Two-Way/Two-Way Operations Minneapolis 2006â2018 14 73 2 1 1 0 Philadelphia 2013â2018 15 85 0 0 3 0 Combined 2006â2018 29 158 2 1 4 0 Three-Leg Signal Control Intersections with Two-Way/Two-Way Operations Minneapolis 2006â2018 15 86 6 2 7 0 Philadelphia 2013â2018 20 117 15 1 5 0 Combined 2006â2018 35 203 21 3 12 0 Four-Leg Stop Control Intersections with Two-Way/Two-Way Operations Minneapolis 2006â2018 10 42 2 0 4 0 Philadelphia 2013â2018 1 3 0 0 0 0 Combined 2006â2018 11 45 2 0 4 0 Four-Leg Signal Control Intersections with Two-Way/Two-Way Operations Minneapolis 2006â2018 84 441 110 10 57 3 Philadelphia 2013â2018 43 204 96 4 19 1 Combined 2006â2018 127 645 206 14 76 4 Four-Leg Signal Control Intersections with One-Way/Two-Way Operations Minneapolis 2006â2018 58 259 74 6 28 2 Philadelphia 2013â2018 11 54 19 0 9 0 Combined 2006â2018 69 313 93 6 37 2 NOTE: FS = fatal and suspected serious injury. TableÂ 84. Pedestrian and bicycle crash frequency by city, severity, and intersection type.

142 Pedestrian and Bicycle Safety Performance Functions per year as a function of one or more explanatory variables. This is a common approach to modeling roadway segment and intersection crash frequencies (e.g., Miaou 1994; Shankar, Mannering, and Barfield 1995; Poch and Mannering 1996; El-Basyouny and Sayed 2006) because it accounts for the overdispersion that is often observed in crash data. Overdispersion results from the variance exceeding the mean in the crash frequency distribution. The general func- tional form of the negative binomial regression model is: (3-6)ln Xi i im b f= + where: Î»i = predicted number of crashes on roadway segment or intersection i, Î² = vector of estimable regression parameters, Xi = vector of geometric design, traffic volume, and other site-specific data, and Îµi = gamma-distributed error term. The mean-variance relationship for the negative binomial distribution is: (3-7)E y E yVar y 1i ii a= +` ` `j j j: D where: Var (yi) = variance of observed crashes y occurring on roadway segment or intersection i, E(yi) = predicted crash frequency on roadway segment or intersection i, and Î± = overdispersion parameter. The appropriateness of the negative binomial regression model is based on the significance of the overdispersion parameter. When Î± is not significantly different from zero, the negative binomial model reduces to the Poisson model. For all the models that were estimated, the esti- mate of Î± is reported to verify the appropriateness of the negative binomial approach. The method of maximum likelihood is used to estimate the model parameters. This method estimates model parameters by selecting those that maximize a likelihood function that describes FigureÂ 12. Seventy-five-ft intersection influence zone for pedestrian and bicycle crashes.

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 143Â Â the underlying statistical distribution assumed for the regression model. The likelihood function for the negative binomial model that was used in this study is shown in Equation 3-8. (3-8) !y y L i i i N i i i i 1 i m i i i m i i m m C C = + + + i = yJ L K K J L K K` ` ` N P O O N P O Oj j j% where: N = total number of roadway segments or intersections in the sample, Î = gamma function, and Î¸ = 1/Î±. To apply the negative binomial regression models estimated in this study, the following functional form was used for roadway segments: (3-9)e L AADT NMVol ei Xn n0 1 2 3 4 4# # # #m = gb b b b b b+ +X` j where: Î»i = predicted number of crashes on roadway segment i, e = exponential function, Î²0 = regression coefficient for constant, L = roadway segment length (mi), AADT = annual average daily traffic (veh/day), NMVol = annual average nonmotorized traffic volume (i.e., AADP or AADB) (ped/day or bike/day), Î²1 = regression coefficient for segment length. Î²2 = regression coefficient for AADT, Î²3 = regression coefficient for nonmotorized volume (i.e., AADP or AADB), Î²4, . . . , Î²n = regression coefficients for explanatory variables, i = 3, . . . , n, and X4, Xn = vector of geometric design, traffic volume, and other site-specific data. The following functional forms were considered for the intersection SPFs: (3-10),e AADT AADT NMVol NMVol ei maj maj Xmin min , ,nm nm n n0 1 2 1 2 3 3# # # # #m = gb b b b b b b+ +X ` ` ` ` `j j j j j or (3-11)e AADT NMVol ei total Xcrossing T NM n n0 3 3# # #m = gb b b b b+ +X ` ` `j j j where: Î»i = predicted number of pedestrian or bicycle crashes at intersection i, e = exponential function, Î²0 = regression coefficient for constant, AADTmaj = annual average daily traffic volume for the major road (veh/day), AADTmin = annual average daily traffic volume for the minor road (veh/day), AADTtotal = sum of annual average daily traffic volumes for the major and minor roads (= AADTmaj + AADTmin) (veh/day), NMVolmaj = annual average daily nonmotorized volume (i.e., AADP or AADB) for major roadway (ped/day or bike/day), NMVolmin = annual average daily nonmotorized volume (i.e., AADP or AADB) for minor roadway (ped/day or bike/day),

144 Pedestrian and Bicycle Safety Performance Functions NMVolcrossing = sum of daily nonmotorized volume crossing all intersection legs (ped/day or bike/day), Î²1, Î²2 = regression coefficients for major and minor road AADT, respectively, Î²1,nm, Î²2,nm = regression coefficients for major- and minor-road nonmotorized volume (i.e., AADP or AADB), respectively, Î²T = regression coefficient for total entering-intersection traffic volume (veh/day), Î²NM = regression coefficient for total entering-intersection nonmotorized volume (ped/day or bike/day), Î²3, . . . ,Î²n = regression coefficients for explanatory variables, i = 3, . . . , n, and X3, . . . , Xn = vector of geometric design and other site-specific data. For roadway segments, consideration was given to modeling pedestrian and bicycle crashes by particular crash types. For example, different pedestrian crash types (e.g., walking along roadway, dash/dart out, etc.) were considered for modeling purposes, but available crash datasets yielded insufficient and/or conflicting details to confidently distinguish between various pedestrian crash types; so in the end, all pedestrian crashes assigned to a site were included together in the predictive models for roadway segments. Similarly, consideration was given to modeling bicycle crashes that occurred along the roadway and driveway-related crashes separately. However, available crash datasets yielded insufficient and/or conflicting details to confidently distinguish between these two crash types; in the end, all bicycle crashes assigned to a site were included together in the predictive models for roadway segments. The same was true for modeling intersection crashes. Consideration was given to modeling different crash types at intersections (e.g., crashes involving through movements or left-turning movements by motor vehicle traffic), but again, available crash datasets yielded insufficient and/or conflicting details to confidently distinguish between specific crash types at inter sections, so all crashes assigned to an intersection were included together in the predictive models for intersections. Due to sample size issues, data for several roadway segments and intersection types were combined for modeling purposes. For roadways, data for four-lane undivided roads and four- lane divided roads were combined for modeling purposes; data for one-lane, one-way roads; two-lane, one-way roads; and three-lane, one-way roads were also combined for modeling purposes. For intersections, data for three-leg stop control intersections with two-way/two- way operations and four-leg stop control intersections with two-way/two-way operations were combined for modeling purposes; data for three-leg signal control intersections with two-way/ two-way operations and four-leg signal control intersections with one-way/two-way operations were combined for modeling purposes. These roadway segments and intersections were combined for modeling purposes as they were perceived to have the most similar operational characteristics among the site types being evaluated. For both roadway segment and intersection models, initial models were developed consider- ing only motor vehicle volumes (i.e., AADT) and nonmotorized volumes (i.e., AADP or AADB) (i.e., reduced models), and then additional models were developed considering motor vehicle volumes (i.e., AADT), nonmotorized volumes (i.e., AADP or AADB), and other relevant site characteristics (i.e., expanded models) as potential significant predictors of crashes. It was antic- ipated that the reduced models would primarily be considered for potential use in network screening, and the expanded models would be considered for potential use in the predictive methods for urban and suburban arterials. Models were developed for all severity level crashes combined (i.e., total) and for fatal and suspected serious (FS) injury crashes, when feasible. All models were estimated in R (i.e., a free software environment for statistical computing and graphics) using the GLM package. A forward-selection process was used to estimate the

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 145Â Â SPFs. In this method, each potential independent variable considered was tested to examine its impact on the dependent variable and the most impactful variable selected. Nonlinear trans- formations were considered for the continuous independent variables when such transfor- mation improved the overall model fit and to identify the best functional form for each of the potential independent variables. Then, all other variables were sequentially added to determine the best pair. This process was repeated until the addition of no other variables could significantly improve the overall model fit. Categorical variables were combined so that a sufficient number of observations were included within each group in the models. Indicator variables for the state (Pennsylvania versus Minnesota) were also included to account for unobserved regional differences in safety performance (e.g., reporting differences, environment, etc.). Typically, a validation procedure would be performed in which just a subset of the data was used to develop a preliminary model so that the modelâs effectiveness could be quantified using the remaining data. However, the low sample sizes (both in terms of the number of segments/ intersections and crashes) made this impossible. Instead, the research team randomly selected some of the data to develop the model and repeated this process multiple times to confirm that parameter estimates were relatively stable. 3.4 Analysis Results This section presents the final pedestrian and bicycle SPFs developed for urban and suburban roadway segments and intersections that included pedestrian and bicycle exposure data for potential consideration in HSM2. SectionÂ 3.4.1 presents the final models for roadway segments, and SectionÂ 3.4.2 presents the final models for intersections. 3.4.1 Roadway Segment Models Pedestrian and bicycle SPFs were developed for three types of urban and suburban roads: â¢ Two-lane undivided roads (2U). â¢ Four-lane undivided and divided roads (4U and 4D). â¢ One-way roads (OW) (including one-lane, two-lane, and three-lane roads). As indicated previously, data for four-lane undivided roads and four-lane divided roads were combined for modeling purposes. An indicator variable for median type (undivided or divided) was used in the models to account for differences between the two types of roads. Similarly, data for one-lane, one-way roads; two-lane, one-way roads; and three-lane, one-way roads were combined for modeling purposes, and an indicator variable for number of lanes was used in the models to account for differences in the number of lanes. Pedestrian SPFs are presented in Section 3.4.1.1, and bicycle SPFs are presented in SectionÂ 3.4.1.2 3.4.1.1 Pedestrian Models The final pedestrian SPFs for estimating pedestrian crashes along a roadway, by road type, are shown first in tabular form presenting conventional statistical output results, followed by the same models converted to a form more suitable for inclusion in the HSM. For each road type, three levels of models are presented as appropriate: â¢ A reduced model to estimate total pedestrian crashes (i.e., all severity levels combined) that primarily includes exposure measures for motor vehicles (i.e., AADT) and pedestrians (i.e., AADP). â¢ An expanded model to estimate total pedestrian crashes (i.e., all severity levels combined) that includes exposure measures for motor vehicles (i.e., AADT) and pedestrians (i.e., AADP)

146 Pedestrian and Bicycle Safety Performance Functions as well as other geometric and site characteristic features found to be significant predictors of total pedestrian crashes. â¢ A reduced model to estimate FS pedestrian crashes (i.e., fatal and suspected serious injury crashes) that primarily includes exposure measures for motor vehicles (i.e., AADT) and pedestrians (i.e., AADP). Comparisons of the expanded models with existing models in the HSM PartÂ C illus- trate potential differences in future editions of the HSM (if the models from this research are integrated into the HSM) and potential compatibility issues with existing HSM models. 3.4.1.1.1 Pedestrian Models (Conventional Statistical Output) All models were estimated in R using the GLM package. For all roadway segment models, the number of years and segment length were entered as offset variables (in log form) so that their effect is proportional on predicted crash frequency. TableÂ 85 presents a reduced model for total pedestrian crashes (i.e., all severity levels combined) for two-lane undivided roads (2U) that primarily includes exposure variables. The table shows the model coefficients, their standard error, and associated p-values. The table also shows the inverse of the overdispersion parameter and its standard error, the 2 x log-likelihood at convergence, the total number of crashes, and the total number of sites associated with the model. In R, the output from the GLM package provides the inverse of the traditional over- dispersion parameter. Therefore, TableÂ 85 and the other tables in SectionÂ 3.4.1.1.1 provide the inverse of the overdispersion and standard error for the overdispersion inverse, consistent with model outputs from R. In this reduced model and other models presented below, an indicator variable is included to account for differences between sites in Minneapolis and Philadelphia. FigureÂ 13 graphically presents the SPF shown in TableÂ 85 for various motor vehicle (AADT) and pedestrian (AADP) volumes. TableÂ 86 presents an expanded model for total pedestrian crashes (i.e., all severity levels combined) for two-lane undivided roads (2U). The table shows the model coefficients, their standard error, and associated p-values; the inverse of the overdispersion parameter and its stan- dard error; the 2 x log-likelihood at convergence; the total number of crashes; and the total number of sites associated with the model. In addition to exposure measures for motor vehicles and pedestrians, significant predictors of total pedestrian crashes on two-lane undivided roads include: â¢ Presence/absence of a sidewalk buffer greater than 0Â ft. â¢ Lane width. â¢ Number of bus/transit stops within 1,000Â ft. FigureÂ 14 graphically presents the SPF shown in TableÂ 86 for various motor vehicle (AADT) and pedestrian (AADP) volumes. Variable Coefficient Std. Error p-Value Constant â5.214 1.137 < 0.001 Natural log of AADT volume (veh/day) 0.327 0.117 0.005 Natural log of AADP volume (ped/day) 0.224 0.087 0.010 Indicator for roadway segment within Philadelphia 0.563 0.116 < 0.001 Inverse of overdispersion parameter 0.789 0.162 â 2 x log-likelihood at convergence â2,214.687 Total number of crashes 413 Total number of sites 2,027 NOTE: â = Not applicable. TableÂ 85. SPF for total pedestrian crashes on two-lane undivided roads (reduced model).

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 147Â Â FigureÂ 13. Graphical representation of the SPF for predicted average total pedestrian crashes per year on two-lane undivided roads (reduced model). Variable Coefficient Std. Error p-Value Constant â4.0290 1.190 <0.001 Natural log of AADT volume (veh/day) 0.3470 0.119 0.004 Natural log of AADP volume (ped/day) 0.1140 0.091 0.206 Indicator for sidewalk buffer greater than 0 ft â0.6650 0.116 < 0.001 Average lane width (ft) â0.0510 0.015 <0.001 Number of bus/transit stops within 1,000 ft 0.0178 0.006 0.004 Indicator for roadway segment within Philadelphia 0.5000 0.120 <0.001 Inverse of overdispersion parameter 1.0550 0.255 â 2 x log-likelihood at convergence â2,162.878 Total number of crashes 413 Total number of sites 2,027 NOTE: â = Not applicable. TableÂ 86. SPF for total pedestrian crashes on two-lane undivided roads (expanded model). No model to predict FS pedestrian crashes (i.e., fatal and suspected serious injury crashes) for two-lane undivided roads is provided because either the model did not converge or the coefficient(s) for one or more of the exposure measures (i.e., AADT or AADP) was counter- intuitive or not statistically significant. TableÂ 87 presents a reduced model for total pedestrian crashes (i.e., all severity levels com- bined) for four-lane undivided and divided roads. The table shows the model coefficients, their standard error, and associated p-values. The table also shows the inverse of the overdispersion parameter and its standard error, the 2 x log-likelihood at convergence, the total number of crashes, and the total number of sites associated with the model. In this reduced model and the other models for four-lane undivided and divided roads, an indicator variable is included to account for the presence/absence of a median. FigureÂ 15 graphically presents the SPF shown in TableÂ 87 for various motor vehicle (AADT) and pedestrian (AADP) volumes.

148 Pedestrian and Bicycle Safety Performance Functions TableÂ 88 presents an expanded model for total pedestrian crashes (i.e., all severity levels com- bined) for four-lane undivided and divided roads. The table shows the model coefficients, their standard error, and associated p-values. The table also shows the inverse of the overdispersion parameter and its standard error, the 2 x log-likelihood at convergence, the total number of crashes, and the total number of sites associated with the model. In addition to exposure mea- sures for motor vehicles and pedestrians and the presence/absence of a median, the number of schools within 1,000Â ft is a significant predictor of total pedestrian crashes on four-lane undivided and divided roads. FigureÂ 16 graphically presents the SPF shown in TableÂ 88 for various motor vehicle (AADT) and pedestrian (AADP) volumes. TableÂ 89 presents a reduced model for FS pedestrian crashes (i.e., fatal and suspected serious injury crashes) for four-lane undivided and divided roads (4U and 4D). The table shows the model coefficients, their standard error, and associated p-values; the inverse of the overdisper- sion parameter and its standard error; the 2 x log-likelihood at convergence; the total number of FigureÂ 14. Graphical representation of the SPF for predicted average total pedestrian crashes per year on two-lane undivided roads (expanded model). Variable Coefficient Std. Error p-Value Constant â11.154 3.235 < 0.001 Natural log of AADT volume (veh/day) 0.902 0.319 0.005 Natural log of AADP volume (ped/day) 0.236 0.132 0.073 Indicator for a divided roadway â1.343 0.389 < 0.001 Indicator for roadway segment within Philadelphia 0.222 0.383 0.562 Inverse of overdispersion parameter 0.539 0.175 â 2 x log-likelihood at convergence â540.279 Total number of crashes 113 Total number of sites 550 NOTE: â = Not applicable. TableÂ 87. SPF for total pedestrian crashes on four-lane undivided and divided roads (reduced model).

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 149Â Â FigureÂ 15. Graphical representation of the SPF for predicted average total pedestrian crashes per year on four-lane undivided (4U) and divided (4D) roads (reduced model). Variable Coefficient Std. Error p-Value Constant â9.826 3.287 0.003 Natural log of AADT volume (veh/day) 0.787 0.324 0.015 Natural log of AADP volume (ped/day) 0.160 0.136 0.238 Indicator for a divided roadway â1.283 0.389 < 0.001 Number of schools within 1,000 ft 0.160 0.067 0.017 Indicator for roadway segment within Philadelphia 0.293 0.380 0.441 Inverse of overdispersion parameter 0.581 0.192 â 2 x log-likelihood at convergence â535.138 Total number of crashes 113 Total number of sites 550 NOTE: â = Not applicable. TableÂ 88. SPF for total pedestrian crashes on four-lane undivided and divided roads (expanded model). crashes; and the total number of sites associated with the model. FigureÂ 17 graphically presents the SPF shown in TableÂ 89 for various motor vehicle (AADT) and pedestrian (AADP) volumes. TableÂ 90 presents a reduced model for total pedestrian crashes (i.e., all severity levels combined) for one-way roads. The table shows the model coefficients, their standard error, and associated p-values. The table also shows the inverse of the overdispersion parameter and its standard error, the 2 x log-likelihood at convergence, the total number of crashes, and the total number of sites associated with the model. In this reduced model and the other models for one-way roads, an indicator variable is included to account for the number of lanes. FigureÂ 18 graphically presents the SPF shown in TableÂ 90 for various motor vehicle (AADT) and pedestrian (AADP) volumes. TableÂ 91 presents an expanded model for total pedestrian crashes (i.e., all severity levels combined) for one-way roads (OW). The table shows the model coefficients, their standard error,

150 Pedestrian and Bicycle Safety Performance Functions and associated p-values; the inverse of the overdispersion parameter and its standard error; the 2 x log-likelihood at convergence; the total number of crashes; and the total number of sites associated with the model. In addition to exposure measures for motor vehicles and pedestrians and indicator variables for the number of lanes, significant predictors of total pedes- trian crashes on one-way roads include: â¢ Presence/absence of a sidewalk buffer greater than 0Â ft. â¢ Lane width. â¢ Number of alcohol sales establishments within 1,000Â ft. FigureÂ 19 graphically presents the SPF shown in TableÂ 91 for various motor vehicle (AADT) and pedestrian (AADP) volumes. No model to predict FS pedestrian crashes (i.e., fatal and suspected serious injury crashes) for one-way roads is provided because either the model did not converge or the coefficient(s) for one or more of the exposure measures (i.e., AADT or AADP) was counterintuitive or not statistically significant. FigureÂ 16. Graphical representation of the SPF for predicted average total pedestrian crashes per year on four-lane undivided (4U) and divided (4D) roads (expanded model). Variable Coefficient Std. Error p-Value Constant â26.576 5.900 < 0.001 Natural log of AADT volume (veh/day) 2.125 0.569 < 0.001 Natural log of AADP volume (ped/day) 0.638 0.198 0.001 Indicator for a divided roadway â0.380 0.556 0.495 Indicator for roadway segment within Philadelphia â0.421 0.772 0.585 Inverse of overdispersion parameter 0.419 0.248 â 2 x log-likelihood at convergence â231.484 Total number of crashes 39 Total number of sites 550 NOTE: â = Not applicable. TableÂ 89. SPF for FS pedestrian crashes on four-lane undivided and divided roads (reduced model).

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 151Â Â FigureÂ 17. Graphical representation of the SPF for predicted average FS pedestrian crashes per year on four-lane undivided and divided roads (reduced model). 3.4.1.1.2 Pedestrian Models (Form Suitable for Integration into the HSM) This section presents the pedestrian SPFs from TableÂ 85 through TableÂ 91 converted to a form more suitable for integration within the HSM. In HSM Part C, the components of the predictive models generally consist of an SPF, one or more adjustment factors, and a calibration factor. Using existing HSM notation and variable definitions as appropriate, Equation 3-12 and Equation 3-13 present the predictive models for roadway segments from the HSM Part C, ChapterÂ 12 on urban and suburban arterials but slightly rearranged to properly integrate the new pedestrian and bicycle SPFs into the equations. (3-12)N+N= + pedrN Npredicted rs br r biker#C` j (3-13)= AF AF AFNbr spf rs r r yr1 2# # # #fN ` j where: Npredicted rs = predicted average crash frequency for an individual roadway for the selected year (crashes/year), Nbr = predicted average crash frequency of an individual roadway segment (excluding pedestrian and bicycle crashes) (crashes/year), Cr = calibration factor for roadway segment type r to adjust prediction to local conditions, Npedr = predicted average crash frequency of pedestrian crashes for an individual roadway segment (crashes/year), Nbiker = predicted average crash frequency of bicycle crashes for an individual road- way segment (crashes/year), Nspf rs = predicted total average crash frequency of an individual roadway segment for base conditions (excluding pedestrian and bicycle crashes) (crashes/year), and AF1r . . . AFyr = adjustment factors specific to roadway segment type r and specific geometric design and traffic control features y.

152 Pedestrian and Bicycle Safety Performance Functions Variable Coefficient Std. Error p-Value Constant â9.339 2.160 <0.001 Natural log of AADT volume (veh/day) 0.897 0.229 <0.001 Natural log of AADP volume (ped/day) 0.207 0.104 0.047 Indicator for sidewalk buffer greater than 0 ft â0.569 0.228 0.013 Average lane width (ft) â0.071 0.032 0.028 Indicator for two-lane roadway â0.484 0.297 0.102 Indicator for three-lane roadway â0.804 0.344 0.019 Number of alcohol sales establishments within 1,000 ft 0.013 0.009 0.126 Indicator for roadway segment within Philadelphia 0.452 0.225 0.045 Inverse of overdispersion parameter 0.709 0.263 â 2 x log-likelihood at convergence â748.164 Total number of crashes 129 Total number of sites 982 NOTE: â = Not applicable. TableÂ 91. SPF for total pedestrian crashes on one-way roads (expanded model). Variable Coefficient Std. Error p-Value Constant â10.651 2.021 < 0.001 Natural log of AADT volume (veh/day) 0.829 0.224 < 0.001 Natural log of AADP volume (ped/day) 0.337 0.087 < 0.001 Indicator for two-lane roadway â0.475 0.282 0.092 Indicator for three-lane roadway â0.646 0.337 0.055 Indicator for roadway segment within Philadelphia 0.549 0.221 0.013 Inverse of overdispersion parameter 0.661 0.245 â 2 x log-likelihood at convergence â759.995 Total number of crashes 129 Total number of sites 982 NOTE: â = Not applicable. TableÂ 90. SPF for total pedestrian crashes on one-way roads (reduced model). FigureÂ 18. Graphical representation of the SPF for predicted average total pedestrian crashes per year on one-way roads (reduced model).

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 153Â Â FigureÂ 19. Graphical representation of the SPF for predicted average total pedestrian crashes per year on one-way roads (expanded model). Each component of Equation 3-12 (i.e., Nbr, Npedr, and Nbiker) is estimated separately to obtain the estimated total crash frequency for an individual roadway segment, and then a calibration factor is applied. A comprehensive model for estimating the pedestrian component [i.e., pedes- trian crashes (Npedr)] on roadway segments would take the following form: (3-14)C# #N= ...N AF AF, ,pedr pedbase rs p r p r r1 2#` j where: Npedr = predicted average crash frequency of pedestrian crashes for an individual roadway segment (crashes/year), Npedbase rs = predicted average crash frequency of pedestrian crashes of an individual roadway segment for base conditions of the SPF developed for site type r (crashes/year), AF1p,r . . . AFnp,r = adjustment factors for pedestrian crashes specific to SPF for site type r, and Cr = calibration factor to adjust SPF for local conditions for site type r. The SPF for pedestrian crashes on roadway segments is in the form of (3-15)= exp ln ln lnN a b AADT c AADP Lpedbase rs # #+ + +b ` ` `j j jl where: AADT = annual average daily traffic (AADT) volume on the segment (veh/day), AADP = annual average daily pedestrian (AADP) volume on the segment (ped/day), L = length of roadway segment (mi), and a, b, c = regression coefficients.

154 Pedestrian and Bicycle Safety Performance Functions The coefficients of a, b, and c; the overdispersion parameters; and the base conditions for the SPFs to predict total pedestrian crashes and FS pedestrian crashes on roadway segments are provided in TableÂ 92 and TableÂ 93, respectively. In TableÂ 92 and TableÂ 93, the combined models for four-lane undivided and divided roads from TableÂ 87, TableÂ 88, and TableÂ 89 are provided as separate models for 4U and 4D. The adjustment factors for geometric design features of road segments applicable to the pedestrian SPFs in TableÂ 92 are as follows. 3.4.1.1.2.1 AFs for Total Pedestrian Crashes on One-Way Roads (Reduced Model) AF1p,r 2 Number of Lanes Adjustment factors applicable to the reduced model for total pedestrian crashes on one-way roads, accounting for the number of lanes, are presented in TableÂ 94. Road Type Intercept AADT AADP Overdispersion Parameter Reduced models 2U â5.214 0.327 0.224 1.267 4U â11.154 0.902 0.236 1.855 4D â12.497 0.902 0.236 1.855 OW â10.651 0.829 0.337 1.513 Expanded models 2U â4.029 0.347 0.114 0.948 4U â9.826 0.787 0.160 1.721 4D â11.109 0.787 0.160 1.721 OW â9.339 0.897 0.207 1.410 Base conditions for reduced models: 2U No base conditions. 4U No base conditions. 4D No base conditions. OW One lane. Base conditions for expanded models: 2U Sidewalk buffer is 0 ft, average lane width is 12 ft, no bus/transit stops within 1,000 ft. 4U No schools within 1,000 ft. 4D No schools within 1,000 ft. OW One lane, sidewalk buffer is 0 ft, average lane width is 12 ft, no alcohol sales establishments within 1,000 ft. NOTE: 2U = two-lane undivided; 4U = four-lane undivided; 4D = four-lane divided; OW = one-lane, one-way. TableÂ 92. SPF coefficients for total pedestrian crashes on roadway segments. 4U â26.576 2.125 0.638 2.387 4D â26.956 2.125 0.638 2.387 Base conditions: 4U No base conditions. 4D No base conditions. NOTE: 4U = four-lane undivided; 4D = four-lane divided. TableÂ 93. SPF coefficients for FS pedestrian crashes on roadway segments.

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 155Â Â 3.4.1.1.2.2 AFs for Total Pedestrian Crashes on Two-Lane Undivided Roads (Expanded Model) AF1p,r 2 Presence of Sidewalk Buffer Adjustment factors applicable to the expanded model for total pedestrian crashes on two-lane undivided roads, accounting for the presence of a sidewalk buffer, are presented in TableÂ 95. AF2p,r 2 Average Lane Width The adjustment factor applicable to the expanded model for total pedestrian crashes on 2U roads, accounting for the average lane width, is calculated as follows: (3-16).AF LW 0 0512 1exp,p r2 #= - -b` j l where: LW = average lane width (ft). AF3p,r 2 Number of Bus/Transit Stops The adjustment factor applicable to the expanded model for total pedestrian crashes on two-lane undivided roads, accounting for the number of bus/transit stops located near the roadway segment is calculated as follows: (3-17).BSAF 0 0178exp,p r3 #= ` j where: BS = number of bus/transit stops within 1,000Â ft of the center of the roadway segment. 3.4.1.1.2.3 AFs for Total Pedestrian Crashes on Four-Lane Undivided and Four-Lane Divided Roads (Expanded Model) AF1p,r 2 Number of Schools The adjustment factor applicable to the expanded model for total pedestrian crashes on four-lane undivided and divided roads, accounting for the number of schools located near the roadway segment, is calculated as follows: (3-18).SAF 0 160exp,p r1 #= ` j where: S = number of schools within 1,000Â ft of the center of the roadway segment. Number of Lanes One lane 1 Two lanes 0.622 Three lanes 0.524 TableÂ 94. Adjustment factors for number of lanes (total pedestrian crashes on one-way roads â reduced model). Sidewalk Buffer None 1 Greater than 0 ft 0.514 TableÂ 95. Adjustment factors for presence of a sidewalk buffer (total pedestrian crashes on two-lane undivided roads â expanded model).

156 Pedestrian and Bicycle Safety Performance Functions 3.4.1.1.2.4 AFs for Total Pedestrian Crashes on One-Way Roads (Expanded Model) AF1p,r 2 Number of Lanes Adjustment factors applicable to the expanded model for total pedestrian crashes on one-way roads, accounting for the number of lanes, are presented in TableÂ 96. AF2p,r 2 Presence of Sidewalk Buffer Adjustment factors applicable to the expanded model for total pedestrian crashes on one-way roads, accounting for the presence of a sidewalk buffer, are presented in TableÂ 97. AF3p,r 2 Average Lane Width The adjustment factor applicable to the expanded model for total pedestrian crashes on one- way roads, accounting for the average lane width is calculated as follows: (3-19).AF LW 012 071exp,p r4 #= - -b` j l where: LW = average lane width (ft). AF4p,r 2 Number of Alcohol Sales Establishments The adjustment factor applicable to the expanded model for total pedestrian crashes on one- way roads, accounting for the number of alcohol sales establishments located near the roadway segment is calculated as follows: (3-20).ASEAF 0 013exp,p r5 #= ` j where: ASE = number of alcohol sales establishments within 1,000Â ft of the center of the roadway segment. 3.4.1.1.3 Pedestrian Models (Comparison with Existing HSM Models) Following development of the pedestrian SPFs for urban roadway segments, compatibility testing of the new models was conducted to check the reasonableness of the results. FigureÂ 13 through FigureÂ 19 provide some sense of the reasonableness of the new models for application and potential incorporation in the HSM; however, to gain a better understanding of the poten- tial use of these models within the HSM, output results from the new models were compared to output from existing models in HSM Part C ChapterÂ 12 (Predictive Method for Urban and Number of Lanes One lane 1 Two lanes 0.616 Three lanes 0.448 TableÂ 96. Adjustment factors for number of lanes (total pedestrian crashes on one-way roads â expanded model). Sidewalk Buffer None 1 Greater than 0 ft 0.566 TableÂ 97. Adjustment factors for presence of a sidewalk buffer (total pedestrian crashes on one-way roads â expanded model).

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 157Â Â Suburban Arterials). Note, for this comparison, the new models and the existing HSM models were not calibrated using a single agencyâs data; and unlike most of the existing HSM models, pedes- trian exposure is accounted for in the new models so some differences are expected. Nonetheless, output results from the new models were compared to output from existing models in the HSM to gain a sense of the reasonableness of the new models for application and potential incor- poration into HSM2. Comparisons were made between the expanded models developed as part of this research and HSM Part C models, as the expanded models and accompanying adjustment factors are most applicable for potential integration into HSM Part C. FigureÂ 20 provides a comparison of the predicted average total pedestrian crash frequency per year from the expanded two-lane undivided roads model and the predicted average total pedes- trian crash frequency and total crashes (i.e., multiple-vehicle + single-vehicle + pedestrian + bicycle crashes) from the existing HSM Part C model for urban two-lane, undivided roads. The predicted average crash frequencies are for the base conditions for both models. Comparisons with the expanded model are shown for four pedestrian volume levels: 50, 100, 300, and 800Â ped/day. Based on the comparison, it looks like the new model predicts significantly higher pedestrian crash frequencies compared to the existing HSM model. If the expanded model is integrated into the HSM Part Câs predictive chapter for urban and suburban arterials, not only would this likely result in higher average predicted pedestrian crash frequencies for urban two-lane undivided roads, but pedestrian crashes would also likely become a higher percentage or proportion of the average predicted total crashes for this type of roadway. The default values in the HSM show that pedestrian crashes are about 0.5Â percent of the total crashes on urban two-lane undivided roads. There is some concern with the expanded model at lower AADTs (i.e., less than 2,000 or 3,000Â veh/day). The predicted average pedestrian crash frequency may be higher than the predicted average total crash frequency, including nondriveway-related multiple-vehicle, single-vehicle, and bicycle crashes combined. This would not be a logical result. FigureÂ 20. Comparison of SPF for predicted average total pedestrian crashes per year on two-lane undivided roads (expanded model) and the existing HSM model.

158 Pedestrian and Bicycle Safety Performance Functions Based on this assessment, the expanded model for total pedestrian crashes (i.e., all severity levels combined) on two-lane undivided roads may or may not be compatible with the existing HSM Part C model for urban and suburban two-lane undivided roads at lower AADTs. It is difficult to determine at this time without both models being calibrated together. Calibration of the models will likely be critical. One possibility may be that the pedestrian model be recommended for use with two-lane undivided roads (2U) above a certain AADT (e.g., 6,000Â veh/day or more). FigureÂ 21 provides a comparison of the predicted average total pedestrian crash fre- quency from the expanded four-lane undivided roads model and the predicted average total pedestrian crash frequency and total crashes (i.e., multiple-vehicle + single-vehicle + pedestrian + bicycle crashes) from the existing HSM PartÂ C model for urban four-lane undivided roads. The predicted average crash frequencies are for the base conditions for both models. Comparisons with the expanded model are shown for three pedestrian volume levels: 50, 300, and 800Â ped/day. For this road type, predicted average pedestrian crashes are similar for both models. Based on this assessment, the expanded model for total pedestrian crashes (i.e., all severity levels combined) on four-lane undivided roads appears compatible with the existing HSM PartÂ C model for urban and suburban four-lane undivided roads; however, calibration of the models will likely be critical. FigureÂ 22 provides a comparison of the predicted average total pedestrian crash frequency from the expanded four-lane divided roads model and the predicted average total pedestrian crash frequency and total crashes (i.e., multiple-vehicle + single-vehicle + pedestrian + bicycle crashes) from the existing HSM Part C model for urban four-lane, divided roads. The predicted average crash frequencies are for the base conditions for both models. Comparisons with the expanded model are shown for three pedestrian volume levels: 50, 300, and 800 ped/day. FigureÂ 21. Comparison of SPF for predicted average total pedestrian crashes per year on four-lane undivided roads (expanded model) and the existing HSM model.

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 159Â Â For this road type, predicted average pedestrian crashes are similar for both models. Based on this assessment, the expanded model for total pedestrian crashes (i.e., all severity levels combined) on four-lane divided roads appears compatible with the existing HSM PartÂ C model for urban and suburban four-lane divided roads; however, calibration of the models will likely be critical. There is no existing HSM model for urban and suburban one-way roads for comparison purposes with the expanded model for total pedestrian crashes (i.e., all severity levels com- bined) for one-way roads, although it is anticipated that HSM2 will include new models for this road type. Thus, no comparison is made here between the new expanded pedestrian model for one-way roads and an existing HSM model. 3.4.1.2 Bicycle Models The final bicycle SPFs for estimating bicycle crashes along a roadway by road type are shown first in tabular form presenting conventional statistical output results, followed by the same models converted to a form more suitable for inclusion in the HSM. For each road type, three levels of models are presented as appropriate: â¢ A reduced model to estimate total bicycle crashes (i.e., all severity levels combined) that primarily includes exposure measures for motor vehicles (i.e., AADT) and bicycles (i.e., AADB). â¢ An expanded model to estimate total bicycle crashes (i.e., all severity levels combined) that includes exposure measures for motor vehicles (i.e., AADT) and bicycles (i.e., AADB) as well as other geometric and site characteristic features found to be significant predictors of total bicycle crashes. â¢ A reduced model to estimate FS bicycle crashes (i.e., fatal and suspected serious injury crashes) that primarily includes exposure measures for motor vehicles (i.e., AADT) and bicycles (i.e., AADB). FigureÂ 22. Comparison of SPF for predicted average total pedestrian crashes per year on four-lane divided roads (expanded model) and existing HSM model.

160 Pedestrian and Bicycle Safety Performance Functions Comparisons of the expanded models with existing models in HSM PartÂ C illustrate poten- tial differences in future editions of the HSM if the models from this research are integrated into the HSM and potential compatibility issues with existing HSM models for motor vehicle crashes, excluding pedestrian and bicycle crashes. 3.4.1.2.1 Bicycle Models (Conventional Output) All models were estimated in R using the GLM package. For all roadway segment models, the number of years and segment length were entered as offset variables (in log form) so that their effect is proportional on predicted crash frequency. TableÂ 98 presents a reduced model for total bicycle crashes (i.e., all severity levels combined) for two-lane undivided roads that primarily includes exposure variables. The table shows the model coefficients, their standard error, and associated p-values. The table also shows the inverse of the overdispersion parameter and its standard error, the 2 x log-likelihood at convergence, the total number of crashes, and the total number of sites associated with the model. In R, the output from the GLM package provides the inverse of the traditional overdispersion parameter. There- fore, TableÂ 98 and the other tables in SectionÂ 3.4.1.2.1 provide the inverse of the overdispersion and standard error for the overdispersion inverse, consistent with model outputs from R. In this reduced model and other models presented below, an indicator variable is included to account for differences between sites in Minneapolis and Philadelphia. FigureÂ 23 graphically presents the SPF shown in TableÂ 98 for various motor vehicle (AADT) and bicycle (AADB) volumes. TableÂ 99 presents an expanded model for total bicycle crashes (i.e., all severity levels combined) for two-lane undivided roads. The table shows the model coefficients, their standard error, and associated p-values; the inverse of the overdispersion parameter and its standard error; the 2 x log-likelihood at convergence; the total number of crashes; and the total number of sites associated with the model. In addition to exposure measures for motor vehicles and bicycles, significant predictors of total bicycle crashes on two-lane undivided roads include: â¢ Presence/absence of a buffered bike lane. â¢ Lane width. â¢ Number of bus/transit stops within 1,000Â ft. â¢ Number of schools stops within 1,000Â ft. FigureÂ 24 graphically presents the SPF shown in TableÂ 99 for various motor vehicle (AADT) and bicycle (AADB) volumes. No model to predict FS bicycle crashes (i.e., fatal and suspected serious injury crashes) for two-lane undivided roads is provided because either the model did not converge or the coefficient(s) for one or more of the exposure measures (i.e., AADT or AADB) was counter- intuitive or not statistically significant. Variable Coefficient Std. Error p-Value Constant â9.647 2.121 < 0.001 Natural log of AADT volume (veh/day) 0.546 0.199 0.006 Natural log of AADB volume (bike/day) 0.493 0.150 0.001 Indicator for roadway segment within Pennsylvania 0.457 0.206 0.026 Inverse of overdispersion parameter 0.348 0.101 â 2 x log-likelihood at convergence â1,089.866 Total number of crashes 157 Total number of sites 2,027 NOTE: â = Not applicable. TableÂ 98. SPF for total bicycle crashes on two-lane undivided roads (reduced model).

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 161Â Â FigureÂ 23. Graphical representation of the SPF for predicted average total bicycle crashes per year on two-lane undivided roads (reduced model). Variable Coefficient Std. Error p-Value Constant â8.422 2.159 <0.001 Natural log of AADT volume (veh/day) 0.528 0.201 0.009 Natural log of AADB volume (bike/day) 0.359 0.153 0.019 Indicator for presence of a buffered bike lane on one or more sides â1.609 1.075 0.134 Average lane width â0.058 0.023 0.012 Number of bus/transit stops within 1,000 ft 0.028 0.010 0.006 Number of schools within 1,000 ft 0.151 0.084 0.072 Indicator for roadway segment within Pennsylvania 0.386 0.209 0.065 Inverse of overdispersion parameter 0.426 0.134 â 2 x log-likelihood at convergence â1,074.293 Total number of crashes 157 Total number of sites 2,027 NOTE: â = Not applicable. TableÂ 99. SPF for total bicycle crashes on two-lane undivided roads (expanded model). TableÂ 100 presents a reduced model for total bicycle crashes (i.e., all severity levels combined) for four-lane undivided and divided roads. The table shows the model coefficients, their standard error, and associated p-values. The table also shows the inverse of the overdispersion parameter and its standard error, the 2 x log-likelihood at convergence, the total number of crashes, and the total number of sites associated with the model. In this reduced model and the other models for four-lane undivided and divided roads, an indicator variable is included to account for the presence/absence of a median. FigureÂ 25 graphically presents the SPF shown in TableÂ 100 for various motor vehicle (AADT) and bicycle (AADB) volumes. TableÂ 101 presents an expanded model for total bicycle crashes (i.e., all severity levels com- bined) for four-lane undivided and divided roads. The table shows the model coefficients, their standard error, and associated p-values. The table also shows the inverse of the over- dispersion parameter and its standard error, the 2 x log-likelihood at convergence, the total

162 Pedestrian and Bicycle Safety Performance Functions FigureÂ 24. Graphical representation of the SPF for predicted average total bicycle crashes per year on two-lane undivided roads (expanded model). Variable Coefficient Std. Error p-Value Constant â14.935 3.148 < 0.001 Natural log of AADT volume (veh/day) 0.996 0.304 0.001 Natural log of AADB volume (bike/day) 0.691 0.233 0.003 Indicator for a divided roadway â0.761 0.319 0.017 Indicator for roadway segment within Pennsylvania 0.747 0.408 0.067 Inverse of overdispersion parameter 0.900 0.345 â 2 x log-likelihood at convergence â549.505 Total number of crashes 118 Total number of sites 550 NOTE: â = Not applicable. TableÂ 100. SPF for total bicycle crashes on four-lane undivided and divided roads (reduced model). number of crashes, and the total number of sites associated with the model. In addition to exposure measures for motor vehicles and bicycles and the presence/absence of a median, sig- nificant predictors of total bicycle crashes on four-lane undivided and divided roads include: â¢ Speed limit. â¢ Number of alcohol sales establishments within 1,000Â ft. FigureÂ 26 graphically presents the SPF shown in TableÂ 101 for various motor vehicle (AADT) and bicycle (AADB) volumes. No model to predict FS bicycle crashes (i.e., fatal and suspected serious injury crashes) for four-lane undivided and divided roads is provided because either the model did not converge or the coefficient(s) for one or more of the exposure measures (i.e., AADT or AADB) was counter- intuitive or not statistically significant.

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 163Â Â FigureÂ 25. Graphical representation of the SPF for predicted average total bicycle crashes per year on four-lane undivided and divided roads (reduced model). Variable Coefficient Std. Error p-Value Constant â13.864 3.187 < 0.001 Natural log of AADT volume (veh/day) 0.960 0.299 0.001 Natural log of AADB volume (bike/day) 0.607 0.272 0.026 Indicator for a divided roadway â0.760 0.327 0.020 Indicator for speed limit greater than 25 mph 0.384 0.443 0.385 Number of alcohol sales establishments within 1,000 ft 0.018 0.010 0.076 Indicator for roadway segment within Pennsylvania 0.516 0.458 0.260 Inverse of overdispersion parameter 0.951 0.373 â 2 x log-likelihood at convergence â545.780 Total number of crashes 118 Total number of sites 550 NOTE: â = Not applicable. TableÂ 101. SPF for total bicycle crashes on four-lane undivided and divided roads (expanded model). TableÂ 102 presents a reduced model for total bicycle crashes (i.e., all severity levels combined) for one-way roads. The table shows the model coefficients, their standard error, and associated p-values. The table also shows the inverse of the overdispersion parameter and its standard error, the 2 x log-likelihood at convergence, the total number of crashes, and the total number of sites associated with the model. In this reduced model and the other models for one-way roads, an indicator variable is included to account for the number of lanes. FigureÂ 27 graphically presents the SPF shown in TableÂ 102 for various motor vehicle (AADT) and bicycle (AADB) volumes. TableÂ 103 presents an expanded model for total bicycle crashes (i.e., all severity levels combined) for one-way roads. The table shows the model coefficients, their standard error, and associated

164 Pedestrian and Bicycle Safety Performance Functions FigureÂ 26. Graphical representation of the SPF for predicted average total bicycle crashes per year on four-lane undivided and divided roads (expanded model). Variable Coefficient Std. Error p-Value Constant â10.388 2.365 < 0.001 Natural log of AADT volume (veh/day) 0.446 0.241 0.064 Natural log of AADB volume (bike/day) 0.749 0.207 < 0.001 Indicator for two-lane roadway â0.691 0.314 0.028 Indicator for three-lane roadway â0.242 0.336 0.472 Indicator for roadway segment within Pennsylvania 0.595 0.262 0.023 Inverse of overdispersion parameter 540 5714 â 2 x log-likelihood at convergence â527.69 Total number of crashes 80 Total number of sites 982 NOTE: â = Not applicable. TableÂ 102. SPF for total bicycle crashes on one-way roads (reduced model). p-values; the inverse of the overdispersion parameter and its standard error; the 2 x log-likelihood at convergence; the total number of crashes; and the total number of sites associated with the model. In addition to exposure measures for motor vehicles and bicycles and an indicator variable for number of lanes, the number of alcohol sales establishments within 1,000Â ft of the center of the roadway segment is a significant predictor of total bicycle crashes on one- way roads. FigureÂ 28 graphically presents the SPF shown in TableÂ 103 for various motor vehicle (AADT) and bicycle (AADB) volumes. No model to predict FS bicycle crashes (i.e., fatal and suspected serious injury crashes) for one-way roads is provided because either the model did not converge or the coefficient(s) for

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 165Â Â one or more of the exposure measures (i.e., AADT or AADB) was counterintuitive or not statistically significant. 3.4.1.2.2 Bicycle Models (Form Suitable for Integration into the HSM) This section presents the bicycle SPFs from TableÂ 98 through TableÂ 103 converted to a form more suitable for integration within the HSM. In HSM Part C, the components of the predic- tive models generally consist of an SPF, one or more adjustment factors, and a calibration factor. Previously, Equation 3-12 and Equation 3-13 present the predictive models for roadway segments from the HSM urban and suburban arterial chapter (i.e., ChapterÂ 12), but slightly rearranged to properly integrate the new pedestrian and bicycle SPFs into the equations (see Section 3.4.1.1.2). FigureÂ 27. Graphical representation of the SPF for predicted average total bicycle crashes per year on one-way roads (reduced model). Variable Coefficient Std. Error p-Value Constant â9.363 2.441 <0.001 Natural log of AADT volume (veh/day) 0.444 0.248 0.074 Natural log of AADB volume (bike/day) 0.579 0.219 0.008 Indicator for two-lane roadway â0.913 0.331 0.006 Indicator for three-lane roadway â0.437 0.355 0.217 Number of alcohol sales establishments within 1,000 ft 0.023 0.008 0.003 Indicator for roadway segment within Pennsylvania 0.545 0.266 0.041 Inverse of overdispersion parameter 711 6455 â 2 x log-likelihood at convergence â519.509 Total number of crashes 80 Total number of sites 982 NOTE: â = Not applicable. TableÂ 103. SPF for total bicycle crashes on one-way roads (expanded model).

166 Pedestrian and Bicycle Safety Performance Functions Each component of Equation 3-12 (i.e., Nbr, Npedr, and Nbiker) is estimated separately to obtain the estimated total crash frequency for an individual roadway segment, and then a calibration factor is applied. A comprehensive model for estimating bicycle crashes on roadway segments would take the following form: (3-21)C# #N= fN AF AF, , ,rs b r b r nb r r1 2biker bikebase # AF` j where: Nbiker = predicted average crash frequency of bicycle crashes for an individual road- way segment (crashes/year), Nbikebase rs = predicted average crash frequency of bicycle crashes of an individual roadway segment for base conditions of the SPF developed for site type r (crashes/year), AF1b,r . . . AFnb,r = adjustment factors for bicycle crashes specific to SPF for site type r, and Cr = calibration factor to adjust SPF for local conditions for site type r. The SPF for bicycle crashes on segments is in the form of: (3-22)exp=N ln ln lna b AADT c AADB Lrsbikebase # #+ + +b ` ` `j j jl where: AADT = annual average daily traffic volume on the segment (veh/day), AADB = annual average daily bicycle volume on the segment (bike/day), L = length of roadway segment (mi), and a, b, c = regression coefficients. The coefficients of a, b, and c; the overdispersion parameters; and the base conditions for the SPFs to predict total bicycle crashes on roadway segments are provided in TableÂ 104. FigureÂ 28. Graphical representation of the SPF for predicted average total bicycle crashes per year on one-way roads (expanded model).

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 167Â Â In TableÂ 104, the combined models for four-lane undivided and divided roads (from TableÂ 100 and TableÂ 101) are provided as separate models for four-lane undivided and divided roads. The AFs for geometric design features of road segments applicable to the bicycle SPFs in TableÂ 104 are as follows. 3.4.1.2.2.1 AFs for Total Bicycle Crashes on One-Way Roads (Reduced Model) AF1b,r 2 Number of Lanes AFs applicable to the reduced model for total bicycle crashes on one-way roads, accounting for the number of lanes, are presented in TableÂ 105. 3.4.1.2.2.2 AFs for Total Bicycle Crashes on 2U Roads (Expanded Model) AF1b,r 2 Presence of Buffered Bike Lane AFs applicable to the expanded model for total bicycle crashes on two-way undivided roads, accounting for the presence of a buffered bike lane, are presented in TableÂ 106. Reduced models 2U â9.647 0.546 0.493 2.873 4U â14.935 0.996 0.691 1.111 4D â15.696 0.996 0.691 1.111 OW -10.388 0.446 0.749 0.002 Expanded models 2U â8.422 0.528 0.359 2.347 4U â13.864 0.960 0.607 1.052 4D â14.624 0.960 0.607 1.052 OW â9.363 0.444 0.579 0.001 Base conditions for reduced models: 2U No base conditions. 4U No base conditions. 4D No base conditions. OW One lane. Base conditions for expanded models: 2U No buffered bike lane on either side of the road, average lane width is 12 ft, no bus/transit stops within 1,000 ft, no schools within 1,000 ft. 4U Speed limit less than or equal to 25 mph, no alcohol sales establishments within 1,000 ft. 4D Speed limit less than or equal to 25 mph, no alcohol sales establishments within 1,000 ft. OW One lane, no alcohol sales establishments within 1,000 ft. NOTE: 2U = two-lane undivided; 4U = four-lane undivided; 4D = four-lane divided; OW = one-lane, one-way. TableÂ 104. SPF coefficients for total bicycle crashes on roadway segments. Number of Lanes One lane 1 Two lanes 0.501 Three lanes 0.785 TableÂ 105. Adjustment factors for number of lanes (total bicycle crashes on one-way roads â reduced model).

168 Pedestrian and Bicycle Safety Performance Functions AF2b,r 2 Average Lane Width The AF applicable to the expanded model for total bicycle crashes on two-way undivided roads, accounting for the average lane width is calculated as follows: (3-23).AF LW 0 05812exp,b r2 #= - -b` j l where: LW = average lane width (ft). AF3b,r 2 Number of Bus/Transit Stops The AF applicable to the expanded model for total bicycle crashes on two-lane undivided roads, accounting for the number of bus/transit stops located near the roadway segment is calculated as follows: (3-24).BSAF 0 0 82exp,b r3 #= ` j where: BS = number of bus/transit stops within 1,000Â ft of the center of the roadway segment. AF4b,r 2 Number of Schools The AF applicable to the expanded model for total bicycle crashes on 2U roads, accounting for the number of schools located near the roadway segment is calculated as follows: (3-25).SAF 0 151exp,b r4 #= ` j where: S = number of schools within 1,000Â ft of the center of the roadway segment. 3.4.1.2.2.3 AFs for Total Bicycle Crashes on Four-Lane Undivided and Four-Lane Divided Roads (Expanded Model) AF1b,r 2 Speed Limit AFs applicable to the expanded model for total bicycle crashes on four-lane divided and four-lane undivided roads, accounting for the speed limit along the road, are presented in TableÂ 107. Presence of Buffered Bike Lane None 1 Present on at least one side of the road 0.20 TableÂ 106. Adjustment factors for the presence of a buffered bike lane (total bicycle crashes on two-way undivided roads â expanded model). Speed Limit Less than or equal to 25 mph 1 Greater than 25 mph 1.468 TableÂ 107. Adjustment factors for speed limit (total bicycle crashes on four-land divided and four-laned undivided roads â expanded model).

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 169Â Â AF2b,r 2 Number of Alcohol Sales Establishments The AF applicable to the expanded model for total bicycle crashes on four-lane undivided and four-lane divided roads, accounting for the number of alcohol sales establishments located near the roadway segment is calculated as follows: (3-26).ASEAF 0 018exp,b r2 #= ` j where: ASE = number of alcohol sales establishments within 1,000Â ft of the center of the roadway segment. 3.4.1.2.2.4 AFs for Total Bicycle Crashes on One-Way Roads (Expanded Model) AF1b,r 2 Number of Lanes AFs applicable to the expanded model for total bicycle crashes on one-way roads, accounting for the number of lanes, are presented in TableÂ 108. AF2b,r 2 Number of Alcohol Sales Establishments The AF applicable to the expanded model for total bicycle crashes on one-way roads, accounting for the number of alcohol sales establishments located near the roadway segment, is calculated as follows: (3-27).ASEAF 0 023exp,b r2 #= ` j where: ASE = number of alcohol sales establishments within 1,000Â ft of the center of the roadway segment. 3.4.1.2.3 Bicycle Models (Comparison with Existing HSM Models) Following development of the bicycle SPFs for urban roadway segments, compatibility testing of the new models was conducted to check the reasonableness of the results. FigureÂ 23 through FigureÂ 28 provide some sense of the reasonableness of the new models for application and potential incorporation in the HSM; but, to gain a better understanding of the potential use of these models within the HSM, output results from the new models were compared to output from existing models in HSM Part C ChapterÂ 12 (Predictive Method for Urban and Suburban Arterials). Note, for this comparison, the new models and the existing HSM models were not calibrated using a single agencyâs data; and, unlike the existing HSM models, bicycle exposure is accounted for in the new models so some differences are expected. Nonetheless, output results from the new models were compared to output from existing models in the HSM to gain a sense of the reasonableness of the new models for application and potential incorporation in the HSM. Comparisons were made between the expanded models developed as part of this research and HSM Part C models, as the expanded models and accompanying adjustment factors are most applicable for potential integration into HSM PartÂ C. Number of Lanes One lane 1 Two lanes 0.401 Three lanes 0.646 TableÂ 108. Adjustment factors for number of lanes (total bicycle crashes on one-way roads â expanded model).

170 Pedestrian and Bicycle Safety Performance Functions FigureÂ 29 provides a comparison of the predicted average total bicycle crash frequency per year from the expanded two-lane undivided roads model and the predicted average total bicycle crash frequency and total crashes (i.e., multiple-vehicle + single-vehicle + pedestrian + bicycle crashes) from the existing HSM Part C model for urban two-lane undivided roads. The predicted average crash frequencies are for the base conditions for both models. Comparisons with the expanded model are shown for four bicycle volume levels: 50, 100, 300, and 800Â bike/day. Based on the comparison, it looks like the new model predicts a significantly greater number of bicycle crash frequencies compared to the existing HSM model. If the expanded model is integrated into the HSM PartÂ Câs predictive chapter for urban and suburban arterials, not only would it likely result in higher average predicted bicycle crash frequencies for urban two-lane undivided roads, but bicycle crashes would also likely become a higher percentage or pro- portion of the average predicted total crashes for this type of roadway. The default values in the HSM show that bicycle crashes are about 0.4Â percent of the total crashes on urban two-lane undivided roads. Based on this assessment, the expanded model for total bicycle crashes (i.e., all severity levels combined) on two-lane undivided roads may be compatible with the existing HSM Part C model for urban and suburban two-lane undivided roads, but calibration of the models will likely be critical. FigureÂ 30 provides a comparison of the predicted average total bicycle crash frequency from the expanded four-lane undivided roads model and the predicted average total bicycle crash frequency and total crashes (i.e., multiple-vehicle + single-vehicle + pedestrian + bicycle crashes) from the existing HSM Part C model for urban four-lane undivided roads. The predicted average crash frequencies are for the base conditions for both models. Comparisons with the expanded model are shown for three bicycle volume levels: 50, 300, and 800 bike/day. Similar to the comparison of models for two-lane undivided roads, it looks like the new model for four-lane undivided roads predicts significantly higher bicycle crash frequencies FigureÂ 29. Comparison of SPF for predicted average total bicycle crashes per year on two-lane undivided roads (expanded model) and existing HSM model.

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 171Â Â compared to the existing HSM model for roads with higher bicycle volumes. For roads with low bicycle volumes, the results of the two models are similar in magnitude. If this expanded model is integrated into the HSM Part Câs predictive chapter for urban and suburban arterials, not only would this likely result in higher average predicted bicycle crash frequencies for urban four-lane undivided roads with higher bicycle volumes, but bicycle crashes would also likely become a higher percentage or proportion of the average predicted total crashes for this type of roadway. The default values in the HSM show that bicycle crashes are about 0.2Â percent of the total crashes on urban four-lane undivided roads. Based on this assessment, the expanded model for total bicycle crashes (i.e., all severity levels combined) on four-lane undivided roads may be compatible with the existing HSM PartÂ C model for urban and suburban four-lane undivided roads, but calibration of the models will likely be critical. FigureÂ 31 provides a comparison of the predicted average total bicycle crash frequency from the expanded four-lane divided roads model and the predicted average total bicycle crash fre- quency and total crashes (i.e., multiple-vehicle + single-vehicle + pedestrian + bicycle crashes) from the existing HSM Part C model for urban four-lane divided roads. For roads with low bicycle volumes, the results of the two models are similar in magnitude; whereas, for roads with higher bicycle volumes, the expanded model predicts higher average bicycle crash frequencies than the existing HSM model. If this expanded model is integrated into the HSM Part Câs predic- tive chapter for urban and suburban arterials, not only would this likely result in higher average predicted bicycle crash frequencies for urban four-lane divided roads with higher bicycle volumes, but bicycle crashes would also likely become a higher percentage or proportion of the average predicted total crashes for this type of roadway. The default values in the HSM show that bicycle crashes are about 0.5Â percent of the total crashes on urban four-lane divided roads. Based on this assessment, the expanded model for total bicycle crashes (i.e., all severity levels combined) on four-lane divided roads may be compatible with the existing HSM PartÂ C model for urban and suburban four-lane divided roads, but calibration of the models will likely be critical. FigureÂ 30. Comparison of SPF for predicted average total bicycle crashes per year on four-lane undivided roads (expanded model) and existing HSM model.

172 Pedestrian and Bicycle Safety Performance Functions There is no existing HSM model for urban and suburban one-way roads for comparison purposes with the expanded model for total bicycle crashes (i.e., all severity levels combined) for one-way roads, although it is anticipated that HSM2 will include new models for this road type. Thus, no comparison is made here between the new expanded bicycle model for one-way roads and an existing HSM model. 3.4.2 Intersection Models Pedestrian and bicycle SPFs were developed for three types of urban and suburban intersections: â¢ Three-leg stop control intersections with two-way/two-way operations (3ST 2Ã2) and four-leg stop control intersections with two-way/two-way operations (4ST 2Ã2) combined. â¢ Three-leg signal control intersections with two-way/two-way operations (3SG 2Ã2) and four- leg signal control intersections with one-way/two-way operations (4SG 1Ã2) combined. â¢ Four-leg signal control intersections with two-way/two-way operations (4SG 2Ã2). As previously indicated, data for three-leg stop control intersections with two-way/two-way operations and four-leg stop control intersections with two-way/two-way operations were combined together for modeling purposes. An indicator variable for number of legs was used in the models to account for differences between three-leg and four-leg intersections. Similarly, data for three-leg signal control intersections with two-way/two-way operations and four-leg signal control intersections with one-way/two-way operations were combined together for modeling purposes; and again, an indicator variable for number of legs was used in the models to account for differences between three-leg and four-leg intersections. Pedestrian SPFs are presented in Section 3.4.2.1, and bicycle SPFs are presented in SectionÂ 3.4.2.2. 3.4.2.1 Pedestrian Models The final pedestrian SPFs for estimating pedestrian crashes at intersections by intersection type are shown first in tabular form presenting conventional statistical output results, followed FigureÂ 31. Comparison of SPF for predicted average total bicycle crashes per year on four-lane divided roads (expanded model) and existing HSM model.

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 173Â Â by the same models converted to a form more suitable for inclusion in the HSM. For each inter- section type, three levels of models are presented as appropriate: â¢ A reduced model to estimate total pedestrian crashes (i.e., all severity levels combined) that primarily includes exposure measures for motor vehicles (i.e., AADT) and pedestrians (i.e., AADP). â¢ An expanded model to estimate total pedestrian crashes (i.e., all severity levels combined) that includes exposure measures for motor vehicles (i.e., AADT) and pedestrians (i.e., AADP) as well as other geometric, traffic control, and site characteristic features found to be significant predictors of total pedestrian crashes. â¢ A reduced model to estimate FS pedestrian crashes (i.e., fatal and suspected serious injury crashes) that primarily includes exposure measures for motor vehicles (i.e., AADT) and pedestrians (i.e., AADP). Comparisons of the reduced and/or expanded models with existing models in the HSM PartÂ C illustrate potential differences in future editions of the HSM if the models from this research are integrated into the HSM and potential compatibility issues with existing HSM models for motor vehicle crashes, excluding pedestrian and bicycle crashes. 3.4.2.1.1 Pedestrian Models (Conventional Statistical Output) All models were estimated in R using the GLM package. For all intersection models, the number of years was entered as offset variables (in log form) so that the effect is proportional on predicted crash frequency. TableÂ 109 presents a reduced model for total pedestrian crashes (i.e., all severity levels com- bined) at three-leg and four-leg stop control intersections with two-way/two-way operations that primarily includes exposure variables. The table shows the model coefficients, their standard error, and associated p-values. The table also shows the inverse of the overdispersion parameter and its standard error, the 2 x log-likelihood at convergence, the total number of crashes, and the total number of sites associated with the model. In R, the output from the GLM package pro- vides the inverse of the traditional overdispersion parameter. Therefore, TableÂ 109 and the other tables in Section 3.4.2.1.1 provide the inverse of the overdispersion and standard error for the overdispersion inverse, consistent with model outputs from R. In this reduced model and other models presented below, an indicator variable is included to account for differences between sites in Minneapolis and Philadelphia. FigureÂ 32 graphically presents the SPF shown in TableÂ 109 for various motor vehicle entering (AADTtotal) and pedestrian crossing (AADPcrossing) volumes. No expanded model to predict total pedestrian crashes (i.e., all severity levels combined) at three-leg and four-leg stop control intersections to predict FS pedestrian crashes (i.e., fatal and Variable Coefficient Std. Error p-Value Constant â53.670 25.141 0.033 Natural log of daily entering-intersection traffic (AADTtotal) volume (veh/day) 4.293 2.582 0.096 Natural log of daily pedestrian volumes crossing all intersection legs (AADPcrossing) (ped/day) 1.655 1.079 0.125 Indicator for roadway segment within Pennsylvania â24.297 10,844.181 0.998 Inverse of overdispersion parameter 7,814 257,934 â 2 x log-likelihood at convergence â13.727 Total number of crashes 4 Total number of sites 37 NOTE: â = Not applicable. TableÂ 109. SPF for total pedestrian crashes at three-leg and four-leg stop control intersections with two-way/two-way operations (reduced model).

174 Pedestrian and Bicycle Safety Performance Functions suspected serious injury crashes) is provided because either the models did not converge or the coefficient(s) for one or more of the measures was counterintuitive or not statistically significant. TableÂ 110 presents a reduced model for total pedestrian crashes (i.e., all severity levels com- bined) at three-leg signal control intersections with two-way/two-way operations and four-leg signal control intersections with one-way/two-way operations that primarily includes exposure variables. The table shows the model coefficients, their standard error, and associated p-values. The table also shows the inverse of the overdispersion parameter and its standard error, the 2 x log-likelihood at convergence, the total number of crashes, and the total number of sites associ- ated with the model. In this reduced model, an indicator variable is included to account for the number of legs. FigureÂ 33 graphically presents the SPF shown in TableÂ 110 for various motor vehicle entering (AADTtotal) and pedestrian crossing (AADPcrossing) volumes. FigureÂ 32. Graphical representation of the SPF for predicted average total pedestrian crashes per year at three-leg and four-leg stop control intersections with two-way/two-way operations (reduced model). Variable Coefficient Std. Error p-Value Constant â12.750 2.986 < 0.001 Natural log of daily entering-intersection traffic (AADTtotal) volume (veh/day) 0.961 0.299 0.001 Natural log of daily pedestrian volumes crossing all intersection legs (AADPcrossing) (ped/day) 0.112 0.097 0.246 Indicator for four-leg intersection 0.999 0.329 0.002 Indicator for roadway segment within Pennsylvania 0.672 0.297 0.024 Inverse of overdispersion parameter 2.240 1.030 â 2 x log-likelihood at convergence â279.106 Total number of crashes 114 Total number of sites 104 NOTE: â = Not applicable. TableÂ 110. SPF for total pedestrian crashes at three-leg signal control intersections with two-way/two-way operations and four-leg signal control intersections with one-way/two-way operations (reduced model).

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 175Â Â No expanded model to predict total pedestrian crashes (i.e., all severity levels combined) or FS pedestrian crashes (i.e., fatal and suspected serious injury crashes) at three-leg signal con- trol intersections with two-way/two-way operations or four-leg signal control intersections with one-way/two-way operations is provided because either the models did not converge or the coefficient(s) for one or more of the measures was counterintuitive or not statistically significant. TableÂ 111 presents a reduced model for total pedestrian crashes (i.e., all severity levels com- bined) at four-leg signal control intersections with two-way/two-way operations that primar- ily includes exposure variables. The table shows the model coefficients, their standard error, and FigureÂ 33. Graphical representation of the SPF for predicted average total pedestrian crashes per year at three-leg signal control intersections with two-way/two-way operations and four-leg signal control intersections with one-way/two-way operations (reduced model). Variable Coefficient Std. Error p-Value Constant â19.085 2.964 < 0.001 Natural log of daily entering-intersection traffic (AADTtotal) volume (veh/day) 1.518 0.288 < 0.001 Natural log of daily pedestrian volumes crossing all intersection legs (AADPcrossing) (ped/day) 0.395 0.165 0.017 Indicator for roadway segment within Pennsylvania 1.201 0.223 < 0.001 Inverse of overdispersion parameter 1.924 0.579 â 2 x log-likelihood at convergence â395.059 Total number of crashes 206 Total number of sites 127 NOTE: â = Not applicable. TableÂ 111. SPF for total pedestrian crashes at four-leg signal control intersections with two-way/two-way operations (reduced model).

176 Pedestrian and Bicycle Safety Performance Functions associated p-values. e table also shows the inverse of the overdispersion parameter and its stan- dard error, the 2 x log-likelihood at convergence, the total number of crashes, and the total number of sites associated with the model. FigureÂ 34 graphically presents the SPF shown in TableÂ 111 for various motor vehicle entering (AADTtotal) and pedestrian crossing (AADPcrossing) volumes. TableÂ 112 presents an expanded model for total pedestrian crashes (i.e., all severity levels combined) at four-leg signal control intersections with two-way/two-way operations. e table shows the model coecients, their standard error, and associated p-values; the inverse of the FigureÂ 34. Graphical representation of the SPF for predicted average total pedestrian crashes per year at four-leg signal control intersections with two-way/two-way operations (reduced model). Variable Coefficient Std. Error p-Value Constant 19.9410 3.139 < 0.001 Natural log of daily entering-intersection traffic (AADTtotal) volume (veh/day) 1.6830 0.308 < 0.001 Natural log of daily pedestrian volumes crossing all intersection legs (AADPcrossing) (ped/day) 0.2680 0.186 0.151 Indicator for right-turn-on-red being prohibited at one or more intersection approach 0.1980 0.319 0.535 Indicator for some level of protected left-turning movements at all intersection approaches 0.5690 0.375 0.129 Number of alcohol sales establishments within 1,000 ft 0.0189 0.018 0.281 Indicator for roadway segment within Pennsylvania 1.2930 0.243 < 0.001 Inverse of overdispersion parameter 2.1680 0.696 â 2 x log-likelihood at convergence 390.810 Total number of crashes 206 Total number of sites 127 NOTE: â = Not applicable. TableÂ 112. SPF for total pedestrian crashes at four-leg signal control intersections with two-way/two-way operations (expanded model).

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 177Â Â overdispersion parameter and its standard error; the 2 x log-likelihood at convergence; the total number of crashes; and the total number of sites associated with the model. In addition to exposure measures for motor vehicles and pedestrians, significant predictors of total pedestrian crashes at four-leg signal control intersections with two-way/two-way operations include: â¢ Prohibition of right-turn-on-red. â¢ Protection of left-turn movements. â¢ Number of alcohol sales establishments within 1,000Â ft. Note, prohibiting right-turn-on-red has a relatively high p-value, indicating this parameter is not statistically significant. Nevertheless, it was kept in the final model since the parameter esti- mate makes sense and is in line with engineering expectations (e.g., eliminating right-turn-on- red should reduce pedestrian-vehicle interactions and thus reduce the potential for pedestrian crashes). One reason for the high p-value might be the small number of intersections with this feature in the dataset (13, or approximately 10Â percent). An improved sample size should verify these findings. FigureÂ 35 graphically presents the SPF shown in TableÂ 112 for various motor vehicle entering (AADTtotal) and pedestrian crossing (AADPcrossing) volumes. No expanded model to predict FS pedestrian crashes (i.e., fatal and suspected serious injury crashes) at four-leg signal control intersections with two-way/two-way operations is provided because either the model did not converge or the coefficient(s) for one or more of the measures was counterintuitive or not statistically significant. 3.4.2.1.2 Pedestrian Models (Form Suitable for Integration into the HSM) This section presents the pedestrian SPFs from TableÂ 109 through TableÂ 112 converted to a form more suitable for integration within the HSM. In HSM Part C, the components of the predictive models generally consist of an SPF, one or more adjustment factors, and a calibration FigureÂ 35. Graphical representation of the SPF for predicted average total pedestrian crashes per year at four-leg signal control intersections with two-way/two-way operations (expanded model).

178 Pedestrian and Bicycle Safety Performance Functions factor. Using existing HSM notation and variable definitions as appropriate, Equation 3-28 and Equation 3-29 present the predictive models for intersections from the HSM Part C, ChapterÂ 12 on urban and suburban arterials but slightly rearranged to properly integrate the new pedestrian and bicycle SPFs into the equations. (3-28)N+N= +C#N Npredicted bi i pedi bikeiint ` j (3-29)#N=N AF AF AFbi spf i i yi1 2int # # #f` j where: Npredicted int = predicted average crash frequency of an individual intersection for the selected year (crashes/year), Nbi = predicted average crash frequency of an individual intersection (excluding pedestrian and bicycle crashes) (crashes/year), Ci = calibration factor for intersection type i to adjust prediction to local conditions, Npedi = predicted average crash frequency of pedestrian crashes for an individual intersection (crashes/year), Nbikei = predicted average crash frequency of bicycle crashes for an individual inter- section (crashes/year), Nspf int = predicted total average crash frequency of an individual intersection for base conditions (excluding pedestrian and bicycle crashes) (crashes/year), and AF1i . . . AFyi = adjustment factors specific to intersection type i and specific geometric design and traffic control features y. Each component of Equation 3-28 (i.e., Nbi, Npedi, and Nbikei) is estimated separately to obtain the estimated total crash frequency for an individual intersection. A comprehensive model for estimating the pedestrian component [i.e., pedestrian crashes (Npedi)] at intersections would take the following form: (3-30)#N C#=N AF AF, ,pedi pedbase p i p i i1 2int # f` j where: Npedi = predicted average crash frequency of pedestrian crashes for an individual intersection (crashes/year), Npedbase int = predicted average crash frequency of pedestrian crashes for an individual inter- section for base conditions of the SPF developed for site type i (crashes/year), AF1p,i . . . AFnp,i = adjustment factors for pedestrian crashes specific to SPF for site type i, and Ci = calibration factor to adjust SPF for local conditions for site type i. The SPF for pedestrian crashes at intersections is in the form of (3-31)= exp ln lnN a b AADT c AADPpedbase totalint crossing# #+ +b ` `j jl where: AADTtotal = sum of annual average daily traffic volumes for the major and minor roads (=AADTmaj + AADTmin) (veh/day), AADPcrossing = sum of daily pedestrian volumes crossing all intersection legs (ped/day), and a, b, c = regression coefficients. The coefficients of a, b, and c; the overdispersion parameters; and the base conditions for the SPFs to predict total pedestrian crashes at intersections are provided in TableÂ 113. In TableÂ 113, the combined model for total pedestrian crashes (i.e., all severity levels combined) at three-leg

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 179Â Â and four-leg stop control intersections with two-way/two-way operations from TableÂ 109 is provided as separate models. Similarly, the combined model for total pedestrian crashes (i.e., all severity levels combined) at three-leg signal control intersections with two-way/two-way operations and four-leg signal control intersections with one-way/two-way operations from TableÂ 110 is provided as separate models in TableÂ 113. The AFs applicable to the expanded model for total pedestrian crashes at four-leg signal control intersections with two-way/two-way operations in TableÂ 113 are as follows. AF1p,i 2 Right-Turn-on-Red AFs applicable to the expanded model for total pedestrian crashes at four-leg signal control intersections with two-way/two-way operations, accounting for right-turn-on-red operations are presented in TableÂ 114. AF2p,i 2 Type of Left-Turn Signal Phasing AFs applicable to the expanded model for total pedestrian crashes at four-leg signal control intersections with two-way/two-way operations, accounting for type of left-turn signal phasing are presented in TableÂ 115. Intersection Type Intercept ( ) AADTtotal ( ) AADPcrossing ( ) Overdispersion Parameter ( ) Reduced models 3ST 2Ã2 â53.670 4.293 1.655 1.28E-04 3SG 2Ã2 â12.750 0.961 0.112 0.446 4ST 2Ã2 â53.670 4.293 1.655 1.28E-04 4SG 2Ã2 â19.085 1.518 0.395 0.520 4SG 1Ã2 â11.751 0.961 0.112 0.446 Expanded models 4SG 2Ã2 â19.941 1.683 0.268 0.461 Base conditions for reduced models: 3ST 2Ã2 No base conditions. 3SG 2Ã2 No base conditions. 4ST 2Ã2 No base conditions. 4SG 2Ã2 No base conditions. 4SG 1Ã2 No base conditions. Base conditions for expanded models: 4SG 2Ã2 Right-turn-on-red allowed on all approaches, all approaches have permissive left turns, no alcohol sales establishments within 1,000 ft. NOTE: 3ST 2Ã2= three-leg stop control intersections with two-way/two-way operations; 3SG 2Ã2= three-leg signal control intersections with two-way/two-way operations; 4ST 2Ã2 = four-leg stop control intersections with two-way/two-way operations; 4SG 2Ã2 = four-leg signal control intersections with two-way/two-way operations; 4SG 1Ã2 = four-leg signal control intersections with one-way/two-way operations combined. TableÂ 113. SPF coefficients for total pedestrian crashes at intersections. Right-Turn-on-Red Allowed on all approaches 1 Prohibited at one or more approach 0.787 TableÂ 114. Adjustment factors for right-turn-on-red (total pedestrian crashes at four-leg signal control intersections with two-way/two-way operations â expanded model).

180 Pedestrian and Bicycle Safety Performance Functions AF3p,i 2 Number of Alcohol Sales Establishments The AF applicable to the expanded model for total pedestrian crashes at four-leg signal con- trol intersections with two-way/two-way operations, accounting for the number of alcohol sales establishments within 1,000Â ft of the center of the intersection is as follows: (3-32).AF ASE 0 0189exp,p i3 #= ` j where: ASE = Number of alcohol sales establishments within 1,000Â ft of the center of the intersection. 3.4.2.1.3 Pedestrian Models (Comparison with Existing HSM Models) Following development of the pedestrian SPFs for urban intersections, compatibility testing of the new models was conducted to check the reasonableness of the results. FigureÂ 32 through FigureÂ 35 provide some sense of the reasonableness of the new models for application and potential incorporation in the HSM, but to gain a better understanding of the potential use of these models within the HSM, output results from the new models were compared to output from existing models in HSM PartÂ C, ChapterÂ 12 (Predictive Method for Urban and Suburban Arterials). Note, for these comparisons, the new models and the existing HSM models were not calibrated using a single agencyâs data; and unlike most of the existing models in the HSM, pedestrian exposure is accounted for in the new models, so some differences are expected. Nonetheless, output results from the new models were compared to output from existing models in the HSM to gain a sense of the reasonableness of the new models for application and potential incorporation in the HSM. If an expanded model was available, a comparison was made between the expanded model and existing HSM model. Otherwise, a comparison was made using the reduced model. FigureÂ 36 provides a comparison of the predicted total pedestrian crash frequency from the reduced model for total pedestrian crashes (i.e., all severity levels combined) at three-leg stop control intersections with two-way/two-way operations and the predicted average total pedes- trian crash frequency and total crashes (i.e., multiple-vehicle + single-vehicle + pedestrian + bicycle crashes) from the existing HSM Part C model for three-leg stop control intersections. The predicted average crash frequencies are for the base conditions for both models. Compari- sons with the reduced model are shown for three pedestrian crossing volume levels: 100, 300, and 800 ped/day. For the existing HSM model, two scenarios were tested: â¢ 85Â percent of the total daily entering-intersection traffic (i.e., AADTtotal) volume assigned to the major-road approach and 15Â percent assigned to the minor-road approach (85/15). â¢ 60Â percent of the total daily entering-intersection traffic (i.e., AADTtotal) volume assigned to the major-road approach and 40Â percent assigned to the minor-road approach (60/40). Based on the comparison, at low pedestrian crossing volumes around 100 ped/day or less, the new model predicts pedestrian crash frequencies similar to estimates from the existing HSM model. Although it is not clearly visible in FigureÂ 36, the predicted average total pedestrian crash Type of Left-Turn Signal Phasing All permissive 1 Protected/permissive or protected/protected 0.552 TableÂ 115. Adjustment factors for type of left-turn signal phasing (total pedestrian crashes at four-leg signal control intersections with two-way/two-way operations â expanded model).

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 181Â Â frequencies for the 60/40 scenario are basically equivalent to estimates for the 85/15 scenario up to a total entering-intersection volume of 22,000 veh/day. As pedestrian crossing volumes increase above 100 ped/day, the new model predicts significantly higher pedestrian crash frequencies compared to estimates from the existing HSM model; and at a total entering- intersection traffic volume (AADTtotal) around 29,000 veh/day and above and pedestrian crossing volumes of 800 ped/day, the new model predicts more pedestrian crashes than total crashes from the existing HSM model. Based on this assessment, the reduced model for total pedestrian crashes (i.e., all severity levels combined) at three-leg stop control intersections with two-way/two-way operations does not appear compatible with the existing HSM Part C model for urban and suburban three-leg stop control intersections. FigureÂ 37 provides a comparison of the predicted total pedestrian crash frequency from the reduced model for total pedestrian crashes (i.e., all severity levels combined) at four-leg stop con- trol intersections with two-way/two-way operations and the predicted average total pedestrian crash frequency and total crashes (i.e., multiple-vehicle + single-vehicle + pedestrian + bicycle crashes) from the existing HSM PartÂ C model for four-leg stop control intersections. The pre- dicted average crash frequencies are for the base conditions for both models. Comparisons with the reduced model are shown for three pedestrian crossing volume levels: 100, 300, and 800 ped/day. For the existing HSM model, two scenarios were tested: â¢ 85Â percent of the total daily entering-intersection traffic (i.e., AADTtotal) volume assigned to the major-road approach and 15Â percent assigned to the minor-road approach (85/15). â¢ 60Â percent of the total daily entering-intersection traffic (i.e., AADTtotal) volume assigned to the major-road approach and 40Â percent assigned to the minor-road approach (60/40). The results of this comparison for four-leg stop control intersections are similar to the com- parison for three-leg stop control intersections. Thus, the reduced model for total pedestrian crashes FigureÂ 36. Comparison of SPF for predicted average total pedestrian crashes per year at three-leg stop control intersections with two-way/two-way operations (reduced model) and existing HSM model.

182 Pedestrian and Bicycle Safety Performance Functions (i.e., all severity levels combined) at four-leg stop control intersections with two-way/two-way operations does not appear compatible with the existing HSM PartÂ C model for urban and suburban four-leg stop control intersections. FigureÂ 38 provides a comparison of the predicted total pedestrian crash frequency from the reduced model for total pedestrian crashes (i.e., all severity levels combined) at three-leg signal control intersections with two-way/two-way operations and the predicted average total pedes- trian crash frequency and total crashes (i.e., multiple-vehicle + single-vehicle + pedestrian + bicycle crashes) from the existing HSM PartÂ C model for three-leg signal control intersections. As a reminder, the HSM PartÂ C model for three-leg signal control intersections already includes a comprehensive model to predict pedestrian crashes. Estimates from this model are used in this comparison. In this comparison, predicted average crash frequencies are for the base conditions for both models. Comparisons of the models are shown for three pedestrian crossing volume levels: 100, 300, and 800Â ped/day. For the existing HSM model, two scenarios were tested: â¢ 85Â percent of the total daily entering-intersection traffic (i.e., AADTtotal) volume assigned to the major-road approach and 15Â percent assigned to the minor-road approach (85/15). â¢ 60Â percent of the total daily entering-intersection traffic (i.e., AADTtotal) volume assigned to the major-road approach and 40Â percent assigned to the minor-road approach (60/40). The results of this comparison show that for three-leg signal control intersections, the reduced model consistently predicts higher pedestrian crash frequencies than the existing HSM model for three-leg signal control intersections, but the estimates are of similar magnitude compared to total crashes. Thus, it appears that the reduced model for total pedestrian crashes (i.e., all severity levels combined) at three-leg signal control intersections with two-way/two-way opera- tion is suitable for integration into the HSM for use with three-leg signal control intersections, but calibration of the models will likely be critical. FigureÂ 37. Comparison of SPF for predicted average total pedestrian crashes per year at four-leg stop control intersections with two-way/two-way operations (reduced model) and existing HSM model.

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 183Â Â There is no existing HSM model for four-leg signal control intersections with one-way/two-way operations so no comparison is made here between the reduced model for total pedestrian crashes (i.e., all severity levels combined) at four-leg signal control intersections with one-way/ two-way operations and an existing HSM model. FigureÂ 39 provides a comparison of the predicted total pedestrian crash frequency from the expanded model for total pedestrian crashes (i.e., all severity levels combined) at four-leg signal control intersections with two-way/two-way operation and the predicted average total pedes- trian crash frequency and total crashes (i.e., multiple-vehicle + single-vehicle + pedestrian + bicycle crashes) from the existing HSM PartÂ C model for four-leg signal control intersections. Again, as a reminder, the HSM PartÂ C model for four-leg signal control intersections already includes a comprehensive model to predict pedestrian crashes, and estimates from this model are used in this comparison. The predicted average crash frequencies in this comparison are for the base conditions for both models. Comparisons with the models are shown for three pedestrian crossing volume levels: 100, 300, and 800Â ped/day. For the existing HSM model, two scenarios were tested: â¢ 85Â percent of the total daily entering-intersection traffic (i.e., AADTtotal) volume assigned to the major-road approach and 15Â percent assigned to the minor-road approach (85/15). â¢ 60Â percent of the total daily entering-intersection traffic (i.e., AADTtotal) volume assigned to the major-road approach and 40Â percent assigned to the minor-road approach (60/40). The results of this comparison for four-leg signal control intersections with two-way/two-way operations are similar to the comparison for three-leg signal control intersections with two-way/ two-way operations although the new model for four-leg signal control intersections predicts higher pedestrian crash frequencies compared to the existing HSM model. Thus, it appears that the expanded model for total pedestrian crashes (i.e., all severity levels combined) at four-leg FigureÂ 38. Comparison of SPF for predicted average total pedestrian crashes per year at three-leg signal control intersections with two-way/two-way operations (reduced model) and existing HSM model.

184 Pedestrian and Bicycle Safety Performance Functions signal control intersections with two-way/two-way operation may be suitable for integration into the HSM for use with four-leg signal control intersections, but calibration of the models will likely be critical. 3.4.2.2 Bicycle Models e nal bicycle SPFs for estimating bicycle crashes at intersections by intersection type are shown rst in tabular form presenting conventional statistical output results, followed by the same models converted to a form more suitable for inclusion in the HSM. For each intersection type, three levels of models are presented as appropriate: â¢ A reduced model to estimate total bicycle crashes (i.e., all severity levels combined) that pri- marily includes exposure measures for motor vehicles (i.e., AADT) and bicycles (i.e., AADB). â¢ An expanded model to estimate total bicycle crashes (i.e., all severity levels combined) that includes exposure measures for motor vehicles (i.e., AADT) and bicycles (i.e., AADB) as well as other geometric, trac control, and site characteristic features found to be signicant pre- dictors of total bicycle crashes. â¢ A reduced model to estimate FS bicycle crashes (i.e., fatal and suspected serious injury crashes) that primarily includes exposure measures for motor vehicles (i.e., AADT) and bicy- cles (i.e., AADB). Comparisons of the reduced and/or expanded models with existing models in the HSM Part C illustrate potential dierences in future editions of the HSM if the models from this research are integrated into the HSM and potential compatibility issues with existing HSM models. 3.4.2.2.1 Bicycle Models (Conventional Statistical Output) All models were estimated in R using the GLM package. For all intersection models, the number of years was entered as oset variables (in log form) so that the eect is proportional on predicted crash frequency. FigureÂ 39. Comparison of SPF for predicted average total pedestrian crashes per year at four-leg signal control intersections with two-way/two-way operations (expanded model) and existing HSM model.

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 185Â Â TableÂ 116 presents a reduced model for total bicycle crashes (i.e., all severity levels combined) at three-leg and four-leg stop control intersections with two-way/two-way operations that pri- marily includes exposure variables. The table shows the model coefficients, their standard error, and associated p-values. The table also shows the inverse of the overdispersion parameter and its standard error, the 2 x log-likelihood at convergence, the total number of crashes, and the total number of sites associated with the model. In R, the output from the GLM package pro- vides the inverse of the traditional overdispersion parameter. Therefore, TableÂ 116 and the other tables in Section 3.4.2.2.1 provide the inverse of the overdispersion and standard error for the overdispersion inverse, consistent with model outputs from R. In this reduced model and other models presented below, an indicator variable is included to account for differences between sites in Minneapolis and Philadelphia. FigureÂ 40 graphically presents the SPF shown in TableÂ 116 for various motor vehicle entering (AADTtotal) and bicycle crossing (AADBcrossing) volumes. No expanded model to predict total bicycle crashes (i.e., all severity levels combined) at three- leg and four-leg stop control intersections with two-way/two-way operations or to predict FS Variable Coefficient Std. Error p-Value Constant â38.443 10.456 < 0.001 Natural log of daily entering-intersection traffic (AADTtotal) volume (veh/day) 1.577 1.097 0.150 Natural log of daily bicycle volumes crossing all intersection legs (AADBcrossing) (bike/day) 3.179 1.084 0.003 Indicator for roadway segment within Pennsylvania 2.189 2.319 0.345 Inverse of overdispersion parameter 11,340 34,1914 â 2 x log-likelihood at convergence â19.84 Total number of crashes 8 Total number of sites 37 NOTE: â = Not applicable. TableÂ 116. SPF for total bicycle crashes at three-leg and four-leg stop control intersections with two-way/two-way operations (reduced model). FigureÂ 40. Graphical representation of the SPF for predicted average total bicycle crashes per year at three-leg and four-leg stop control intersections with two-way/two-way operations (reduced model).

186 Pedestrian and Bicycle Safety Performance Functions bicycle crashes (i.e., fatal and suspected serious injury crashes) is provided because either the models did not converge or the coefficient(s) for one or more of the measures was counter- intuitive or not statistically significant. TableÂ 117 presents a reduced model for total bicycle crashes (i.e., all severity levels combined) at three-leg signal control intersections with two-way/two-way operations and four-leg signal control intersections with one-way/two-way operations that primarily includes exposure variables. The table shows the model coefficients, their standard error, and associated p-values. The table also shows the inverse of the overdispersion parameter and its standard error, the 2 x log-likelihood at convergence, the total number of crashes, and the total number of sites associated with the model. In this reduced model, an indicator variable is included to account for the number of legs. FigureÂ 41 graphically presents the SPF shown in TableÂ 117 for various motor vehicle entering (AADTtotal) and bicycle crossing (AADBcrossing) volumes. No expanded model to predict total bicycle crashes (i.e., all severity levels combined) or FS pedestrian crashes (i.e., fatal and suspected serious injury crashes) at three-leg signal control intersections with two-way/two-way operations or four-leg signal control intersections with one-way/two-way operations is provided because either the models did not converge or the coefficient(s) for one or more of the measures was counterintuitive or not statistically significant. TableÂ 118 presents a reduced model for total bicycle crashes (i.e., all severity levels combined) at four-leg signal control intersections with two-way/two-way operations that primarily includes exposure variables. The table shows the model coefficients, their standard error, and associated p-values. The table also shows the inverse of the overdispersion parameter and its standard error, the 2 x log-likelihood at convergence, the total number of crashes, and the total number of sites associated with the model. FigureÂ 42 graphically presents the SPF shown in TableÂ 118 for various motor vehicle entering (AADTtotal) and bicycle crossing (AADBcrossing) volumes. TableÂ 119 presents an expanded model for total bicycle crashes (i.e., all severity levels com- bined) at four-leg signal control intersections with two-way/two-way operations. The table shows the model coefficients, their standard error, and associated p-values; the inverse of the overdispersion parameter and its standard error; the 2 x log-likelihood at convergence; the total number of crashes; and the total number of sites associated with the model. In addition to expo- sure measures for motor vehicles and bicycles, significant predictors of total bicycle crashes at four-leg signal control intersections with two-way/two-way operations include: â¢ Presence of bicycle facilities entering the intersection. â¢ Protection of left-turn movements. â¢ Number of schools stops within 1,000Â ft. Variable Coefficient Std. Error p-Value Constant â8.644 4.496 0.055 Natural log of daily entering-intersection traffic (AADTtotal) volume (veh/day) 0.379 0.419 0.365 Natural log of daily bicycle volumes crossing all intersection legs (AADBcrossing) (bike/day) 0.342 0.283 0.228 Indicator for four-leg intersection (4SG 1Ã2) 0.450 0.434 0.300 Indicator for roadway segment within Pennsylvania 0.154 0.411 0.709 Inverse of overdispersion parameter 1.550 1.110 â 2 x log-likelihood at convergence â186.646 Total number of crashes 49 Total number of sites 104 NOTE: â = Not applicable. TableÂ 117. SPF for total bicycle crashes at three-leg signal control intersections with two-way/two-way operations and four-leg signal control intersections with one-way/ two-way operations (reduced model).

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 187Â Â FigureÂ 43 graphically presents the SPF shown in TableÂ 119 for various motor vehicle entering (AADTtotal) and bicycle crossing (AADBcrossing) volumes. No expanded model to predict FS bicycle crashes (i.e., fatal and suspected serious injury crashes) at four-leg signal control intersections with two-way/two-way operations is provided because either the model did not converge or the coefficient(s) for one or more of the measures was counterintuitive or not statistically significant. 3.4.2.2.2 Bicycle Models (Form Suitable for Integration into the HSM) This section presents the bicycle SPFs from TableÂ 116 through TableÂ 119 converted to a form more suitable for integration within the HSM. In HSM PartÂ C, the components of the predictive models generally consists of an SPF, one or more adjustment factors (AFs), and a calibration FigureÂ 41. Graphical representation of the SPF for predicted average total bicycle crashes per year at three-leg signal control intersections with two-way/two-way operations and four-leg signal control intersections with one-way/two-way operations (reduced model). Variable Coefficient Std. Error p-Value Constant â12.135 3.672 < 0.001 Natural log of daily entering-intersection traffic (AADTtotal) volume (veh/day) 0.843 0.333 0.011 Natural log of daily bicycle volumes crossing all intersection legs (AADBcrossing) (bike/day) 0.289 0.249 0.245 Indicator for roadway segment within Pennsylvania â0.087 0.301 0.771 Inverse of overdispersion parameter 4.450 5.720 â 2 x log-likelihood at convergence â249.641 Total number of crashes 76 Total number of sites 127 NOTE: â = Not applicable. TableÂ 118. SPF for total bicycle crashes at four-leg signal control intersection with two-way/two-way operations (reduced model).

188 Pedestrian and Bicycle Safety Performance Functions FigureÂ 42. Graphical representation of the SPF for predicted average total bicycle crashes per year at four-leg signal control intersections with two-way/two-way operations (reduced model). Variable Coefficient Std. Error p-Value Constant 13.829 3.670 < 0.001 Natural log of daily entering-intersection traffic (AADTtotal) volume (veh/day) 0.958 0.354 0.007 Natural log of daily bicycle volumes crossing all intersection legs (AADBcrossing) (bike/day) 0.404 0.277 0.146 Indicator for at least one bicycle facility entering the intersection 0.442 0.281 0.115 Indicator for some level of protected left-turning movements at all intersection approaches 0.592 0.407 0.146 Number of schools within 1,000 ft 0.110 0.080 0.170 Indicator for roadway segment within Pennsylvania 0.010 0.296 0.973 Inverse of overdispersion parameter 50 603 â 2 x log-likelihood at convergence 243.26 Total number of crashes 76 Total number of observations 127 NOTE: â = Not applicable. TableÂ 119. SPF for total bicycle crashes at four-leg signal control intersections with two-way/two-way operations (expanded model).

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 189Â Â factor. Equation 3-28 and Equation 3-29 present the predictive models for intersections from the HSM ChapterÂ 12 on urban and suburban arterials but slightly rearranged to properly integrate the new pedestrian and bicycle SPFs into the equations (see Section 3.4.2.1.2.). Each component of Equation 3-28 (i.e., Nbi, Npedi, and Nbikei) is estimated separately to obtain the estimated total crash frequency for an individual intersection, and then a calibration factor is applied. A comprehensive model for estimating bicycle crashes at intersections would take the following form: (3-33)C# #N= AFN AF ,,bike ii b ibikebase b i1 2int # f` j where: Nbikei = predicted average crash frequency of bicycle crashes for an individual inter- section (crashes/year), Nbikebase int = predicted average crash frequency of bicycle crashes for an individual inter- section for base conditions of the SPF developed for site type i (crashes/year), AF1b,i . . . AFnb,i = adjustment factors for bicycle crashes specific to SPF for site type i, and Ci = calibration factor to adjust SPF for local conditions for site type i. The SPF for bicycle crashes at intersections is in the form of (3-34)= lnexp ln c AADBa b AADTN sinbikebase cros gtotalint # #+ +b ` `j jl where: AADTtotal = sum of annual average daily traffic volumes for the major and minor roads (=AADTmaj + AADTmin) (veh/day). FigureÂ 43. Graphical representation of the SPF for predicted average total bicycle crashes per year at four-leg signal control intersections with two-way/two-way operations (expanded model).

190 Pedestrian and Bicycle Safety Performance Functions AADBcrossing = sum of daily bicycle volumes crossing all intersection legs (ped/day). a, b, c = regression coefficients. The coefficients of a, b, and c; the overdispersion parameters; and the base conditions for the SPFs to predict total bicycle crashes at intersections are provided in TableÂ 120. In TableÂ 120, the combined model for total bicycle crashes (i.e., all severity levels combined) at three-leg and four-leg stop control intersections with two-way/two-way operations is provided as separate models. Similarly, the combined model for total bicycle crashes (i.e., all severity levels combined) at three-leg signal control intersections with two-way/two-way operations and four-leg signal control intersections with one-way/two-way operations from TableÂ 117 is provided in TableÂ 120 as separate models. Adjustment factors applicable to the expanded model for total bicycle crashes at four-leg signal control intersections with two-way/two-way operations in TableÂ 120 are as follows. AF1b,i 2 Presence of Bicycle Facilities Entering the Intersection Adjustment factors applicable to the expanded model for total bicycle crashes at four-leg signal control intersections with two-way/two-way operations, accounting for the presence of bicycle facilities entering the intersection, are presented in TableÂ 121. Intersection Type Intercept ( ) AADTtotal ( ) AADBcrossing ( ) Overdispersion Parameter ( ) Reduced models 3ST 2Ã2 â38.443 1.577 3.179 8.82E-05 3SG 2Ã2 â8.644 0.379 0.342 0.645 4ST 2Ã2 â38.443 1.577 3.179 8.82E-05 4SG 2Ã2 â12.135 0.843 0.289 0.225 4SG 1Ã2 â8.194 0.379 0.342 0.645 Expanded models 4SG 2Ã2 â13.829 0.958 0.404 0.02 Base conditions for reduced models: 3ST 2Ã2 No base conditions. 3SG 2Ã2 No base conditions. 4ST 2Ã2 No base conditions. 4SG 2Ã2 No base conditions. 4SG 1Ã2 No base conditions. Base conditions for expanded models: 4SG 2Ã2 No bicycle facilities entering the intersection, all approaches have permissive left turns, no schools within 1,000 ft NOTE: 3ST 2Ã2= three-leg stop control intersections with two-way/two-way operations; 3SG 2Ã2= three-leg signal control intersections with two-way/two-way operations; 4ST 2Ã2 = four-leg stop control intersections with two-way/two-way operations; 4SG 2Ã2 = four-leg signal control intersections with two-way/two-way operations; 4SG 1Ã2 = four-leg signal control intersections with one-way/two-way operations combined. TableÂ 120. SPF coefficients for total bicycle crashes at intersections. Presence of at Least One Bicycle Facility Entering the Intersection No 1 Yes 0.611 TableÂ 121. Adjustment Factors for presence of bicycle facilities entering the intersection (total bicycle crashes at four-leg signal control intersections with two-way/two-way operations â expanded model).

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 191Â Â AF2b,i 2 Type of Left-Turn Signal Phasing Adjustment factors applicable to the expanded model for total bicycle crashes at four-leg signal control intersections with two-way/two-way operations, accounting for type of left-turn signal phasing are presented in TableÂ 122. AF3b,i 2 Number of Schools The adjustment factor applicable to the expanded model for total bicycle crashes at four- leg signal control intersections with two-way/two-way operations, accounting for number of schools within 1,000Â ft of the center of the intersection is as follows: (3-35).AF S 0 110exp,b i3 #= ` j where: S = number of schools within 1,000Â ft of the center of the intersection. 3.4.2.2.3 Bicycle Models (Comparison with Existing HSM Models) Following development of the bicycle SPFs for urban intersections, compatibility testing of the new models was conducted to check the reasonableness of the results. FigureÂ 40 through FigureÂ 43 provide some sense of the reasonableness of the new models for application and potential incorporation in the HSM, but to gain a better understanding of the potential use of these models within the HSM, output results from the new models were compared to output from existing models in HSM PartÂ C, ChapterÂ 12 (Predictive Method for Urban and Suburban Arterials). Note, for these comparisons, the new models and the existing HSM models were not calibrated using a single agencyâs data; and, unlike the existing models in the HSM, bicycle expo- sure is accounted for in the new models, so some differences are expected. Nonetheless, output results from the new models were compared to output from existing HSM models to gain a sense of the reasonableness of the new models for application and potential incorporation in the HSM. If an expanded model was available, a comparison was made between the expanded model and the existing HSM model. Otherwise, a comparison was made using the reduced model. FigureÂ 44 provides a comparison of the predicted total bicycle crash frequency from the reduced model for total bicycle crashes (i.e., all severity levels combined) at three-leg stop con- trol intersections with two-way/two-way operations and the predicted average total bicycle crash frequency and total crashes (i.e., multiple-vehicle + single-vehicle + pedestrian + bicycle crashes) from the existing HSM PartÂ C model for three-leg stop control intersections. The predicted average crash frequencies are for the base conditions for both models. Comparisons with the reduced model are shown for three bicycle crossing volume levels: 100, 300, and 800 bike/day. For the existing HSM model, two scenarios were tested: â¢ 85Â percent of the total daily entering-intersection traffic (i.e., AADTtotal) volume assigned to the major-road approach and 15Â percent assigned to the minor-road approach (85/15). â¢ 60Â percent of the total daily entering-intersection traffic (i.e., AADTtotal) volume assigned to the major-road approach and 40Â percent assigned to the minor-road approach (60/40). Type of Left-turn Signal Phasing All permissive 1 Protected/permissive or protected/protected 0.583 TableÂ 122. Adjustment factors for type of left-turn signal phasing (total bicycle crashes at four-leg signal control intersections with two-way/two-way operations â expanded model).

192 Pedestrian and Bicycle Safety Performance Functions Based on the comparison, at lower bicycle crossing volumes around 500Â ped/day or less, the new model predicts bicycle crash frequencies similar to estimates from the existing HSM model. Although it is not clearly visible in FigureÂ 44, the predicted average total bicycle crash frequencies for the 60/40 scenario are basically equivalent to estimates for the 85/15 scenario up to a total entering-intersection volume of 22,000Â veh/day. As bicycle crossing volumes increase above 500Â bike/day, the new model predicts significantly higher bicycle crash frequencies compared to estimates from the existing HSM model. Based on this assessment, the reduced model for total bicycle crashes (i.e., all severity levels combined) at three-leg stop control intersections with two-way/two-way operations appears compatible with the existing HSM PartÂ C model for urban and suburban three-leg stop control intersections although calibration of the models will likely be critical. FigureÂ 45 provides a comparison of the predicted total bicycle crash frequency from the reduced model for total bicycle crashes (i.e., all severity levels combined) at four-leg stop control intersections with two-way/two-way operations and the predicted average total bicycle crash frequency and total crashes (i.e., multiple-vehicle + single-vehicle + pedestrian + bicycle crashes) from the existing HSM PartÂ C model for four-leg stop control intersections. The predicted average crash frequencies are for the base conditions for both models. Comparisons with the reduced model are shown for three bicycle crossing volume levels: 100, 300, and 800Â bike/day. For the existing HSM model, two scenarios were tested: â¢ 85Â percent of the total daily entering-intersection traffic (i.e., AADTtotal) volume assigned to the major-road approach and 15Â percent assigned to the minor-road approach (85/15). â¢ 60Â percent of the total daily entering-intersection traffic (i.e., AADTtotal) volume assigned to the major-road approach and 40Â percent assigned to the minor-road approach (60/40). The results of this comparison for four-leg stop control intersections are similar to the com- parison for three-leg stop control intersections. Thus, the reduced model for total bicycle crashes FigureÂ 44. Comparison of SPF for predicted average total bicycle crashes per year at three-leg stop control intersections with two-way/two-way operations (reduced model) and existing HSM model.

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 193Â Â (i.e., all severity levels combined) at four-leg stop control intersections with two-way/two-way operations appears compatible with the existing HSM PartÂ C model for urban and suburban four-leg stop control intersections, although calibration of the models will likely be critical. FigureÂ 46 provides a comparison of the predicted total bicycle crash frequency from the reduced model for total bicycle crashes (i.e., all severity levels combined) at three-leg signal control intersections with two-way/two-way operation and the predicted average total bicycle crash frequency and total crashes (i.e., multiple-vehicle + single-vehicle + pedestrian + bicycle crashes) from the existing HSM PartÂ C model for three-leg signal control intersections. The pre- dicted average crash frequencies are for the base conditions for both models. Comparisons with the models are shown for three bicycle crossing volume levels: 100, 300, and 800Â bike/day. For the existing HSM model, two scenarios were tested: â¢ 85Â percent of the total daily entering-intersection traffic (i.e., AADTtotal) volume assigned to the major-road approach and 15Â percent assigned to the minor-road approach (85/15). â¢ 60Â percent of the total daily entering-intersection traffic (i.e., AADTtotal) volume assigned to the major-road approach and 40Â percent assigned to the minor-road approach (60/40). The results of this comparison show that for three-leg signal control intersections, the reduced model consistently predicts bicycle crash frequencies similar in magnitude to estimates from the existing HSM model for three-leg signal control intersections. Thus, it appears that the reduced model for total bicycle crashes (i.e., all severity levels combined) at three-leg signal control inter- sections with two-way/two-way operation is suitable for integration into the HSM for use with three-leg signal control intersections, although calibration of the models will likely be critical. There is no existing HSM model for four-leg signal control intersections with one-way/two-way operations so no comparison is made here between the reduced model for total bicycle crashes (i.e., all severity levels combined) at four-leg signal control intersections with one-way/two-way operations and an existing HSM model. FigureÂ 45. Comparison of SPF for predicted average total bicycle crashes per year at four-leg stop control intersections with two-way/two-way operations (reduced model) and existing HSM model.

194 Pedestrian and Bicycle Safety Performance Functions FigureÂ 47 provides a comparison of the predicted total bicycle crash frequency from the expanded model for total bicycle crashes (i.e., all severity levels combined) at four-leg signal con- trol intersections with two-way/two-way operation and the predicted average total bicycle crash frequency and total crashes (i.e., multiple-vehicle + single-vehicle + pedestrian + bicycle crashes) from the existing HSM PartÂ C model for four-leg signal control intersections. The predicted average crash frequencies are for the base conditions for both models. Comparisons with the models are shown for three bicycle crossing volume levels: 100, 300, and 800Â bike/day. For the existing HSM model, two scenarios were tested: â¢ 85Â percent of the total daily entering-intersection traffic (i.e., AADTtotal) volume assigned to the major-road approach and 15Â percent assigned to the minor-road approach (85/15). â¢ 60Â percent of the total daily entering-intersection traffic (i.e., AADTtotal) volume assigned to the major-road approach and 40Â percent assigned to the minor-road approach (60/40). The results of this comparison for four-leg signal control intersections with two-way/two-way operations are similar to the comparison for three-leg signal control intersections with two-way/ two-way operations although the new model for four-leg signal control intersections predicts higher bicycle crash frequencies compared to the existing HSM model. Thus, it appears that the expanded model for total bicycle crashes (i.e., all severity levels combined) at four-leg signal con- trol intersections with two-way/two-way operation is suitable for integration into the HSM for use with four-leg signal control intersections, although calibration of the models will likely be critical. 3.5 Calibration of the Models As indicated in the comparison of the new pedestrian and bicycle SPFs with existing HSM models, calibration will likely be necessary to integrate the new pedestrian and bicycle SPFs with existing HSM models for multiple- and single-vehicle crashes. The purpose of calibration is to FigureÂ 46. Comparison of SPF for predicted average total bicycle crashes per year at three-leg signal control intersections with two-way/two-way operations (reduced model) and existing HSM model.

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 195Â Â adjust predictive models which were developed with data from one jurisdiction for application in another jurisdiction. Calibration provides a means to account for differences between juris- dictions in various factors such as climate, driver populations, and crash reporting thresholds. For the Part C models to provide results that are meaningful and accurate for a jurisdiction, the SPFs should be calibrated using data from the given jurisdiction. Calibration procedures and calibration of predictive models are likely even more critical as predictive models are being combined in the predictive methods of HSM PartÂ C. In Equation 3-12 for urban and suburban roadway segments and Equation 3-28 for urban and suburban intersections, a calibration factor is applied to the component Nbr for roadway segments and Nbi for intersections. These calibration factors should be based strictly on multiple- and single-vehicle crashes only, excluding pedestrian and bicycle crashes. Then in Equations 3-14, 3-21, 3-30, and 3-33, the calibration factors for the pedestrian and bicycle SPFs used to calculate Npedr, Nbiker, Npedi, and Nbikei should be developed strictly using observed pedestrian and bicycle crashes, respectively. 3.6 Summary and Recommendations Several pedestrian and bicycle SPFs incorporating pedestrian and bicycle exposure measures into the models were developed for urban and suburban roadway segments and intersections using data from Minneapolis and Philadelphia. These SPFs were developed for potential consideration in the HMS PartÂ C, ChapterÂ 12 on predictive methods for urban and suburban arterials and/or for potential use in the PartÂ B network screening chapter of HSM2. Pedestrian and bicycle SPFs were developed for the following site types: â¢ Urban and suburban roadway segments â Two-lane undivided roads (2U) â Four-lane undivided roads (4U) FigureÂ 47. Comparison of SPF for predicted average total bicycle crashes per year at four-leg signal control intersections with two-way/two-way operations (expanded model) and existing HSM model.

196 Pedestrian and Bicycle Safety Performance Functions â Four-lane divided roads (4D) â One-way roads (OW) (including one-lane, two-lane, and three-lane roads) â¢ Urban and suburban intersections â Three-leg stop control (3ST 2Ã2) â Three-leg signal control (3SG 2Ã2) â Four-leg stop control (4ST 2Ã2) â Four-leg signal control (4SG 2Ã2) â Four-leg signal control (4SG 1Ã2) For each site type, three levels of models were developed, as appropriate: â¢ A reduced model to estimate total pedestrian/bicycle crashes (i.e., all severity levels combined) that primarily includes exposure measures for motor vehicles and pedestrians/bicycles. â¢ An expanded model to estimate total pedestrian/bicycle crashes (i.e., all severity levels com- bined) that includes exposure measures for motor vehicles and pedestrians/bicycles as well as other geometric, traffic control, and site characteristic features found to be significant pre- dictors of total pedestrian/bicycle crashes. â¢ A reduced model to estimate FS pedestrian/bicycle crashes (i.e., fatal and suspected serious injury crashes) that primarily includes exposure measures for motor vehicles and pedestrians/ bicycles. Consideration was given to modeling pedestrian and bicycle crashes by crash types. For example, for roadway segments, consideration was given to modeling pedestrian crashes that occurred while walking along a roadway separately from crossing crashes, and consideration was given to modeling bicycle crashes that occurred along the roadway and driveway-related crashes separately. For intersections, consideration was given to modeling crashes involving through movements by motor vehicles separately from crashes involving left-turning movements by motor vehicles. However, limitations in the crash data prohibited such analyses. In the end, all pedestrian crashes assigned to a site were included together in the development of the predictive models and similarly for bicycle crashes. Expanded models were developed specifically for potential use in the HSM Part C, ChapterÂ 12 on urban and suburban arterial predictive methods. If an expanded model is not available for a given site type, then a reduced model could be considered in its place for incorporation into the urban and suburban arterial predictive method chapter. Reduced models were also developed for potential consideration in the network screening chapter of HSM2, which will include network-screening-level SPFs. Comparisons of the pedestrian and bicycle SPFs with existing HSM Part C models demonstrated that several of the new models may not be compatible with existing HSM models, and calibra- tion will likely be critical for all the new pedestrian and bicycle SPFs to be compatible with the existing HSM models. Based on sample sizes used in model development and compatibility testing of the new models to check for the reasonableness of their results, TableÂ 123, TableÂ 124, TableÂ 125, and TableÂ 126 summarize the recommendations concerning potential integration of the final pedestrian and bicycle SPFs presented in this section into HSM2. As indicated in TableÂ 123, the reduced models for total pedestrian crashes on roadway segments for four-lane undivided and divided roads, and one-way roads are recommended for potential inclusion in the network screening chapter of HSM2 Part B. In addition, the expanded models for four-lane undivided and divided roads and one-way roads are recommended for potential inclusion in the urban and suburban arterial predictive chapter of HSM2 Part C. The reduced and expanded models for two-lane undivided roads may be considered for inclusion in the HSM2 Part B and Part C chapters, respectively, but compatibility issues with the existing HSM model for two-lane undivided roads have been

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 197Â Â Reduced models (total crashes) 2U â5.214 0.327 0.224 1.267 Yesa No 4U â11.154 0.902 0.236 1.855 Yes No 4D â12.497 0.902 0.236 1.855 Yes No OW â10.651 0.829 0.337 1.513 Yes No Expanded models (total crashes) 2U â4.029 0.347 0.114 0.948 No Yesa 4U â9.826 0.787 0.160 1.721 No Yes 4D â11.109 0.787 0.160 1.721 No Yes OW â9.339 0.897 0.207 1.410 No Yes Reduced models (FS crashes) 4U â26.576 2.125 0.638 2.387 Yes No 4D â26.956 2.125 0.638 2.387 Yes No Base conditions for reduced models (total crashes): 2U No base conditions. 4U No base conditions. 4D No base conditions. OW One lane. Base conditions for expanded models (total crashes): 2U Sidewalk buffer is 0 ft, average lane width is 12 ft, no bus/transit stops within 1,000 ft. 4U No schools within 1,000 ft. 4D No schools within 1,000 ft. OW One lane, sidewalk buffer is 0 ft, average lane width is 12 ft, no alcohol sales establishments within 1,000 ft. Base conditions for reduced models (FS crashes): 4U No base conditions. 4D No base conditions. NOTE: 2U = two-lane undivided; 4U = four-lane undivided; 4D = four-lane divided; OW = one-lane, one-way. aUse of this model in HSM2 will be contingent based on compatibility with the models to predict multiple- and single-vehicle motor vehicle crashes (excluding pedestrian and bicycle crashes). TableÂ 123. SPF coefficients for pedestrian crashes on roadway segments for consideration in HSM2. identified and likely need to be resolved for these models to be included in HSM2. In addition, the reduced models for FS pedestrian crashes for four-lane undivided and divided roads are recommended for potential inclusion in the network screening chapter of HSM2 Part B. As indicated in TableÂ 124, the reduced models for total bicycle crashes on roadway segments for two-way undivided roads, four-way undivided and divided roads, and one-way roads are recommended for potential inclusion in the network screening chapter of HSM2 Part B. The expanded models for two-way undivided roads, four-way undivided and divided roads, and one-way roads are recommended for potential inclusion in the urban and suburban arterial chapter of HSM2 Part C. As indicated in TableÂ 125, the reduced models for total pedestrian crashes at intersections for three-leg signal control intersections with two-way/two-way operations and four-leg signal control intersections with one-way/two-way operations are recommended for potential inclu- sion in both the network screening chapter of HSM2 Part B and the urban and suburban arterial predictive chapter of HSM2 Part C. The reduced model for four-leg signal control intersections

Reduced models (Total crashes) 2U â9.647 0.546 0.493 2.873 Yes No 4U â14.935 0.996 0.691 1.111 Yes No 4D â15.696 0.996 0.691 1.111 Yes No OW â10.388 0.446 0.749 0.002 Yes No Expanded models (Total crashes) 2U â8.422 0.528 0.359 2.347 No Yes 4U â13.864 0.960 0.607 1.052 No Yes 4D â14.624 0.960 0.607 1.052 No Yes OW â9.363 0.444 0.579 0.001 No Yes Base conditions for reduced models (Total crashes): 2U No base conditions. 4U No base conditions. 4D No base conditions. OW One lane. Base conditions for expanded models (Total crashes): 2U No buffered bike lane on either side of the road, average lane width is 12 ft, no bus/transit stops within 1,000 ft, no schools within 1,000 ft. 4U Speed limit less than or equal to 25 mph, no alcohol sales establishments within 1,000 ft. 4D Speed limit less than or equal to 25 mph, no alcohol sales establishments within 1,000 ft. OW One lane, no alcohol sales establishments within 1,000 ft. NOTE: 2U = two-lane undivided; 4U = four-lane undivided; 4D = four-lane divided; OW = one-lane, one-way. TableÂ 124. SPF coefficients for bicycle crashes on roadway segments for consideration in HSM2. Intersection Type Intercept ( ) AADTtotal ( ) AADPcrossing ( ) Overdispersion Parameter ( ) Recommended for Use in HSM2 Network Screening Chapter Urban/Suburban Arterial Chapter Reduced models (total crashes) 3ST 2Ã2 â53.670 4.293 1.655 1.28E-04 No No 3SG 2Ã2 â12.750 0.961 0.112 0.446 Yes Yes 4ST 2Ã2 â53.670 4.293 1.655 1.28E-04 No No 4SG 2Ã2 â19.085 1.518 0.395 0.520 Yes No 4SG 1Ã2 â11.751 0.961 0.112 0.446 Yes Yes Expanded models (total crashes) 4SG 2Ã2 â19.941 1.683 0.268 0.461 No Yes Base conditions for reduced models (total crashes): 3ST 2Ã2 No base conditions. 3SG 2Ã2 No base conditions. 4ST 2Ã2 No base conditions. 4SG 2Ã2 No base conditions. 4SG 1Ã2 No base conditions. Base conditions for expanded models (total crashes): 4SG 2Ã2 Right-turn-on-red allowed on all approaches, all approaches have permissive left turns, no alcohol sales establishments within 1,000 ft. NOTE: 3ST 2Ã2= three-leg stop control intersections with two-way/two-way operations; 3SG 2Ã2= three-leg signal control intersections with two-way/two-way operations; 4ST 2Ã2 = four-leg stop control intersections with two-way/two-way operations; 4SG 2Ã2 = four-leg signal control intersections with two-way/two-way operations; 4SG 1Ã2 = four-leg signal control intersections with one-way/two-way operations combined. TableÂ 125. SPF coefficients for pedestrian crashes at intersections for consideration in HSM2.

Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data 199Â Â Intersection Type Intercept ( ) AADTtotal ( ) AADBcrossing ( ) Overdispersion Parameter ( ) Recommended for Use in HSM2 Network Screening Chapter Urban/Suburban Arterial Chapter Reduced models (total crashes) 3ST 2Ã2 â38.443 1.577 3.179 8.82E-05 Yes Yes 3SG 2Ã2 â8.644 0.379 0.342 0.645 Yes Yes 4ST 2Ã2 â38.443 1.577 3.179 8.82E-05 Yes Yes 4SG 2Ã2 â12.135 0.843 0.289 0.225 Yes No 4SG 1Ã2 â8.194 0.379 0.342 0.645 Yes Yes Expanded models (total crashes) 4SG 2Ã2 â13.829 0.958 0.404 0.02 No Yes Base conditions for reduced models (total crashes): 3ST 2Ã2 No base conditions. 3SG 2Ã2 No base conditions. 4ST 2Ã2 No base conditions. 4SG 2Ã2 No base conditions. 4SG 1Ã2 No base conditions. Base conditions for expanded models (total crashes): 4SG 2Ã2 No bicycle facilities entering the intersection, all approaches have permissive left turns, no schools within 1,000 ft. NOTE: 3ST 2Ã2= three-leg stop control intersections with two-way/two-way operations; 3SG 2Ã2= three-leg signal control intersections with two-way/two-way operations; 4ST 2Ã2 = four-leg stop control intersections with two-way/two-way operations; 4SG 2Ã2 = four-leg signal control intersections with two-way/two-way operations; 4SG 1Ã2 = four-leg signal control intersections with one-way/two-way operations combined. TableÂ 126. SPF coefficients for bicycle crashes at intersections for consideration in HSM2. with two-way/two-way operations is recommended for potential inclusion in the network screen- ing chapter in HSM2 Part B. The expanded model for four-leg signal control intersections with two-way/two-way operations is recommended for potential inclusion in the urban and suburban arterial predictive chapter of HSM2 Part C. The reduced models for three-leg and four-way stop control intersections with two-way/two-way operations are not recommended for inclusion in the HSM2 Part B and Part C chapters due to compatibility issues with the existing HSM models in the urban and suburban arterial predictive chapter. As indicated in TableÂ 126, the reduced models for total bicycle crashes at intersections for three-way stop control and signal control intersections with two-way/two-way operations, four-way stop control and signal control intersections with two-way/two-way operations, and four-way signal control with one-way/two-way operations are recommended for potential inclusion in the network screening chapter of HSM2 Part B. The reduced models for three-way stop control and signal control intersections with two-way/two-way operations, four-way stop control intersections with two-way/two-way operations, four-way signal control intersections with one-way/two-way operations, and the expanded model for four-way signal control inter- sections with two-way/two-way operations are recommended for potential inclusion in the urban and suburban arterial predictive chapter of HSM2 Part C. Several adjustment factors to be used in conjunction with the pedestrian and bicycle SPFs are also recommended for consideration in the urban and suburban arterial predictive chapter of HSM2 Part C, including adjustment factors for: â¢ Pedestrian SPFs for roadway segments: â Presence of sidewalk buffers (TableÂ 95 and TableÂ 97). â Lane width (Equations 3-16 and 3-19). â Number of lanes (TableÂ 96).

200 Pedestrian and Bicycle Safety Performance Functions â Number of bus/transit stops within 1,000Â ft of the center of the roadway segment (Equation 3-17). â Number of schools within 1,000Â ft of the center of the roadway segment (Equation 3-18). â Number of alcohol sales establishments within 1,000Â ft of the center of the roadway segment (Equation 3-20). â¢ Bicycle SPFs for roadway segments: â Presence of a buffered bike lane (TableÂ 106). â Lane width (Equation 3-23). â Speed limit (TableÂ 107). â Number of lanes (TableÂ 108). â Number of bus/transit stops within 1,000Â ft of the center of the roadway segment (Equation 3-24). â Number of schools within 1,000Â ft of the center of the roadway segment (Equation 3-25). â Number of alcohol sales establishments within 1,000Â ft of the center of the roadway seg- ment (Equations 3-26 and 3-27). â¢ Pedestrian SPFs for intersections: â Right-turn-on-red (TableÂ 114). â Left-turn signal phasing (TableÂ 115). â Number of alcohol sales establishments within 1,000Â ft of the center of the intersection (Equation 3-32). â¢ Bicycle SPFs for intersections: â Presence of bicycle facilities (TableÂ 121). â Left-turn signal phasing (TableÂ 122). â Number of schools within 1,000Â ft of the center of the intersection (Equation 3-35). HSM ChapterÂ 12 already includes a comprehensive SPF to predict pedestrian crashes at three-leg and four-leg signal control intersections. A decision needs to be made on whether the pedestrian SPFs for three-leg and four-leg signal control intersections developed as part of this project should replace the existing HSM models or whether the existing pedestrian models for three-leg and four-leg signal control intersections should be retained. The primary reasons for retaining the existing models include: (a) HSM users are already familiar with the models, (b)Â the existing models incorporate more adjustment factors than the new models, and (c) no compat- ibility issues have been identified to date regarding the integration of these models with the other models used to predict multiple-vehicle, single-vehicle, bicycle crashes, etc. The primary reason for replacing the existing HSM pedestrian models for three-leg and four-leg signal control intersections with new models developed as part of this research is that all of the pedestrian (and bicycle) models in HSM2 would be developed with data from Minneapolis and Philadelphia. It is recommended that the existing pedestrian SPFs for three-leg and four-leg signal control intersections in HSM be removed and replaced in HSM2 with the new models developed as part of this research for consistency with the other pedestrian and bicycle SPFs. Because the pedestrian and bicycle SPFs presented in this section were developed using nega- tive binomial regression, which produces an overdispersion parameter, these models can be used in EB procedures to combine predicted crash frequencies from SPFs and observed crash data to estimate an expected crash frequency.