National Academies Press: OpenBook

Pedestrian and Bicycle Safety Performance Functions (2023)

Chapter: Section 2 - Literature Review and Survey of Practice

« Previous: Section 1 - Introduction
Page 9
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 9
Page 10
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 10
Page 11
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 11
Page 12
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 12
Page 13
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 13
Page 14
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 14
Page 15
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 15
Page 16
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 16
Page 17
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 17
Page 18
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 18
Page 19
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 19
Page 20
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 20
Page 21
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 21
Page 22
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 22
Page 23
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 23
Page 24
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 24
Page 25
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 25
Page 26
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 26
Page 27
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 27
Page 28
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 28
Page 29
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 29
Page 30
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 30
Page 31
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 31
Page 32
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 32
Page 33
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 33
Page 34
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 34
Page 35
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 35
Page 36
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 36
Page 37
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 37
Page 38
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 38
Page 39
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 39
Page 40
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 40
Page 41
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 41
Page 42
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 42
Page 43
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 43
Page 44
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 44
Page 45
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 45
Page 46
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 46
Page 47
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 47
Page 48
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 48
Page 49
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 49
Page 50
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 50
Page 51
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 51
Page 52
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 52
Page 53
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 53
Page 54
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 54
Page 55
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 55
Page 56
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 56
Page 57
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 57
Page 58
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 58
Page 59
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 59
Page 60
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 60
Page 61
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 61
Page 62
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 62
Page 63
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 63
Page 64
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 64
Page 65
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 65
Page 66
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 66
Page 67
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 67
Page 68
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 68
Page 69
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 69
Page 70
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 70
Page 71
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 71
Page 72
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 72
Page 73
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 73
Page 74
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 74
Page 75
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 75
Page 76
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 76
Page 77
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 77
Page 78
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 78
Page 79
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 79
Page 80
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 80
Page 81
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 81
Page 82
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 82
Page 83
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 83
Page 84
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 84
Page 85
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 85
Page 86
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 86
Page 87
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 87
Page 88
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 88
Page 89
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 89
Page 90
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 90
Page 91
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 91
Page 92
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 92
Page 93
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 93
Page 94
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 94
Page 95
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 95
Page 96
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 96
Page 97
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 97
Page 98
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 98
Page 99
Suggested Citation:"Section 2 - Literature Review and Survey of Practice." National Academies of Sciences, Engineering, and Medicine. 2023. Pedestrian and Bicycle Safety Performance Functions. Washington, DC: The National Academies Press. doi: 10.17226/27294.
×
Page 99

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.

9   This section summarizes the state of practice and other information related to the development and use of SPFs for pedestrians and bicycles. Section 2.1 summarizes the published literature reviewed relevant to the objectives of the research. Section 2.2 summarizes information gathered from a survey of transportation agencies, practitioners, and researchers to learn about their expe- riences related to the development and use of pedestrian and bicycle SPFs and interviews with selected agencies to gather more detailed information about their pedestrian and bicycle count programs, inventory datasets, and crash datasets; and Section 2.3 summarizes key issues learned from the literature review, the survey of practice, and interviews with selected agencies. 2.1 Literature Review The published literature is summarized according to the following topics: • Pedestrian and bicycle crash statistics. • Barriers to collecting pedestrian and bicycle safety performance data and developing per- formance-based decisions in the United States (e.g., legal, privacy restrictions, liability, data ownership, missing data, data incompatibility). • Domestic and international best practices that overcome barriers to collecting pedestrian and bicycle safety data, including traditional, nontraditional, and surrogate data sources at system- wide and local levels. • Processes used to access and link various pedestrian and bicycle data sources, including count data, trip data, hospital data, and crash data. • Methods for estimating pedestrian and bicycle exposure. • Methods for estimating pedestrian and bicycle safety performance. • Pedestrian and bicycle countermeasures. 2.1.1 Pedestrian and Bicycle Crash Statistics Pedestrian and bicycle crash statistics are summarized here to demonstrate the need for pedes- trian and bicycle safety to be explicitly incorporated in the HSM crash prediction methodology based on the magnitude of pedestrian and bicycle crashes throughout the United States. The summary statistics also serve to prioritize the types of locations and site types that should be given priority for SPF development, based upon where pedestrian and bicycle crashes typically occur on the network. Table 1 presents information on fatalities from motor vehicle traffic crashes in the United States from 2006 through 2019 (NHTSA 2021a). The table shows the number of total fatalities, the number of pedestrian fatalities, pedestrian fatalities as a percentage of total fatalities, the number of bicyclist fatalities, and bicyclist fatalities as a percentage of total fatalities. Since 2016, S E C T I O N 2 Literature Review and Survey of Practice

10 Pedestrian and Bicycle Safety Performance Functions pedestrian fatalities have eclipsed 6,000 annually. Before 2016, the last year that more than 6,000 pedestrian fatalities occurred in a year was 1990 when 44,599 fatalities occurred on the nation’s highways in the United States. Since 2006, the percentage of pedestrian fatalities to total fatalities and the percentage of bicyclist fatalities to total fatalities has been increasing. Figure 1 illustrates the environmental characteristics (i.e., land use and location) of where pedestrian and bicycle fatalities occurred in 2019 and 2018, respectively (NHTSA 2021b; Year Total Fatalities Pedestrian Fatalities Percentage of Pedestrian Fatalities to Total Fatalities Bicyclist Fatalities Percentage of Bicyclist Fatalities to Total Fatalities 2006 42,708 4,795 11% 772 1.8% 2007 41,259 4,699 11% 701 1.7% 2008 37,423 4,414 12% 718 1.9% 2009 33,883 4,109 12% 628 1.9% 2010 32,999 4,302 13% 623 1.9% 2011 32,479 4,457 14% 682 2.1% 2012 33,782 4,818 14% 734 2.2% 2013 32,893 4,779 15% 749 2.3% 2014 32,744 4,910 15% 729 2.2% 2015 35,484 5,494 15% 818 2.3% 2016 37,806 6,080 16% 853 2.3% 2017 37,473 6,075 16% 806 2.2% 2018 36,835 6,374 17% 871 2.4% 2019 36,096 6,205 17% 846 2.3% Table 1. Totals of pedestrian and bicyclist fatalities in traffic crashes from 2006 to 2019 (NHTSA 2021a). *Based on the location of the pedestrian or bicyclist struck at the time of the crash. NOTE: Percentages may not add up to 100 percent due to independent rounding. Unknowns were removed before calculating percentages. 2019 Land Use and Pedestrian Location 2018 Land Use and Bicyclist Location Figure 1. Percentage of pedestrian and bicyclist fatalities in relation to land use and location (NHTSA 2021b; NHTSA 2020).

Literature Review and Survey of Practice 11   NHTSA 2020). A majority of pedestrian fatalities occurred in urban areas (82 percent) and at nonintersection locations (73 percent). In this context, nonintersection locations mainly consist of roadway stretches between intersections (i.e., midblock locations), whereas locations such as parking lanes, bicycle lanes, and sidewalks are included in the “Other” category. Similarly, a majority of bicyclist fatalities occurred in urban areas (79 percent) and at nonintersection loca- tions (60 percent). 2.1.2 Barriers to Collecting Pedestrian and Bicycle Safety Performance Data This section describes barriers to collecting pedestrian and bicycle safety performance data and developing performance-based decisions in the United States (e.g., legal, privacy restrictions, liability, data ownership, missing data, and data incompatibility). It is important to understand these barriers so that the crash prediction methods developed as part of this project are consis- tent with the capabilities and data that the majority of agencies readily have available to them or can easily collect. It would be undesirable to develop SPFs that require data only a few agencies have access to or could collect because those SPFs would not be useful to a majority of agencies. A review of the research literature suggests that most barriers are related to the collection of pedestrian and bicyclist exposure data. The specific barriers to collecting pedestrian and bicyclist exposure data that an agency might face include: • Barriers related to various counting tools and methodologies. • Barriers in terms of the amount of time that sites can be counted and the number of sites that can be counted. • Difficulty developing or maintaining consistent documentation or database of counts. • Lack of available funding dedicated to such efforts. • Barriers related to permitting and other legal considerations. Each of these barriers is discussed below, after a short discussion of the specific types of data collection methods used to gather pedestrian and bicyclist exposure data. Additionally, barriers related to collecting crash data and infrastructure data are also presented. 2.1.2.1 Exposure Data 2.1.2.1.1 Methods to Collect Pedestrian and Bicyclist Exposure Data Various methods and technologies are used to collect pedestrian and bicyclist exposure data. Examples include manual on-site counts, manual video counts, infrared sensors, computer pro- cessing of video, inductive loops, pneumatic tubes, and piezometric pads. Many studies have been conducted to identify the advantages and disadvantages of these various data collection methods, which generally fall into the following categories: cost, labor, maintenance, accuracy, and appli- cability in different areas and under different conditions. Table 2 summarizes the advantages and disadvantages of various counting methods found by Ryan and Lindsey (2013). Among the methods and technologies described in Table 2, state transportation agencies most commonly use manual on-site counts, passive infrared, and manual video counts for count- ing pedestrians and pneumatic tubes for counting bicycles (FHWA 2011). FHWA conducted a webinar survey of over 100 practitioners (mostly from state transportation agencies and metro- politan planning organizations) to inquire about methods for collecting pedestrian and bicycle information. When surveyed, 61 percent of respondents said they use manual counting methods for pedestrians, and 49 percent said they use manual counting methods for bicyclists. Further, respondents who used technologies to conduct counts in the past were asked which technologies their agency used. The results are shown in Figure 2 and Figure 3. In addition to the approaches and technologies illustrated for collecting counts, several agen- cies have been experimenting with innovative approaches to collecting pedestrian and bicycle

12 Pedestrian and Bicycle Safety Performance Functions Counting Method Accuracy Cost Advantages Disadvantages Methods to Overcome Barriers Manual on-site counts Depends on training and human error (1%–25%) Labor - High Equipment - Low Ease of implementation • Requires trained observers • Only for short-term counts • Standardized training • Volunteers Video (Manual) High Labor - High Equipment - Medium Can capture user characteristics and directional counts N/A Generate funding and interest Video (Automatic) Depends on algorithms Labor - Low Software - High Good in crowded areas Requires higher data analysis skills Generate funding and interest Active infrared Undercounting due to occlusion (5%– 15%) Labor - Low Equipment - High • Portable • Can be used for long-term counts • Visible objects may affect counts • Directional counts not possible Generate funding and interest Passive infrared Undercounting due to occlusion (5%– 50%) Labor - Low Equipment - High Widely tested N/A Generate funding and interest Inductive loops Varies depending on installation Labor - Low Equipment - High Can be used for long-term counts • Can only be used for bicycle counts • Difficult to use in shared lanes Generate funding and interest Pneumatic tubes High Labor - Low Equipment - High • Can be used for long-term counts • Can provide speed estimates Potential tripping hazard Generate funding and interest Piezometric pads May not count groups accurately Labor - Low Equipment - High N/A N/A N/A NOTE: N/A = Not applicable. Table 2. Considerations for counting methods and technologies (adapted from Ryan and Lindsey 2013). Piezometric Pad 3% Active Infrared 23% Passive Infrared 26%Video (Automatic) 10% Video (Manual) 28% Other 10% Technology Usage for Pedestrian Counts Figure 2. Technologies used to conduct pedestrian counts (FHWA 2011).

Literature Review and Survey of Practice 13   count data. Some agencies have been using passive and/or active monitoring of mobile devices. Passive monitoring occurs when smartphone users do not initiate an application from their smartphone to be tracked, while active monitoring relies on smartphone users to initiate a fitness- based application (such as Strava) to gather walking, jogging, running, or bicycling activity. Both of these smartphone monitoring approaches have their strengths and limitations. Passive moni- toring can provide larger and less biased samples of pedestrians and bicyclists because it does not require the smartphone user to initiate location monitoring, but in some situations, it may be difficult to differentiate between travel modes (e.g., bicyclists in slow-moving motor vehicle traffic). Active monitoring can provide more detailed data about activity type (e.g., walking, jogging, bicycling, off-road bicycling) as well as demographics, but the activity data are from a much smaller and more biased sample (e.g., recreation-based activity) (S. Turner et al. 2017). Additionally, bikeshare programs collect large amounts of data on bicyclists’ activity, which could provide another means of estimating the change in bike usage over time (Martin and Shaheen 2014). However, bikeshare data often fail to capture all bicyclist activity (e.g., recreational trips) and thus cannot be used alone to estimate total bicycle exposure. Several agencies have also examined the feasibility of using pedestrian crosswalk push- buttons to estimate pedestrian counts at intersections. In this approach, the number of crosswalk actuations is factored up to account for cases in which more than one pedestrian crosses during each pedestrian crosswalk phase. While signal actuation counts are a noisy estimate of pedes- trian counts, this method may be useful for the measurement of relative changes in exposure (S. Turner et al. 2017). Numerous barriers prevent transportation agencies from collecting and documenting pedes- trian and bicyclist exposure data. Many of these barriers are associated with the chosen counting method, while others are related to temporal or spatial limitations, legal issues, documentation, and funding issues. 2.1.2.1.2 Barriers Specific to Counting Method NCHRP Report 797 (Ryus et al. 2014a) and associated online documents (Ryus et al. 2014b) discuss methods and technologies used to collect pedestrian and bicycle volume data. According to these documents, the two most commonly cited barriers to collecting pedestrian and bicycle data are cost (specifically, the costs of procuring instruments, setup, and maintenance) and staff Inductive Loops 15% Pneumatic Tubes 25% Active Infrared 10% Passive Infrared 17% Video (Automatic) 6% Video (Manual) 17% Other 10% Technology Usage for Bicycle Counts Figure 3. Technologies used to conduct bicycle counts (FHWA 2011).

14 Pedestrian and Bicycle Safety Performance Functions time (which is significantly high for manual counts). Exact costs associated with different meth- ods of data collection are not well-documented, and a case study of pedestrian and bicycle data collection in the U.S. found that there are misconceptions among practitioners about the costs associated with different data collection methods (Schneider et al. 2005). These misconceptions can lead to the selection of less cost-efficient methods or complete dismissal of a counting project. The misconception about costs often arises when comparing labor and technology costs. While manual on-site counts and manual video counts have very high labor costs, they have lower setup and maintenance costs than other technologies such as inductive loops and pneumatic tubes, which require significant investment to purchase and must be maintained regularly. With regards to automated counters, infrared sensors have relatively lower initial and maintenance costs than other technologies (FHWA 2011; Nordback et al. 2016; Ryan and Lindsey 2013; Ryus et al. 2014a). These automated counters can often be cheaper in the long run than prolonged manual counting. However, the trade-off between staff and technology costs is typically mis- judged when considering automatic technologies. Other barriers associated with the various count methods include error rates, detection issues, and usage limitations. While manual counts are the most common, the accuracy of this method can be affected by observer fatigue, and it cannot be used for long-term or permanent counts (Nordback et al. 2016; Ryus et al. 2014a). One study that compared simultaneous manual on-site counts to manual video counts indicated an average undercount error rate of 15 percent associated with manual on-site counts due to distraction, fatigue, and inexperience (Diogenes et al. 2007). Infrared sen- sors (both passive and active) have larger error rates than manual counts due to occlusion, which is the failure to detect individuals in groups or side-by-side pedestrians or bicyclists (FHWA 2011; Nordback et al. 2016; Ryan and Lindsey 2013; Ryus et al. 2014a). Infrared sensors have also been shown to have trouble detecting bicycles traveling at speeds higher than 10 to 15 miles per hour (mph) (Hudson, Qu, and Turner 2010). Additionally, active infrared sensors can experience interference from rain, leaves, animals, etc., and are best suited for indoor use (FHWA 2011). Inductive loop detectors experience accuracy issues because they count based on the detection of metal objects. For this reason, these detectors undercount bicycles made of composite materials and overcount if other vehicles pass over the sensors (FHWA 2011). Inductive loop detectors also frequently undercount bicycles traveling either near the edge of the lane or outside of the lane completely due to improper installation (Kothuri et al. 2012). These technological limitations, along with cost and staff time, are some of the primary bar- riers to conducting pedestrian and bicycle counts. Difficulty identifying reliable technologies for long-term counts is another limitation of counting methods that leads to temporal issues (Nordback et al. 2016). 2.1.2.1.3 Temporal Barriers for Exposure Data Manual, 2-hour counts are currently the most common method to obtain pedestrian and bicyclist exposure data. However, short-term counts do not provide daily and annual pedestrian volume data to assess the safety performance of a facility (ISW8 2017). To account for time- of-day variations (e.g., commuting vs. noncommuting hours), longer-duration counts should be conducted and appropriate adjustment factors developed to extrapolate 2-hour counts to 24-hour counts. The TMG recommends that 12-hour counts be used to create a time-of-day profile since 2-hour counts may yield high error rates (FHWA 2016). Additionally, volumes will vary for facilities based on day of the week (e.g., weekday vs. weekend) and time of year (e.g., summer vs. winter). Adjustment factors to account for these variations cannot be developed without access to permanent count data over the course of a week or year. The NCHRP Report 797 recommends, at the most basic level, permanent counts for at least one location in a jurisdic- tion to observe these temporal patterns (Ryus et al. 2014a).

Literature Review and Survey of Practice 15   In a survey of over 60 practitioners, FHWA found that 87 percent had no experience extrapo- lating short-term counts over longer periods of time (FHWA 2011). Hence, additional training is necessary to not only conduct accurate counts but on how to develop and use adjustment factors to extrapolate short-term counts to long-term volumes. The costs and technological limitations associated with conducting long-term counts and extrapolating short-term counts are temporal barriers to obtaining accurate pedestrian and bicyclist exposure data. Similar to the temporal barriers arising from when counts are conducted, spatial barriers arise due to the locations chosen for counts. 2.1.2.1.4 Spatial Barriers for Exposure Data In a webinar featuring more than 300 attending practitioners, identifying count locations was one of the top responses received to the question “What problems have you encountered in trying to count pedestrians?” (Nordback et al. 2016). Since it is not feasible to conduct counts at every intersection or along every road, specific locations must be selected for conducting counts, and these will cover only a small portion of the jurisdiction’s road network. When continuous count stations are not available, locations chosen for short-term counts should be representative of the rest of the jurisdiction, which frequently requires input from stakeholders and agencies. Biased site selection can also be a barrier to conducting pedestrian and bicycle counts. Often, counts are only conducted at sites where a specific project is being considered for funding from the transportation agencies. Counts that reveal too few pedestrians or bicyclists at a location often make it difficult to justify additional spending or improvements at that location (Schneider et al. 2005). This presents an issue in that low numbers of pedestrian and bicycle traffic could indicate intimidating roadway conditions that warrant improvements, but these locations are not likely to be included in a count due to low volumes. Nordback et al. (2016) interviewed various practitioners about their experiences with count- ing nonmotorized traffic, and multiple interviewees commented that the TMG should include pedestrian-specific site selection advice. The lack of standardization methods for identifying sites for short- or long-term counts presents a significant barrier to conducting counts by making it difficult to justify spending and garner support for programs. Additionally, specific geometric constraints may inhibit the use of specific equipment at locations of interest. Count location also brings about a legal barrier related to permitting and rights-of-way. 2.1.2.1.5 Legal Barriers for Exposure Data The primary legal barrier to conducting pedestrian and bicycle counts is permitting. In prepar- ing to conduct pedestrian and bicycle counts, permits often need to be applied for and purchased to install counting devices (Ryus et al. 2014a). Permits may also be required for right-of-way encroachment, pavement cutting (in the case of inductive loop counting), and landscaping. The permitting process can be burdensome as it adds both time and cost to a project and requires advanced planning as well as support for the counting project. Another legal barrier that could arise is privacy concerns with respect to video technologies. Some jurisdictions may not permit the use of video technologies for counting, which may pose another barrier if this were the selected or preferred counting method (Ryus et al. 2014a). 2.1.2.1.6 Documentation Barriers for Exposure Data The lack of standard methods for data reporting and data storage poses a barrier to collecting pedestrian and bicyclist exposure data. In 2016 the Travel Monitoring Analysis System of FHWA was expanded to include bicycle and pedestrian counts; however, this database is not yet compre- hensive (FHWA 2016). Hence, a detailed national database for bicycle or pedestrian counts does

16 Pedestrian and Bicycle Safety Performance Functions not currently exist, and existing databases cannot accommodate continuous counts (Nordback et al. 2016). Only a few agencies in the United States (e.g., Los Angeles County) have created their own online clearinghouse for bicycle count data. Unfortunately, a common database cannot be achieved without standard reporting procedures, including a list of data that must be collected (Griffin et al. 2014). A lack of standardization and a national database for reporting and collect- ing count results presents a barrier to developing an association between nonmotorized traffic exposure data and safety performance. 2.1.2.1.7 Funding Barriers for Exposure Data While cost is the primary barrier to collecting pedestrian and bicyclist exposure data, it is magnified by funding issues. It is often difficult to secure funding for projects because of low management interest and a lack of understanding of the value of counting pedestrians (Nordback et al. 2016). In 2012, students at the University of Minnesota conducted a study of all 50 state transportation agency websites for information about pedestrian and bicycle counting initiatives. The results revealed that three states (Colorado, Vermont, and Washington) showed extensive support for nonmotorized traffic monitoring, 30 states showed some support, and 17 states showed little or no support (Lindsey et al. 2013). With such little support from state transportation agencies, establishing permanent and institutionalized count programs is difficult. 2.1.2.2 Practices for Overcoming Barriers to Collecting Pedestrian and Bicycle Count Data While many cities and counties have begun developing pedestrian and/or bicycle counting programs, most transportation agencies in the U.S. do not yet have mature programs. In a recent survey, FHWA found that 72 percent of transportation agency personnel reported three or fewer years of collecting motorized or nonmotorized count data, and of this, only 56 percent of agen- cies reported that they have started collecting pedestrian or bicycle count data (FHWA 2011). Despite the low numbers of agencies conducting pedestrian and bicycle counts, various practices have been established to reduce costs, build support for counting programs, make the process of conducting counts easier and more comprehensive, and improve count accuracy. 2.1.2.2.1 Practices for Decreasing Costs Since costs are the biggest barrier to conducting pedestrian and bicycle counts, it is important to establish cost-efficient methods to conduct pedestrian/bicycle counts, as well as alternative ways of securing funding to implement counting programs. The most prevalent way that agencies have saved money on counts is by outsourcing labor. Many agencies have had success recruiting volunteers to conduct counts. Volunteers can be recruited from local universities, community groups, advocacy groups, and other places. For example, Cheyenne, Wyoming, recruited local Boy Scouts to conduct over 40 pedestrian counts on the Cheyenne Greenway in 1996 and 1997 with no direct costs to the city (Schneider et al. 2005). In addition to volunteers, the Minnesota DOT has engaged a variety of stakeholders, including the Department of Public Works, Transit for Livable Communities (a nonprofit organization), recreational agencies, and the Minneapolis Park and Recreation Board, to perform counts (Lindsey et al. 2013). Nonprofit organizations and other state or city agencies and groups can be great places to find outside labor. Another way labor can be outsourced is through crowdsourcing. In a study conducted at Washington University in St. Louis, the Amazon Mechanical Turk website was used to crowd- source manual video counts. Workers were paid a minimal amount to count pedestrians and bicyclists in video frames, and each frame was counted five unique times (Hipp et al. 2013). This provides both inexpensive and accurate counts. Outsourcing labor to perform counts can

Literature Review and Survey of Practice 17   significantly reduce the costs of conducting manual counts and engage citizens and grow support for counting programs. 2.1.2.2.2 Practices for Building Support for Counting Programs and Procuring Funding Public and political support is a vital part of successful pedestrian and bicycle counting pro- grams because funding is needed to support the program. It was previously discussed that only three states show signs of extensive support for counting programs, while 17 states show little or no support. Despite these low numbers, some transportation agencies have developed practices aimed at garnering support from the government and communities. Colorado was one of the three states showing extensive support for nonmotorized traffic mon- itoring. Colorado has developed its counting programs around the concept of active transport. A report prepared for Kaiser Permanente Colorado’s Active Transportation program suggests establishing a website with implementation guidelines, instructions, and a data clearinghouse to build local support for counts. By promoting counting initiatives as measures of active transport, municipalities can gather public support for their programs by stressing the health, livability, and environmental implications of the counts (Krizek and Forsyth 2012). North Carolina Department of Transportation (NCDOT) uses a model of partnering with local agencies to install, monitor, and maintain continuous count stations. This program includes training for agencies and coordinating installation among agencies. NCDOT installs and owns the counters for the first 2 years. During these 2 years, NCDOT conducts quality assurance and quality control to determine the need for maintenance and conducts routine maintenance. The local agencies own and maintain the equipment after the 2 years are over (O’Brien et al. 2016). Another important aspect of building support for counting programs, and ultimately procur- ing funding, is being able to justify spending. Agencies use different methods to do this, primarily tracking data usage and developing clear political purposes for their data collection projects. Iowa DOT has recently been proactive in long-range planning for bicycle transportation. To justify spending on facilities, they track their usage of count data to show its importance and to justify the need for ongoing collection (Schneider et al. 2005). Alternatively, for new studies, identifying a clear political purpose for data collection can garner support from local govern- ments. For example, many cities and counties calculate the economic impacts of pedestrian and bicycle facilities, provide objective data about facility usage, and evaluate the need for improved facilities to justify spending and ensure continued support of programs (Schneider et al. 2005). The previous section on barriers identified that it is difficult to obtain support and funding for counts at locations with low volumes. One practitioner who was involved in FHWA’s webinar recommended installing counters for the first time in locations with high activity levels to help build political support for the counting program. By doing so, support can be garnered early on from the large volumes counted, and low-volume locations can be added later to increase net- work coverage (Nordback et al. 2016). Finally, though somewhat difficult to establish, institutionalized counting programs help build community and political support through continuity. Minnesota Department of Transportation (MnDOT) has established an annual count program with volunteers, which includes a statewide count once a year. This program has increased community interest in coordinated count efforts (Lindsey et al. 2013). Data collection over time also provides an opportunity for monitoring progress and ensures that data are available to staff, officials, and the public (Schneider et al. 2005). The City of Portland, Oregon, uses bicycle loop detectors to continuously count bicycles and archives data for future analysis. This practice allowed researchers from Portland State University to access the data and observe various trends related to am and pm peak travel and seasonal- and weather-related trends (Kothuri et al. 2012). Access to this type of data can be

18 Pedestrian and Bicycle Safety Performance Functions useful for operations and planning purposes (e.g., time-dependent signal timing plans or plan- ning for special events). While building support is one of the best ways to secure funding, some agencies have developed nongovernmental methods of funding. One example of securing alternative funding options comes from MnDOT. MnDOT has received grants from the State Health Improvement Program to conduct their counts by proposing the counts as a means of measuring active transport and healthy living (Lindsey et al. 2013). While some sources of funding will focus on improving exist- ing systems, other sources of funding (such as innovation funds) can be sought when looking to establish new programs (Krizek and Forsyth 2012). Securing funding is the most important step to launching a counting program, and ensuring future funding is equally important. For this reason, it is important to establish ways of building support for counting programs. 2.1.2.2.3 Practices for Making Counts Easier and More Comprehensive Agencies and practitioners have developed a variety of practices aimed at easing the counting process to establish long-term programs and make counts and data collection as a whole more comprehensive. Many agencies have begun by integrating pedestrian and bicycle counts into existing motor vehicle count programs (Schneider et al. 2005). Additionally, some agencies have found that nonmotorized data have already been collected on past projects and traffic studies and simply were not stored due to the lack of formal means for reporting (Ryus et al. 2014a). By integrating pedestrian and bicyclist counts into existing programs, these counts can be obtained with little additional cost and time investment. Another way transportation agencies have eased the counting process is by developing stan- dard training materials and count forms for volunteers. Using volunteer labor is shown to decrease costs for a count; however, volunteers can require significant training before conduct- ing the count. MnDOT developed training materials and standard forms for manual counts of pedestrians and bicyclists and then held webinars to train volunteers (Lindsey et al. 2013). Having standard training materials and reporting sheets speeds up the process of training volunteers and reduces staff time spent for each count. The development of standard reporting sheets for statewide use is also a good practice to aid the development of a common storage database. The City of Seattle has conducted manual, quarterly bicycle counts at 50 locations since 2011 (Nordback et al. 2017). All of these counts are accessible on the website for Seattle’s Open Data Program, which seeks to improve understand- ing of city operations, generate opportunities for people to benefit from data like bicycle counts, and encourage development of innovative technology solutions to improve the quality of life (Matmiller 2016). Counts on this site are organized into simple tables showing location and observed volume. Counts are always conducted in the same four months in the same three time- frames to generate similar output formats that can easily be stored in a common database. This type of standardization is an example of how an organization can successfully store count data. Last, many agencies recommend linking data from multiple sources to develop a more robust and comprehensive database of pedestrian and bicycle performance data. NCHRP Report 797 recommends taking advantage of data from transportation agencies that conduct traffic studies and already have developed count databases with georeferenced count locations (Ryus et al. 2014a). In their Consensus Recommendations for Pedestrian Injury Surveillance report, the Injury Surveillance Workgroup 8 (ISW8) recommended linking data from behavioral surveys, emer- gency room data, and police reports as well (ISW8 2017). Multiple agencies have successfully used surveys to aid their data collection efforts. For example, Colorado uses randomly sampled, local-level surveys as an inexpensive way to reach a wide population and gather data about pedestrian and bicycle trips (Krizek and Forsyth 2012).

Literature Review and Survey of Practice 19   Pinellas County, Florida, has also used surveys to determine trail usage for transportation pur- poses (Schneider et al. 2005). The Institute for Transportation Research and Education (ITRE) also collected data from continuous counters and linked it to manual counts and survey data to obtain a richer dataset on four shared-use paths in North Carolina (ITRE 2018). Linking data from multiple sources is good practice when seeking to collect more information about pedes- trian and bicycle usage other than counts. 2.1.2.2.4 Practices for Improving Count Accuracy Quality assurance practices are not frequently used in conducting pedestrian and bicycle counts due to added cost. One easy method of quality control is to have more than one counter at a loca- tion, but this is rarely used due to high associated labor costs (Hudson, Qu, and Turner 2010). Though there are not many ways of improving the accuracy of counting technologies, quality assurance practices can be used to flag potential errors in count data. For example, NCDOT has a quality assurance and quality control program that involves ensuring quality during site selec- tion, procurement, installation, and onboarding. Their program checks the data quarterly and flags for potential equipment malfunction using gap checks, identifying consecutive zeros, iden- tifying data outside of the standard range, and considering directional distribution (Jackson et al. 2017). In another study, researchers working with the National Park Service and San Antonio River Authority used interquartile ranges to compare northbound and southbound counts at a location to flag outliers in the data by identifying cutoff points of large differences between the counts (S. Turner and Lasley 2013). It is recommended that, at a minimum, a visual inspection of the data be conducted to identify potential count errors. Overall, many practices have been developed to reduce costs, build support for funding, and improve the data collection process for monitoring pedestrian and bicycle traffic. A variety of actions can be taken to accomplish these goals, but many of these practices still suffer from downfalls, primarily when it comes to count accuracy and reporting and storing data. However, these practices are good ways to establish new counting programs and will eventually help the development of long-term count programs. 2.1.2.3 Barriers to Collecting Pedestrian and Bicyclists Crash Data Similar to motorized modes, crash data for nonmotorized modes are primarily collected using police crash records. However, underreporting of such data, especially if there are no injuries, is well-documented (Elvik and Mysen 1999). Since the majority of pedestrian or bicycle crashes result in injuries—due to the vulnerable nature of these road users—underreporting might be less prevalent for these nonmotorized modes compared to motorized modes (Ward, Lyons, and Thoreau 2006). For example, studies estimate that between 60 to 90 percent of motor vehicle crashes are underreported (Tin et al. 2010), while early studies conservatively estimate that underreporting of nonmotorized crashes by police may be as large as 20 percent for pedestrians and 10 percent for bicyclists (Agran, Castillo, and Winn 1990). One of the solutions to overcome underreporting of pedestrian and bicyclist crashes is to sup- plement police crash records with additional data sources. Hospital data are one potential source since people with major injuries often visit the hospital, and this information can then be linked to crash databases. However, there are barriers to linking multiple databases together to obtain a more comprehensive crash database and fill the gap in crash records (Children’s Hospital of Philadelphia Research Institute 2018). One of the main obstacles is the lack of unique identify- ing information that can link records across the various databases (Conderino et al. 2017; Cook et al. 2015; Children’s Hospital of Philadelphia Research Institute 2018). Another significant obstacle is the issue of privacy since linking hospital data and crash records could reveal iden- tifying information about specific individuals. Hence, combining crash and hospital datasets

20 Pedestrian and Bicycle Safety Performance Functions requires careful use of data and institutional review board (IRB) approvals. In the United States, the Health Insurance Portability and Accountability Act prohibits the use of names, social secu- rity numbers, or vehicle identification numbers in medical records unless waived by an IRB (Clark 2004). Other obstacles and barriers to linking crash data with additional data sources include (Nebraska’s Traffic Records Coordination Committee 2016; Milani et al. 2015; Curry, Pfeiffer, and Metzger 2017): • Lack of uniformity across local jurisdictions or lack of an electronic database (both of which are often the case for traffic citation information that could help identify minor crashes). • Lack of resources or knowledge to link the very different datasets. • Lags in obtaining data. • Lack of documented procedures. • Administrative barriers. Best practices to overcome these barriers, especially those related to linking hospital data and crash records, are further discussed in Section 2.1.3.3. 2.1.2.4 Barriers to Collecting Infrastructure Data Another barrier to collecting pedestrian and bicyclist performance data is related to the lack of detailed infrastructure data for pedestrian and bicyclist facilities. This includes the locations of bicycle paths, bicycle lanes, sidewalks, pedestrian crossing signals, bicycle signals, etc. These elements are often not included in roadway inventory databases maintained by state or metro- politan agencies. Instead, they are generally only known to local agencies, which may or may not have detailed databases to store this information. In cases when these data do exist, it is often incomplete due to a lack of standard methods for storing this information. Thus, pedestrian and bicyclist infrastructure information may be available for only a portion of the roadway network. If collected, local agencies often do not update this information regularly; instead, data elements may be updated based on specific projects that are being performed within a municipality. The quality of the data is often variable, with some elements/locations having more updated infor- mation than others. Even in cases where databases do exist, they often do not contain all of the pertinent details about these infrastructure elements (e.g., width of sidewalks or bicycle lanes), and they may not be well maintained such that new changes to the infrastructure are not repre- sented in these databases. The availability of high-quality aerial photographs through online platforms (such as Google Maps or similar software) helps alleviate these barriers to some extent. The presence of many infrastructure elements can be observed using aerial photography (e.g., presence of sidewalk or bicycle lanes) or even “street view” photos (e.g., pedestrian countdown signals or bicycle signals), although doing so requires considerable time and resources to update and maintain a comprehensive database. Construction records can also be used to provide this information. Even though the time and resources required to do this are considerably larger, these records can provide detailed information that might not be otherwise available from aerial photography or without a site visit. 2.1.3 Processes to Link Pedestrian and Bicycle Data Sources This section briefly summarizes approaches to extrapolating short-term count data to develop estimates of weekly, monthly, or annual exposure; several nontraditional data sources that have been used to enhance traditional count data; and information on linking hospital data with police crash reports.

Literature Review and Survey of Practice 21   2.1.3.1 Count Data Most commonly available pedestrian and bicycle count data are collected for a short period around motor vehicle peak daytime travel periods. These short-duration counts are less directly useful for safety analyses because they primarily represent exposure for the time period and time of year when the data were collected. Longer-duration counts or estimates (e.g., daily volumes) are preferred for exposure because they can be better compared with annual crash data. Short- term counts, however, can be adjusted based on long-term counts at similar locations to develop estimates of weekly, monthly, or annual exposure. A number of studies have used automated count data from a limited number of locations to create expansion factors to extrapolate short- term counts at other locations to long-term estimates (Schneider, Arnold, and Ragland 2009a; Schneider et al. 2012; Ryus et al. 2014a; Liggett et al. 2016). This process is based on the assump- tion that similar sites have similar patterns in hourly, weekly, and monthly volume trends. Even if the magnitude of volumes is different, the proportional trends can be similar. Ryus et al. (2014a, 2014b) provides information on the development of expansion factors using long-term duration counts, but general expansion factors are also available from the National Bicycle and Pedestrian Documentation Project for agencies that do not have access to their own automated count data (Alta Planning + Design 2016). However, general expansion factors should be used with caution. 2.1.3.2 Trip Data Trip data have traditionally been available only from self-reported survey sources, like the NHTS or regional surveys used to develop regional travel models. More recently, trip data have become available from smartphone tracking applications that can provide more detailed and accurate information on trips. Strava, a smartphone application for tracking running and bicy- cling activity, makes an aggregate dataset called Strava Metro that is commercially available. By overlaying all bicycle trips in an area, Strava can provide volumes on road segments. While these data are biased in favor of recreational trips, they can be fused with other sources to enhance volume estimates (Proulx and Pozdnukhov 2017). Another example of a crowdsourced dataset is data collected by the CycleTracks smartphone application developed by the San Francisco County Transportation Authority. Griffin and Jiao (2014) integrated automated counts at five locations with Strava Metro and CycleTracks data to evaluate the suitability of these crowd- sourced data for volume estimates and found them to be promising. Also, bikeshare programs’ data could be used as an estimate of changes in bicycle usage. 2.1.3.3 Hospital and Crash Data Linking hospital and crash data has several benefits for safety analysis, including connecting crash conditions to injury outcomes and providing insights on the accuracy of crash severity documented on policy crash reports. This approach has been taken by international public health researchers for many years (Bull and Roberts 1973; Rosman 1996; Cercarelli, Rosman, and Ryan 1996). Table 3 summarizes the information on database linking for some domestic and international studies that looked at pedestrians and bicyclists. Most of the international studies do not specify a linking method used. Where linking methods were described, in most cases, efforts were made to link police crash reports and hospital records using data such as age, gender, date, time, and location. In addition to the studies listed in Table 3, the National Highway Traffic Safety Administration (NHTSA) facilitated the Crash Outcome Data Evaluation System (CODES) program in the United States from 1992 to 2013 to establish a probabilistic linking methodology for augment- ing state crash data with medical outcome data (Cook et al. 2015). CODES uses a probabilistic methodology to link crash records to injury outcome records collected at the scene and en route by emergency medical services, by hospital personnel after arrival at the emergency depart- ment or admission as an inpatient, and/or at the time of death on the death certificate. State CODES programs became fully autonomous in 2013. Fifteen states maintain an active CODES

22 Pedestrian and Bicycle Safety Performance Functions program. The CODES program addresses all motor vehicle crash types, including pedestrian and bicycle crashes. Another solution to overcoming the barrier of underreporting crashes is to use crowdsourced data. Information from the community can be obtained to improve upon the knowledge of crashes (Schnedier, Khattak, and Zegeer 2001; Medury et al. 2017), or social networks can be monitored to obtain more data on bicycle and pedestrian crashes (Schulz, Ristoski, and Paulheim 2013; Mai and Hranac 2013). 2.1.4 Methods for Estimating Pedestrian and Bicycle Exposure As described above, highway agencies experience several common barriers related to collecting pedestrian and bicycle safety performance data, including a lack of pedestrian and bicyclist expo- sure data. When and where counts are conducted, they are often short-term counts at a few select locations on the network. Sometimes these short-term counts are expanded to weekly, monthly, or annual exposure using expansion factors, but they are still only applicable to specific locations. To overcome some of the barriers to collecting count data and to obtain estimates for a larger portion of the network, several methods have been developed for estimating pedestrian and bicycle volumes at intersections and along roadway segments. This section describes several methods for estimating pedestrian and bicyclist exposure at intersections and along roadway segments. NCHRP Report 770: Estimating Bicycling and Walking for Planning and Project Development: A Guidebook provides a summary of pedestrian and bicycle demand modeling research, high- lighting three general categories of models that provide facility-level volume estimates at road- way intersections and network segments (Kuzmyak et al. 2014). • Trip generation and flow models. This approach estimates the number of pedestrian or bicycle trips between small areas, such as block faces or analysis zones. These models follow a traditional travel modeling approach since they estimate trip generation, trip distribution, and network assignment. Ultimately, trips assigned to the pedestrian or bicycle network are totaled for specific intersections and segments. One study applied the traditional travel model approach to block-sized pedestrian analysis zones in central Baltimore, Maryland (Clifton et al. Reference Country Road User Type Linked Databases Linking Method Pedestrians Sciortino et al. 2005 United States Pedestrians Police and hospital Age, gender, date, and time Loo and Tsui 2009 China Pedestrians Police and hospital Linked database Tarko and Azam 2011 United States Pedestrians Police and hospital Not specified Bicyclists Leonard, Beattie, and Gorman 1999 Scotland Child bicyclists Police and hospital Not specified Langley et al. 2003 New Zealand Bicyclists Police and hospital Probabilistic record linkage (AUTOMATCH) Veisten et al. 2007 Norway Bicyclists Police and hospital Not specified Isaksson-Hellman 2012 Sweden Bicyclists Insurance, police, and hospital Not specified Lopez et al. 2012 United States Bicyclists Police and hospital Probabilistic record linkage Tin, Woodward, and Ameratunga 2013 New Zealand Bicyclists Police, hospital, insurance, and mortality Name, gender, date of birth, and address Pedestrians and Bicyclists Stutts and Hunter 1998 United States Pedestrians and bicyclists Police and hospital Age, gender, location, date, and time Cryer et al. 2001 England Pedestrians and bicyclists Police and hospital Name, date, and location Table 3. Pedestrian and bicycle studies linking police and injury data.

Literature Review and Survey of Practice 23   2008), and a similar approach was developed using small grid cell pedestrian analysis zones in Portland, Oregon (Clifton et al. 2015). These models require travel survey data. • Network simulation models. This category of models, including Space Syntax, develops volume estimates for each part of a pedestrian network based on network characteristics such as connectivity and sight lines. In some cases, these network variables are combined with land- use variables to estimate pedestrian volumes (Raford and Ragland 2004; Raford and Ragland 2005). These models are often complex and require proprietary software to apply. • Direct demand models. These models estimate volumes along roadway segments and inter- sections using site and surrounding area characteristics. Street block face or midblock count data have been used to model pedestrian volumes in New York City (Pushkarev and Zupan 1971), Milwaukee, Wisconsin (Benham and Patel 1977), and Minneapolis, Minnesota (Han- key et al. 2012; Hankey and Lindsey 2016). However, more recent direct demand pedestrian volume models have been developed from intersection crossing counts (Pulugurtha and Repaka 2008; Schneider, Arnold, and Ragland 2009b; Liu and Griswold 2009; Haynes and Andrzejewski 2010; Jones et al. 2010; Miranda-Moreno and Fernandes 2011; Schneider et al. 2012; Grembek et al. 2014) and bicycle intersection volumes (Griswold, Medury, and Schnei- der 2011; Strauss and Miranda-Moreno 2013). The direct demand modeling approach is most commonly used to estimate exposure because the models are simple to understand, do not require complex computer applications to execute, and are straightforward to apply. Typical steps used in the direct demand approach to estimate pedestrian exposure are described below. 1. Pedestrian counts are taken at a sample of locations in a community. These counts are often collected manually over short periods of time, but automated detection techniques that collect data over weeks, months, or even years can also be used. 2. Short-term counts may be expanded to represent annual volume estimates, using the approach described in Ryus et al. (2014a). 3. The annual (or other duration) pedestrian volumes are used as the dependent variable in a predictive model. Statistical software is used to identify significant relationships between the pedestrian volumes at each study location and explanatory variables describing the charac- teristics of the study location (e.g., land-use characteristics, transportation system features, demographic factors, or any other factors thought to be relevant to pedestrian volumes). 4. The preferred statistical model can be used to estimate pedestrian volumes in other locations throughout the community. Table 4 summarizes several recent direct demand pedestrian and bicycle volume models. Many of these models are based on short counts (ranging from 2 to 12 hours) and are only appropriate for estimating pedestrian or bicycle volumes during the specific times of day (e.g., afternoon peak period) or seasons of the year when the counts were taken. However, models developed for San Francisco (Schneider et al. 2012) and the California state highway system (Grembek et al. 2014) are based on counts extrapolated to annual volumes to produce estimates of full-year pedestrian volumes. Many early applications of this approach used linear regression modeling. While simple to understand, this approach can produce unrealistic, negative volume estimates, so most recent studies have used log-linear and negative binomial model structures. Explanatory variables that have been most commonly statistically significant in these models include (Pulugurtha and Repaka 2008; Schneider, Arnold, and Ragland 2009b; Liu and Griswold 2009; Haynes and Andrzejewski 2010; Jones et al. 2010; Miranda-Moreno and Fernandes 2011; Schneider et al. 2012; Hankey and Lindsey 2016): • Surrounding population density. • Surrounding jobs or employment density. • Proximity to transit.

24 Pedestrian and Bicycle Safety Performance Functions General Information Count Information Model Information Model Location Source Type of CountSites Count Period(s) Used for Model Model Output Model Type Charlotte, NC UNC Charlotte (Pulugurtha and Repaka 2008) Signalized intersections 7 a.m.–7 p.m. Total pedestrians approaching intersections from 7 a.m.–7 p.m. (shorter periods also modeled) Linear Alameda County, CA UC Berkeley SafeTREC (Schneider, Arnold, and Ragland 2009b) Signalized and unsignalized arterial and collector roadway intersections Tu, W, or Th: 12–2 p.m. or 3–5 p.m.; Sa: 9–11 a.m., 12–2 p.m., or 3–5 p.m. Total pedestrian crossings at arterial and collector roadway intersections during a typical week Linear San Francisco, CA San Francisco State (Liu and Griswold 2009) Signalized and unsignalized intersections Weekdays 2:30–6:30 p.m. Total pedestrian crossings at intersections from 2:30– 6:30 p.m. on typical weekday Linear Santa Monica, CA Fehr and Peers (Haynes and Andrzejewski 2010) Signalized and unsignalized intersections Weekdays 5–6 p.m. Total pedestrian crossings at intersections from 5-6 p.m. on typical weekday Linear San Diego, CA Alta Planning + Design (Jones et al. 2010) Signalized and unsignalized intersections (includes some trail/roadway intersections) Weekdays 7–9 a.m. Total pedestrians approaching intersections from 7–9 a.m. Log-linear Montreal, Quebec McGill University (Miranda-Moreno and Fernandes 2011) Signalized intersections Weekdays –-9 a.m., 11 a.m.–1 p.m., and 3:30– 6:30 p.m. Total pedestrian crossings at intersections over 8 count hours (shorter periods also modeled) Log-linear (also used negative binomial) Alameda County, CA UC Berkeley SafeTREC (Griswold et al. 2011) Signalized and unsignalized arterial and collector roadway intersections Tu, W, or Th: 12–2 p.m. or 3–5 p.m.; Sa: 9–11 a.m., 12–2 p.m., or 3–5 p.m. Total bicycle volume at arterial and collector roadway intersections during a typical week Log-linear San Francisco, CA UC Berkeley SafeTREC (Schneider et al. 2012) Signalized and unsignalized intersections Tu, W, or Th: 4–6 p.m., extrapolated to annual volumes Total pedestrian crossings at intersections during a full year Log-linear Minneapolis, MN University of Minnesota (Hankey et al. 2012) Midblock locations along sidewalks and multiuse trails September 12-hour (6:30 a.m.–6:30 p.m.) counts and 2-hour counts (4–6 p.m.) extrapolated to 12-hour counts Total pedestrian and bicycle volumes on roadway and trail segments during a 12-hour period in September Negative binomial Montreal, Quebec McGill University (Strauss and Miranda-Moreno 2013) Signalized intersections Weekdays 6–9 a.m., 11 a.m.–1 p.m., and 3:30– 6:30 p.m. Average seasonal daily bicycle flows Unknown California UC Berkeley SafeTREC (Grembek et al. 2014) Intersections along urban arterials in the State Highway System Various 2-hour and 4- hour periods on weekdays and weekends extrapolated to annual volumes Total pedestrian crossings at intersections during a full year Log-linear Minneapolis, MN University of Minnesota (Hankey and Lindsey 2016) Midblock locations along sidewalks and multiuse trails September: 4–6 p.m. Total pedestrian and bicycle volumes on roadway and trail segments from 4-6 p.m. on September days Log-linear NOTE: UNC = University of North Carolina; UC = University of California; TREC = Transportation and Research Center. Table 4. Direct demand pedestrian and bicycle volume models.

Literature Review and Survey of Practice 25   2.1.5 Methods for Estimating Pedestrian and Bicycle Safety Performance The research team identified several examples of pedestrian and bicycle SPFs developed directly from crash data for application at the local, state, and national levels. Additional examples of agen- cies that have applied other approaches to estimating pedestrian and bicycle safety performance were also identified, along with several tools developed to prioritize pedestrian and bicycle infra- structure improvements. These examples are summarized in Sections 2.1.5.1 through 2.1.5.3. 2.1.5.1 Pedestrian Safety Prediction Methodology from NCHRP Project 17-26, “Methodology to Predict the Safety Performance of Urban and Suburban Arterials” In Phase III of NCHRP 17-26, Harwood et al. (2008) developed a pedestrian safety prediction methodology for urban and suburban signalized intersections for use in the HSM. The SPFs for estimating pedestrian crashes at three- and four-leg signalized intersections have been incorpo- rated into Chapter 12 of the HSM. Data for pedestrian crashes at signalized intersections on urban and suburban arterials were obtained from Toronto, Ontario, Canada, and Charlotte, North Carolina. These cities were chosen primarily because they had databases of both vehicle volumes and pedestrian volumes at the intersections as well as crash records for pedestrian crashes. Through a combination of available databases, field visits, and a review of aerial images of the intersections, the following variables were obtained for consideration in the modeling process: • Number of intersection legs. • Intersection skew angle. • Number of through lanes. • Number of right-turn lanes. • Number of left-turn lanes. • Presence of marked crosswalks. • Presence of median. • Presence and type of pedestrian signal. • Presence and type of turn restrictions. • Annual average daily traffic (AADT) by intersection leg. • Pedestrian crossing volume by leg. Separate safety prediction models using negative binomial regression were developed for Toronto and Charlotte. Eight functional forms were considered including various combinations of vehicle and pedestrian volume variables. For example, some models used a ratio of minor- approach volume to major-approach volume, while others used these volumes as separate vari- ables. After these volume-only models were developed, additional variables were considered and tested, including: • Maximum number of lanes crossed by a pedestrian on any intersection leg. • Number of legs with refuge islands. • Number of intersection legs with marked crosswalks. • Presence of skewed intersection leg. The only variable that was statistically significant with an effect in the expected direction was the maximum number of lanes crossed by a pedestrian on any intersection leg considering the presence of refuge islands. Eventually, the models developed for Toronto and Charlotte were combined, and adjustment factors were developed for the number of bus stops, schools, and alcohol sales establishments within 1,000 ft of the intersection.

26 Pedestrian and Bicycle Safety Performance Functions Equation 2-1 is the base model for three-leg signalized intersections in the final form. (2-1)exp= . . . . .ln ln lnN AADT AADT AADT PedVol n6 60 0 05 0 24 0 41 0 09minped tot maj lanesx- + + + + J L K KK J L K K N P O O N P O OO Equation 2-2 is the base model for four-leg signalized intersections in the final form. exp= . . . . .ln ln lnN AADT AADT AADT PedVol n9 53 0 40 0 26 0 45 0 04minped tot maj lanesx- + + + + J L K KK J L K K N P O O N P O OO (2-2) where: Nped = Predicted average crash frequency of pedestrian crashes, AADTtot = Sum of the annual average daily traffic volumes (veh/day) for the major and minor roads (= AADTmaj + AADTmin), AADTmaj = Annual average daily traffic volume (veh/day) for major road (both directions of travel combined), AADTmin = Annual average daily traffic volume (veh/day) for minor road (both directions of travel combined), PedVol = Sum of daily pedestrian volumes (ped/day) crossing all intersection legs, and nlanesx = Maximum number of traffic lanes crossed by a pedestrian in any crossing maneuver at the intersection considering the presence of refuge islands. These base models apply to signalized intersections with the following base conditions: • No bus stops within 1,000 ft of the intersection. • No schools within 1,000 ft of the intersection. • No alcohol sales establishments within 1,000 ft of the intersection. 2.1.5.2 SPFs for Estimating Bicycle Crashes at Intersections for Boulder, Colorado Nordback, Marshall, and Janson (2014) developed a methodology to estimate bicycle crashes and applied the method using data from Boulder, Colorado. The research team followed similar methods used to develop the SPFs in the HSM. Boulder’s bicycle mode share is approximately 12 percent, and Boulder has a long history of counting bicycles. Exposure metrics used in modeling were AADT and annual average daily bicyclists (AADB). Only bicycle intersection crashes were included in the crash databases used to develop the SPFs. The SPFs were modeled as a negative binomial model using a generalized linear model with a log link. For bicyclist traffic, two methods were used to estimate AADB from turning movement counts: a factor method, which used computed monthly and daily factors based on continuous bicycle counts; and a statistical model dependent upon time, weather, and one spatial variable. This second method is described in a paper titled Estimating Annual Average Daily Bicyclists and Analyzing Cyclist Safety at Urban Intersections by Nordback (2012). Two SPFs were modeled: one for the years 2001 to 2005 and the other for the years 2008 to 2011. The 2001 to 2005 Model is shown in Equation 2-3. , .e AADT AADB k 0 54= =N . . .bike 9 07 0 64 0 53- (2-3) The 2008 to 2011 Model is shown in Equation 2-4. , .e AADT AADB k 0 36= =N . . .bike 8 94 0 58 0 65- (2-4)

Literature Review and Survey of Practice 27   Nbike is the predicted number of intersection bicycle crashes during the study period, and k is the overdispersion parameter. The models are not specific to any particular intersection type or configuration. The coefficients in both models are statistically significant at the 95 percent confidence level. 2.1.5.3 SPFs for Estimating Pedestrian and Bicycle Crashes at Intersections for Seattle, Washington To proactively identify locations for pedestrian and bicycle safety improvements, the Seattle Department of Transportation (SDOT) conducted research to develop SPFs to estimate both pedestrian (Thomas et al. 2017a) and bicycle (Thomas et al. 2017b) crashes at intersections in Seattle. The research team first developed a crash database of police-reported pedestrian and bicycle crashes over an 8-year period (2007 to 2014), which included 3,726 pedestrian crashes and 3,120 bicycle crashes. Crash type was not a variable available for these crashes in the police reports, so the research team defined crash types based on two or more variables that described specific crash events. Once the unique combinations were determined, the research team defined several crash types that accounted for a large majority of the crashes. These types were further grouped into two pedestrian crash types and three bicycle crash types for which SPFs were devel- oped. The two pedestrian crash types—all intersection pedestrian crashes and crashes involving a crossing pedestrian and a through-moving vehicle—accounted for 69 percent of all pedestrian crashes. The three bicycle crash types included: total bicycle intersection crashes, crashes in which the bicycle and vehicle were moving in opposite directions at an intersection, and angle bicycle intersection crashes. The three types accounted for 56 percent of all bicycle crashes. Using Seattle’s spatial database that includes variables related to crashes, intersections, seg- ments, signalization, and other roadway descriptors, the research team identified where pedes- trian and bicycle crashes were occurring and what factors seemed to be overrepresented in fatal and serious injury crashes. This analysis led to the identification of high-frequency crash types and location scenarios for further evaluation. Next, the research team developed a comprehensive intersection database that included all intersections with three or more legs. This database was supplemented with several other data elements, including building footprints, land-use variables, elevation data, school locations, and transit data. While estimated pedestrian and bicycle counts were available for the intersections, AADTs were not always available. Arterial classification was shown in previous research to be a good surrogate for AADT and was used in SPF development. A multivariate analysis was conducted in two phases to develop pedestrian and bicycle SPFs. In the first phase, a Conditional Random Forest (CRF) was used to examine a wide range of factors for potential inclusion in the models and to narrow the list down to those most likely to have an impact on crashes. The CRF identified the variables that tended to be overrepresented at locations where crashes of interest had occurred. In the second phase, binomial regression was used to develop SPFs for each of the five crash types of interest. Table 5 and Table 6 present the model statistics of the SPFs developed for estimating pedestrian and bicycle crashes, respectively. Once the models were developed, the researchers created four systems of ranking sites for improvement. These were historic crash frequency (using observed crashes only); predicted crash frequency (using the SPF developed in the research); Empirical Bayes (EB) analysis (using the predicted safety from the SPF weighted with crash history to obtain expected crash fre- quency); and potential for safety improvement (measured as the difference between expected and predicted crashes). 2.1.5.4 SPFs for Estimating Pedestrian and Bicycle Crashes in Michigan Multiple studies have been conducted in Michigan to estimate pedestrian and bicycle crashes. The Transportation Research Center for Livable Communities conducted research to develop

28 Pedestrian and Bicycle Safety Performance Functions Effect Category Total Pedestrian Crashes Crashes Involving Pedestrians Crossing and Motorists Going Straight Estimate Std. Error Estimate Std. Error Intercept N/A −10.8298 1.0123 −11.6231 1.5733 Number of commercial properties within 0.1 mi of intersection N/A 0.0248 0.0036 0.0313 0.0055 Daily buses within 150 ft N/A 0.0016 0.0003 0.0022 0.0004 Total building volume (0.1 mi) N/A 3.2E-08 0.0000 0.0000 0.0000 Natural log of annual average daily pedestrians (AADP) N/A 0.7199 0.1378 0.6440 0.2102 AADP N/A -0.0917 0.0189 −0.0741 0.0294 Commercial only, building density/sq. footage within 0.1 mi of intersection N/A −3.2E-08 0.0000 0.0000 0.0000 Local streets proportion N/A −1.0584 0.1595 −0.7632 0.2600 Mean income area residents N/A 0.0000 0.0000 0.0000 0.0000 Average slope of terrain within 0.5 mi surrounding intersection N/A −0.0976 0.0472 N/A Total population (0.1 mi) N/A 0.0002 0.0001 N/A Presence of traffic signal Yes 1.0818 0.0792 0.2849 0.1290No Base cond. Base cond. Number of legs 4 0.6052 0.0873 0.8687 0.1128 ≥ 5 0.6719 0.1449 0.6630 0.2296 3 Base cond. Base cond. Total lanes at intersection 7–8 0.4440 0.1231 N/A 9–26 0.4770 0.1681 N/A 3–6 Base cond. N/A Total lanes (largest leg) 3–4 −0.0624 0.1114 0.0691 0.1249 5–12 0.4053 0.1312 0.3907 0.1562 1–2 Base cond. Base cond. Highest arterial class Collector 0.6537 0.1661 1.4117 0.2610 Major 1.2129 0.1578 1.9278 0.2563 Minor 1.1736 0.1504 1.8381 0.2423 Undesignated Base cond. Base cond. Presence of parking Yes 0.2230 0.0749 N/ANo Base cond. N/A Scale Model parameter 0.7994 0.0676 1.3819 0.2126 NOTE: Base cond. = base condition; N/A = Not applicable. Table 5. SPFs for estimating total pedestrian crashes and crashes involving pedestrians crossing and motorists going straight at intersections in Seattle, Washington (Thomas et al. 2017b). SPFs as well as other methods for predicting pedestrian and bicycle crashes along road segments and at intersections in Michigan (Gates et al. 2016). The models were based solely on AADT and did not incorporate any exposure measure for nonmotorists. The models are similar in form to SPFs shown for motor vehicle crashes in the HSM and showed that pedestrian and bicycle crashes tend to increase with increasing traffic volumes. Table 7 and Table 8 show model coefficients (a, b) and overdispersion factors (k) for estimat- ing pedestrian and bicycle crashes (respectively) for urban segments. Individual models were developed for eight roadway types and three severity levels [total, fatal-and-injury (FI), and property-damage-only (PDO)]. Similarly, Table 9 and Table 10 show model coefficients (a, b, c) and overdispersion factors (k) for estimating pedestrian and bicycle crashes at urban intersections for four intersection types and three severity levels (total, FI, and PDO). Dolatsara (2014) developed SPFs using negative binomial regression to estimate pedestrian and bicycle crashes using pedestrian and bicyclist exposure data for urban signalized intersections

Literature Review and Survey of Practice 29   Effect Category Bicycle Crashes: Total Intersection Crashes Bicycle Crashes: Opposite Direction Intersection Crashes Bicycle Crashes: Angle Intersection Crashes Estimate Std. Error Estimate Std. Error Estimate Std. Error Intercept N/A −17.1605 1.0408 −22.7245 1.7445 −16.5371 1.3443 Natural log of AADB N/A 1.1398 0.0888 1.3770 0.1252 1.0226 0.1091 Number of commercial properties within 0.1 mi of intersection N/A 0.0120 0.0042 0.0171 0.0067 0.0135 0.0053 Natural log of AADP N/A 0.7388 0.1444 0.9188 0.2427 0.8249 0.1880 AADP N/A −0.0888 0.0205 −0.1091 0.0340 −0.1213 0.0283 Commercial only: building density/sq. footage within 0.1 mi of intersection N/A N/A N/A −2.628E-08 0 N/A N/A Average slope of terrain within 0.5 mi surrounding intersection N/A N/A N/A 0.1789 0.0850 −0.2149 0.0618 Number of transit stops within 150 ft of intersection N/A N/A N/A N/A N/A 0.0011 0.0004 Local streets proportion 0–1 −0.9132 0.1896 N/A N/A −1.4070 0.2277 Presence of traffic signal Yes 0.5131 0.0955 0.08985 0.1417 N/A N/A No (Base cond.) Base cond. N/A Base cond. N/A N/A N/A Highest arterial class Collector 0.2979 0.1717 0.9889 0.3157 0.0419 0.1987 Major: Interstate 0.8888 0.1612 1.6218 0.2748 0.5133 0.1994 Minor 0.9072 0.1547 1.9267 0.2663 0.4058 0.1840 Local (Base cond.) Base cond. N/A Base cond. N/A Base cond. N/A Bike lanes on any leg Yes 0.1796 0.0833 N/A N/A N/A N/A No (Base cond.) Base cond. N/A N/A N/A N/A N/A Shared-lane markings on any leg Yes 0.1547 0.0725 N/A N/A N/A N/A No (Base cond.) Base cond. N/A N/A N/A N/A N/A Number of legs 4 0.5780 0.0774 N/A N/A 0.8918 0.1008 ≥ 5 0.6765 0.1516 N/A N/A 0.9974 0.1953 3 (Base cond.) Base cond. N/A N/A N/A Base cond. N/A Presence of parking Yes 0.1800 0.0859 0.5398 0.1526 N/A N/A No (Base cond.) Base cond. N/A Base cond. N/A N/A N/A Total lanes (largest leg) 3–4 0.0055 0.0982 0.2751 0.1574 −0.0497 0.1254 5–12 0.3077 0.1359 0.5542 0.2130 0.3558 0.1648 1–2 (Base cond.) Base cond. N/A Base cond. N/A Base cond. N/A Two-way center turn lane on any leg Yes 0.4232 0.1142 0.3925 0.1822 0.3135 0.1476 No (Base cond.) Base cond. N/A Base cond. N/A Base cond. N/A Scale Model parameter 1.1835 0.1106 1.4569 0.2979 1.6067 0.2165 Sample size N/A 1753 462 870 NOTE: Base cond. = Base condition; N/A = Not applicable. Table 6. SPFs for estimating bicycle crashes at intersections in Seattle, Washington: total crashes, opposite direction crashes, and angle crashes (Thomas et al. 2017a).

30 Pedestrian and Bicycle Safety Performance Functions Total 2U −19.530 0.38* 1.86E-14 3T −3.480* −0.03* 7.16E-08 4U −21.040 1.87 2.00E-03 5T -9.280 0.69 0.12 4D −8.558 0.42* 1.03E-16 6D −5.520* 0.27* 1.58 8D -89.570 0.63* 1.04 OW −7.420* 0.36* 0.00 FI 2U −21.050 0.54* 2.46E-15 3T −3.480* −0.03* 7.16E-08 4U −22.490 2.00 0.00 5T −10.650 0.81 0.03 4D −8.150* 0.37* 9.92E-11 6D −4.600* 0.17* 0.87 8D −10.810 0.81 0.81 OW -0.900* −0.37* 0.00 PDO 2U −12.780 −0.65* 1.00 3T N/A N/A N/A 4U 14.640* 1.00* 0.00 5T −1.380* −0.34* 2.96E-07 4D −20.040* 1.34* 1.00 6D N/A N/A N/A 8D 1.680* −0.65- 0.00 OW −178.870* 17.48* 0.00 NOTE: 2U = two-lane undivided arterials. 3T = three-lane arterials including a center two-way left-turn lane (TWLTL). 4U = four-lane undivided arterials. 4D = four-lane divided arterials (including a raised or depressed median). 5T = five-lane arterials including a center TWLTL. 6D = six-lane divided arterials (including a raised or depressed median). 8D = eight-lane divided arterials (including a raised or depressed median). OW = one-way arterials including two-lane, three-lane, and three-lane segments. N/A = Not applicable. *Variable was not significant at a 95 percent confidence interval. Table 7. Michigan-specific AADT-only pedestrian urban segment crash models (Gates et al. 2016). in Michigan. The database included 164 intersections in four Michigan cities. In addition to vehicle volumes and pedestrian and bicyclist exposure data, Dolatsara considered several vari- ables related to the geometry of the intersection for inclusion in the SPFs. The SPF to predict pedestrian crashes included motor vehicle AADT, pedestrian volume, number of left-turn lanes, presence of on-street parking, presence of speed signs, and presence of a bus stop within 0.1 mi of the intersection. Table 11 shows the model coefficients for the urban signalized intersection pedestrian crash prediction model. The SPF to estimate bicycle crashes included motor vehicle AADT, bicycle volume, number of left-turn lanes, presence of bike lanes, presence of a bus stop within 0.1 mi of the intersection, and city indicator variables. Table 12 shows the model coefficients for the urban signalized intersection bicycle crash prediction model. Oh et al. (2013) conducted a study for the Michigan Department of Transportation (MDOT) to develop a systematic approach to determine performance measures for nonmotorized safety and to identify the need for countermeasures when designing facilities. The primary objectives of the research were to: 1. Build an inventory database for nonmotorized safety analysis and provide a guideline for data collection, storage, and management.

Literature Review and Survey of Practice 31   Total 2U −25.170 0.960 0.00 3T −4.110* 0.090* 0.00 4U -6.510* 0.360* 0.64 5T −13.340 1.050 0.00 4D −17.722 1.381 0.00 6D −11.325 0.830* 0.00 8D −3.160* −0.020* 0.04 OW −0.240* −0.320* 1.00 FI 2U −26.880 1.130 0.00 3T −5.470* 0.220* 0.00 4U −5.610* 0.240* 2.62 5T −14.450 1.150 0.00 4D −20.046 1.610 0.00 6D −11.672 0.850* 0.06 8D −4.050* 0.060* 0.62 OW −3.920* 0.070* 1.00 PDO 2U −15.580* −0.380* 0.00 3T 0.080* −0.570* 0.00 4U −10.980* 0.690* 0.00 5T −9.670* 0.490* 0.00 4D −8.440* 0.180* 0.00 6D −11.060* 0.550* 0.00 8D 1.510* −0.710* 0.00 OW 12.790* −2.004* 0.00 NOTE: 2U = two-lane undivided arterials. 3T = three-lane arterials including a center two-way left-turn lane (TWLTL). 4U = four-lane undivided arterials. 4D = four-lane divided arterials (including a raised or depressed median). 5T = five-lane arterials including a center TWLTL. 6D = six-lane divided arterials (including a raised or depressed median). 8D = eight-lane divided arterials (including a raised or depressed median). OW = one-way arterials including two-lane, three-lane, and three-lane segments. *Variable was not significant at a 95 percent confidence interval. Table 8. Michigan-specific AADT-only bicycle urban segment crash models (Gates et al. 2016). Total 3ST 15.512 0.765 0.386 2.143 3SG -9.044 0.402* 0.187 1.057 4ST −11.613 0.547 0.269 2.254 4SG −7.578 0.364 0.173 0.959 FI 3ST −15.099 0.742 0.338 1.000 3SG -9.223 0.418* 0.182* 1.354 4ST −11.520 0.529 0.271 2.712 4SG −7.583 0.366 0.157 0.779 PDO 3ST −20.711 0.886 0.661 1.168E-13 3SG −10.221 0.158* 0.283* 1.431E-16 4ST −16.547 0.793* 0.247* 0.000 4SG −10.535 0.316 0.311 0.977 NOTE: 3ST = three-leg intersection with stop control on the minor-road approach. 3SG = three-leg signalized intersection. 4ST = four-leg intersection with stop control on the minor-road approaches. 4SG = four-leg signalized intersection. *Variable was not significant at a 95 percent confidence interval. Table 9. Michigan-specific AADT-only pedestrian urban intersection crash models (Gates et al. 2016).

32 Pedestrian and Bicycle Safety Performance Functions Total 3ST −14.744 0.778 0.394 1.214 3SG −11.092 0.575 0.232 1.000 4ST −11.173 0.618 0.188 1.184 4SG −-6.958 0.256 0.227 0.884 FI 3ST −15.567 0.873 0.353 0.939 3SG −10.889 0.551 0.204 1.000 4ST −11.555 0.659 0.157 0.083 4SG −7.834 0.340 0.203 0.702 PDO 3ST −13.646 0.340* 0.591 1.648E-07 3SG −14.180 0.654* 0.331* 7.56E-11 4ST −11.718 0.408* 0.313 1.000 4SG −6.087 −0.072* 0.323 0.749 NOTE: 3ST = three-leg intersection with stop control on the minor-road approach. 3SG = three-leg signalized intersection. 4ST = four-leg intersection with stop control on the minor-road approaches. 4SG = four-leg signalized intersection. *Variable was not significant at a 95 percent confidence interval. Table 10. Michigan-specific AADT-only bicycle urban intersection crash models (Gates et al. 2016). Significant variables Coefficient Std. Error Z P>│z│ AADT 0.0002510 8.22E-06 3.05 0.002 Pedestrian volume (exposure) 0.0000910 0.0000344 2.65 0.008 Number of left-turn lanes 0.2296894 0.0821085 2.80 0.005 Presence of on-street parking (0: no on-street parking; 1: if just one corridor of an intersection has on-street parking) 0.5712769 0.2418822 2.36 0.180 Presence of speed signs (0: if no speed sign; 1: if just one corridor of an intersection has a speed sign) −0.4470537 0.2291633 −1.95 0.051 Presence of bus stop (0: no bus stop; 1: if just one corridor of an intersection has a bus stop within 0.1 mi of the intersection) 0.9400843 0.4297579 2.19 0.290 Constant −2.6607680 0.5042898 −5.28 0.000 NOTE: Z: the test statistic for the regression coefficients. P>|z|: P-value for the Z-test. Number of Observations: the number of observations used in the regression model. LR Chi2: likelihood-ratio chi-squared test statistic. Prob > Chi2: P-value for the chi-squared test. Log-likelihood: the log likelihood of the fitted model. Pseudo R2: McFadden's pseudo R-squared. Table 11. Michigan urban signalized intersection pedestrian crash prediction model (Dolatsara 2014). 2. Conduct a detailed analysis of high-crash and low-crash communities to identify factors affecting crashes involving pedestrians and bicyclists and develop applicable performance measures. 3. Evaluate performances of recent pedestrian and bicycle improvement projects through before-and-after studies and cost-benefit analyses to quantify their effectiveness. 4. Identify cultural issues associated with pedestrian incidents and determine what issues can and cannot be addressed by engineering solutions. 5. Develop systematic guidance for adjusting performance measurements by comparing the nationwide nonmotorized performance measurements and analysis results. 6. Develop a user guide for using performance measures and determine the need for nonmotorized countermeasures.

Literature Review and Survey of Practice 33   To accomplish the objectives of the project, data were collected in four Michigan cities including Ann Arbor, East Lansing, Flint, and Grand Rapids. Data collected included: • Nonmotorized crash data. • Pedestrian and bicycle volumes. • Nonmotorized facility inventory. • Nonmotorized improvement projects. • Activity locations. • Socioeconomic and demographic data. • Crime rates. • Land-use data. • Traffic volume data. Safety performance was measured at three levels: city, census tract, and corridor. Accordingly, data were processed at these three levels. As this study used a modeling approach in estimating pedestrian and bicycle volumes, all necessary data, including socioeconomic data, had to be processed for individual intersections. The census-tract-level analysis required processing all necessary data for each census tract. The corridor-level analysis required more detailed data processing efforts. In this study, two types of pedestrian and bicycle volume data were collected. The first type was 12-hour data collected at select locations using automated pedestrian and bicycle sensors, and the second type was 1-hour data collected manually at coverage locations. The 12-hour counts were used to extrapolate the 1-hour counts for coverage locations. For each city, 12-hour counts were collected at three locations, and 1-hour counts were collected at 20 locations. In addition to count data, pedestrian and bicycle volumes at signalized intersections were estimated from models developed using data from 91 signalized intersections in the four cities (i.e., Ann Arbor, East Lansing, Flint, and Grand Rapids). The pedestrian and bicycle volume estimation models included facility, geometry/design, land-use, demographic, and socioeconomic data. Socioeconomic, demographic, exposure, and physical feature variables were considered in the crash analysis. Analyses were conducted at three levels: city, census tract, and corridor. For cor- ridor analyses, the safety performance of intersection and midblock components were evaluated separately. Several negative binomial and Poisson regression models were fitted. The final inter- section and midblock SPFs developed to estimate pedestrian and bicycle crashes were as follows. Significant variables Coefficient Std. Error Z P>│z│ AADT 0.0004190 7.87E-06 5.32 0.000 Bicycle volume (exposure) 0.0008022 0.0001504 5.33 0.000 Number of left-turn lanes 0.1566364 0.0730476 2.14 0.032 Presence of bike lanes (0: no bike lane; 1 if just one corridor of an intersection has a bike lane) 0.5408297 0.2150191 2.52 0.012 Presence of bus stop (0: no bus stop; 1 if just one corridor of an intersection has a bus stop within 0.1 mi of the intersection) 0.9032806 0.4139433 2.18 0.029 Constant −3.3775220 0.4935776 −6.84 0.000 NOTE: Z: the test statistic for the regression coefficients. P>|z|: P-value for the Z-test. Number of Observations: the number of observations used in the regression model. LR Chi2: likelihood-ratio chi-squared test statistic. Prob > Chi2: P-value for the chi-squared test. Log-likelihood: the log likelihood of the fitted model. Pseudo R2: McFadden's pseudo R-squared. Table 12. Michigan urban signalized intersection bicycle crash prediction model (Dolatsara 2014).

34 Pedestrian and Bicycle Safety Performance Functions Equations 2-5 and 2-6, Intersection Models exp= . . . . . . N NLN AADT Ped Vol Number of bars Grad Degree 0 043449 0 000018 0 000056 0 0455736 0 0035416 0 043991 minped roador- + + + - + ` j (2-5) exp= . . . . . . . N D NLN AADT Bike Vol Dum Dum 0 26783 0 0913841 0 0000211 0 0006378 0 6660973 0 5015012 0 4236471 bike major right turn lanes road bus stop land use minor business - + + + + - -` j (2-6) where: Nped = Predicted number of pedestrian crashes at intersections, Nbike = Predicted number of bicycle crashes at intersections, NLNminor road = Total number of lanes on minor roads, Dmajor right-turn lanes = Number of right-turn lanes on the major road, AADT = Annual average daily traffic approaching the intersection, Ped Vol = Number of pedestrians crossing the intersection, Bike Vol = Number of bicyclists crossing the intersection, Number of Bars = Number of bars, Grad Degree = Number of people who have a graduate degree within ¼ mi, Dumbus stop = 1 if a bus stop exists; 0 otherwise, and Dumbusiness land use = 1 if business area; 0 otherwise. Equations 2-7 and 2-8, Midblock Models exp= . . . . . . N Access AADT Ped Vol Speed Limit Corridor Length s0 017 0 00004 0 00005 0 105 0 76 2 04 Pointped - + + - + + ` j (2-7) (2-8) exp= . . . . . . . . N AADT Bike Vol Speed Limit No Bus Stop Bike Commuter Bike Lane Corridor Length 0 00006 0 011 0 117 0 073 0 026 0 31 0 72 2 04 bike - + - + + - + + ` j where: Nped = Predicted number of pedestrian crashes at midblock, Nbike = Predicted number of bicycle crashes at midblock, Access Points = Number of access points, AADT = Annual average daily traffic of two ends, Ped Vol = Average of pedestrian volumes, Bike Vol = Average of bicycle volumes, Speed Limit = Speed limit of the arterial, No Bus Stop = Number of bus stops along the corridor, Bike Commuter = Number of employees commuting by bicycle, Bike Lane = 1 if bicycle lane exits; 0 otherwise, and Corridor Length = Length of corridor. McArthur, Savolainen, and Gates (2014) investigated pedestrian and bicycle crashes involving children aged 5 to 14 located within 1 mi of a school that included students from kindergarten to eighth grade in Michigan. In addition to the crash data, demographic and socioeconomic factors were obtained from the U.S. Census Bureau, including child population, kindergarten through 8th-grade enrollment, if the school was located on a local roadway, average family size, population density, median family income, the average number of parents per household, and the portion of households with residents who are not White people.

Literature Review and Survey of Practice 35   A random effects negative binomial model was developed using 5 years of crash data (2007 to 2011) to predict the number of pedestrian and bicycle crashes combined. Model results are presented in Table 13. 2.1.5.5 SPFs for Estimating Pedestrian and Bicycle Crashes in New Zealand and Australia SPFs have been developed for estimating pedestrian and bicycle crashes in several jurisdic- tions and cities across New Zealand and Australia over the last 15 years. This section presents a selection of the models. The models were developed using generalized linear modeling (GLM) methods with a negative binomial or Poisson error structure. A number of test statistics and graphing methods were used to assess goodness-of-fit. 2.1.5.5.1 Traffic-Flow-Only SPFs for Pedestrians and Cycles (S. Turner, Roozenburg, and Francis 2006) The first set of SPFs for crashes involving pedestrians and bicycles in New Zealand was pub- lished in 2006. The models covered crossing pedestrians and through cyclists on midblock arte- rial sections and crossing pedestrians and cyclists at signalized intersections and roundabouts. The study also included interviews at a Christchurch public hospital with pedestrians and bicy- clists involved in serious injury crashes to collect more detailed information about injuries sus- tained and to better understand reporting rates of such crashes. The models developed, shown in Equation 2-9, have the following form: (2-9)N b AADT AADP AADBb b0 1 2= where: N = predicted number of crashes for an approach or midblock section, AADT/AADP/AADB = annual average daily traffic of vehicles, pedestrians, or bicyclists, respectively, in a set time period, and bn = model coefficients. The key outcome of this study was the strong safety-in-numbers effect for both pedestrian and bicycle crashes. While the safety-in-numbers effect was well known at the national level (countries with high bicycle mode share generally have low crash rates per bicyclist), there was Parameter Estimate Std. Error p-Value Marginal Effect Constant 0.933 0.654 0.153 N/A Standard deviation 0.635 0.021 < 0.001 N/A Log (child population ages 5–14) 0.228 0.020 < 0.001 0.168 Log (kindergarten–8th grade enrollment) 0.066 0.014 < 0.001 0.049 School located on local roadway 0.172 0.038 < 0.001 0.126 Average family size 0.326 0.127 0.010 0.239 Population density (persons per sq. mi.) 0.292 0.013 < 0.001 0.215 Median family income ($1,000) −0.017 0.001 < 0.001 −0.012 Average parents per household −2.214 0.403 < 0.001 −1.630 Proportion of households without White residents −1.644 0.217 < 0.001 −1.210 Overdispersion parameter 20.183 7.365 0.006 N/A N/A = Not applicable. Table 13. SPF estimating child pedestrian and bicycle crashes near schools (McArthur, Savolainen, and Gates 2014).

36 Pedestrian and Bicycle Safety Performance Functions limited research at the time to demonstrate that it also occurred at an element level (intersection and link) and for specific crash types. For pedestrians, the main crash type at traffic signals is pedestrians crossing at 90 degrees to traffic, with either the motorists running a red signal or the pedestrians not crossing on the green. Equation 2-10 for this crash type shows a safety-in-numbers effect on the pedestrian flow count (AADP). (2-10). AADT AADP7 28 10#=N . .ped 6 0 63 0 39- The main crash type for bicyclists at traffic signals is through-movement bicyclists being hit by a right-turning driver from the opposite leg (i.e., left-turning vehicle for right-hand drive). The safety-in-numbers effect for bicyclists is shown in Equation 2-11 by the low parameter on the through bicycle volume (AADB): (2-11). AADT AADB4 404 10#=N . .bike 4 0 34 0 2- Another key outcome of this study (from the hospital survey) was evidence of the low rate of crash reporting for serious (i.e., hospitalized) pedestrian and bicycle crashes in the police crash database. Also, there are a lot of pedestrian-only crashes (tripping on footpaths—especially the elderly) and bicyclist-only crashes (off-road and on-road cyclists riding into objects including parked vehicles) that go unreported. The models developed in this study have been used in a variety of ways in New Zealand. The primary application has been in economic evaluation (appraisal), as the models provide a future crash prediction for new and modified intersections and road links. The pedestrian models are also used in the New Zealand pedestrian crossing facilities tool that was developed to assist pro- fessionals in selecting the most efficient and safe crossing facility. 2.1.5.5.2 Midblock and Signalized Bicycle Facilities (S. Turner, Binder, and Roozenburg 2009) This research involved a more detailed study of bicycle crashes at midblock locations and signalized intersections using data from three New Zealand cities (Christchurch, Hamilton, and Palmerston North) that due to their relatively flat, wide roads have a relatively high propor- tion of bicycle trips. The study considered the bicycle and traffic volumes along with several site characteristics, including the length of each link, parking (e.g., with and without parking and level of parking utilization), presence of bicycle lanes (no separators), and painted median (called flushed medians in New Zealand). The models were generally of the following form (Equation 2-12): (2-12)b AADT AADB L AF=Nbike b b b LayoutFactor0 1 2 3 where: Nbike = predicted number of bicycle crashes, AADT = traffic volume on the link, AADB = bicycle volume of the link, L = length of the link (between major intersections), AFLayout Factor = Adjustment factors used in conjunction with the model, and bn = model coefficients. As an example, Equation 2-13 presents the general model for midblock, nonturning crashes (bicyclists traveling straight through): (2-13). AADT AADB L2 28 10#=N . . .bike 4 0 31 0 50 0 27-

Literature Review and Survey of Practice 37   The key findings from this model include: 1. Bicycle crash potential per kilometer increases as the link length between major intersections reduces. This is most likely due to additional factors that occur on shorter city street blocks, which typically occur in the heart of cities or on mixed-use arterials. It appears to act as a surrogate for the additional complexity of these mixed-use arterials compared with suburban midblocks. 2. The models show the safety-in-numbers effects for bicyclists, but the level of this effect depends on the crash type. 3. Painted (flushed) medians (often called two-way left-turn lanes in North America) reduce all midblock bicycle crashes by approximately 37 percent and nonintersection/driveway crashes by approximately 50 percent. The extra width (of the median) seems to increase the distance between vehicles and bicyclists midblock. 4. The absence of parking reduces bicycle and all vehicle midblock crashes by approximately 75 percent. 5. Routes with low levels of parking tended to have higher crash rates than those with higher levels of parking. This may be the result of bicyclists using the parking lane and then merging into the traffic lane to go around parked vehicles. This merging maneuver by a bicyclist may surprise some drivers. The analysis provided some counterintuitive results in terms of bicycle lanes, showing that the presence of bicycle lanes increases bicycle crash rates. After further analysis, using before-and- after evaluations, it was identified that this was a result of bias-by-selection. In selecting routes for bicycle lanes (a safety treatment), routes were selected with higher-than-average bicycle crash rates. Given this bias, the impact of this variable cannot be included in cross-sectional models and instead is excluded from the model, and an AF is used for bicycle lanes, based on the before- and-after result. The before-and-after study showed a 10 percent reduction in bicycle crashes when a bicycle lane is installed. This was based on narrower bicycle lanes using early design standards. There is more recent evidence that benefits greater than 10 percent can be achieved for wider and high-standard bicycle lanes. 2.1.5.5.3 Bicyclists and Roundabouts (S.Turner, Roozenburg, and Smith 2009) A number of crash prediction modeling studies were undertaken in New Zealand in the early part of the 2000s for different intersection and link types and different crash types. This included a study of urban roundabouts and urban traffic signals. Instead of the traditional approach of focusing on motor-vehicle-only crashes, these studies looked at all the main crash types including pedestrians and bicycles. At roundabouts, especially those in cities with higher bicycle volumes, the main crash types included crashes with bicyclists. Interestingly, pedestrian crashes at roundabouts are fairly rare, especially at multilane roundabouts, as this intersection type is rarely used in areas with a high number of pedestrians. In the limited cases where roundabouts are used in areas with a large number of pedestrians, they are nor- mally part of a traffic-calmed (low-speed) environment, where low speeds make pedestrian crashes rare. This study of roundabouts included traditional layout and traffic flow variables along with approach and circulating speeds and sight distance. The impact of sight distance and approach speed is of particular interest as “new world” design guidelines (United States, Australia, and New Zealand) for roundabouts generally call for high levels of sight distance and relatively high approach speeds, compared with design guidelines in older countries, like those in Europe, where they limit sight distance and try to manage down-approach speeds. In terms of bicycle crashes, the much higher speed of motor vehicles compared with bicyclists, especially on larger roundabouts, is a major concern in terms of not only crashes but also high-severity crashes.

38 Pedestrian and Bicycle Safety Performance Functions The list of variables considered in the models was as follows: • Manual motor vehicle and bicycle counts for each movement, which were scaled up to AADTs. • Speeds of free-flowing vehicles traveling through the roundabout as they entered and circu- lated through the roundabout for each approach (collected in the field). • The sight distance between drivers entering the intersection to vehicles approaching from their right, measured from three locations: (1) at the limit (yield) line, (2) 10 meters back from the limit line, and (3) 40 meters back from the limit line. • Number of lanes for each approach and circulating. • Number of roundabout arms/legs. • Road markings. • Superelevation direction of circulating lanes (whether inward or outward). • Direction of the gradient of approaches. • Lighting. • Pedestrian and bicycle facilities where relevant. • Surrounding land use. • Features that obstruct visibility. • Speed limit on main road [≥ 80 km/h (50 mph) is a high-speed roundabout]. Many of these variables were not found to be statistically significant predictors of bicycle crashes. Speed and sight distance were found to be important. Further analysis was undertaken to examine the relationship between approach and circulating speed and both approach sight distance and geometric design factors. Unfortunately, this analysis was not meaningful, due to the large number of unique and poorly designed roundabouts in the dataset. However, under- standing how speed can be managed at roundabouts, especially those with large numbers of bicyclists, has been an ongoing area of research in New Zealand over the last 10 years. The main model for bicycle crashes involved bicyclists traveling around the roundabout being hit by vehicles entering the roundabout (entering versus circulating crashes). This model is pre- sented in Equation 2-14. (2-14). AADT AADB S3 88 10#=N . . .bike entering e c e5 0 43 0 38 0 49- where: Nbike entering = predicted number of entering crashes involving bicyclists circulating, AADTe = annual average daily traffic entering on the approach, AADBc = annual average daily circulating bicyclist flow perpendicular to the entering motor vehicle flow, and Se = free mean speed of vehicles as they enter the roundabout. A separate study (unpublished) was undertaken of bicycle crashes at roundabouts (and traf- fic signals) in the State of Queensland, Australia. This analysis involved using a sample of local intersections, mainly on the state highway network, combined with the New Zealand data to understand whether the same significant predictors of crashes were present in the Queensland data. Covariate parameters were developed for Queensland and generally showed the same vari- ables were important, with the crash rates generally being higher than in New Zealand for most bicycle crash types. This is likely due to the higher speed limits and larger intersections that exist on the state highway network. 2.1.5.5.4 Pedestrian and Bicycle Safety at Traffic Signals (S. Turner et al. 2010; S. Turner, Singh, and Nates 2012) Two additional research studies looked at pedestrian and bicycle crashes at traffic signals in more detail. The first was an Austroads study (S. Turner et al. 2010) that focused on the safety

Literature Review and Survey of Practice 39   benefits of on-road bicycle facilities at traffic signals. This included bicycle lanes on the approach legs of the intersections, bicycle lanes on the departure legs of the intersections, bicycle storage boxes, and wide curb lanes. The study looked at what facilities were provided as bicyclists nego- tiated traffic signals based on Cumming’s (2000) six elements of continuity. Figure 4 shows the six elements along with bicycle treatments that are typically used. Some of these treatments, like hook turns, are not widely used, so the study could not evaluate their effectiveness. The research team collected data from two cities (Christchurch, New Zealand, and Adelaide, South Australia) for the study. The number of predictor variables used in the study was extensive, as shown in the models below. Models were developed separately for each city and combined for both cities. Two of the key crash types are crashes in the same direction on the approach to the inter section and right-turn crashes where a through bicyclist is hit by a right-turning vehicle (left-turn vehicle Figure 4. Six elements of continuity through an intersection and typical bicycle treatments (Cumming 2000; S. Turner, Singh, and Nates 2010)

40 Pedestrian and Bicycle Safety Performance Functions when right-turn drive). The models for these two crash types are provided in Equation 2-15 and Equation 2-16: Same Direction Bicycle SPF (Christchurch) (2-15) . AADT AADB4 72 10#=N total approach width bicycle lane width curbside lane width F F F ( ) . . . . bike sd Christchurch transition facility shared lanes 5 0 655 0 062 1 30 3 236 storage+ - - - - ` ` j j where: AADT = total one-way daily volume of traffic on approach, AADB = total one-way daily volume of bicyclists on approach, Fstorage = storage treatments present on approach (value = 1.502), Ftransition facility = presence of transition cycle facility on approach (value = 4.304), and Fshared lanes = presence of shared lanes on approach (value = 3.365). Right-Turn Crashes Bicycle SPF (Christchurch) . AADT AADB2 38 10#= .N no of through traffic lanes bicycle lane width depth of bicycle box depth F F F F 1 1 intersection ( ) . . . . . . bike rt Christchurch approach facility shared rt rt 5 0 550 0 079 0 436 1 336 0 023 1 397 painted phasing + + - - - ` ` ` ` j j j j (2-16) where: AADT = daily volume of right-turning traffic for specified movement, AADB = daily volume of through cyclists for specified movement, Fpainted = painted treatments (value = 0.553), Fapproach facility = presence of approach cycle facility (value = 3.388), Fshared rt = shared right-turn (rt) lane on motor vehicle movement approach (RT motor vehicles) (value = 0.933), and Frt phasing = fully/partially protected phasing arrangement at intersection (value = 0.987). Both models show that greater curb-lane width, whether it includes a bicycle lane or not, has a strong safety benefit. However, the models also indicate that the presence of bicycle facilities increases crashes (F-values for storage lane and transition facility are greater than 1). The project team believes this is due to bias-by-selection, which was evident in the earlier models presented for bicycle midblock. Before-and-after studies were undertaken on bicycle facility treatments and generally showed that such treatments reduce crashes. Shared lanes, whether left and through or right and through, also increase crashes. The wider approach widths seem to result in fewer crashes in Christchurch. Other models showed that painted bicycle lanes result in fewer crashes than unpainted bicycle lanes. The analysis showed that for the continuous variables (number of through lanes, width of bicycle lanes, depth of bicycle storage box, and intersection depth), it is fairly consistent across the two cities in terms of whether a factor results in an increase or decrease in crashes. The second study of traffic signals addressed both motor-vehicle-only and pedestrian crashes and was based on data from 238 traffic signals across several cities in New Zealand and in Melbourne, Victoria, Australia (S. Turner, Singh, and Nates 2012). The dataset consisted of three- and four-leg intersections, including some higher speed signals [where the main road speed limit is 80 km/h (50 mph) or above].

Literature Review and Survey of Practice 41   The variables collected for this study were extensive but can be grouped under the following five categories: • Detailed signal layout and geometry. • Signal operation (e.g., signal phasing, cycle times, estimated degree of saturation, and coordination). • Motor vehicle and pedestrian volumes. • Crash data (e.g., by crash type and time of day). • Miscellaneous (e.g., number of signal displays and land use). Traffic signal data were extracted from the traffic signal coordination system, SCATS. In addition to average “green times” and “cycle times,” the number of times the pedestrian phase was called was recorded. The latter was used to estimate the level of pedestrian demand on each leg of an intersection (where this phase was provided). The motor vehicle traffic volumes were also estimated from counts in SCATS. Manual counts were used to estimate the propor- tion of traffic making each maneuver for shared lanes (lanes with one detector) and free-left- turn lanes. The three main crash types involving pedestrians were right-angle crashes, right-turn crashes, and left-turn crashes. For the latter two crash types, the pedestrian was crossing parallel to the green phase, and drivers were turning onto the minor road. The majority of New Zealand/ Melbourne traffic signals allow traffic to filter-turn across the potential path of parallel-crossing pedestrians for at least a portion of the cycle time. Many intersections have early starts for pedes- trians, and others have partially or fully controlled left- and right-turning phases. Left-turning signal phases are less common than right-turning phases. Due to the small number of left- turning pedestrian crashes, a model was not developed for this crash type. Right-angle pedestrian crashes occur either when a vehicle is running a red light or a pedestrian is crossing without the right-of-way. The preferred model for right-angle pedestrian crashes is provided in Equation 2-17: b AADT AADT= .N no of approach lanes all red time cycle time F F F F , . . . . . bike NA NB approach approach shared turns split median islandcycle facilities 0 0 314 0 364 0 16 0 61 0 810 phasing - ` ` ` j j j (2-17) where: Nbike NA,NB = number of predicted NA (near-side) and NB (far-side) injury crashes in 5 years, AADTapproach = total daily traffic volume entering the intersection from the approach, AADPapproach = pedestrian volume level on the approach (on a scale of 1 to 5, with 1 being a low volume of pedestrians and 5 being a high volume), b0 (Auckland) = constant for Auckland (value = 3.84E-05), b0 (Wellington) = constant for Wellington (value = 1.28E-05), b0 (Christchurch) = constant for Christchurch (value = 5.30E-05), b0 (Hamilton) = constant for Hamilton (value = 5.94E-05), b0 (Dunedin) = constant for Dunedin (value = 8.90E-05), b0 (Melbourne) = constant for Melbourne (value = 3.39E-05), Fcycle facilities = presence of facilities for cyclists (e.g., cycle lanes and/or storage boxes) (value = 0.513), Fshared turns = presence of lanes with shared-turning movements (e.g., left/through, right/ through, or both) (value = 1.321), Fsplit phasing = signal coordination with upstream intersection (value = 0.741), and Fmedian island = presence of raised median/central island on approach with pedestrian movement (value = 0.767).

42 Pedestrian and Bicycle Safety Performance Functions The coefficients for motor vehicle and pedestrian volumes in Equation 2-17 are similar, which indicates they are equally important factors. Wider approaches are predicted to have more crashes. The coefficients for cycle time and all-red time suggest that increasing the length of the signal cycle results in more pedestrian crashes, possibly as a result of pedestrian frustra- tion. A split signal phasing sequence, the presence of a raised median, and cycle facilities on the approach result in reduced crash numbers. The presence of lanes with shared-turning move- ments is observed to cause more crashes. Table 14 shows the variables included within the various motor vehicle and pedestrian SPFs along with whether they increased (cells showing an up arrow) or decreased (cells showing a down arrow) the potential for crashes. The table shows there are trade-offs in safety that occur at traffic signals. One measure may decrease crashes for one crash type or road user but increase crashes ©Parameter Motor Vehicle Crashes Pedestrian Crashes Right Angle Right-Turn Against Loss of Control Rear End Other Right Angle Right- Turning O ve ra ll M od el Au ck la nd a nd M el bo ur ne M od el Pe ak -P er io d M od el O ve ra ll M od el Pe ak -P er io d M od el Sm al l I nt er se ct io n M ed iu m In te rs ec tio n La rg e In te rs ec tio n O ve ra ll M od el Au ck la nd a nd M el bo ur ne m od el Higher approaching traffic volume Higher right-turning traffic volume Higher degree of saturation Higher pedestrian volume Larger intersection size (number of approach lanes, intersection depth) More approaching lanes More through lanes Wider approaches Longer cycle time Longer all-red time Longer yellow time Longer lost time (intergreen and all-red time) Full right-turn protection Split phasing Mast arm Coordinated signals Additional advance detectors Shared turns (e.g., left/through and right/through lanes) Shared right-turn/through lane Raised median/central island Longer right-turn bay/lane Free left turn for motor vehicles Presence of merge on intersection exit Cycle facilities Upstream bus bay within 100 meters Upstream parking Higher speed limit (≥ 80 km/h) Commercial land use Residential land use Table 14. Trade-offs in safety that occur at traffic signals (S. Turner, Singh, and Nates 2012).

Literature Review and Survey of Practice 43   for another. Similarly, bicyclists’ safety can be compromised by improvements that are made to an intersection to make it safer for motor vehicles and/or pedestrians. Hence, when designing traffic signals and how they will operate, it is important to understand the types of road users that will use the intersection and the proportion of traffic making turns. A signalized intersection in a built-up area with lots of pedestrians and/or bicyclists should have a different layout and signal phasing than an intersection in a suburban area where the majority of road users are motor vehicles. 2.1.5.6 Systemic Safety Prioritization Method for Pedestrian and Bicycle Crashes in Oregon The Oregon Department of Transportation (ODOT) developed a network screening approach to prioritize corridors with the most potential for reducing pedestrian and bicycle crashes (Bergh et al. 2015). The primary steps in the methodology included: • Identify potential crash-contributing factors (i.e., roadway or location characteristics) present at locations where crashes were reported from 2007 to 2011. • Identify and prioritize locations within the state where one or more crash-contributing factors are present. Crash-contributing factors include a range of roadway characteristics that appear to be asso- ciated with higher frequencies of serious pedestrian or bicycle crashes. Pedestrian and bicycle volumes were not considered in the methodology due to the lack of consistent, statewide data. The systemic network screening methodology was complemented by a secondary network screening method outlined in HSM Part B to account for locations with a known history of frequent and serious pedestrian and bicycle crashes. Results from both screening methods were combined to prioritize candidate project corridors. The systemic analysis was consistent with the steps outlined in the FHWA Systemic Safety Project Selection Tool (Preston et al. 2013): 1. Identify focus crash types and crash-contributing factors. 2. Screen and prioritize candidate locations. 3. Select countermeasures. 4. Identify specific projects at prioritized candidate locations. Potential factors developed by this study were roadway or location characteristics that could contribute to a pedestrian or bicycle crash and were present at locations where target crashes were reported. Based on 2007 through 2011 crash data, the primary crash-contributing factors for pedestrian and bicycle crashes included: • Pedestrian crash-contributing factors – Presence of transit stops. – Undivided four-lane roads in urban areas. – Presence of traffic signals. – Presence of pedestrian-activated flashers or beacons. – Posted speed limit. – Average daily traffic. – Reported crash frequency and severity. • Bicycle crash-contributing factors – Driveway density. – Undivided four-lane roads in urban areas. – Lack of bicycle facility on at least one side of the roadway. – Presence of traffic signals. – Posted speed limit. – Presence of transit stops. – Reported crash frequency and severity.

44 Pedestrian and Bicycle Safety Performance Functions Having identified pedestrian and bicycle crash-contributing factors, ODOT screened the state and urban nonstate networks separately to prioritize locations where multiple crash-contributing factors were present. The network screening process involved five general steps: 1. Develop and segment (0.10 mi increments) a roadway network including information about facility characteristics. 2. Develop a scoring process to apply to each segment, reflecting the relative crash-contributing factor. 3. Apply scoring to prioritize individual roadway segments. 4. Combine individual, priority roadway segments into candidate project corridors. 5. Develop a list of prioritized candidate project corridors based on the average corridor score. There was no quantitative data associating the identified crash-contributing factors and crash frequency, so a subjective scoring system was developed to account for combinations of crash- contributing factors. Figure 5 and Figure 6 illustrate the segment scoring system developed to Pedestrian Segment Data Roadway Name: Start (Nearest Cross Street or Milepost): End (Nearest Cross Street or Milepost): Length (miles): 0.1 Jurisdiction: Instructions: Please complete this form for each 0.1-mi-long segment within the area you would like to evaluate. To develop a score for the corridor, calculate the average score of the corridor by dividing the sum of all scores by the number of 0.1-mi-long segments within the corridor. Use the final corridor score to prioritize projects. Crash-Contributing Factor Data Score Score Methodology Is at least 1 traffic signal located within 100 ft of the segment? 1 point if at least 1 signal is located on the segment or within 100 ft of the segment How many transit stops are located within 100 ft of the segment? 1 point for segments with 1 transit stop located on the segment or within 100 ft of the segment; 2 points for 2 or more transit stops Are there one or more pedestrian- activated beacons or flashers located on the segment? 1 point subtracted (rewarded) for the presence of an enhanced midblock crossing What is the posted speed limit? 2 points for posted speed limit of 35 or 40 mph; 4 points for posted speed limits above 40 mph Is the corridor an undivided, four- lane segment? 2 points if segment is an undivided four- lane segment AADT of corridor 2 points for AADT between 12,000 and 18,000; 4 points for AADT above 18,000 Number of minor or moderate injuries resulting from pedestrian involvementa 2 points if a nonsevere injury was reported; 1 additional point for each additional injury Number of pedestrians involved in a crash but not injureda 2 points if a pedestrian is involved in a crash but not injured; 1 additional point for each additional pedestrian involved but not injured Number of severe injuries resulting from pedestrian-involved crashesa 4 points if a severe injury was reported; 2 additional points for each additional injury Number of fatalities resulting from pedestrian-involved crashesa 4 points for fatalities Total Score aAll crash history should reflect the latest 5 years of available data. Figure 5. Systemic-based segment scoring worksheet for pedestrian crash-contributing factors (adapted from Bergh et al. 2015).

Literature Review and Survey of Practice 45   prioritize sites based on pedestrian and bicycle crash-contributing factors. ODOT applied the scoring methodology to rank 9,490 segments (0.1 mi each) on the state network. Individual segments were prioritized based on their total score and grouped into longer candidate project corridors based on the scores of upstream and downstream segments within 0.5 mi. Within each region, sites were prioritized into project corridor lists using the systemic-based and/or crash frequency/severity-based screening methods. Projects focusing on pedestrian and bicycle crashes were prioritized separately, as they each require mitigation countermeasures spe- cific to that mode of travel. Monsere et al. (2017) continued the work in Oregon to improve methods to identify and pri- oritize locations with increased or elevated potential for pedestrian and bicycle crashes with the Bicycle Segment Data Roadway Name: Start: End: Length (miles): 0.1 Jurisdiction: Instructions: Please complete this form for each 0.1-mi-long segment within the area you would like to evaluate. To develop a score for the corridor, calculate the average score of the corridor by dividing the sum of all scores by the number of 0.1-mi-long segments within the corridor. Use the final corridor score to prioritize projects. Crash-Contributing Factor Data Score Score Methodology Is at least 1 traffic signal located within 100 ft of the segment? 1 point if at least 1 signal is located on the segment or within 100 ft of the segment Is the corridor an undivided, four- lane segment? 2 points if segment is an undivided four- lane segment How many transit stops are located within 100 ft of the segment? 1 point for segments with 1 transit stop located on the segment or within 100 ft of the segment; 2 points for 2 or more transit stops Does the left side of the road have a bicycle facility? 2 points for the lack of bicycle facility on the left side of the road Does the right side of the road have a bicycle facility? 2 points for the lack of bicycle facility on the right side of the road AADT of corridor 2 points for AADT between 12,000 and 18,000; 4 points for AADT above 18,000 What is the posted speed limit? 2 points for posted speed limit of 35 or 40 mph; 4 points for posted speed limits above 40 mph How many driveways or alleys are located on the corridor? 2 points for segments with 1 driveway; 3 points for segments with 2 to 3 driveways; 4 points for segments with 4 to 8 driveways; 5 points for segments with more than 8 driveways Number of minor or moderate injuries resulting from bicyclist involveda 2 points if a nonsevere injury was reported; 1 additional point for each additional injury Number of bicyclists involved in a crash but not injureda 2 points if a bicyclist is involved in a crash but not injured; 1 additional point for each additional bicyclist involved but not injured Number of severe injuries resulting from bicyclist-involved crashesa 4 points if a severe injury was reported; 2 additional points for each additional injury Number of fatalities resulting from bicyclist-involved crashesa 4 points for fatalities Total Score aAll crash history should reflect the latest 5 years of available data Figure 6. Systemic-based segment scoring worksheet for bicycle crash-contributing factors (adapted from Bergh et al. 2015).

46 Pedestrian and Bicycle Safety Performance Functions objective to develop a scoring method with weights derived from data analysis, as compared to best judgment or a subjective scoring system. To develop the scoring method, data were collected from 188 random segments and 184 random intersections. Geometric, land use, volume, and crash data were collected from multiple sources. The dataset included 213 bicycle and pedestrian crashes on roadway segments and 238 at intersections from 2009 to 2013. Logistic regression models were developed for both crash occurrence (crash or not) and crash severity. A summary of the significant variables included in both the pedestrian and bicycle models is presented in Table 15. The crash severity models were not robust, most likely due to too few segments and intersections with serious crashes in the dataset, so the crash occurrence models were used to create a scoring tool. The final pedestrian and bicycle scoring method for segments and intersec- tions are presented in Table 16 through Table 19. Pedestrian Models Crash Occurrence Crash Severity Segment Intersection Segment Intersection Traffic directions (one way is base) Total population density (per square mile) Retail density (per acre) Major road. Presence of total traffic lanes Presence of on-street parking Number of transit lines going through the intersection AADT (2014) Three-leg intersection density (per square mile) Presence of TWLTL Major road. Presence of right-turn lanes Presence of lighting Four-leg intersection density (per square mile) Posted speed limit (mph) Major road. AADT 2014 Total population density (per square mile) Total population density (per square mile) Minor road. Presence of median Minor road. Presence of bicycle lanes Total number of traffic lanes Minor road. Presence of right-turn lanes Minor road. Presence of right- turn lanes Minor road. Presence of total traffic lanes Major road. AADT 2014 Household density (per acre) Major road. Presence of right- turn lanes Minor road. Traffic direction Bicycle Models Crash Occurrence Crash Severity Segment Segment Segment Segment Presence of crossing Bikes per day (STRAVA) Crosswalk combination Type B (Type A is base) Total population density (per square mile) AADT 2014 Number of transit lines going through the intersection Crosswalk combination Type C (Type A is base) Household density (per acre) 3-leg intersection density (per square mile) Minor road. Functional class Crosswalk combination Type N (Type A is base) Household size Bikes per day (STRAVA) Minor road. Total number of traffic lanes Width of sidewalk buffer Minor road. Presence of right- turn lanes Major road. Presence of left- turn lanes Minor road. Presence of left- turn lanes NOTE: Type A: Segments with no midblock crossing and a signalized intersection on one end. Type B: Segments with no midblock crossing and signalized intersections on both ends. Type C: Segments with a midblock crossing and a signalized intersection on one end. Type N: Segments with no midblock crossing and unsignalized intersections on both ends. Table 15. Summary of significant variables in pedestrian and bicycle models (adapted from Monsere et al. 2017).

Literature Review and Survey of Practice 47   Variables Levels Internal Weight Score Traffic direction One-way 3.63 17Two-way 1.00 0 On-street parking Yes 3.81 17No 1.00 0 Posted speed limit (mph) < =25 1.00 0 30 1.28 6 35 1.64 8 > 35 2.70 12 Presence of TWLTL Yes 2.92 14No 1.00 0 Total population density (per square mile) < =1,000 1.00 0 1,001–3,000 1.30 6 3,001–5,000 1.84 8 5,001–7,000 2.62 11 > 7,000 4.41 20 Total traffic lanes 2 1.00 0 3 or 4 2.09 10 > 4 4.38 20 NOTE: Bold scores represent the maximum score for a given category/variable. Table 16. Pedestrian segment scoring system (Monsere et al. 2017). Variables Levels Internal Weight Score Total population density (per square mile) < =1,000 1.00 0 1,001–3,000 1.44 5 3,001–5,000 2.30 8 5,001–7,000 3.77 13 > 7,000 6.03 21 Number of transit lines with routes through the intersection 0 1.00 0 1 1.47 6 2 2.15 8 3 3.16 12 > 3 6.79 25 Major AADT (2014) < =5,000 1.00 0 5,001–10,000 1.37 5 10,001–15,000 1.88 7 15,001–20,000 2.57 10 20,001–25,000 3.52 13 > 25,000 4.82 18 Major road. Presence of median Yes 1.00 0 No 3.52 3 Minor road. Presence of right-turn lanes Yes 1.00 0 No 3.71 15 Major road. Presence of right-turn lanes Yes 1.00 0 No 2.19 8 NOTE: Bold scores represent the maximum score for a given category/variable. Table 17. Pedestrian intersection scoring system (Monsere et al. 2017).

48 Pedestrian and Bicycle Safety Performance Functions Variables Levels Internal Weight Score Bikes per day (STRAVA) < =200 1.00 0 201–800 1.50 15 > 800 2.48 25 Major AADT (2014) < =5,000 1.00 0 5,001–10,000 1.07 12 10,001–15,000 1.38 14 15,001–20,000 1.61 16 20,001–25,000 1.89 19 > 25,000 2.40 25 Three-leg intersection density (per square mile) (EPA Smart Location) 1-150 1.00 0 151-200 1.23 13 > 200 1.60 16 Presence of marked crosswalk Yes 1.00 0 No 3.34 34 NOTE: Bold scores represent the maximum score for a given category/variable. Table 18. Bicycle segment scoring system (Monsere et al. 2017). Variables Levels Internal Weight Score Bikes per day (STRAVA) < =200 1.00 0 201–800 2.40 11 > 800 4.30 20 Number of transit lines with routes through the intersection 0 1.00 0 1 1.40 7 2 2.00 10 3 2.80 14 > 3 5.70 27 Minor functional class Collector 1.00 0Arterial 2.30 12 Minor road. Total number of traffic lanes 2 1.00 0 3 1.60 8 4 2.70 12 > 4 7.20 31 Presence of marked crosswalk Yes 1.00 0 No 2.20 10 NOTE: Bold scores represent the maximum score for a given category/variable. Table 19. Bicycle intersection scoring system (Monsere et al. 2017). 2.1.5.7 U.S. Road Assessment Program (usRAP) and International Road Assessment Program (iRAP) iRAP coordinates RAP efforts occurring in Europe, Australia, and the United States and pro- vides software (ViDA) for assessing the safety of a given section of road. ViDA has two primary applications: generating star ratings for 100-meter sections of roadway and developing safer roads investment plans for networks of these roadway segments. usRAP has developed a tool to facilitate the coding of roadway video logs to collect input variables needed for ViDA. usRAP coordinates the safer roads investment plans developed in the United States. Star ratings are assigned to a roadway based on the design features, traffic control, and other characteristics of the roadway that can be observed by visual inspection of a picture or video of the roadway. The ViDA software does not evaluate road segments and intersections separately; however, when a segment includes an intersection, the features of that intersection are incor- porated into the safety estimation. A road safety score is determined for each 100-meter road

Literature Review and Survey of Practice 49   segment, and the star ratings (1 through 5) are assigned for specific bands of the road safety score. One-star roads have the fewest safety-related design and traffic operational features, while five-star roads have many safety-related design and traffic control features. Separate star ratings are assigned for vehicle occupants, motorcyclists, bicyclists, and pedestrians because the features that affect crash frequencies for these various travel modes differ. The star ratings consider fac- tors related to both crash likelihood and crash protection. Star ratings are strongly influenced by traffic speeds on the roadway (whenever possible, represented by the higher of the speed limits and the 85th percentile speed) but are not influenced by traffic volumes. Therefore, the star ratings represent the crash potential for an individual user and compare the “built-in” safety of one road segment to the next. Star ratings should not be used to compare the expected number of crashes between roadway segments. Road safety scores are calculated as the summation of individual crash-type scores developed separately for each category of user. The general form of a crash-type score is as follows: Crash Type Score Likelihood Severity Operating speed External flow influence Median traversability # # # # - = where: Likelihood: refers to road attribute contributing factors that account for the chance that a crash will be initiated. Severity: refers to road attribute contributing factors that account for the severity of a crash. Operating speed: refers to factors that account for the degree to which the potential of a crash changes with speed. External flow influence: accounts for the degree to which a person’s potential of being involved in a crash is a function of another person’s use of the road. Median traversability: accounts for the potential that an errant vehicle will cross a median (only applies to run-off and head-on crashes when calculating the crash-type score for vehicle occupants and motorcyclists). For pedestrians, the crash types for which scores are developed include along-the-roadway crashes (calculated separately for the driver and passenger sides of the roadway), crossing-the- major-road crashes, and crossing-the-minor-road crashes. For bicyclists, the crash types include run-off crashes, along-the-roadway crashes, and intersection crashes. The crash types for pedestrians and bicyclists are a product of estimations of crash likelihood, crash severity, operating speed, and external flow influence. “Likelihood” is a measure of the degree to which a specific roadway feature makes a crash more or less likely than a baseline condition (generally, the absence of a feature). This measure is the equivalent of a CMF, and much of the information used to develop these values comes from the CMF Clearinghouse. Table 20 shows the roadway features that impact the likelihood of pedestrian and bicycle crashes. In addition to road safety scores and star ratings, ViDA can produce Safer Roads Investment Plans (SRIPs). SRIPs are a prioritized list of countermeasures that can cost-effectively improve star ratings and reduce infrastructure-related crash potential. The plans are based on an eco- nomic analysis of a range of countermeasures. The cost of implementing the countermeasure is compared with the reduction in crash costs that would result from its implementation. The reduction in crash costs is based on the reduction in the number of fatal and serious injury crashes estimated for the 100-meter roadway segment due to countermeasures proposed in the SRIP. Like the star ratings, fatality estimates are made separately for each roadway user group. Fatal- ity estimates account for traffic volumes and allow for the performance of roadway segments to

50 Pedestrian and Bicycle Safety Performance Functions Road User Crash Type Roadway Feature Coded into ViDA Pedestrians Along the roadway Sidewalk Curvature Quality of curve Sight distance Lane width Delineation Grade Road condition Speed management, traffic calming Vehicle parking Shoulder rumble strips Street lighting School zone warning Crossing an intersection leg Number of lanes Median type Pedestrian crossing Pedestrian crossing quality Intersection type Intersection quality Pedestrian fencing Skid resistance, grip Street lighting Sight distance Vehicle parking Speed management, traffic calming School zone warning Bicyclists Run-off Lane width Curvature Quality of curve Delineation Street lighting Road condition Grade Skid resistance, grip Along the roadway Facilities for bicycles Curvature Quality of curve Sight distance Lane width Delineation Grade Road condition Speed management, traffic calming Shoulder rumble strips Vehicle parking Skid resistance Street lighting Crossing the major/minor road Intersection type Intersection quality Property access points Skid resistance, grip Facilities for bicycles Street lighting Sight distance Intersection channelization Speed management, traffic calming Pedestrian crossing Table 20. Roadway features that impact the likelihood of pedestrian and bicycle crash types in usRAP.

Literature Review and Survey of Practice 51   be compared. That is, a segment with a lower star rating may be expected to have fewer fatalities than a segment with a higher star rating simply because fewer people are using the facility. The equations for fatality estimates are similar in form to the SPFs used in the HSM. For each user group and crash type, the equation for the fatality estimate takes the following form: (2-18)Fatalities S S a AADP AADB CFR 365 10 b 9# # #= a k where: Fatalities = Number of fatalities expected per year per 100 million vehicle miles trav- eled for the crash type and the user group (e.g., pedestrian crashes along the roadway). SRS = Star rating score for the specific user group and crash type. AADP/AADB = Average annual daily pedestrian or bicycle volume. a, b = Coefficient and power for AADP/AADB term to allow for a nonlinear rela- tionship between AADP/AADB and crash frequency. CF = Calibration factor to local conditions. This factor is calculated as the ratio of actual fatalities observed on the segment to the number of crashes calculated using the ViDA estimation. The calibration factor incorporates the safety context of the road outside of the road’s design features. Elements of the context might include seat belt usage, enforcement, driver education programs, rates of impaired driving, prominence of vehicle safety features in the local fleet, and other characteristics that may differ between locations. The calibration factor also allows for two roadway segments with the same star rating score and similar volumes to have different expected fatality rates due to differences in the safety context. Serious injury estimates are determined based on the number of fatalities calculated as shown in Equation 2-18. The estimate for serious injuries for a given user group and crash type is equal to the estimated number of fatalities multiplied by the ratio of observed serious injuries to observed fatalities for that roadway segment. This ensures that the estimated ratio of fatalities to serious injuries is the same as the observed ratios for these crashes. 2.1.5.8 Methods for Estimating Bicycle Safety Performance in the Netherlands In 2014, the Dutch Institute for Road Safety Research (SWOV) published a study on the first development phase of CycleRAP (Wijlhuizen, Dijkstra, and van Petegem 2014). CycleRAP is a model developed to assess the crash potential of bicyclists and light mobility vehicle users on roads and other facilities (iRAP, n.d.-c). Cycle crash types not involving motor vehicles are cen- tral to the CycleRAP model and underpin the need for a model to specifically assess the safety performance of bicyclists and light mobility vehicle users. The reason that the Netherlands and other countries are increasingly interested in cycle crash types not involving motor vehicles is because • Underreporting crashes that do not involve motor vehicles is high. This is because those that are involved are often taken directly to hospitals without the crashes being reported to the police. Crashes may also be located off the road network (e.g., on a cycling path) that is not routinely considered as part of road safety assessments and audits. • These crash types account for a high proportion of serious injuries and fatalities for this road user group. For example, a study of England’s hospital records between 1999 and 2005 showed that of the 37,504 pedal cyclists injured in traffic collisions in England, 67 percent were involved in a noncollision.

52 Pedestrian and Bicycle Safety Performance Functions SWOV used hierarchical, multiple regression models to investigate the relationship between cycling infrastructure characteristics and bicycle crashes (Wijlhuizen, Dijkstra, and van Petegem 2014). Initial models included the following independent variables: • Total score: number of unsafe 25-meter locations/km. • Quality score: number of unsafe 25-meter locations/km. • Height-and-length profile score: number of unsafe 25-meter locations/km. • Obstacle score: number of unsafe 25-meter locations/km. • Total number of intersections/km. • Number of major intersections/km. • Number of roundabouts/km. • Number of small intersections/km. • Bicycle intensity. • Motor vehicle intensity. Crash data for model development were obtained from ambulance records. Models were developed based on incidents when bicyclists were transported to the hospital and incidents when bicyclists were not transported to the hospital. Variables that proved to have a statistically significant relationship to bicycle crashes included: • Bicycle intensity. • Motor vehicle intensity. • Height-and-length density. • Major intersection density. • Minor intersection density. • Roundabout density. In another effort for the city of Amsterdam, SWOV developed a network safety index (NSI) to estimate the frequency of bicycle crashes and establish priorities for making safety improvements (Wijlhuizen et al. 2017). The NSI was developed for roads with a speed of 50 km/h in Amster- dam because the most serious crashes occur in the city on these types of roads. The NSI score estimates the number of traffic crashes per 1,000 m of street in the period from 2009 to 2012. The formula for the NSI is as follows: _ _ . . . . . . . . . exp TramBusAccess 50 0 480 1 00 36 8 0 572 10 2 2 00 0 541 0 572 10 2 Crashes L NSI MVT intensity Bicycle intensity LengthHeight q Density Obstacles_Density SmallCrossing Points_Density Roundabout_Density intercept Creditable Speeds Limit Score Density TramBusAccess intercept . .0 385 0 465# # # # # # # # # = + + + + + - + - + + - ` ` j j j Figure 7 is an example NSI heat map of Amsterdam showing color-coded streets to distinguish the degree of road safety based on their infrastructure characteristics and the intensity of bicycle and motor vehicle traffic. In the heat map, streets are divided into five equally large groups. The dark blue streets are the 20 percent of streets with the highest NSI score (i.e., the streets that are considered to have the highest potential for crashes). The green-colored streets have the lowest potential for crashes. 2.1.5.9 Pedestrian and Bicyclist Intersection Safety Index In 2006, Carter et al. developed the Pedestrian and Bicycle Intersection Safety Indices (Ped ISI and Bike ISI) tool (Carter et al. 2006). The purpose of the tool was to enable engineers and plan- ners to proactively identify intersection crossings and approach legs that should be the greatest

Literature Review and Survey of Practice 53   priority for undergoing pedestrian and bicycle safety improvements. The Ped and Bike ISI scores are an evaluation of each approach leg of an intersection rather than the intersection as a whole. The tool applies to intersections with the following characteristics: • Three- or four-leg intersections. • Signalized, two-way, or four-way stops. • Traffic volume from 600 to 5,000 veh/day. • One-way or two-way roads. • One to four through lanes. • Speed limit from 15 to 45 mph. To develop the safety indices, researchers collected video and data from 68 pedestrian crossing intersections and 67 bicycle approaches at intersections in Florida, Pennsylvania, California, and Oregon. Safety ratings (i.e., opinions) from experts and bicycle-/pedestrian-motorist interac- tions from video analysis of each site were used to generate multivariate linear regression models for ranking. The resulting models are provided in Table 21 (pedestrians) and Table 22 (bicyclists). For bicyclists, separate models were developed depending on the type of maneuver the bicyclist would be making. With the models, a larger ISI indicates a higher priority for improvement. 2.1.5.10 Level of Traffic Stress (LTS) For bicycle networks, the level of traffic stress (LTS) for users is a measure defined as “the ability of a network to connect travelers’ origins to their destinations without subjecting them Round 0.00 Figure 7. Example NSI heat map of Amsterdam (Wijlhuizen et al. 2017).

54 Pedestrian and Bicycle Safety Performance Functions Ped ISI = 2.372 – 1.867SIGNAL – 1.807STOP + 0.335THRULANES + 0.018SPEED + 0.006(MAINADT×SIGNAL) + 0.238COMM SIGNAL Signal-controlled crossing 0 = no1 = yes STOP Stop-controlled crossing 0 = no1 = yes THRULANES Number of through lanes on street being crossed (both directions) 1,2,3,… SPEED Eighty-fifth percentile speed of street being crossed Speed in mph MAINADT Main street traffic volume ADT in thousands COMM Predominant land use in surrounding area is commercial development (i.e., retail, restaurants) 0 = not predominantly commercial area 1 = predominantly commercial area Table 21. Pedestrian intersection safety index (Carter et al. 2006). Through Bike ISI = 1.13 + 0.019MAINADT + 0.815MAINHISPD + 0.650TURNVEH + 0.470(RTLANES×BL) + 0.023(CROSSADT×NOBL) + 0.428(SIGNAL×NOBL) + 0.200PARKING Right Turn Bike ISI = 1.02 + 0.027MAINADT + 0.519RTCROSS + 0.151CROSSLNS + 0.200PARKING Left Turn Bike ISI = 1.100 + 0.025MAINADT + 0.836BL + 0.485SIGNAL + 0.736(MAINHISPD×BL) + 0.380(LTCROSS×NOBL) + 0.200PARKING BL Bike lane presence 0 = none or wide curb lane (WCL)1 = bike lane (BL) or bike lane crossover (BLX) CROSSADT Cross-street traffic volume ADT in thousands CROSSLANS Number of through lanes on cross street 1, 2, … LTCROSS Number of traffic lanes for bicyclists to cross to make a left turn 0,1, 2, … MAINADT Main street traffic volume ADT in thousands MAINHISPD Main street speed limit ≥ 35 mph 0 = no 1 = yes NOBL No bike lane present 0 = BL or BLX1 = none or WCL PARKING On-street parking on main street approach 0 = no 1 = yes RTCROSS Number of traffic lanes for bicyclists to cross to make a right turn 0,1, 2, … RTLANES Number of right-turn traffic lanes on main street approach 0, 1, 2, … SIGNAL Traffic signal at intersection 0 = no1 = yes TURNVEH Presence of turning vehicle traffic across the path of through bicyclists 0 = no 1 = yes Table 22. Bicyclist intersection safety index (Carter et al. 2006). to unacceptably stressful links” (Furth and Mekuria 2013). The LTS method categorizes streets and intersections from LTS 1 (suitable for children) through LTS 4 (suitable for riders who are comfortable sharing the road with autos traveling at 35 mph or more) based on simple potential crash-contributing factors. The four categories of LTS are as follows: • LTS 1: Anybody would bike on it. • LTS 2: For basic adult bicyclists. • LTS 3 or 4: For advanced bicyclists. Criteria for LTS categorization are listed in Table 23.

Literature Review and Survey of Practice 55   2.1.5.11 ActiveTrans Priority Tool Lagerwey et al. (2015) developed the ActiveTrans Priority Tool, a step-by-step method for pri- oritizing improvements to pedestrian and bicycle facilities, either separately or together as part of a “Complete Streets” evaluation approach. The user selects general categories to consider in the prioritizing scheme, including options for stakeholder input; constraints (both cost and legal); opportunities (upcoming projects); safety; existing conditions; demand; connectivity; equity; and compliance. The user assigns weights to each of these categories to reflect the desired priority in the project selection and ranking process. Within each of the categories, predefined project scoring criteria are selected and entered by the user, or entirely custom criteria can be entered. Scoring criteria data can be converted to scaled scores using a method that best fits the data (e.g., proportionate, inverse proportionate, quantile scaling). For example, the predefined safety vari- ables include total crash frequency, fatal and serious injury crash frequency, and rates for each of these variables for both pedestrian and bicycle crashes. 2.1.5.12 TransPed TransPed is based on an interactive geographic information system (GIS) tool designed to assist in the planning and analysis of pedestrian and bicycle transportation (FDOT 2017a). The tool includes a breadth of traditional transportation data such as existing infrastructure, available routes, traffic counts, forecasts, and crashes, as well as information about land use and socioeco- nomic characteristics pertinent to travel by alternative modes. The data are amalgamated into a Composite Bike/Ped Suitability Index that shows the spectrum of opportunity for active trans- portation and a Bike/Ped Quality of Service grade that can be used for prioritization for infra- structure improvements through spatial or attribute-driven analyses. The prioritization scoring system used within TransPed is presented in Table 24. 2.1.6 Pedestrian and Bicyclist Countermeasures The HSM Part D does not include any countermeasures proven to directly reduce (or increase) pedestrian or bicycle crashes. Additionally, Part D of HSM2 will no longer provide a list of LTS 1 LTS 2 LTS 3 LTS 4 Al on g a pa rk in g la ne Street width (through lanes per direction) 1 N/A 2 or more N/A Reach from curb (sum of bike lane and parking lane width, including marked buffer and paved gutter) 15 ft or more 14 or 14.5 ft 13.5 ft or less N/A Speed limit or prevailing speed 25 mph or less 30 mph 35 mph 40 mph or more Bike lane blockage (typically occurs in commercial areas) Rare N/A Frequent N/A N ot a lo ng a p ar ki ng la ne Street width (through lanes per direction) 1 2 if directions are separated by a median More than 2 or 2 without a median N/A Reach from curb, including marked buffer and paved gutter 6 ft or more 5.5 ft or less N/A N/A Speed limit or prevailing speed 30 mph or less N/A 35 mph 40 mph or more Bike lane blockage (typically applies in commercial areas) Rare N/A frequent N/A N/A = Not applicable. Table 23. Criteria for level of traffic stress in mixed traffic (Furth and Mekuria 2013).

56 Pedestrian and Bicycle Safety Performance Functions HSM-approved CMFs; rather, Part D of HSM2 will provide information on how to select and apply CMFs and direct users to the CMF Clearinghouse to find high-quality CMFs. The follow- ing subsections summarize recently developed CMFs the research team considered for use as adjustment factors with the pedestrian and bicycle SPFs developed as part of this research. The CMFs presented in Sections 2.1.6.1 through 2.1.6.4 address the following treatments: • Installation of a pedestrian countdown timer. • Installation of a pedestrian hybrid beacon (PHB). • Modification of signal phasing (implement a leading pedestrian interval). • Median treatment for ped/bike safety. Section 2.1.6.5 provides the list of countermeasures included in PEDSAFE and BIKESAFE, which are tools developed by FHWA to provide practitioners with the latest information avail- able for improving the safety and mobility of pedestrians and bicyclists. Section 2.1.6.6 lists countermeasures incorporated into the usRAP/iRAP models for both pedestrians and bicyclists. 2.1.6.1 Install Pedestrian Countdown Timer Van Houten, LaPlante, and Gustafson (2012) evaluated the safety effectiveness of pedestrian countdown timers installed in Detroit and Kalamazoo, Michigan. Van Houten, LaPlante, and Gustafson used a time series approach to quantify the safety effectiveness of the treatment, and 449 sites were included in the evaluation. The resulting CMF and star rating are shown in Exhibit 1. Category Criteria Score Range Default Weight Safety and security Number of pedestrian crashes 0 to 50 25%Number of pedestrian fatalities 0 to 50 Agile, resilient, and quality Closing system gap 0 to 50 10%Pavement condition 0 to 25 Pedestrian level of service 0 to 25 Efficient and reliable mobility Pedestrian demand 0 to 50 15%Means of transportation to work 0 to 50 More transportation choices SunRail station, intercity transit (MegaBus, Greyhound, etc.) 0 to 30 20%Bus stops, trails, bikeshare, park & ride 0 to 20 Underserved population composite 0 to 50 Economic competitiveness Employment density 2010 0 to 30 10%Hotel/motel population density 2010 0 to 40 Residential population density 2010 0 to 30 Quality places Schools vicinity 0 to 25 15% Points of interest vicinity 0 to 25 Land use (residential, office, commercial, institutional, recreation) 0 to 25 Urban area vicinity 0 to 25 Environment and conservation Floodplains/wetlands 0 to 50 5%Vehicle level of service 0 to 50 Table 24. Criteria for scoring system within TransPed (FDOT 2017b). Countermeasure CMF Unadjusted Standard Error of CMF Crash Type Crash Severity Star Ratinga Install pedestrian countdown timer 0.3 Vehicle/pedestrian All 4 aThe star rating is based on a scale of 1 to 5, where a 5 indicates the highest or most reliable rating. Exhibit 1. CMF and star ratings for pedestrian countdown timer (Van Houten, LaPlante, and Gustafson 2012).

Literature Review and Survey of Practice 57   2.1.6.2 Install Pedestrian Hybrid Beacons (PHBs) Zegeer et al. (2017) developed CMFs for several types of pedestrian treatments at unsignal- ized pedestrian crossings including rectangular rapid flashing beacons (RRFBs), pedestrian hybrid beacons (PHBs), pedestrian refuge islands, and Advance Yield or STOP markings and signs. Approximately 1,000 treatment and comparison sites were selected from 14 different cities throughout the United States. Most of the study sites were at intersections on urban, multilane streets. Cross-sectional regression models and before-and-after EB analysis techniques were used to determine the crash effects of each treatment type. The resulting CMF and quality of the CMF for installing pedestrian hybrid beacons are provided in Exhibit 2. 2.1.6.3 Modify Signal Phasing (Implement a Leading Pedestrian Interval) Fayish and Gross (2010) investigated the safety effectiveness of implementing a leading pedes- trian interval (LPI) to reduce pedestrian crashes at signalized intersections. Crash data were obtained at 10 treatment sites and 63 nontreatment sites in State College, Pennsylvania. An EB before-and-after approach was used to estimate the safety effectiveness of the LPI implementations. The resulting CMF and quality of the CMF for implementing an LPI are displayed in Exhibit 3. 2.1.6.4 Median Treatment for Pedestrian/Bicycle Safety Zhang et al. (2017) analyzed the impact of median treatments on pedestrian and bicyclist safety based on data from Maryland. Based on data from 1998 to 2016 for sites in urban areas, Zhang et al. used the EB before-and-after approach to estimate the safety effectiveness of median treat- ments. The resulting CMF and quality of the CMF for a median treatment are shown in Exhibit 4. 2.1.6.5 Countermeasures Included in PEDSAFE and BIKESAFE PEDSAFE and BIKESAFE are online tools that provide users with a list of possible engi- neering, education, or enforcement treatments to improve pedestrian and bicyclist safety and Countermeasure CMF Unadjusted Standard Error of CMF Crash Type Crash Severity Star Ratinga Install pedestrian hybrid beacons 0.453 0.167 Vehicle/pedestrian All 4 Install pedestrian hybrid beacon (PHB or HAWK) with advanced yield or stop markings and signs 0.432 0.134 Vehicle/pedestrian All 4 NOTE: An adjustment factor addressing PHBs is included in the predictive method for pedestrian crashes based on usRAP (see Section 4.2.1.16). HAWK = High-intensity Active crossWalK. aThe star rating is based on a scale of 1 to 5, where a 5 indicates the highest or most reliable rating. Exhibit 2. CMF and star ratings for pedestrian hybrid beacons (Zegeer et al. 2017). Countermeasure CMF Unadjusted Standard Error of CMF Crash Type Crash Severity Star Ratinga Modify signal phasing (implement a leading pedestrian interval) 0.630 0.193 Vehicle/pedestrian All 3 Modify signal phasing (implement a leading pedestrian interval) 0.554 0.235 Vehicle/pedestrian All 3 NOTE: An adjustment factor addressing LPIs is included in the predictive method for pedestrian crashes based on usRAP (see Section 4.2.1.16). aThe star rating is based on a scale of 1 to 5, where a 5 indicates the highest or most reliable rating. Exhibit 3. CMF and star ratings for LPI (Fayish and Gross 2010).

58 Pedestrian and Bicycle Safety Performance Functions mobility. PEDSAFE includes 67 engineering, education, and enforcement countermeasures in the toolkit. The most relevant pedestrian-related treatments that appear reasonable for consid- eration in the development of pedestrian SPFs for the HSM are listed in Table 25. BIKESAFE includes 46 engineering, education, and enforcement countermeasures in the toolkit. The most relevant bicycle-related treatments that appear reasonable for consideration in the development of bicycle SPFs for the HSM are listed in Table 26. 2.1.6.6 Countermeasures Included in usRAP/iRAP A total of 94 countermeasures are incorporated into the usRAP/iRAP model. Table 27 lists the countermeasures and estimated safety effectiveness of the countermeasures that address pedes- trian crashes. Similarly, Table 28 lists the countermeasures that address bicycle crashes. Countermeasure CMF Unadjusted Standard Error of CMF Crash Type Crash Severity Star Ratinga Median treatment for pedestrian and bicycle safety 1.12 0.18 Vehicle/pedestrian, vehicle/bicycle All 3 aThe star rating is based on a scale of 1 to 5, where a 5 indicates the highest or most reliable rating. Exhibit 4. CMF and star rating for median treatment (Zhang et al. 2017). Category Treatment/Countermeasure Along the roadway • Sidewalks, walkways, and paved shoulders At crossing locations • Curb ramps • Marked crosswalks and enhancements • Curb extensions • Crossing islands • Raised pedestrian crossing • Lighting and illumination • Parking restrictions • Pedestrian overpass/underpass • Automated pedestrian detection • Leading pedestrian interval • Advance yield/stop lines Transit • Transit stop improvements • Access to transit • Bus bulb-outs Roadway design • Bicycle lanes • Lane narrowing • Lane reduction (road diet) • Driveway improvements • Raised medians • One-way/two-way street conversions • Improved right-turn slip-lane design Intersection design • Modified T-intersections • Intersection median barriers • Curb radius reduction • Modified skewed intersections Signals and signs • Traffic signals • Pedestrian signals • Pedestrian signal timing • Traffic signal enhancements • Right-turn-on-red restrictions • Advanced stop lines at traffic signals • Left-turn phasing • Push buttons and signal timing • Pedestrian hybrid beacon (PHB) • Rectangular rapid flashing beacon (RRFB) Table 25. Relevant pedestrian treatments in PEDSAFE (FHWA, n.d.).

Literature Review and Survey of Practice 59   Countermeasure Estimated Crash Reduction Central hatching 10%–25% Pedestrian crossing – unsignalized 25%–40% Pedestrian fencing 25%–40% School zones 10%–25% Sight distance (obstruction removal) 25%–40% Skid resistance 25%–40% Pedestrian footpath 40%–60% Pedestrian refuge island 25%–40% Regulate roadside commercial activity 10%–25% Parking improvements 10%–25% Intersection – signalize 25%–40% Shoulder sealing 25%–40% Speed management 25%–40% Street lighting 10%–25% Pedestrian crossing – signalized 25%–40% Traffic calming 25%–40% Restrict/combine direct access points 25%–40% Pedestrian crossing – grade separation 60% or more Service road 25%–40% Table 27. Pedestrian treatments in usRAP/iRAP (iRAP, n.d.-b). Category Treatment/Countermeasure Shared roadway • Roadway surface improvements • Lighting improvements • Parking treatments • Median/crossing island • Driveway improvements • Lane reductions (road diet) • Lane narrowing • Street track treatments On-road bike facilities • Bike lanes • Wide curb lanes • Paved shoulders • Shared bus-bike lanes • Contraflow bike lanes • Separated bike lanes Intersection treatments • Curb radius reduction • Intersection markings • Turn restrictions Markings, signs, and signals • Optimizing signal timing for bicyclists • Bike-activated signal detection • Sign improvements for bicyclists • Pavement marking improvements • School zone improvements • Rectangular rapid flashing beacons (RRFBs) • Pedestrian hybrid beacons (PHBs) • Bicycle signal heads Table 26. Relevant bicycle treatments in BIKESAFE (FHWA, n.d.).

60 Pedestrian and Bicycle Safety Performance Functions 2.2 Survey of Practice The research team conducted a survey of transportation agencies, practitioners, and researchers to learn about their experiences related to the development and use of pedestrian and bicycle SPFs. The research team also contacted select agencies to gather information about their pedes- trian and bicycle count programs, inventory datasets, and crash datasets. The results of these activities are reported in Sections 2.2.1 and 2.2.2. 2.2.1 Web-Based Survey The research team conducted an online survey of transportation agencies, practitioners, and researchers involved in pedestrian and bicycle activities. The objectives of the survey were to: • Assess barriers to collecting pedestrian and bicycle safety performance data and best practices to overcome these barriers. • Understand why and how many transportation agencies use and implement the HSM and, in particular, HSM Part C procedures. • Identify and prioritize site types for which pedestrian and bicycle SPFs are needed in future editions of the HSM. • Identify and prioritize design elements, traffic control features, and land-use information con- sidered by engineers and planners during the project development process and that should potentially be addressed in the new models. • Identify pedestrian and bicycle safety databases and potential agencies (and contacts) to work with toward the completion of the research objectives. The survey was conducted online using the Qualtrics software. Invitations to participate in the survey were sent to each of the following: • Fifty state bicycle and pedestrian coordinators. • FHWA bicycle and pedestrian coordinators. • Bicycle and pedestrian advisory commissions in California. • District bicycle and pedestrian coordinators in Virginia and California. • Agency contacts in select cities, counties, and metropolitan planning organizations (MPOs). Table 28. Bicycle treatments in usRAP/iRAP (iRAP, n.d.-a). Countermeasure Estimated Crash Reduction Central hatching 10%–25% Delineation 10%–25% Pedestrian fencing 25%–40% School zones 10%–25% Bicycle facilities 25%–40% Sight distance (obstruction removal) 25%–40% Parking improvements 10%–25% Regulate roadside commercial activity 10%–25% Shoulder sealing 25%–40% Speed management 25%–40% Intersection – signalize 25%–40% Road surface rehabilitation 25%–40% Street lighting 10%–25% Lane widening 25%–40% Traffic calming 25%–40% Restrict/combine direct access points 25%–40%

Literature Review and Survey of Practice 61   Select international pedestrian and bicycle safety experts: • Chairs of the following TRB standing committees, sections, and task forces for distribution to committee members and friends: – Standing Committee on Transportation Issues in Major Cities (ABE30). – Standing Committee on Accessible Transportation and Mobility (ABE60). – Pedestrians and Cycles section (ANF00). – Standing Committee on Pedestrians (ANF10). – Standing Committee on Bicycle Transportation (ANF20). – Task Force on Arterials and Public Health (ADD55T). – Standing Committee on Transportation and Land Development (ADD30). – Standing Committee on Safety, Data, Analysis, and Evaluation (ANB20). – Standing Committee on Urban Transportation Data and Information Systems (ABJ30). – Standing Committee on Highway Safety Performance (ANB25). An email explaining the overall objective of the research and a link to the survey was sent to all potential participants. Instructions invited recipients to respond to the survey or forward the email with the survey link to a person in their organization qualified to respond to questions related to pedestrian or bicycle safety performance. Multiple responses from the same agency were allowed. The research team received 57 responses to the web-based survey, including responses from 12 state agencies and 14 local agencies. Of those 57 responses, 41 respondents completed the entire survey, and 30 provided their agency information. The survey questions were categorized according to the following topics: • Ongoing data collection efforts and best practices. • Prior experience with the HSM. • Need for pedestrian and bicycle SPFs or other predictive methods. • Data collection efforts, data availability, and willingness to participate in the research. Responses to the survey are summarized below. Responses to categorical questions are summarized by showing both the percentage of the responses and the frequency/number of responses shown in parentheses. For those questions where it makes sense, the categorical responses are ordered from the highest to the lowest number of responses. For those ques- tions that asked agencies to provide further detail, verbatim responses are provided in bul- leted form following the question. As previously mentioned, not all agencies responded to every question. A. Ongoing Data Collection Efforts and Best Practices 1. How does your agency quantify pedestrian and bicycle safety performance on its road- ways? [Select all that apply] We use observed crash frequencies on individual facilities 79.5% (39) We rely on community complaints/requests regarding sites that are unsafe for pedestrians or bicyclists 32.7% (16) Other 24.5% (12) We estimate crash risk by identifying environmental, geometric, roadway, or other features at a location known to represent a risk to pedestrian and bicycle safety performance 18.4% (9) We currently do not have any procedures in place to quantify pedestrian and bicycle safety performance 12.2% (6) We calculate predicted and/or expected crash frequencies using SPFs 4.1% (2) Total responses 100.0% (49)

62 Pedestrian and Bicycle Safety Performance Functions Other responses: • We conducted a bike/pedestrian network screening to identify crash hot spot locations or roadways and intersections with worse safety performance than comparable facilities based on crash type, history, and crash rate. We analyzed the level of traffic stress for cyclists for all roads in our Transportation Planning Organization (TPO) area using a methodology adapted from ‘Low-Stress Neighborhood Bikeability Assessment to Prioritize Bicycle Infrastructure.’ • Our state has a Model Minimum Uniform Crash Criteria (MMUCC)-compliant electronic crash reporting system. We are able to quantify pedestrian and bicycle safety performance on roadways using observed crash frequencies on individual facilities. Identifying and locating pedestrian and bicycle crashes from the existing crash database is not an issue for Connecti- cut (CT). In our crash reports, pedestrian and bicyclist behaviors prior to the crash are also reported. However, we are not able to calculate predicted/expected crash frequencies using SPFs because pedestrian or bicycle traffic volumes are not currently available in CT. • We use the Signal Four Analytics (http://s4.geoplan.ufl.edu/), which maps and includes crash reports to analyze crashes. We also coordinate with FDOT and Miami-Dade County to identify and remedy high-frequency crash locations. • We do not have specific SPFs for pedestrian and bicycle crashes due to the lack of exposure data or counts. We do use SPFs in network screening and analyze the resulting locations for overrepresentations in collision with pedestrian or collision with bicycle collision types. • We obtain collision data from the state’s database (SWTRS) and further clean/refine this data within our own database. • In the past, we have developed pedestrian and bicycle crash frequencies by route ID in each jurisdiction. Recently we are using GIS to prepare geospatial “heat” maps of pedestrian crash densities. Further, we are now identifying “corridors” or route segments using multi-variate GIS data for social-economic, land-use, roadway, and crashes to define “priority” locations within clusters to review for pedestrian and bicycle accommodations and to propose projects. • The State makes crash statistics database (CARE) available to all practitioners but limits disclosure to public and publishing any data. State, local engineers, MPOs, and consultants may use. • At this time, Pennsylvania DOT (PennDOT) counts crash data at a statewide level and does not address predictive modeling. We will evaluate crash clusters when mapping iden- tifies them. PennDOT will also work with local governments to do roadway safety audits on areas of concern. • In my role, we rely on community complaints/requests; however, our agency does have personnel that deal with the issue more directly. • Road Safety Audits. • We have recently tried near-miss video detection analysis at two locations to measure turn- ing speed and frequency of close-calls before and after a geometric safety improvement. 2. What are the primary methods that your agency uses to collect pedestrian volume data? [Select all that apply] Infrared scanners (passive or active) 60.4% (29) Manual counts 52.1% (25) Video cameras 27.1% (13) Other 20.8% (10) We do not collect pedestrian volume data. 20.8% (10) Combination of methods 16.7% (8) Laser scanners 2.1% (1) Radio beam 0.0% (0) LIDAR 0.0% (0) Microwave radar 0.0% (0) Total responses 100.0% (48)

Literature Review and Survey of Practice 63   Combination responses: • Pedestrian data collection is episodic, not routine. When done, we use manual and video. • STRAVA data and a few scanners. Very limited though—not comprehensive. • We have a couple of spots with IR and loops to distinguish bikes and peds. Everything else is manual for pedestrians. Portland State is investigating pedestrian call buttons. • The Memphis MPO received an FHWA grant and retrieved six automated counters including three for pedestrian and bicycle counts. • Only collected on traffic volume counts. • This is for specific projects. There is no statewide effort to count pedestrians. Other responses: • We would only collect pedestrian data on a case-by-case basis for specific project needs. We have used third-party video algorithms for this purpose. • Some data collection is done by others via traffic studies so may be manual or video. • Manual counts by consultants in private traffic studies. • If at all, we collect by cameras. Otherwise, we estimate pedestrian volumes per HSM guidance. • We collect data on self-reported walking trips through our GreenTrips program. GreenTrips rewards participants aged 18 and over for logging nonsingle-occupant vehicle trips on the program’s online platform. We lend infrared counters to the Chattanooga Department of Transportation to conduct pedestrian counts. Agency staff volunteers for manual counts led by the Chattanooga Department of Transportation. 3. What are the primary methods that your agency uses to collect pedestrian infra- structure data (e.g., presence of crosswalks and sidewalks, sidewalk and crosswalk widths, etc.)? [select all that apply] Aerial imagery (e.g., from Google Maps or similar software) 62.2% (28) On-site inspections 57.8% (26) Construction plans and/or as-built drawings 35.6% (16) Video photologs 28.9% (13) Other (please specify) 28.9% (13) We do not collect infrastructure data 8.9% (4) Crowd-sourced maps (e.g., OpenStreets or similar software) 6.7% (3) Total responses 100.0% (45) Other responses: • GIS Mapping using consultant services. • ADA elements are collected through on-site inspections and put in an official database, but other sidewalk/crosswalk data are collected via the other methods checked above. • Request information from TPO jurisdictions about recently constructed facilities. • We’re currently in the process of collecting infrastructure data but only on state-owned facilities. • Regarding the pedestrian infrastructure data, CT does not have a complete inventory for crosswalks and sidewalks yet. However, CT is in the process of creating a complete inter- section inventory, in which the presence of crosswalks on each approach of an intersection is to be collected through aerial imagery (via a Bing maps-based customized software). In the meanwhile, CT DOT has several projects related to midblock pedestrian crossing safety and thus has begun locating the midblock pedestrian crossings on both the state and local road systems and collecting Model Inventory of Roadway Elements (MIRE)- compliant safety data elements. Locating the midblock crosswalks is either through on-site inspection or construction plans, while collecting MIRE attributes for those crosswalks will be on-site or through aerial imagery. CT does not have a pedestrian and bicycle sidewalk

64 Pedestrian and Bicycle Safety Performance Functions inventory but has video log systems from which pedestrian/bicycle infrastructure data can be collected through image processing or manual collections. However, no such effort has been carried out, due to lack of staff time, funding, or video processing expertise. • Paid Fugro to collect data elements on state routes and some MPOs add subcontracts to collect local road data. Data is updated with construction plans. Independent transition plan collects and tracks ADA deficiencies. • We use information from our Transportation Improvement Program from updated proj- ects and our new Bicycle and Pedestrian Report questionnaire. • Although most of our pedestrian infrastructure information is gathered on a per-project basis, in FY18 we have a project to do a more comprehensive data-gathering effort. • PennDOT does not own facilities outside of the curb line (sidewalks and similar), so we don’t track that information. • We collect some pedestrian infrastructure data at the beginning of an improvement project. Typically, this is done using photolog or aerial imagery. We do not have pedestrian infra- structure data statewide. This is similar to how we collect pedestrian volume data. • We are in the process of collecting GIS layers from local jurisdictions. Then we will deter- mine where we do not have information and hire a consultant to create a GIS sidewalk layer through an analysis of aerial imagery. 4. What are the primary methods that your agency uses to collect bicycle volume data? [select all that apply] Manual counts 51.1% (24) Pneumatic tube 40.4% (19) Infrared scanners (active or passive) 29.8% (14) Inductive loops 27.7% (13) Video cameras 21.3% (10) We do not collect bicycle volume data. 19.1% (9) Other (please specify) 19.1% (9) Combination of methods (please specify) 14.9% (7) LIDAR 2.1% (1) Laser scanners 2.1% (1) Piezoelectric strips 0.0% (0) Microwave radar 0.0% (0) Radio beam 0.0% (0) Total responses 100.0% (47) Combination responses: • Video and regional planning data. • Bike data collection is episodic, not routine. When done, we use manual and video. • Also, use STRAVA—open source data. • Only collected during traffic volume counts. • We collect bicycle volumes with our manual turning movement counts. Those counts are completed at a limited number of intersections statewide, and we currently do not have a statewide inventory for these turning movement counts. We have also done some limited counts with pneumatic tubes and infrared scanners by Eco-Counter. Other responses: • Bicycles are included in our annual vehicle counts if they are riding in a lane but often won’t get counted through this method if on the shoulder. • Video sensor technology. • We do not routinely collect bicycle traffic data but would on a case-by-case basis. I am aware of third-party apps that collect user data that can be used to estimate bicycle traffic but do not believe our agency uses that data.

Literature Review and Survey of Practice 65   • Strava. • Currently, we rely on our local planning partners to do the counts as part of their planning process. • Other than that, the Department does not do bicycle counts to estimate statewide numbers. 5. What are the primary methods that your agency uses to collect bicycle infrastructure data (e.g., presence of bicycle lanes, width of bicycle lanes, dedicated bicycle signage, etc.) [select all that apply] Aerial imagery (e.g., from Google Maps or similar software) 58.7% (27) On-site inspections 54.3% (25) Construction plans and/or as-built drawings 47.8% (22) Other (please specify) 32.6% (15) Video photologs 26.1% (12) Crowdsourced maps (e.g., OpenStreets or similar software) 8.7% (4) We do not collect bicycle infrastructure data. 0.0% (0) Total responses 100.0% (46) Other responses: • We (MPO) maintain a database and request info from the local governments annually to update it. • We use all these methods but do not have one centralized database. • Request information from TPO jurisdictions about recently constructed facilities. Currently developing an inventory of state-owned facilities. • Inter-agency coordination of facilities. • Using work orders for maintenance/installation of bike signs and sharrow installation. • Very limited inventory on state system. • Although most of our bicycle infrastructure info is gathered on a per-project basis, in FY18 we have a project to do a more comprehensive data-gathering effort. • There may be some infrastructure data at the beginning of an improvement project. Typi- cally, this is done using photolog or aerial imagery. We do not have infrastructure data statewide. This is similar to how we collect bike volume data. For bike lane or shared-lane markings a permit and maintenance agreement is required so there is some limited infor- mation on bicycle markings. • We collect maps and GIS layers from local jurisdictions that we input into our regional GIS bicycle layer. • Houston Bike plan. 6. What are the biggest barriers/challenges that your agency faces in collecting pedes- trian safety performance data (e.g., crash frequencies) [select all that apply] Pedestrian behavior prior to the crash is not available in existing crash databases. 45.8% (22) Other (please specify). 43.8% (21) Pedestrian crashes are generally not reported to the police. 27.1% (13) Pedestrian safety performance is not a higher priority than other safety issues. 25.0% (12) Pedestrian crashes are difficult to identify in existing crash databases. 20.8% (10) Pedestrian crashes are difficult to accurately match to specific locations on the roadway. 18.8% (9) Total responses 100.0% (48) Other responses: • To calculate actual safety, we need crash RATES, and we do not have 24/7/365 count data on all roads. • Most databases focus on fatalities but more useful would be any injury or fatality pedes- trian crash.

66 Pedestrian and Bicycle Safety Performance Functions • If we have any issues at all, it’ll be related to the fact that Oregon is a self-reporting state. Overall, our crash data is very good, where it exists. We also don’t collect data on ped vs. bike crashes. • They are rare and random events that do not happen frequently (thankfully), but this makes it hard to know where the next will occur or where to best invest funds to have the greatest impact on reducing pedestrian crashes. Plus there are a lot of pedestrian “inter- actions” that do not get coded because no one gets hurt. • Gross pedestrian crash numbers are very low; thus, statistical analysis methods do not work. Also, lack of pedestrian volume data means crash rates cannot be calculated. • Lower-severity crashes may not be reported. Bike-on-ped crashes not reported/recordable. • Data related to pedestrian crashes appears to be improving in accuracy in the state-maintained crash databases that we work with. • No significant barriers. • Crash reports are often incomplete and inconsistent with respect to pedestrian behavior prior to the crash and are sometimes biased against pedestrians due to the fact that the pedestrians are often not in any condition to be interviewed. • Typically, only get data from serious/fatal crashes. Pedestrians who were near-missed or lightly injured do not stick around for reports. Police have difficulties completing data elements for pedestrian crashes due to fatality of pedestrian, missing pedestrian, wrong information from pedestrian or driver, conflict of narrative. Often there is little to no follow up on missing data elements in pedestrian-related crashes. Number of crashes are relatively small and randomly distributed. • Driver and pedestrian behavior preceding and during the incidence is self-reported and/or not complete on the crash reports. Normalizing by pedestrian volumes except in a few key areas in a few cities is not feasible, thus the above-mentioned geospatial crash analysis and land-use methods. • Without the ability to obtain wide-scale pedestrian volumes, it’s not possible to calculate pedestrian crash rates at enough locations to make the data meaningful. • SDOT (Seattle DOT) relies on SPD (Seattle Police Department) collision reports for the majority of our pedestrian safety data. Pedestrians tend to not report to SPD unless they are injured and require the report to file an insurance claim. SPD reports that are gener- ated by pedestrians, rather than by officers, are not able to automatically compile certain data fields in SPD’s database, making it possible to miss out on some pedestrian collisions. • We use the Tennessee Highway and Homeland Security database called Titan for Shelby County and Fayette. It allows us to look at areas of the intersection, the reason for the crash, mph, frequencies, etc. • Pedestrian crash data are available, but there is no pedestrian count or infrastructure data available. • Pedestrian crash data are available, but no pedestrian count data to properly quantify on a systemic level. • New Hampshire is a generally rural state with relatively small numbers of peds, with the exception of the few urbanized areas. • Pre-crash data are not necessarily a barrier; it’s just that analyzing police report dia- grams and narratives are time-consuming manual processes, and those data items are not included in our electronic DBs. We maintain multi-year crash databases and GIS layers for each of the two states we work with. • Also, TxDOT only keeps data for 5 years; this creates a problem if you want to look at data from a longer duration of time. Because of this, we are maintaining our own database of crash data in our region. • Staff time. • Quantifying the pedestrian demand if an improvement is made is somewhat subjective.

Literature Review and Survey of Practice 67   What steps have been taken to overcome these barriers to collecting pedestrian safety perfor- mance data? (Please specify if you have access to any internal or external reports that provide best practices to overcome these challenges.) • We hope to work on a project for how we might do extrapolations to do estimated crash rates. • State now has a standard crash reporting system that allows for searchable data. • Outreach on the responsibilities of drivers/pedestrians to report crashes. • We have produced a report trying to use both risk factors and pedestrian crashes to determine locations that could potentially benefit from investments. In addition, we are researching risk factors to try to determine which risk factors are most important to the selection of improvement sites. • We have tried to work with the various police departments in our region and the Department of Transportation to best gather the data on pedestrian incidents. Data from police reports are not often well-documented or properly explained. It is a challenge. We’ve expected the DOT to continue to push for better data entry. • We used a risk-based analysis to identify pedestrian crash risk factors such as motor vehicle volumes, speeds, crossing distance, presence of transit stops, etc. • Our Traffic Safety Unit has been verifying each reported collision using the actual police report to extract more accurate information. • We devote staff time to reading crash report narratives and confirming that a pedestrian was involved in the crash, the severity of the crash, the geolocation of the crash, and whether or not the crash occurred on private property. We have developed a ‘how to’ document for cleaning the data. • We’re looking at developing specific pedestrian performance measures to indicate how we’re providing for pedestrians. We’re still in the infancy of this effort. • I have reached out to our police office and have been given permission to conduct bike/ped specific training for the officers. I am currently developing the training. • Educating officers on the latest process and technology is an ongoing process that we are always trying to improve. As pedestrian crashes increase in frequency, we anticipate data will be better compiled and screened. • We have conducted a detailed review of the fatal and a sample of the injury crash report diagrams and narratives with Google Images to define the sequence of events and infra- structure in the area of the crash to define trends and needs for a 4E (i.e., enforcement, education, emergency response, and engineering) approach. • Reliance on bike/ped advocacy groups to speak up. • We are developing a website where people can report unreported crashes and near misses. • The State makes crash statistics database (CARE) available to all practitioners but limits disclosure to public and publishing any data. State, local engineers, MPOs, and consultants may use. • Outreach to pedestrian advisory groups promoting the concept of reporting ALL pedes- trian collisions, not just injury collisions. SDOT is also pursuing near-miss study technol- ogy to help analyze locations where collisions are inferred to occur at but have no collision history. SDOT employees are also advised to update our internal collision database when reports are inaccurate. • We are keeping a repository of data and trying to be able to track and map changes that we retrieve from Tennessee DOT and Mississippi DOT. • At the Houston-Galveston Area Council, we are maintaining our own database of crash data. • Work with Police Department on observed issues through frequent discussions/collaboration and monthly coordination meetings with Public Works/transportation staff. • I do not know of any steps being taken. We did discuss possibly using near-miss data for determining high-risk locations.

68 Pedestrian and Bicycle Safety Performance Functions • Do not know of any, but there may be steps taken by others in the agency. • We are discussing it but no concrete plans yet. • Not aware of any. • None. • Nothing. My State DOT does the bare minimum. • Nothing at this time. • I am not aware of an effort to overcome these barriers. 7. What are the biggest barriers/challenges that your agency faces in collecting data on pedestrian volumes, counts, or exposure [select all that apply] Lack of staff time to perform manual counts/set up automated counters 68.1% (32) Funding availability to expand existing count programs 51.1% (24) Funding availability to initiate count programs 48.9% (23) Other (please specify) 27.7% (13) Lack of data collection experience 21.3% (10) Total responses 100.0% (47) Other responses: • Our state mostly relies on manual counts, which rely on volunteers. This usually only takes place in September for a few days and only a few hours within a given day. Not ideal time/frequency for counts. • Lack of management commitment to develop routine counting program. • They don’t seem to care for this data, believe me, I’ve tried. • FHWA doesn’t require pedestrian counts, so the emphasis always has been, and continues to be on motorized traffic. Fighting the mindset that pedestrians and bikes are traffic too and need to be counted. • Rural/small town commitment to a count program. • Current technology has some shortcomings to collect data at all locations. For example, Pyro Box of Eco-Counter just can be tied to the pole at the curb of sidewalk and face to wall. • Infrared counters are only useful at bottleneck locations and not the most convenient to install. • Some cities are beginning programs. Virginia DOT maintains 85% of the network (all county routes outside of cities) a count program would need to be shown as beneficial to safety/infrastructure efforts targeted. Our count program budget was recently reduced to fund other maintenance needs, so adding more nonmotorized counting would be difficult at this time. • Difficult to identify key locations for pedestrian counters. Requires tandem counts with pneumatic tubes to remove bike counts from ped counts. • This is a task that falls to the planning partners. PennDOT does not do counts for estimat- ing statewide pedestrian volumes. Only for some local projects. • Lack of staff time to analyze data. What steps have been taken to overcome these barriers to collecting data on pedestrian volumes, counts, or exposure? (Please specify if you have access to any internal or external reports that provide best practices to overcome these challenges.) • Recently approved multiple MPOs’ requests to purchase bicycle and pedestrian counting equipment. Once the MPOs have the equipment, they can begin to collect 24/7 data on their networks. We have also enforced survey/data requirements within our Safe Routes to Schools (SRTS) projects and programs. This will allow us to collect and review student travel surveys and parents’ perceptions of active transportation.

Literature Review and Survey of Practice 69   • Have investigated some methods to collect counts but have not implemented or tried any technologies. • Reliance on Regional Planning and DOT data. • We have integrated a more robust methodology and template for counting bicyclists and pedestrians, but we are still limited in bandwidth to the need to perform this data collection using manual counts. • Interns. We get them to collect counts by hand (Jamar board) when we don’t have the money or staff time to get them done through our count’s unit. • Working with other agencies to share data collected by each. • We own the proper counters but, in most cases, do not have the authority to install the equipment without help from the DOT or certain localities. They need to initiate or agree to help perform counters that use equipment. • Identified as a need in a policy document. • Commitment from management. We have a good start to a count program and data collec- tion methodology. • They don’t seem to care for this data, believe me, I’ve tried. • We support Chattanooga, the largest jurisdiction in our TPO area with manual and infra- red counts in several locations. • Trying to initiate a counting program but very difficult. Unlike motorists, pedestrians can walk in any direction and any path. It’s hard to know when and where to count them for modeling purposes. • We do currently have a bike and pedestrian count program (https://itre.ncsu.edu/focus /bike-ped/nc-nmvdp/). At present, funding is sufficient to maintain the program, though, the level of funding only permits a phased approach to serving the full state. • Earlier this year, we were the beneficiary of a Technology for Healthy Communities grant, which enabled us to pilot the use of video sensor technology from a start-up company called Numina. Since the pilot has ended, we are in talks with Numina to find a price point that is workable for our budget so that we can keep the sensors running beyond the pilot. • We’ve looked at several software and hardware systems used to assist in collecting pedes- trian count data statewide. We are currently waiting on the results of a Louisiana research project that will identify the best approach to deploying pedestrian counting statewide and how to scale current practices. • Some cities are beginning programs. VDOT maintains 85% of the network (all county routes outside of cities). A count program would need to be shown as beneficial to safety/ infrastructure efforts targeted. Our count program budget was recently reduced to fund other maintenance needs, so adding more nonmotorized counting would be difficult at this time. • Reliance on bike/ped advocacy groups to speak up. • We have recently purchased two portable infrared scanners that collect pedestrian volume information. These scanners communicate wirelessly and have a suite of reports built into the interface. We are making the data available to agency transportation staff. • No steps have been taken that I know of. • Vision Zero group is currently looking to invest in permanent counters for pedestrian facilities. Requires identifying key locations that are representative of the larger area to help reduce the amount of locations needed. This will help develop our existing pedestrian exposure model the safety group uses to help weight SPFs. • We are keeping a repository of data and trying to be able to track and map changes that we retrieve from TDOT and MDOT. Looking at working with other jurisdictions on their counting software and also looking at the projects through our reports that jurisdictions submit to our MPO. When we have our annual call for applications for the Transporta- tion Alternatives Program, we also get an idea of what pedestrian volumes and accident

70 Pedestrian and Bicycle Safety Performance Functions reports. There are also only two staff members that focus on Bicycle and Pedestrian efforts. So we lack the staff time to perform those counts. • None. • We have looked at opportunities to work with metropolitan planning organizations (MPOs) and regional planning commissions (RPCs) to coordinate counting efforts at local and regional levels. • We have hired new staff to assist with data collection. It is challenging to convince upper management that this is essential to our work. • Hiring of consultant to prepare collision history/assessment of last 5 years of collision data. 8. What are the biggest barriers/challenges that your agency faces in collecting pedes- trian infrastructure data (e.g., presence of crosswalks and sidewalks, sidewalk and crosswalk widths, etc.) [select all that apply] Lack of staff time to create/update infrastructure database 64.4% (29) Lack of funding to collect or update data 35.6% (16) Lack of available data sources (e.g., video photologs) to create infrastructure database 31.1% (14) Other (please specify) 28.9% (13) Poor data quality or lack of specificity within existing databases 26.7% (12) Existing data sources are updated too infrequently to maintain accurate database 24.4% (11) Total responses 100.0% (45) Other responses: • We have all this data. • Data are collected, but management support and commitment to incorporate into corpo- rate database and provide ongoing upkeep is negligible. • They don’t seem to care for this data, believe me, I’ve tried. • The state doesn’t have jurisdiction over much of the pedestrian infrastructure. That’s at the local level, and some cities have an inventory, others do not. The DOT is currently collect- ing inventory on state-owned facilities. • We have sufficient resources to collect and update pedestrian facility data on state roads currently. If this changes, any of the above could be factors. • We are collecting some with a new ADA compliance effort that is using a new asset man- agement system (AMS). However, we maintain 56,000 centerline miles, so we are prioritiz- ing where to collect asset data. • The quickness with which things change on the system and the failure to have a systematic way of reporting changes by developers and maintenance. • We have good infrastructure data. • Seattle is changing too rapidly to keep our Master drawings up-to-date on a daily basis. Citywide inventories are done infrequently and are spaced far apart. • We would like to see how other MPOs retrieve this data. We would have to request this information from local jurisdictions. We currently do not have access to that updated data. • We collected this data as part of our ADA transition plan and now have a complete inventory. • PennDOT does not own or maintain these pieces of infrastructure and so does not track them. They are the responsibility of the local governments. • Simple user interface to manage inventory is needed. What steps have been taken to overcome these barriers to collecting pedestrian infrastructure data? (Please specify if you have access to any internal or external reports that provide best practices to overcome these challenges.) • We are beginning to make some changes to our Transportation Alternatives Program that will allow us to consistently collect data on new infrastructure. We will also attempt to update our information from previously awarded projects.

Literature Review and Survey of Practice 71   • Some minimum level of effort has been completed. Using usRAP to document pedestrian activity along the functional classified routes. • Participation in DOT program for enhancing pedestrian signing as a safety measure. • We have taken a more proactive approach to update our asset inventories. • Created a database and inventoried pedestrian infrastructure in transit corridors. • We keep unofficial, individual region (not statewide) databases as we collect the data. • We are beginning to collect this data as part ADA and will put in place some systematic ways to report changes, plus we have been looking at using machine collection methods, but it is still very preliminary. • We were sued under the Americans with Disabilities Act. This is the push that may change the status of pedestrian facility data within the agency. • We have an infrastructure collection program now. Our Linear Referencing System is being updated to include pedestrian infrastructure data. • We devote staff time to update infrastructure databases, but it is a significant undertaking. We intend to update databases annually moving forward to minimize the time required for each update. • In initial talks with other NCDOT units to obtain federal state planning and research funds to more comprehensively develop our pedestrian (and bike) facility geodatabase. • Nothing yet beyond alerting our new Chief of Traffic Engineering that there is a problem. • Coordinating with data collection and other departments and districts to identify resources. • We are collecting some with a new ADA compliance effort that is using a new AMS. How- ever, we maintain 56,000 centerline miles, so we are prioritizing where to collect asset data. • Barriers are still in place. • All MPOs in State have some Bike/Ped plan and have some sort of inventory. Not all are to same level of detail. Many do not review state of repair. • Assets group recently invested heavily in doing a citywide analysis of multiple pedes- trian facilities (curb ramps, sidewalk conditions, crosswalks, etc.). Traffic Operations division is looking to hire on more engineers in order to keep the Master drawings more up-to-date. • Most of the projects that are related to bike/ped funding are not always updated or com- municated with the MPO. We do request that information by retrieving it from our local jurisdictions. So forming regular and updated reports has not been completed consistently. • No steps have been taken that I am aware of. • We’ve created a long-term project to begin comprehensive pedestrian infrastructure data collection. • We are in the process right now of creating a regional sidewalk layer in GIS; currently, it does not exist. The challenge is that most local jurisdictions do not have a sidewalk layer in GIS. At the regional level, it is difficult to manage data if your local partners are not col- lecting data and mapping it. • Hiring interns, temp staff to collect inventory. Still a slow process. 9. What are the biggest barriers/challenges that your agency faces in collecting bicycle safety performance data (e.g., crash frequencies) [select all that apply] Bicyclist behavior prior to the crash is not available in existing crash databases 48.9% (23) Other (please specify) 34.0% (16) Bicycle crashes are generally not reported to the police 25.5% (12) Bicycle crashes are difficult to identify in existing crash databases 23.4% (11) Bicycle safety performance is not a priority because there are too few bicycle crashes reported 23.4% (11) Bicycle crashes are difficult to accurately match to specific locations on the roadway 17.0% (8) Total responses 100.0% (47)

72 Pedestrian and Bicycle Safety Performance Functions Other responses: • We have good data but need additional counts for more accurate crash *rate* calculations. • If we have any issues at all, it’ll be related to the fact that Oregon is a self-reporting state. Overall, our crash data is very good, where it exists. We also don’t collect data on ped vs. bike crashes. • Gross bicycle crash numbers are very low; thus, statistical analysis methods do not work. Also - lack of bicycle volume data means crash rates cannot be calculated. • Bike-on-bike or bike-only crashes not reported. Lower-severity bike-vehicular crashes possibly not reported. • They don’t seem to care for this data, believe me, I’ve tried. • Reports are inconsistent. Accuracy depends on the police officer completing the report (if one is completed at all). • Crash reports are often incomplete and inconsistent with respect to bicyclist behavior prior to the crash and are sometimes biased against bicyclists due to common misconcep- tions about bicyclist behavior and the fact that the bicyclists are often not in any condition to be interviewed. • Not enough for crash frequency ratings. • Again, these are rare and random. Everything said about pedestrians can be repeated here. Plus, just bike crashes with no vehicle are not reported. • Without the ability to obtain wide-scale bicyclist volumes, it’s not possible to calculate bicyclist crash rates at enough locations to make the data meaningful. • As with pedestrian collisions, SDOT relies on SPD reports and bicyclists tend to not report collisions that do not cause injury or bike damage. SDOT can use the Seattle Fire Depart- ment (SFD) locational data as a supplement, but the data therein is lacking (only location and time, no other info). Also, like pedestrian collisions, self-reported events are not auto- matically processed by SPD database, so these have to be manually processed by SDOT employees whenever possible. We also use SFD bicycle reports to create bicycle collision heatmaps. This information only has time, date, latitude, and longitude. Due to the lack of information including movements and causes of collision, we cannot use these for in-depth safety analysis. • Bicycle crash data are available, but there is no bicycle count data available. • Pre-crash data are not necessarily a barrier. It’s just that analyzing police report diagrams and narratives are time-consuming manual processes, and those data items are not included in our electronic DBs. We maintain multi-year crash databases and GIS layers for each of the two states we work with. • Bicycle crashes not involving a motor vehicle aren’t reported to the police. We don’t have any idea of how many of these happen or where. • No significant barriers. What steps have been taken to overcome these barriers to collecting bicycle safety performance data? (Please specify if you have access to any internal or external reports that provide best practices to overcome these challenges.) • Searching of crash history in DOT database for city streets. • Outreach on the responsibilities of drivers/pedestrians to report crashes. • Not sure about the steps. The gap again is police reporting disparity, as with pedestrian crashes. • We used a risk-based analysis to identify bicycle crash risk factors such as motor vehicle volumes, speeds, crossing distance, presence of transit stops, etc. • None. We’ve discussed partnering with the department of health to try and get at ER records.

Literature Review and Survey of Practice 73   • We devote staff time to reading crash report narratives and confirming that a cyclist was involved in the crash, the severity of the crash, the geolocation of the crash, and whether or not the crash occurred on private property. We have developed a ‘how to’ document for cleaning the data. • Trying to develop a consistent way to report crashes and to ensure all crashes are reported. • I have reached out to our police office and have been given permission to conduct bike/ped specific training for the officers. I am currently developing the training. • Improving training and awareness regarding bike crashes. • One way that was trialed was a crowdsourcing application that collected bicycle comments on the safety and rideability of roads in Portland Oregon (by Portland State University); this gave a source of more data on the actual riders’ perception of the roadway. • We are developing a website where people can report unreported crashes and near misses. • Many MPOs have initiated crowdsource documenting of routes. Bicycle clubs have evalu- ated some routes. • Same as pedestrian safety data - promote self-reports for all events, pursue near-miss tech- nology, and update internally when capable. • We use the Tennessee Highway and Homeland Security database called Titan for Shelby County and Fayette. It allows us to look at areas of the intersection, the reason for the crash, mph, frequencies, etc. • We are just beginning discussions internally on how to approach trauma centers and emer- gency rooms for data-sharing efforts. • At the regional level, we cannot direct the local police agencies to do their work differently. We are having these discussions at the regional level through our Regional Safety Council and the Pedestrian-Bicyclist Subcommittee. Hopefully, our Regional Safety Plan will highlight certain solutions for different agencies (plan should be released next year). • Work with Police Department on observed issues through frequent discussions/collaboration and monthly coordination meetings with Public Works/transportation staff. • No steps taken that I know of. • I don’t believe there are efforts being taken to overcome this. • Barriers are still in place. • Do not know. • Not aware of any. • Nothing at this time. 10. What are the biggest barriers/challenges that your agency faces in collecting bicycle volumes, counts, or exposure? [select all that apply] Lack of staff time to perform manual counts/set up automated counters 60.9% (28) Funding availability to expand existing count programs 56.5% (26) Funding availability to initiate count programs 50.0% (23) Lack of funding to collect or update data 39.1% (18) Lack of data collection experience 21.7% (10) Poor data quality or lack of specificity within existing databases 21.7% (10) Other (please specify) 17.4% (8) Total responses 100.0% (46) Other responses: • Lack of management commitment to develop a routine and consistent count program. • They don’t seem to care for this data, believe me, I’ve tried. • Quality control is an issue in our office, as well as lack of staff to ensure counters are work- ing correctly and to analyze data. • Rural/small town commitment to a count program. • Some places cannot collect data due to difficulty of installation of the current technology.

74 Pedestrian and Bicycle Safety Performance Functions • SDOT is ramping up exposure data collection for bicyclists throughout the city. Previous data was ad hoc and has been found to be lacking in our desires moving forward, so we have to develop a new standard for data before accuracy can be confirmed. • PennDOT has historically viewed regular counts as part of the local planning process, not something the agency does. What steps have been taken to overcome these barriers to collecting data on bicycle volumes, counts, or exposure? (Please specify if you have access to any internal or external reports that provide best practices to overcome these challenges.) • Recently approved multiple MPOs’ requests to purchase bicycle and pedestrian counting equipment. Once the MPOs have the equipment, they can begin to collect 24/7 data on their networks. • Have investigated some methods to collect counts but have not implemented or tried any technologies. • Participation in regional count program and data exchange. • Interns. We get them to collect counts by hand (Jamar board) when we don’t have the money or staff time to get them done through our counts unit. • We have the two counter devices, assessing video as an option as well. • Identified as a need in a policy document. • Management support. • We support Chattanooga, the largest jurisdiction in our TPO area with manual and infrared/ tube counts in several locations. • Trying to get additional staff or work with interns. Time-consuming and not very efficient. • We do currently have a bike and pedestrian count program (https://itre.ncsu.edu/focus /bike-ped/nc-nmvdp/). At present, funding is sufficient to maintain the program, though, the level of funding only permits a phased approach to serving the full state. • Again, we share data. Portland State is working on a common database for sharing bike data. We are slowly adding a few permanent bike counters. • Earlier this year, we were the beneficiary of a Technology for Healthy Communities grant, which enabled us to pilot the use of video sensor technology from a start-up company called Numina. Since the pilot has ended, we are in talks with Numina to find a price point that is workable for our budget so that we can keep the sensors running beyond the pilot. • See above-mentioned research project on collecting counts for bicyclists and pedestrians. In general, we have too few bicyclists on state routes to condone doing counts except in highly urbanized areas. We have used Strava data for project in the past and anticipate purchasing it again for wider spread use on both state and local routes. • Like ped volumes some cities are counting bike volumes but only on high-volume facilities at targeted locations. • Expansion of bike counters (sensors and tubes), especially as bicycle facilities are incorpo- rated. Improvements to the quality of data (accuracy, resiliency, and constancy) in order to be more applicable in the future. Citywide approach to bicycle exposure data requires citywide volume modeling. • We are keeping a repository of data and trying to be able to track and map changes that we retrieve from TDOT and MDOT. Looking at working with other jurisdictions on their counting software and also looking at the projects through our reports that jurisdictions submit to our MPO. When we have our annual call for applications for the Transportation Alternatives Program we also get an idea of what pedestrian volumes and accident reports. There are also only two staff members that focus on Bicycle and Pedestrian efforts. So we lack the staff time to perform those counts. • We are considering developing funding for the local MPO/RPC to purchase counters for their planning efforts. It is under discussion right now. • We have looked at opportunities to work with MPO/RPCs to coordinate counting efforts at local and regional levels.

Literature Review and Survey of Practice 75   • We have hired additional staff, and we are increasing our current budget to purchase more Eco Counters (permanent counter locations) and maintain our existing temporary counter program that utilizes Trafx counters (temporary counters are deployed at various locations across the region for approximately 2-week periods). • Hiring of consultant to prepare collision history/assessment of last 5 years of collision data. • Barriers are still in place. • No steps taken that I know of. • None. 11. What are the biggest barriers/challenges that your agency faces in collecting bicycle infrastructure data (e.g., presence of bicycle lanes, width of bicycle lanes, dedicated bicycle signage, etc.)? [select all that apply] Lack of staff time to create/update infrastructure database 67.4% (29) Lack of available data sources (e.g., video photologs) to create infrastructure database 32.6% (14) Existing data sources updated too infrequently to maintain accurate database 30.2% (13) Other (please specify) 25.6% (11) Total responses 100.0% (43) Other responses: • We have all this data. • Data are collected, but management support and commitment to incorporate into corpo- rate database and provide ongoing upkeep are negligible. • No singular central database. • Cost is expensive, and even when the inventory is done, it will only reflect state-owned facilities. • We have bicycle infrastructure data and are continuously compiling it, but it is challenging to keep up with, particularly signage, which can change frequently and may not have been documented appropriately. • We have fairly good data. • Bicycle infrastructure on state highways is generally maintained by host municipality under a Maintenance Agreement and records are spotty at best. • Limited amount of separate facilities—many shared roadways and uncertainty as to how to quantify. • Very little infrastructure has been developed outside of our Class 1 and 2 cities. And few PennDOT projects, until recently, added more than a wide shoulder. So there was nothing to count. • This doesn’t seem to be a problem. What steps have been taken to overcome these barriers to collecting bicycle infrastructure data? (Please specify if you have access to any internal or external reports that provide best practices to overcome these challenges.) • Have been collecting usRAP data, including bicycle activity, for functionally classified routes. • Consultant contract for GIS update. • We have a fairly accurate and updated inventory of bicycle infrastructure data. • We have a fairly up-to-date database. We pester the local governments annually to update it. • We keep unofficial, individual region (not statewide) databases as we collect the data. • At this time, we will have to continue with the equipment we currently own and lend in our program. • Identified as a need in a policy document. • We are creating a singular central database to house these things.

76 Pedestrian and Bicycle Safety Performance Functions • We devote staff time to update infrastructure databases, but it is a significant undertaking. We intend to update databases annually moving forward to minimize the time required for each update. • In initial talks with other NCDOT units to obtain federal state planning and research funds to more comprehensively develop our bike (pedestrian) facility geodatabase. • Nothing yet beyond alerting our new Chief of Traffic Engineering that there is a problem. • We are manually checking projects with certain attributes, searching maintenance and purchase order items for bicycle signing and striping, and reviewing data collection efforts for missed routes. • As part of our Vision Zero Action Plan, we will be collecting data on facilities that will support systematic evaluation. • Project Managers are now required to include infrastructure data updates as part of delivery. Data improvement and updating are expected during project development phases as well. Annual site visits are expected for certain facility types (trails, protected bike lanes, etc.). • We would like to see how other MPOs retrieve this data. We would have to request this information from local jurisdictions. We currently do not have access to that updated data. • We’ve instituted a new policy for applying to construct bicycle lanes on state roads that should significantly lower the bar for applicants. Part of that online process will allow the Department to track when, where applications arrive from and which are approved. It will hopefully automate some of the tracking process. • We hired additional staff and increased funding, but more work needs to be done in order to elevate the importance of this work with our leadership. • Hiring interns, temp staff to collect inventory. Still a slow process. • No steps have been taken that I know of. • None at this time. • Barriers are still in place. B. Prior Experience with Highway Safety Manual (HSM) 1. Has your agency incorporated and/or implemented the HSM into existing policies, practices, or procedures? Yes 49.0% (25) No 27.5% (14) Unsure 23.5% (12) Total responses 100.0% (51) 2. If YES or UNSURE to Question B1, which parts of the HSM are currently being uti- lized by your agency: [select all that apply] Part D – Crash Modification Factors 54.3% (19) Part B – Roadway Safety Management Process 40.0% (14) Part C – Predictive Method 40.0% (14) Unsure 40.0% (14) Part A – Introduction, Human Factors, and Fundamentals 25.7% (9) Total responses 100.0% (35) 3. Has your agency made use of any of the following electronic tools for implementing HSM procedures (select all that apply): Unsure 48.6% (17) Spreadsheet-based tools 34.3% (12) IHSDM 20.0% (7) Safety Analyst 20.0% (7) Others 20.0% (7) Total responses 100.0% (35) NOTE: IHSDM = Interactive Highway Safety Design Model

Literature Review and Survey of Practice 77   Other responses: • Safety Analyst was used and has been discontinued. • We have developed in-house tools for network screening using SPFs on segments. We are currently working on SPFs for intersections and roadway departures. • We attempted Safety Analyst (SA), but our updated GIS-based linear reference system (LRS) has precluded using, so we now apply Virginia-specific SPF for segments and inter- sections to our LRS (without SA functionality). • Developed an investigations manual that uses proportions of crashes from HSM. • Most are case-by-case not programmatic. C. Need for Pedestrian and Bicycle Safety Performance Functions (SPFs) or Other Predictive Methods 1. As a part of this project, the research team will develop (or update existing) pedestrian- and bicycle-specific SPFs (or other predictive methods) that can be used to predict crash frequencies involving pedestrians and bicyclists for incorporation into HSM Part C (Predictive Method). Considering the list of site types with existing or planned SPFs for vehicular collisions in the HSM, on a scale from 1 (lowest priority) to 5 (highest priority), please indicate how important it would be for your agency to have compre- hensive SPFs to predict pedestrian or bicycle crashes for the given site types. Pedestrians: Predictive Method for Rural Two-Lane, Two-Way Roads 1 (Low) 2 3 4 5 (High) Total Responses Wt. Avg Response Four-leg intersections with signal control 13.5% (5) 10.8% (4) 24.3% (9) 29.7% (11) 21.6% (8) 37 3.4 Four-leg intersections with stop control on minor-road approaches 13.5% (5) 13.5% (5) 27.0% (10) 32.4% (12) 13.5% (5) 37 3.2 Undivided two-lane, two-way roadway segments 19.4% (7) 13.9% (5) 30.6% (11) 16.7% (6) 19.4% (7) 36 3.0 Three-leg intersections with signal control 16.7% (6) 19.4% (7) 27.8% (10) 19.4% (7) 16.7% (6) 36 3.0 Three-leg intersections with stop control on minor-road approaches 18.9% (7) 18.9% (7) 29.7% (11) 24.3% (9) 8.1% (3) 37 2.8 Four-leg intersections with all-way stop control 21.6% (8) 16.2% (6) 32.4% (12) 16.2% (6) 13.5% (5) 37 2.8 Single-lane roundabouts 27.0% (10) 16.2% (6) 35.1% (13) 2.7% (1) 18.9% (7) 37 2.7 Multilane roundabouts 32.4% (12) 18.9% (7) 21.6% (8) 8.1% (3) 18.9% (7) 37 2.6 Three-leg intersections with all- way stop control 24.3% (9) 27.0% (10) 32.4% (12) 5.4% (2) 10.8% (4) 37 2.5 Predictive Method for Multilane Highways, Rural 1 (Low) 2 3 4 5 (High) Total Responses Wt. Avg Response Four-leg intersections with signal control 11.4% (4) 22.9% (8) 28.6% (10) 11.4% (4) 25.7% (9) 35 3.2 Undivided four-lane roadway segments 19.4% (7) 13.9% (5) 25.0% (9) 22.2% (8) 19.4% (7) 36 3.1 Four-leg intersections with stop control on minor-road approaches 16.7% (6) 13.9% (5) 33.3% (12) 19.4% (7) 16.7% (6) 36 3.1 Divided roadway segments 27.8% (10) 13.9% (5) 25.0% (9) 22.2% (8) 11.1% (4) 36 2.8 Three-leg intersections with signal control 16.7% (6) 27.8% (10) 27.8% (10) 11.1% (4) 16.7% (6) 36 2.8 Three-leg intersections with stop control on minor-road approaches 19.4% (7) 19.4% (7) 38.9% (14) 13.9% (5) 8.3% (3) 36 2.7 Multilane roundabouts 30.6% (11) 16.7% (6) 27.8% (10) 13.9% (5) 11.1% (4) 36 2.6 Single-lane roundabouts 30.6% (11) 19.4% (7) 27.8% (10) 11.1% (4) 11.1% (4) 36 2.5

78 Pedestrian and Bicycle Safety Performance Functions Predictive Method for Urban and Suburban Arterials 1 (Low) 2 3 4 5 (High) Total Responses Wt. Avg Response Four-leg intersections with signal control 2.9% (1) 0.0% (0) 22.9% (8) 20.0% (7) 54.3% (19) 35 4.2 Four-lane undivided arterials 2.8% (1) 5.6% (2) 22.2% (8) 27.8% (10) 41.7% (15) 36 4.0 Four-leg intersections with stop control on minor-road approaches 0.0% (0) 2.9% (1) 31.4% (11) 25.7% (9) 40.0% (14) 35 4.0 Two-lane undivided arterials 0.0% (0) 2.9% (1) 34.3% (12) 31.4% (11) 31.4% (11) 35 3.9 Four-lane divided arterials 5.7% (2) 5.7% (2) 25.7% (9) 22.9% (8) 40.0% (14) 35 3.9 Five-lane arterials including a center two-way, left-turn lane (TWLTL) 8.8% (3) 5.9% (2) 17.6% (6) 26.5% (9) 41.2% (14) 34 3.9 Three-leg intersections with signal control 2.9% (1) 2.9% (1) 35.3% (12) 20.6% (7) 38.2% (13) 34 3.9 Three-lane arterials including a center TWLTL 0.0% (0) 2.9% (1) 40.0% (14) 31.4% (11) 25.7% (9) 35 3.8 Four-leg intersections with signal control on high-speed roads (speed limits of 50 mph and above) 8.8% (3) 5.9% (2) 20.6% (7) 26.5% (9) 38.2% (13) 34 3.8 Three-leg intersections with stop control on minor-road approaches 2.9% (1) 5.9% (2) 41.2% (14) 20.6% (7) 29.4% (10) 34 3.7 Three-leg intersections with stop control on minor-road approaches on high-speed roads (speed limits of 50 mph and above) 6.1% (2) 6.1% (2) 24.2% (8) 36.4% (12) 27.3% (9) 33 3.7 Four-leg intersections with stop control on minor-road approaches on 8.8% (3) 5.9% (2) 20.6% (7) 35.3% (12) 29.4% (10) 34 3.7high-speed roads (speed limits of 50 mph and above) Three-leg intersections with signal control on high-speed roads (speed limits of 50 mph and above) 8.8% (3) 5.9% (2) 23.5% (8) 29.4% (10) 32.4% (11) 34 3.7 Six-lane arterials 15.2% (5) 12.1% (4) 12.1% (4) 21.2% (7) 39.4% (13) 33 3.6 Four-leg intersections with all-way stop control 8.8% (3) 5.9% (2) 32.4% (11) 23.5% (8) 29.4% (10) 34 3.6 One-way arterials 11.4% (4) 14.3% (5) 31.4% (11) 20.0% (7) 22.9% (8) 35 3.3 Three-leg intersections with all-way stop control 8.8% (3) 14.7% (5) 38.2% (13) 17.6% (6) 20.6% (7) 34 3.3 Multilane roundabouts 20.0% (7) 11.4% (4) 20.0% (7) 20.0% (7) 28.6% (10) 35 3.3 Single-lane roundabouts 20.0% (7) 11.4% (4) 20.0% (7) 31.4% (11) 17.1% (6) 35 3.1 Predictive Method for Ramps 1(Low) 2 3 4 5 (High) Total Responses Wt. Avg Response Four-leg ramp terminals at four- quadrant partial cloverleafs 11.8% (4) 5.9% (2) 50.0% (17) 14.7% (5) 17.6% (6) 34 3.2 Three-leg ramp terminals at two- quadrant partial cloverleafs 11.8% (4) 8.8% (3) 47.1% (16) 14.7% (5) 17.6% (6) 34 3.2 Four-leg ramp terminals with diagonal ramps 8.8% (3) 20.6% (7) 38.2% (13) 20.6% (7) 11.8% (4) 34 3.1 Ramp terminals at single-point diamond interchange 11.8% (4) 11.8% (4) 38.2% (13) 26.5% (9) 11.8% (4) 34 3.1 Three-leg ramp terminals with diagonal exit ramps 11.8% (4) 20.6% (7) 38.2% (13) 17.6% (6) 11.8% (4) 34 3.0 Three-leg ramp terminals with diagonal entrance ramps 11.8% (4) 20.6% (7) 35.3% (12) 20.6% (7) 11.8% (4) 34 3.0

Literature Review and Survey of Practice 79   Bicycles: Predictive Method for Rural Two- Lane, Two-Way Roads 1 (Low) 2 3 4 5 (High) Total Responses Wt. Avg Response Undivided two-lane, two-way roadway segments 17.6% (6) 14.7% (5) 26.5% (9) 17.6% (6) 23.5% (8) 34 3.1 Four-leg intersections with signal control 14.7% (5) 14.7% (5) 38.2% (13) 11.8% (4) 20.6% (7) 34 3.1 Four-leg intersections with stop control on minor-road approaches 14.7% (5) 14.7% (5) 44.1% (15) 8.8% (3) 17.6% (6) 34 3.0 Three-leg intersections with stop control on minor-road approaches 14.7% (5) 23.5% (8) 38.2% (13) 5.9% (2) 17.6% (6) 34 2.9 Three-leg intersections with signal control 17.6% (6) 23.5% (8) 35.3% (12) 5.9% (2) 17.6% (6) 34 2.8 Four-leg intersections with all-way stop control 20.6% (7) 20.6% (7) 38.2% (13) 5.9% (2) 14.7% (5) 34 2.7 Three-leg intersections with all-way stop control 20.6% (7) 29.4% (10) 35.3% (12) 2.9% (1) 11.8% (4) 34 2.6 Single-lane roundabouts 26.5% (9) 29.4% (10) 23.5% (8) 8.8% (3) 11.8% (4) 34 2.5 Multilane roundabouts 30.3% (10) 24.2% (8) 21.2% (7) 9.1% (3) 15.2% (5) 33 2.5 Predictive Method for Multilane Highways, Rural 1 (Low) 2 3 4 5 (High) Total Responses Wt. Avg Response Undivided four-lane roadway segments 17.6% (6) 17.6% (6) 26.5% (9) 17.6% (6) 20.6% (7) 34 3.1 Four-leg intersections with stop control on minor-road approaches 17.6% (6) 17.6% (6) 32.4% (11) 14.7% (5) 17.6% (6) 34 3.0 Divided roadway segments 17.6% (6) 20.6% (7) 29.4% (10) 17.6% (6) 14.7% (5) 34 2.9 Three-leg intersections with stop control on minor-road approaches 17.6% (6) 20.6% (7) 35.3% (12) 11.8% (4) 14.7% (5) 34 2.9 Four-leg intersections with signal control 17.6% (6) 23.5% (8) 26.5% (9) 14.7% (5) 17.6% (6) 34 2.9 Three-leg intersections with signal control 20.6% (7) 29.4% (10) 26.5% (9) 11.8% (4) 11.8% (4) 34 2.6 Single-lane roundabouts 23.5% (8) 32.4% (11) 29.4% (10) 5.9% (2) 8.8% (3) 34 2.4 Multilane roundabouts 29.4% (10) 29.4% (10) 20.6% (7) 8.8% (3) 11.8% (4) 34 2.4 Predictive Method for Urban and Suburban Arterials 1 (Low) 2 3 4 5 (High) Total Responses Wt. Avg Response Three-lane arterials including a center two-way, left-turn lane (TWLTL) 0.0% (0) 0.0% (0) 32.4% (11) 35.3% (12) 32.4% (11) 34 4.0 Four-lane undivided arterials 0.0% (0) 2.9% (1) 38.2% (13) 17.6% (6) 41.2% (14) 34 4.0 Five-lane arterials including a center TWLTL 2.9% (1) 5.9% (2) 26.5% (9) 23.5% (8) 41.2% (14) 34 3.9 Four-leg intersections with signal control 0.0% (0) 6.1% (2) 27.3% (9) 33.3% (11) 33.3% (11) 33 3.9 Four-lane divided arterials 0.0% (0) 8.8% (3) 32.4% (11) 26.5% (9) 32.4% (11) 34 3.8 Four-leg intersections with stop control on minor-road approaches 0.0% (0) 5.9% (2) 32.4% (11) 35.3% (12) 26.5% (9) 34 3.8 Two-lane undivided arterials 2.9% (1) 8.8% (3) 26.5% (9) 35.3% (12) 26.5% (9) 34 3.7 Three-leg intersections with signal control 0.0% (0) 12.1% (4) 33.3% (11) 30.3% (10) 24.2% (8) 33 3.7 Three-leg intersections with stop control on minor-road approaches 0.0% (0) 15.2% (5) 33.3% (11) 30.3% (10) 21.2% (7) 33 3.6 Four-leg intersections with stop control on minor-road approaches on high-speed roads (speed limits of 50 mph and above) 5.9% (2) 14.7% (5) 26.5% (9) 20.6% (7) 32.4% (11) 34 3.6 Four-leg intersections with signal control on high-speed roads (speed limits of 50 mph and above) 5.9% (2) 14.7% (5) 26.5% (9) 17.6% (6) 35.3% (12) 34 3.6 (continued on next page)

80 Pedestrian and Bicycle Safety Performance Functions Six-lane arterials 11.8% (4) 8.8% (3) 29.4% (10) 20.6% (7) 29.4% (10) 34 3.5 Three-leg intersections with signal control on high-speed roads (speed limits of 50 mph and above) 5.9% (2) 17.6% (6) 26.5% (9) 20.6% (7) 29.4% (10) 34 3.5 Four-leg intersections with all-way stop control 9.1% (3) 9.1% (3) 33.3% (11) 24.2% (8) 24.2% (8) 33 3.5 Three-leg intersections with stop control on minor-road approaches on high-speed roads (speed limits of 50 mph and above) 5.9% (2) 17.6% (6) 32.4% (11) 17.6% (6) 26.5% (9) 34 3.4 One-way arterials 2.9% (1) 20.6% (7) 35.3% (12) 26.5% (9) 14.7% (5) 34 3.3 Multilane roundabouts 20.6% (7) 11.8% (4) 20.6% (7) 17.6% (6) 29.4% (10) 34 3.2 Three-leg intersections with all-way stop control 8.8% (3) 17.6% (6) 35.3% (12) 17.6% (6) 20.6% (7) 34 3.2 Single-lane roundabouts 20.6% (7) 11.8% (4) 32.4% (11) 14.7% (5) 20.6% (7) 34 3.0 Predictive Method for Urban and Suburban Arterials 1 (Low) 2 3 4 5 (High) Total Responses Wt. Avg Response Predictive Method for Ramps 1(Low) 2 3 4 5 (High) Total Responses Wt. Avg Response Four-leg ramp terminals at four- quadrant partial cloverleafs 11.8% (4) 20.6% (7) 41.2% (14) 5.9% (2) 20.6% (7) 34 3.0 Three-leg ramp terminals at two- quadrant partial cloverleafs 11.8% (4) 20.6% (7) 41.2% (14) 5.9% (2) 20.6% (7) 34 3.0 Ramp terminals at single-point diamond interchange 11.8% (4) 20.6% (7) 41.2% (14) 11.8 % (4) 14.7% (5) 34 3.0 Three-leg ramp terminals with diagonal exit ramps 14.7% (5) 29.4% (10) 32.4% (11) 8.8% (3) 14.7% (5) 34 2.8 Three-leg ramp terminals with diagonal entrance ramps 14.7% (5) 29.4% (10) 29.4% (10) 11.8 % (4) 14.7% (5) 34 2.8 Four-leg ramp terminals with diagonal ramps 11.8% (4) 32.4% (11) 35.3% (12) 5.9% (2) 14.7% (5) 34 2.8 2. Which pedestrian-related design elements, traffic control features, and/or land-use char- acteristics would your agency most want incorporated into HSM predictive methods or other predictive methods for estimating pedestrian crash frequency? [select all that apply] Roadway Segments: Presence of midblock crosswalk 85.3% (29) Presence of rectangular rapid flash beacons 85.3% (29) Presence of pedestrian sidewalk or shared-use path 82.4% (28) Presence of pedestrian refuge island 79.4% (27) Presence of pedestrian hybrid beacon 73.5% (25) Presence of raised median 73.5% (25) Adjacent land-use types (e.g., commercial, industrial, institutional, residential) 67.6% (23) Driveway density 64.7% (22) Presence of bus stops 64.7% (22) Paved shoulder 52.9% (18) Presence of on-street parking 52.9% (18) Buffer width (distance between sidewalk and travel way) 50.0% (17) Presence of raised curb 50.0% (17) Width of sidewalk or shared-use path 41.2% (14) Other (please specify) 23.5% (8) Total responses 100.0% (34) Other responses: • Signage: posted speed limits, crossing warnings, etc. • Presence of schools specifically. Roadway traffic volumes. Roadway speed.

Literature Review and Survey of Practice 81   • Presence of marked crosswalk and/or beacons. • Any measure that is used to enhance crossing would be most important (but would like to see all the ones above). • Segment length and lighting. • All of the above. • No preference. Intersections: Presence of marked crosswalks 85.3% (29) Crossing distance 82.4% (28) Presence of pedestrian signal 73.5% (25) Presence of pedestrian refugee island 73.5% (25) Presence of lighting 73.5% (25) Presence of leading or lagging pedestrian interval 70.6% (24) Presence of channelized right-turn lane 70.6% (24) Presence of curb extensions 67.6% (23) Presence of raised pedestrian crossing 64.7% (22) Presence of rectangular rapid flash beacons (RRFBs) 64.7% (22) Presence of right-turn-on-red restrictions 64.7% (22) Presence of protected/permissive left-turning (PPLT) phases at signal 64.7% (22) Presence of bus stops 64.7% (22) Presence of pedestrian-actuated signal 61.8% (21) Presence of protected pedestrian phase 61.8% (21) Presence of pedestrian countdown signal 58.8% (20) Presence of pedestrian hybrid beacon 58.8% (20) Presence of advance yield/stop lines 47.1% (16) Presence of advanced yield or stop signs 44.1% (15) Presence of curb ramps 41.2% (14) Crosswalk width 35.3% (12) Other (please specify) 32.4% (11) Total responses 100.0% (34) Other responses: • On-street vehicle parking landscape designs. Electronic traffic enforcement. Lighting Sig- naling: bicycle or pedestrian phase signal cycle. • Pedestrian delay. • Adjacent land use (schools, bars, apartment buildings). • Curb radii. • The most challenging is not the signalized, it is RRFBs and getting them to put median refuges in. Also, PPLT using Flashing Yellow arrow and measures that help pedestrians on side streets probably not pedestrian or leading pedestrian. • Signal head type (size, reflectivity, backing, etc.). • All of the above. • No preference. 3. Which bicycle-related design elements, traffic control features, and/or land-use char- acteristics would your agency most want incorporated into HSM predictive methods or other predictive methods for estimating bicycle crash frequency? [select all that apply]

82 Pedestrian and Bicycle Safety Performance Functions Roadway Segments: Presence of dedicated bicycle lane 85.3% (29) Presence of protected/separated bicycle lane 76.5% (26) Presence of dedicated turn lanes 67.6% (23) Presence of share lane markings and/or signs 64.7% (22) Driveway density 64.7% (22) Adjacent land-use types (e.g., commercial, industrial, institutional, residential) 64.7% (22) Presence and width of buffer (between bike and motor vehicle lanes or between bike and parking lanes) 61.8% (21) Presence of shared path 61.8% (21) Width of dedicated bicycle lane 58.8% (20) Presence and placement of parking lane 58.8% (20) Presence of paved shoulders 58.8% (20) Width of motor vehicle lanes 58.8% (20) Presence of colored bicycle lane 52.9% (18) Presence of bus stops 50.0% (17) Presence of bicycle boulevard 38.2% (13) Other (please specify) 20.6% (7) Total responses 100.0% (34) Other responses: • Landscape design, Signage: Bikes May Use Full Lane vs. Share the Road, electronic traffic enforcement. • All of the above plus information on heavy vehicle operations (e.g., loading and also how much activity there is regarding ride-sharing). • Roadway traffic volumes, roadway speed. • Number of travel lanes, posted/design speed. • Roadway slope. • All of these. • No preference. Intersections: Presence of dedicated bicycle lane 82.4% (28) Presence of green striping in conflict zone/weaving areas 70.6% (24) Presence of marked bicycle box 67.6% (23) Presence of dedicated turn lanes 67.6% (23) Presence of colored bicycle lane 64.7% (22) Presence of shared-use path 64.7% (22) Presence of bicycle-activated signal detection 64.7% (22) Presence of bicycle signal 61.8% (21) Presence of protected intersection 61.8% (21) Crossing distance 55.9% (19) Presence of bus stops 47.1% (16) Presence of bicycle boulevard 41.2% (14) Other (please specify) 23.5% (8) Total responses 100.0% (34) Other responses: • Lighting, vehicle parking, landscape design, electronic traffic enforcement. • Traffic volume, speeds, median refuge, similar phasing choices to pedestrian section. • Presence of protected/permissive left-turning phases at signal. • Two-stage left turn? Presence of shared right-turn lane (bike lane previous to right-turn lane then goes away near intersection).

Literature Review and Survey of Practice 83   • Lighting, signal head info, and complexity of intersection (number of potential conflicts, number of potential movements, etc.). • All of the above. • No preference. D. Data Collection Efforts, Data Availability, and Willingness to Participate in Research 1. Does your agency have (or has had in the past) any programs to measure pedestrian and/or bicycle volumes or counts? Yes 60.0% (24) No 27.5% (11) Unsure 12.5% (5) Total responses 100.0% (40) 2. If YES or UNSURE to Question D1, for what type(s) of roadway segments and inter- sections has your agency collected pedestrian data? [select all that apply] Urban signalized intersections 71.4% (20) Urban street segments 60.7% (17) Urban unsignalized intersections 50.0% (14) Rural signalized intersections 25.0% (7) Rural unsignalized intersections 17.9% (5) None 14.3% (4) Rural street segments 10.7% (3) Total responses 100.0% (28) 3. If any option in Question D2 is selected, about how many individual sites have pedestrian data been collected for each of these site types? Urban Rural Street Segments Signalized Intersection Unsignalized Intersections Street Segments Signalized Intersection Unsignalized Intersections Los Angeles DOT 100 1,000 100 0 0 0 City of Toronto 15 2,500 0 0 0 0 Seattle DOT 20 200 100 0 0 0 Houston-Galveston Area Council (H-GAC) 100 0 0 0 0 0 Richmond Regional Transportation Planning Organization 50 25 25 0 0 0 Louisiana DOTD 20 20 0 0 0 0 City of Jacksonville, Florida 1 21 8 0 0 0 Louisville, Kentucky 0 20 0 0 0 0 City of Miami 20 0 0 0 0 0 North Carolina DOT Division of Bicycle and Pedestrian Transportation 10 0 0 0 0 0 Roanoke Valley-Alleghany Regional Commission 0 5 5 0 0 0 4. Where are pedestrian volumes counted? Midblock on the sidewalks or shared-use paths 56.5% (13) At intersections for the entire intersection as a whole 47.8% (11) At intersections by approach 52.2% (12) Other (please specify) 26.1% (6) Total number of sites 100.0% (23)

84 Pedestrian and Bicycle Safety Performance Functions Other responses: • All of our pedestrian counts are done through individual projects and are not done on a regular basis. • At bottlenecks. • Too many to list. • We collect pedestrian volumes with our manual turning movement counts. Those counts are completed at a limited number of intersections statewide, and we currently do not have a statewide inventory for these turning movement counts. 5. If any option in Question D4 is selected, on average, what are the durations of these pedestrian counts? [hours/day] 6. If any option in Question D4 is selected, on average, how many years of pedestrian count data are available for each of these site types? Midblock At Intersections OtherFor the Entire Intersection By Approach Seattle DOT 24 N/A 12 N/A Delaware Valley Regional Planning Commission (DVRPC) N/A N/A N/A 24 Houston-Galveston Area Council (H-GAC) 24 N/A N/A N/A North Carolina DOT Division of Bicycle and Pedestrian Transportation 24 N/A N/A N/A Oregon DOT N/A N/A 16 16 City of Toronto 8 N/A 8 N/A Los Angeles DOT N/A 6 6 N/A City of Miami N/A 4 N/A 4 Louisiana DOTD 4 4 N/A N/A South Carolina DOT N/A N/A 4 N/A Richmond Regional Transportation Planning Organization 2 2 2 N/A Louisville, Kentucky N/A 2 N/A N/A Roanoke Valley-Alleghany Regional Commission N/A N/A 2 N/A City of Jacksonville, Florida 2 2 N/A N/A Midblock At Intersections OtherFor the Entire Intersection By Approach City of Toronto 1 0 15 0 Los Angeles DOT 0 10 10 0 Louisville, Kentucky 0 10 0 0 City of Miami 0 4 0 4 Richmond Regional Transportation Planning Organization 3 3 3 0 Louisiana DOTD 3 3 0 0 Roanoke Valley-Alleghany Regional Commission 0 0 3 0 Seattle DOT 2 0 1 0 North Carolina DOT Division of Bicycle and Pedestrian Transportation 2 0 0 0 Oregon DOT 0 0 1 1 City of Jacksonville, Florida 1 1 0 0 Delaware Valley Regional Planning Commission (DVRPC) 0 0 0 1 Houston-Galveston Area Council (H-GAC) 1 0 0 0 New Jersey DOT 0 0 0 0

Literature Review and Survey of Practice 85   7. For what type(s) of roadway segments and intersections have bicycle data been collected? [select all that apply] Urban street segments 25.3% (19) Urban signalized intersections 25.3% (19) Urban unsignalized intersections 20.0% (15) Rural street segments 8.0% (6) Rural signalized intersections 8.0% (6) Rural unsignalized intersections 6.7% (5) None 6.7% (5) Total responses 100.0% (75) 8. If any option in Question D7 is selected, about how many individual sites have bicycle data been collected for each of these site types? Urban Rural Street Segments Signalized Intersection Unsignalized Intersections Street Segments Signalized Intersection Unsignalized Intersections Seattle DOT 100 200 100 20 0 0 Los Angeles DOT 20 100 100 0 0 0 Houston-Galveston Area Council (H-GAC) 100 0 0 50 0 0 Richmond Regional Transportation Planning Organization 45 25 25 0 0 0 Louisville, Kentucky 10 10 10 0 0 0 City of Jacksonville, Florida 1 21 8 0 0 0 City of Miami 20 0 0 0 0 0 Salt Lake City Corp. 4 17 0 0 0 0 North Carolina DOT Division of Bicycle and Pedestrian Transportation 10 0 0 0 0 0 Roanoke Valley-Alleghany Regional Commission 0 5 5 0 0 0 9. Where are bicycle volumes counted? Midblock 60.0% (15) At intersections for the entire intersection as a whole 52.0% (13) At intersections by approach 48.0% (12) Other 28.0% (7) Total responses 100.0% (25) Other responses: • Bikes in the individual turn or straight movements and using each crosswalk. • Bottlenecks. • Along trails (rural street segments in database). • We collect bicycle volumes with our manual turning movement counts. Those counts are completed at a limited number of intersections statewide and we currently do not have a statewide inventory for these turning movement counts. We have also done some limited counts with pneumatic tubes and infrared scanners by Eco-Counter. • In bike lanes, separated paths, on streets without bike lanes.

86 Pedestrian and Bicycle Safety Performance Functions 10. If any option in Question D9 is selected, on average what are the durations of these bicycle counts? [hours/day] 11. If any option in Question D9 is selected, on average, how many years of bicycle count data are available for each of these site types? 12. Does your agency use scaling factors to adjust short-duration counts and/or 24-hour counts to adjust for time of day, day of week, and month to estimate 24-hour volumes? No 58.3% (14) Yes, we have both pedestrian and bicycle scaling factors 33.3% (8) Yes, we have pedestrian scaling factors only 4.2% (1) Yes, we have bicycle scaling factors only 4.2% (1) Total responses 100.0% (24) 13. Does your agency use nontraditional or surrogate measures of volumes, counts, or exposure for pedestrian activities (e.g., vehicular traffic volumes combined with mode splits based on national averages to estimate pedestrian volumes)? No 73.1% (19) Yes 26.9% (7) Total responses 100.0% (26) Midblock At Intersections Other For the Entire Intersection By Approach Louisville, Kentucky 24 24 N/A N/A Seattle DOT 24 N/A 12 24 New Jersey DOT 24 24 24 N/A Delaware Valley Regional Planning Commission (DVRPC) N/A N/A N/A 24 Houston-Galveston Area Council (H-GAC) 24 N/A N/A N/A North Carolina DOT Division of Bicycle and Pedestrian Transportation 24 N/A N/A N/A Oregon DOT 16 16 16 16 Los Angeles DOT N/A 6 6 N/A Roanoke Valley-Alleghany Regional Commission N/A N/A 2 N/A City of Miami N/A 4 N/A 4 Louisiana DOTD 4 4 N/A N/A Richmond Regional Transportation Planning Organization 2 2 2 N/A City of Jacksonville, Florida 2 2 N/A N/A Salt Lake City Corp. 2 2 N/A N/A Midblock At Intersections Other For the Entire Intersection By Approach Los Angeles DOT 0 10 10 0 Louisville, Kentucky 7 10 0 0 Salt Lake City Corp. 5 5 0 0 Oregon DOT 5 1 1 1 City of Miami 4 0 4 Richmond Regional Transportation Planning Organization 3 3 3 0 Roanoke Valley-Alleghany Regional Commission 0 0 3 0 Louisiana DOTD 3 3 0 0 Seattle DOT 2 0 1 2 North Carolina DOT Division of Bicycle and Pedestrian Transportation 2 0 0 0 City of Jacksonville, Florida 1 1 0 0 Delaware Valley Regional Planning Commission (DVRPC) 0 0 0 1 Houston-Galveston Area Council (H-GAC) 1 0 0 0 New Jersey DOT 0 0 0 0

Literature Review and Survey of Practice 87   13.1 If YES to Question D13, please describe the types of surrogate volume data that are collected for pedestrian activities. Response: 1 2 Vehicular traffic volumes Unofficial counts done with the Jamar board Known Estimated mode split based on state or national averages N/A Known Estimated mode split calibrated to individual jurisdictions N/A Sometimes Counts obtained for a subset of the population (e.g., ride tracking app) N/A Known Surrounding land use N/A Known Socioeconomic data N/A Knowable Transit access N/A Known Other (please specify) Total responses (2) 13.2 For what type(s) of roadway segments and intersections have these surrogate measures been estimated for pedestrian activities? Urban street segments 66.7% (2) Urban signalized intersections 66.7% (2) Urban unsignalized intersections 66.7% (2) Rural street segments 33.3% (1) Rural signalized intersections 33.3% (1) Rural unsignalized intersections 33.3% (1) Shared-use path segments 33.3% (1) Total responses 100.0% (3) 14. Does your agency use nontraditional or surrogate measures of volumes, counts, or exposure for bicycle activities (e.g., vehicular traffic volumes combined with mode splits based on national averages to estimate bicycle volumes)? No 88.0% (22) Yes 12.0% (3) Total responses 100.0% (26) 14.1 Please describe the types of surrogate volume data that are collected for bicycle activities. Response: 1 Vehicular traffic volumes Known Estimated mode split based on state or national averages Known Estimated mode split calibrated to individual jurisdictions Sometimes Counts obtained for a subset of the population (e.g., ride tracking app) Known Surrounding land use Known Socioeconomic data Knowable Transit access Sometimes Other (please specify) Total responses (1)

88 Pedestrian and Bicycle Safety Performance Functions 14.2 For what type(s) of roadway segments and intersections have these surrogate measures been estimated for bicycle activities? Urban street segments 66.7% (1) Urban signalized intersections 66.7% (1) Urban unsignalized intersections 66.7% (1) Rural street segments 33.3% (1) Rural signalized intersections 33.3% (1) Rural unsignalized intersections 33.3% (1) Shared-use path segments 33.3% (0) Total responses 100.0% (1) 15. What variables related to pedestrian or bicycle infrastructure elements are included in your agency’s electronic roadway inventory database? [please select all that apply] Presence of sidewalks or shared-use paths on roadway segments 53.8% (14) Presence of dedicated bicycle lanes 46.2% (12) Presence of rectangular rapid flashing beacons (RRFBs) 38.5% (10) Sidewalk width 23.1% (6) Presence of pedestrian signal heads 23.1% (6) Presence of pedestrian push buttons at signals 23.1% (6) Bicycle lane width 23.1% (6) Other (please specify) 23.1% (6) Presence of marked crosswalks at intersections 19.2% (5) Presence of Accessible Pedestrian Signals (APS) 19.2% (5) Presence of buffer (between bicycle lane and motor vehicle lane, between bicycle lane and parking lane, or between travel way and sidewalk) 19.2% (5) Sidewalk condition 19.2% (5) Presence of pedestrian hybrid beacon (PHB) 19.2% (5) None 19.2% (5) Presence of shared-lane markings (sharrows) 15.4% (4) Crosswalk marking style (e.g., high-visibility or traversing lines) 15.4% (4) Crosswalk width 11.5% (3) Presence of bicycle signage 11.5% (3) Presence of advanced crossing signage or markings 11.5% (3) Presence of pedestrian islands 7.7% (2) Presence of bicycle boxes at intersections 7.7% (2) Presence of bicycle corral 3.8% (1) Total responses 100.0% (26) Other responses: • Shoulder of 3 ft or larger. • Wide shoulders, curb ramps. • ADA Amenities. • I am unsure of what, if anything, is in the database. • Lane widths, lane types, roadway slope, crosswalk condition, paving condition. • We collect ped infrastructure data at the beginning of an improvement project. Typically, this is done using Photolog or Aerial Imagery. We do not have ped/bike infrastructure data statewide. 16. Does your agency maintain an electronic crash database in which crashes involv- ing pedestrians and/or bicyclists can be easily identified? If so, are crash locations identified using a constant referencing system that can be linked to your electronic roadway inventory? Yes 74.1% (20) No 18.5% (5) Unsure 7.4% (2) Total responses 100.0% (27)

Literature Review and Survey of Practice 89   17. Does your agency have access to any nontraditional data sources that might indicate frequency of crashes involving pedestrians and/or bicycles (e.g., police reports or hospital records)? Yes (if so, please indicate what type of information is available) 42.9% (12) No 39.3% (11) Unsure 17.9% (5) Total responses 100.0% (27) Data Sources: • Trauma center data. • Crash reports. • Vital statistics and police reports are utilized. • Mentioned Portland State Crowdsourcing app that collected some nontraditional data. • Police reports are available. • We use SPD collision reports for the majority of our safety data. This includes a bevy of information that helps us establish movements, locations, contributing factors, injury outcomes, etc. • Police reports. • Police reports and hospital records. • Police reports and we hope to start working with regional hospital system to acquire hos- pital data. The following key insights were obtained from the survey of state and local agencies: • Although state agencies tended to maintain well-developed databases of crash infor- mation, they were not as likely to contain databases of pedestrian/bicycle volumes or infrastructure. One reason is that state agencies may not own or be responsible for main- taining this infrastructure (e.g., sidewalks or walking/shared-use paths). Local (either city or regional) agencies were more likely to have information on pedestrian/bicycle volumes and infrastructure. • Pedestrian and bicycle safety performance is primarily quantified using observed crash fre- quencies at individual locations or community complaints, and this information is not likely to be compared to predicted crash frequencies based on local characteristics. A lack of volume data makes obtaining crash rates difficult for many agencies. Inaccurate or incomplete crash reports, the time required to analyze crash reports, and the infrequency of pedestrian and bicycle crashes (other than fatal or major injury crashes) make quantifying pedestrian and bicycle safety performance more difficult. • Although some automated count measures are in use (particularly infrared scanners, pneu- matic tubes, or video-based detection), manual counts are the most common sources of pedes- trian and bicycle volume data. The lack of staff or volunteer time to collect data, funding for automated counting equipment, and lack of management commitment to pedestrian and bicycle safety appear to be barriers to accurately obtaining pedestrian and bicycle volume data. Coordination between agencies within a region can relieve some of this burden. • Aerial imagery (from publicly available sites such as Google or Bing) appears to be the primary source of pedestrian and bicycle infrastructure data; however, recent infrastructure changes, such as the addition of a sidewalk or bike lane, may not be reflected in these aerial images. Some agencies are updating roadway inventory databases to include pedestrian and bicycle infrastructure or creating new databases. • Pedestrian and bicycle SPFs are primarily desired for urban and suburban roadway segments and intersections, as opposed to rural roadway segments and intersections and ramps. SPFs for roundabouts are not as high of a priority as those for traditional intersection forms in urban and suburban areas.

90 Pedestrian and Bicycle Safety Performance Functions • Almost all proposed CMFs were strongly desired for both pedestrian and bicycle crashes at both intersections and roadway segments. The top responses (> 64% approval) were: – Pedestrian crashes at roadway segments: ◾ Presence of crosswalks, RRFBs, sidewalks/shared-use paths, pedestrian refuge island, pedestrian hybrid beacon, and raised median, as well as adjacent land-use types, drive- way density, and presence of bus stops. – Pedestrian crashes at intersections: ◾ Crossing distance and presence of marked crosswalks, pedestrian signal, pedestrian ref- ugee island, lighting, leading or lagging pedestrian interval, channelized right-turn lane, raised pedestrian crossing, RRFBs, right-turn-on-red restrictions, PPLT phases, and bus stops. – Bicycle crashes at roadway segments: ◾ Presence of dedicated bicycle lane, protected or separated bicycle lane, dedicated turn lanes, and shared-lane marking and/or signs, as well as adjacent land-use types and driveway density. – Bicycle crashes at intersections: ◾ Presence of dedicated bicycle lane, green striping in conflict zone/weaving areas, marked bicycle box, dedicated turn lanes, colored bicycle lane, shared-use paths, and bicycle- activated signal detection. 2.2.2 Interviews with Selected Agencies Based on information gathered through the literature review, an internet search, and the online survey, the research team contacted several agencies to gather more information about their pedestrian and bicycle count programs, inventory datasets, and crash datasets. The agencies contacted included: • City of Minneapolis, Minnesota/Minnesota DOT, • City of St. Paul, Minnesota, • City of Seattle, Washington, • City of Bellevue, Washington, • Delaware Valley Regional Planning Commission (DVRPC), Philadelphia, Pennsylvania, and • Oregon DOT. Details about the respective agencies’ pedestrian and bicycle count programs, inventory data- sets, and crash datasets are summarized below. 2.2.2.1 Location/Agency: City of Minneapolis, Minnesota, and Minnesota DOT Since 2007, the Minneapolis Public Works Department has conducted annual bicyclist and pedestrian traffic counts at locations throughout the city. Each year, the Public Works Depart- ment collects 2-hour counts at select locations across Minneapolis. Count observations are con- ducted by trained volunteers using a standardized form and methodology. Counts are conducted on three weekdays in September from 4:00 to 6:00 p.m., and a simple extrapolation factor is used to create estimated daily traffic (EDT) over 24 hours. Automated counts are also used to validate the model that extrapolates EDT from 2-hour counts and to understand bicycle and pedestrian patterns over longer periods of time. The Public Works Department has conducted counts at approximately 575 locations in Minneapolis. Thirty sites were chosen in 2007 and designated as annual or benchmark sites. These sites are counted annually to provide baseline data on bicycling and walking volumes in Minneapolis. The benchmark sites are located at natural pinch points for pedestrians and bicy- clists (such as bridges and trails) and on streets with bicycle improvements anticipated in the

Literature Review and Survey of Practice 91   coming years. Roughly 380 sites are regular count locations, which the Public Works Department counts in a 3- to 4-year cycle. The sites are spaced across the city, usually as pairs at intersections and with multiple sites along busy corridors. The remaining sites are special count locations and have only been counted once or twice. These locations are counted to measure data before and after pedestrian or bicycle infrastructure improvements or for project-specific purposes. In any given year, counts are conducted at all benchmark sites, 25 to 33 percent of regular sites, and special sites based on additional data needs. A guide documenting the Public Works’ nonmotor- ized count program and methodology, including the full methodology on EDT extrapolation, is available online at www.minneapolismn.gov/pedestrian/data/pedcounts (Minneapolis Public Works Department 2013). The Public Works Department has an inventory database that includes the type of bicycle facility present on the roadway. Most of the locations within the city have a 30 mph posted speed limit. Limited inventory information is available for pedestrian facilities, such as sidewalks. Crash data can be obtained from the Public Works Department. 2.2.2.2 St. Location/Agency: City of Paul, Minnesota St. Paul uses a counting methodology identical to that of Minneapolis, which was established by Transit for Livable Communities (TLC), a local nonprofit organization that conducts nonmo- torized counts. TLC is the local manager of the federally funded Nonmotorized Transportation Pilot Program (NTP) grant and conducts counts as part of its ongoing evaluation of NTP projects in Minneapolis. While TLC’s count efforts primarily focus on NTP evaluation, data are shared with Minneapolis Public Works, as the counting methods are similar. The count methodology uses volunteers to manually count pedestrians and bicyclists between 4 and 6 p.m. on a weekday (Tuesday, Wednesday, or Thursday) during the second week of September. The week may shift if inclement weather prevents counts during that time. The count sites are gen- erally 40–60 ft back from an intersection and use a “screen line” approach that counts pedestrians and bicyclists that cross the screen line on both sides of the road in both directions of travel. In most cases, counts for each direction of travel are not recorded separately but are recorded as one count for that location. Counts are generally not conducted at intersections unless needed for a specific project. Two or three intersection counts may be conducted each year. Some sites were counted as early as 2007, but the counting program increased substantially in 2013. Since 2013, approximately 160–170 sites have been counted. About 30 sites serve as benchmarks for bicycle counts, and 25 sites serve as benchmarks for pedestrians and are counted each year. The remaining sites shift from year to year based on project needs and the location of available volunteers. In 2015, three permanent count stations were installed. Two of the stations are along the same roadway but in opposite directions of travel. A median made two separate counters necessary. The first full year of data for these continuous sites was 2016. In addition, there are 48-hour video counts conducted at about 10 to 20 sites per year. These counts are gener- ally conducted to support specific projects, such as bike lane installation. St. Paul does not use factors to scale 2-hour counts to an annual average count. It does estimate a 24-hour September weekday count, based on the permanent counters and the 48-hour video counts. Counts are mapped as point locations in a shapefile. They are not extrapolated to roadway segments. St. Paul staff may do this extrapolation anecdotally as needed. St. Paul also maintains a bike facility shapefile that shows the bike facilities present along its network as well as the installation year for the facility. This file is updated annually. Maps are available on a website. A walking facility map is also produced, but it may be updated less frequently.

92 Pedestrian and Bicycle Safety Performance Functions Vehicle annual average daily traffic (AADT) is recorded but may be in computer-aided design (CAD) files rather than shapefiles. Overlaying pedestrian and bicycle counts on facility maps is straightforward. Including AADT information on the same maps requires additional conversion. The St. Paul Police Department maintains a database of pedestrian and bicycle crashes that is more current than state databases and may pick up incidents that the state database misses. The data are available on their website. 2.2.2.3 Location/Agency: City of Seattle, Washington SDOT has an extensive count program for pedestrian and bicycle volumes. The data include both permanent and temporary counts. Permanent counts are performed at 10 locations. This information is obtained using infrared scanners. The permanent count locations are typically bike trails or bridges with large volumes of bicyclists. Additionally, two on-street locations were installed with continuous counters; however, only 30 days of data are available from these loca- tions. Data from the remaining permanent count stations are available for the last year. Semiper- manent counts are also conducted using pneumatic tubes at approximately 100 locations. These counts are collected for 24 hours for multiple days, typically lasting a week. Reliable semiperma- nent data are available since 2014. Temporary pedestrian and bicycle counts are also conducted as part of traffic studies. These data are typically collected using video cameras at a peak hour or sometimes for 24 hours. These are typically obtained from individual projects that require this information, and the data are then stored in PDF files for individual studies. Due to how the counts are obtained (on a project basis), there is little regularity to the count data. However, a list of traffic data studies from 2014 to the present is available. Recently, SDOT (2016) conducted a study to assess the safety of bicycles and pedestrians in Seattle. SDOT found that most bicycle and pedestrian crashes happen at intersections along arterial streets. The most common crash type is a left hook, followed by an angle, and then a right hook, whereas the most common crash type for pedestrians is when pedestrians are crossing or left hook. A comparison to exposure was also conducted. Strava data were used for bicycle counts, while a trip generation model was used as a proxy for pedestrian exposure. The details are not provided in the public document, and the Strava data cannot be shared due to privacy issues. Infrastructure data are stored and available from SDOT’s website (https://data.seattle.gov/). This dataset includes all information about the road network, including bike lanes; GIS data for the collisions layer can be used for safety analyses. Crash data are also publicly available and can be found online. 2.2.2.4 Location/Agency: City of Bellevue, Washington The City of Bellevue is in the process of implementing a video camera data collection effort throughout the city. The data will include motor vehicle, pedestrian, and bicyclist counts and will be automatically counted from the video recordings. The motor vehicle counts have already been established and can count with 95 percent accuracy; however, the pedestrian and bicyclist count- ing algorithms still require improvement. The city anticipated that the pedestrian and bicyclist data would be available by the end of 2018 at 95 percent accuracy. Currently, five intersections are equipped with cameras, and they anticipate the number to increase to 100 by the end of 2019. The city is also working to automatically collect surrogate measures of safety, such as near collisions. There is also limited Eco-Counter-level data available for a couple of locations, but the data do not have the same level of detail that the video cameras will be able to provide.

Literature Review and Survey of Practice 93   2.2.2.5 Location/Agency: Delaware Valley Regional Planning Commission (DVRPC), Philadelphia, Pennsylvania The DVRPC is the federally designated Metropolitan Planning Organization (MPO) for a nine- county region in two states: Bucks, Chester, Delaware, Montgomery, and Philadelphia counties in Pennsylvania; and Burlington, Camden, Gloucester, and Mercer counties in New Jersey. The DVRPC has an extensive count program for pedestrian and bicycle volumes, and all count infor- mation is publicly available through their website (https://www.dvrpc.org/traffic/). This website contains all count data, locations, and time periods. The DVRPC data include both permanent and temporary counts. Permanent counts are performed at 10 locations along mixed-use (walking and bicycling) trails. This information is obtained primarily from pneumatic tubes installed along the trails. Historical data exist for at least the past 2 years for eight of the 10 sites, while at least 1 year of data are available for the remaining two sites. For each of these locations, pedestrian and bicycle volumes are available at 15-minute intervals, and directional data are available. Temporary pedestrian and bicycle counts are provided across the Philadelphia region and are also available on the DVRPC website. These are typically obtained from individual projects that require this information, and the data are then stored on the website. Data are available for over 100 individual sites. A quick review of the available data suggests that at each location, counts are typically performed for about a 5- to 7-day period to obtain an average daily pedestrian or bicycle volume. For each of these, a detailed report is available that provides hourly counts for each day within the count period as well as the range of observed temperatures and weather. The detailed report also includes two factors—pedestrian/bicycle seasonal factor and equipment factor—to scale the counts. These seasonal factors were obtained by looking at counts on similar days of the week/months of the year within previous years. Thus, the average daily volumes are already scaled to account for seasonable variation in the count data. New daily seasonal scaling factors have been calculated using the permanent counting station information, but these have yet to be applied to the data available on the website. Due to how the counts are obtained (on a project basis), there is little regularity to the count data. Some locations have multiple counts across a long period (5+ years), while some are just a single instance. These temporary counts are typically performed at midblock locations and do not include detailed turning movement data. A picture of each count location is generally provided. Infrared scanners and video cameras are used for pedestrian counts; pneumatic tubes, infrared scanners, and video cameras are used for bicycle counts. The DVRPC does not maintain a pedestrian and bicycle infrastructure database, but the DVRPC representative indicated that the City of Philadelphia might have this information. Crash data are available from the DVRPC, but since the DVRPC spans two states—Pennsylvania and New Jersey—the data are not in a consistent format. Roadway inventory databases are also inconsistent in format. 2.2.2.6 Location/Agency: Oregon DOT Oregon DOT does not have permanent bicycle or pedestrian counting programs. It does have some temporary counts available. Temporary counts are done on an as-needed basis for specific projects and are typically conducted using cameras or manual counts. The counts are typically done for 2 to 16 hours; however, even if the data are collected for 16 hours using video cameras, only 2 to 4 hours of the videos are postprocessed. The counts are available in a database that dates back to 2009. Due to how the counts are obtained (on a project basis), there is little consistency in the count data. These temporary counts are typically done at intersections and sometimes include turning movements.

94 Pedestrian and Bicycle Safety Performance Functions Oregon DOT conducted a study, Risk Factors for Pedestrian and Bicycle Crashes (Monsere et al. 2017). Monsere et al. used data from Strava for bicycle counts and used total population density as a proxy for pedestrian counts. Logistic regression models to predict crash occurrence and crash severity were developed as a function of variables related to roadway characteristics and vehicle volumes. A pedestrian and bicycle infrastructure database is not maintained at the state level. Each region maintains its own databases. Crash data summaries could also be obtained and are avail- able online. 2.3 Summary of Key Issues 2.3.1 Priorities for Pedestrian and Bicycle SPF Development A majority of pedestrian and bicyclist fatalities occur in urban locations, and most respon- dents to the survey ranked urban roadway segments and intersections as the highest priority for developing pedestrian and bicycle SPFs over roadway segments and intersections in rural areas and crossroad ramp terminals. Thus, based on crash statistics and survey responses, priority should be given to developing pedestrian and bicycle SPFs for urban roadway segments and intersections. Additionally, pedestrian and bicycle count data are primarily available in urban areas. Very little count data are available for rural areas. 2.3.2 Summary of Barriers to Collecting Pedestrian and Bicycle Safety Performance Data Various methods and technologies are used to collect pedestrian and bicyclist exposure data. Examples include manual on-site counts, manual video counts, infrared sensors, computer pro- cessing of video, inductive loops, pneumatic tubes, and piezometric pads. The primary barrier to collecting pedestrian and bicycle performance data is cost. Cost is affected by the selected counting method, count duration, location of counts, and permitting, and is even more of a bar- rier due to funding issues associated with these types of data collection. Additionally, each of the factors affecting cost presents its own barriers and limitations that make collecting pedestrian and bicycle data difficult. Standardization and increased guidelines from transportation agencies would help practitioners avoid some of these barriers and ease the process of conducting counts. Despite all of these barriers to conducting pedestrian and bicycle counts, many agencies and practitioners have developed counting practices that help decrease costs and have established programs in their jurisdictions. 2.3.3 Summary of Practices for Overcoming Barriers to Collecting Pedestrian and Bicycle Count Data Most transportation agencies in the United States do not have mature pedestrian and bicycle count programs. Cost is often the biggest barrier to conducting pedestrian and bicycle counts. One of the most prevalent ways that agencies try to save money on counts is by outsourcing labor and, in several cases, using volunteers to conduct counts. Public and political support for funding is a vital part of successful pedestrian and bicycle counting programs. To garner support from government and communities, some transportation agencies have developed count programs around the concept of active transport and track usage data to show its importance and to justify the need for ongoing collection. To make the counting process easier and more efficient, sev- eral transportation agencies have integrated pedestrian and bicycle counts into existing motor vehicle count programs. Another way transportation agencies have eased the counting process

Literature Review and Survey of Practice 95   is by developing standard training materials and count forms for volunteers. Standard reporting sheets are also a good practice to aid the development of a common way to store data. Quality assurance practices are not frequently used in conducting pedestrian and bicycle counts due to added cost. At a minimum, visual inspection of the data should be conducted to identify poten- tial errors. 2.3.4 Summary of Processes to Link Pedestrian and Bicycle Data Sources Most pedestrian and bicycle count data are collected over short times around peak daytime travel periods. Short-term counts can be combined with long-term counts at similar locations to develop estimates of weekly, monthly, or annual exposure. The process is based on the assump- tion that similar sites have similar patterns in hourly, weekly, and monthly volume trends. Even if the magnitude of volumes is different, the proportional trends can be similar. Guidance is avail- able for developing expansion factors using long-duration counts, but general expansion factors are also available from the National Bicycle and Pedestrian Documentation Project. Trip data have traditionally been available from self-reported survey sources like the NHTS or regional surveys. More recently, trip data have become available from smartphones. While the active smartphone data are biased in favor of recreational trips, both passive and active trip data can be fused with count data to enhance volume estimates. Linking hospital and crash data has several benefits for safety analysis, including connecting crash conditions to injury outcomes and providing insights on the accuracy of crash severity documented on police crash reports. In most cases where efforts were made to link police crash reports and hospital records, linkages were made based on age, gender, date, time, and location. In addition, in the United States, 15 states maintain an active CODES program, which uses a probabilistic methodology to link crash records to injury outcome records collected at the scene and en route by emergency medical services, by hospital personnel after arrival at the emergency depart- ment or admission as an inpatient, and/or the time of death on the death certificate. The CODES program addresses all motor vehicle crash types, including pedestrian and bicycle crashes. 2.3.5 Summary of Methods for Estimating Pedestrian and Bicycle Exposure Several methods for estimating pedestrian and bicycle exposure at intersections and along roadway segments include (1) trip generation and flow models, (2) network simulation models, and (3) direct demand models. Direct demand models are most commonly used because they are simple to understand, do not require complex computer applications to execute, and are straight- forward to apply. Explanatory variables that have been most commonly statistically significant in these models include surrounding population density, surrounding jobs or employment density, and proximity to transit. 2.3.6 Summary of Methods for Estimating Pedestrian and Bicycle Safety Performance Multiple studies used negative binomial regression models to develop SPFs to predict pedes- trian and bicycle crashes. Some models were developed that did not include pedestrian and bicycle volumes, while other models included pedestrian and bicycle volumes as predictors. Geometric, signal timing/phasing, land-use, demographic, and socioeconomic characteristics have also been included in models as predictors. Most SPFs were developed to predict crashes at intersections,

96 Pedestrian and Bicycle Safety Performance Functions but some models were developed for roadway segments. In some instances, SPFs were devel- oped for specific crash types such as pedestrians and bicyclists crossing and motorists traveling straight through the intersection or pedestrians and bicyclists crossing and motorists making a left turn. All crash prediction models were developed to predict crashes in urban areas. None were developed to predict pedestrian or bicycle crashes in rural areas. In addition, all models were developed specifically for a given mode. SPFs were developed to predict either pedestrian crashes or bicycle crashes. No SPFs were developed to predict pedestrian and bicycle crashes that combined them in the same model. Several other types of models have been developed for estimating pedestrian and bicycle safety performance. Rather than outputting an estimate of predicted crashes, the models result in a safety score or an index reflective of the safety performance. These other types of models have been developed with varying levels of complexity and scientific rigor. 2.3.7 Summary of Factors Contributing to Pedestrian and Bicycle Crashes Table 29 provides a list of factors identified in the literature that potentially contribute to pedestrian and bicycle crashes. The factors are categorized into three groups: roadway geometric and operational characteristics, land use, and demographics. The table indicates whether the factor is expected to contribute to pedestrian crashes, bicycle crashes, or both and whether it is expected to contribute to crashes along a roadway or at an intersection. In reviewing the lit- erature, it was not always evident if the factor is expected to contribute to pedestrian or bicycle crashes or if it is expected to contribute to crashes along a roadway or at an intersection, so some assumptions were made in developing this summary table. The contributing factors listed in Table 29 were considered as potential elements for incorporation into the pedestrian and bicycle SPFs developed as part of this research. 2.3.8 Summary of Pedestrian and Bicycle Countermeasures The first edition of the HSM does not include any countermeasures proven to directly reduce (or increase) pedestrian and bicycle crashes. Thus, the research team searched for other sources of reliable pedestrian and bicycle countermeasures that could potentially be used in conjunc- tion with the pedestrian and bicycle SPFs developed as part of this research. Sources that were considered included recently completed research; three-, four-, and five-star CMFs included in the CMF Clearinghouse; countermeasures included in PEDSAFE and BIKESAFE; and adjust- ment factors included in usRAP and iRAP. Table 30 provides a summary of pedestrian-related countermeasures in the CMF Clearinghouse (three stars or above) PEDSAFE and usRAP and iRAP for potential use with the pedestrian and bicycle SPFs developed as part of this research. Similarly, Table 31 provides a corresponding summary of bicycle-related countermeasures.

Literature Review and Survey of Practice 97   Category Contributing Factor Pedestrian Bicyclist Roadway Segment Intersection Roadway geometric and operational characteristics Number of lanes Lane width Shoulder width Presence of bicycle lanes Presence of on-street parking Presence of median Vertical grade/slope Horizontal curve Quality of curve Presence of shared-use paths Sidewalk width Presence of bus stops Number of driveways Presence of midblock crossing Presence of lighting Delineation Shoulder rumble strips Sight distance Road condition Posted/operating speed Speed management/traffic calming Distance to closest marked crosswalk or intersection Distance to nearest traffic signal Functional class One-way versus two-way traffic Number of through lanes being crossed Number of left-turn lanes Number of right-turn lanes Presence of marked crosswalk Type of traffic control Pedestrian crossing quality Pedestrian fencing Skid resistance/grip Presence of pedestrian-activated flashers or beacons AADT Pedestrian volume Bicycle volume Number of left- and right-turning vehicles Percent heavy vehicles Pedestrian delay Land use Land use (commercial, industrial, institutional, residential) Block size Presence of school zone Presence of alcohol sales establishments Urban/rural Demographics Age Population density Household size Mean household income Ethnicity Single-family residential Vehicle numbers in housing units Table 29. Summary of factors contributing to pedestrian and bicycle crashes.

98 Pedestrian and Bicycle Safety Performance Functions Category Treatment/Countermeasure Along the roadway • Sidewalks, walkways, and paved shoulders • Pedestrian fencing • Regulated roadside commercial activity • Shoulder sealing • Restrict/combine direct access points At crossing locations • Curb ramps • Marked crosswalks and enhancements • Curb extensions • Crossing islands • Raised pedestrian crossing • Lighting and illumination • Parking restrictions • Pedestrian overpass/underpass • Automated pedestrian detection • Leading pedestrian interval • Advance yield/stop lines • Central hatching • School zones Transit • Transit stop improvements • Access to transit • Bus bulb-outs Roadway design • Bicycle lanes • Lane narrowing • Lane reduction (road diet) • Driveway improvements • Raised medians • One-way/two-way street conversions • Improved right-turn slip-lane design • Skid resistance • Speed management Intersection design • Modified T-intersections • Intersection median barriers • Curb radius reduction • Modified skewed intersections Signals and signs • Traffic signals • Pedestrian signals • Pedestrian signal timing • Traffic signal enhancements • Right-turn-on-red restrictions • Advanced stop lines at traffic signals • Left-turn phasing • Push buttons and signal timing • Pedestrian hybrid beacon (PHB) • Rectangular rapid flashing beacon (RRFB) • Pedestrian countdown timer Table 30. Summary of pedestrian-related countermeasures in CMF Clearinghouse PEDSAFE and usRAP/iRAP.

Literature Review and Survey of Practice 99   Category Treatment/Countermeasure Shared roadway • Roadway surface improvements • Lighting improvements • Parking treatments • Median/crossing island • Driveway improvements • Lane reductions (road diet) • Lane narrowing • Street track treatments • Pedestrian fencing • Regulate roadside commercial activity • Shoulder sealing • Speed management • Road surface rehabilitation • Restrict/combine direct access points On-road bike facilities • Bike lanes • Wide curb lanes • Paved shoulders • Shared bus-bike lanes • Contraflow bike lanes • Separated bike lanes Intersection treatments • Curb radius reduction • Intersection markings • Turn restrictions • Crosshatching • Sight distance Markings, signs, and signals • Optimizing signal timing for bicyclists • Bike-activated signal detection • Sign improvements for bicyclists • Pavement marking improvements • School zone improvements • Rectangular rapid flashing beacons (RRFBs) • Pedestrian hybrid beacons (PHBs) • Bicycle signal heads Table 31. Summary of bicycle-related countermeasures in CMF Clearinghouse BIKESAFE and usRAP/iRAP.

Next: Section 3 - Development of Pedestrian and Bicycle Models Incorporating Available Pedestrian and Bicyclist Exposure Data »
Pedestrian and Bicycle Safety Performance Functions Get This Book
×
 Pedestrian and Bicycle Safety Performance Functions
Buy Paperback | $121.00
MyNAP members save 10% online.
Login or Register to save!
Download Free PDF

Each year, national crash studies have estimated that while overall traffic fatalities are decreasing, the percentages of those fatalities among pedestrians and cyclists are increasing.

NCHRP Research Report 1064: Pedestrian and Bicycle Safety Performance Functions, from TRB's National Cooperative Highway Research Program, presents state departments of transportation and other transportation professionals with an update of pedestrian and bicycle safety performance functions (SPFs).

Supplemental to the report are three spreadsheet tools that address SPFs on rural multilane roads, rural two-lane roads, and urban/suburban arterials.

READ FREE ONLINE

  1. ×

    Welcome to OpenBook!

    You're looking at OpenBook, NAP.edu's online reading room since 1999. Based on feedback from you, our users, we've made some improvements that make it easier than ever to read thousands of publications on our website.

    Do you want to take a quick tour of the OpenBook's features?

    No Thanks Take a Tour »
  2. ×

    Show this book's table of contents, where you can jump to any chapter by name.

    « Back Next »
  3. ×

    ...or use these buttons to go back to the previous chapter or skip to the next one.

    « Back Next »
  4. ×

    Jump up to the previous page or down to the next one. Also, you can type in a page number and press Enter to go directly to that page in the book.

    « Back Next »
  5. ×

    To search the entire text of this book, type in your search term here and press Enter.

    « Back Next »
  6. ×

    Share a link to this book page on your preferred social network or via email.

    « Back Next »
  7. ×

    View our suggested citation for this chapter.

    « Back Next »
  8. ×

    Ready to take your reading offline? Click here to buy this book in print or download it as a free PDF, if available.

    « Back Next »
Stay Connected!