Safety Prediction Models for Six-Lane and One-Way Urban and Suburban Arterials (2022)

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.

88 CHAPTER 5. PREDICTIVE MODELS FOR URBAN AND SUBURBAN ROADWAY SEGMENTS WITH SIX OR MORE LANES This chapter describes the activities undertaken to calibrate safety predictive models for urban and suburban roadway segments with six or more lanes. Each model consists of an SPF and a family of CMFs. The SPF is derived to estimate the crash frequency with specified design elements and operating conditions. The CMFs are used to adjust the SPF estimate whenever one or more elements or conditions deviate from those that are specified. The calibrated safety predictive models were used to develop the two-way arterial roadway segment safety predictive method. This method describes how to use the models to evaluate the safety of two-way six-or-more-lane arterials, as may be influenced by road geometry, roadside features, and traffic volume. Collectively, the predictive models for roadway segments in this chapter address the following facilities. • Divided two-way arterials with six lanes (including a raised or depressed median) (6D). • Undivided two-way arterials with six lanes (6U). • Two-way arterials with six lanes and a TWLTL in the middle (7T). • Divided two-way arterials with eight lanes (including a raised or depressed median) (8D). This chapter is divided into four sections. The first section provides a brief background related to segmentation, database development, and modeling approach. The second section summarizes the details of calibration data. The third section describes the calibration of the models to predict FI, PDO, vehicle-pedestrian, and vehicle-bicycle crash frequency. The fourth section provides a list of CMFs. BACKGROUND The road segment boundaries are typically defined by intersections or by a change in the cross- section. Each segment is homogenous with respect to characteristics such as traffic volumes and key roadway design characteristics and traffic control features. Figure 14 shows the segment length, L, for a single homogenous roadway segment occurring between two intersections. However, several homogenous roadway segments can occur between two intersections.

89 Figure 14. Definition of roadway segments and intersections (AASHTO, 2010). A new (unique) homogenous segment begins at the center of each intersection and where there is a change in at least one of the following characteristics of the roadway: • AADT (vehicles per day). • Number of through lanes. • Presence/type of median (undivided, divided by raised or depressed median, center TWLTL). • Lane width. • Outside shoulder width. • Median width. • Speed category. A cross-sectional (as opposed to panel) database was created for developing the regression models. The database includes a five-year study period for all observations. Study duration in years is represented as an offset variable in the regression model. One reason for using cross- sectional data for model calibration relates to the accuracy of the AADT values in most highway safety databases. Segment AADT is frequently extrapolated by a state DOT from partial-year counts taken at temporary count stations located several miles from the subject segment. Thus, there are accuracy implications associated with this temporal and spatial extrapolation (Bonneson et al., 2012). Moreover, state DOT practice when a current count is not available for a segment is sometimes used to adjust the AADT from the last year it was counted (which could be several years previous); other times, the practice is to leave the variable as missing. Thus, averaging each segment’s AADT over years minimizes the variability in AADT, which, based on the aforementioned observations, is considered largely random. More generally, cross- sectional data provide a more robust predictive model than panel data when the year-to-year variability in the independent variables is largely random. A second reason for using cross-sectional data for model calibration is to minimize the problems associated with overrepresentation of segments or intersections with zero crashes. Statistical methods have been developed to improve the fit of a model to these zero-inflated or excess zero data. However, Lord et al. (2005) and Geedipally et al. (2012) indicate that when these methods

90 have been applied to highway crash data, they have (a) an inherent tendency to over-fit the data, (b) a theoretic explanation of dual state highway safety that is problematic with one of the two states that has a long-term mean equal to zero (i.e., the mean of the Poisson distribution is always equal to zero), and (c) the potential to obfuscate the interpretation of predictive model trend and coefficient meaning. Thus, summing each segment’s crashes over years minimizes the proportion of segments or intersections with zero crashes in the database and precludes the need for a dual state distribution. Separate models were developed for FI and PDO crashes. Experience with regression-based calibration of SPFs and CMFs using total crashes and using only FI crashes indicates that the calibration coefficients often vary among model types for common variables. Some of this variation is likely due to the fact that geometric elements often have a different effect on FI crashes than on PDO crashes. As a result, the search for correlation and possible causation is challenged when using total crash data to build total crash prediction models because total crashes combine FI and PDO crashes. It is widely recognized that PDO crash counts vary widely on a regional basis due to significant variation in the reporting threshold. When crash frequency varies systematically from county to county, district to district, and state to state because of formal and informal differences in the reporting threshold, the use of PDO crash data to build PDO crash prediction models is problematic. This observation suggests that PDO-based and total crash models are likely to include regional biases and added uncertainty due to variation in reporting thresholds. Based on these issues, the following model-building process was developed. Researchers rationalized that (a) FI crash data are likely to provide the most accurate insight into regression model structure and factors influencing safety, and (b) PDO-based models are preferred to total crash models. However, because of under-reporting, the development of PDO regression models is problematic and the variable effect provided by these models may not be accurate. Therefore, the FI regression model structure was developed first and then used as a starting point for the development of the PDO regression model. By doing this, the research team could estimate the PDO crashes at the same base conditions. Some geometric variables that were significant in the FI model were less significant in the PDO model. Specifically, the standard error was increased for those geometric variables that varied more among counties than within counties. Unfortunately, it is not known whether the among-county variation is due to differences in reporting threshold (as may be informally applied at different levels within a state) or because of differences in geometry. This approach often resulted in the PDO model having fewer geometric variables than the FI model. Since FI models were more accurate than PDO models, the CMFs were developed from FI crash data only but were used for both FI and PDO crashes. CALIBRATION DATA The database assembly for these types of facilities focused on Texas and two HSIS states: California and Illinois. Although a general description of all the data collected was provided in the previous chapter, the descriptions provided here and the subsequent chapters are tailored specifically for the models documented in the corresponding chapter and include only the variables that were considered in the final models. The data are summarized in Table 47 and Table 48 for electronic and supplemental variables, respectively.

91 Table 47. Variables acquired from state databases for six-or-more-lane arterials. Data Variable Description AADT Two-way annual average daily traffic volume (veh/day) during the study period Segment length Length of the homogenous segment (mi) in the state database Lane width Average width (ft) of the through lanes Inside shoulder width (divided segments) Average width of the inside shoulder (ft) in the two directions of travel Outside shoulder width Average width of the outside shoulder (ft) in the two directions of travel Median width Average median width (ft) along the segment

92 Table 48. Supplemental data collected for six-or-more-lane arterials. Data Variable Description Bus or HOV lane presence Presence of bus-only or HOV lanes. Bicycle lane presence Presence of bicycles lanes. Sidewalks Presence of sidewalks along each side of the roadway segment: 0: No sidewalk. 1: Sidewalk on one side of the roadway. 2: Sidewalks on both sides of the roadway. Lighting Presence of lighting along each side of the roadway segment: 0: No lighting. 1: Lighting on one side of the roadway. 2: Lighting on both sides of the roadway. Parallel parking proportion Proportion of the length of segment with parallel parking (considered in both directions of travel for two-way streets). Angle parking proportion Proportion of the length of segment with angle parking (considered in both directions of travel for two-way streets). Speed limit Posted speed limit (mph) as observed from speed limit signs. Median barrier Presence of concrete barriers in the median. Railroad crossings Number of railroad-highway crossings within the limits of the roadway segment. Driveway density Density of driveways along the length of the segment (driveways/mile), classified consistently with the HSM Chapter 12 driveway categories: Major commercial driveways. Minor commercial driveways. Major industrial/institutional driveways. Minor industrial/institutional driveways. Major residential driveways. Minor residential driveways. Other driveways. Roadside fixed-object density Density of fixed roadside objects (objects/mile) within 30 ft of the edge of traveled way (in both directions of travel for two-way streets). In absence of marked edge lines, edge of traveled way was considered to be 2.0 ft from the face of the curb. Fixed objects were counted using the same method as required for application of the HSM CMF for roadside fixed objects (described on pages 12–41 of the HSM). Roadside fixed-object offset Average distance from the edge of traveled way to the roadside fixed objects (as defined above). Inside curb proportion Ratio of the two-way total length (ft) of curb present along the inside (median side) of the segment to twice the length of the segment. Outside curb proportion Ratio of the two-way total length (ft) of curb present along the outside (right shoulder side) of the segment to twice the length of the segment. MODEL DEVELOPMENT—SIX-OR-MORE-LANE ARTERIALS The regression model form that was used to predict the average crash frequency on an individual roadway segment is as follows: 𝑁 = (𝑁 𝐼 + 𝑁 𝐼 ) × 𝐶𝑀𝐹 × 𝐶𝑀𝐹 × 𝐶𝑀𝐹 × 𝐶𝑀𝐹 (141) with,

93 𝑁 = 𝑁 × 𝐶𝑀𝐹 _ × 𝐶𝑀𝐹 _ × 𝐶𝑀𝐹 _ × 𝐶𝑀𝐹 , (142) 𝑁 = 𝑁 × 𝐶𝑀𝐹 × 𝐶𝑀𝐹 , (143) 𝑁 = 𝐿 × 𝑛 × 𝑒 ( ) (144) 𝑁 = 𝐿 × 𝑛 × 𝑒 ( ) (145) 𝐶𝑀𝐹 = 𝑒 ( ) (146) 𝐶𝑀𝐹 = 𝑒 ( . ) (147) 𝐶𝑀𝐹 = 𝑒 ( ) (148) 𝐶𝑀𝐹 = 𝑒 / (149) 𝐶𝑀𝐹 _ = 𝑒 _ ( _ ) (150) 𝐶𝑀𝐹 _ = 𝑒 _ ( _ ) (151) 𝐶𝑀𝐹 _ = 𝑒 _ ( _ ) (152) 𝐶𝑀𝐹 , = 𝑒 , ; j= mv, sv (153) 𝐶𝑀𝐹 = 1.0 + 0.01𝐷𝑒 (154) where, 𝑁 = predicted annual average crash frequency for model j (j=mv, sv). 𝑁 = predicted annual average multiple-vehicle crash frequency. 𝑁 = predicted annual average single-vehicle crash frequency. 𝐼 = crash indicator variable (= 1.0 if multiple-vehicle crash data, 0.0 otherwise). 𝐼 = crash indicator variable (= 1.0 if single-vehicle crash data, 0.0 otherwise). 𝐿 = segment length, mi. 𝑛 = number of years of crash data. 𝐴𝐴𝐷𝑇 = average annual daily traffic, veh/day. 𝐼 = California state indicator variable (= 1.0 if site is in California, 0.0 if not). 𝐼 = Illinois state indicator variable (= 1.0 if site is in Illinois, 0.0 if not). 𝐶𝑀𝐹 = lane width CMF. 𝐶𝑀𝐹 = shoulder width CMF. 𝐶𝑀𝐹 = median width CMF. 𝐶𝑀𝐹 = railroad crossing CMF. 𝐶𝑀𝐹 _ = major commercial driveways CMF. 𝐶𝑀𝐹 _ = major industrial driveways CMF. 𝐶𝑀𝐹 _ = minor driveways CMF. 𝐶𝑀𝐹 , = median barrier CMF. 𝐶𝑀𝐹 = roadside fixed objects CMF. 𝑊 = average lane width, ft. 𝑊 = average outside shoulder width, ft. 𝑊 = median width, ft. 𝑛 = number of railroad crossings on the segment. 𝑛 _ = major commercial driveway density, driveways/mile. 𝑛 _ = major industrial driveway density, driveways/mile. 𝑛 _ = minor driveway density, driveways/mile.

94 𝐼 = median barrier presence indicator variable (= 1.0 if present, 0.0 if absent). 𝑂 = roadside fixed-object offset, ft. 𝐷 = roadside fixed-object density, fixed objects/mile. 𝑝 = roadside fixed-object collisions as a proportion of total crashes. 𝑏 = calibration coefficient for variable i. The inverse dispersion parameter, K (which is the inverse of the overdispersion parameter k), is allowed to vary with the segment length. The inverse dispersion parameter is calculated using Equation 155: 𝐾 = 𝐿 × 𝑒 , ; 𝑗 = 𝑚𝑣, 𝑠𝑣 (155) where, 𝐾 = inverse dispersion parameter. 𝛿 = calibration coefficient for inverse dispersion parameter. The predictive model calibration process consisted of the simultaneous calibration of multiple- vehicle and single-vehicle crash models and CMFs using the aggregate model represented by the equations above. The simultaneous calibration approach was needed because several CMFs were common to multiple-vehicle and single-vehicle crash models. The database assembled for calibration included two replications of the original database. The dependent variable in the first replication was set equal to the multiple-vehicle crashes. The dependent variable in the second replication was set equal to the single-vehicle crashes. Table 49 and Table 50 summarize the modeling results for two-way arterial segments for FI and PDO crashes, respectively. The variables with the corresponding p-values less than 0.05 can be considered statistically significant (at the significance level α = 0.05). For those few variables where the p-value was greater than 0.05, it was decided that the variable was important to the model, and its trend was found to be consistent with previous research findings (even if the specific value was not known with a great deal of certainty when applied to this database).

95 Table 49. Calibrated coefficients for FI crashes on six-or-more-lane arterials. Coefficient Variable Facility Estimate Std. Error t-statistic p-value 𝑏 Intercept for MV crashes 6U −15.4189 2.7868 −5.53 <0.0001 6D −11.5649 0.5438 −21.27 <0.0001 7T −11.4439 0.5362 −21.34 <0.0001 8D −11.3817 0.5805 −19.61 <0.0001 𝑏 AADT on MV crashes 6U 1.6329 0.2625 6.22 <0.0001 6D 1.2399 0.0526 23.59 <0.0001 7T 1.2399 0.0526 23.59 <0.0001 8D 1.2399 0.0526 23.59 <0.0001 𝑏 Intercept for SV crashes 6U −4.5419 1.3489 −3.37 0.0008 6D −5.2579 0.8626 −6.10 <0.0001 7T −4.5419 1.3489 −3.37 0.0008 8D −5.3556 0.9281 −5.77 <0.0001 𝑏 AADT on SV crashes 6U 0.3694 0.1309 2.82 0.0048 6D 0.4631 0.0835 5.55 <0.0001 7T 0.3694 0.1309 2.82 0.0048 8D 0.4631 0.0835 5.55 <0.0001 𝑏 Lane width All −0.0219 0.0138 −1.58 0.1144 𝑏 Outside shoulder width All −0.0285 0.0045 −6.30 <0.0001 𝑏 Median width 6D/8D −0.0057 0.0012 −4.65 <0.0001 𝑏 , Median barrier on MV crashes 6D/8D −0.5106 0.1550 −3.29 0.0010 𝑏 , Median barrier on SV crashes 6D/8D 0.6766 0.2099 3.22 0.0013 𝑏 Railroad crossing presence All 0.0388 0.0218 1.78 0.0747 𝑏 _ Major commercial driveway density on MV crashes All 0.0350 0.0038 9.20 <0.0001 𝑏 _ Major industrial driveway density on MV crashes All 0.0107 0.0085 1.25 0.2105 𝑏 _ Minor driveway density on MV crashes All 0.0054 0.0015 3.72 0.0002 𝑏 Roadside fixed-object density on SV crashes All 0.1310 0.0366 3.58 0.0004 𝑏 Added effect of Illinois All −0.3808 0.0475 −8.02 <0.0001 𝛿 Inverse dispersion parameter for MV crashes 6U 2.8668 0.2825 10.15 <0.0001 6D 2.0469 0.0586 34.96 <0.0001 7T 1.2993 0.1198 10.84 <0.0001 8D 2.4932 0.1738 14.35 <0.0001 𝛿 Inverse dispersion parameter for SV crashes 6U 3.0797 0.9312 3.31 0.0010 6D 1.4992 0.1316 11.39 <0.0001 7T 3.0797 0.9312 3.31 0.0010 8D 2.0078 0.3753 5.35 <0.0001 Observations 2229 segments (6U=92; 6D=1759; 7T=222; 8D=113) Note: MV = multiple vehicle; SV = single vehicle.

96 Table 50. Calibrated coefficients for PDO crashes on six-or-more-lane arterials. Coefficient Variable Facility Estimate Std. Error t-statistic p-value 𝑏 Intercept for MV crashes 6U −15.6792 2.2895 −6.85 <0.0001 6D −9.2080 0.5054 −18.22 <0.0001 7T −9.1980 0.4998 −18.40 <0.0001 8D −8.8445 0.5459 −16.20 <0.0001 𝑏 AADT on MV crashes 6U 1.6966 0.2160 7.85 <0.0001 6D 1.0611 0.0490 21.68 <0.0001 7T 1.0611 0.0490 21.68 <0.0001 8D 1.0611 0.0490 21.68 <0.0001 𝑏 Intercept for SV crashes 6U −3.9795 1.3071 −3.04 0.0023 6D −4.7118 0.6937 −6.79 <0.0001 7T −3.9795 1.3071 −3.04 0.0023 8D −4.3443 0.7476 −5.81 <0.0001 𝑏 AADT on SV crashes 6U 0.3429 0.1269 2.70 0.0068 6D 0.4341 0.0671 6.47 <0.0001 7T 0.3429 0.1269 2.70 0.0068 8D 0.4341 0.0671 6.47 <0.0001 𝑏 Lane width All −0.0516 0.0138 −3.75 0.0002 𝑏 Outside shoulder width All −0.0278 0.0044 −6.26 <0.0001 𝑏 Median width 6D/8D −0.0035 0.0011 −3.11 0.0019 𝑏 , Median barrier on MV crashes 6D/8D −0.7651 0.1517 −5.04 <0.0001 𝑏 , Median barrier on SV crashes 6D/8D 0.5723 0.1545 3.70 0.0002 𝑏 Railroad crossing presence All 0.0420 0.0187 2.25 0.0255 𝑏 _ Major commercial driveway density on MV crashes All 0.0479 0.0040 11.89 <0.0001 𝑏 _ Major industrial driveway density on MV crashes All 0.0091 0.0083 1.09 0.2709 𝑏 _ Minor driveway density on MV crashes All 0.0069 0.0015 4.48 <0.0001 𝑏 Roadside fixed-object density on SV crashes All 0.1461 0.0305 4.79 <0.0001 𝑏 Added effect of Illinois All 0.7871 0.0420 18.73 <0.0001 𝛿 Inverse dispersion parameter for MV crashes 6U 2.9953 0.1894 15.82 <0.0001 6D 1.9099 0.0412 46.41 <0.0001 7T 1.0820 0.1077 10.04 <0.0001 8D 1.6689 0.1367 12.20 <0.0001 𝛿 Inverse dispersion parameter for SV crashes 6U 1.9732 0.2607 7.57 <0.0001 6D 1.9997 0.0888 22.52 <0.0001 7T 1.9732 0.2607 7.57 <0.0001 8D 1.8385 0.2282 8.06 <0.0001 Observations 2229 segments (6U=92; 6D=1759; 7T=222; 8D=113) Indicator variables were included for the states of California and Illinois. However, only the coefficient for Illinois was statistically significant. This means that the magnitude of the crashes between Texas and California are about the same, but Illinois experiences fewer FI crashes and more PDO crashes for the same conditions and exposure. The trend could not be explained by difference in road design among the states. It is likely that the differences between states are due to unobserved variables such as vertical grade, signing, pavement condition, weather, reporting accuracy, and speed limit.

97 The mixed nonlinear regression procedure (NLMIXED) in the Statistical Analysis System (SAS) software was used to estimate the proposed model coefficients. This procedure was used because the proposed predictive model is both nonlinear and discontinuous. The log-likelihood function for the NB distribution was used to determine the best-fit model coefficients. Figure 15 and Figure 16 show the relationship between the number of FI crashes and traffic flow for six-or-more-lane segments for multi-vehicle and single-vehicle crashes, respectively. Figure 15 shows that divided facilities experience fewer multi-vehicle crashes than undivided facilities. Figure 16 shows that six-lane divided facilities experience slightly more single-vehicle FI crashes than do eight-lane divided and six-lane undivided arterials. Figure 15. Graphical form of the SPF for FI multiple-vehicle collisions, six-or-more-lane arterials. Figure 16. Graphical form of the SPF for FI single-vehicle collisions, six-or-more-lane arterials.

98 Figure 17 and Figure 18 show the relationship between the number of PDO crashes and traffic flow for six or more lanes for multi-vehicle and single-vehicle crashes, respectively. Figure 17 shows that eight-lane divided facilities experience more multi-vehicle PDO crashes than do six- lane undivided and divided facilities. Figure 18 shows that eight-lane divided facilities experience more single-vehicle PDO crashes than do six-lane divided and six-lane undivided facilities. Figure 17. Graphical form of the SPF for PDO multiple-vehicle collisions, six-or-more-lane arterials. Figure 18. Graphical form of the SPF for PDO single-vehicle collisions, six-or-more-lane arterials.

99 The proportions in Table 51 are used to separate multiple-vehicle crashes into components by collision type for arterials with six or more lanes. Table 51. Distribution of multiple-vehicle collisions for roadway segments by manner of collision type. Collision Type Proportion of Crashes by Severity Level for Specific Road Types 6U 6D 7T 8D FI PDO FI PDO FI PDO FI PDO Rear-end collision 0.752 0.586 0.769 0.591 0.694 0.588 0.746 0.647 Head-on collision 0.037 0.008 0.012 0.012 0.034 0.012 0.006 0.000 Angle collision 0.064 0.052 0.091 0.081 0.148 0.092 0.147 0.093 Sideswipe, same direction 0.083 0.302 0.087 0.262 0.072 0.255 0.073 0.236 Sideswipe, opposite direction 0.028 0.005 0.011 0.020 0.020 0.024 0.011 0.012 Other multiple-vehicle collisions 0.037 0.046 0.030 0.033 0.031 0.029 0.017 0.012 Source: HSIS data for California (2006–2010). The proportions in Table 52 are used to separate single-vehicle crashes into components by crash type for arterials with six or more lanes. Table 52. Distribution of single-vehicle crashes for roadway segments by collision type for arterials with six or more lanes. Collision Type Proportion of Crashes by Severity Level for Specific Road Types 6U 6D 7T 8D FI PDO FI PDO FI PDO FI PDO Collision with fixed object—left 0.100 0.174 0.296 0.353 0.158 0.248 0.167 0.273 Collision with fixed object— right 0.350 0.413 0.332 0.397 0.495 0.481 0.611 0.591 Collision with other object 0.050 0.130 0.032 0.073 0.011 0.037 0.000 0.045 Other single-vehicle collision 0.500 0.283 0.339 0.177 0.337 0.234 0.222 0.091 Source: HSIS data for California (2006–2010). VEHICLE-PEDESTRIAN COLLISIONS The number of vehicle-pedestrian crashes per year for a roadway segment is estimated using Equation 156. 𝑁 = 𝑁 × 𝑓 (156) where, 𝑁 = predicted average crash frequency of an individual roadway segment (excluding vehicle-pedestrian and vehicle-bicycle collisions). 𝑁 = predicted average crash frequency of vehicle-pedestrian collisions for a roadway segment. 𝑓 = pedestrian crash adjustment factor.

100 The pedestrian crash adjustment factor is estimated by dividing the vehicle-pedestrian crashes by the total segment crashes (excluding vehicle-pedestrian and vehicle-bicycle collisions) for each segment type. Table 53 presents the values of f . All vehicle-pedestrian collisions are considered FI crashes. The HSM adjustment factors (from the original Table 12-16 in HSM Chapter 12) are also displayed for comparison. Pedestrian crash adjustment factors are developed using Equation 157. 𝑓 = 𝑁𝑁 (157) where, 𝑁 = crash frequency of vehicle-pedestrian collisions for an individual roadway segment. 𝑁 = crash frequency of an individual roadway segment (excluding vehicle-pedestrian and vehicle-bicycle collisions). Table 53. Pedestrian crash adjustment factor for two-way roadway segments. Source Road Type Pedestrian Crash Adjustment Factor (fpedr) Posted Speed 30 mph or Lower Posted Speed Greater Than 30 mph Number of Segments Total Pedestrian Crashes Total MV and SV Crashesa 𝒇𝒑𝒆𝒅𝒓 Number of Segments Total Pedestrian Crashes Total MV and SV Crashesa 𝒇𝒑𝒆𝒅𝒓 H SM C h. 1 2 2U 0.036 0.005 3T 0.041 0.013 4U 0.022 0.009 4D 0.067 0.019 5T 0.030 0.023 Pr op os ed 6U 22 10 549 0.018 72 18 1359 0.013 6D 106 69 2377 0.029 1661 369 24720 0.015 7T 16 11 324 0.034 250 138 10016 0.014 8D 1 1 612 122 150 6623 0.023 Note: Shaded cell = data not available. a Excludes pedestrian and bicycle crashes. VEHICLE-BICYCLE COLLISIONS The number of vehicle-bicycle collisions per year for an intersection is estimated using Equation 158. 𝑁 = 𝑁 × 𝑓 (158) where,

101 𝑁 = predicted average crash frequency of an individual intersection (excluding vehicle-pedestrian and vehicle-bicycle collisions). 𝑁 = predicted average crash frequency of vehicle-bicycle collisions for an intersection. 𝑓 = bicycle crash adjustment factor. The bicycle crash adjustment factor is estimated by dividing the vehicle-bicycle crashes by the sum of single-vehicle and multiple-vehicle crashes for each intersection type. Table 54 presents the values of 𝑓 . All vehicle-bicycle collisions are considered FI crashes. The HSM adjustment factors (from HSM Table 12-17) are also displayed for comparison. The adjustment factors are developed using Equation 159. 𝑓 = 𝑁𝑁 (159) where, 𝑁 = crash frequency of an individual roadway segment (excluding vehicle- pedestrian and vehicle-bicycle collisions). 𝑁 = crash frequency of vehicle-bicycle collisions for an individual roadway segment. Table 54. Bicycle crash adjustment factor for two-way roadway segments. Source Road Type Bicycle Crash Adjustment Factor (fbiker) Posted Speed 30 mph or Lower Posted Speed Greater Than 30 mph Number of Segments Total Bicycle Crashes Total MV and SV Crashesa 𝒇𝒃𝒊𝒌𝒆𝒓 Number of Segments Total Bicycle Crashes Total MV and SV Crashesa 𝒇𝒃𝒊𝒌𝒆𝒓 H SM C h. 1 2 2U 0.018 0.004 3T 0.027 0.007 4U 0.011 0.002 4D 0.013 0.005 5T 0.050 0.012 Pr op os ed 6U 22 7 549 0.013 72 9 1359 0.007 6D 106 16 2377 0.007 1661 190 24720 0.008 7T 16 8 324 0.025 250 46 10016 0.001 8D 1 0 612 122 92 6623 0.014 Note: Shaded cell = data not available. a Excludes pedestrian and bicycle crashes. CMFS FOR SIX-OR-MORE-LANE ARTERIALS Several CMFs were calibrated in conjunction with the SPFs. All of them were calibrated using the FI crash data. Collectively, they describe the relationship between various geometric factors and crash frequency. These CMFs are described in this section and, where possible, compared with the findings from previous research as a means of model validation. Many of the CMFs

102 found in the literature are typically derived from (and applied to) the combination of multiple- vehicle and single-vehicle crashes. That is, one CMF is used to indicate the influence of a specified geometric feature on total crashes. In contrast, the models developed for this project include several CMFs that are calibrated for a specific crash type. If the standard errors of the CMFs are desired, then Equations 133–140 can be used to compute them. This section shows figures of the CMFs developed from the regression models described above for six-or-more-lane arterials. Where available, other CMFs from the literature are used for comparison purposes. Lane Width CMF The lane width CMF is described using Equation 160: 𝐶𝑀𝐹 = 𝑒 . ( ) (160) The base condition for this CMF is a 12-ft lane width. The lane width used in this CMF is an average for all through lanes on the segment. The lane width CMF is shown in Figure 19 using a thick, solid trend line. The lane widths used to calibrate this CMF range from 9 to 16 ft. This CMF is applicable to both multi-vehicle and single-vehicle crashes. Also shown in Figure 19 are CMFs developed by other researchers. Broken lines are used to differentiate these CMFs from the one proposed in this research project. The proposed CMF closely tracks the CMFs developed by Bonneson and Pratt (2009) for nonrestrictive median segments. The CMF developed by Petritsch et al. (2007) is shown to be more sensitive to lane width than the proposed CMF or the CMF developed by Bonneson and Pratt (2009). Figure 19. Lane width CMF, six-or-more-lane arterials.

103 Outside Shoulder Width CMF The outside shoulder width CMF is described using Equation 161. 𝐶𝑀𝐹 = 𝑒 . ( . ) (161) The base condition for this CMF is a 1.5-ft outside shoulder width. The shoulder width used in this CMF is an average of two roadbeds on the segment. The outside shoulder width CMF is shown in Figure 20 using a thick, solid trend line. The outside shoulder widths used to calibrate this CMF range from 0 to 14 ft. This CMF is applicable to both multi-vehicle and single-vehicle crashes. The outside shoulder width CMFs developed in different studies are compared in Figure 6. Thin lines or broken lines are used to differentiate these CMFs from the one developed for this research project. The CMF proposed in this research closely tracks the CMF for restrictive median segments developed by Bonneson and Pratt (2009) and for PDO crashes developed by Petritsch et al. (2007). Figure 20. Outside shoulder width CMF, six-or-more-lane arterials. Median Width CMF The median width CMF is described using Equation 162. 𝐶𝑀𝐹 = 𝑒 . ( ) (162) The base condition for this CMF is a 15-ft median width. The median width CMF is shown in Figure 21 using a thick, solid trend line. The median widths used to calibrate this CMF range from 0 to 60 ft. This CMF is applicable to both multi-vehicle and single-vehicle crashes. The

104 CMF proposed in this research is compared with the CMF in HSM Chapter 12 and CMFs developed by other researchers in Figure 21. The HSM Chapter 12 CMF applies only to traversable medians without traffic barriers, not including TWLTLs. As shown, there is considerable variation in the median width CMFs. This variation is likely due to other factors that are correlated with median type. For example, a restrictive median reduces the effective number of driveways by preventing through and left-turn movements into or out of driveways. Figure 21. Median width CMF, six-or-more-lane arterials. Median Barrier CMF The median barrier CMF is applicable to cable barriers and concrete barriers on roadway segments. The base condition is a median with no barrier. The calibrated median barrier CMF has two forms, depending on which component model is being used. The median barrier CMF for multiple-vehicle crashes is described using Equation 163. 𝐶𝑀𝐹 , = 𝑒 . × (163) Figure 22 shows the change in the median barrier CMF value for multiple-vehicle crashes with the presence of a median barrier. The results suggest that the presence of a median barrier reduces multiple-vehicle crash frequency. In general, a median barrier prevents vehicles from entering into opposing traffic on the other roadbed and thus reduces the number of crashes.

105 Figure 22. Median barrier CMF for multiple-vehicle crashes, six-or-more-lane arterials. The median barrier CMF for single-vehicle crashes is described using Equation 164. 𝐶𝑀𝐹 , = 𝑒 . × (164) Figure 23 shows the change in the median barrier CMF value for single-vehicle crashes with the presence of a median barrier. The results suggest that the presence of a median barrier increases single-vehicle crash frequency. Although a median barrier prevents a vehicle from entering into opposing traffic on the other roadbed, the vehicle will still be involved in a collision with the barrier.

106 Figure 23. Median barrier CMF for single-vehicle crashes, six-or-more-lane arterials. Railroad Crossing Presence CMF The railroad crossing presence CMF is described using Equation 165. 𝐶𝑀𝐹 = 𝑒 . × / (165) The base condition for this CMF is the absence of a railroad crossing on the segment. This CMF is applicable to both multi-vehicle and single-vehicle crashes. The change in the CMF with the increase in railroad crossings is shown in Figure 24. The crashes increase by 4 percent with each railroad crossing on the segment. Figure 24. Railroad crossing CMF, six-or-more-lane arterials.

107 Driveway CMF The driveway CMF is applicable to multiple-vehicle crashes only. Major commercial, major industrial, and minor driveways are found to be significant in influencing crashes. Minor driveways include all driveway types. Major driveways are those that serve sites with 50 or more parking spaces. Minor driveways are those that serve sites with fewer than 50 parking spaces. Commercial driveways provide access to establishments that serve retail customers. Industrial/institutional driveways serve factories, warehouses, schools, hospitals, churches, offices, public facilities, and other places of employment. Residential driveways serve single- and multiple-family dwellings. The major commercial driveway CMF is described using Equation 166. 𝐶𝑀𝐹 _ = 𝑒 . ( _ ) (166) The major industrial driveway CMF is described using Equation 167. 𝐶𝑀𝐹 _ = 𝑒 . ( _ ) (167) The base condition for the commercial driveway CMF is two driveways per mile, whereas it is one driveway per mile for the industrial driveway CMF. The comparison of CMFs is shown in Figure 25. It can be seen that commercial driveways are associated with more multiple-vehicle crashes than are industrial driveways. Figure 25. Major driveway CMF, six-or-more-lane arterials. The minor driveway CMF is described using Equation 168. 𝐶𝑀𝐹 _ = 𝑒 . ( _ ) (168)

108 The base condition for the minor driveway CMF is 10 driveways per mile. The change in CMF with the increase in the driveways is shown in Figure 26. Figure 26. Minor driveway CMF, six-or-more-lane arterials. Table 55 shows the comparison of percentage increase in crashes associated with the presence of a driveway on an example 1-mi urban street segment. The related percentage crash increase found in this research is similar to the increase found by other various researchers. Table 55. Increase in crashes with driveways. Source Crash Severities Percent Increase in Crashes per Driveway Petritsch et al. (2007) BC 0.2 Petritsch et al. (2007) PDO 0.2 Sawalha & Sayed (2001) KABCO 1.7 Bonneson & McCoy (1997) KABCO 0.5 Proposed, major commercial KABCO 4.0 Proposed, major industrial KABCO 1.0 Proposed, minor KABCO 0.5 Roadside Fixed-Object CMF The roadside fixed-object CMF is applicable to single-vehicle crashes only. It is described using Equation 169. 𝐶𝑀𝐹 = 1.0 + 0.01𝐷𝑒 . (169) The base condition for the roadside fixed-object CMF is absence of roadside objects. The change in the roadside fixed-object CMF with the increase in the offset distance for a segment with 50 roadside objects per mile is shown in Table 56.

109 Table 56. Roadside fixed-object CMF, six-or-more-lane arterials. Offset to Fixed Objects (𝑶𝒇𝒐) (ft) CMF (Proposed) 0 1.50 2 1.38 5 1.26 10 1.13 15 1.07 20 1.04 25 1.02 30 1.01