Median Cross-Section Design for Rural Divided Highways (2014)

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.

46 C H A P T E R 4 The safety analysis involved consideration of roadway and crash data for two types of medians on rural divided high- ways. These median types were traversable medians and medians with barriers. A traversable median is defined as a median in which there is no median barrier and few fixed objects or steep slopes that would hinder an out-of-control vehicle from crossing the entire median and entering the opposing travel lanes. A traversable median may have short lengths of roadside bar- rier at individual fixed objects or terrain features, but does not have a continuous median barrier. If an out-of-control vehicle enters a traversable median, three outcomes are pos- sible; the vehicle will either return to the original traveled way, come to rest in the median due to the driver’s action or of its own accord, or cross the median and enter the oppos- ing lanes. The objective of the safety analysis of traversable medians is to develop a methodology to predict the safety performance of traversable medians, specifically expected crash frequency, and severity, as a function of key median cross-section design variables including median width and median slopes. The definition of a barrier median is self-evident; such medi- ans contain a barrier placed by the highway agency to mini- mize the possibility that out-of-control vehicles will cross the median. A barrier median is one with a continuous median barrier for an extended section. A traversable median with short lengths of barrier at individual fixed objects and terrain features is not considered a barrier median. 4.1 Target Crashes The safety performance measures used in the statistical modeling for median cross sections included total crashes and crash severity levels. The primary safety analysis consid- ered median-related crashes and included crashes in travers- able medians, and crashes in medians with barriers. 4.1.1 Median-Related Crashes Median-related crashes are defined as crashes in which one or more of the involved vehicles departs from the left side of one of the divided roadways and enters the median. Median- related crashes were classified into specific crash types by the crash outcome. The basic crash types that can occur in any median type include • Rollover crashes in the median, • Fixed-object crashes in which one or more vehicles strike a fixed-object (other than median barrier), • Crash-involved vehicle comes to a rest in the median, and • Vehicle departs from, and then returns to, the original trav- eled way. In addition, there are median-related crash types that can occur in only specific median types. These include • Cross-median crashes that can only occur in traversable medians or in barrier medians where a vehicle crosses or penetrates the barrier and • Barrier crashes in which one or more of the involved vehi- cles strike the median barrier. Each of these crash types is addressed in the following discussion of the median types in which they may occur. 4.1.2 Crashes in Traversable Medians All of the four basic crash types that can occur in any median type are possible on traversable medians. In addition, the following specific crash types that will primarily occur on roadways with traversable medians include • Crossed median, entered opposing traveled way and collided with an opposing vehicle; and • Crossed median, entered opposing traveled way, but did not collide with an opposing vehicle. Safety Analysis of Median Cross-Section Design

47 The first of these crash types, which involves a collision with an opposing vehicle, is often referred to as a cross-median crash (CMC) or a crossover crash. The second of these crash types, which involves a vehicle that enters or crosses the oppos- ing roadway but does not collide with an opposing vehicle, has been referred to as noncollision cross-median crashes (NCMC). CMCs have high severity because they involve a col- lision between vehicles traveling in opposite directions. Con- cern about the frequency of CMCs has led to the addition of barriers to traversable medians. The four basic crash types listed in Section 4.1.1 can be readily identified from computerized crash records. CMCs and NCMCs are difficult to identify without a review of hardcopy police crash reports. Most state accident data do not include a code that indicates explicitly whether a vehicle entered an opposing roadway or collided with an opposing vehicle. Procedures were developed to use com- puterized crash data in identifying candidate crashes, but review of hardcopy police crash reports was done for some crashes to ensure that CMCs and NCMCs were correctly identified. 4.1.3 Crashes in Medians with Barriers Specific median-related crash types that primarily occur on roadways that have a median barrier include the following: • Hit barrier; • Went over or through barrier, hit opposing vehicle (CMC); • Went over or through barrier, entered opposing roadway but did not collide with opposing vehicle (NCMC); • Rollover; • Fixed object crash; and • Other median-related crash. The research team assigned crashes to these categories based primarily on computerized crash data, but a limited review of hardcopy police crash reports from Ohio and Pennsylvania was used to test the methodology for properly classifying some crashes. A review of hardcopy police crash reports was con- ducted to verify that all barrier-involved crashes could be classified with computerized crash data. 4.1.4 Crash Sorting Sequence Electronic crash databases vary by state, but the general procedure for identifying median-related crashes had the fol- lowing steps: 1. Obtain all crashes for a rural divided highway segment. 2. Sort crashes to distinguish median-related crashes from on-roadway crashes and run-off-road crashes involving the right side of the roadway. Fields such as first harm- ful event, most harmful event, and sequence of events, as well as the overall crash classification, were used to identify crashes in which a vehicle did run off the road into the median. The median-related crashes were then culled to remove ramp, backing, and wrong-way crashes, and crashes in which vehicles purposefully entered or crossed the median. 3. Sort the median-related crashes to identify overturn, struck-barrier, vaulted-, or broke-through-barrier, CMCs, and NCMCs. 4. Determine the proportion of severe crashes in each crash category. (The sorting sequences for each state are shown in Appen- dix C which is available on the TRB website.) 4.2 Analysis of Traversable Medians The objective of the analysis was to determine the effects of key median cross-section design features and roadway char- acteristics on the safety performance of traversable medians. The key median cross-section design features of interest are median width and median slopes. The key roadway charac- teristic of interest is traffic volume. Other roadway charac- teristics considered are median shoulder types and widths, locations of interchange ramps, particularly entrance ramps, locations of horizontal curves, and presence of median shoul- der rumble strips. Past studies of the safety performance of geometric design features of roadways have used two fundamentally different approaches: before-after evaluation and statistical model development. Before-after evaluation was not a feasible approach to analysis of traversable medians in this research because highway agencies seldom implement projects that change the width or slope of a traversable median without making other more extensive changes as part of the same project. Therefore, a statistical modeling approach based on negative binomial regression was used in the analysis of traversable medians. A commonly used term for the type of analysis in highway research is cross-sectional analysis. How- ever, this term may be confusing in the context of this proj- ect, because it refers to a statistical cross-sectional approach, rather than to analysis of the highway median cross section; therefore, the term cross-sectional analysis is not used in this research. On many rural divided highways, median widths are consistent over extended sections of roadway, although median widths for some roadways with curvilinear align- ment are designed to vary continuously. In contrast, on

48 many divided roadways median slopes may vary along the roadway as the cross-section design is adapted to specific terrain features. Roadway sections used to assess the safety performance of traversable medians are defined on the basis of their median cross-section design policies. Median cross-section designs for traversable medians are characterized by median width (measured from edge of traveled way to edge of traveled way) and the steepness of the median slopes (1V:4H, 1V:6H, 1V:8H, 1V:10H, etc.). Specific combinations of median width and median slope can be referred to either as a typical cross- section or a median cross-section design policy. The objective of the analysis of traversable medians was to determine and compare the expected crash frequency and crash severity distribution for specific median cross-section designs. The remainder of this section presents the safety mea- sures (dependent variables), median cross-section and road- way characteristics (independent variables), data collected, and data analyses performed. 4.2.1 Safety Measures (Dependent Variables) The safety measures for the analysis of traversable medi- ans are the crash frequencies (overall and by crash severity level) for specific median typical cross sections. Specific crash severity levels of interest include the following: • All crash severity levels combined, • Fatal-and-injury crashes, and • Property-damage-only crashes. Decisions concerning appropriate median cross-section designs are driven primarily by the frequencies of fatal-and- injury crashes. However, the frequency of less severe injury crashes must be considered because of their importance in comparison of the differences in safety performance between traversable and barrier medians. Property-damage-only crashes were considered, particularly because placement of a barrier would be expected to increase property-damage-only crashes. Specific crash types of interest are as follows: • Median-related crashes, • CMCs, • NCMCs, • Rollover crashes, • Fixed-object crashes, and • Other median-related crashes. 4.2.2 Median Cross-Section and Roadway Characteristics (Independent Variables) The median cross-section and roadway characteristics of interest are as follows: • Traffic volume (ADT) for mainline roadways; • Median width (edge of traveled way to edge of traveled way); • Typical median slope category (e.g., 1V:4H, 1V:6H, 1V:10H, etc.); • Number of lanes by direction of travel; • Inside shoulder type and width; • Presence of entrance and exit ramps; • Presence of horizontal curves; and • Presence of rumble strips. Independent variables in addition to ADT and median width were incorporated in models where they are found to significantly affect median crash frequencies or severities. 4.2.3 Data Collection Steps in collecting the data for analysis of traversable medi- ans included choosing cooperating states that could furnish crash and roadway data needed, selecting sites for each agency, and collecting field and other supplementary data needed to ensure a credible study. Participating States Several state highway agencies agreed to furnish data for this research. Of those who were willing to furnish data the researchers identified six states (California, Missouri, North Carolina, Ohio, Pennsylvania, and Washington) with reliable roadway and crash data and appropriate sites for the research. These states were selected for the following reasons: • California, North Carolina, Ohio, and Washington are par- ticipants in FHWA’s Highway Safety Information System (HSIS). Their data are, therefore, readily accessible to the research team and known to be of high quality. • Missouri was selected because of MoDOT’s extensive recent installation of cable barrier in freeway medians and because of their extensive mileage of rural divided nonfreeways. In addition, the research team has worked extensively with MoDOT data in past studies, and has an established agree- ment with MoDOT for online access to their roadway, crash, and video log data.

49 • Pennsylvania was selected because their existing comput- erized roadway data files already classify divided highway medians into traversable, nontraversable, or barrier catego- ries. The research team has used Pennsylvania data exten- sively in previous median cross-section safety research for PennDOT. These states had sufficient mileage sites available to furnish the required sample sizes and had medians with various typi- cal cross sections. Data Set Collected from Computerized Files The following list identifies the variables collected from existing computerized data files provided by state highway agencies (or obtained from HSIS to save state highway agency labor): • Median width, • Presence/absence of barrier, • Type of barrier, • Barrier location relative to traveled way, • Number of lanes by direction of travel, • Median shoulder width (ft), • Presence of rumble strips on median shoulder, • Segment length (mi), • Posted speed limit, • ADT (veh/day), • Ramp locations, • Mileposts or other location reference data that ties to crash locations, and • Electronic crash records. Note, some data was obtained or verified from state high- way agency videologs or verified during field data collection. Median widths were available from existing computerized roadway inventory databases, but review of typical cross sec- tions on as-built plans was not a good source of median slope data. Median slope data had to be determined from field data collection as described in this subsection. Data collection for the basic data set for traversable medi- ans was conducted jointly with similar activities for the basic data set for barrier medians. Medians with barriers were cat- egorized by the type of barrier and the location of the barrier within the median, usually near the edge of the shoulder, in the center of the median, or 2.4 m (8 ft) upslope from the median ditch line. At least 5 years of crash data were obtained as well as yearly ADT data corresponding to the crash data. For the states of California, North Carolina, Ohio, and Washington, crash data was obtained from the FHWA HSIS system. The research team had on-line access to Missouri crash data. Pennsylvania crash data was obtained through arrange- ments with PennDOT. The crash data period varied from state to state. The years of crash data used from each state are as follows: • California—2001 to 2005, • Missouri—2002 to 2007, • North Carolina—2000 to 2004, • Ohio—2001 to 2005, • Pennsylvania—2002 to 2006, and • Washington—2001 to 2005. Field Data Collection The following two methods were used to collect field data: • Use of Penn State’s scanning laser system developed dur- ing this research. Penn State’s system, described in more detail in Appendix D of this report (available on the TRB website), consists of a scanning laser mounted on the rear of a vehicle. This system has the capability to create a digi- tal terrain map of the median. This system was used to collect detailed field data in California, North Carolina, Ohio, Pennsylvania, and Washington State. In all partici- pating states except Pennsylvania, data sections coincided with mile markers and were 1.6 km (1 mi) in length. In Pennsylvania, reference markers are placed every 0.8 km (0.5 mi), so study sections were 0.8 km (0.5 mi) in length. • Manual measurements of median cross-section features including median slopes. In Missouri, manual measurement of median cross-section features was made by a team of two trained field personnel. Slope measurements were made with an electronic level and dimensions of cross-section features were measured with a measuring wheel or tape. Point or continuous objects in the median, other than median barrier, were measured over a 60-m (200-ft) section, 30 m (100 ft) either side of the sample measurement point. Field data collectors also categorized the horizontal or vertical curves at each measurement point. Presence of shoulder rumble strips was verified. A pilot test was conducted as part of the research to estimate field data collection costs. Measure- ments of the median cross section for a 3.2-km (2-mi) sec- tion required approximately 3 hours for a two-person team. The manual field measurement method was very labor intensive and the scanning laser system was used in five of the six participating states to minimize the cost of collecting a detailed data set. Automated data collection was shown to

50 be both accurate and less expensive than manual field data collection. 4.2.4 Crash Statistics for Traversable Medians The mileage of freeway and nonfreeway sites is shown in Table 4-1. The total roadway length considered was 3,250.8 km (2,020.4 mi). There were 1,976.4 km (1,228.3 mi) with travers- able medians. The sample of rural freeways totaled 1,444.7 km (897.9 mi) in length, while rural nonfreeways totaled 355.0 km (220.6 mi). There were 176.7 km (109.8 mi) of traversable medians on urban roadways. The frequency of crashes for traversable sites on rural freeways is shown in Table 4-2. This table also shows the number and proportion of median-related crashes. On rural freeways with traversable medians, about 26 percent of total crashes were median related. Table 4-2 also gives the frequency and proportion of each crash type that con- stitutes median-related crashes on traversable medians. Rollover crashes are about 35 percent of median-related crashes on traversable medians. Fixed-object crashes are about 33 percent, and other median-related crashes are about 27 percent of median-related crashes. The CMCs and NCMCs total less than 5 percent of median-related crashes on rural freeways. The frequency of crashes for traversable sites on rural nonfreeways is shown in Table 4-3. This table is formatted in the same way as Table 4-2. On rural nonfreeways with traversable medians, about 31 percent of total crashes were median related. Fixed-object crashes were 27 percent, and other median-related crashes were about 35 percent. CMCs and NCMCs totaled almost 7 percent of the median-related crashes on rural nonfreeways. State Area type Road type Traversable median (mi) Barrier median (mi) Combined (mi) CA Rural Freeway 263.17 52.00 315.17 Nonfreeway 37.12 3.86 40.98 Urban Freeway 14.00 12.00 26.00 MO Rural Freeway 55.00 65.50 120.50 Nonfreeway 150.50 0.00 150.50 Urban Freeway 0.00 3.00 3.00 Nonfreeway 6.00 0.00 6.00 NC Rural Freeway 17.00 385.81 402.81 Nonfreeway 0.00 16.55 16.55 Urban Freeway 20.79 92.85 113.64 Nonfreeway 0.00 5.00 5.00 OH Rural Freeway 126.81 45.08 171.89 Urban Freeway 37.87 22.72 60.59 PA Rural Freeway 254.59 32.47 287.06 WA Rural Freeway 181.33 43.24 224.57 Nonfreeway 33.02 0.00 33.02 Urban Freeway 31.14 11.97 43.11 Total 1,228.34 792.05 2,020.39 Table 4-1. Total roadway mileage for project database. State Total crashes Crash type Median- related crashes (percent of total) Rollover crashes (percent of median- related) CMCs (percent of median-related) NCMCs (percent of median-related) Fixed-object crashes (percent of median-related) Other median- related crashes (percent of median- related) CA 4,651 1,516 (32.59) 698 (46.04) 56 (3.69) 4 (0.26) 273 (18.00) 485 (31.99) MO 878 297 (33.82) 75 (25.25) 13 (4.37) 15 (5.05) 61 (20.53) 133 (44.78) NC 595 73 (12.26) 7 (9.58) 0 (0.00) 1 (1.36) 26 (35.61) 39 (53.42) OH 5,497 1,184 (21.53) 198 (16.72) 33 (2.78) 1 (0.08) 459 (38.76) 493 (41.63) PA 3,467 844 (24.34) 137 (16.23) 28 (3.31) 24 (2.84) 543 (64.33) 112 (13.27) WA 3,298 890 (26.98) 589 (66.17) 29 (3.25) 10 (1.12) 211 (23.70) 51 (5.73) Total 18,386 4,804 (26.12) 1,704 (35.47) 159 (3.30) 55 (1.14) 1,573 (32.74) 1,313 (27.33) Table 4-2. Crash frequency for traversable medians on rural freeways.

51 State Severity level Total crashes Crash type Median-related crashes (percent of total crashes) Rollover crashes (percent of median-related crashes) CMCs (percent of median- related crashes) NCMCs (percent of median- related crashes) Hit-fixed-object crashes (percent of median-related crashes) Other median- related crashes (percent of median-related crashes) CA Fatal 175 102 (58.28) 61 (59.80) 13 (12.74) 1 (0.98) 9 (8.82) 18 (17.64) Injury 1,745 762 (43.66) 449 (58.92) 33 (4.33) 2 (0.26) 60 (7.87) 218 (28.60) PDO 2,731 652 (23.87) 188 (28.83) 10 (1.53) 1 (0.15) 204 (31.28) 249 (38.19) MO Fatal 20 10 (50.00) 3 (30.00) 3 (30.00) 0 (0.00) 1 (10.00) 3 (30.00) Injury 240 136 (56.66) 48 (35.29) 7 (5.14) 2 (1.47) 22 (16.17) 57 (41.91) PDO 618 151 (24.43) 24 (15.89) 3 (1.98) 13 (8.60) 38 (25.16) 73 (48.34) NC Fatal 6 0 (0.00) 0 (0.00) 0 (0.00) 0 (0.00) 0 (0.00) 0 (0.00) Injury 197 29 (14.72) 5 (17.24) 0 (0.00) 0 (0.00) 10 (34.48) 14 (48.27) PDO 392 44 (11.22) 2 (4.54) 0 (0.00) 1 (2.27) 16 (36.36) 25 (56.81) OH Fatal 26 17 (65.38) 7 (41.17) 6 (35.29) 0 (0.00) 3 (17.64) 1 (5.88) Injury 994 384 (38.63) 124 (32.29) 16 (4.16) 0 (0.00) 109 (28.38) 135 (35.15) PDO 4,477 783 (17.48) 67 (8.55) 11 (1.40) 1 (0.12) 347 (44.31) 357 (45.59) PA Fatal 73 29 (39.72) 15 (51.72) 6 (20.68) 0 (0.00) 8 (27.58) 0 (0.00) Injury 1,453 333 (22.91) 85 (25.52) 14 (4.20) 15 (4.50) 171 (51.35) 48 (14.41) PDO 1,941 482 (24.83) 37 (7.67) 8 (1.65) 9 (1.86) 364 (75.51) 64 (13.27) WA Fatal 52 22 (42.30) 18 (81.81) 3 (13.63) 0 (0.00) 1 (4.54) 0 (0.00) Injury 1,223 416 (34.01) 343 (82.45) 15 (3.60) 1 (0.24) 47 (11.29) 10 (2.40) PDO 2,023 452 (22.34) 228 (50.44) 11 (2.43) 9 (1.99) 163 (36.06) 41 (9.07) Total Fatal 352 180 (51.13) 104 (57.77) 31 (17.22) 1 (0.55) 22 (12.22) 22 (12.22) Injury 5,852 2,060 (35.20) 1,054 (51.16) 85 (4.12) 20 (0.97) 419 (20.33) 482 (23.39) PDO 12,182 2,564 (21.04) 546 (21.29) 43 (1.67) 34 (1.32) 1,132 (44.14) 809 (31.55) Total 18,386 4,804 (26.12) 1,704 (35.47) 159 (3.30) 55 (1.14) 1,573 (32.74) 1,313 (27.33) Table 4-4. Crash severity distribution for traversable medians on rural freeways. State Total crashes Crash type Median-related crashes (percent of total) Rollover crashes (percent of median- related) CMCs (percent of median- related) NCMCs (percent of median- related) Fixed-object crashes (percent of median- related) Other median- related crashes (percent of median- related) CA 221 68 (30.76) 29 (42.64) 6 (8.82) 1 (1.47) 13 (19.11) 19 (27.94) MO 1,843 579 (31.41) 143 (24.69) 21 (3.62) 16 (2.76) 171 (29.53) 228 (39.37) WA 228 55 (24.12) 42 (76.36) 3 (5.45) 0 (0.00) 9 (16.36) 1 (1.81) Total 2,292 702 (30.62) 214 (30.48) 30 (4.27) 17 (2.42) 193 (27.49) 248 (35.32) Table 4-3. Crash frequency for traversable medians on rural nonfreeways. The crash severity distribution for traversable medians on rural freeways is shown in Table 4-4. In traversable medians, CMCs are much more severe than fixed-object or rollover crashes. Although CMCs are only 3.3 percent of total median- related crashes, they represent over 17 percent of the fatal crashes in traversable medians on rural freeways. Rollover crashes also are severe and represent 58 percent of fatal and 52 percent of injury crashes on traversable medians on rural freeways. The crash severity distribution for traversable medians on rural nonfreeways is shown in Table 4-5. Again, CMCs are much more severe than fixed-object or rollover crashes. Although CMCs are only 4 percent of crashes, they represent almost 17 percent of the fatalities on rural nonfreeway tra- versable median sites. Rollover crashes are also severe. They are about 30 percent of crashes, but represent almost 42 per- cent of fatal crashes.

52 4.2.5 Statistical Analysis/Modeling of Crash Data for Traversable Medians Regression models, or safety performance functions (SPFs), were developed to estimate the effect of cross-section design features on traversable median safety performance. Safety per- formance was estimated as a function of traffic volume, median width, and median slope, as well as four other independent variables found to have a statistically significant relationship with safety. The analysis focused on frequency and severity— total and fatal-and-injury—of the following crash types: all median related; rollover in the median; hit fixed object in median (other than barrier); entered opposing traveled way and collided with an opposing vehicle (CMC); entered oppos- ing traveled way, but did not collide with an opposing vehicle (NCMC); all cross-median collision types; and other median- related collisions. Thus, a total of 14 dependent variables were considered for 7 target crash types and 2 crash severity levels. SPFs were developed with multiple regression, assuming a negative binomial (NB) error distribution of accident fre- quencies, using data from all traversable median sites and barrier sites for the period before barrier installation. The deci- sion to include the before period of barrier sites was made to maximize the amount of information used to develop the func- tions. Separate models were developed for each state as well as combined state models for each of the following three clas- sifications of roadways: • Four-lane freeway, • Four-lane nonfreeway, and • Six-lane freeway. Two generalized linear modeling techniques were used to fit the data. The first method used a repeated measures correla- tion structure to model yearly crash counts for a site. In this method, the covariance structure, assuming compound sym- metry, is estimated before final regression parameter estimates are determined by general estimating equations. Consequently, model convergence for this method is dependent on the covari- ance estimates as well as parameter estimates. When the model failed to converge for the covariance estimates, an alternative method was considered. In this method, yearly crash counts for a site were totaled and ADT values were averaged to create one summary record for a site. Regression parameter estimates were then directly estimated by maximum likelihood, without an additional covariance structure being estimated. Both methods also produced an estimate of the over- dispersion parameter, or the estimate for which the variance exceeds the mean. Overdispersion occurs in traffic data when a number of sites being modeled have zero accident counts, which creates variation in the data. In this study, most sites tended to have positive accident counts because they were a mile in length, particularly when accident counts were aggre- gated across multiple years. When the estimate for dispersion was very small or even slightly negative, the model was re-fit assuming a constant value. The best estimates of the safety effectiveness of highway design features are likely to be determined from observational before-after studies of projects that changed the design feature in question (and only that design feature). Such evaluations focus solely on the effect of that particular design feature and can be performed using the Empirical Bayes (EB) method Table 4-5. Crash severity distribution for traversable medians on rural nonfreeways. State Severity level Total crashes Crash type Median- related crashes (percent of total crashes) Rollover crashes (percent of median- related crashes) CMCs (percent of median- related crashes) NCMCs (percent of median- related crashes) Fixed-object crashes (percent of median- related crashes) Other median- related crashes (percent of median- related crashes) CA Fatal 6 2 (33.33) 0 (0.00) 1 (50.00) 0 (0.00) 0 (0.00) 1 (50.00) Injury 92 39 (42.39) 21 (53.84) 4 (10.25) 1 (2.56) 2 (5.12) 11 (28.20) PDO 123 27 (21.95) 8 (29.62) 1 (3.70) 0 (0.00) 11 (40.74) 7 (25.92) MO Fatal 42 19 (45.23) 8 (42.10) 3 (15.78) 0 (0.00) 4 (21.05) 4 (21.05) Injury 548 257 (46.89) 89 (34.63) 13 (5.05) 9 (3.50) 52 (20.23) 94 (36.57) PDO 1,253 303 (24.18) 46 (15.18) 5 (1.65) 7 (2.31) 115 (37.95) 130 (42.90) WA Fatal 8 3 (37.50) 2 (66.66) 0 (0.00) 0 (0.00) 1 (33.33) 0 (0.00) Injury 78 26 (33.33) 24 (92.30) 0 (0.00) 0 (0.00) 2 (7.69) 0 (0.00) PDO 142 26 (18.30) 16 (61.53) 3 (11.53) 0 (0.00) 6 (23.07) 1 (3.84) Total Fatal 56 24 (42.85) 10 (41.66) 4 (16.66) 0 (0.00) 5 (20.83) 5 (20.83) Injury 718 322 (44.84) 134 (41.61) 17 (5.27) 10 (3.10) 56 (17.39) 105 (32.60) PDO 1,518 356 (23.45) 70 (19.66) 9 (2.52) 7 (1.96) 132 (37.07) 138 (38.76) Total 2,292 702 (30.62) 214 (30.48) 30 (4.27) 17 (2.42) 193 (27.49) 248 (35.32)

53 to compensate for the potential bias due to regression to the mean. A before-after study of the effects of installing median barrier is presented later in this report in Section 4.3.6. How- ever, before-after evaluations are not often feasible for other roadside design features because such features are seldom changed without accompanying changes to other roadway or roadside features. Where before-after studies are not feasible, cross-sectioned studies based in regression modeling are usually the best avail- able alternative. The results of cross-sectional studies should be used cautiously because correlations between the study variables may produce regression coefficients whose values do not necessarily represent the incremental effects of those variables on crash frequency. A cross-sectional approach of this type has been applied below to the evaluation of median cross-section design features. Models produced for both methods, yearly and summary, estimate crashes per mile per year. To include this rate cal- culation, the first method used site length as an offset in the model while the second method used the number of years of crash data multiplied by site length as the offset. When both methods produce models, parameter estimates are nearly identical. However, confidence intervals for estimates pro- duced with yearly crash counts tend to be wider due to yearly variation. Consequently, statistical significance of the esti- mates is harder to achieve with yearly models. Both methods were accomplished with the GENMOD procedure of SAS. The primary difference between methodologies occurred for the combined models from more than one state. The treatment of the state effect differed between methodolo- gies. For the repeated measures or yearly method, the covari- ance structure used captured most of the variations in the data. Consequently, no additional adjustment was needed to account for the state effect. On the other hand, the summary record method treated state as a random effect in the model. The mixed model for this method was generated with the GLIMMIX procedure of SAS. Of the independent variables summarized in the previous section, an attempt was made to incorporate as many variables, in addition to ADT, in the SPF to obtain the best possible func- tion to predict crashes at traversable median sites. ADT was included in all models, regardless of its significance, as long as its coefficient was positive. Selection of the remaining vari- ables in the model was performed by evaluating the statistical significance at the 80 percent confidence level. Additionally, the following three independent variables had the additional crite- rion that the resulting relationship between the characteristic and safety was meaningful in engineering terms: • Presence of on-ramps increases crashes, • Presence of horizontal curves increases crashes, and • Presence of shoulder rumble strips decreases crashes. All regression models for median-related collision types were developed to predict target crash frequencies per mile per year in the following form: ( )= exp + lnADT + +…+ (2)0 1 2 2N b b b X b Xn n where: N = predicted accident frequency per mile per year ADT = average daily traffic volume (veh/day) b0, . . . , bn = regression coefficients determined by model fitting X1, . . . , Xn = independent design characteristics A typical model of this type has the form:    = exp lnADT + MW + MS + SW + CP + OR + RS (3) 0 1 2 3 4 5 6 7 N b + b b b b b b b where: MW = median width (ft) MS = median slope ratio SW = shoulder width (ft) CP = curve presence (= 0 for tangent sites; = 1 for sites on horizontal curves) OR = on-ramp presence (= 0 for sites with no on-ramp present; = 1 for sites with an on-ramp present) RS = rumble strip presence (= 0 for sites with no shoulder rumble strip present; = 1 for sites with a shoulder rumble strip present) The median slope ratio is the ratio of the horizontal com- ponent of the median foreslope to the vertical component. For example, if the median foreslope is 1V:4H then MS = 4. The modeling process often produced more than one suitable model for the purposes of this research. When this occurred, final models were selected from all possible models by the following considerations: 1. Select the model with the most significant variables. 2. When deciding between models with an equal number of independent variables, select the model with more impor- tant variables (e.g., median slope took precedence over number of ramps). 3. As part of this process, also consider variable effect size. The regression models developed with this approach are presented next, where each of the elements of design are fully discussed. Tables 4-6 and 4-7 present the final SPFs for traversable medians for the total and fatal-and-injury crash severity levels, respectively. Model coefficients including the stan- dard error for each coefficient, as well as the overdispersion parameter for the model, are presented for each crash type

Median-related crash type Road type Model coefficient (standard error) Comment Intercept ADTa Median width Median slope ratio Shoulder width Curve presence On-ramp presence Rumble strip presence Overdispersion parameter All median-related crashes Four-lane freeway –7.9411 0.7946 0.0027 –0.0241 – – 0.2655 – 0.2851 b All years combined (0.6817) (0.0656) (0.0009) (0.0116) – – (0.0652) – (0.0256) All states combined Four-lane nonfreeway –12.2034 1.3205 – –0.0969 – – 0.3844 – 0.6436 Yearly data (3.5107) (0.3682) – (0.0618) – – (0.2090) – (0.5840) Washington data Six-lane freeway –9.4117 0.8980 – 0.0562 – – 0.2012 – 0.2780 Yearly data (1.7015) (0.1590) – (0.0383) – – (0.1463) – (0.0543) All states combined CMCs + NCMCs Four-lane freeway –24.0562 1.9119 – 0.1100 – – – – 0.0598 All years combined (5.5618) (0.5318) – (0.0795) – – – – (0.3924) California data Four-lane nonfreeway –21.7518 2.4317 – –0.5406 – – – – 0.0100 All years combined (12.4365) (1.3494) – (0.3270) – – – – (0.0000) Washington data Six-lane freeway –20.0770 1.5599 –0.0160 0.1810 – – – – 0.0100 b Yearly data (6.0467) (0.5660) (0.0073) (0.1044) – – – – (0.0000) All states combined CMCs Four-lane freeway –29.5036 2.0385 – 0.5523 – – – – 0.0100 All years combined (16.1063) (1.5020) – (0.1660) – – – – (0.0000) Ohio data Four-lane nonfreeway –21.7518 2.4317 – –0.5406 – – – – 0.0100 Yearly data (12.4365) (1.3494) – (0.3270) – – – – (0.0000) California data Six-lane freeway –23.1034 1.8886 –0.0226 0.1474 – – – – 0.0100 b All years combined (6.5012) (0.6074) (0.0081) (0.1087) – – – – (0.0000) All states combined Rollover crashes Four-lane freeway –11.0162 1.1212 – –0.0596 –0.0661 0.1706 – – 0.3531 All states combined (1.7951) (0.1696) – (0.0243) (0.0449) (0.1278) – – (0.0759) California data Four-lane nonfreeway –4.7235 0.6149a – –0.2551 – – – – 0.0100 Yearly data (9.1502) (1.0287) – (0.0748) – – – – (0.0000) California data Six-lane freeway –6,3372 0.4671 a – – – – – – 0.2230 Yearly data (4.0087) (0.3614) – – – – – – (0.3099) Ohio data Fixed-object crashes Four-lane freeway –13.9439 1.0902 – 0.1420 – – 0.8715 – 0.9733 Yearly data (4.4369) (0.4011) – (0.0789) – – (0.3525) – (0.3036) North Carolina data Four-lane nonfreeway –10.3363 0.9632 – 0.0470 –0.1067 – – – 0.0100 Yearly data (2.8848) (0.3008) – (0.0255) (0.0459) – – – (0.0000) Missouri data Six-lane freeway –14.4190 1.3915 –0.0107 – –0.0484 – 0.4502 – 0.1626 All years combined (3.5309) (0.3101) (0.0040) – (0.0316) – (0.1452) – (0.0658) Ohio data Other median- related crashes Four-lane freeway –12.9170 1.1565 0.0088 –0.0561 – 0.2330 – – 0.3001 All years combined (1.8619) (0.1852) (0.0045) (0.0267) – (0.1353) – – (0.0842) California data Four-lane nonfreeway –12.9183 1.2754 –0.0050 – –0.0537 – – – 1.4785 Yearly data (2.3924) (0.2407) (0.0038) – (0.0390) – – – (0.4316) All states combined Six-lane freeway –15.8412 1.4852 –0.0045 – – – 0.4425 –0.2820 0.2692 Yearly data (2.6138) (0.2341) (0.0032) – – – (0.1586) (0.1863) (0.0911) All states combined a Coefficient not significant at 80% level. b State variance less than 0.08. Table 4-6. Safety performance models for all crash severity levels combined for roadways with traversable medians.

Median-related crash type Road type Model coefficient (standard error) Comment Intercept ADTa Median width Median slope ratio Shoulder width Curve presence On-ramp presence Rumble strip presence Overdispersion parameter All median-related crashes Four-lane freeway –11.9079 1.1423 0.0070 –0.0542 – – – –0.1359 0.2516 All years combined (1.4800) (0.1498) (0.0035) (0.0218) (0.1036) (0.0570) California data Four-lane nonfreeway –11.0049 1.1362 – –0.1137 – – – – 0.0135 Yearly data (3.4332) (0.3391) (0.0691) (0.6390) Washington data Six-lane freeway –11.5339 0.9801 – 0.0875 – – – – 0.2149 Yearly data (3.0763) (0.2874) (0.0444) (0.0949) Ohio data CMCs + NCMCs Four-lane freeway –23.9308 1.9937 – 0.1171 –0.2219 – – – 0.1982 All years combined (6.7601) (0.6371) – (0.0896) (0.1538) (0.4933) California data Four-lane nonfreeway –12.1576 0.9228a – – – – – – 0.0100 Yearly data (14.1086) (1.4807) (0.0000) California data Six-lane freeway –19.3150 1.6090 –0.0153 – – – – – 0.9550 Yearly data (7.4574) (0.667) (0.0061) (1.1955) All states combined CMCs Four-lane freeway –20.2012 1.8093 –0.0205 – – – – – 0.1453 All years combined (5.3011) (0.5482) (0.0105) (0.4660) California data Four-lane nonfreeway –10.7480 0.4192a – 0.2759 – – – – 0.0100 Yearly data (18.5931) (2.0183) (0.1683) (0.0000) California data Six-lane freeway –18.6668 1.5600 –0.0177 – – – – – 1.1219 Yearly data (7.6442) (0.6846) (0.0063) – (0.2927) All states combined Rollover crashes Four-lane freeway –11.4640 0.9661 0.0216 –0.0328 –0.1390 0.2587 – – 0.1959 All states combined (1.8352) (0.1786) (0.0046) (0.0243) (0.0472) (0.125) (0.0723) California data Four-lane nonfreeway –4.4858 0.5537 – 0.2469 – – – – 0.0100 Yearly data (10.2028) (1.1427) (0.0866) (0.0000) California data Six-lane freeway –6.2621 0.3234 a – 0.1285 – – – – 0.1616 Yearly data (5.3068) (0.4954) (0.0828) (0.4834) Ohio data Fixed-object crashes Four-lane freeway –9.7775 1.1178 –0.0310 –0.1835 – – – – 1.0308 Yearly data (3.9487) (0.41352) (0.0108) (0.0549) (0.9865) California data Four-lane nonfreeway –9.0654 1.0969a – –0.5954 – – – – 0.0100 All years combined (13.3950) (1.5240) (0.3332) (0.0000) Washington State Six-lane freeway –14.1827 1.2062 a –0.0092 – – – 0.6540 – 0.6676 All years combined (5.6230) (0.5198) (0.0064) (0.2738) (0.2929) Ohio data Other median- related crashes Four-lane freeway –10.960 0.9733 – –0.0569 – – – – 0.3384 All years combined (2.3981) (0.2327) (0.0336) (0.1438) California data Four-lane nonfreeway –8.0851 0.6674 – 0.0631 –0.1068 – – – 0.0100 All years combined (3.2892) (0.3451) (0.0260) (0.0474) (0.0000) Missouri data Six-lane freeway –21.2765 1.9104 –0.0065 – – – 0.4578 –0.5298 0.1657 Yearly data (4.3737) (0.3918) (0.0045) (0.2104) (0.2832) (0.1637) All states combined a Coefficient not significant at 80% level. Table 4-7. Safety performance models for fatal-and-injury crashes for roadways with traversable medians.

56 and roadway type combination. Where no coefficient value is given in the tables, the coefficient was not statistically sig- nificant and should be treated as equal to zero in Equations 2 through 4. All model coefficients shown in Tables 4-6 and 4-7 are statistically significant at the 80 percent confidence level unless otherwise noted. A comment indicating the regression method and dataset used is also provided. For example, the SPF in Table 4-6 for all median-related crashes on four-lane freeways shown in the top line of the table is: = exp –7.9411 + 0.7946 ln ADT + 0.0027MW – 0.024MS + 0.2655OR (4)4FTN     where: N4FT = predicted median-related crash frequency per mile per year on four-lane freeways with traversable medians ADT = average daily traffic volume (veh/day) MW = median width (ft) MS = median slope ratio OR = on-ramp presence The signs of the coefficients indicate how the crash rate changes when changes are made in a design variable such as median width, median slope, etc. Equation 3 indicates that median-related crashes increase as median widths increase (positive coefficient), and decrease as median slopes are flat- tened (negative coefficient). The safety effect for each design variable is the percent change in crash rate that can be expected for each unit change in a design variable (holding all others constant). For exam- ple, the safety effect for median width is shown in Table 4-8. The top line in this table says that for all median-related crashes on four-lane freeways with traversable medians, the rate increases 0.27 percent for each additional foot of median width. The equation used to determine the safety effect is as follows: [ ]( )Safety effect 100 exp coefficient – 1 (5)= This number gives the percent change in crash rate due to an independent variable, provided all other variables remain constant. The modeling results are described below with cautious wording. Many of the observed effects are logical and repre- sent reasonable findings, but their magnitude and statistical significance may be influenced by correlations between some of the variables studied. Median Width The distribution of median widths on the study sites is shown in Table 4-9. The meaning of the statistically significant safety effects for median width shown in Tables 4-7 and 4-8 is as follows: 1. On four-lane freeways, total median-related crashes increased with wider medians. Fatal-and-injury median- related crashes also increased with wider medians. The safety effects were a 0.3 percent increase for each additional foot of median width for all median-related crashes and a 0.7 percent increase for each additional foot of median width for fatal-and-injury median-related crashes. Median-related crash type Road type Percent change in crash frequency per unit change in median width Total severity F & I severity Effect LB UB Effect LB UB All median-related crashes Four-lane freeway 0.27 0.08 0.45 0.71 0.03 1.39 Four-lane nonfreeway Six-lane freeway CMCs + NCMCs Four-lane freeway Four-lane nonfreeway Six-lane freeway –1.58 –3.01 –0.14 –1.52 –2.69 –0.33 CMCs Four-lane freeway –2.03 –4.02 0.00 Four-lane nonfreeway Six-lane freeway –2.23 –3.79 –0.65 –1.76 –2.97 –0.53 Rollover crashes Four-lane freeway 2.18 1.26 3.11 Four-lane nonfreeway Six-lane freeway Fixed-object crashes Four-lane freeway –3.05 –5.08 –0.98 Four-lane nonfreeway Six-lane freeway –1.07 –1.84 –0.29 –0.92 –2.16 0.34 Other median-related crashes Four-lane freeway 0.88 0.00 1.77 Four-lane nonfreeway –0.50 –1.24 0.25 Six-lane freeway –0.45 –1.06 0.17 –0.65 –1.52 0.22 Table 4-8. Safety effect of median width for roadways with traversable medians.

57 2. On six-lane freeways, total CMCs plus NCMCs decreased with wider medians, as did fatal-and-injury CMCs plus NCMCs. The safety effects were a 1.6 percent decrease in CMCs plus NCMCs for each additional foot of median width and a 1.5 percent decrease in fatal-and-injury CMCs plus NCMCs for each additional foot of median width. 3. On four-lane freeways, CMC fatal-and-injury crashes decreased with wider medians. The safety effect was a 2.0 percent decrease in CMC fatal-and-injury crashes for each additional foot of median width. 4. On six-lane freeways, CMCs also decreased with wider medians. The safety effects were a 2.2 percent decrease for each additional foot of median width for CMCs and a 1.8 percent decrease for each additional foot of median width for fatal-and-injury CMCs. 5. For four-lane freeways, fatal-and-injury rollover crashes increased with wider medians. The safety effect was a 2.2 percent increase for each additional foot of median width. 6. For four-lane freeways, the fatal-and-injury fixed-object crashes decreased with wider medians. The safety effect was a 3.1 percent decrease for each additional foot of median width. 7. On six-lane freeways, total fixed-object and fatal-and- injury fixed-object crashes decreased with wider medi- ans. The safety effects were a 1.1 percent decrease in total fixed-object crashes for each additional foot of median width and a 0.9 percent decrease in fatal-and-injury fixed-object crashes for each additional foot of median width. 8. On four-lane freeways, other median-related crashes increased with wider medians. The safety effect was a 0.9 per cent increase for each additional foot of median width. 9. On four-lane nonfreeways, other median-related crashes decreased with wider medians. The safety effect was a 0.5 percent decrease for each additional foot of median width. 10. On six-lane freeways, other median-related crashes and fatal-and-injury other median-related crashes decreased with wider medians. The safety effects were a 0.5 percent decrease for other median-related crashes and a 0.7 per- cent decrease for fatal-and-injury other median-related crashes for each additional foot of median width. In summary, increasing median widths were found to have partially offsetting effects by crash type. Total median- related crashes appear to increase with wider medians. CMCs and NCMCs decreased with wider medians while rollover crashes increased. Fixed-object crashes also decreased with wider medians. Few significant relationships were found for Table 4-9. Median width distribution for study sites. State Median type No. of lanes Road type Roadway length (mi) for sites in categories of median width (ft) < = 15 15 to 35 35 to 55 55 to 75 75 to 95 95 to 115 115+ CA Traversable 4 Freeway 2 17 51 143 50 Nonfreeway 1 13 1 12 11 6 Freeway 2 Barrier 4 Freeway 12 4 24 Nonfreeway 4 6 Freeway 1 5 4 MO Traversable 4 Freeway 56 157 Nonfreeway 7 270 17 5 2 Barrier 4 Freeway 1 81 49 NC Traversable 4 Freeway 34 5 6 Barrier 4 Freeway 25 79 180 51 16 15 Nonfreeway 7 3 7 6 Freeway 3 4 7 OH Traversable 4 Freeway 1 28 79 6 Freeway 7 17 6 Barrier 4 Freeway 2 20 6 Freeway 1 18 1 1 PA Traversable 4 Freeway 14 61 46 194 154 28 13 6 Freeway 5 1 Barrier 4 Freeway 9 40 2 3 6 1 3 6 Freeway 1 WA Traversable 4 Freeway 46 29 110 1 8 Nonfreeway 26 1 1 5 6 Freeway 1 1 3 5 1 6 Barrier 4 Freeway 1 3 1 8 1 6 Freeway 1

58 nonfreeways. Relationships for fatal-and-injury crashes fol- lowed the same trends as all crashes. The vehicle dynamics simulation analysis presented in Chapter 5 includes results that help to explain the crash analy- sis results for median width presented here. Median Slope Safety effects of median slopes are shown in Table 4-10. Each median slope effect found to be statistically significant is discussed below. The distribution of median slopes for the study sites is shown in Table 4-11. 1. Flatter slopes appear to decrease median-related crashes for both four-lane freeways and nonfreeways. The safety effects were a 2.4 percent decrease for each unit change in the median slope ratio for four-lane freeways and a 9.2 percent decrease for each unit change in median slope ratio on nonfreeways. For fatal-and-injury crashes, the corresponding safety effects were a 5.3 percent decrease on four-lane freeways and a 10.8 percent decrease for nonfreeways. 2. Flatter slopes appear to increase total median-related crashes for six-lane freeways. The safety effect was a 5.8 per- cent increase for each unit change in the median slope ratio. The safety effect for fatal-and-injury median-related crashes was a 9.2 percent increase for each unit change in median slope ratio. 3. Flatter slopes increase both CMCs plus NCMCs and CMCs on both four- and six-lane freeways. The safety effect for CMCs plus NCMCs was an 11.5 percent per unit increase in median slope ratio for four-lane freeways and a 19.8 per- cent per unit increase in median slope ratio for six-lane freeways. The safety effect for fatal-and-injury CMCs plus NCMCs on four-lane freeways was a 12.4 percent per unit increase in median slope ratio. The observed safety effect for CMCs was a 73.1 percent increase for four-lane free- ways and 15.9 percent increase for six-lane freeways. 4. For nonfreeways, flatter slopes appeared to decrease both CMCs plus NCMCs and CMCs. The safety effect was a 41.8 percent decrease per unit change for both CMCs plus NCMCs and CMCs. However, for fatal-and-injury CMCs, flatter slopes appeared to increase crashes. The safety effect was a 31.8 percent increase per unit change in the median slope ratio. 5. As expected, flatter slopes appear to decrease rollover crashes for four-lane freeways and nonfreeways. The same finding was seen for fatal-and-injury rollover crashes on four-lane freeways. The safety effect for rollover crashes was a 5.8 percent decrease per unit change in median slope ratio for four-lane freeways and a 22.5 percent decrease per unit change in median slope ratio for nonfreeways. Fatal- and-injury rollover crashes had a 3.2 percent decrease for four-lane freeways and a 21.9 percent decrease for nonfreeways. 6. On six-lane freeways, fatal-and-injury rollover crashes appeared to increase with flatter slopes. The safety effect was a 13.7 percent increase in crashes per unit change in slope ratio. 7. On four-lane freeways and nonfreeways, fixed-object crashes appeared to increase with flatter slopes. However, fatal-and-injury fixed-object crashes appeared to decrease Median-related crash type Road type Percent change in crash frequency per unit change in median slope ratio Total severity F & I severity Effect LB UB Effect LB UB All median-related crashes Four-lane freeway –2.39 –4.59 –0.13 –5.28 –9.23 –1.15 Four-lane nonfreeway –9.24 –19.59 2.45 –10.75 –22.06 2.20 Six-lane freeway 5.78 –1.87 14.03 9.15 0.06 19.07 CMCs + NCMCs Four-lane freeway 11.53 –4.48 30.44 12.42 –5.69 34.01 Four-lane nonfreeway –41.76 –69.32 10.54 Six-lane freeway 19.84 –2.62 47.50 CMCs Four-lane freeway 73.72 25.46 140.54 Four-lane nonfreeway –41.76 –69.32 10.54 31.78 –5.25 83.27 Six-lane freeway 15.88 –6.66 43.86 Rollover crashes Four-lane freeway –5.79 –10.16 –1.20 –3.23 –7.73 1.49 Four-lane nonfreeway –22.52 –33.08 –10.28 –21.88 –34.08 –7.43 Six-lane freeway 13.71 –3.32 33.74 Fixed-object crashes Four-lane freeway 15.26 –1.25 34.53 –16.76 –25.26 –7.30 Four-lane nonfreeway 4.81 –0.30 10.19 –44.87 –71.31 5.94 Six-lane freeway Other median-related crashes Four-lane freeway –5.45 –10.28 –0.37 –5.53 –11.55 0.90 Four-lane nonfreeway 6.52 1.22 12.10 Six-lane freeway Table 4-10. Safety effect of median slope ratio for roadways with traversable medians.

59 with flatter slopes. The safety effect for fixed-object crashes was a 15.3 percent increase per unit change in median slope ratio on four-lane freeways and a 4.8 percent increase per unit change in median slope ratio for nonfreeways. Fixed-object fatal-and-injury crashes showed the opposite safety effects of a 16.8 percent decrease per unit change in median slope ratio for four-lane freeways and a 44.9 per- cent decrease per unit change in median slope ratio for nonfreeways. 8. On four-lane freeways, other median-related crashes appeared to decrease with flatter slopes. Fatal-and-injury other median-related crashes also appeared to decrease with flatter slopes. The safety effects were a 5.5 percent decrease in other median-related crashes per unit change in median slope ratio and a 5.5 percent decrease in fatal- and-injury other median-related crashes per unit change in median slope ratio. 9. On nonfreeways, fatal-and-injury other median-related crashes appeared to increase with flatter slopes. The observed safety effect was a 6.5 percent increase in crashes per unit change in median slope ratio. In summary, flatter slopes were found to increase both CMCs plus NCMCs and CMCs on four- and six-lane freeways. However, flatter slopes generally decreased rollover crashes and other median-related crashes. Fixed-object crashes gen- erally increased with flatter slopes, but fatal-and-injury fixed- object crashes decreased. All median-related crashes decreased with flatter slopes on four-lane freeways and nonfreeways, but increased for six-lane freeways. Fatal-and-injury median- related crashes followed the same trend. Median Shoulder Width Safety effects of median shoulder widths are shown in Table 4-12. Each median shoulder width effect found to be statistically significant is discussed below. The distribution of inside (median) shoulder widths is shown in Table 4-13. 1. Fatal-and-injury CMCs plus NCMCs appear to decrease on four-lane freeways with wider median shoulders. The safety effect was a 19.9 percent decrease for each addi- tional foot of median shoulder width. 2. On four-lane freeways, rollover crashes and fatal-and- injury rollover crashes also decreased with wider median shoulders. The safety effects were a 6.4 percent decrease in all rollover crashes for each additional foot of median shoulder width and a 13.0 percent decrease in fatal-and- injury rollover crashes for each additional foot of median shoulder. 3. Fixed-object crashes decreased on nonfreeways and six- lane freeways with wider median shoulders. The safety State Median type No. of lanes Road type Roadway length (mi) for sites in categories of median slope ratio Missing 1:4 1:6 1:10 1:14 1:14+ CA Traversable 4 Freeway 19 2 11 28 204 1 Nonfreeway 15 3 20 6 Freeway 2 Barrier 4 Freeway 3 24 13 Nonfreeway 4 6 Freeway 1 9 MO Traversable 4 Freeway 5 97 97 12 2 Nonfreeway 18 138 126 12 7 Barrier 4 Freeway 7 61 58 4 1 NC Traversable 4 Freeway 7 2 35 1 Barrier 4 Freeway 113 4 91 158 Nonfreeway 3 14 6 Freeway 5 5 4 OH Traversable 4 Freeway 1 3 97 7 6 Freeway 28 2 Barrier 4 Freeway 2 13 7 6 Freeway 3 15 3 PA Traversable 4 Freeway 34 6 126 267 77 6 Freeway 1 4 1 Barrier 4 Freeway 47 4 12 1 6 Freeway 1 WA Traversable 4 Freeway 1 12 167 14 Nonfreeway 4 20 0 6 Freeway 1 13 3 Barrier 4 Freeway 1 3 10 6 Freeway 1 Table 4-11. Median slope ratio distribution for study sites.

60 effects were a 10.1 percent decrease on nonfreeways for each additional foot of median shoulder width and a 4.7 percent decrease on six-lane freeways for each addi- tional foot of median shoulder width. 4. On four-lane nonfreeways, wider median shoulders were found to reduce other median-related crashes. The safety effect was a 5.2 percent decrease for each additional foot of median shoulder width. The safety effect for fatal-and- injury other median-related crashes was a 10.1 percent decrease for each additional foot of median shoulder width. In summary, wider median shoulders decreased all types of crashes where significant effects were discovered. The types of crashes with statistically significant safety effects include Median-related crash type Road type Percent change in crash frequency per unit change in shoulder width inside Total severity F & I severity Effect LB UB Effect LB UB All median-related crashes Four-lane freeway Four-lane nonfreeway Six-lane freeway CMCs + NCMCs Four-lane freeway –19.90 –40.75 8.28 Four-lane nonfreeway Six-lane freeway CMCs Four-lane freeway Four-lane nonfreeway Six-lane freeway Rollover crashes Four-lane freeway –6.40 –14.28 2.22 –12.98 –20.66 –4.55 Four-lane nonfreeway Six-lane freeway Fixed-object crashes Four-lane freeway Four-lane nonfreeway –10.12 –17.86 –1.66 Six-lane freeway –4.72 –10.45 1.37 Other median-related crashes Four-lane freeway Four-lane nonfreeway –5.23 –12.21 2.30 –10.13 –18.10 –1.39 Six-lane freeway Table 4-12. Safety effect of shoulder width for roadways with traversable medians. State Median type No. of lanes Road type Roadway length (mi) for sites in categories of inside shoulder width (ft) 0 2 3 4 5 6 7 8 9 10 11 12 12+ CA Traversable 4 Freeway 190 190 2 3 64 1 3 Nonfreeway 32 1 5 6 Freeway 2 Barrier 4 Freeway 27 1 4 3 1 4 Nonfreeway 4 6 Freeway 9 1 MO Traversable 4 Freeway 30 121 55 7 Nonfreeway 15 108 28 19 46 46 14 18 5 2 Barrier 4 Freeway 29 74 22 6 NC Traversable 4 Freeway 26 6 13 Barrier 4 Freeway 12 91 117 146 Nonfreeway 10 7 6 Freeway 4 10 OH Traversable 4 Freeway 71 26 7 1 1 2 6 Freeway 6 17 7 Barrier 4 Freeway 12 8 1 1 6 Freeway 1 17 1 2 PA Traversable 4 Freeway 290 4 1 83 111 21 6 Freeway 5 1 Barrier 4 Freeway 3 9 1 4 29 16 2 6 Freeway 1 WA Traversable 4 Freeway 1 186 6 1 Nonfreeway 33 6 Freeway 1 12 2 2 Barrier 4 Freeway 1 13 6 Freeway 1 Table 4-13. Distribution of inside shoulder widths for study sites.

61 fatal-and-injury CMC plus NCMC, rollover, fixed-object, and other median-related crashes. Presence of Horizontal Curves Safety effects of the presence of horizontal curves are shown in Table 4-14. Each horizontal curve effect found to be statistically significant is discussed below. The distribution of horizontal curve presence is shown in Table 4-15. 1. On four-lane freeways, the presence of a horizontal curve appears to increase rollover and fatal-and-injury rollover crashes. The safety effects were a 18.6 percent increase in rollover crashes where a horizontal curve is present, and a Table 4-15. Distribution of horizontal curve presence for study sites. State Median type No. of lanes Road type Roadway length (mi) for sites with and without horizontal curves None Curve present CA Traversable 4 Freeway 194 69 Nonfreeway 25 13 6 Freeway 2 Barrier 4 Freeway 31 9 Nonfreeway 2 2 6 Freeway 6 4 MO Traversable 4 Freeway 272 29 Nonfreeway 272 29 Barrier 4 Freeway 123 8 NC Traversable 4 Freeway 30 15 Barrier 4 Nonfreeway 17 Freeway 297 69 6 Freeway 14 OH Traversable 4 Freeway 86 22 6 Freeway 25 5 Barrier 4 Freeway 17 5 6 Freeway 15 6 PA Traversable 4 Freeway 287 223 6 Freeway 2 4 Barrier 4 Freeway 31 33 WA Traversable 4 Freeway 164 30 Nonfreeway 20 13 6 Freeway 14 3 Barrier 4 Freeway 6 8 Median-related crash type Road type ObsUsed Percent change in crash frequency between sites No. of with and without horizontal curves Total severity F&I severity Effect LB UB Effect LB UB All median-related crashes Four-lane freeway 1,194 Four-lane nonfreeway 33 Six-lane freeway 96 CMCs + NCMCs Four-lane freeway 244 Four-lane nonfreeway 31 Six-lane freeway 90 CMCs Four-lane freeway 107 Four-lane nonfreeway 31 Six-lane freeway 90 Rollover crashes Four-lane freeway 244 18.60 –7.69 52.37 29.52 1.37 65.49 Four-lane nonfreeway 38 Six-lane freeway 66 Fixed-object crashes Four-lane freeway 61 Four-lane nonfreeway 313 Six-lane freeway 66 Other median-related crashes Four-lane freeway 244 26.24 –3.16 64.57 Four-lane nonfreeway 384 Six-lane freeway 96 Table 4-14. Safety effect of curve presence for roadways with traversable medians.

62 29.5 percent increase in fatal-and-injury rollover crashes where a horizontal curve is present. 2. The presence of a curve increased other median-related crashes on four-lane freeways. The only observed safety effect statistically significant was a 26.2 percent increase in other median-related crashes when a curve was present. In summary, few statistically significant safety effects were found for the presence of horizontal curves. Curve presence did increase rollover and other median-related crashes. Presence of On-Ramps Safety effects of the presence of on-ramps are shown in Table 4-16. Each on-ramp effect found to be statistically sig- nificant is discussed below. The distribution of ramp pres- ence is shown in Table 4-17. 1. The presence of an on-ramp appears to increase median- related crashes for all road types considered. The safety effects of the presence of an on-ramp were a 30.4 percent increase for four-lane freeways, a 46.9 percent increase for nonfreeways, and a 22.3 percent increase for six-lane freeways. 2. No statistically significant effects were found for CMC plus NCMC, CMC, or rollover crashes. 3. On freeways, the presence of an on-ramp appears to increase fixed-object crashes. The safety effect for four- lane freeways was a 139.1 percent increase in fixed-object median-related crashes when a ramp was present and the safety effect for six-lane freeways was a 56.9 percent increase in fixed-object median-related crashes when a ramp was present. Fatal-and-injury fixed-object crashes on six-lane freeways increased 92.3 percent when a ramp was present. These observed effects may result from a greater likelihood of fixed objects in the median in the vicinity of on-ramps. 4. On six-lane freeways, other median-related crashes increased when an on-ramp was present. The safety effects were a 55.7 percent increase in all other median-related crashes and a 58.1 percent increase in fatal-and-injury other median-related crashes. In summary, there were large safety effects for the presence of an on-ramp for all median-related crashes, fixed-object crashes, and other median-related crashes. No statistically significant effects were found for CMC plus NCMC, CMC, or rollover crashes. Presence of Rumble Strips Safety effects of the presence of shoulder rumble strips are shown in Table 4-18. Each rumble strip presence effect found to be statistically significant is discussed below. The distribution of shoulder rumble strip presence is shown in Table 4-19. 1. On four-lane freeways, fatal-and-injury median-related crashes decreased when shoulder rumble strips were pres- ent. The safety effect was a 12.7 percent decrease in crashes when rumble strips were present. 2. For six-lane freeways, other median-related crashes and fatal-and-injury other median-related crashes decreased when rumble strips were present. The safety effects were a 24.6 percent decrease for other median-related crashes and a 41.1 percent decrease for fatal-and-injury other median- related crashes. Table 4-16. Safety effect of on-ramp presence for roadways with traversable medians. Median-related crash type Road type Percent change in crash frequency between sites with and without on-ramps Total severity F & I severity Effect LB UB Effect LB UB All median-related crashes Four-lane freeway 30.40 14.75 48.19 Four-lane nonfreeway 46.87 –2.50 121.25 Six-lane freeway 22.29 –8.21 62.91 CMCs + NCMCs Four-lane freeway Four-lane nonfreeway Six-lane freeway CMCs Four-lane freeway Four-lane nonfreeway Six-lane freeway Rollover crashes Four-lane freeway Four-lane nonfreeway Six-lane freeway Fixed-object crashes Four-lane freeway 139.06 19.80 377.04 Four-lane nonfreeway Six-lane freeway 56.86 18.00 108.52 92.32 12.45 228.91 Other median-related crashes Four-lane freeway Four-lane nonfreeway Six-lane freeway 55.67 14.08 112.42 58.06 4.64 138.74

63 3. No statistically significant effects were found for CMC plus NCMC, CMC, rollover, or fixed-object crashes. In summary, shoulder rumble strips were found to decrease total median-related crashes, but no statistically significant effects were found for CMC plus NCMC, CMC, rollover, or fixed-object crashes. 4.3 Analysis of Medians with Barriers The objective of the safety analysis of medians with barri- ers is to develop a methodology to predict the safety perfor- mance of barrier medians, analogous to the methodology for traversable medians described in Section 4.2. The safety Table 4-17. Distribution of on-ramp presence for study sites. State Median type No. of lanes Road type Roadway length (mi) for sites with and without on-ramps Missing None On-ramp present CA Traversable 4 Freeway 65 175 23 Nonfreeway 13 25 6 Freeway 2 Barrier 4 Freeway 8 29 3 Nonfreeway 2 2 6 Freeway 3 4 3 MO Traversable 4 Freeway 182 31 Nonfreeway 270 31 Barrier 4 Freeway 107 24 NC Traversable 4 Freeway 31 14 Barrier 4 Nonfreeway 293 73 Freeway 12 5 6 Freeway 11 3 OH Traversable 4 Freeway 88 20 6 Freeway 22 8 Barrier 4 Freeway 18 4 6 Freeway 17 4 PA Traversable 4 Freeway 473 37 6 Freeway 6 Barrier 4 Freeway 58 6 6 Freeway 1 WA Traversable 4 Freeway 33 149 12 Nonfreeway 10 22 1 6 Freeway 10 7 Barrier 4 Freeway 3 11 6 Freeway 1 Table 4-18. Safety effect of rumble strip presence for roadways with traversable medians. Median-related crash type Road type Percent change in crash frequency between sites with and without shoulder rumble strips Total severity F & I severity Effect LB UB Effect LB UB All median-related crashes Four-lane freeway –12.71 –28.75 6.95 Four-lane nonfreeway Six-lane freeway CMCs + NCMCs Four-lane freeway Four-lane nonfreeway Six-lane freeway CMCs Four-lane freeway Four-lane nonfreeway Six-lane freeway Rollover crashes Four-lane freeway Four-lane nonfreeway Six-lane freeway Fixed-object crashes Four-lane freeway Four-lane nonfreeway Six-lane freeway Other median-related crashes Four-lane freeway Four-lane nonfreeway Six-lane freeway –24.58 –47.65 8.66 –41.13 –66.21 2.56

64 prediction methodology for barrier medians will combine the results from two analysis approaches: a cross-sectional analy- sis using regression models similar to the analysis approach used for traversable medians and a before-after evaluation of the installation of barriers in existing traversable medians. Typical cross sections for barrier medians are defined in terms of the road type, barrier type, and the barrier placement policy. Barriers are placed in medians primarily to prevent CMCs where a vehicle from one roadway crosses the median and collides with a vehicle on the opposing roadway. However, although the placement of a median barrier prevents most CMCs, it may increase the total number of crashes in the median due to collisions with the barrier. Some vehicles that would have stopped in the median or recovered in the median and returned to the original traveled way will instead collide with the median barrier. The increase in the number of CMCs and their high sever- ity has resulted in a substantial increase in the use of barri- ers in medians in the last 10 years. Because of the retrofit of many existing medians with barriers, especially cable barri- ers, developing a safety prediction methodology for barrier medians should be directed toward comparing the safety performance of alternative barrier types and placements and to use together with the result of the analysis of traversable medians, the analysis results to compare median cross-section designs with and without barriers. 4.3.1 Safety Measures (Dependent Variables) The safety measures for analysis of barrier medians are similar to those for traversable medians. Specific crash sever- ity levels are the same as in the analysis of traversable medi- ans, but there are additional crash types of interest. Specific median-related crash types of interest primarily in barrier medians include the following: • Hit barrier; • Went over or through barrier; hit opposing vehicle (CMC); • Went over or through barrier; entered opposing roadway but did not collide with opposing vehicle (NCMC); • Rollover; • Fixed object; and • Other median related. The median-related crash types involving vehicles that do not reach the barrier are the same for barrier medians as for traversable medians. State Median type No. of lanes Road type Roadway length (mi) for sites with and without shoulder rumble strips None Rumble strip present CA Traversable 4 Freeway 98 165 Nonfreeway 13 25 6 Freeway 2 Barrier 4 Freeway 9 31 Nonfreeway 2 2 6 Freeway 5 5 MO Traversable 4 Freeway 20 193 Nonfreeway 50 251 Barrier 4 Freeway 10 121 NC Traversable 4 Freeway 21 24 Barrier 4 Freeway 115 251 Nonfreeway 7 10 6 Freeway 7 7 OH Traversable 4 Freeway 26 82 6 Freeway 2 28 Barrier 4 Freeway 3 19 6 Freeway 1 20 PA Traversable 4 Freeway 320 190 6 Freeway 2 4 Barrier 4 Freeway 55 9 6 Freeway 1 WA Traversable 4 Freeway 23 171 Nonfreeway 10 23 6 Freeway 10 7 Barrier 4 Freeway 3 11 6 Freeway 1 Table 4-19. Distribution of rumble strip presence for study sites.

65 4.3.2 Median Cross-Section and Roadway Characteristics (Independent Variables) The median cross-section and roadway characteristics of interest are the same for barrier medians as presented for tra- versable medians in Section 4.2 except that additional char- acteristics specific to the median barrier will be added: the barrier type, the offset from the traveled way to the barrier, and the placement policy used to locate the barrier within the median. The barrier types are flexible (cable) barrier, semi- rigid (guardrail), and rigid barrier (concrete safety shaped). 4.3.3 Data Collection The data collected for barrier medians was the same as for traversable medians shown in the list provided in Section 4.2.3. The barrier type as well as the placement policy was determined from the responses to the state summary. The automated ter- rain mapping system (described in Appendix D which is avail- able on the TRB website) measured the sideslope and distance to the barrier in all states except Missouri. The specific type of cable barrier was verified by visual inspection. Participating States The participating states for barrier median sections on rural freeways are California, Missouri, North Carolina, Ohio, Pennsylvania, and Washington. The researchers did not include barrier medians on rural divided nonfreeways because there was not sufficient mileage to provide an ade- quate sample size. Field Data Collection Data collection for barrier median sites was conducted jointly with the data collection for traversable medians. The scanning laser system was used in California, North Carolina, Ohio, Pennsylvania, and Washington states. At least 5 years of crash data was obtained as well as yearly ADT data corre- sponding to the crash data. For the states of California, North Carolina, Ohio, and Washington, crash data were obtained from the FHWA HSIS system. Missouri and Pennsylvania crash data were obtained directly from the Missouri and Pennsylvania DOTs. 4.3.4 Crash Statistics for Barrier Medians The mileage of traversable and barrier median sites has been presented in Table 4-1. The total roadway length for all study sites was 3,250.8 km (2,020.4 mi). There were 1,277.4 km (792.1 mi) of roadway with barrier medians, including rural freeways. There is a limited mileage of barrier median sites on nonfreeways; the total length of rural nonfreeway sites with barrier medians was 32.8 km (20.4 mi). Because the non- freeway mileage was minimal, no analysis was made of non- freeway barrier median sites. The frequency of crashes for barrier median sites is shown in Table 4-20. This table also shows the number and propor- tion of crashes that were median-related crashes. On rural freeway barrier median sites, about 37 percent of the crashes were median related. Thus, a larger percentage of crashes are median related in barrier medians as compared to travers- able medians (37 versus 26 percent). Table 4-20 also gives the frequency and proportion of each crash type that constitutes median-related crashes on barrier medians. Hit-barrier crashes represent more than half of median- related crashes on barrier medians. There are few CMCs or NCMCs, and they total less than 1 percent of barrier median crashes. Rollover crashes vary widely from state to state, but average about 7 percent of median-related crashes. Fixed- object crashes average about 17 percent of median-related crashes, but have one reporting variation that should be noted. The Missouri crash database does not have a code Table 4-20. Crash frequency for barrier medians on rural freeways. State Total crashes Crash type Median-related crashes (percent of total crashes) Hit-barrier crashes (percent of median- related) CMCs (percent of median- related) NCMCs (percent of median- related) Rollover crashes (percent of median- related) Fixed-object crashes (percent of median- related) Other median- related crashes (percent of median-related) CA 1,732 586 (33.83) 294 (50.17) 18 (3.07) 0 (0.00) 121 (20.64) 37 (6.31) 116 (19.79) MO 2,525 687 (27.20) 0 (0.00) 14 (2.03) 14 (2.03) 67 (9.75) 451 (65.64) 141 (20.52) NC 14,110 5,954 (42.19) 3,702 (62.17) 12 (0.20) 2 (0.03) 275 (4.61) 727 (12.21) 1,236 (20.75) OH 3,085 721 (23.37) 178 (24.68) 11 (1.52) 0 (0.00) 50 (6.93) 183 (25.38) 299 (41.47) PA 418 187 (44.73) 179 (95.72) 4 (2.13) 1 (0.53) 1 (0.53) 0 (0.00) 2 (1.06) WA 699 151 (21.60) 61 (40.39) 0 (0.00) 2 (1.32) 61 (40.39) 24 (15.89) 3 (1.98) Total 22,569 8,286 (36.71) 4,414 (53.27) 59 (0.71) 19 (0.22) 575 (6.93) 1,422 (17.16) 1,797 (21.68)

66 for “hit-barrier,” so all barrier crashes are coded as fixed- object crashes. In Table 4-20, Missouri has zero hit-barrier crashes, but fixed-object crashes are more than 65 percent of median-related crashes. The fixed-object struck in many of these crashes is probably the median barrier, but the research team was unable to separate those crashes from other fixed- object crashes. The crash severity distribution for barrier medians is shown in Table 4-21. The small number of CMCs that rep- resent less than 1 percent of the median-related crashes in barrier medians represent over 13 percent of the fatal median-related crashes. Rollover crashes also are severe; they are about 7 percent of the total median-related crashes but account for over 31 percent of fatal median-related crashes. Hit-barrier and fixed-object crashes, on the other hand, represent 52 and 17 percent of total median-related crashes, but just 25 and 9 percent of fatal median-related crashes. 4.3.5 Statistical Analysis/Modeling of Crash Data for Barrier Medians Regression models, or safety performance functions (SPFs), were developed to estimate the effect of cross-section design features on barrier median safety performance as was done for traversable medians. Safety performance was estimated as a function of traffic volume, median width, and median slope, as well as four other independent variables found to have a statistically significant relationship with safety. The analysis focused on frequency and severity—total and fatal- and-injury—of the following crash types: all median-related; hit-barrier crashes; rollover in the median; hit-fixed-object in median (other than barrier); hit any fixed object; and other median-related collisions. Thus, a total of 12 dependent vari- ables were considered for 6 target crash types and 2 crash severity levels. (CMCs and NCMCs were not modeled due to the small number of these crashes occurring in barrier medians.) SPFs were developed with multiple regression, assuming a negative binomial (NB) error distribution of accident fre- quencies, using data from all traversable median sites and barrier sites for the period before barrier installation. The decision to include the before period of barrier sites was made to maximize the amount of information used to develop the functions. Separate models were developed for each state as well as combined state models for each of the following two classifications of roadways: • Four-lane freeway and • Six-lane freeway. Models were developed for three types of median barrier, as follows: • Flexible barriers (e.g., cable barriers); • Semi-rigid barriers (e.g., steel guardrail); and • Rigid barriers (e.g., concrete barriers). Two generalized linear modeling techniques were used to fit the data. The first method used a repeated measures cor- relation structure to model yearly crash counts for a site. In this method, the covariance structure, assuming compound symmetry, is estimated before final regression parameter estimates are determined by general estimating equations. Consequently, model convergence for this method is depen- dent on the covariance estimates as well as parameter esti- mates. When the model failed to converge for the covariance estimates, an alternative method was considered. In this method, yearly crash counts for a site were totaled and ADT values were averaged to create one summary record for a site. Regression parameter estimates were then directly estimated by maximum likelihood, without an additional covariance structure being estimated. Both methods also produced an estimate of the overdis- persion parameter, or the estimate for which the variance exceeds the mean. Overdispersion occurs in traffic data when a number of sites being modeled have zero accident counts, which creates variation in the data. In this research, most sites tended to have positive accident counts because they were a mile in length, particularly when accident counts were aggre- gated across multiple years. When the estimate for dispersion was very small or even slightly negative, the model was re-fit assuming a constant value. Models produced for both methods estimate per mile, per year crashes. To include this rate calculation, the first method used site length as an offset in the model while the second method used the number of years of crash data multiplied by site length as the offset. When both methods produce models, parameter estimates are nearly identical. However, confidence intervals for estimates produced with yearly crash counts tend to be wider due to yearly variation. Consequently, statistical significance of the estimates is harder to achieve. Both meth- ods were accomplished with the GENMOD procedure of SAS. The primary difference between methodologies occurred for the combined models from more than one state. The treat- ment of the state effect differed between methodologies. For the repeated measures method, the covariance structure used captured most of the variations in the data. Consequently, no additional adjustment was needed to account for the state effect. On the other hand, the summary record method treated state as a random effect in the model. The mixed model for this method was generated with the GLIMMIX procedure of SAS.

State Severity level Total accidents Crash type Median-related crashes (percent of total crashes) Hit-barrier crashes (percent of median-related crashes) CMCs (percent of median- related crashes) NCMCs (percent of median-related crashes) Rollover crashes (percent of median-related crashes) Fixed-object crashes (percent of median-related crashes) Other median- related crashes (percent of median-related crashes) CA Fatal 56 29 (51.78) 7 (24.13) 7 (24.13) 0 (0.00) 9 (31.03) 2 (6.89) 4 (13.79) Injury 692 287 (41.47) 138 (48.08) 6 (2.09) 0 (0.00) 78 (27.17) 15 (5.22) 50 (17.42) PDO 984 270 (27.43) 149 (55.18) 5 (1.85) 0 (0.00) 34 (12.59) 20 (7.40) 62 (22.96) MO Fatal 41 11 (26.82) 0 (0.00) 4 (36.36) 1 (9.09) 3 (27.27) 0 (0.00) 3 (27.27) Injury 602 143 (23.75) 0 (0.00) 8 (5.59) 5 (3.49) 34 (23.77) 46 (32.16) 50 (34.96) PDO 1,882 533 (28.32) 0 (0.00) 2 (0.37) 8 (1.50) 30 (5.62) 405 (75.98) 88 (16.51) NC Fatal 159 76 (47.79) 23 (30.26) 3 (3.94) 0 (0.00) 25 (32.89) 7 (9.21) 18 (23.68) Injury 4,300 1,900 (44.18) 930 (48.94) 4 (0.21) 0 (0.00) 174 (9.15) 218 (11.47) 574 (30.21) PDO 9,651 3,978 (41.21) 2,749 (69.10) 5 (0.12) 2 (0.05) 76 (1.91) 502 (12.61) 644 (16.18) OH Fatal 10 5 (50.00) 0 (0.00) 2 (40.00) 0 (0.00) 1 (20.00) 1 (20.00) 1 (20.00) Injury 669 246 (36.77) 59 (23.98) 5 (2.03) 0 (0.00) 37 (15.04) 51 (20.73) 94 (38.21) PDO 2,406 470 (19.53) 119 (25.31) 4 (0.85) 0 (0.00) 12 (2.55) 131 (27.87) 204 (43.40) PA Fatal 4 2 (50.00) 1 (50.00) 1 (50.00) 0 (0.00) 0 (0.00) 0 (0.00) 0 (0.00) Injury 200 103 (51.50) 19 (18.44) 2 (1.94) 1 (0.97) 0 (0.00) 0 (0.00) 81 (78.64) PDO 214 82 (38.31) 79 (96.34) 1 (1.21) 0 (0.00) 1 (1.21) 0 (0.00) 1 (1.21) WA Fatal 11 2 (18.18) 0 (0.00) 0 (0.00) 0 (0.00) 1 (50.00) 1 (50.00) 0 (0.00) Injury 249 50 (20.08) 10 (20.00) 0 (0.00) 0 (0.00) 33 (66.00) 6 (12.00) 1 (2.00) PDO 439 99 (22.55) 51 (51.51) 0 (0.00) 2 (2.02) 27 (27.27) 17 (17.17) 2 (2.02) Total Fatal 281 125 (44.48) 31 (24.80) 17 (13.60) 1 (0.80) 39 (31.20) 11 (8.80) 26 (20.80) Injury 6,712 2,729 (40.65) 1,156 (42.35) 25 (0.91) 6 (0.21) 356 (13.04) 336 (12.31) 850 (31.14) PDO 15,576 5,432 (34.87) 3,147 (57.93) 17 (0.31) 12 (0.22) 180 (3.31) 1,075 (19.79) 1,001 (18.42) Total 22,569 8,286 (36.71) 4,334 (52.30) 59 (0.71) 19 (0.22) 575 (6.93) 1,422 (17.16) 1,877 (22.65) Table 4-21. Crash severity distribution for barrier medians on rural freeways.

68 Of the independent variables summarized in the previous section, an attempt was made to incorporate as many vari- ables, in addition to ADT, in the SPF to obtain the best possible function to predict crashes at barrier median sites. ADT was included in all models, regardless of its significance, as long as its coefficient was positive. Selection of the remaining vari- ables in the model was performed by evaluating the statistical significance at the 80 percent confidence level. Additionally, the following three independent variables had the additional criterion that the resulting relationship between the charac- teristic and safety was meaningful in engineering terms: • Presence of on-ramps increases crashes, • Presence of horizontal curves adversely affects crashes, and • Presence of shoulder rumble strips decreases crashes. The modeling process often produced more than one suitable model for the purposes of this research. When this occurred, final models were selected from all possible models by the following considerations: 1. Select the model with the most significant variables. 2. When deciding between models with equal number of independent variables, select the model with the more important variables (e.g., median slope took precedence over number of ramps). 3. As part of this process, also consider variable effect size. All regression models for median-related collision types were developed to predict target crash frequencies per mile per year in the following form: ( )= exp lnADT + +…+ (6)0 1 2 2N b + b b X b Xn n where: N = predicted accident frequency per mile per year ADT = average daily traffic volume (veh/day) b0, . . . , bn = regression coefficients determined by model fitting X1, . . . , Xn = independent design characteristics The regression models developed with this approach are presented next, where each of the elements of design are discussed. Tables 4-22 and 4-23 present the final SPFs for barrier medians for the total and fatal-and-injury crash severity levels, respectively. Model coefficients, including the stan- dard error for each coefficient, as well as the overdispersion parameter for the model, are presented for each crash type and roadway type combination. Where no coefficient value is given in the tables, the coefficient was not statistically sig- nificant and should be treated as equal to zero in Equations 6 and 7. All model coefficients shown in Tables 4-22 and 4-23 are statistically significant at the 80 percent confidence level unless otherwise noted. A comment indicating the regression method and dataset used also is provided. As an example the SPF in Table 4-22 for all median-related crashes on four-lane freeways shown in the top line of the table is = + − − + +  exp –5.8329 0.7093 ln ADT 0.0047 MW 0.0185 MS 0.0341 SW 0.1087 OR (7)4FFN where: N4FF = predicted median-related crash frequency per mile, per year on four-lane freeways with flexible median barriers ADT = average daily traffic volume (veh/day) MW = median width (ft) MS = median slope ratio OR = on-ramp presence SW = inside shoulder width (ft) The signs of the coefficients indicate how the crash rate changes when changes are made in a design variable such as median width, median slope, etc. Equation 7 indicates that median-related crashes decrease as median widths increase (negative coefficient) and decrease as median slopes are flat- tened (negative coefficient). The safety effect for each design variable is the percent change in crash rate that can be expected for each unit change in a design variable. For example, the safety effect for median width is shown in Table 4-24. The top line in Table 4-24 says that for all median-related crashes on four-lane freeways with flexible median barriers, the rate decreases 0.47 percent for each additional foot of median width. The equation used to determine the safety effect is [ ]( ) −Safety effect 100 exp coefficient 1 (8)= This number gives the percent change in crash rate per median width unit increase and all other variables remain constant. Because the ultimate findings of the research were based on the results of the before-after evaluation presented later in this section, rather than on the regression models presented in Tables 4-22 and 4-23, the regression models are reviewed below in summary form, rather than in detail. Median Width In summary, wider medians were found to decrease total median-related crashes, hit-barrier crashes, fixed-object crashes, and hit-fixed-object crashes (including barrier crashes). On four-lane freeways with flexible barrier, wider medians appear to increase rollover crashes; however, for four-lane freeways with rigid barrier, wider medians decreased rollover crashes. (text continued on page 76)

Median- related crash type Road type Barrier type Model coefficient (standard error) Comment Intercept ADTa Median width Median slope ratio Shoulder width Curve presence On-ramp presence Rumble strip presence Over- dispersion parameter All median- related crashes Four- lane freeway Flexible barrier –5.8329 0.7093 –0.0047 –0.0185 0.0341 – 0.1087 – 0.4111 Yearly data (0.6651) (0.0665) (0.0014) (0.012) (0.0113) – (0.0765) – (0.0308) All states combined Semi- rigid barrier –8.7702 0.9924 –0.0054 –0.0403 – – – – 0.2622 Yearly data (1.4092) (0.1295) (0.0038) (0.021) – – – – (0.0381) All states combined Rigid barrier –10.8816 1.3769 – –0.1816 –0.2346 – – – 0.0100 Yearly data (9.3988) (0.9373) – (0.0567) (0.1099) – – – (0.0000) Washington data Six-lane freeway Flexible barrier –5.7171 0.6420 0.0580 –0.2710 0.3022 Yearly data (3.8602) (0.3770) (0.0415) (0.1250) (0.0548) All states combined Semi- rigid barrier –10.2198 1.0990 –0.0483 0.0768 All years combined (5.7299) (0.5544) (0.0323) (0.0523) California data Rigid barrier –22.6335 2.1336 –0.0234 –0.0317 0.4198 0.0100 Yearly data (0.6399) (0.0619) (0.0013) (0.0051) (0.0224) (0.0000) All states combined Hit-barrier crashes Four- lane freeway Flexible barrier –7.3188 0.5179 0.2557 –0.2607 1.0063 Yearly data (1.0030) (0.1038) (0.0208) (0.1470) (0.0833) All states combined Semi- rigid barrier –12.9287 1.3718 –0.0128 0.0639 –0.1079 0.2498 0.3771 Yearly data (2.2728) (0.2229) (0.0061) (0.0305) (0.0347) (0.1642) (0.0726) North Carolina data Rigid barrier –9.1685 0.9703 0.0551 –0.2001 0.2758 Yearly data (2.6535) (0.2532) (0.0180) (0.0577) (0.1266) All states combined Six-lane freeway Flexible barrier –19.4212 1.7994 –0.0376 0.3138 –0.1020 0.4469 0.8599 Yearly data (6.7768) (0.5761) (0.0076) (0.0642) (0.0763) (0.2242) (0.1941) All states combined Semi- rigid barrier –7.2653 0.8525 –0.0584 –0.0668 0.1797 Yearly data (5.5769) (0.5412) (0.0404) (0.0358) (0.1274) All states combined Rigid barrier –24.7498 2.4352 –0.0203 –0.0282 0.0100 Yearly data (1.0909) (0.1147) (0.0019) (0.0071) (0.0000) All states combined a Coefficient not significant at 80% level. Table 4-22. Safety performance models for all crash severity levels combined for roadways with median barriers. (continued on next page)

Median- related crash type Road type Barrier type Model coefficient (standard error) Comment Intercept ADTa Median width Median slope ratio Shoulder width Curve presence On-ramp presence Rumble strip presence Over- dispersion parameter Rollover crashes Four- lane freeway Flexible barrier –10.0023 0.8419 0.0196 –0.1494 0.3577 All years combined (2.0666) (0.1971) (0.0043) (0.0387) (0.1290) All states combined Semi- rigid barrier –9.1963 0.9591 –0.1059 0.0100 All years combined (8.5721) (0.8672) (0.0566) (0.0000) All states combined Rigid barrier –9.3545 1.1202 –0.1327 –0.3715 0.4131 Yearly data (8.0629) (0.8019) (0.0597) (0.1774) (0.4668) All states combined Six-lane freeway Flexible barrier –16.4753 1.2897 –0.0811 0.2591 0.0100 Yearly data (11.5655) (0.9859) (0.0619) (0.1490) (0.0000) All states combined Semi- rigid barrier –22.4091 1.8843 0.0100 Yearly data (6.4691) (0.5642) (0.0000) California data Rigid barrier –12.1599 0.9274 0.0100 Yearly data (6.8980) (0.6045) (0.0000) All states combined Fixed-object crashes Four- lane freeway Flexible barrier –12.5218 1.2352 –0.0046 1.8716 Yearly data (1.2021) (0.1223) (0.0029) (0.1589) All states combined Semi- rigid barrier –5.2205 0.4598 –0.0887 1.2600 Yearly data (3.8314) (0.3882) (0.0483) (0.3745) All states combined Rigid barrier –15.6971 1.4237 –0.0500 0.6306 Yearly data (4.8691) (0.4687) (0.0304) (0.9698) All states combined Six-lane freeway Flexible barrier –3.6234 0.5469 –0.0162 –0.1956 0.3674 1.0642 0.2238 All years combined (4.2008) (0.4306) (0.0121) (0.1043) (0.2338) (0.3754) (0.1569) All states combined a Coefficient not significant at 80% level. Table 4-22. (Continued)

Median- related crash type Road type Barrier type Model coefficient (standard error) Comment Intercept ADTa Median width Median slope ratio Shoulder width Curve presence On-ramp presence Rumble strip presence Over- dispersion parameter Hit-any- fixed-object crashes Four- lane freeway Flexible barrier –7.0339 0.7027 0.0413 0.0495 0.1863 0.4570 Yearly data (0.7116) (0.0678) (0.0182) (0.0126) (0.1073) (0.0450) North Carolina data Semi- rigid barrier –13.1563 1.3325 –0.0451 0.2771 0.2966 0.5180 Yearly data (1.7826) (0.1661) (0.0313) (0.1924) (0.1145) (0.0705) All states combined Rigid barrier –16.0751 1.4516 0.0926 0.3472 0.3318 Yearly data (5.7491) (0.5733) (0.0369) (0.0508) (0.2735) All states combined Six-lane freeway Flexible barrier –11.1972 1.1812 –0.0154 0.0619 0.2794 –0.4438 0.4390 Yearly data (3.7262) (0.3403) (0.0056) (0.0370) (0.1739) (0.1705) (0.0856) All states combined Semi- rigid barrier –5.0610 0.6189 –0.0503 0.0100 Yearly data (4.7205) (0.4631) (0.0374) (0.0000) California data Rigid barrier –21.9059 2.1052 –0.0236 –0.0311 0.2748 0.0100 Yearly data (0.9910) (0.0963) (0.0021) (0.0082) (0.0346) (0.0000) Other median- related crashes Four- lane freeway Flexible barrier –9.8892 0.9274 0.0061 –0.0425 0.0353 0.2943 Yearly data (0.9143) (0.0915) (0.0018) (0.0157) (0.0154) (0.0670) All states combined Semi- rigid barrier –13.6199 1.2921 –0.0032 –0.0324 0.1623 Yearly data (2.0145) (0.1969) (0.0024) (0.0180) (0.1005) All states combined Rigid barrier –17.5930 1.7248 –0.0462 –0.7422 0.0100 Yearly data (3.363) (0.3324) (0.0259) (0.3123) (0.0000) California data Six-lane freeway Flexible barrier –11.2964 0.8519 0.0148 0.1871 0.0636 Yearly data (6.6773) (0.5387) (0.0106) (0.0755) (0.1287) Ohio data Semi- rigid barrier –16.7856 1.6695 –0.0342 0.0880 Yearly data (4.7485) (0.4625) (0.0107) (0.2242) All states combined Rigid barrier –16.0186 1.3739 –0.0097 0.1168 0.3133 Yearly data (9.2481) (0.8151) (0.0075) (0.0571) (0.3409) All states combined a Coefficient not significant at 80% level. Table 4-22. (Continued)

Median-related crash type Road type Barrier type Model coefficient (standard error) Comment Intercept ADTa Median width Median slope ratio Shoulder width Curve presence On-ramp presence Rumble strip presence Over- dispersion parameter All median- related crashes Four-lane freeway Flexible barrier –14.3896 1.4652 –0.1247 0.4172 Yearly data (7.5048) (0.7175) (0.0517) (0.2084) Missouri data Semi-rigid barrier –11.1040 0.9845 0.0367 0.0656 0.2254 Yearly data (2.2380) (0.2098) (0.0225) (0.0363) (0.0725) All states combined Rigid barrier –13.8073 1.3157 0.0358 –0.0657 0.0100 Yearly data (2.5907) (0.2362) (0.0148) (0.0446) (0.0000) All states combined Six-lane freeway Flexible barrier –10.4765 1.0742 –0.1168 0.2763 0.0157 Yearly data (3.4162) (0.3481) (0.0845) (0.2108) (0.1207) Ohio data Semi-rigid barrier –10.7386 1.1117 –0.0645 0.1061 All years combined (7.3769) (0.7146) (0.0419) (0.0941) California data Rigid barrier –13.3590 1.2615 0.2243 (4.8250) (0.4247) (0.1502) Hit-barrier crashes Four-lane freeway Flexible barrier –8.3364 0.6959 –0.0069 0.0668 0.4128 All years combined (1.0014) (0.0960) (0.0030) (0.0283) (0.1366) North Carolina data Semi-rigid barrier –10.5963 0.9739 –0.0139 0.0593 0.5638 Yearly data (2.7821) (0.2551) (0.0085) (0.0431) (0.1800) All states combined Rigid barrier –17.1734 1.6936 –0.0241 0.0494 –0.1212 0.0100 Yearly data (3.3608) (0.3482) (0.0174) (0.0159) (0.0770) (0.0000) All states combined Six-lane freeway Flexible barrier –20.4821 1.6574 –0.0182 0.2334 0.6483 Yearly data (5.3176) (0.4924) (0.0087) (0.0794) (0.3901) All states combined Semi-rigid barrier –9.9114 1.1557 –0.0208 –0.0739 0.0241 Yearly data (7.7448) (0.7506) (0.0124) (0.0258) (0.1872) All states combined Rigid barrier –18.8340 1.7143 0.0944 (4.6550) (0.4010) (0.1587) a Not significant at 20% level. Table 4-23. Safety performance models for fatal-and-injury crashes for roadways with median barriers.

Median-related crash type Road type Barrier type Model coefficient (standard error) Comment Intercept ADTa Median width Median slope ratio Shoulder width Curve presence On-ramp presence Rumble strip presence Over- dispersion parameter Rollover crashes Four-lane freeway Flexible barrier –6.8911 0.5274 0.0170 –0.1532 0.2655 All years combined (2.2656) (0.2182) (0.0046) (0.0410) (0.1643) All states combined Semi-rigid barrier –4.4884 0.6264 –0.1332 –0.2433 0.4432 Yearly data (6.0381) (0.5976) (0.0462) (0.0432) (0.3828) Washington data Rigid barrier –8.1647 0.9959 –0.4888 0.0100 All years combined (35.3887) (3.5398) (0.3052) (0.0000) Washington data Six-lane freeway Flexible barrier –7.0493 0.6549 –0.1753 0.0346 All years combined (6.277) (0.5985) (0.0883) (0.1725) All states combined Semi-rigid barrier –27.5126 2.2787 0.5836 Yearly data (8.9451) (0.7853) (1.0540) All states combined Rigid barrier –14.4923 1.1226 0.0100 Yearly data (7.9921) (0.6991) (0.0000) All states combined Fixed-object crashes Four-lane freeway Flexible barrier –13.2253 1.2809 –0.2008 1.3065 Yearly data (2.4501) (0.2335) (0.0300) (0.4154) All states combined Semi-rigid barrier –9.2486 0.7475 –0.1525 0.2822 All years combined (5.9057) (0.5507) (0.0775) (0.3841) North Carolina data Rigid barrier –18.1491 1.4580 0.1765 All years combined (6.7195) (0.6374) (0.5148) California data Six-lane freeway Flexible barrier –8.9646 0.7097 0.0100 Yearly data (4.7498) (0.4294) (0.0000) Washington, Ohio data Semi-rigid barrier –21.1554 1.6562 9.2433 Yearly data (9.7736) (0.8580) (8.9500) All states combined Rigid barrier –3.3447 0.2188 –0.0379 5.0976 Yearly data (16.3475) (1.5036) (0.0140) (7.1034) All states combined a Not significant at 20% level. Table 4-23. (Continued) (continued on next page)

Median-related crash type Road type Barrier type Model coefficient (standard error) Comment Intercept ADTa Median width Median slope ratio Shoulder width Curve presence On-ramp presence Rumble strip presence Over- dispersion parameter Hit-any-fixed- object crashes Four-lane freeway Flexible barrier –8.5609 0.7349 –0.0047 0.0558 0.4952 Yearly data (0.9977) (0.0993) (0.0027) (0.0172) (0.1135) All states combined Semi-rigid barrier –14.1843 1.3338 –0.0382 0.6337 Yearly data (2.3132) (0.2112) (0.0278) (0.1557) All states combined Rigid barrier –16.9305 1.4933 0.0604 0.0100 Yearly data (2.4229) (0.2366) 0.0184 0.0000 All states combined Six-lane freeway Flexible barrier –11.3440 0.9409 0.0703 0.4176 –0.2677 0.0970 All years combined (4.5754) (0.4274) (0.0544) (0.2630) (0.2136) (0.0940) All states combined Semi-rigid barrier –7.9611 0.8718 –0.0887 0.3199 All years combined (10.1298) (0.9815) (0.0590) (0.1998) California data Rigid barrier –13.2134 1.3236 –0.0336 0.2214 (11.2166) (1.0773) (0.0225) (0.2998) Other median- related crashes Four-lane freeway Flexible barrier –17.0825 1.4361 –0.0462 0.1402 0.0428 All years combined (2.6603) (0.2644) (0.0302) (0.0493) (0.1608) Missouri data Semi-rigid barrier –11.8700 0.9536 0.0348 0.3767 Yearly data (8.1875) (0.7653) (0.0144) (0.4118) All states combined Rigid barrier –23.5437 2.1624 –0.0283 0.0100 Yearly data (9.8672) (0.9150) (0.0210) (0.0000) North Carolina data Six-lane freeway Flexible barrier –35.1204 3.1728 –0.0277 0.5139 (10.9298) (0.9841) (0.0055) (0.7632) Semi-rigid barrier –14.3896 1.4652 –0.1247 0.4172 Yearly data (7.5048) (0.7175) (0.0517) (0.2084) California data Rigid barrier –11.1040 0.9845 0.0367 0.0656 0.2254 Yearly data (2.2380) (0.2098) (0.0225) (0.0363) (0.0725) All states combined a Not significant at 20% level. Table 4-23. (Continued)

75 Median- related crash type Road type Barrier type Percent change in crash frequency per unit change in median width Total severity F & I severity Effect LB UB Effect LB UB All median- related crashes Four-lane freeway Flexible barrier –0.47 –0.75 –0.19 Semi-rigid barrier –0.53 –1.27 0.20 Rigid barrier Six-lane freeway Flexible barrier Semi-rigid barrier Rigid barrier –2.31 –2.56 –2.07 Hit-barrier crashes Four-lane freeway Flexible barrier –0.69 Semi-rigid barrier –1.27 –2.44 –0.09 –1.39 Rigid barrier –2.38 Six-lane freeway Flexible barrier –3.69 –5.12 –2.23 –1.80 Semi-rigid barrier –2.06 Rigid barrier –2.01 –2.37 –1.65 Rollover crashes Four-lane freeway Flexible barrier 1.98 1.12 2.86 1.71 Semi-rigid barrier Rigid barrier Six-lane freeway Flexible barrier Semi-rigid barrier Rigid barrier Fixed-object crashes Four-lane freeway Flexible barrier –0.46 –1.02 0.10 Semi-rigid barrier Rigid barrier Six-lane freeway Flexible barrier –1.61 –3.92 0.76 Semi-rigid barrier Rigid barrier –3.72 Hit-any-fixed- object crashes Four-lane freeway Flexible barrier –0.47 Semi-rigid barrier Rigid barrier Six-lane freeway Flexible barrier –1.53 –2.60 –0.45 Semi-rigid barrier Rigid barrier –2.33 –2.74 –1.92 Other median- related crashes Four-lane freeway Flexible barrier 0.62 0.27 0.96 –3.31 Semi-rigid barrier –0.32 –0.79 0.15 Rigid barrier 3.54 Six-lane freeway Flexible barrier 1.49 –0.61 3.63 Semi-rigid barrier –3.37 –5.37 –1.32 –2.79 Rigid barrier –0.96 –2.40 0.50 –2.73 –1.28 –3.02 –5.66 –3.45 –4.41 0.80 –6.33 –1.00 –7.48 0.66 –6.71 –3.78 –0.09 0.27 1.01 –0.12 0.36 2.64 –1.03 0.06 1.06 6.50 1.28 –1.68 Table 4-24. Safety effect of median width for roadways with median barriers.

76 Table 4-25. Safety effect of median slope ratio for roadways with median barriers. Median- related crash type Road type Barrier type Percent change in crash frequency per unit change in median slope ratio Total severity F & I severity Effect LB UB Effect LB UB All median- related crashes Four-lane freeway Flexible barrier –1.64 –3.91 0.69 –11.72 –20.23 –2.30 Semi-rigid barrier –3.95 –7.83 0.09 3.74 –0.74 8.42 Rigid barrier –16.60 –25.38 –6.79 3.65 0.68 6.70 Six-lane freeway Flexible barrier 5.98 –2.30 14.96 –11.02 –24.60 5.00 Semi-rigid barrier –4.71 –10.55 1.51 –6.25 –13.65 1.78 Rigid barrier –3.12 –4.09 –2.14 Hit-barrier crashes Four-lane freeway Flexible barrier 29.14 23.97 34.52 6.91 1.14 13.01 Semi-rigid barrier 6.60 0.42 13.16 6.11 –2.49 15.47 Rigid barrier 5.67 2.01 9.46 5.06 1.84 8.39 Six-lane freeway Flexible barrier 36.86 20.67 55.22 26.29 8.08 47.57 Semi-rigid barrier –5.67 –12.85 2.09 –7.13 –11.71 –2.31 Rigid barrier –2.78 –4.12 –1.42 Rollover crashes Four-lane freeway Flexible barrier –13.88 –20.19 –7.06 –14.20 –20.84 –7.00 Semi-rigid barrier –10.05 –19.49 0.49 –12.47 –20.04 –4.18 Rigid barrier –12.42 –22.62 –0.88 –38.66 –66.28 11.55 Six-lane freeway Flexible barrier –7.79 –18.32 4.10 –16.08 –29.67 0.13 Semi-rigid barrier Rigid barrier Fixed-object crashes Four-lane freeway Flexible barrier –18.19 –22.87 –13.23 Semi-rigid barrier –8.49 –16.75 0.60 –14.15 –26.24 –0.07 Rigid barrier –4.87 –10.38 0.97 Six-lane freeway Flexible barrier –17.76 –32.97 0.89 Semi-rigid barrier Rigid barrier Hit-any-fixed- object crashes Four-lane freeway Flexible barrier 4.22 0.56 8.01 5.74 2.23 9.37 Semi-rigid barrier –4.41 –10.10 1.65 –3.75 –8.85 1.63 Rigid barrier 9.70 2.05 17.93 6.23 2.47 10.12 Six-lane freeway Flexible barrier 6.38 –1.06 14.39 7.28 –3.78 19.62 Semi-rigid barrier –4.90 –11.63 2.34 –8.49 –18.47 2.72 Rigid barrier –3.06 –4.61 –1.49 Other median- related crashes Four-lane freeway Flexible barrier –4.16 –7.06 –1.16 Semi-rigid barrier –3.19 –6.55 0.29 –4.52 –10.01 1.31 Rigid barrier –4.51 –9.25 0.47 Six-lane freeway Flexible barrier Semi-rigid barrier Rigid barrier Results were mixed for other median-related crash types. Fatal-and-injury crashes followed the same trends. Median Slope Safety effects of median slopes are summarized in Table 4-25. In summary, flatter slopes appear to reduce the number of roll- over and fixed-object crashes, but increase hit-barrier crashes. The safety effects for hit-barrier crashes are larger for flexible barriers, but are generally uniform for all barrier types for roll- over crashes. Hit-fixed-object crashes (including hit-barrier crashes) increased with flatter slopes where flexible barriers were present, but decreased where semi-rigid and rigid barriers were in place. Total median-related crashes generally declined with flat- ter slopes, and more severe fatal-and-injury median-related crashes generally decreased with flatter slopes. Median Shoulder Width Safety effects for median shoulder width statistically signifi- cant are shown in Table 4-26. In summary, wider median shoul- ders generally appear to decrease the number of hit-barrier crashes. Rollover crashes decreased with wider median shoul- ders on four-lane freeways but increased on six-lane freeways. (text continued from page 68)

77 The results for fixed-object crashes (including hit-barrier crashes) were mixed, but the majority of results showed increased crashes with wider shoulders. Overall, on four-lane freeways, median-related crashes appeared to increase with wider shoulders for flexible and semi-rigid barriers, but decreased for rigid barriers. On six-lane freeways with rigid barrier, median-related crashes increased with wider median shoulders. Presence of Horizontal Curves Safety effects for presence of horizontal curves that are statistically significant are shown in Table 4-27. In summary, horizontal curve presence was associated with increased rollover and hit-fixed-object crashes (including hit-barrier crashes). Safety effects were more often significant at sites with flexible barriers. Presence of On-Ramps Safety effects of on-ramp presence that are statistically significant are shown in Table 4-28. In summary, the pres- ence of an on-ramp generally increases crashes where sta- tistically significant relationships were found. The amount of increase due to the presence of an on-ramp ranged from 11 to 189 percent. Table 4-26. Safety effect of inside shoulder width for roadways with median barriers. Median- related crash type Road type Barrier type Percent change in crash frequency per unit change in inside shoulder width Total severity F & I severity Effect LB UB Effect LB UB All median- related crashes Four-lane freeway Flexible barrier 3.47 1.19 5.79 Semi-rigid barrier 6.78 –0.56 14.66 Rigid barrier –20.92 –36.24 –1.91 –6.35 –14.20 2.21 Six-lane freeway Flexible barrier Semi-rigid barrier Rigid barrier 52.17 45.64 59.00 Hit-barrier crashes Four-lane freeway Flexible barrier Semi-rigid barrier –10.23 –16.13 –3.92 Rigid barrier –18.14 –26.89 –8.34 –11.41 –23.82 3.01 Six-lane freeway Flexible barrier –9.69 –22.24 4.87 Semi-rigid barrier –6.46 –12.80 0.33 Rigid barrier Rollover crashes Four-lane freeway Flexible barrier Semi-rigid barrier –21.59 –27.95 –14.67 Rigid barrier –31.03 –52.27 –0.35 Six-lane freeway Flexible barrier 29.58 –3.24 73.53 Semi-rigid barrier Rigid barrier Fixed-object crashes Four-lane freeway Flexible barrier Semi-rigid barrier Rigid barrier Six-lane freeway Flexible barrier Semi-rigid barrier Rigid barrier Hit-any-fixed- object crashes Four-lane freeway Flexible barrier 5.08 2.51 7.71 Semi-rigid barrier Rigid barrier 41.51 28.10 56.32 Six-lane freeway Flexible barrier Semi-rigid barrier Rigid barrier 31.63 23.00 40.87 Other median- related crashes Four-lane freeway Flexible barrier 3.60 0.52 6.77 Semi-rigid barrier 15.05 4.45 26.72 Rigid barrier Six-lane freeway Flexible barrier 20.57 3.98 39.81 Semi-rigid barrier Rigid barrier 12.39 0.49 25.70

78 Median- related crash type Road type Barrier type Percent change in crash frequency between sites with and without horizontal curves Total severity F & I severity Effect LB UB Effect LB UB All median- related crashes Four-lane freeway Flexible barrier Semi-rigid barrier Rigid barrier Six-lane freeway Flexible barrier Semi-rigid barrier Rigid barrier Hit-barrier crashes Four-lane freeway Flexible barrier Semi-rigid barrier 28.38 –6.94 77.11 Rigid barrier Six-lane freeway Flexible barrier Semi-rigid barrier Rigid barrier Rollover crashes Four-lane freeway Flexible barrier Semi-rigid barrier Rigid barrier Six-lane freeway Flexible barrier Semi-rigid barrier Rigid barrier Fixed-object crashes Four-lane freeway Flexible barrier Semi-rigid barrier Rigid barrier Six-lane freeway Flexible barrier 44.40 –8.69 128.35 Semi-rigid barrier Rigid barrier Hit-any-fixed- object crashes Four-lane freeway Flexible barrier 20.48 –2.37 48.69 Semi-rigid barrier 31.93 –9.51 92.35 Rigid barrier Six-lane freeway Flexible barrier 51.83 –10.28 156.94 Semi-rigid barrier Rigid barrier Other median- related crashes Four-lane freeway Flexible barrier Semi-rigid barrier Rigid barrier Six-lane freeway Flexible barrier Semi-rigid barrier Rigid barrier Table 4-27. Safety effect of curve presence for roadways with median barriers.

79 Median- related crash type Road type Barrier type Percent change in crash frequency between sites with and without on-ramps Total severity F & I severity Effect LB UB Effect FB UB All median- related crashes Four-lane freeway Flexible barrier 11.48 –4.03 29.51 Semi-rigid barrier Rigid barrier Six-lane freeway Flexible barrier 31.82 –12.79 99.25 Semi-rigid barrier Rigid barrier Hit-barrier crashes Four-lane freeway Flexible barrier Semi-rigid barrier Rigid barrier Six-lane freeway Flexible barrier 56.35 0.75 142.63 Semi-rigid barrier Rigid barrier Rollover crashes Four-lane freeway Flexible barrier Semi-rigid barrier Rigid barrier Six-lane freeway Flexible barrier Semi-rigid barrier Rigid barrier Fixed-object crashes Four-lane freeway Flexible barrier Semi-rigid barrier Rigid barrier Six-lane freeway Flexible barrier 189.84 38.86 504.98 Semi-rigid barrier Rigid barrier Hit-any-fixed- object crashes Four-lane freeway Flexible barrier Semi-rigid barrier 34.53 7.48 68.38 Rigid barrier Six-lane freeway Flexible barrier 32.23 –5.96 85.92 Semi-rigid barrier Rigid barrier Other median- related crashes Four-lane freeway Flexible barrier Semi-rigid barrier Rigid barrier Six-lane freeway Flexible barrier Semi-rigid barrier Rigid barrier Table 4-28. Safety effect of on-ramp presence for roadways with median barriers. Presence of Shoulder Rumble Strips Safety effects for the presence of shoulder rumble strips found to be statistically significant are shown in Table 4-29. In summary, the presence of rumble strips appeared to reduce hit-barrier and hit-fixed-object crashes (including hit-barrier crashes). No significant safety effects were observed for roll- over crashes. 4.3.6 Before-After Evaluation of Median Barrier Installation An alternative analysis of barrier median crashes was con- ducted with observational before-after evaluations, com- parisons of safety performance before and after place ment of barriers in existing medians. Not all sites in the data base of barrier medians were usable for this analysis because the construction date when barrier was added to the median was not known in all cases. Sites were used in the states of Missouri, North Carolina, and Ohio. In Missouri, all sites included in the analysis had flexible barrier installed in the median and the years of construction were 2004, 2005, or 2006. In North Carolina, barriers were installed in the years of 1998 through 2001, and in Ohio barrier was installed in 2004. The before-after evaluations were conducted using the Empirical Bayes (EB) method. This method compensates for the potential bias due to regression to the mean. Regression

80 Median- related crash type Road type Barrier type Percent change in crash frequency between sites with and without shoulder rumble strips Total severity F&I severity Effect LB UB Effect LB UB All median- related crashes Four-lane freeway Flexible barrier Semi-rigid barrier Rigid barrier Six-lane freeway Flexible barrier –23.74 –40.30 –2.58 Semi-rigid barrier Rigid barrier Hit-barrier crashes Four-lane freeway Flexible barrier –22.95 –42.23 2.76 Semi-rigid barrier Rigid barrier Six-lane freeway Flexible barrier Semi-rigid barrier Rigid barrier Rollover crashes Four-lane freeway Flexible barrier Semi-rigid barrier Rigid barrier Six-lane freeway Flexible barrier Semi-rigid barrier Rigid barrier Fixed-object crashes Four-lane freeway Flexible barrier Semi-rigid barrier Rigid barrier Six-lane freeway Flexible barrier Semi-rigid barrier Rigid barrier Hit-any-fixed- object crashes Four-lane freeway Flexible barrier Semi-rigid barrier Rigid barrier Six-lane freeway Flexible barrier –35.84 –54.07 –10.38 –23.49 –50.09 17.30 Semi-rigid barrier Rigid barrier Other median- related crashes Four-lane freeway Flexible barrier Semi-rigid barrier Rigid barrier –52.39 –74.19 –12.19 Six-lane freeway Flexible barrier Semi-rigid barrier Rigid barrier Table 4-29. Safety effect of rumble strip presence for roadways with median barriers.

81 Road type Median-related crash type Barrier type installed Sites Percent decrease (increase) in crashes after barrier installation Standard Error Significance at 5% level Four-lane freeway All median-related crashes All 302 (225) 10.55 Yes Flexible 238 (277) 14.94 Yes Semi-rigid 63 (152) 14.09 Yes Rigid 1 (140) 68.21 * Six-lane freeway All median-related crashes Flexible 25 66 14.37 Yes Four-lane freeway CMCs All 302 97 1.18 Yes Flexible 238 96 1.44 Yes Semi-rigid 63 98 1.83 Yes Rigid 1 100 - * Six-lane freeway CMCs Flexible 25 73 27.27 Yes Four-lane freeway CMCs + NCMCs All 302 69 6.71 Yes Flexible 238 55 10.66 Yes Semi-rigid 63 89 6.13 Yes Rigid 1 100 - * Six-lane freeway CMCs + NCMCs Flexible 25 74 25.9 Yes Four-lane freeway Rollover crashes All 302 79 2.23 Yes Flexible 238 74 3.24 Yes Semi-rigid 63 88 2.71 Yes Rigid 1 100 - * Six-lane freeway Rollover crashes Flexible 25 23 31.53 No Four-lane freeway Hit-fixed-object crashes (including hit-barrier crashes) All 302 (604) 42.66 Yes Flexible 238 (720) 62.61 Yes Semi-rigid 63 (426) 52.64 Yes Rigid 1 (892) 465.36 * Six-lane freeway Hit-fixed-object crashes Flexible 25 (209) 31.17 Yes Four-lane freeway Other median-related crashes All 302 (103) 9.66 Yes Flexible 238 (128) 14.11 Yes Semi-rigid 63 (72) 13.01 Yes Rigid 1 (40) 74.80 * Six-lane freeway Other median-related crashes Flexible 25 16 15.45 No * Not stated due to small sample size. Table 4-30. Crash reduction factors: percent decrease (increase) in median-related crashes after the installation of median barrier. to the mean occurs when a countermeasure is installed at a site with high short-term crash experience such that the crash experience would have decreased in the after study period whether the countermeasure had been implemented or not. Before-after evaluations must be carefully structured to assure that the effect of regression to the mean is not mis- taken for an effect of the countermeasure. The EB method controls regression to the mean by com- bining the results from predictive models, such as nega- tive binomial (NB) regression, with actual observed crash frequencies for specific sites. The predicted and observed values are combined in a weighting procedure, where the weights are based on the overdispersion parameter from the NB regression model. The NB models used in these steps were developed from a group of reference sites similar to the before-period condition at the sites where barrier medi- ans are installed. The safety prediction models for travers- able medians whose development is described in Section 4.2 served this purpose. The primary output from this type of evaluation is an estimate of the overall safety effectiveness of the barrier installation, usually expressed as a percent change in crash frequency. The resulting percent increase or decrease in total median-related crashes after the installation of median bar- rier, along with their standard errors, are shown in Table 4-30. All median-related crashes increased after the installation of barrier. The increase was greatest for sites where flexible bar- rier was installed, having a 277 percent increase in the num- ber of crashes. The analysis of specific crash types indicates that CMC and NCMC were virtually eliminated, however hit-fixed-object crashes increased six to nine times. Fixed-object crashes have a reporting oddity that should be noted. The Missouri crash data- base does not have a code for “hit-barrier,” so all barrier crashes are coded as fixed-object crashes. Therefore, in Table 4-30 the category for hit-fixed-object crashes includes hit-barrier crashes. Rollover crashes decreased after installation of median barrier. The decrease in rollover crashes was nearly 80 percent on four-lane freeways and 23 percent on six-lane freeways. It is evident that a median barrier eliminates cross-median crashes and decreases rollover at the cost of greatly increased fixed- object crashes. The percent increase or decrease in fatal-and-injury crashes after the installation of median barrier is shown in Table 4-31.

82 All median-related fatal-and-injury crashes increased by 55 percent after the installation of median barrier. Cross- median crashes were virtually eliminated and rollover crashes were decreased by nearly 70 percent on four-lane freeways and 30 percent on six-lane freeways. Fatal-and- injury hit-fixed-object crashes more than doubled. Road type Median-related crash type Barrier type installed Sites Percent decrease (increase) in crashes after barrier installation Standard Error Significance at 5% level Four-lane freeway All median-related crashes All 302 (55) 6.24 Yes Flexible 238 (60) 8.60 Yes Semi-rigid 63 (50) 9.15 Yes Rigid 1 (8) 51.29 * Six-lane freeway All median-related crashes Flexible 25 (17) 20.07 No Four-lane freeway CMCs All 302 96 1.81 Yes Flexible 238 92 3.43 Yes Semi-rigid 63 100 - * Rigid 1 100 - * Six-lane freeway CMCs Flexible 25 69 31.20 Yes Four-lane freeway CMCs + NCMCs All 302 67 8.20 Yes Flexible 238 62 10.11 Yes Semi-rigid 63 83 12.02 Yes Rigid 1 100 - * Six-lane freeway CMCs + NCMCs Flexible 25 69 31.20 Yes Four-lane freeway Rollover crashes All 302 61 4.98 Yes Flexible 238 57 6.35 Yes Semi-rigid 63 69 7.77 Yes Rigid 1 100 - * Six-lane freeway Rollover crashes Flexible 25 31 34.71 No Four-lane freeway Hit-fixed-object crashes (including hit-barrier crashes) All 302 (123) 14.84 Yes Flexible 238 (132) 19.82 Yes Semi-rigid 63 (113) 22.17 Yes Rigid 1 (23) 85.97 * Six-lane freeway Hit-fixed-object crashes Flexible 25 (128) 53.78 Yes Four-lane freeway Other median-related crashes All 302 (57) 9.21 Yes Flexible 238 (67) 12.80 Yes Semi-rigid 63 (47) 13.38 Yes Rigid 1 14 62.78 * Six-lane freeway Other median-related crashes Flexible 25 14 26.11 No * Not stated due to small sample size. Table 4-31. Crash reduction factors: percent decrease (increase) in fatal-and-injury median-related crashes after installation of median barrier. The interpretation of these results is strongly influenced by the findings (based on Tables 4-4 and 4-21) that CMCs, in particular, are much more severe than other crash types, including rollover and fixed-object crashes. The application of the before-after evaluation results is addressed in Chapter 6 of this report.