Factors Contributing to Median Encroachments and Cross-Median Crashes (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.

6S E C T I O N 2 Past research on median safety has investigated the factors that caused vehicle encroachments, median crash frequency or severity, and countermeasures to prevent median encroach- ments. This section of the report summarizes the history of median safety research by reviewing past encroachment and crash studies, potential countermeasures, human fac- tors research related to median encroachments, and scanning tours aimed at identifying median encroachment contributing factors. 2.1 Median Encroachments A median encroachment is any vehicle maneuver in which all or a portion of a vehicle unintentionally crosses the edge line on the left side of the traveled way for one roadway of a divided highway and enters the median area. Median encroachments range from momentary excursions into the median area and returns to the roadway, to maneuvers in which a vehicle leaves the roadway and then comes to rest or overturns within the median area, to maneuvers in which a vehicle collides with an object in the median area. Some median encroachment maneuvers extend so far that the vehi- cle enters or crosses the roadway in the opposing direction of travel. The total frequency of median encroachments is very dif- ficult to measure. One of the founding documents of roadside safety, the Hutchison and Kennedy study (2), estimated road- side encroachment frequency by periodic monitoring of fresh vehicle tracks in the snow on rural freeway medians in Illinois during winter months and assessed which tracks appeared to be attributable to uncontrolled entries into the median. There have been other studies over the years that have attempted to quantify roadside encroachment rates (3, 4), but most such studies have been limited in scope because of the very low fre- quency of encroachments and the resulting very high cost of observing a sufficient number of encroachments to reliably estimate their frequency. A limitation of these encroachment studies has been the difficulty in deciding whether observed encroachments were accidental or deliberate. Early median safety studies sought to determine and quantify factors that caused vehicle encroachments into the median area on divided highways. In the early 1960s, Hutchinson and Kennedy (2) studied vehicle encroach- ments along I-74 and the Kingery Expressway (I-57) in Illinois. Each facility was a four-lane divided highway. I-74 had a depressed median width of 40 feet, while the Kingery Expressway had a depressed median width of only 18 feet. After 6 years of data collection, four relationships were observed, each containing ADT as one-half of the rela- tion. One of the relationships examined was ADT versus encroachment rate, which is based on encroachments per vehicle-mile traveled. It was shown that for ADT volumes of 4,000 vehicles per day and less, the encroachment rate was stable and slightly above 400 encroachments per 100 million vehicle-miles traveled (MVMT). As the ADT increased from 4,000 to 5,000 vehicles per day, there was a sharp decline in the encroachment rate to approximately 150 encroach- ments per 100 MVMT. As the ADT volume continued to increase, the encroachment rate then stayed relatively con- stant at 150 encroachments per 100 million vehicle-miles traveled. Figure 2-1 shows the relationship between ADT and encroachment rates. The driving environment was considered a primary reason for the fluctuation in encroachment rates in relation to traf- fic volumes. At low traffic volumes, drivers are less attentive. There is more freedom of movement within the travel lanes and the only restrictions are the physical features of the road- way (2). Therefore, it is likely that vehicles tend to sway off the traveled way and eventually into the median area. As traffic volumes increase, driver alertness also increases and the per- centage of “lateral veering” vehicles is greatly reduced because of the decreased vehicle spacing within the traffic stream. In addition, with the presence of other vehicles, a “follow-the- leader” phenomenon results in which vehicles farther back in Review of Median Safety Studies

7 the traffic stream tend to position in the same vehicle path as those farther downstream. A 1980 study by Cooper (3) attempted to build on the research by Hutchinson and Kennedy in determining the factors involved in roadside encroachments and their effect on the severity of the encroachment using Canadian pro- vincial data. Data were collected over a 5-month period in 5 provinces on 59 roadway sections with lengths between 37 and 62 miles. The author stated that transitory evidence (e.g., vehicle tracks off the roadway) was used in collect- ing the field data that supplemented police crash records so that unreported encroachments might be considered in the modeling. The factors that were considered included posted speeds, traffic volumes, horizontal alignment, vertical align- ment, departure direction, ditch slope, shoulder type, and shoulder width. Since the information was not available in the raw data, the author did have to make an assumption of the roadway type based on the posted speed limit. Roads with speed limits less than or equal to 90 km/h (56 mph) were assumed to be two-lane roads; divided highways and freeways were represented by roads with speed limits greater than 90 km/h (56 mph). During the analysis, there was no strong correlation that could ever be drawn between any geo- metric feature (or series of geometric features) and roadside encroachments. Some other interesting conclusions drawn by the author include the following: • Since only 25 to 30 percent of the roadside encroachment rate variance could be explained by the factors included in the study, the author postulated that external factors to this study (i.e., weather, light conditions, and road surface characteristics) could play a part in explaining a large per- centage of roadside encroachments. • The highest encroachment rates occurred with mid-ranged shoulder widths. • Vertical alignment played a larger role in contributing to roadside encroachments than did horizontal alignment. • Higher speed roadways experienced higher encroach- ment rates. • Although the analysis yielded an encroachment rate vs. vehicle volume graph similar to Hutchinson and Kennedy (Figure 2-1), the author was skeptical of the relationship’s accuracy and hypothesized that the effect was caused by a flawed data reduction technique. An analysis conducted by Davis and Morris (5) reexamined the data from the Hutchinson and Kennedy study using sta- tistical tools that were not available to the original authors to determine if the conclusions drawn by the authors were accurate with regard to encroachment rates vs. ADT relation- ships. Of particular interest was the unanticipated decrease in encroachments that was originally modeled between ADT values of 4,000 and 5,000 vehicles per day. Davis and Morris hypothesized that other factors could have played a part in the encroachment rates, including winter weather and roadway maturation, which would account for the anomaly in the data. The authors used the 10 observation periods originally analyzed, as shown in Table 2-1, and plotted the data including the standard errors as shown in Figure 2-2. This illustration is similar to the relationship demonstrated in Figure 2-1, except that the standard errors were included to illustrate the confidence bounds for the true encroachment rate. The standard error bars indicate whether the true rates are the same between observation periods. Of concern were the high encroachment rates observed in Periods 1 through 3 at lower ADT values that did not appear to be in sync with the remainder of the study. Because these data were collected along a newer freeway (I-74), there is concern that drivers may have been encroaching on the median as they adjusted to a new driving environment. An incremental increase of Figure 2-1. Encroachment rate for Interstate 74 and Kingery Expressway (2).

80.46 encroachments per MVMT was derived for winter weather conditions and attributed to the effect that pavement conditions have on driver behavior. The authors concluded that while there is a strong relationship between encroach- ments and ADT, Periods 1 through 3 are not recommended for consideration and that encroachments can be expected to increase during winter months. The results of this research would substantially lower the expected encroachments on lower volume roads and place more emphasis on encroach- ments occurring on roadways with ADT values above 5,000. Another relationship studied by Hutchinson and Kennedy related ADT to the average encroachment angle. As ADT increased from 2,000 to 6,000 vehicles per day, the degree of angle encroachment also increased from 9 to 14 degrees. Figure 2-3 shows the relationship between ADT and aver- age encroachment angle. The theory behind these observa- tions is that there is an increase in vehicle conflict as traffic volumes increase. As a result, a driver may suddenly leave the traveled way and enter the median area because another vehicle unexpectedly merges into the occupied lane. Hutchinson and Kennedy also examined the relationship between ADT and the percent of vehicles that entered into (i.e., encroached on) the median. As shown in Figure 2-4, as ADT increased from 4,000 to 6,000 vehicles per day, the per- centage of vehicles crossing into the median increased. The final relationship studied by Hutchinson and Kennedy was that between ADT and lateral distance traveled by encroach- ing vehicles on I-74. The ADT volumes range from 2,000 to 6,000 vehicles per day. These vehicles traveled an average of 19 to 27 feet into the median area over this range of ADTs. This relationship is shown in Figure 2-5. In 1978, data from several Canadian provinces were collected to investigate single-vehicle run-off-the-road crashes on divided and undivided rural highways. A mul- tiple regression analysis of 1,937 encroachments explained only 30 percent of the variance between crashes and traffic volumes (6). Factors such as alcohol, weather, and driver variables were considered to have a significant effect on the models developed. This study showed no significant cor- relation between ADT volumes and encroachment rates; however, when the data were forced into 2,000 vehicles per day groupings and averaged over a set of ranges, the results were nearly identical to the Hutchinson and Kennedy (2) study discussed previously. This study also showed that, Figure 2-2. Encroachment rates including standard errors for the Hutchinson and Kennedy study periods (5). Table 2-1. Hutchinson and Kennedy encroachment data for 10 observation periods (5). Figure 2-3. Relationship between ADT and average encroachment angle (2). Figure 2-4. Relationship between ADT and percent of vehicles encroaching on median (2).

9 An investigation into the factors causing median encroach- ments and cross-median crashes was conducted by Sicking et al. (10). This involved a review of more than 43,000 Kansas freeway crash reports from 2002 to 2006, which identified 8,233 crashes that involved a vehicle entering the median. Of these, 525 events involved vehicles traversing the median and crossing into the opposing lanes of traffic, and 115 involved vehicles crossing the median and colliding with a vehicle trav- eling in the opposite direction. By looking at the reported snow and ice conditions over the 5-year study period, it was determined that winter driving conditions were present less than 12.5 percent of the time, yet nearly 25 percent of the total cross-median events occurred in winter conditions indi- cating an overrepresentation of crashes in these conditions. However, only 7 percent of cross-median crashes in winter conditions involved a fatality compared to 22 percent in non- winter conditions. These data seem to suggest that although winter weather conditions are prone to cause more median encroachments and cross-median events, they are less likely to be fatal; which could be due, in part, to the lower speeds that vehicles travel during inclement weather. The relationship between cross-median events and traffic volumes appears to have a constant relationship for the Kansas data (10). An average of 2.2 cross-median events per 100 million vehicle-miles of travel was observed for ADT ranges between approximately 4,000 and 17,000 vehicles per day. This con- stant relationship of events involving median encroachments to vehicle exposure is similar to the recommendations made in the research conducted by Davis and Morris (5). 2.2 Median-Related Crashes Median-related crashes are traffic crashes in which one or more of the involved vehicles leaves the left side of one road- way of a divided highway and enters the median. A median- related crash in which one or more of the involved vehicles leaves the left side of one roadway of a divided highway, enters the median, crosses the entire median width, enters the opposing roadway of the divided highway, and collides with a vehicle in that opposing roadway is referred to as a cross-median collision (CMC). If a vehicle involved in a crash enters the opposing roadway of a divided highway, but does not collide with an opposing vehicle, this is referred to as a non-collision cross-median crash (NCMC). In an NCMC, the vehicle that enters the opposing roadway may come to rest or roll over in the opposing roadway; may cross the entire opposing roadway, come to rest, roll over, or strike an object on the far side of the opposing roadway; or may enter the opposing roadway and then return to the median area. Crosby (11) evaluated the cross-median crash experi- ence on the New Jersey Turnpike over a 7-year period (1952 through 1958, inclusive). The crash data were from the original on average, the ratio of observed to reported crashes was 3.75:1 for two-lane undivided highways and 5:1 for multi- lane divided highways. A study by Miaou (7) published in 2001, sought to model encroachment rates on rural two-lane highways based on factors deemed to be responsible for such encroachments. The model utilized crash data from Washington State for a 3-year period and was built upon previous research by the author (8, 9). The developed model is shown in Equation 1. E 365 AADT 1000000 exp 0.04 AADT 1000 Ln st = + ( )p p pβ – f Hazf 0.12 HC 0.05 VG + + +  p p ( )1 where E = expected number of roadside encroachments per mile per year AADT = average annual daily traffic (in number of vehicles) from 1,000 to 12,000 bst = state constant (with a default value of -0.42) Lnf = 0, 0.20, and 0.44, respectively, for road segments with 12-foot, 11-foot, and 10-foot-wide lanes Hazf = 0.4 to 0.5 (with a default value of 0.45) HC = horizontal curvature (in degrees per 100-foot arc) from 0 to 30 degrees VG = vertical grade (in percent) from 0 to 10 percent The major drawback of this approach is that the modeling was based exclusively on reported crashes. Since encroach- ments involving a collision are more likely to be reported than encroachments that do not involve a collision, it can therefore be assumed that minor roadside encroachments are not well represented in the model. It is important to note that the author found traffic volume, lane width, horizontal cur- vature, and vertical grade to be significant factors to include in the model. Figure 2-5. Relationship between ADT and encroachment distance (2).

10 However, Garner and Deen also stated that other median ele- ments such as cross slopes and the presence of obstructions can have a greater effect on median crash experience than the median width. Garner and Deen (12) indicated that median cross slopes have a substantial impact on median crash experience. They indicated that deeply depressed medians with cross slopes of 1V:4H and 1V:3H for a 36-foot wide median have been shown to have a significantly higher crash rate than the raised and depressed medians with flatter cross slopes for widths of 20, 30, and 60 feet. Medians with steep slopes may not provide reasonable recovery areas (12). In addition, steep slopes also increase the likelihood of vehicle rollover. It was shown that the roadways studied with cross slopes of 1V:4H and 1V:3H had crash rates of 10.3 and 16.5 crashes per 100 million vehicle-miles traveled, respectively, whereas the average crash rates for roadways with flatter median slopes were found to be 3.4 crashes per 100 million vehicle-miles. The raised median design analyzed in the Garner and Deen study also was shown to have some downfalls. This design seemed to have a higher number of crossover crashes. It was concluded that when drivers hit the median, they tend to overreact, which causes them to lose control of the vehicle. There are other disadvantages associated with raised medi- ans. Raised medians do not provide an adequate storage area for snow removal. Also, water tends to migrate onto the roadway, which allows icy spots to form during cold weather. Garner and Deen also concluded that irregular medians, which have a varying median width and nature, have higher median crash rates, total crash rates, and severity rates. Foody and Culp (13) studied the safety aspects between mound and depressed medians, each having an 84-foot design width. They observed the crash frequency and severity of single-vehicle crashes on four-lane divided Interstates in Ohio from 1969 to 1971 for each median type. They observed 125 miles of highway with the mound median design and 103 miles of highway with the depressed median design. The depressed median had side slopes of 1V:8H. The mound median had 1V:8H foreslopes and 1V:3H backslopes. The study detailed single-vehicle median crashes, crash severity, vehicle path encroachments, and median rollover crashes. The following summarizes the study results obtained by Foody and Culp: • The crash rate is slightly higher for the raised median than for the depressed median section. • There is no difference in injury-related crashes between the two median types. • There is no difference between the two types of medians in the number of median encroachments. • There is no significance in the difference of rollover fre- quency between the two median designs. 118-mile New Jersey Turnpike before the installation of median guardrails. In 1958, 18 miles of median guardrail were installed on sections with variable median widths of 6 to 26 feet. The data used for the research were limited to the through travel lanes excluding those within service areas, interchanges, and their interconnecting roadways and ramps. During the analy- sis period, 48 of 158 (30.4 percent) fatal crashes were consid- ered cross-median crashes. During the analysis period, there were a total of 455 cross-median crashes. They constituted approximately 8.3 percent of all crashes on the New Jersey Turnpike during the analysis period (455 of 5,473 total col- lisions). The cross-median crash rate was higher when the medians were narrower. When grouping roadways together based on traffic volumes, the crash rates were calculated as follows: • 10.8 cross-median crashes per 100 million vehicle-miles traveled on medians 26-foot wide with ADT between 13,400 and 14,800 vehicles per day; • 7.3 cross-median crashes per 100 million vehicle-miles traveled on medians 26-foot wide with ADT between 18,800 and 27,400 vehicles per day; • 0.4 cross-median crashes per 100 million vehicle-miles traveled on medians 24-foot wide with ADT between 23,100 and 36,900 vehicles per day; • 6.2 cross-median crashes per 100 million vehicle-miles traveled on medians 20-foot wide with ADT between 36,000 and 37,600 vehicles per day; and • 5.5 cross-median crashes per 100 million vehicle-miles traveled on medians 20-foot wide with ADT between 23,600 and 24,400 vehicles per day. Garner and Deen (12) compared various median types on divided, four-lane Interstate highways with similar geometric features in Kentucky. The two variables that were the primary focus in the study were median width and median cross sec- tion. For the routes studied, variables such as pavement width and shoulder width remained constant. The types of medians that were analyzed in the study were raised, depressed, deeply depressed, and irregular medians. The results of the Garner and Deen study verified previous conclusions from other researchers that wider medians had fewer CMC crashes. Their data indicated that the percent- age of vehicles crossing the median decreases as the median width increases. Their data also indicated that the relationship between crash rate and median width was not clear. However, deeply depressed medians had a higher crash rate than raised medians. Garner and Deen suggested that the beneficial effects of wide medians can be offset by steep median side-slopes. As such, they recommended slopes of 1V:6H or flatter when the median is 60 feet wide. Additionally, median widths of 30 to 40 feet were recommended on high-speed divided highways.

11 increments. As for the Illinois data, there is a sharp decline in relative effects between median widths in the range from 10 to 40 feet. The largest decline was between the interval of 10 and 20 feet, in which there was a 45-percent decrease in the relative effects of increasing the median width. From 20 to 40 feet, the average decline in relative effects was 42 percent. For median widths greater than 40 feet, the relative effect of increasing the median width for head-on collisions stayed fairly constant around 0.10, or a 10 percent reduction in the total crash rate. The validity of the results observed from the Illinois and Utah HSIS study are controlled by the variables that were used. There are clearly other variables that were either not measured by the database or not used in the final model sim- ply because of the need to limit the model to as few vari- ables as possible (14). Other variables that could have been included in the model are median slope, type of traffic, envi- ronmental factors, and other geometric factors. The general results of this study indicate that crash rates decrease as the median width is increased. It is apparent from the data that there is little decrease in crash rates for medians less than 20 to 30 feet. Therefore, increases in safety effects are not seen until the median reaches at least 20 to 30 feet in width. Even greater safety benefits can be seen for median widths up to 65 to 80 feet, at which point the safety effects begin to level off. Mason et al. (15) used crash and roadway inventory data to characterize CMC crashes on Pennsylvania Interstates and expressways. In 5 years, 267 of these crashes occurred; 15 percent resulted in fatalities and 72 percent resulted in reported injuries. When compared to all crash types on Inter- states and expressways, the severity level of CMC collisions is A study by Kniuman et al. (14) investigated median safety using Highway Safety Information System (HSIS) data from Utah and Illinois. In Utah, the total crash rate was found to decline from 650 crashes per 100 million vehicle-miles for medians with zero width, to 111 crashes per 100 mil- lion vehicle-miles for median widths in the range of 85 to 110 feet. In Illinois, the data suggest a similar trend, with a crash rate of 692 crashes per 100 million vehicle-miles for the medians with zero width and 53 crashes per 100 million vehicle-miles where the median was 85 to 110 feet wide. It also was reported that the average rate of head-on collisions for median widths greater than 55 feet was 1 and 3 crashes per 100 million vehicle-miles for the Utah and Illinois data, respectively. For the Utah data, single-vehicle crashes do not decline as the median width increases from a range between 1 and 24 feet to a range between 85 and 110 feet. For the Illinois data, single-vehicle crashes were found to decline by almost half as the median width increased from a range of 1 to 24 feet to a range of 85 to 110 feet. It was shown that little reduction in crash rate was obtained for median widths in the range of 0 to 25 feet. The most apparent decline in total crash rate was found to occur roughly between 20 and 30 feet. For medians between 60 and 80 feet, the decline in crash rates seems to level off. All of the previously discussed results can be found in Table 2-2, which is an excerpt from the study results. Relationships were developed between the type of colli- sion and the relative effects of the median width. The type of crash most affected by the increase in median width was head-on collisions. For the Utah data, the relative effects were fairly linear. It showed an approximate 17 percent decrease in the relative effects of increasing the median width in 10-foot Median width (ft) Average crash rate (crashes per 100 MVMT) Category Mean N Single- vehicle Head-on Rollover Total Utah 0 0.0 176 127 10 14 650 1 to 10 9.4 257 97 10 5 618 11 to 29 14.9 213 89 8 7 462 30 to 54 46.3 52 109 1 29 159 55 to 84 71.7 179 106 1 22 137 85 to 110 101.0 105 93 0 29 111 All 32.0 982 103 6 14 424 Illinois 0 0.0 567 86 21 5 692 1 to 24 12.8 199 69 12 8 647 25 to 34 29.8 176 92 3 15 292 35 to 44 39.7 479 51 2 6 129 45 to 54 49.2 200 61 2 7 127 55 to 64 63.8 450 27 1 3 45 65 to 84 71.9 239 40 1 5 59 85 to 110 88.9 171 36 1 6 53 All 39.4 2,481 58 7 6 283 Table 2-2. Relationship between median width and crash rate in Utah and Illinois (14).

12 roadway sections were expected to increase the median bar- rier crash frequency, holding all other variables constant. A unit increase in the median barrier offset was expected to decrease the median barrier crash frequency by 3.5 percent, holding all other variables constant. A lower posted speed limit was expected to decrease the median barrier crash fre- quency while the absence of interchange entrance ramp was also expected to decrease the expected median barrier crash frequency, holding all other variables constant. Donnell and Mason (18) predicted the severity of both CMC and median barrier crashes using crash event and roadway inventory data from Pennsylvania Interstate high- ways. Three severity levels (fatal, injury, and property dam- age only) were considered. In the CMC crash severity model, an ordered response was used while the multinomial logit was used to estimate median barrier crash severity. In the CMC crash severity model, the use of drugs or alcohol and the direction of the horizontal curve influenced severity. The predicted probabilities of a fatal CMC crash were between 9.8 and 24.3 percent when considering the various catego- ries of independent variables. The predicted probabilities of an injury CMC crash were between 68.1 and 70.5 percent when considering the various categories of the independent variables. The assumption of parallel regression lines was vio- lated when predicting the severity of median barrier crashes. As such, a nominal response was considered. The indepen- dent variables that influenced crash severity included pave- ment surface condition, drug or alcohol use, the presence of an interchange ramp, and ADT. The predicted severity prob- abilities were as follows: • Fatal: 0.5 to 0.8 percent, • Injury: 53.5 to 60.2 percent, and • Property damage only: 39.0 to 46.0 percent. Donnell and Mason (19) used both CMC and median bar- rier crash frequency and severity models to evaluate existing median barrier warrant criteria in Pennsylvania. Interstate highways with and without median barrier were compared using roadway inventory and crash data. The economic eval- uation consisted of benefits derived from changes in crash costs and the costs were derived from barrier installation, maintenance, and user costs. The benefit-cost analysis results are shown in Figure 2-6. In Figure 2-6(a), the concrete barrier was assumed to be installed only in the center of a median. The gray-shaded area represents benefit-cost (B/C) ratios that exceed 1.0 and where the data used in the analysis represent most CMC and median barrier crashes. The outlined region also contains B/C ratios that exceed 1.0. The frequency of crashes was very low in the outlined region and, therefore, a site-specific evaluation significantly more severe. Additionally, nearly 63 percent of CMC crashes occurred during daylight conditions, 58 per- cent occurred during wet or snowy and icy conditions, and 12 percent involved drugs or alcohol usage. Limited field data collection found that median shoulder width, roadway grade, median cross slopes, the presence and degree of horizontal curvature, presence of roadside obstacles, and vehicle type did not statistically influence CMC crashes. However, there was preliminary evidence to conclude that the presence of interchange entrance ramps does increase the likelihood of CMC crashes. Using the CMC crash data from Pennsylvania, Donnell et al. (16) estimated models of crash frequency for Interstate highways. The model took the form shown in Equation 2. 0.2 (2)–18.203 1.770 –0.0165N e L AADT eCMC MW= × × × × where NCMC = number of CMC crashes per year for one direc- tion of travel L = segment length (mi) AADT = average annual daily traffic (veh/day) MW = median width (ft) All of the parameters were statistically significant (p < 0.0001). However, the model explained only a small proportion of the variation in CMC crash frequency. Interpretation of the AADT parameter (1.77) suggests that CMC crashes are a two-stage process. First, the likelihood that a vehicle loses control and enters the median when traveling in one direc- tion of travel is roughly proportional to the traffic volume (AADT) in that direction. The out-of-control vehicle must then traverse the median, enter the opposing traveled way, and collide with a vehicle traveling in the opposite direction. The likelihood of this occurring should be roughly proportional to the one-way traffic volume (AADT) in the opposing travel lanes. Since the traffic volumes on opposing roadways are typically quite similar on Interstate highways, it seems logical that the likelihood of a CMC crash be roughly proportional to the square of the one-way AADT. A one-unit increase in the median width decreases the CMC crash frequency by approxi- mately 1.7 percent. Donnell and Mason (17) used negative binomial regres- sion to predict the frequency of median barrier crashes on Pennsylvania Interstate highways. There were a total of 4,416 median barrier collisions that occurred during the 5-year study period (1994 through 1998) on 738 miles of divided highway that were protected with a longitudinal barrier. The ADT, presence of an interchange entrance ramp, posted speed limit, horizontal curve indicator, and median barrier offset from the left edge of the traveled way were all statistically sig- nificant predictors of median barrier crash frequency. Curved

13 • Roadway friction factor if number of horizontal curves is greater than 0.67 per km (1.08 per mile); and • Section location (Interstate Route 90, Interstate Route 205, U.S. Route 2, or State Route 16). These indicator variables had the value of 0 when the condi- tion specified was not present or not applicable and had the value of 1 (or a specified length, width, or number of curves) if the condition specified was present or applicable. A comparison of the model output for NM, NB, and RENB is shown in Table 2-3. The results of the negative multinomial regression model indicate that crash frequencies are lower along road sections with lower traffic volumes. The predicted median crossover crash frequency decreases as the number of horizontal curves per km increases. However, the indica- tor for the difference between the maximum and minimum shoulder width (greater than 4 feet) and number of horizon- tal curves is greater than two per section, and this suggests that the crash frequency increases as the curve frequency and shoulder width difference increases. The section length vari- ables were all positive in the negative multinomial model. Miaou et al. (4) presented predictive models of crash fre- quency and severity as well as B/C analysis results for a cross- sectional with/without median barrier study in Texas. Two years of data (1998 and 1999) were collected from Interstates, freeways, and expressways with four or more lanes and a posted speed limit of 55 mph or greater. Only divided highway sections with ADT less than 150,001 vehicles per day were con- sidered in the analysis as were sections with medians between 15 and 150 feet wide. There were 346 reported cross-median crashes in 52 Texas counties during the 2-year analysis period. An additional 3,064 median-related crashes were reported on sections with no longitudinal median barrier. There were 3,672 median-related crashes included in the analysis time period along sections with longitudinal median barrier. Of these 3,672 reported crashes, 2,714 crashes (74 percent) were defined as hit-median-barrier crashes. using the methodology described was recommended. In Fig- ure 2-6(b), two numerical values are shown in each cell. The value on top represents the B/C ratio for center placement location, while the value on the bottom represents a 4-foot offset from the edge of the traveled way. Because the strong- post W-beam guiderail used along medians in Pennsylvania has design deflection ranging from 2 to 4 feet, it is not used when the median is less than 10-foot wide. The “NB” in Fig- ure 2-6(b) represents a condition where no benefits were found by considering a longitudinal barrier. Either the crash severity or frequency did not change enough when compar- ing the with/without median barrier scenario to show a net benefit in crash cost. Ulfarsson and Shankar (20) estimated a predictive model of median crossover crash frequencies with a multiyear panel of cross-sectional roadway data. The study compared three different count regression models, including negative multi- nomial (NM), negative binomial (NB), and random-effects negative binomial (RENB). The results showed that the nega- tive multinomial model outperformed the other two due to the existence of section-specific correlation in the panel. Vari- ables considered in the model included indicator variables for the following conditions: • ADT less than 5,000 vehicles per day; • ADT between 5,000 and 10,000 vehicles per day per lane; • Median width between 9 and 12 m (30 to 40 feet); • Number of horizontal curves per km; • Length of section (km) if median width is less than 12 m (40 feet); • Length of section (km) if median width is between 12 and 18 m (40 feet to 60 feet); • Length of section (km) if median width is greater than 18 m (60 feet); • Difference between maximum and minimum shoulder width is greater than 1.2 m (4 feet) and the number of horizontal curves is greater than two per section; (a) Concrete Median Barrier (b) W-Beam Guiderail Median Barrier Figure 2-6. Benefit-cost ratios for median barrier installation (19).

14 parameter estimates. The roadway inventory and traffic vol- ume variables included in the models were as follows: • Median width (ft); • Logarithm of ADT; • Number of lanes; • Posted speed limit (dummy variable for 60 mph, dummy variable for 65 mph, dummy variable for 70 mph); and • A dummy variable for the year 1999. Four median crash types were considered in the frequency and severity models: cross-median crashes on sections with no barrier, other median-related crashes on sections with no barrier, all median-related crashes on sections with a barrier, and hit-median-barrier-only crashes on sections with a bar- rier. A Poisson-gamma model, using a full Bayes approach, was used to specify and estimate the crash frequency prediction model. The advantage of using such a modeling technique is that it accounts for the uncertainty associated with the model Variable NB RENB NM Constant –1.551 (0.181)† –0.118 (0.391) –1.500 (0.251)† ADT less than 5,000 vehicles per lane daily, indicator –1.398 (0.186)† –1.373 (0.190)† –1.381 (0.312)† ADT between 5,000 and 10,000 vehicles per lane daily, indicator –0.233 (0.158) –0.266 (0.157)‡ –0.298 (0.290) Median width between 30 and 40 ft, indicator 0.463 (0.206)† 0.368 (0.215)‡ 0.432 (0.309) Number of horizontal curves per kilometer –0.309 (0.128)† –0.325 (0.141)† –0.502 (0.262)‡ Length of section (km) if median width is less than 40 ft, 0 otherwise 0.281 (0.047)† 0.278 (0.062)† 0.175 (0.052)† Length of section (km) if median width is between 40 and 60 ft, 0 otherwise 0.526 (0.065)† 0.502 (0.070)† 0.292 (0.068)† Length of section (km) if median width is greater than 60 ft, 0 otherwise –0.358 (0.060)† –0.343 (0.065)† 0.105 (0.026)† Difference between maximum and minimum shoulder width is > 4 ft and the number of horizontal curves is greater than 2 per section, indicator 0.542 (0.321)‡ 0.489 (0.285)‡ 0.486 (0.580) Roadway friction factor if number of horizontal curves is greater than 1.08 per mi, 0 otherwise 0.011 (0.004)† 0.010 (0.005)† 0.009 (0.006) Washington State Route 2, indicator –2.093 (1.098)‡ –1.973 (1.371) 0.271 (0.587) Washington State Route 16, indicator –1.338 (0.581)† –1.290 (0.792) –1.188 (0.746) Washington State Route 90, indicator –0.722 (0.199)† –0.732 (0.195)† –0.560 (0.341) Washington State Route 205, indicator –1.814 (1.055)‡ –1.756 (1.150) –8.815 (0.533)† 0.447 (0.172)† 0.258 (1.074) 128.780 (312.380) 34.514 (90.241) ** ** –827.556 In L at NB values — — –883.746 In L at convergence –711.931 –715.801 –613.078 NOTE: Standard errors are given in parentheses. An “indicator” variable is 1 or a specified quantity if the condition holds and 0 otherwise. The NB and RENB model results presented elsewhere (6) are presented here for comparison with the NM model results. † = Significance at the 95% level by the two-tailed t-test: ‡ = significance at the 90% level by the two-tailed test. a,b = parameters of the beta distribution used in the RENB model: ** = information not available. Parameter α Parameter a Parameter b In L(β=0, α=1), naïve model Table 2-3. NM model coefficient estimation results for median crossover crash frequency (20).

15 • Posted speed limit (dummy variable for 60 mph, dummy variable for 65 mph, dummy variable for 70 mph). None of the explanatory variables used in the crash sever- ity models were found to be statistically significant; there- fore, the observed crash severity distributions were used in the economic analysis. The severity distributions for each of the four crash types are shown in Table 2-5. The crash frequency models and severity data were used to estimate B/C ratios for both concrete and high-tension cable median barrier in Texas. Figure 2-7 shows a potential guideline for concrete median barrier based on B/C ratios. As shown, the B/C ratios increase from lower-left to upper-right. Zone No. 4 includes divided, limited-access roadways with low traffic volumes and the entire range of median widths considered in the study. The B/C ratios in Zone No. 4 were Results of the crash frequency modeling effort are shown in Table 2-4. As shown, the median width is negatively cor- related with crash frequency in all models. This indicates that as the median width increases, the crash frequency decreases. Ordered multinomial logit models were used to develop crash severity models for all four crash types described previ- ously. The variables considered in these models included the following: • Five levels of crash severity (K: fatal injury; A: incapacitating injury; B: nonincapacitating injury; C: possible injury; O: property damage only); • Dummy variable for year 1999; • Median width (feet); • Logarithm of ADT; • Number of lanes; and Covariate (coefficient) Crash frequency model No barrier With barrier Cross-median crashes Other median- related crashes All median- related crashes Hit-median-barrier crashes Offset = exposure (in MVMT) = v1 (=365*AADT*Segment Length/1,000,000) —* — — — Intercept term Overall intercept (β0) Dummy variable for 1999: 1 if 1999 and 0 if 1998 (β1) –3.779 (±0.48) 1.163 (±0.14) –2.239 (±0.07) –0.068 (±0.05) –1.771 (±0.07) –0.031 (±0.001) –1.740 (±0.99) –0.018 (±0.06) Median width (in ft) (β2) –0.011 (±0.003) –0.002 (±0.001) –0.006 (±0.001) –0.013 (±0.002) Log (AADT) (β3) (AADT in 1,000s) — — — — Number of lanes (=β4) –0.293 (±0.09) — — — Posted speed limit (mph) Dummy variable for 60 mph (=1 if 60 mph; =0 if otherwise) (β5) Dummy variable for 65 mph (=1 if 65 mph; =0 if otherwise) (β6) Dummy variable for 70 mph (=1 if 70 mph; =0 if otherwise) (β7) –0.139 (±0.54) 0.500 (±0.16) 0.284 (±0.18) –0.342 (±0.17) –0.126 (±0.06) –0.079 (±0.07) –0.575 (±0.08) –0.075 (±0.07) –0.007 (±0.07) –0.063 (±0.10) –0.188 (±0.09) 0.004 (±0.09) Inverse dispersion parameter Inverse dispersion parameter for this model (Ψ) Inverse dispersion parameter for worst possible model of crash frequency (Ψ0freq) 0.727 (±0.17) 0.158 (±0.02) 1.388 (±0.12) 0.429 (±0.02) 1.956 (±0.16) (0.466 (±0.02) 1.464 (±0.13) 0.367 (±0.02) Goodness-of-fit measures Deviance information criterion/sample size (DIC/n) 0.39 0.78 1.71 0.69 2.54 0.76 2.14 0.75 NOTE: All models were structured using the full Bayes framework with noninformative priors (or hyperpriors). Parameters (β and Ψ) were estimated by using Markov chain Monte Carlo techniques, and the values shown in the table are their posterior means. Values in parentheses are the estimated 1 standard error of parameters to their left based on the posterior density of the parameter. *—indicates not statistically significant at the10% significance level. )/)(/(R freqfreq 01 111 ΨΨΨ −= Table 2-4. Posterior mean and standard error of estimated parameters of Texas median safety crash frequency models (4).

16 B/C ratio. Table 2-6 shows the calculated favorability ratios for various median widths and traffic volumes. A favorability ratio of 1.0 indicated that concrete and high-tension-cable barriers had the same mean B/C ratio and higher ratios suggested increased favorability of using the high-tension- cable barrier over the concrete barrier in terms of the mean B/C ratios. Miaou et al. (4) recommended considering high-tension-cable barriers only when the favorability ratio exceeded 2. Noyce and McKendry (21) investigated the magnitude of, and factors affecting, median crossover crashes in Wisconsin using data from freeways and expressways. In 3 years (2001 through 2003), there were 631 median crossover crashes on four Interstates and 17 other freeways and expressways in Wisconsin. Of these, 81 percent (511 of 631) were single- vehicle crashes. In such instances, single-vehicle crashes involve motorists running off the road to the left and enter- less than 2.0; thus, the combination of traffic volume and median width was considered a lower priority for longitu- dinal barrier consideration than the other zones. Zone No. 1 includes average annual daily traffic volumes between 70,000 and 125,000 vehicles per day and median widths between 0 and 60 feet. In Zone No. 1, various median width-traffic volume combinations produced B/C ratios greater than 10. As such, divided highways in Zone No. 1 without longitudi- nal median barrier were considered the highest priority for median barrier installation. Further, it was recommended that road sections with a mean B/C ratio greater than 10 be given the highest priority when installing concrete median barriers. To develop a potential guideline for the installation of high-tension-cable barriers, a favorability ratio was developed. A favorability ratio was defined as the ratio of the high-tension- cable barrier’s mean B/C ratio over the concrete barrier’s mean Barrier and Crash Type N Severity Type K % A % B % C % PDO % No Median Barrier Cross-Median 346 73 21.1 73 21.1 82 23.7 58 16.8 60 17.3 Other Median-Related 3,046 71 2.3 272 8.9 639 20.9 734 23.9 1,348 44.0 With Median Barrier All Median-Related 3,672 36 1.0 190 5.2 681 18.5 1,098 29.9 1,667 45.4 Hit-Median-Barrier 2,714 13 0.5 128 4.7 490 18.0 835 30.8 1,248 46.0 N = total number of crashes K = fatal A = Incapacitating injury B = Nonincapacitating injury C = Possible injury PDO = Property damage only Table 2-5. Texas median crash severity distribution (4). Figure 2-7. Benefit-cost ratios based on Texas study (4).

17 ing the median; however, a collision with a vehicle traveling in the opposing travel lanes did not result. The crossover crash severity distribution was as follows: • Fatal: 6.5 percent (41 of 631), • Injury: 53.2 percent (336 of 631), and • Property damage only: 40.3 percent (254 of 631). The most common initial cause of median crossover crashes was lost control due to weather (44.0 percent), lost control on dry pavement (41.7 percent), and vehicle collision (11.1 percent). Fitzpatrick et al. (22) used crash data from Texas along approximately 1,200 miles of roadway to develop crash modification factors for freeway and rural multilane high- way median-related crashes. Models that quantify the effect of median-related factors on the cross-median crashes were created for medians with rigid barriers, urban medians with- out barriers, and rural medians without barriers. In NCHRP Project 22-21, Graham et al. (1) studied crash statistics of rural highways with and without median barriers to determine guidelines for designing typical cross sections for medians (i.e., width, slope, and barrier). Simulation of vehicle incursions into medians of various designs also was conducted. The fatal-and-injury crash analysis results for rural four-lane freeways generally indicate that CMC crashes decrease with wider medians, while rollover crashes generally increase with wider medians. These two effects are of almost equal magnitude but in opposite directions. The vehicle dynamics simulation results indicate that, at a median width in the range from 50 to 60 feet, there is a boundary at which the probability of a CMC crash becomes less than the probability of a rollover crash. This suggests that when the lower severity of rollover crashes is taken into account, that there are diminish- ing returns in continuing to make the median wider. Crash prediction models for rural four-lane freeways show that flatter slopes are associated with more CMC crashes and fewer rollover crashes. The models indicate that flatter slopes on freeways are associated with fewer fixed-object crashes. The vehicle dynamics simulations show an inter- action between median slope and median width that was not evident in the crash analysis. For median slopes in the range from 1V:4H to 1V:7H, the boundary between medians where CMC crashes are most prevalent and those for which rollover crashes are most prevalent falls in the median width range from 50 to 55 feet. For median slopes of 1V:8H or flatter, that boundary falls at 60 feet. Thus, the vehicle dynamics simu- lations indicate that the concerns about high-severity CMC crashes are greatest for median widths less than 60 feet and for median slopes steeper than 1V:8H. Furthermore, these results suggest that the likelihood of CMC crashes does not continue increasing as the median slope becomes flatter than 1V:8H. In considering median barrier use, crash prediction mod- els and a before/after evaluation estimated crash modification factors for flexible, semi-rigid, and rigid barriers. A benefit- cost analysis showed that all of these barrier types can be Table 2-6. Favorability ratios from Texas study (4).

18 cost-effective in reducing severe CMC crashes. However, they increase the frequency of less severe fixed-object crashes. The cost-benefit analysis results indicate that flexible barriers may be cost-effective even at lower traffic volumes than shown in current AASHTO median barrier warrants. 2.3 Crash Countermeasures for Improving Median Safety One of the most frequently used countermeasures to pre- vent median encroachments is shoulder rumble strips, which may be considered for implementation on a range of road- way types, including urban and rural freeways, multilane divided highways, multilane undivided highways, and two- lane roads. Recent research by Torbic et al. (23) has provided some of the most reliable and comprehensive estimates to date of the safety effectiveness of shoulder rumble strips. The safety effectiveness estimates for shoulder rumble strips and the standard errors (SE) for the estimates are as follows: • Urban/Rural Freeways Rolled shoulder rumble strips [based on results from Griffith (24)] – 18 percent reduction in single-vehicle run-off-the-road (SVROR) crashes (SE = 7) – 13 percent reduction in SVROR fatal-and-injury (FI) crashes (SE = 12) • Rural Freeways Shoulder rumble strips [based on combined results from Torbic et al. (23) and Griffith (24)] – 11 percent reduction in SVROR crashes (SE = 6) – 16 percent reduction in SVROR FI crashes (SE = 8) • Rural Two-Lane Roads Shoulder rumble strips [based on results from Torbic et al. (23) and Patel et al. (25)] – 15 percent reduction in SVROR crashes (SE = 7) – 29 percent reduction in SVROR FI crashes (SE = 9) Estimates on the safety effectiveness of shoulder rumble strips along rural multilane divided highways are also avail- able but are not considered as reliable as the estimates for freeways and rural two-lane roads. The safety estimates for rural multilane divided highway are as follows: • Rural Multilane Divided Highways Shoulder rumble strips [based on results from Carrasco et al. (26)] – 22 percent reduction in SVROR crashes – 51 percent reduction in SVROR FI crashes The lack of reliable estimates on the safety effectiveness of shoulder rumble strips along other roadway types does not indicate that shoulder rumble strips are ineffective on these roadway types. Rather, their safety effects are not known at this time. Torbic et al. (23) also conducted a review of safety evalu- ations of shoulder rumble strips that have been conducted in many states, and in some cases the evaluations included data from multiple states. Table 2-7 summarizes the results of these safety evaluations, along with results from several unpublished materials. Table 2-7 shows the state/location of the evaluation, the type of facility where the rumble strips were installed, the types of collisions included in the analysis, the estimated safety effectiveness of the rumble strip applica- tion, and the type of analysis that was performed (i.e., if it could be determined from the reference material). Several key findings are as follows: • Most of the studies evaluated the safety effectiveness of shoulder rumble strips installed along freeway facilities. Only a limited number of studies investigated the safety effectiveness of shoulder rumble strips along lower class roadways (i.e., nonfreeways). • Most of the evaluations were limited to those collision types most directly affected by the installation of shoulder rumble strips (i.e., SVROR type crashes). However, several studies did investigate the safety impact of shoulder rum- ble strips on total crashes. • SVROR crashes were reduced by 10 to 80 percent due to shoulder rumble strips. The simple average percent reduc- tion in SVROR crashes from these studies is 36 percent. • Total crashes were reduced by 13 to 33 percent due to shoulder rumble strips. The simple average percent reduc- tion in total crashes from these studies is 21 percent. Concerning the safety effectiveness of shoulder rumble strips, NCHRP Report 617: Accident Modification Factors for Traffic Engineering and ITS Improvements (40), summarizes the current status of crash reduction factors for a variety of treatments. In preparing NCHRP Report 617, a panel of safety experts assigned a level of predictive certainty to each crash modification factor based upon a critical review of the published research. In assigning a single value or values of the safety effectiveness of shoulder rumble strips, the panel only referenced the 1999 study by Griffith (24) and assigned a medium-high level of predictive certainty to these estimates. NCHRP Report 617 specifically states that the estimated safety effects are only applicable to freeways and no other types of roads (i.e., two-lane or multilane roads). Although the employment of rumble strips has been shown to have a positive effect on roadside encroachment, there is research to suggest that they might have a secondary effect that could still eventually lead to roadside encroachment. Spain- hour and Mishra (41) conducted an investigation of run-off

19 State/location Type of facility Type of collisions targeted Percent decrease (–) or percent increase (+) in target collision frequency from application of shoulder rumble strips (standard deviation) Type of analysis Arizona (27) Interstate SVROR –80% Cross–sectional comparison California (28) Interstate SVROR –49% Before–after with comparison sitesTotal –19% Connecticut (29) Limited–access roadways SVROR –32% Before–after with comparison sites Florida (27) Fixed object –41% Naïve before–after Ran–into–water –31% Illinois and California (24) Freeways SVROR (total) –18% (±6.8%) Before–after with marked comparison sites and a comparison group SVROR (injury) –13% (±11.7%) Rural freeways SVROR (total) –21.1% (±10.2%) SVROR (injury) –7.3% (±15.5 %) Kansas (30) Freeways SVROR –34% Unknown Maine (31) Rural freeways Total Inconclusive Before–after with comparison sites Massachusetts (30) SVROR –42% Unknown Michigan (32) SVROR –39% Cross–sectional comparison Minnesota (26) Rural multilane divided highways Total –16% Naïve before–after Injury –17% SVROR (total) –10% SVROR (injury) –22% Total –21% Before–after with comparison sitesInjury –26% SVROR (total) –22% SVROR (injury) –51% Minnesota (25) Rural two–lane roads SVROR (total) –13% (8%) Before–after EB analysis with a reference group SVROR (injury) –18% (12%) Montana (33) Interstate and primary highways SVROR –14% Before–after with comparison sites New Jersey (30) SVROR –34% Unknown New York (34) Interstate Parkway SVROR –65% to 70% Naïve before–after Pennsylvania (35) Interstate SVROR –60% Naïve before–after Tennessee (36) Interstate SVROR –31% Unknown Utah (37) Interstate SVROR –27% Before–after with comparison sitesTotal –33% Virginia (38) Rural freeways SVROR –52% Before–after with comparison sites Washington (39) Total –18% Naïve before–after Multistate (27) Rural freeways SVROR –20% Before–after with comparison sites Table 2-7. Summary of safety benefits attributed to the installation of shoulder rumble strips (23). road crashes in Florida as a result of driver overcorrection. In one-third of the crashes where overcorrection has occurred the vehicle actually ended up encroaching on the opposite road- side of its initial encroachment. The most often cited known factors for causing drivers to initially leave the travel lanes, in order of occurrence in crash reports, were alcohol, exces- sive speeds, driver inattention, and fatigue. In attempting to establish a regression model for predicting crashes due to overcorrection, the authors found that the presence of rum- ble strips has a strong association with this crash type. To this end, it was concluded that while rumble strips are effective at preventing cars from leaving the roadway in the initial direc- tion, they can also cause panic to the driver resulting in an overcorrection and possible loss of vehicle control.

20 2.4 Road Safety Audits Agencies will occasionally conduct road safety audits (RSAs) to formally examine the safety performance of the highways within their jurisdictions. RSAs from two agencies were identified that were focused on locations with excessive median-related crashes. The Massachusetts Highway Department (MassHighway) has conducted three RSAs on Interstate roadways. The first report reviewed (42) a site on I-195 in Westport, Massachu- setts. The purpose of the review was to identify current safety issues on the section under study and to recommend counter- measures for these safety issues. The section chosen was 1.4 miles long and included Inter- changes 9, 10, and 11 on I-195 in southeastern Massachusetts. An 11-person team made the reviews and included traffic engineers, highway designers, safety management profession- als, state police, and RSA consultants. The analysis procedure was based on FHWA’s Road Safety Audit Guidelines (43) and included the following steps: • Obtaining and reviewing crash and other traffic character- istics data and available roadway plans, • Conducting site reviews including photographing and video taping current roadway conditions, • Identifying potentially hazardous issues, and • Identifying and evaluating countermeasures to correct or lessen the noted issues. Issues were categorized by a frequency and severity rating scale. The two ratings scales are shown in Tables 2-8 and 2-9. The relative risk of each identified issue, which combines the frequency and severity rating, is shown in Table 2-10. Site reviews were guided by a prompt list developed for median crossover RSAs by the consultant. Seven contributing Estimated Expected crash frequency (per audit item) Frequency rating Exposure Probability high high 5 or more crashes per year Frequent medium high high medium 1 to 4 crashes per year Occasional medium medium low high high low Less than 1 crash per year, but more than 1 crash every 5 years Infrequent low medium medium low Less than 1 crash every 5 years Rare low low Table 2-8. Frequency rating used in road safety audits (43). Typical crashes expected (per audit item) Expected crash severity Severity rating High-speed crashes; head-on and rollover crashes Probable fatality or incapacitating injury Extreme Moderate speed crashes; fixed object or off-road crashes Moderate to severe injury High Crashes involving medium to low speeds; lane changing or sideswipe crashes Minor to moderate injury Moderate Crashes involving low to medium speeds; typical of rear-end or sideswipe crashes Property damage only or minor injury Low Table 2-9. Severity rating used in road safety audits (43). Frequency rating Severity ratings Low Moderate High Extreme Frequent C D E F Occasional B C D E Infrequent A B C D Rare A A B C Crash risk ratings: A: minimum risk level D: significant risk level B: low risk level E: high risk level C: moderate risk level F: extreme risk level Table 2-10. Crash risk assessment used in road safety audits (43).

21 factors to the median-related crashes were rated by the RSA team. The specific factors were related to travel speeds, advanced warning of merging traffic, and guidance through the interchange in the middle of the study section. Although lack of a median barrier along most of the site was noted and discussed, the RSA team did not recommend adding barrier in the study section. Countermeasures were recommended for each of the noted issues and included installation of overhead sign warning of the major merge movement, addition of chevron sign in an on-ramp curve, improvement of sight distance to merging areas, and increased enforcement patrols. MassHighway has completed at least two other RSAs using the rating scheme shown in Tables 2-8, 2-9, and 2-10. The University of Wisconsin Traffic Operations and Safety Laboratory (TOPS) has researched the history of cross-median crashes in Wisconsin. The TOPS lab reviewed crash data for the period of 2001–2005 (21). During this period, they reviewed 227 cross-median crashes that resulted in 63 fatalities. The TOPS lab developed a top 25 list of roadway segments within Wisconsin that have high rates of cross-median crashes. A consultant was hired by Wisconsin DOT to develop an RSA process and to audit the top five sites from the TOPS list (44). The RSA process developed, including the following steps: • Site selection—Five representative sites were selected from the TOPS list. • Data assembly and analysis—Traffic information and site information were analyzed with 5 years of crash reports for each of the five sites. • Site review—The five-member RSAR team met with local officials who were familiar with the site, and conducted a comprehensive audit of each site. • Report preparation—Results of the audit, including rec- ommended countermeasures, were documented. The contributing factors and recommended counter- measures for each of the five sites are shown in Table 2-11. Inter- change spacing was investigated at the five sites and revealed that three of the five sites had minimal spacing between inter- changes. Further study of weaving and merging distances was recommended. Some of the summary points from these audits included the following: • At four of the five sites, over 70 percent of the crashes occurred in one direction. • Many of the cross-median crashes occur after or just before bridges. • Many drivers involved in cross-median crashes were in the right lane or shoulder prior to the crash. • There were clusters of crashes at bridges, on/off ramps, and weave sections. 2.5 Median Design Practice A WISDOT Transportation Synthesis Report, Putting the Brakes on Crossover Crashes: Median Barrier Research and Practice in the U.S. (45), outlines a survey sent to all AASHTO Research Advisory Committee members. The survey covered three topics related to median barrier policy in the state agencies. The first was to determine which states use the AASHTO Roadside Design Guide (46) to decide where to consider installation of median barriers. The second was to determine which states have their own policies or guide- lines for deciding where to install median barriers or the Site Contributing factors Countermeasures 2 Driving too fast for weather conditions 1. Flatten vertical curves 2. Flatten bridge deck 3. Install safety edge 4. Install rumble strips 5. Install pull-off areas 5 Driver inattention Vehicle malfunction Congestion 1. Investigate drainage issues 2. Increase law enforcement 3. Install exit signs 4. Flatten median side slopes 17 Driving too fast for weather conditions Vehicle malfunction 1. Install de-icing technology on bridge decks 2. Provide enforcement pull offs 3. Install safety edge 20 Driver inattention Loss of control Vehicle malfunction 1. Install safety edge 2. Remove fixed objects 3. Review ramp design 24 Driver inattention Avoiding an animal 1. Install safety edge 2. Repave Table 2-11. Contributing factors and countermeasures for sites from road safety audits performed in Wisconsin (44).

22 appropriate barrier type to use. The third was to determine which states have acceptance/rejection criteria for temporary concrete barriers used in work zones. Survey responses were received from 22 state DOTs and 3 Canadian agencies. The key findings were that most agencies (76 percent) primarily use the AASHTO Roadside Design Guide for guidance on where to install median barriers. Although the Roadside Design Guide recommends median closure for median widths of 50 feet or less, Oregon DOT is adopting a 60-foot median width as a closure warrant, and wider medians may be closed with evidence of cross- median crashes. The Arizona DOT adopted a median width policy in 1999 similar to the current guidance in Chapter 6 of the Roadside Design Guide, but with some modifications. More than half of the agencies (60 percent) have developed their own policy or guidelines for deciding where to install median barriers and which barrier type to use. The Indiana DOT’s Design Manual incorporates guidelines for decid- ing where to install median barriers and which types to use, and the DOT also is developing design guidelines for high-tension-cable barriers. Guidelines developed by Alberta Infra structure and Transportation indicate the most forgiv- ing system that will serve the purpose should be used; that is, flexible systems are preferred over rigid systems. Most agencies (80 percent) have criteria for accepting or reject- ing temporary concrete barriers used in work zones. New Hampshire DOT uses material certifications for acceptance of all concrete barriers and performs a visual inspection to ensure that the proper connection is used and that no chips to the concrete are problematic. New York State DOT recently issued a new engineering instruction intended to address fabrication issues.