Safety Prediction Methodology and Analysis Tool for Freeways and Interchanges (2021)

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.

5 CHAPTER 2: LITERATURE REVIEW This chapter describes the findings from a review of the literature related to freeway and interchange safety. The objective of this review was to identify the highway infrastructure- related factors that influence the safety of freeways and interchanges. The findings were used to identify design and operational elements that have a significant effect on safety, knowledge gaps, and alternative model forms. This information was used to develop a methodological framework for safety prediction. The framework is described in Chapter 3. This chapter consists of two parts. The first part provides background information on interchange design, operation, and safety in the United States. The second part describes the findings from a review of the literature on freeway and interchange safety. BACKGROUND This part of the chapter provides background information on the interchanges that are in service on the U.S. highway system. The objectives of this section are to provide some context for the discussion in subsequent parts of the chapter and to establish a vocabulary for this discussion. Topics addressed include interchange safety, interchange type, and ramp configuration. The last section describes the freeway and interchange area in terms of its disaggregated component parts. The purpose of this section is to identify components that have a unique operational character. Data from Highway Statistics for 2007 (Federal, 2009) were used to characterize the distribution of freeway mileage by functional system and area type. These data are shown in the top part of Table 1. They indicate that there are about 46,700 miles on the Interstate Highway System and 13,400 miles on other access-controlled highways. About 58 percent of this mileage is in rural areas. An examination of Highway Statistics data for recent years indicates that freeway mileage has increased by about 1 percent per year since 2004, almost exclusively in urban areas. The number of interchanges on the highway system is not indicated in Highway Statistics. However, an estimate of this number can be obtained using the average interchange spacing values for rural and urban areas derived by Torbic et al. (2007). The resulting distribution of interchanges is shown in the middle part of Table 1. It is estimated that there are about 17,800 interchanges on the Interstate Highway System and 6,900 interchanges on other access-controlled highways. In contrast to the freeway mileage distribution, only about 32 percent of interchanges are in rural areas. Torbic et al. (2007) also derived estimates of the proportion of freeway-to-freeway (i.e., system) interchanges by area type. These estimates were used to compute the number of system and service interchanges on freeways. The distribution is shown in the bottom part of Table 1. The data shown suggest that there are about 3,400 system interchanges and 21,300 service interchanges on the freeway system.

6 TABLE 1. Freeway mileage and number of interchanges in the United States Attribute Functional System Rural Urban Total Freeway length, mi 1 Interstate 30,313 16,396 46,709 Non-interstate 2 4,575 8,809 13,384 Total: 34,888 25,205 60,093 Number of interchanges 3 Interstate 6,900 10,900 17,800 Non-interstate 1,000 5,900 6,900 Total: 7,900 16,800 24,700 Number of interchanges 4 Freeway-to-freeway (system) 400 3,000 3,400 Freeway-to-arterial (service) 7,500 13,800 21,300 Total: 7,900 16,800 24,700 Notes: 1 - Data from Highway Statistics for 2007 (Federal, 2009). 2 - Includes “Other Freeways and Expressways” and “Other Principal Arterials” with full access control. 3 - Based on an average spacing of 4.4 and 1.5 miles per interchange for rural and urban areas, respectively. 4 - Based on an estimated 5.3 and 17.6 percent freeway-to-freeway interchanges in rural and urban areas, respectively. Interchange Safety The characteristics of crashes in interchange areas were tabulated by Torbic et al. (2007). They examined the Fatality Analysis Reporting System and the General Estimates System for the years 2000 through 2004. The data they reported indicate that interchange-related crashes represent 22 percent of all fatal crashes occurring on freeways. It is noted that the interchange area constitutes about 20 percent of the freeway mileage (at 0.5 mi/interchange). They estimated that there is an average 0.05 fatal crashes/yr per interchange and 12.5 total crashes/yr per interchange (all severities). Interchange Types Typical interchange types are illustrated in Figure 1 (Policy, 2004). The ramps associated with these interchanges vary widely in their geometry. Sharp curvature is common at one or more points along the ramp’s length. Also, some ramps that serve two traffic movements (e.g., right- and left-turn from freeway to crossroad) have an intermediate ramp-to-ramp merge (or diverge) point that makes their operation different from those serving only one movement. A survey of state departments of transportation (DOTs) was conducted by Garber and Fontaine (1999) for the purpose of developing guidelines for interchange selection. They received completed surveys from 36 of the 50 state DOTs. One survey question related to the types of interchanges being used in the respondent’s jurisdiction. The responses to this question are summarized in Figure 2.

7 Other 3% SPU 1% Trumpet 4% Directional 6% Full Cloverleaf 8% Partial Cloverleaf 16% Diamond 62% Figure 1. Typical interchange types. Figure 2. Distribution of interchange types used in the United States. The trends in Figure 2 suggest that the diamond is the most widely used interchange type, followed by the partial cloverleaf (or “parclo”). Together, these two interchange types account for 78 percent of all interchanges. Based on the previous estimate of 24,700 interchanges in the U.S., the distribution in Figure 2 indicates that there are about 15,300 diamond interchanges, 4,000 partial cloverleaf interchanges, 2,000 full cloverleaf interchanges, 1,000 trumpet interchanges, 250 single-point urban interchanges (SPUIs), and 700 “other” interchanges.

8 Typical variations of the diamond and parclo interchange types are shown in Figure 3. It is noted that the SPUI is generally considered to be a diamond-type interchange, but was separately categorized in the distribution shown in Figure 2. Frontage road variations exist for the tight urban diamond and SPUI but are not shown in Figure 3. The parclo AB (with loop ramps on the same side of the crossroad) is also not shown. Diamond Interchanges Parclo Interchanges Figure 3. Typical diamond and parclo interchange types. Ramp Configuration Typical ramp configurations used at interchanges in non-frontage road settings are shown in Figure 4. The ramp configurations used in frontage road settings are not shown, but can be described as slip, buttonhook, and scissor. The diamond ramp configuration in Figure 4 is shown to have two alternative alignments. The alignment shown using the solid line is straight with a skew at the crossroad ramp terminal and possibly one short curve near the freeway ramp terminal. The alignment shown using dashed lines has two curves and negligible skew at the crossroad ramp terminal. These alternatives are likely to have a different influence on the safety of the diamond configuration and possibly the ramp terminal. The parclos shown in Figure 3 (and the directional interchanges shown in Figure 1) each have two or more ramps with a ramp-to-ramp junction at some point along their length. This treatment is not shown in Figure 4 but is found on ramps with the diamond or connector configuration. The ramp-to-ramp merge or diverge point associated with this ramp configuration is likely to have a negative influence on ramp safety, relative to a ramp without such points.

9 Figure 4. Typical ramp configurations. The full cloverleaf interchange is sometimes designed to have a collector-distributor (C- D) road configuration that removes ramp-related weaving from the main lanes. The geometry of this configuration is shown in the lower right corner of Figure 4. It is often used with the outer connection ramp. Freeway and Interchange Design Components A section of freeway with one or more interchanges can be disaggregated into the following components: ● freeway segment, ● interchange ramp, ● crossroad ramp terminal, ● freeway speed-change lane, ● crossroad segment, and ● crossroad speed-change lane. These components are shown in Figure 5 for a hypothetical freeway section. The section has a constant number of basic lanes. Various combinations of entrance and exit ramps are shown on the west side of the facility (similar ramps are provided on the east side but are not shown to simplify the figure). Some of the ramps are associated with a conventional interchange and other ramps serve surface roadways but do not represent one of the typical interchange types shown in Figure 1. Buttonhook Collector-Distributor Parclo Loop (Non-Free-Flow) Free-Flow Loop Outer Connection Direct Connection (used at directional interchanges) Semi-Direct Connection (used at directional interchanges) Diamond (Diagonal) One curve Two curves

10 Figure 5. Freeway and interchange design components. Freeway Segment and Speed-Change Lane Figure 5 illustrates several types of freeway components. For example, there are two speed-change lanes shown, each is associated with a ramp entrance (with length Len). There are two speed-change lanes associated with a ramp exit (with length Lex). The freeway section is shown to consist of several segments. There is a segment formed by an entrance-exit ramp pair Diamond Crossroad Segment Freeway Segment Freeway Speed-Change Lane: Acceleration Lane Freeway Auxiliary Lane: Weaving Section Freeway Speed-Change Lane: Deceleration Lane Crossroad Speed- Change Lanes: Acceleration Lane Deceleration Lane Notes: 1. Interchange ramp proper not shown for any ramps. 2. Right-turn movements not shown at crossroad ramp terminals. 3. Frontage road through movements are not shown at crossroad ramp terminals. 4. Speed-change lanes not shown on west side of the freeway. 5. Lengths measured between marked/painted ramp gore and start (or end) of taper. N Controlled Crossroad Ramp Terminal Crossroad Ramp Terminal Traffic Movements Len Len-en Lw ev Lex-ex Lex Lex-en Len Len-ex Lex Parclo A Parclo A (2-quad) Parclo B Parclo B (2-quad) SPUI

11 with a weaving section (with length Lwev). There is a segment between two ramp entrances (with length Len-en), between two ramp exits (with length Lex-ex), between a ramp entrance and exit (with length Len-ex), and between a ramp exit and entrance (with length Lex-en). Within the lengths of roadway between ramps, lane changes can occur directly or indirectly as a result of the entering (or exiting) ramp traffic. Lane-change frequency and concentration (i.e., intensity) is likely correlated with ramp volume, length of roadway between ramps (i.e., Len-en, Lex-ex, Len-ex, Lex-en), and main lane volume. Crash frequency is likely correlated with lane-change intensity. Crossroad Ramp Terminal The crossroad segment shown in Figure 5 has two “controlled” crossroad ramp terminals. These terminals are typically associated with service interchanges, such as the diamond, parclo, SPUI, and one-quadrant interchange types. A service interchange is used at the intersection of a major and minor roadway, and has one or more traffic movements that are stop- or signal- controlled. The right-hand side of Figure 5 illustrates the traffic movements that intersect at the controlled crossroad ramp terminals of the diamond and parclo interchanges. The north and south legs represent ramps. Right-turn movements are not shown. They may be served at the intersection or externally using a channelized speed-change lane. If Figure 5 is extended to include frontage road settings, then through movements would also be shown on the ramp approaches at each crossroad ramp terminal. With one exception, the traffic movement patterns at crossroad ramp terminals are not the same as found at traditional 3-leg or 4-leg intersections. Hence, the use of a safety prediction model calibrated using traditional 3-leg or 4-leg intersections to estimate ramp terminal crash frequency should be viewed with skepticism until it can be shown statistically that this extrapolation yields acceptably accurate predictions. The one exception is the parclo A (2-quad) terminal, which has traffic patterns that are similar to a traditional 3-leg intersection. Interchange Ramp Ramp configurations are not shown in Figure 5. However, schematic drawings for these ramps were provided previously in Figure 4. The interchange ramp alignment is comprised of tangent and curved segments. These segments are shown in Figure 6 for a diamond ramp. Logically, the crash history of a ramp is likely to be influenced by the number of ramp curves and their radii. It may also be influenced by the operating speed on the ramp segments. This speed is likely to change along the length of the ramp.

12 Curve 1* Tangent 2 Tangent 1 Curve 2 Tangent 3 Curve 1 Tangent 2 Tangent 1 Curve 2* Tangent 3 Exit Ramp Entrance Ramp * - controlling curve Figure 6. Disaggregated alignment for a diamond ramp. FREEWAY AND INTERCHANGE SAFETY This part of the chapter describes the findings from a review of the literature related to the topic of freeway and interchange safety. In many instances, the findings are described in terms of safety performance functions (SPFs) or crash modification factors (CMFs). An SPF describes the relationship between traffic volume and crash frequency for a roadway or intersection. A CMF is used to describe the safety effect of a geometric design or traffic control device. The review is focused on the following topics: ramp proper, freeway speed-change lane, freeway segment, and high-occupancy-vehicle (HOV) facilities. Relatively little information was found on crossroad ramp terminal safety. Table 2 provides some insight as to the relative intensity of crashes for seven selected ramp configurations and area types. The crash rates shown are based on data provided in Appendix B of the report by Bauer and Harwood (1998). Traffic volume data were not provided in the report so comparisons among columns are not fully normalized by exposure. Nevertheless, the following trends are noted from inspection of the rates in Table 2: ● A normal deceleration lane tends to have a lower crash rate than the ramp proper. ● A diverge area on a direct or semi-direct connection ramp tends to have a higher crash rate than the ramp proper. ● The crash rate for a normal acceleration lane tends to be larger than the ramp proper rate when the likely speed increase is large, and vice versa when the speed increase is small. ● The rural diamond exit ramp has a higher crash rate on the ramp proper and at the crossroad ramp terminal approach than the rural diamond entrance ramp. An early study of interchange crash data was undertaken by Cirillo (1968). She examined crash data at 718 urban interchanges and 942 rural interchanges on the Interstate Highway System. The crash rates reported for the ramp components are listed in Table 3. The following trends are noted from inspection of the rates in this table: ● A deceleration lane tends to have a lower crash rate than the ramp proper. ● The rural exit ramp has a higher crash rate on the ramp proper than the rural entrance ramp; however, the reverse trend holds for the urban ramp combination. ● An acceleration lane tends to have a lower crash rate than the ramp proper.

13 TABLE 2. Crash distribution among interchange ramp components Interchange Ramp Component Rural Urban Entrance Exit Ramp Diamond Diamond Diamond Parclo Free- Flow Loop Outer Con- nection Direct & Semi- Direct Total Crashes (All Severities) per Mile per Year by Ramp Type 1 Normal deceleration lane -- 0.34 3.2 5.8 5.9 2.7 2.8 Deceleration lane with mainline lane drop -- -- 12 -- -- 2.1 Ramp proper 0.09 0.76 8.4 13 6.1 11 6.2 Normal acceleration lane 0.48 -- -- -- 7.5 2.1 3.1 Acceleration lane with mainline lane addition 0.0 -- -- -- -- -- 2.9 Diverge area on ramp -- -- -- -- -- -- 11 Merge area on ramp -- -- -- -- -- -- 5.8 Two-way road on ramp -- -- -- 6.1 -- 7.6 -- Total crashes: 79 184 998 50 183 422 511 Total miles: 84.8 91.3 47.6 1.9 9 16 34.8 Total Crashes (All Severities) per Ramp per Year by Ramp Type 1 Crossroad ramp terminal (stop) 0.11 0.25 1.72 2.8 -- 0.97 -- Crossroad ramp terminal (free) -- -- -- -- 0.91 1.1 -- Total crashes: 64 155 721 59 101 275 -- Total ramps: 194 204 139 7 37 45 68 Note: 1 - Based on three years of crash data. Crashes cited occurred in the speed-change lane, ramp proper, or ramp approach to the crossroad ramp terminal. Rates are not computed for interchange components for which there is less than 0.3 miles of total length for all components combined. Source: Bauer and Harwood (1998). TABLE 3. Crash distribution among interchange ramp components Interchange Ramp Component Rural Urban Exit Entrance Exit Entrance Total Crashes (All Severities) per 100 mvm by Ramp Type 1 Deceleration lane 137 -- 186 -- Ramp proper 346 161 370 719 Acceleration lane -- 76 -- 174 Total crashes: 547 375 1,635 2,620 Total 100 million-vehicle-miles: 3.11 4.27 7.33 10.0 Note: 1 - mvm: million-vehicle-miles. Source: Cirillo (1968).

14 Interchange Ramp Proper This subsection describes the findings from a review of the safety literature related to the interchange ramp proper. The ramp proper is defined to be the portion of the ramp between the freeway speed-change lane and the crossroad ramp terminal. Travel along a ramp requires the driver to change speed and direction relatively often and to varying degrees. The frequency and extent of these changes varies by ramp configuration, traffic composition, crossroad ramp terminal control, and the design speed differential of the two intersecting roadways. The ramp design elements of note are represented by horizontal curvature, superelevation rate, ramp grade, and ramp-to-ramp merge (or diverge) points. In addition to these changing conditions, travel along the ramp presents the driver with complex conditions and multiple decision points. These changes and complexities can increase the potential for conflict or crash, especially for larger vehicles (Garber et al., 1992). Basic Descriptors This subsection describes various fundamental descriptors of ramp proper design and operation. Topics addressed include: area type, ramp type, entrance/exit side, ramp configuration, number of lanes, and ramp length. Area Type. This descriptor indicates the population density in the vicinity of the ramp. The categories used are urban and rural. The rates listed in Table 3 indicate that urban ramps have a higher crash rate than rural ramps; however, the relative increase varies by ramp type (i.e., exit or entrance). More recently, Bauer and Harwood (1998) examined crash data for five ramp configurations on Washington freeways. All total, 551 ramps are represented in the database. Their regression analysis of the data indicated that ramps in urban areas have 40 percent more crashes than those in rural areas, given the same traffic volume and ramp configuration. The ISAT acknowledges a possible difference between urban and rural ramps; however, the SPFs in ISAT are identical for both area types (Torbic et al., 2007). Ramp Type. This descriptor indicates whether the ramp is used to enter or exit from the freeway. The categories used are entrance and exit. The ramp proper rates listed in Table 3 indicate that rural exit ramps have a higher crash rate than rural entrance ramps. However, the reverse trend applies to urban ramp combinations. A report by Lundy (1966) describes crash and exposure data for ten ramp configurations on California freeways. All total, 582 ramps were represented in the database. These data were re-analyzed for this report using regression analysis. The objective of this analysis was to quantify the relationship between ramp type and crash frequency. Based on this analysis, it was determined that exit ramps have about 42 percent more crashes than entrance ramps, given the same traffic volume and ramp configuration. This trend was consistent for all ten of the ramp configurations considered (although the relative increase varied by configuration).

15 More recently, Khorashadi (1998) examined crash data for nine ramp configurations found in California. All total, 13,325 ramps were included in the database, representing a mixture of rural and urban areas. A re-analysis of the reported crash data indicates that exit ramps have about 64 percent more crashes than entrance ramps, given the same traffic volume and ramp configuration. Bauer and Harwood’s (1998) regression analysis of Washington ramp data indicated that exit ramps have about 65 percent more crashes than entrance ramps, given the same traffic volume and ramp configuration. This percentage is consistent with that found in the data reported by Khorashadi (1998) for ramps in California. Bauer and Harwood offered that this trend was partly due to crashes occurring on the exit ramp that are related to the operation of the crossroad ramp terminal (most notably rear-end crashes) and thus, are not directly attributable to ramp geometry. The SPFs in ISAT indicate that exit ramps have more crashes than entrance ramps when the configuration is a diamond, free-flow loop, or parclo loop. The amount of increase varies with traffic volume. They also indicate that there is no safety difference between entrance and exit ramps for the direct or semi-direct connection configurations. Entrance/Exit Side. This descriptor indicates whether the ramp is entered on the right side of the freeway and curves to the right, or is entered on the left side and curves to the left. The dominance of right-hand ramps on the freeway system has led drivers to develop an expectation for right-hand ramps. As a result, left-hand ramps are unexpected by unfamiliar drivers and tend to be associated with an increased level of lane changing and speed change. These attributes are believed to explain the increased frequency of crashes observed with left-hand ramps, relative to right-hand ramps, especially in the vicinity of the freeway speed-change lane. A survey of ramp configurations on the Interstate Highway System taken in the late 1960s indicates that left-hand ramps represented about 4.7 percent of the 5,088 ramps found in 22 states (Yates, 1970). This percentage did not vary appreciably for ramps in urban or rural areas. A more recent survey of ramp configurations on freeways in North Carolina was undertaken by Moon and Hummer (2009). They found that left-hand ramps represent about 1.8 percent of the 1837 ramps surveyed. North Carolina was one of the states included in the survey by Yates (1970) so it can be used to examine the change in left-hand ramp presence over time. An examination of Yates’ data indicates that left-hand ramps represented 3.7 percent of the 696 ramps found in North Carolina in the late 1960s. By comparing the two percentages, it appears that about half of the left-hand ramps have been removed from North Carolina freeways in the last 40 years. Two reports were found that address ramp safety and that distinguish between the ramp entrance and exit side in the analysis. However, in these two reports, it is difficult to determine the extent to which crashes in the speed-change lane have also been included in the reported data. Hence, the findings cited in the next two paragraphs reflect a mix of crashes on both the ramp proper and the speed-change lane. A discussion of entrance/exit side and its influence on crashes in the speed-change lane is provided in a subsequent section titled Freeway Speed-Change Lane. Yates (1970) examined crash data for loop and outer connection ramps in 22 states. Only 17 loop ramps with a left-hand entrance or exit were identified, so his analysis focused on the 223

16 outer connection ramps with a left-hand entrance or exit. The reported data were re-analyzed for this report. The results indicate that total crash frequency is 25 percent lower on left-hand outer connection ramps. This finding is contrary to the aforementioned concern about driver expectancy; however, it is likely dominated by crash trends on the ramp proper (as opposed to the speed-change lane) and reflects only the outer connection ramp. Lundy (1966) also examined total crash data for a range of ramp configurations and ramp types, some of which were designated as “left-side” ramps. The data were specific to under- crossing and over-crossing interchanges and are believed to include speed-change-related crashes. The typical under-crossing interchange has the freeway at grade and the crossroad below the freeway. The typical over-crossing interchange has the freeway at grade and the crossroad above the freeway. A re-analysis of the reported data indicates that left-hand ramps at under-crossing interchanges have about 95 percent more crashes than right-hand ramps, given the same volume and ramp type. Similarly, left-hand ramps at over-crossing interchanges have about 54 percent more crashes than right-hand ramps. Ramp Configuration. This descriptor describes the general shape of the ramp. Typical configurations are shown in Figure 4. Several previous studies have examined the frequency and severity of crashes on interchange ramps. These examinations focused on the individual ramps (as opposed to the overall interchange) because of the unique influence of individual ramp element design on crash potential. The findings of many of these studies are summarized by Twomey et al. (1992). For example, they reported total crash rates that compared “curved” and “straight” ramp alignments. Their data indicate that curved ramps have 14 percent more crashes than straight ramps. SPFs have been developed by Bauer and Harwood (1998) and by Torbic et al. (2007). Several configuration-based SPFs were also developed for this report using the data reported by Khorashadi (1998). These SPFs are compared in Figure 7 for exit ramps in rural and urban areas. Similar trends are obtained for entrance ramps. Crash frequency is defined to include total crashes occurring on the ramp (i.e., all severities). A typical ramp length of 0.20 mi was used with the SPFs developed by Torbic et al. An examination of Figure 7 indicates that a consistent trend among ramp configurations is difficult to find. The only trend that is consistent among all SPFs is that the diamond ramp has fewer crashes than the parclo loop ramp, given the same traffic volume. A comparison of the Bauer and Harwood SPFs with those from the Khorashadi data indicates that parclo loops tend to have more frequent crashes than other ramp configurations. Also, diamond ramps tend to be in the “middle” in terms of having fewer crashes than some ramp configurations and more than others. The Khorashadi SPFs tend to predict fewer crashes than the Bauer and Harwood SPFs for a given volume and configuration. The Torbic et al. SPFs suggest that direct connection ramps have more crashes than other configurations, which is in contrast to the trend indicated for this ramp by the other two researchers.

17 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3,000 5,000 7,000 9,000 11,000 13,000 15,000 Average Daily Traffic Demand, veh/day C ra sh F re qu en cy , c ra sh es /y r Parclo Loop Outer Free-Flow Loop Diamond Urban Exit Ramp Direct Conn. 0.0 0.1 0.2 0.3 0.4 0.5 0 500 1,000 1,500 2,000 2,500 3,000 Average Daily Traffic Demand, veh/day C ra sh F re qu en cy , c ra sh es /y r Parclo Loop Outer Connection Diamond and Direct Conn. Rural Exit Ramp Free-Flow Loop 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3,000 5,000 7,000 9,000 11,000 13,000 15,000 Average Daily Traffic Demand, veh/day C ra sh F re qu en cy , c ra sh es /y r Parclo Loop Outer Connection Free-Flow Loop Diamond and Direct Conn. Urban Exit Ramp 0.0 0.1 0.2 0.3 0.4 0.5 0 500 1,000 1,500 2,000 2,500 3,000 Average Daily Traffic Demand, veh/day C ra sh F re qu en cy , c ra sh es /y r Direct Conn. Free-Flow and Parclo Loop Diamond Rural Exit Ramp 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3,000 5,000 7,000 9,000 11,000 13,000 15,000 Average Daily Traffic Demand, veh/day C ra sh F re qu en cy , c ra sh es /y r Direct Connection Free-Flow and Parclo Loop Diamond Urban Exit Ramp a. Bauer and Harwood rural ramp SPFs. b. Bauer and Harwood urban ramp SPFs. c. Khorashadi rural ramp SPFs. d. Khorashadi urban ramp SPFs. e. ISAT rural ramp SPFs. f. ISAT urban ramp SPFs. Figure 7. Ramp-configuration-based SPFs based on total crashes. 0.0 0.1 0.2 0.3 0.4 0.5 0 500 1,000 1,500 2,000 2,500 3,000 Average Daily Traffic Demand, veh/day C ra sh F re qu en cy , c ra sh es /y r Parclo Loop Outer Direct Connection DiamondRural Exit Ramp Free-Flow Loop

18 The trends shown in Figure 7 suggest that ramp configuration alone may not provide a sufficient basis upon which to develop SPFs. For example, ramp configuration does not distinguish between ramps that do, and do not, have a ramp-to-ramp merge or diverge point along their length. Moreover, traffic volume is typically not available for the ramp segments before and after the merge or diverge point. This limitation makes the calibration of an SPF for an entire ramp problematic (Bauer and Harwood, 1998; Bahar et al., 2001). The variation in curvature that can exist among ramps with the same configuration (see Figures 4 and 6) may also explain some of the variations observed in Figure 7. Number of Lanes. This descriptor indicates the number of lanes in the ramp cross section. Data reported by Bauer and Harwood (1998) indicate that 84 percent of interchange ramps in Washington have one lane and 16 percent have two lanes. They developed a regression model that indicates single-lane ramps have more than twice as many crashes as two-lane ramps, for a given traffic volume and configuration. Ramp Length. This descriptor indicates the length of the ramp, as measured along a portion (or all) of the ramp proper. Most ramp SPFs found in the literature do not include ramp length. Similarly, most reported ramp crash rates are not based on ramp length. Bauer and Harwood (1998) developed a regression model for ramp segments (i.e., a piece of the ramp proper that is shorter than its total length) that included a variable for segment length. They found that crash frequency increased with segment length. The SPFs in ISAT include a variable for ramp length. The form of the ISAT SPFs indicates a one-to-one correlation between ramp length and crash frequency. Elements with Quantified Relationship This subsection describes various associations between safety and ramp proper design or operation that have been quantified through an analysis of crash data. Topics addressed include horizontal curve radius, grade, lane width, and ramp meter operation. Horizontal Curve Radius. In the late 1960s, Yates (1970) gathered data for 5,088 ramps on the Interstate Highway System in 22 states. One element of his examination was the relationship between crash rate and ramp curve radius. His examination focused on the ramp curve with the sharpest radius. These data were re-analyzed for this report using regression analysis. The results are shown in Figure 8 using the trend line labeled “Ramp.” The lines shown in Figure 8 illustrate the relationship between crash risk and curve radius. The line labeled “Ramp” is based on data reported by Yates (1970). The line labeled “Two-Lane Highway” is obtained from an equation derived by Harwood et al. (2000) and is based on the curve having a 90-degree deflection angle. It is shown to provide a basis of comparison with the ramp relationship. The two lines suggest that a two-lane highway curve has more crash risk than a ramp curve for any radius smaller than 2,000 ft. It is likely that this difference between the two curves can be explained by the lower speeds and different driver expectations on ramp curves, relative to two-lane highway curves.

19 1.0 1.2 1.4 1.6 1.8 2.0 0 500 1,000 1,500 2,000 2,500 3,000 Ramp Radius, ft C ra sh M od ifi ca tio n Fa ct or . Two-Lane Highway Ramp Figure 8. Relationship between CMF value and ramp radius based on total crashes. Ramp Grade. Bahar et al. (2007) synthesized safety research on the effect of various geometric and traffic control elements. Their review indicated that some research has been conducted on the topic of grade for two-lane highways. The research cited indicates that crash risk increases with increasing grade (up or down). Lundy (1966) examined crash data for 582 ramps on California freeways. The data were specific to under-crossing and over-crossing interchanges. The typical under-crossing interchange has the freeway at grade and the crossroad below the freeway (i.e., exit ramp on downgrade). The typical over-crossing interchange has the freeway at grade and the crossroad above the freeway (i.e., exit ramp on upgrade). The reported data were re-analyzed for this report. The analysis separately considered ramp type (i.e., entrance or exit) and grade (i.e., up or down). The findings indicate that grade does not have a significant effect on ramp crash frequency. Lane Width. Bauer and Harwood (1998) examined the effect of ramp lane width and found that wider lanes were associated with fewer crashes on exit ramps. They calibrated a regression model for ramp segments using all ramp crashes, except those identified as rear end. They rationalized that rear-end crashes on exit ramps were more likely related to the crossroad ramp terminal than the ramp geometry. The average lane width was 15 ft. A CMF was derived from the reported regression model. It is shown in Figure 9 using the trend line labeled “Ramp.” The lines shown in Figure 9 illustrate the relationship between crash risk and average ramp lane width. The line labeled “Ramp” is based on the factor derived from the Bauer and Harwood (1998) model. The two lines labeled “Two-Lane Highway” were obtained from an equation derived by Harwood et al. (2000) and are based on a proportion of “related crashes” equal to 0.55. This proportion represents the proportion of single-vehicle crashes on exit ramps reported by Bauer and Harwood (1998). The equation derived by Harwood et al. was “shifted” to a base lane width of 15 ft to facilitate its comparison with the line labeled “Ramp.”

20 0.6 0.8 1.0 1.2 1.4 1.6 10 12 14 16 18 20 22 Lane Width, ft C ra sh M od ifi ca tio n Fa ct or . Two-Lane Highway AADT > 2,000 veh/day Ramp Two-Lane Highway AADT < 400 veh/day Base Width = 15 ft Figure 9. Relationship between CMF value and ramp lane width based on total crashes. The two-lane highway lines are shown in Figure 9 to provide a basis of comparison with the ramp relationship. These lines suggest that a two-lane highway with an annual average daily traffic (AADT) volume in excess of 2,000 veh/day has more crash risk than a ramp curve. The reverse trend is true for a highway with an AADT volume less than 400 veh/day. Ramp Meter Operation. Ramp metering is a freeway traffic management strategy that is generally recognized to improve freeway operation and safety. However, its effect on ramp safety and operation has not been as closely examined. One research project by Upchurch and Cleavenger (1999) specifically examined the effect of ramp metering on ramp-related crashes. They conducted a before-after study using the non-metered hours as a comparison group. They examined ramp-related crashes during a period of three years before and three years after ramp meters were installed on nine ramps of an Arizona freeway. A log-odds analysis of the reported data was conducted for this report. The results indicate that total ramp-related crashes increased 500 percent during the hours when ramp metering was operational. The increase is largely related to the increase in rear-end crashes on the ramp. Related Topics This subsection describes the influence of ramp configuration on manner of collision. Khorashadi (1998) examined crash data for nine ramp configurations found in California. All total, 13,325 ramps were represented in the database. Included in the reported data was the number of multiple-vehicle crashes. These data were re-analyzed for this report. Single-vehicle crash frequency was estimated by subtracting multiple-vehicle crashes from the reported total crash frequency. A regression model was fit to each data set. These models are shown in Figure 10.

21 0.0 0.4 0.8 1.2 1.6 2.0 0 5,000 10,000 15,000 Average Daily Traffic Demand, veh/day C ra sh F re qu en cy , c ra sh es /y r Parclo Loop Diamond Exit Ramp Single-Vehicle Multiple-Vehicle Figure 10. Relationship between ramp crash frequency and crash type based on total crashes. The lines in Figure 10 indicate that the distribution of collision type varies with traffic volume. This trend is consistent among all ramp configurations. It suggests that multiple-vehicle crashes account for a larger portion of all crashes when ramp volume exceeds about 8,000 veh/day. These trends logically reflect the effect of traffic exposure, such that single- vehicle crashes are more common when there are few vehicles on the road and multiple-vehicle crashes are more common when there are more vehicles on the road. Freeway Speed-Change Lane This subsection describes the findings from a review of the literature related to the freeway speed-change lane. The AASHTO document A Policy on Geometric Design of Highways and Streets (Green Book) (2004) defines the speed-change lane to be an auxiliary lane, including tapered areas, that provides for the acceleration or deceleration of vehicles entering or exiting the freeway lanes. Speed-change lanes have two design types, the parallel design and the taper design. Both designs are shown in Figure 11. The acceleration and deceleration lengths referenced in the figure are defined in the Green Book. The acceleration length and speed-change lane length begin at the end of the controlling curve on an entrance ramp. The deceleration length and speed- change lane length end at the start of the controlling curve on an exit ramp. Hereafter in this chapter, the entrance ramp speed-change lane is referred to as an “acceleration lane” and the exit ramp speed-change lane is referred to as a “deceleration lane.” Alternative speed-change lane lengths and terms were developed for use in Chapters 3 to 9 because of the difficulty of consistently locating the end of the controlling curve in the field or on aerial photographs. The terms used in Chapters 3 to 9 are “ramp entrance length” and “ramp exit length.” They are shown in Figure 11. They are defined by the gore point and taper point.

22 AASHTO Speed-Change Lane Length AASHTO Acceleration Length AASHTO Deceleration Length AASHTO Speed-Change Lane Length 12 ft Exit Ramp with Taper Design Entrance Ramp with Parallel Design Ramp Exit Length Ramp Entrance Length * * * Point where marked gore is 2 ft wide (gore point) End of controlling curve on entrance ramp (start of curve on exit ramp) Taper point Taper point Figure 11. Typical speed-change lanes. Basic Descriptors This subsection describes various descriptors of speed-change lane design and operation. Topics addressed include area type, ramp type, entrance/exit side, and speed-change lane design. Area Type. This descriptor indicates the population density in the vicinity of the speed- change lane. The categories used are urban and rural. The rates listed in Table 3 indicate that speed-change lanes in rural areas have a lower crash rate than those in urban areas. Bauer and Harwood (1998) examined crash data representing 551 ramps on Washington freeways. They specifically identified crashes that occurred in entrance and exit ramp speed- change lanes. Their regression analysis of the data indicated that deceleration lanes in rural areas have 70 percent fewer crashes than those in urban areas, given the same traffic volume. Similarly, acceleration lanes in rural areas have 45 percent fewer crashes than those in urban areas, given the same traffic volume and length. Ramp Type. This descriptor indicates whether the ramp is used to enter or exit from the freeway. The categories used are entrance and exit. The previous examination of crash rates in Tables 2 and 3 indicated that the crash rate for the speed-change lane tends to be lower than that of the ramp proper. One exception is that the crash rate for acceleration lanes can be higher than that of the ramp proper when the speed increase is large.

23 Sarhan et al. (2006) examined crash data for 26 interchanges on a freeway in Canada. The crash data included crashes on the freeway segment within the interchange area plus those in the speed-change lanes. They developed a regression model that included a sensitivity to acceleration length and a second model that included a sensitivity to deceleration length. By comparing these two models, it was found that there were more crashes associated with the acceleration lane, relative to the deceleration lane, for daily freeway traffic volumes in excess of about 70,000 veh/day and equal speed-change lane lengths. This trend is consistent with the crash rates shown in Table 3. Entrance/Exit Side. This descriptor indicates whether the ramp is entered on the right side of the freeway and curves to the right, or is entered on the left side and curves to the left. The focus of this discussion is on crashes that occur in the speed-change lane and in an adjacent segment of the freeway in which ramp-related lane changes occur. A discussion of the association between entrance/exit side and crashes on the ramp is provided in a previous section titled Interchange Ramp Proper. Worrall (1969) examined total crash rates for 139 ramps (of which 29 had a left-side entrance or exit) on urban freeways in Illinois. He found that the left-side entrance ramps had a 60 percent higher crash rate than right-side entrance ramps. Left-side exit ramps had a 90 percent higher crash rate than right-side exit ramps. Based on an examination of operational characteristics for a series of case-study locations, he found that left-side entrance ramps were particularly problematic when the ramp served a large number of trucks. He also found that left- side ramps located just beyond a crest curve or within a short distance of a right-side ramp were also problematic. Moon and Hummer (2009) gathered crash data for 158 ramps (of which 33 ramps had a left-side entrance or exit) on freeways in North Carolina. Crashes in the speed-change lane and on the freeway for a distance up to 1,500 ft from the ramp gore were included in the database. They calibrated a regression model using the data. Model coefficients indicate that left-side entrances or exits have 70 to 150 percent more total crashes than right-side entrances or exits. Zhao and Zhou (2009) gathered crash data for 19 ramps (of which four had a left-side exit) on freeways in Florida. Crashes in the speed-change lane and on the freeway for a distance up to 1,000 ft from the start of the deceleration length were included in the database. A re- analysis of these data was undertaken for this report to quantify the correlation between entrance/exit side and crash frequency. The analysis indicates that left-side exits have 180 percent more total crashes than right-side exits. This percentage is slightly larger than the range reported by Moon and Hummer. Speed-Change Lane Design. As noted in the discussion of Figure 11, speed-change lanes have two design types—the parallel design and the taper design. A key difference between these two designs is the ramp entrance length and the ramp exit length. No research reports or papers were identified that explicitly examined the association between crashes and speed-change lane design. However, there were several documents that described an examination of the relationship between acceleration length (or deceleration length)

24 and crash frequency. The findings reported in these documents are reviewed in the next subsection. Koepke (1993) conducted a survey of 45 state DOTs regarding their ramp design practice and existing ramp designs. He found that 9 percent of the DOTs use the parallel design, 24 percent use the taper design, and 67 percent use both designs. He pointed out that the 1990 Green Book noted a decided trend toward the use of the taper type, although some DOTs use the taper design for exits and the parallel design for entrances. Elements with Quantified Relationship This subsection describes various associations between safety and speed-change lane design that have been quantified through an analysis of crash data. Topics addressed include acceleration length, deceleration length, and ramp exits with a lane drop. Acceleration Length. Five research reports or papers were identified that addressed the relationship between acceleration length and crash frequency. Cirillo (1970) assembled crash data for 3,516 acceleration lanes at interchanges on the Interstate Highway System in 22 states. The analysis unit was a freeway segment with a speed-change lane. Hence, the database included both freeway segment crashes and speed-change-related crashes. The Cirillo data were re-analyzed for the purpose of deriving a CMF for acceleration length. The derivation of this equation is described in the section titled Freeway Segments. The calibrated equation for total crashes is provided in Equation 1. It is shown in Figure 12. The equation converges to a value of 1.0 as acceleration length increases. ( )enrL enr ramp enr eL AADT CMF 3.22 83.0 0.1 3.22 )001.0( 0.1 −−+= (1) where, CMFenr = crash modification factor for acceleration length; AADTramp = ramp AADT volume, veh/day; and Lenr = acceleration length, mi. Equation 1 predicts a factor value of 1.12 for an acceleration length of 2,000 ft. A length of 2,000 ft or more is generally recognized as more than adequate for most freeway ramp entrances. However, the fact that the value exceeds 1.0 for this length and longer indicates that there are additional crashes occurring in the freeway mainlines as a result of ramp-related lane changes that occur downstream of the ramp entrance. Bauer and Harwood (1998) gathered three years of crash data for 276 acceleration lanes and 192 deceleration lanes at interchanges in Washington. Only crashes identified as occurring in the speed-change lane were included in the database. They used a regression analysis to examine the relationship between acceleration length and total crash frequency. They found that speed- change lane crash frequency increased with increasing acceleration length. Bauer and Harwood (1998) also examined acceleration length using total crash data for the combined ramp proper and speed-change lane. In this analysis, they found that the combined

25 0.8 1.0 1.2 1.4 1.6 1.8 300 500 700 900 Acceleration Length, ft C ra sh M od ifi ca tio n Fa ct or . HSM (Highway, 2010) Sarhan et al. (2009) Bauer and Harwood (1998) Cirillo (1970) Ramp AADT = 1,000 veh/day crash frequency decreased as speed-change lane length increased. A CMF was derived from their regression model, as shown in Figure 12. It yields a factor value of 1.0 at the average speed- change lane length of 950 ft. The factor converges to a value of 0.0 as length increases. Figure 12. Relationship between CMF value and acceleration length based on total crashes. Sarhan et al. (2006) examined total crash data for 26 interchanges on a freeway in Canada. The crash data included crashes on the freeway segment within the interchange area plus those in the speed-change lanes. They developed a regression model that included a sensitivity to acceleration length. A CMF was derived from their model, as shown in Figure 12. It yields a value of 1.0 at the average length of 950 ft. The factor converges to a value of 0.0 as entrance length increases. The HSM (Highway, 2010) provides a CMF for acceleration lane length. It is shown in Figure 12. A review of the reference sources for this model indicates that it was derived from the aforementioned Bauer and Harwood (1998) regression model. Hence, the “acceleration length” in reference is actually the ramp entrance length shown in Figure 11. In addition, the base length is stated as 528 ft, which is about one-half of the average length reported by Bauer and Harwood (1998). The factor converges to a value of 0.0 as entrance length increases. Deceleration Length. Four research reports or papers were identified that addressed the relationship between deceleration length and crash frequency. Cirillo (1970) assembled crash data for 3,516 deceleration lanes at interchanges on the Interstate Highway System. The analysis unit was a freeway segment with a speed-change lane. Hence, the database included both freeway segment crashes and speed-change-related crashes. The Cirillo data were re-analyzed for the purpose of deriving a CMF for deceleration length. The derivation of this equation is described in the section titled Freeway Segments. The calibrated equation is provided in Equation 2. It is shown in Figure 13. The equation converges to a value of 1.0 as deceleration length increases.

26 0.8 1.0 1.2 1.4 1.6 1.8 200 400 600 800 1,000 Deceleration Length, ft C ra sh M od ifi ca tio n Fa ct or . HSM (Highway, 2010) Sarhan et al. (2009) Bauer and Harwood (1998) Cirillo (1970) Ramp AADT = 1,000 veh/d ( )exrL exr ramp exr eL AADT CMF 9.22 35.0 0.1 9.22 )001.0( 0.1 −−+= (2) where, CMFexr = crash modification factor for deceleration length; and Lexr = deceleration length, mi. Figure 13. Relationship between CMF value and deceleration length based on total crashes. Bauer and Harwood (1998) gathered three years of crash data for 276 acceleration lanes and 192 deceleration lanes at interchanges in Washington. Only crashes identified as occurring in the speed-change lane were included in the database. They used regression analysis to examine the relationship between deceleration length and crash frequency. The correlation with deceleration length was not statistically significant, so it was not quantified. Bauer and Harwood (1998) also examined deceleration length using crash data for the combined ramp proper and speed-change lane. In this analysis, they found that the combined crash frequency decreased with increasing speed-change lane length. A CMF was derived from their regression model, as shown in Figure 13. It yields a factor value of 1.0 at the average speed- change lane length of 950 ft. The factor converges to a value of 0.0 as length increases. Sarhan et al. (2006) examined total crash data for 26 interchanges on a freeway in Canada. The crash data included crashes on the freeway segment within the interchange area plus those in the speed-change lanes. They developed a regression model that included a sensitivity to deceleration length. A CMF was derived from their model, as shown in Figure 13. It yields a factor value of 1.0 at the average length of 950 ft. The factor converges to a value of 0.0 as exit length increases. The HSM (Highway, 2010) provides information about the effect of deceleration length. In Exhibit 15-5, it indicates that the extension of a lane by 100 ft corresponds to a 7 percent

27 reduction in crashes, provided that the lane does not exceed 690 ft. The guidance in the manual is not clear whether this factor applies to the speed-change lane, speed-change lane and ramp proper, or speed-change lane and freeway segment. The guidance is reproduced as a CMF in Figure 13. Exit Ramps with a Lane Drop. A lane drop at an exit ramp is often a result of the need for two lanes to serve ramp traffic demand. The additional lane is added to the freeway segment as an auxiliary lane in advance of the ramp and then dropped from the freeway cross section at the ramp. This approach maintains the basic number of lanes through the interchange area. Occasionally, the outside lane on the freeway segment is dropped at a single-lane ramp (or two outer lanes are dropped at a dual-lane exit ramp). This approach does not maintain lane balance. Regardless of whether the basic number of lanes and lane balance are maintained, lane drops at exit ramps are not typical and are a source of relatively frequent erratic maneuvers (Taylor and McGee, 1973). Chen et al. (2009) examined crash data for 326 freeway segments in Florida. Each segment had an exit ramp. The collective set of segments represented right-side exit ramps with four different geometric designs, of which three designs had a lane drop. The analysis unit was a freeway segment with a speed-change lane. Hence, the database included both freeway segment crashes and speed-change-related crashes. The crash rates computed by Chen et al. for each of the four designs are reproduced in Table 4. The last column provides a ratio that can be used to make some judgment about the possible influence of a lane drop. The basis of comparison for this ratio is the first ramp listed in the table. This ramp represents the typical design with a single lane ramp, no lane drop, and freeway lane balance. The crash rate ratio indicates that the exits with a lane drop have more crashes, for the same traffic volume. TABLE 4. Comparison of total crash rates for exit ramps with a lane drop Exit Ramp Design Crash Rate crash/mvm Crash Rate Ratio 1 Type Ramp Lanes Speed-Change Lane Design Lane Drop? Lane Balance? 1 1 No Yes 0.34 -- 2 1 Yes No 0.57 1.68 3 2 Yes Yes 0.46 1.35 4 2 Yes No 0.86 2.53 Note: 1 - Ratio uses the crash rate for the type 1 ramp as the basis of comparison. Freeway Segments

28 This subsection describes the findings from a review of the safety literature related to freeway segments. Topics of discussion include freeway segments that are distant from interchange ramps as well as segments in the vicinity of ramp entrances or exits. Unless specifically noted otherwise, the segment is defined to include both travel directions. Basic Descriptors This subsection describes various fundamental descriptors of freeway segment design and operation. Topics addressed include number of lanes and area type. Number of Through Lanes. The relationship between number of lanes and safety is complex such that generalization is difficult. It is generally believed that an additional lane effectively increases the separation between vehicles and that this separation reduces the potential for conflict. Evidence to support this belief is reported by McCasland and Biggs (1980) in their examination of the change in safety associated with a change in cross section at 15 freeway segments in seven states. In each instance, a lane was added within the existing cross section by reducing the lane or shoulder widths. Each site experienced a reduction in its total crash rate, with an average reduction of 29 percent. More recent research that has examined the relationship between number of lanes and safety indicates that cross sections with more lanes are associated with more crashes, for the same traffic volume. Milton and Mannering (1998) used a cross-sectional study to examine crash trends on principal arterial highways in Washington. They found that crashes were more frequent on those highways with more lanes. They rationalized that this trend may be due to an increase in lane changing that occurs on cross sections with more lanes. A similar finding was described by Abdel-Aty and Radwan (2000) in their examination of urban roadway sections. Kononov et al. (2008) examined crash data for urban freeways in three states in an effort to understand the relationship between number of lanes and safety. Based on their examination of the data for a wide range of traffic volumes, they found that number of lanes had a unique influence on the relationship between traffic demand and crash rate. A series of regression analyses resulted in the consistent trend for an ogive shape in the calibrated SPF, as shown in Figure 14a. Separate SPFs were calibrated for differing numbers of lanes. Each trend line in Figure 14a illustrates the observed relationship between traffic demand and crash rate for a given number of lanes. Each trend line is shown to “flatten” (or converge) to a specific large number of crashes at higher traffic demand. Kononov et al. attribute this convergence to a high-density, congested freeway operation where speed and movement is constrained.

29 0 10 20 30 40 50 0 20 40 60 80 100 120 Average Daily Traffic Demand (1000s), veh/day C ra sh R at e, c ra sh es /m i/y r Freeway 1.0-mile segment length 6 Lanes 4 Lanes Safety Improvement a. Relationship between crash rate and lanes. b. Correlation between volume and length. Figure 14. Relationship between crash rate, traffic demand, and number of lanes. The effect of number of lanes is observed by comparing the two trend lines shown in Figure 14a. The six-lane trend line is shown to have a steeper slope than that for four lanes. Kononov et al. rationalize that the steeper slope is a consequence of the increased number of opportunities for lane changes, which is associated with an increased number of lanes. The regression analysis by Kononov et al. is based on “crashes per mile per year” as the dependent variable, as opposed to “crashes per year.” An examination of the relationship between AADT volume and segment length for freeways in several states indicates that there is a consistent negative correlation between segment length and AADT volume. This trend is shown in Figure 14b for freeway segments in California. Segment length and AADT volume have a correlation of -0.24 in this figure. This correlation was found to partially explain the flattening of the trend lines in Figure 14a. The flattening is not evident when using “crashes per year” as the dependent variable. Several researchers have developed SPFs for freeway segments with a specified number of lanes. These SPFs are compared in Figure 15. They are based on fatal-and-injury (FI) crashes. The SPFs shown in Figures 15c and 15d were developed by Torbic et al. (2007) for the ISAT software. The SPFs shown in Figures 15e and 15f were developed by Bonneson and Pratt (2008) using data for Texas freeways. The SPFs shown in these four figures are based on freeway segments that are distant from interchanges, such that there is negligible effect of lane changes associated with ramp entry or exit. It was not possible to verify this “isolation” of interchange influence in the SPFs represented in Figures 15a and 15b. An examination of the trends in Figure 15 indicates that there is little agreement among researchers on the association between number of lanes and crash frequency. 0.0 0.5 1.0 1.5 2.0 2.5 0 50 100 150 200 250 300 350 Average Daily Traffic Demand (1000s), veh/day Se gm en t L en gt h, m i

30 0 5 10 15 20 0 20 40 60 80 100 120 140 160 Average Daily Traffic Demand (1000s), veh/day Fa ta l + In ju ry C ra sh F re qu en cy , . c ra sh es /y r Urban Freeway 1.0-mile segment length 6 Lanes, Persaud & Dzbik (1993) 6 Lanes, Hadi et al. (1995) 4 Lanes, Persaud & Dzbik (1993) 4 Lanes, Hadi et al. (1995) 0 1 2 3 4 5 0 20 40 60 80 Average Daily Traffic Demand (1000s), veh/day Fa ta l + In ju ry C ra sh F re qu en cy , . cr as he s/ yr Rural Freeway 1.0-mile segment length 4 Lanes, Hadi et al. (1995) 0 5 10 15 20 0 20 40 60 80 100 120 140 160 Average Daily Traffic Demand (1000s), veh/day Fa ta l + In ju ry C ra sh F re qu en cy , . c ra sh es /y r Urban Freeway 1.0-mile segment length 6 Lanes 8 Lanes 4 Lanes 0 1 2 3 4 5 0 20 40 60 80 Average Daily Traffic Demand (1000s), veh/day Fa ta l + In ju ry C ra sh F re qu en cy , . cr as he s/ yr Rural Freeway 1.0-mile segment length 6 Lanes 4 Lanes 0 5 10 15 20 0 20 40 60 80 100 120 140 160 Average Daily Traffic Demand (1000s), veh/day In ju ry + F at al C ra sh F re qu en cy , cr as he s/ yr 1.0 Mile Segment Length, No Barrier 2 Entrance Ramps, 2 Exit Ramps 4 Lanes 8 Urban Freeway 6 10 0 1 2 3 4 5 0 20 40 60 80 Average Daily Traffic Demand (1000s), veh/day In ju ry + F at al C ra sh F re qu en cy , cr as he s/ yr 1.0 Mile Segment Length, No Barrier 2 Entrance Ramps, 2 Exit Ramps 4 Lanes Rural Freeway 6 a. Two urban freeway SPFs. b. One rural freeway SPF. c. ISAT urban freeway SPFs. d. ISAT rural freeway SPFs. e. Texas urban freeway SPFs. f. Texas rural freeway SPFs. Figure 15. Freeway segment SPFs based on FI crashes.

31 During the development of the SPFs shown in Figures 15e and 15f, Bonneson and Pratt (2008) noted that an examination of crash rates for freeway segments revealed that crash rate tended to increase as the number of lanes increased, similar to the finding by Milton and Mannering (1998). However, the reverse trend is shown in Figures 15e and 15f, similar to the finding by McCasland and Biggs (1980). This reversal stemmed from the use of a regression model that included factors to account for the effect of barrier presence, median width, and horizontal clearance. They found that a greater portion of the segment length was protected by a barrier (in both the median and along the roadside) on freeways with more lanes. Similarly, they found that the horizontal clearance and median width was smaller on freeways with more lanes. Both of these trends were rationalized to explain the observed increase in crash rate, as opposed to the increase in number of lanes. Area Type. This descriptor indicates the population density in the vicinity of the freeway segment. The categories used are urban and rural. The trends shown in Figure 15 indicate that urban freeway segments have a higher crash frequency than rural segments, for a given traffic volume. Elements with Quantified Relationship This subsection describes various associations between safety and freeway segment design or operation that have been quantified through an analysis of crash data. Topics addressed include horizontal curve radius, grade, lane width, outside shoulder width, inside shoulder width, median width, shoulder rumble strips, horizontal clearance, and ramp meter operation. Horizontal Curve Radius. The correlation between curve radius and freeway segment crash frequency has not received considerable attention in the research literature. One examination of this relationship was undertaken by Raff (1953). He examined total crash data for curves on undivided, divided, and controlled-access rural highways in 15 states. The data cited for controlled-access highways were re-analyzed for this report for the purpose of deriving a CMF. The relationship found is shown in Figure 16. Also shown in Figure 16 is a horizontal curve CMF developed by Bonneson and Pratt (2008) for FI crashes on highway curves. This CMF includes a sensitivity to speed limit. When used with a speed limit of 50 mi/h (which was the average highway speed in the early 1950s), it compares favorably with the trend attributed to the Raff data. A third horizontal curve CMF is also shown in Figure 16. This CMF is documented by Harwood et al. (2000) and is applicable to total crashes on two-lane highway curves. It is shown only to illustrate the trend. It was derived using data from the mid-1980s for curves in Washington. The national speed limit at this time was 55 mi/h.

32 1.0 1.1 1.2 1.3 1.4 1,500 2,000 2,500 3,000 3,500 4,000 4,500 5,000 Curve Radius, ft C ra sh M od ifi ca tio n Fa ct or . Two-Lane Highway Harwood et al. (2000) Def. Angle = 20 deg. Bonneson and Pratt (2008) 55 mi/h Derived from Raff (1953) Controlled Access Highway Bonneson and Pratt (2008) 50 mi/h Figure 16. Relationship between CMF value and curve radius. Grade. As with horizontal curvature, research on the relationship between grade and freeway crash frequency is relatively limited. Dunlap et al. (1978) examined the relationship between grade and crash rate on turnpikes in Ohio and Pennsylvania. They found that crash rates increased on downgrade segments, relative to level or upgrade segments. Milton and Mannering (1998) calibrated a regression model that predicted total crash frequency as a function of grade and other geometric and traffic variables. It was calibrated using data for 2,700 miles of urban and rural multilane highways in Washington. The highways were classified as principal arterials. For highways in eastern Washington, they found that when the grade exceeds 2.5 percent (uphill or downhill) crash frequency increases by 2 percent. For highways in western Washington, they found a 6 percent higher crash frequency for segments with a grade in excess of 1.0 percent. Lane Width. Hadi et al. (1995) calibrated a series of regression models for predicting crash frequency on Florida streets and highways. The models were developed using four years of crash data. The model developed for total crashes on freeway segments included a variable for lane width. The associated regression coefficient was used to derive a CMF for this report. It is shown in Figure 17. The slope of the trend line suggests that a 1-ft reduction in lane width results in a 35 to 40 percent increase in crashes. This increase is unrealistically large and is probably a result of lane width being correlated with other influential variables that are not in the safety prediction model developed by Hadi et al.

33 0.8 1.0 1.2 1.4 1.6 1.8 10 10.5 11 11.5 12 Lane Width, ft C ra sh M od ifi ca tio n Fa ct or Urban, 4-Lane, Freeway Urban, 6-Lane, Freeway (Hadi et al., 1995) Rural Multilane Divided Highway HSM (Highway, 2010) Figure 17. Relationship between CMF value and lane width based on total crashes. In Figure 17, the CMF that is attributed to the HSM (Highway, 2010) was developed for rural multilane divided highways. It illustrates a more likely relationship between lane width and freeway safety, as compared to that attributed to Hadi et al. (1995). It is rationalized that the design and operational differences between a freeway and a divided highway (as related solely to the correlation between lane width and crash frequency) are not so distinct as to invalidate this comparison. Research by Bonneson et al. (2007a) and by Gross et al. (2009) documents an interaction between lane width and shoulder width on rural two-lane highways. As a result of this interaction, the relationship between crash risk and a change in lane width is influenced by the width of the adjacent shoulder (and vice versa). The combined lane width and shoulder-width CMF developed by Bonneson et al. is shown in Figure 18. The relationship identified by Gross et al. is very similar. Research investigating a similar interaction for freeways was not found during the literature review. Outside Shoulder Width. Several researchers have developed regression models relating freeway segment crash frequency with outside shoulder width (among other variables). A CMF for shoulder width was derived from each of these models for this report. They are shown in Figure 19. The data used by Knuiman et al. (1993) represent a mixture of freeway and highway segments in Ohio and Illinois. Three years of crash data were assembled for Illinois segments and four years of data were assembled for the Ohio segments. The segments total 3,055 miles, with about two-thirds of the miles in Illinois. Collectively, the segments represent both urban and rural areas. The data assembled by Harkey et al. (2008) represent access-controlled highways in California. Ten years of crash data were assembled for 993 miles of rural segments and 501 miles of urban segments.

34 0.8 1.0 1.2 1.4 1.6 1.8 9 10 11 12 13 Lane Width, ft C ra sh M od ifi ca tio n Fa ct or Combined AMF (1.5 ft shoulder width) Combined AMF (8 ft shoulder width) 0.9 1.0 1.1 1.2 1.3 6 7 8 9 10 11 12 Outside Shoulder Width, ft C ra sh M od ifi ca tio n Fa ct or . Rural Access Controlled Highway Urban Access Controlled Highway (Harkey et al., 2008) Freeway and Divided Highway (Knuiman et al., 1993) Urban, 6-Lane, Freeway (Hadi et al., 1995) Figure 18. Illustrative interaction between shoulder width and lane width. Figure 19. Relationship between CMF value and outside shoulder width based on total crashes. The trend lines shown in Figure 19 indicate a general agreement among the various alternative CMFs. However, the slopes are sufficiently different among sources as to suggest that other factors may be exerting some influence. For example, it is possible that the CMFs with steeper slopes are associated with roadways with narrower traffic lanes. They may also reflect roadways with more frequent roadside barrier sections, shorter offset to roadside barrier, or both. Inside Shoulder Width. A review of the literature revealed two independent efforts to quantify the relationship between inside shoulder width and freeway crash frequency. Both research projects calibrated a regression model that included a variable for inside shoulder width

35 0.9 1.0 1.1 1.2 1.3 4 5 6 7 8 9 10 Inside Shoulder Width, ft C ra sh M od ifi ca tio n Fa ct or . Principal Arterial Highways Milton and Mannering (1998) Rural Freeway (Hadi et al., 1995) 0.8 0.9 1.0 1.1 1.2 20 30 40 50 60 70 80 Median Width, ft C ra sh M od ifi ca tio n Fa ct or . Urban Access Controlled Highway Rural Access Controlled Highway (Harkey et al., 2008) Freeway and Divided Highway (Knuiman et al., 1993) Urban Freeway (Hadi et al, 1995) 0.8 0.9 1.0 1.1 1.2 20 30 40 50 60 70 80 Median Width, ft C ra sh M od ifi ca tio n Fa ct or . Barrier Offset 2 ft from Both Shoulders Barrier in Center of Median Freeway (Bonneson and Pratt, 2008) Barrier or Bridge Rail for 5% of Segment 10-ft Inside Shoulder Width, 6 or more lanes on a freeway segment. A CMF was derived from each model and is shown in Figure 20. A visual comparison of the trends in Figures 19 and 20 suggests that a change in inside shoulder width has slightly smaller influence on the CMF value than a similar change in outside shoulder width. Figure 20. Relationship between CMF value and inside shoulder width based on total crashes. Median Width. CMFs for median width were derived from regression models developed by Hadi et al. (1995), Knuiman et al. (1993), Harkey et al. (2008), and Bonneson and Pratt (2008). The CMFs derived from the Hadi and Bonneson models were calibrated using freeway crash data. The Knuiman model was based on crash data for a mixture of freeways and major highways in Illinois. The Harkey model was based on crash data for controlled-access highways. The derived CMFs are shown in Figure 21. a. CMFs for three researchers - total crashes. b. CMFs with barrier influence - FI crashes. Figure 21. Relationship between CMF value and median width.

36 The trend lines in Figure 21a apply to total crashes. They show general agreement that narrower medians are associated with more frequent crashes. However, the slopes of the trend lines vary among sources and may be a consequence of other, unexplained factors such as inside shoulder width or barrier presence. For example, Figure 21b illustrates the influence of barrier presence on median width. The trend lines shown are applicable when a barrier is present in the median for 5 percent of the segment length. The lines shift upward with increasing barrier percentage. The effect of median barrier presence on crash frequency and severity was examined by Tarko et al. (2008). They found that the conversion of a depressed median to a flush median with a rigid barrier increased total single-vehicle crashes by 120 percent, while reducing total same- direction crashes only 20 percent. This trend was also found in by Bonneson and Pratt (2008) and is reflected in the CMF shown in Figure 21b. Tarko et al. also found that the inclusion of a barrier in the median increased the likelihood of severe crashes (although it nearly eliminated fatal crashes). For all of the CMFs shown in Figure 21, the median width is measured from the near edges of the travel way of the opposing roadbeds. Thus, this width includes the width of the inside shoulder. The CMFs shown in Figure 21a do not account for the effect of inside shoulder width. Thus, the same factor value is obtained when median width is reduced by two feet and there is either a one-foot reduction of both inside shoulders or a two-foot reduction of the area between shoulders. Logically, a reduction in the inside shoulder width should be accompanied by a larger median-width factor value than a reduction of width between the inside shoulders. Shoulder Rumble Strips. Griffith (1999) investigated the correlation between the presence of continuous, rolled-in rumble strips and crash frequency on urban and rural freeways in California and Illinois. The focus of his examination was single-vehicle run-off-road crashes. The reported crash data were used to calculate rumble strip CMFs. These calculations and the resulting CMFs are shown in Table 5. TABLE 5. CMFs for shoulder rumble strips Data Source State Severity Level Treated Site Crashes 1 Comparison Site Crashes CMF (frs)2 Standard Deviation3 After Before After Before Griffith (1993) Illinois Total/all 1895 2801 1833 2288 0.84 0.036 California Total/all 469 579 364 417 0.93 0.088 Combined Total/all 2364 3380 2197 2705 0.86 0.034 Illinois Injury 877 1135 765 874 0.88 0.059 Notes: 1 - Analysis applies to single-vehicle run-off-road crashes. 2 - frs = Aftertreated × Beforecomp / (Beforetreated × Aftercomp). 3 - Standard deviation = frs × (1/Aftertreated + 1/Beforecomp + 1/Beforetreated + 1/Aftercomp)0.5. Horizontal Clearance. The clear zone concept is based on the rationale that crash frequency and severity will be reduced by increasing the lateral offset to vertical obstructions (or

37 1.00 1.02 1.04 1.06 1.08 1.10 10 15 20 25 30 Horizontal Clearance, ft C ra sh M od ifi ca tio n Fa ct or . Barrier or Bridge Rail for 5% of Segment (barrier offset 2 ft from shoulder) Roadside has No Barrier or Bridge Rail Rural Freeway, 6-Lanes 10-ft Outside Shoulder Width non-traversable ditch cross sections) along the roadside. Objects that cannot be relocated or removed from the clear zone are protected by barrier or made to operate in a break-away manner in the event of a collision. Research on the relationship between horizontal clearance and FI crash frequency was examined by Bonneson and Pratt (2008). The CMF that they derived is illustrated in Figure 22. Figure 22. Relationship between CMF value and horizontal clearance based on FI crashes. The trend lines in Figure 22 indicate that barrier presence along the roadside has some influence on crash risk. The barrier itself is a fixed object, but one that is designed to reduce fatalities associated with roadside crashes. This CMF is applicable when there is a barrier or bridge rail on the roadside (it can consist of several short barrier sections on a portion of the segment or one barrier that extends the length of the segment). The barrier can be rigid or semi- rigid. It can be located adjacent to one roadbed, both roadbeds, or at a specified distance from the edge of traveled way. Ramp Meter Operation. Ramp metering is a freeway traffic-management strategy that is generally recognized to improve freeway operation and safety. Research is cited by Everall (1972) of reductions in total crashes that range from 40 to 70 percent following the introduction of ramp meters on freeway sections in Atlanta, Chicago, and Houston. More recently, Upchurch and Cleavenger (1999) examined the effect of ramp metering on freeway crash frequency. They conducted a before-after study using the non-metered hours as a comparison group. They examined freeway crashes during a period of three years before and three years after ramp meters were installed on nine ramps of an Arizona freeway. A log-odds analysis of the reported data was conducted for this report. This analysis indicates that total crashes on the freeway were reduced by 40 percent when ramp metering was operational.

38 In 2000, Minnesota DOT disabled ramp meters on its freeways at the direction of the State Legislature. A subsequent analysis of total crash data indicated that freeway crash frequency increased 26 percent as a result of the cessation of the ramp meter operation (Cambridge 2001). Illumination. The presence of lighting along a freeway segment has been found to improve safety. Griffith (1994) compared daytime and nighttime crash rates on urban freeway segments with and without continuous lighting. The database contained 54.6 mi of data for segments with continuous lighting and 35.5 mi of data for segments with lighting only at the interchanges. He found that segments without continuous lighting had 12 percent larger nighttime crash rates (based on total crashes). Elvik and Vaa (2004) conducted a meta analysis of several published studies focused on the effect of illumination on safety. They found that injury crash frequency increased 17 percent when lighting level was reduced. More recently, Monsere and Fischer (2008) found that injury crash frequency increased 39 percent when freeway segment lighting was reduced (i.e., lighting fixtures eliminated) along 5.5 mi of interstate highway in Oregon. Related Topics This subsection describes various topics related to the safety of the freeway segment. The focus is on freeway design and operational elements. The topics addressed in this section include ramp entrance/exit-related lane changing, weaving, interchange spacing, truck lane restrictions, and volume-to-capacity ratio. The first three topics relate to lane changing that occurs on the freeway segment in the vicinity of interchange ramps, but not at the speed-change lane. This lane changing is a result of drivers maneuvering from an inside lane to an outside lane to access an exit ramp, or maneuvering from an outside lane to an inside lane to avoid vehicles entering the freeway from an entrance ramp. Ramp Entrance/Exit-Related Lane Changing. The presence of a ramp entrance or exit creates a large number of lane changes on the freeway and a notable variation in lane volume. SPFs were developed by Kiattikomol et al. (2008) in part to examine the influence of interchange ramp presence on freeway crash frequency. They collected three years of crash data for urban freeway segments in North Carolina and Tennessee. At total of 377 miles were represented in the database, with slightly more of the miles (204) on Tennessee freeways. Segments located more than 1,500 ft from the middle of the interchange were considered to be “non-interchange” segments. Total crash rates of 42 and 82 crashes per 100 million vehicle-miles (100 mvm) were found for non-interchange segments in North Carolina and Tennessee, respectively. These rates were found to increase by about 200 percent on interchange segments. Torbic et al. (2007) also developed separate SPFs for freeway segments that were located either within or outside of an interchange area. Segments located more than 0.3 mi from the nearest ramp gore were considered to be “outside” interchange segments. A comparison of the SPFs for both segment types indicates that “within” interchange segments have more crashes than “outside” interchange segments. The amount of increase varies from 0 percent at low volume to more than 100 percent at high-volume. Torbic et al. rationalized that this increase is due to the weaving and lane changing associated with the interchange ramps.

39 0 1 2 3 4 5 6 1,200 1,400 1,600 1,800 2,000 2,200 2,400 Distance from Gore, ft Pe rc en t L an e C ha ng es Upstream of Exit Ramp 1,400 ft Speed-Change Lane Length Downstream of Entrance Ramp Research by Goswami and Bham (2006) was examined to determine the extent of lane- changing activity in the vicinity of an interchange ramp terminal. They used vehicle trajectory data for a segment of I-80 in California for this purpose. They decomposed the freeway segment passing through the interchange into 200-ft zones, with separate zones for each lane. They counted the lane volume and the lane changes that occurred within each zone. The lane-change frequency for entrance and exit ramps is shown in Figure 23. It is expressed as a percentage of freeway lane volume. The trend lines in this figure represent a regression model fit to the data points shown. The trend lines suggest a gradual reduction in lane change activity with distance from the ramp gore. Figure 23. Percent of lane changes as a function of distance from ramp gore. Cirillo (1968) examined total crash rates for segments of the Interstate Highway System that were in the vicinity of an interchange. Several years of crash data were obtained from 20 state DOTs. The database represented more than 9,000 mile-years of data. Crash rates were computed for segments that were located within these distances from the ramp gore: 0.2 mi, 0.2 to 0.4 mi, 0.5 to 0.9 mi, 1.0 to 1.9 mi, 2.0 to 3.9 mi, and 4.0 to 7.9 mi. Crashes on the freeway in the vicinity of the speed-change lane were excluded. Crash rates were separately calculated for entrance and exit ramps in urban and rural areas. These crash rates for the urban segments are shown in Figure 24. The trend line shows that crash rate gradually reduces as distance from the ramp gore increases. It is likely that the lane-change activity associated with the ramp has a direct influence on the trends in crash rate shown in Figure 24. This influence is reflected in a CMF of the following form. xbb rampx eAADTCMF 10)001.0(0.1 −+= (3) where, CMFx = crash modification factor for lane changes at a distance x from the ramp gore; b0, b1 = regression coefficients; and x = distance from ramp gore, mi.

40 0 20 40 60 80 100 120 140 0.0 0.5 1.0 1.5 2.0 2.5 3.0 Distance from Gore, mi C ra sh R at e, c r/1 00 m vm Upstream of Exit Ramp Downstream of Entrance Ramp Urban Freeway Figure 24. Total crash rate as a function of distance from ramp gore. Equation 3 can be integrated to obtain an average CMF value for a segment that extends a distance xb to xe from the ramp gore. The result of this integration is shown in Equation 4. The CMF value converges to 1.0 as the subject segment length or location approaches infinity. ( )eb e b e b xbxb be b ramp x x x x x lc ee xxb AADT dx dxCMF CMF 11 0 )( )001.0( 0.1 1 −− − − += =   (4) where, CMFlc = crash modification factor for ramp-related lane changes; xb = distance from ramp gore to start of segment, mi; and xe = distance from ramp gore to end of segment (xe > xb), mi. Nonlinear regression was used to fit Equation 4 (using a simple SPF) to the data reported by Cirillo (1968). A factor was included in the SPF to account for differences between urban and rural segments. The b1 regression coefficient was allowed to have a unique value for exit ramps and for entrance ramps. The calibrated CMF is shown in Figure 25 for a freeway segment that starts 2,000 ft from the ramp terminal. Other trend lines can be computed for other starting locations. If the subject freeway segment starts 2,000 ft from the gore, then Figure 25 can be used to obtain a factor value for any specified segment length. For example, if the segment is 1,500 ft long and downstream of an entrance ramp, then the end of the segment is at 3,500 ft and the factor value is 1.65. This value can also be obtained by inspection of the trends in Figure 24.

41 Figure 25. Relationship between CMF value and distance from gore. Weaving Section. A weaving section presents a combination of merging and diverging maneuvers along a relatively short length of the freeway. Speed differentials between weaving and non-weaving vehicles can be significant. During heavy traffic demand periods, a weaving section often becomes a bottleneck and increases the potential for rear-end crashes. Cirillo (1970) examined weaving sections at 646 full cloverleaf interchanges on the Interstate Highway System. The weaving sections ranged from 400 to 800 ft in length and included all freeway lanes. An examination of the data indicated that the crash rate for the weaving section decreased as weaving length increased. The amount of decrease was found to vary with freeway traffic demand, with a larger percent decrease associated with lower volume. The data were used to develop a CMF for this report. This equation is shown in Figure 26. It is based on total crashes. Bonneson and Pratt (2008) collected crash data for 588 freeway segments in Texas. The database included freeway segments with and without weaving sections. All weaving sections were between interchanges, included one auxiliary lane, and required one lane change for both the entering and exiting vehicles. The regression model included a variable to account for weaving length. Ramp traffic demand was not available, so it was not included in the model. They derived the CMF that is shown in Figure 26. It yields a factor value of 1.0 as weaving length increases. It is based on FI crashes. Some weaving sections require two or more lane changes for one or more of the weaving movements. These weaving sections are not as common as those that require one lane change for both weaving movements. Research cited in the HSM (Highway, 2010) indicates that the conversion of “two-lane-change” weaving sections to “one-lane-change” sections reduces total crashes in the merging lanes by 32 percent. 1.50 1.55 1.60 1.65 1.70 1.75 1.80 0 1,000 2,000 3,000 4,000 5,000 6,000 7,000 Distance from Gore for X e, ft C ra sh M od ifi ca to n Fa ct or fo r Fr ee w ay S eg m en t Upstream of Exit Ramp Downstream of Entrance Ramp xb x b = 2,000 ft

42 1.0 1.2 1.4 1.6 400 500 600 700 800 900 1,000 Length of Weaving Section, ft C ra sh M od ifi ca tio n Fa ct or Cirillo (1970) Exit Ramp ADT = 1,000 veh/day Entrance Ramp ADT = 1,000 veh/day Bonneson and Pratt (2008) Figure 26. Relationship between CMF value and weaving length. Interchange Spacing. Bared et al. (2006) evaluated the relationship between total crash frequency and interchange spacing using a regression model. They found that segment crash frequency, when expressed on a “per mile” basis, declined with increasing interchange spacing. The following CMF for interchange spacing was derived from this model. 221.0361.0 200,340.3             = − rsp sp AADTLCMF (5) where, CMFsp = crash modification factor for interchange spacing; AADTr = sum of the AADT volumes for all four ramps (two entrance ramps, two exit ramps), veh/day; and Lsp = spacing between ramp terminals of adjacent interchanges (see Figure 27), mi. Equation 5 is shown in Figure 27 using a thick trend line. The AADTramp for the trend line shown is equal to 34,200 veh/day. Equation 5 is derived to yield a CMF of 1.0 at a spacing of 3.0 mi. Two CMFs were derived from other sources and are also shown in Figure 27. The CMF attributed to Bonneson and Pratt (2008) is derived from the safety prediction methodology they developed for Texas DOT. The derivation of this CMF was more complicated given the formulation of the models. It has the following form. It is based on FI crashes. aggenraggwevocsb svmv exrenr sp CMFCMFCMFCC CC CMF || ][ 0.1        + + +=  (6) where, Cenr = ramp entrance crash frequency; Cexr = ramp exit crash frequency; Cmv = multiple-vehicle non-ramp crash frequency;

43 Spacing Freeway Crossroad 0.5 1.0 1.5 2.0 2.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 Spacing, mi C ra sh M od ifi ca tio n Fa ct or . Bared et al. (2006) Bonneson and Pratt (2008) Cirillo (1968) Csv = single-vehicle crash frequency; CMFocsb = outside clearance (some barrier) crash modification factor; CMFwev|agg = aggregated weaving section crash modification factor; and CMFenr|agg = aggregated ramp entrance crash modification factor. Figure 27. Relationship between CMF value and interchange spacing. The outside clearance CMF shown in Equation 6 was incorporated after an analysis of Texas data indicated a trend toward more roadside barrier and reduced horizontal clearance with decreasing interchange spacing. Spacings of less than 0.6 mi were assumed to have a weaving section; those greater than 0.6 mi had ramp entrance lengths of 0.3 mi. An urban eight-lane cross section with an AADT volume of 188,000 veh/day was used for the analysis. The CMF from Equation 6 was normalized to yield a value of 1.0 at a spacing of 3.0 miles for consistency with Equation 5. The CMF attributed to Cirillo (1968) is based on the ramp-related lane-change CMF shown previously as Equation 4. The distance from the ramp gore to the start of the segment xb was set to 0.0 mi. A ramp AADT volume of 1,000 veh/day was assumed. One CMF was calibrated for lane changes that occurred upstream of the exit ramp and a second CMF was calibrated for lane changes that occurred downstream of the entrance ramp. Both were adjusted to yield a value of 1.0 at a spacing of 3.0 mi to facilitate comparison with the other CMFs. The two lane-change CMFs were then multiplied together (since there is both a downstream ramp and an upstream ramp in the weaving section between interchanges) to obtain the trend shown in Figure 27. Truck Lane Restriction. Many states have implemented truck lane restrictions on selected freeways. The typical restriction prohibits trucks from using the inside (left-most) lane of a freeway with three or more lanes in the subject travel direction. The goal is to improve the overall operation and safety of the traffic stream. Research was undertaken by Cate and Urbanik (2004) to examine the safety and operational benefits of truck lane restrictions. They cite a 1997 survey that found 40 percent of state DOTs used truck lane restrictions on Interstate Highway System. They used simulation to investigate the effect of truck lane restrictions. They found that truck lane restrictions

44 substantially reduce lane changes, which they speculate may provide “significant gains in the area of safety and driver comfort.” Kobelo et al. (2008) examined crashes on limited-access highways in Florida. A total of 1,216 mi of highways were represented in the database. At total of 430 mi had a truck lane restriction. The average AADT volume per lane was 18,417 veh/day. One year of crash data were gathered and analyzed in a cross-sectional manner using a regression model. The researchers found that total crash frequency was 4 percent lower on highways with a truck lane restriction, relative to those with no restrictions. Fontaine et al. (2009) examined crashes on interstate highways in Virginia. Six years of crash data were obtained for a time period that spanned the point in time that truck restrictions were implemented. The segments with a truck restriction totaled 237 directional miles. Highway segments for which there was no truck restriction were used as a reference sample to facilitate an empirical Bayes analysis of the before-after data. The reference sample included 154 directional miles. The researchers found that total crash frequency was reduced by 13 percent when the AADT volume per lane was less than 10,000 veh/day. When the AADT volume per lane was higher than 10,000 veh/day, total crash frequency increased by 28 percent. They attributed this trend to the increased difficulty of a lane change with higher lane volume. Volume-to-Capacity Ratio. The volume-to-capacity ratio relates the demand volume to the capacity of a roadway. As the volume nears capacity, average speed tends to decrease and headway is reduced. Logically, these changes have some influence on crash frequency, as well as the crash type (i.e., single vehicle versus multiple vehicle) and crash severity distributions. Some research has been undertaken to examine the relationship between volume-to- capacity ratio and crash character. This research typically compares average hourly volume estimates with the crashes that occur during the same hour for one or more years. In this manner, the analysis is often structured by time of day. There are issues of sample size, day versus night, and autocorrelation that complicate this type of analysis. Hall and Pendleton (1989) gathered data for 41, ten-mile segments on the New Mexico State Highway System. Each segment was selected primarily because it included one of the 50 permanent count stations maintained by the New Mexico DOT. About one-half of the segments had a two-lane cross section and the other one-half had a four-lane cross section. All of the segments were verified to be free of “significant” access points. Three years of crash data were assembled for each segment. The relationship between total crash rate and volume-to-capacity ratio for the “high-volume” segments (which were primarily four-lane highways) is shown in Figure 28. Each data point represents one hour of the average day. As noted by Hall and Pendleton, the segments included in the study did not have sufficient volume to allow the examination of high volume-to-capacity ratios and crash frequency. The trend lines in Figure 28 indicate that there is an increase in crash risk with lower volume-to-capacity ratios. In fact, the trend takes a sharp upward slope during nighttime hours. Further examination of the nighttime data points indicates that those with the highest crash rates occurred between midnight and 5:00 a.m. This finding suggests that lighting level or a lack of

45 0.0 0.5 1.0 1.5 2.0 2.5 3.0 0.00 0.05 0.10 0.15 Volume-to-Capacity Ratio C ra sh R at e, c r/m vm Night Day 0 50 100 150 200 250 300 350 0.0 0.2 0.4 0.6 0.8 1.0 Volume-to-Capacity Ratio C ra sh R at e, c r/1 00 m vm Multiple-Vehicle Single-Vehicle Night Day driver alertness may be the reason for the higher rates during late-night hours (as opposed to a low volume-to-capacity ratio). Figure 28. Relationship between total crash rate and volume-to-capacity ratio for highways. Zhou and Sisiopiku (1997) examined the relationship between volume-to-capacity ratio and total crash rate for three segments of I-94 in Detroit, Michigan. The three segments total 16 miles and experienced 5,047 crashes during a two-year period. The relationship between crash rate and volume-to-capacity ratio in this data is shown in Figure 29. Each data point represents one hour of the average day. Figure 29. Relationship between total crash rate and volume-to-capacity ratio for freeways.

46 0.0 0.1 0.2 0.3 0.4 0.5 0 1,000 2,000 3,000 4,000 Traffic Flow Rate, veh/h C ra sh F re qu en cy , c r/m i/y ea r . Single-Vehicle Multiple-Vehicle Rural Freeway 0.0 0.5 1.0 1.5 2.0 2.5 0.0 0.2 0.4 0.6 0.8 1.0 Volume-to-Capacity Ratio C ra sh F re qu en cy , c r/m i/y ea r . Single-Vehicle b0 = 0.0025, b1 = 1.00, b2 = -5.0 Multiple-Vehicle b0 = 0.00041, b1 = 1.00, b2 = 0.37 Urban Freeway The triangle symbols in Figure 29 represent single-vehicle crash rates. The circular symbols represent multiple-vehicle crash rates. The dark (or solid) symbols represent nighttime hours. The open symbols represent daytime hours. The trend lines shown represent the relationships found for the daytime crash rates. Similar trends were not as clear in the nighttime data. However, if the trend lines are extrapolated to the lower ratios, then the nighttime crash rates would appear to be above those for the daytime (suggesting that crash risk is higher during nighttime hours). In Figure 29, there does not appear to be a difference between nighttime and daytime crash rates for volume-to-capacity ratios above 0.4. These nighttime rates correspond to the hours from 8:00 p.m. to 11:00 p.m. This trend is similar to that for rates of 0.06 or more in Figure 28. It suggests that a lack of driver alertness is the likely reason for the higher rates during late-night hours. Trends similar to those shown in Figure 29 for multiple-vehicle and single-vehicle crashes were also found by Martin (2002). He examined crash rates on interurban motorways in France (primarily tollways). He did not find a significant difference between day and night crash rates. However, he did find that crashes are 1.17 times more severe at night; reflecting frequent high-speed single-vehicle crashes resulting from drowsiness. An analysis of freeway crash data by Lord et al. (2005) also indicates that both crash frequency and crash type vary with volume-to-capacity ratio. Both relationships are shown in Figure 30. The two trend lines shown in each figure indicate that the safety influence of volume- to-capacity ratio varies with crash type. The trends shown in Figure 30a were noted previously in Figure 10 for ramps and have been found by Persaud and Mucsi (1995) for two-lane rural roads. a. Traffic demand and crash frequency. b. Volume-to-capacity ratio and crash frequency. Figure 30. Relationship between traffic demand, crash type, and total crash frequency. The predictive model used by Lord et al. (2005) to fit the trends shown in Figure 30 is represented by the following equation. cv b ho CMFVLYbN /1= (7)

47 0.0 0.5 1.0 1.5 2.0 2.5 0.0 0.2 0.4 0.6 0.8 1.0 Volume-to-Capacity Ratio C ra sh M od ifi ca tio n Fa ct or Single-Vehicle, b2 = -5.0 Multiple-Vehicle, b2 = 1.0 Multiple-Vehicle, b2 = 0.4 Single-Vehicle, b2 = -0.5 with, c Vb cv h eCMF 2/ = (8) where, N = estimate of expected crash frequency, cr; bi = regression coefficients, i = 0, 1, 2; L = segment length, mi; Y = time period of crash estimate, yr; Vh = traffic volume, veh/h; CMFv/c = crash modification factor for volume-to-capacity ratio; and c = capacity, veh/h. In application, Equation 7 is calibrated using volume and crash data representing a common time period (say one year), where the volume is a specified hourly volume for the average day of the year and the crash data represent the count of crashes during the same hour for the entire year. Equation 7 is separately calibrated for single-vehicle crashes and for multiple- vehicle crashes. The use of the exponential relationship in Equation 8 is supported by the trends lines shown in Figures 28 and 29. A review of the values of b2 that provide a best fit to the data in these figures indicates that b2 typically has a value between -0.5 and -5.0 for single-vehicle crashes. In contrast, it has a value between 0.4 and 1.0 for multiple-vehicle crashes. These ranges are consistent with the models calibrated by Lord et al. (2005). The sensitivity of the factor value to the typical range of values for b2 is shown in Figure 31. The structure of Equations 7 and 8 suggests that the regression coefficients associated with the traffic-volume variable will be correlated. This condition requires the thoughtful assembly of calibration data to ensure that b1 and b2 are accurately quantified. Specifically, it requires multiple observations for each hourly volume level, where the capacity varies among sites at each volume level. Figure 31. Relationship between CMF value and volume-to-capacity ratio.

48 HOV Facilities This subsection describes the findings from a review of the safety literature related to HOV facilities on freeways. The focus of the discussion is on HOV facilities that are integral to the freeway cross section. A review of national trends in HOV facility development was undertaken by Fuhs and Obenberger (2002). They found that over 130 HOV facilities were operating on freeways in 23 metropolitan areas as of 2001. About 85 percent of the facilities used a “2+” eligibility policy and 50 percent of the HOV facilities operated on a part-time or time-of-day basis. Fuhs and Obenberger (2002) estimate that HOV facilities represent 1,200 route-miles of freeway (the cumulative lane mileage is twice this amount). Their examination of trends over a 30-year period prior to 2001 indicates that HOV route-miles are growing at a rate of 10 percent per year, with a forecast of 1,800 route-miles for 2009. This forecast mileage would constitute 3 percent of the freeway mileage identified in Table 1. In 2007, the U.S. DOT’s Research and Innovative Technology Administration (Research, 2007) conducted a survey of 94 metropolitan areas. The survey results indicate that 14 metropolitan areas (15 percent) have HOV facilities. These facilities constitute 4 percent of the total freeway mileage that exists in the 94 metropolitan areas. Basic Descriptors This subsection describes various descriptors of HOV lane design and operation. Topics addressed include travel direction (relative to the adjacent general-purpose lane), HOV access type, and lateral separation. Travel Direction. HOV lanes can be described as having concurrent flow operation or contra-flow operation. The distinction between these two types is determined by the nature of the travel direction in the HOV lane relative to that in the adjacent general-purpose lane. Figure 32 illustrates each type of travel direction. The concurrent flow operation has the vehicles in the adjacent lane and the HOV lane traveling in the same direction. The contra-flow operation has vehicles in the HOV lane and the adjacent lane (on the same side of the median) traveling in opposing directions. A review of HOV operation by Fuhs and Obenberger (2002) indicates that about 95 percent of all HOV facilities use concurrent flow operation. Cothron et al. (2004) evaluated crash data for a 5.6-mile contra-flow HOV lane on I-30 in Dallas, Texas. A movable concrete barrier is provided on both sides of the HOV lane. FI crash data were gathered for six years before the HOV lane was opened. It was also collected for nine years after the lane was opened. One year of before data and five years of after data were excluded from the analysis due to major reconstruction projects that occurred in the corridor during those years. The remaining data were analyzed for this report using a regression-based before-after analysis. A 14-percent reduction in FI crashes was found to occur after the HOV lane was opened. Details of the reconstruction projects were not provided, so it is difficult to determine whether these projects contributed to the noted safety improvement.

49 a. Concurrent flow operation. b. Contra-flow operation. Figure 32. HOV travel direction types. Cothron et al. (2004) also evaluated crash data for two freeway segments with concurrent flow HOV lanes in Dallas, Texas. Both segments were 6.5 miles in length, had a three-foot “painted” (i.e., marked) buffer, and one HOV lane in each travel direction. FI crash data were gathered for five years before the HOV lanes were opened and for four years after they were opened. A re-analysis of these data indicate that crashes increased by about 50 percent on each of the two freeway segments. The speed differential between the HOV lane and the adjacent general-purpose lane was cited by Cothron et al. as a contributor to the observed degradation in safety. The reduced lane and shoulder width introduced in the freeway cross section to accommodate the HOV lane was also cited as a possible cause for the crash rate increase. HOV Access Type. Access to the HOV lane can be described as “continuous” or “limited.” Continuous access is provided when vehicles can enter or exit the HOV facility continuously along the freeway segment. Limited access is provided when vehicles can enter or exit the HOV facility only at designated entrance or exit points, as may be defined by pavement markings or physical barriers. A review of national trends in HOV facility design by Fuhs and Obenberger (2002) indicates that 28 percent of HOV facilities have continuous access and 72 percent have limited access. Jang et al. (2009) examined the relative safety of the two HOV access types. They examined crash rates for four freeway segments with continuous access (40.7 mi) and four segments with limited access (50.9 mi). The segments represent six freeways in California. The limited access segments have a buffer width that ranges from 1 to 5 ft. The reported crash rates for FI crashes indicate that facilities with continuous HOV access have 16 percent fewer crashes than facilities with limited HOV access. Lateral Separation. When an HOV lane has limited access, the separation between it and the adjacent general-purpose lane can be described as “barrier” or “buffer.” Barrier separation denotes the use of some type of physical object between the two lanes to discourage or prevent vehicles from entering or exiting the HOV lane. The barrier is usually a concrete safety-shaped barrier. A review of national trends in HOV facility design by Fuhs and Obenberger (2002) indicates that 48 percent of HOV facilities have buffer separation. Some type of barrier separation is provided for another 24 percent and the remaining 28 percent have no buffer (i.e., continuous access). HOV Peak HOV Peak Peak HOV HOV Peak

50 The relative safety of the two HOV access types was examined by Newman et al. (1988). They compared crash rates for three access/separation combinations: ● 15 freeway segments with continuous access (i.e., no buffer), ● 13 segments with limited access and 2-ft buffer, and ● 6 segments with limited access and a 13-ft buffer. The segments collectively represent eight freeways in California. All segments had concurrent operation. The first two combinations were noted to have no inside shoulder. The analysis by Newman et al. (1988) revealed that there was no difference in crash rate between the first two types identified in the previous bullet list. In contrast, the analysis revealed that the segments with a 13-ft buffer had a lower crash rate than the other two types. A re- analysis of these crash rates for this report indicates that the 13-ft buffer reduced total crash rate by 30 percent. The contribution of the inside shoulder width to these findings was not evaluated by Newman et al. Elements with Quantified Relationship This subsection describes various associations between safety and HOV facility design or operation that have been quantified through an analysis of crash data. Topics addressed include speed differential, lane or shoulder width reduction to add an HOV lane, inside shoulder width, HOV access location, and HOV access length. Speed Differential. Speed differential represents the difference between the speed in the HOV lane and that in the adjacent general-purpose lane. Where such differentials exist, the speed in the HOV lane is higher than that of the adjacent lane. Cothron et al. (2004) found that freeway crash rates increased following the opening of two limited access, buffer-separated HOV lane facilities. They rationalized that the increase may be due to the speed differential that was observed to range from 21 to 35 mi/h. Newman et al. (1988) examined crash rates for several freeway segments with HOV facilities, as described previously. One element of the database was speed differential. A regression model was fit to these data for this report. The relationship between speed differential and FI crash rate was statistically significant. It is shown in Figure 33. The trend line suggests that a 25-mi/h speed differential is associated with a 130 percent increase in crash rate.

51 0.0 0.2 0.4 0.6 0.8 1.0 0 5 10 15 20 25 Speed Differential, mi/h Fa ta l-a nd -In ju ry C ra sh R at e, cr as he s/ m vm Figure 33. Relationship between FI crash rate and speed differential. Reducing Lane and Shoulder Width to Add HOV Lanes. HOV lanes are sometimes added to the freeway cross section by reducing the width of the existing lanes or shoulders. This practice was used at two HOV facilities in Dallas, Texas and was cited by Cothron et al. (2004) as possibly contributing to the increase in crash rate following the opening of the facilities. Bauer et al. (2004) examined crash data for 124 freeway segments in southern California that underwent a reduction in lane or shoulder width to add a lane (typically an HOV lane). The treated sites were located on four freeways and totaled 48.9 mi. They found that projects that added one lane to a four-lane (per direction) cross section experienced an 11 percent increase in FI crash frequency. Those projects that added one lane to a five-lane cross section experienced a 7 percent increase in FI crash frequency. Inside Shoulder Width. Jang et al. (2009) examined the relationship between inside shoulder width and crash rate in the HOV lane. They found that, for a common access type, those HOV facilities with a shoulder width of 5 ft or less tended to have a total crash rate that was 100 percent larger than those with a shoulder width of 8 ft or more. HOV Access Location. The location of an HOV lane entrance or exit, relative to the nearest interchange ramp has been speculated to have some influence on crash frequency. This issue of access location is applicable only to HOV lanes with limited access. Jang et al. (2009) examined the relationship between HOV lane access location and freeway segment crash rate. They computed crash rates for 24 different freeway segments having either an HOV lane entrance or exit. When examined in the context of the distance to the nearest ramp, they found that HOV access points within 0.3 mi of the nearest ramp were often associated with a relatively large crash rate. These access points often had a high traffic volume in the HOV lane during the peak demand hours.

52 HOV Access Length. The length of an HOV lane entrance or exit has also been thought to have some influence on crash frequency. This issue of access length is applicable only to HOV lanes with limited access. Jang et al. (2009) computed crash rates for 24 different freeway segments with either an HOV lane entrance or exit. One characteristic of those segments with a relatively large crash rate was the presence of an HOV access length that is characterized as “short” (i.e., less than 0.25 mi long). Crossroad Ramp Terminal This subsection describes the findings from a review of the safety literature related to the crossroad ramp terminal. The discussion in this section is focused on the terminals at service interchanges, as opposed to system interchanges. With a couple of exceptions, the literature review did not identify research that is specific to the safety of crossroad ramp terminals. As a result, the discussion is broadened to include all types of intersections (not just those at interchanges). This discussion is intended to identify the various design and operational elements that have been found to have some safety influence on intersection safety—elements that are likely to have some influence on the safety of crossroad ramps terminals. This type of discussion helped to guide the development of a framework for safety evaluation (as documented in Chapter 3). However, an in-depth discussion of the reported influence of each element on safety is not provided given that its influence at a ramp terminal is unknown. Basic Descriptors This subsection describes various fundamental descriptors of crossroad ramp terminal design and operation. Topics addressed include area type, terminal configuration, and control mode. Area Type. This descriptor indicates the population density in the vicinity of the crossroad ramp terminal. The categories used are urban and rural. A comparison of crash rates for urban and rural intersections in Texas indicates that rural intersections typically have more crashes for common volume levels (Bonneson and Pratt, 2008; Bonneson et al., 2007). In recognition of the different type of highway environment found in urban areas, relative to rural areas, Part C of the HSM (Highway, 2010) describes separate safety predictive methods for urban and rural intersections. Terminal Configuration. The right-hand side of Figure 5 illustrates the configuration of six typical ramp terminals. Also shown are the left-turn and through movements at each terminal. The diamond interchange terminals typically have four legs and the parclo interchange terminals typically have three legs. The number and type of turn movements at each terminal tend to vary among the configurations. They translate into a different number of conflicting travel paths and conflicting volumes for each configuration. These characteristics suggest the possible need for

53 different SPFs for each configuration; however, no research has been identified that confirms this speculation. Bared et al. (2005) developed an SPF for 27 diamond interchanges in Washington. The ramp terminal spacing varied among these interchanges such that conventional, compressed, and tight urban diamonds were represented in the database. The database also was noted to include a mixture of urban and rural interchanges and a mixture of signalized and unsignalized interchanges. Bared et al. indicated that variables reflecting area type and control mode were included in the regression analysis but were found to not have statistical significance. The calibrated SPF for total crashes at diamond interchanges is shown in Figure 34. Figure 34. SPF for two interchange types based on total crashes. Bared et al. (2005) also reported crash data for 13 SPUIs that collectively represent five states. An SPF was fit to the data for this report. The form of the SPF is similar to that used for the diamond interchanges; however, an indicator variable was used to identify the SPUIs from Maryland. The five SPUIs in Maryland had less than half of the crashes that were reported for the other SPUIs, after the effect of traffic volume and crash years were removed. The resulting SPF is shown in Figure 34 for the “non-Maryland” SPUIs. The trends indicate that the SPUI may experience more crashes than does the diamond interchange at higher volume levels. Also shown in Figure 34 is the expected total crash frequency for an interchange with two, signalized four-leg ramp terminals in an urban setting. This estimate was obtained using the SPFs in ISAT. It was rationalized that this combination of conditions was most comparable to the other two interchange SPFs shown. Crashes that occur on the segment between ramp terminals were not included in the estimated expected crash frequency. It is noted that the SPFs provided in ISAT were calibrated using data for conventional intersections (as opposed to crossroad ramp terminals). The ISAT trend line shown in Figure 34 is notably different from the 0 5 10 15 20 0 10 20 30 40 50 Cross Street AADT (1000s), veh/day C ra sh F re qu en cy , c ra sh es /y r Ramp AADT = 0.2 x Cross Street AADT SPUI Diamond ISAT (Two, 4-Leg Urban Signals)

54 other two trend lines and is evidence of a likely difference in the safety of ramp terminals, relative to conventional intersections. Number of Legs. The number of legs at an intersection has been found to have some correlation with crash frequency. More precisely, the influence of leg count is a reflection of the likelihood that each leg facilitates arrival movements, departure movements, or both at the intersection. The number of times each movement crosses the path of another intersection movement represents a conflict, as well as exposure to the possibility of a collision. The number of conflicts for all movement combinations increases exponentially with the number of intersection movements and, thus, with the number of legs that support these movements. A comparison of FI crash rates for three-leg and four-leg intersections in Texas indicates that four-leg intersections typically have 50 to 100 percent more crashes than three-leg intersections for common volume levels (Bonneson and Pratt, 2008; Bonneson et al., 2007). In recognition of this influence, SPFs in Part C of the HSM (Highway, 2010) have been derived separately for three-leg and four-leg intersections. Control Mode. The control mode used at an intersection is generally characterized as two-way stop, all-way stop, and signalized. The control mode can vary among movements on each approach. Of particular note is the use of free or yielding right-turn movements on an approach for which the left-turn and through movements are stop or signal controlled. Research has shown that control mode influences crash frequency, crash severity, and collision type. A comparison of crash rates for two-way-stop-controlled and signalized intersections in Texas indicates that signalized intersections tend to have slightly fewer crashes than two-way- stop intersections in rural areas and for common volume levels (Bonneson et al., 2007). A similar comparison of crash rates for urban intersections indicates that the number of intersection legs has some interaction with control mode such that crash rates were found to be higher at signalized intersections with many lanes than two-way-stop intersections (Bonneson and Pratt, 2008). In contrast, crash rates were lower at signalized intersections with few lanes. In recognition of the influence of control mode, SPFs in Part C of the HSM (Highway, 2010) have been derived separately for signalized and two-way stop-controlled intersections. Elements with Quantified Relationship This subsection describes possible associations between safety and crossroad ramp terminal design or operation. Given the limited amount of information on crossroad ramp terminal safety, the discussion in this section is intended only to identify the design and operational elements that have been found to have an influence on the safety of a conventional intersection. Table 6 provides a summary of these elements and their relationships to safety. The table content is based on a review of the HSM (Highway, 2010) and the reported findings from a previous review by Bonneson et al. (2005).

55 TABLE 6. Elements that may influence the safety of crossroad ramp terminals Category Element Safety Relationship Geometric design Left-turn lane or bay presence Addition of bay correlated with a reduction in crash frequency. Right-turn lane or bay presence Addition of a bay correlated with a reduction in crash frequency. Number of lanes on the major or minor road Additional lanes at a signalized intersections are correlated with a larger crash frequency. The reverse trend applies to two-way stop- controlled intersections. Skew angle Smaller skew angle is correlated with a smaller crash frequency. Median presence and width Wider median at two-way stop-controlled intersections is correlated with a smaller crash frequency. Outside shoulder width Narrow shoulders are correlated with a larger crash frequency. Lane width Narrow lanes are correlated with a larger crash frequency at urban intersections. Right-turn channelization (free right) Urban intersections with right-turn channelization are associated with a larger crash frequency. Access Driveway presence Driveway presence is correlated with an increase in crash frequency. Operation Left-turn signal phasing Intersections with protected or protected-permissive left-turn phasing are correlated with smaller crash frequency. Right turn on red Urban intersection approaches for which right turn on red is prohibited have a smaller crash frequency. Other Truck percentage Rural two-way stop-controlled intersections with a higher percentage of trucks are associated with fewer crashes. The reverse trend applies to signalized intersections. Lighting presence Addition of lighting is correlated with a reduction in nighttime crash frequency. Red light camera operation Use of red light enforcement cameras is associated with a decrease in right-angle crashes and an increase in rear-end crashes. Related Topics This subsection describes wrong-way traffic movements at crossroad ramp terminals and explores possible configurations and design elements that may influence the frequency of these maneuvers. A problem inherent to service interchanges is the potential for wrong-way entry into an exit ramp (Policy, 2004). While the maneuver is not frequent, it has the potential to result in a severe crash. A study by Cirillo et al. (1969) indicates that about 5 percent of all fatalities on the Interstate Highway System in the 1960s were attributable to crashes resulting from wrong-way movements. The problem of wrong-way driving on freeways was examined by several researchers in the 1970s and the findings were summarized by Leisch et al. (1982). The parclo A (2-quad) and

56 Use of sharp corner radii to discourage wrong-way turns. parclo B (2-quad) interchange types (see Figure 3) were noted to be particularly susceptible to wrong-way movements because the ramp approach and departure legs are located on the same side of the crossroad and are typically located very close to one another. A “half-diamond” (with one missing entrance ramp and one missing exit ramp) and other “incomplete” (or partial) interchanges were also noted to be associated with a large number of wrong-way maneuvers. Leisch et al. (1982) noted that about 75 percent of all wrong-way movements occurred at night, during periods of low volume with little traffic on the crossroad to light the roadway and cue other drivers. Countermeasures that were identified include “Do Not Enter,” “Wrong Way,” “No Left Turn,” and “No Right Turn” signs posted at the crossroad ramp terminal. Also identified as a countermeasure was the use of large pavement arrows on the ramp approaches. Design-related techniques identified in the Green Book (Policy, 2004) that have been used to minimize wrong-way maneuvers include channelization and curb return modifications. Figure 35 illustrates how median channelization on the crossroad can be extended slightly into the intersection to provide shadowing of the approach leg to discourage improper left turns into the exit ramp. Figure 35. Designs to discourage wrong-way maneuvers. Another technique to prevent wrong-way maneuvers is to use a short-radius curve, or angular break, at the intersection of the left-edge of the exit ramp approach with the right edge of the crossroad approach. This technique is also shown in Figure 35. It should discourage improper right-turns into the exit ramp. A more recent examination of U.S. fatal crash data for the late 1990s by Moler (2002) indicates that about 0.8 percent of all fatalities (i.e., 350 persons/yr) are attributable to wrong- way maneuvers. This percentage represents a significant reduction in wrong-way fatalities since the 1960s and likely reflects agency use of the aforementioned countermeasures.