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.

69 CHAPTER 4: DATABASE DEVELOPMENT This chapter describes a summary of the data assembled for safety prediction model development and calibration. The database is founded on the road inventory data obtained from the HSIS. These data were enhanced through the inclusion of road inventory data extracted from aerial photographs. The enhanced database was then combined with crash data (also obtained from HSIS) to form the highway safety database needed for model development and calibration. This chapter consists of two parts. The first part summarizes the procedures used to assemble the database. The second part summarizes the database contents using categorical descriptions and various statistics. It also provides a discussion about some trends observed in the data. DATA COLLECTION PROCEDURES The data collection process consisted of a series of activities that culminated in the assembly of a highway safety database suitable for the development of a comprehensive safety prediction methodology for freeways and interchanges. It consisted of the following activities: ● Develop data reduction procedures guide. ● Develop enhanced data collection process and software. ● Assemble road inventory database from HSIS data. ● Enhance safety database. ● Merge road inventory and crash data. Each of these activities is described briefly in the following sections. Develop Data Reduction Procedures Guide It was determined that a robust safety prediction methodology would require the use of a cross-sectional study approach and that the data would need to be attribute-rich and of high quality. To facilitate the development of this database, supplemental road inventory data were added to the HSIS databases. A series of data reduction procedures guides were developed to ensure consistency in the collection of supplemental road inventory data. One guide was developed for freeway segments and speed-change lanes. A second guide was developed for interchange ramp segments. A third guide was developed for crossroad ramp terminals. Each guide described the variables obtained from supplemental data sources (i.e., aerial photographs and interchange diagrams). These data were collected as part of the database enhancement activity, described in the next section. Each guide included a definition for each variable as well as the technique for measuring it. The technicians that comprise the data collection team were trained from this document to ensure consistency in the data collection process.

70 Develop Enhanced Data Collection Process and Software Data enhancement consisted of using supplemental data sources to acquire additional data for each freeway segment, ramp segment, and crossroad ramp terminal in the state database. A key element of this process is the use of aerial photography to extract additional road inventory data for each segment or terminal. These photographs are available from several Internet sources. The data being collected include the width of key cross section elements, barrier location, horizontal curvature, ramp configuration, turn bay presence, median type, etc. Data extraction proceeded on a segment-by-segment basis. The database is represented in a spreadsheet. One row of the spreadsheet represents one segment or crossroad ramp terminal. Each variable is assigned to a column. The supplemental data for each segment or intersection are extracted from an aerial photograph or a road-level photograph. The aerial photographs were obtained from Google Earth and the road-level photographs were obtained from its companion tool, Street View. Google Earth is software available from Google ©. Two procedures were developed for the collection of supplemental data. One procedure was used for data that required the technician to make a determination of presence or condition (e.g., presence of rumble strips, intersection control type, presence of an HOV lane, presence of a channelizing right-turn lane). This data is typically categorical. This procedure is referred to as the “manual” procedure. The other procedure is based on the digitization of key roadway design elements using aerial photographs and software tools. Using this process, the road alignments, cross sections, barrier pieces, and speed-change lanes were manually “digitized” such that relevant points are located using geodetic coordinates. The digitized points were saved to an electronic file and then processed by software to produce the desired widths, lengths, and radii. The data is typically continuous. This procedure is referred to as the “semi-automatic” procedure. It is described more fully in Appendix B. Assemble Road Inventory Database from HSIS Data The database assembly activities consist of three tasks. The first task involved processing the HSIS data to construct a road inventory database with a wide range of variables. The second task involved subsetting the processed data such that the resulting database included a wide range of segment representation (e.g., number of lanes, urban/rural). The need to subset the data was motivated by the recognition that project resources would limit the subsequent collection of supplemental data to only a subset of the full database. The third task involved digitizing the road alignments for the purpose of defining the geodetic coordinates of each segment’s begin milepost. Database Construction There were several steps associated with the database construction task. It produced nine separate databases. Each database represents one combination of state (California, Maine, Washington) and roadway component (freeway segment, interchange ramp proper, crossroad

71 ramp terminal). Speed-change lane data were included in the freeway segment database. This task included the following steps: ● Merge the roadlog, curve, grade, special-use-lane, and other road- or traffic-volume- related files to form a unified file that fully describes each road segment. ● Rename variables for consistency across state databases. ● Convert data for common variables to a common format for consistency across databases. ● Verify that there were no changes in geometry during the analysis years. ● Convert, combine, or manipulate database variables to compute the variables identified in Chapter 3 as being associated with a desired SPF or CMF. ● Identify freeway segments, ramp segments, and ramp terminals for which all data are complete. ● Identify freeway segments and ramp segments that have special-use lanes. ● Identify freeway segments adjacent to ramps and link relevant ramp attributes. ● Identify crossroad ramp terminals by manually linking the intersecting ramps and crossroad segment. ● Confirm that all selected segments satisfy the criteria for minimum length, minimum exposure, and maximum length (described in the following paragraphs). Analysis Period. A three-year analysis period was established for the freeway segment and the crossroad ramp terminal databases. A five-year period was established for the interchange ramp segments. In general, a three-year duration is preferred because it minimizes the potential for changes in geometry, traffic, or environment over the analysis period. However, the frequent occurrence of segments with zero crashes can degrade the accuracy of the statistical analysis. For this reason, a longer, five-year period was used for ramps because of their low volume and shorter length. Site Selection Criteria. The freeway and ramp segments were selected in part based on a minimum-length and a maximum-length criterion. Specifically, each segment had to have a length that equaled or exceeded the precision of their crash location variables. Based on this consideration, the minimum segment length for California and Washington data was 0.01 miles. For the Maine data, the minimum length was 0.10 miles. It was rationalized that the geometric homogeneity of a segment would become more suspect with increasing segment length. For this reason, a maximum-length criterion was somewhat subjectively established at 1.6 miles. This length was established based on experience gained by evaluating the homogeneity of many segments using aerial photographs. The presence of managed lanes was used as a site selection criterion. Freeway segments with managed lanes were not included in the database. This action was taken to provide a focus on basic freeway segments with general-purpose lanes. The entrances, exits, and separation elements associated with managed-lane facilities was reasoned to provide a complicated safety influence that would justify the separate development of a managed-lane safety prediction method. A minimum exposure criterion was also established for freeway segments. Segments with low exposure (i.e., short length or low traffic volume) have a large residual value when they are

72 associated with one or more crashes. This residual often exerts undue leverage on the regression model coefficients and increases the error variance beyond that explained by the negative binomial distribution. To avoid these issues, all candidate segments were screened such that only those segments with a minimum level of exposure were included in the regression database. Equation 9 was used to compute the minimum segment exposure. yBase PRPR E 2 4)2(2 222 min −+−+ = (9) where, Emin = minimum segment exposure, million-vehicle-miles (mvm/yr); PR = prediction ratio (= 3.0); y = analysis period duration, yr; and Base = FI crash rate, crashes/mvm. The prediction ratio PR is the standardized residual for an observation. It is set to 3.0 in Equation 9. This value corresponds to a 99.7th percentile confidence interval. Segments with low exposure and one or more crashes will have a prediction ratio in excess of 3.0—a condition that should occur less than 0.3 percent of the segments that satisfy this criterion. The crash rate used in this equation represents FI crashes for typical freeway segments, as determined during the literature review documented in Chapter 2. These rates and the corresponding minimum exposure values are listed in Table 11. TABLE 11. Base FI crash rates for minimum exposure criteria Area Type Number of Lanes Analysis Period Duration, yr FI Crash Rate, cr/mvm 1 Minimum Exposure, mvm/yr Urban 4 3 0.24 0.13 6 3 0.36 0.08 8 3 0.54 0.06 10 3 0.56 0.05 Rural 4 3 0.14 0.22 6 3 0.21 0.15 8 3 0.27 0.11 Note: 1 - mvm: million vehicle-miles. Manual Assembly Elements. The HSIS data for Maine describes the roadway using a link-node system. Each link for a divided roadway represents one travel direction. In contrast, the HSIS data for California and Washington describe the roadway using two-directional road segments uniquely identified by the state-established milepost referencing system. To facilitate the calibration of models using a common segment definition, the data for Maine had to be manually processed on a link-by-link basis. The manual process was used to convert each link into a set of two-directional road segments with an artificially established milepost referencing system. Once

73 this segment-based file was assembled, the construction of the road inventory database for Maine proceeded in accordance with the steps outlined in the previous bullet list. A manual assembly process was necessary to assemble the crossroad ramp terminal database for each state. The “intersection” files provided in HSIS for these states was inadequate for this purpose because it was tailored to typical intersections of two state routes. With the manual process, candidate ramp terminals were initially identified in the HSIS ramp data. Interchange maps provided by Maine and by Washington were used to locate the individual ramps that comprise a crossroad ramp terminal. Descriptive variables in the ramp data were used with Google Earth to locate ramp terminals with the California data. Then, once the ramps were located, the HSIS data were interrogated to identify the crossroad segment at the point of intersection with the ramps. As a final step, the road inventory data for the intersecting ramps and crossroad segment were extracted from the HSIS database and combined into a common intersection database. At this point, the construction of the road inventory database proceeded in accordance with the steps outlined in the previous bullet list. Level of Effort Indicators. The construction process outlined in the previous bullet list was facilitated using Statistical Analysis Software (SAS) data manipulation and merge procedures. A separate SAS processing code was developed for each of nine databases (i.e., three states and three components for each state). The code reflects the unique nature of each state’s data and its treatment of each roadway component. The lines of code developed for this activity are identified in Table 12. They provide an indication of the magnitude of the development effort. TABLE 12. SAS Processing Effort Roadway Component Task Lines of SAS Code by State California Maine Washington Freeway segment and freeway speed- change lane Identify segments for manual file assembly 0 510 0 Construct road inventory database 1,560 1,590 1,890 Merge crash data and categorize 540 910 760 Total: 2,100 3,010 2,650 Ramp proper (including collector- distributor ramps) Identify segments for manual file assembly 0 0 0 Construct road inventory database 750 470 630 Merge crash data and categorize 560 550 580 Total: 1,310 1,020 1,210 Crossroad ramp terminal Identify intersections for manual file assembly 510 500 970 Construct road inventory database 1,200 1,320 940 Merge crash data and categorize 570 620 530 Total: 2,280 2,440 2,440 Table 12 also lists the lines of SAS code needed for the “merge crash data and categorize” task. This code and related effort is separate from the assembly and construction process described in this section. The crash data merging activity is described in a subsequent section.

74 Database Attributes. Table 13 indicates the number of database variables in each of the nine databases that were assembled. The first row for each roadway component identifies the number of variables obtained from the HSIS database that are considered to be useful to the development of safety prediction models. Examination of Table 13 indicates that relatively few of the HSIS variables for crossroad ramp terminals and for Maine freeways were found to be directly useful for model development. TABLE 13. Database attributes Roadway Component Attribute Attribute Value by State California Maine Washington Freeway segment and freeway speed- change lane Variables from state database 48 4 76 Variables created from state database 31 56 48 Variables added to database 95 109 95 Total: 174 169 219 Ramp proper (including collector- distributor ramps) Variables from state database 26 33 48 Variables created from state database 5 8 5 Variables added to database 53 52 49 Total: 84 93 102 Crossroad ramp terminal Variables from state database 4 2 1 Variables created from state database 47 84 39 Variables added to database 32 75 46 Total: 83 161 86 The second row for each roadway component in Table 13 identifies the number of HSIS variables that were combined using SAS code to create new variables. For example, the horizontal curve database available from HSIS for Washington was used to link curves to segments and to determine the proportion of the curve on each segment. Also, the ramp databases from HSIS for both California and Washington were used to link ramps to freeway segments, such that a variable was added to the freeway segment database that relates the distance between the ramp and segment. The third row for each roadway component identifies the number of variables that were obtained from supplemental sources as part of the data enhancement activity. The details of this activity are described in the next section. The numbers provided in Table 13 indicate that the variables from supplemental sources tended to account for about one-half of the variables in each database. Sampling Technique There are several thousand freeway segments in the HSIS databases for California and Washington. There are also many ramp segments represented in these two databases. However, project resources limited the collection of supplemental data to only a subset of the full database.

75 Thus, a sampling technique was developed to select a cross section of freeway (and ramp) segments, such that a uniform distribution of values for several key variables was obtained. Separate sampling techniques were developed for freeway segments and ramp segments. For freeway segments, the key variables included area type (i.e., urban or rural), number of lanes, right shoulder width, and median type. For ramps, the key variables included area type, ramp type (i.e., entrance, exit), and ramp configuration. A sampling technique was not required for several of the databases. Specifically, the number of freeway and ramp segments in the HSIS database for Maine was sufficiently limited that all segments that satisfied the site selection criteria were included in the assembled databases. No sampling was used for the selection of target crossroad ramp terminals. The need for full representation in the HSIS database tended to limit the number of available ramp terminals such that the entire database for each state had to be examined to maximize the potential for exceeding the minimum sample size. The most limiting requirements for ramp terminal selection were: (1) the adjacent crossroad segment had to be on the state highway system, (2) traffic volume was available for the intersecting ramp terminals, and (3) the ramp terminal was more than 500 ft from the nearest, non-ramp signalized intersection. The first criterion ensured that a complete crash history was available for the terminal. The third criterion reasonably ensured that ramp-terminal-related crashes could be isolated and assigned to the ramp terminal of interest. All of the ramp terminals that satisfied these criteria were considered “target” intersections. Table 14 indicates the number of target segments and crossroad ramp terminals that were assembled. The sample size for Maine represents all available segments in the HSIS database. Similarly, the sample size for crossroad ramp terminals represents all available terminals in the HSIS database. TABLE 14. Database target sample size Category Roadway Component Sample Size by State 1 Calif. Maine Wash. Total Number of segments or ramp terminals Freeway segment and freeway speed-change lane 613 213 1,428 2,254 Ramp proper (including collector-distributor ramps) 412 209 1,292 1,913 Crossroad ramp terminal 223 62 412 697 Total: 1,248 484 3,132 4,864 Total segment length, miles Freeway segment and freeway speed-change lane 254 107 241 602 Ramp proper (including collector-distributor ramps) 66 49 143 258 Total: 320 156 384 860 Average segment length, miles Freeway segment and freeway speed-change lane 0.41 0.50 0.17 0.27 Ramp proper (including collector-distributor ramps) 0.16 0.23 0.11 0.13 Note: 1 - Underlined values represent a subset sample of the HSIS data. All other “sample size” values represent all of the segments or intersections found in the HSIS data that satisfy the site selection criteria.

76 Geo-Location of Segments The third task of the database assembly activity was the geo-location of each segment and ramp terminal in the database. The location process was based on the use of Google Earth and Earth Tools software. Earth Tools software is described in Appendix B. Google Earth was used to digitize the road alignment for a group of adjacent segments. The digitized alignment was then submitted to Earth Tools software along with the segment milepost data. The software used this information computed the geodetic coordinates of each segment begin milepost. The suitability of each segment for subsequent supplemental data collection was also evaluated at this time. Specifically, a segment was considered to be “suitable” if the aerial photograph’s resolution was sufficient to discern the pavement markings and if there was no evidence of construction activity during the analysis period. The geo-location process was needed to facilitate the collection of supplemental data. The coordinates for each segment or intersection were subsequently used by the technicians to “return” to the location using Google Earth so they could collect their assigned data. An added benefit of the digitized alignments is that they could be used (with Earth Tools) to compute horizontal curve geometry for no additional time investment. Enhance Safety Database Database enhancement consisted of using supplemental data sources to acquire additional data for each target segment or crossroad ramp terminal. The data enhancement activity focused on two tasks. One task was the development of data describing for each segment the proportion of hours per day that are congested. Supplemental data were collected and used to derive this proportion for freeway segments. Automatic traffic recorder (ATR) data for the station nearest to each freeway segment were used for this purpose. These ATR data were acquired from the appropriate state agencies. The second task of the data enhancement activity was the use of aerial photography to collect additional data for each segment or ramp terminal. These photographs were obtained from Google Earth. The data collected include the width of key road cross section elements, barrier presence and location, horizontal curvature, ramp configuration, turn bay presence, and median type. The number of variables added to each database is identified in Table 13. Data reduction guidelines and software tools were used to make the collection of supplemental data as consistent and efficient as possible. The development of these guides and tools was described in a previous section. Some of the cross section data and curvature data that were collected using the semi- automated procedure were compared with the equivalent variables provided in the HSIS data. The findings from these comparisons were documented in Appendix B. For the cross section data, the findings indicated that there was a weak correlation between the HSIS data and the measured data. For the curvature data, there was fairly good agreement between the HSIS data and measured data.

77 The extraction and processing of supplemental data was the most time-consuming of all the data collection activities. Technicians were used for most of the extraction activities. Some of the more complicated activities (e.g., developing alignment files for segment location and curvature measurement) were completed by an engineer. It is estimated that the data extraction effort required about 0.45 h (= 0.30 h of technician time and 0.15 h of engineer time) for each target segment or crossroad ramp terminal in the database. It is estimated that the processing and management efforts required about 0.15 h of engineer time for each target segment or crossroad ramp terminal. Merge Road Inventory and Crash Data This activity involved merging the crash data with the enhanced road inventory databases. The crash data were obtained from HSIS. The objective was to assemble the highway safety database that included all of the relevant data for model calibration. A separate safety database was assembled for each combination of state and roadway component. File Merge Issues A key component of this activity was the association of each crash with a roadway component. There were three main issues related to this effort. They are described in the following paragraphs. Link-Node Conversions. One issue that was encountered related to the link-node system in the Maine HSIS data. This system assigns crashes to links and nodes, where a link represents one direction of travel on divided roadways and a node represents the endpoint of a link (often an intersection or ramp terminal). A link is identified by the numbers of the two nodes that bound it. Crash location along a link is identified by its distance from the lowest node number. The node numbers were found to be randomly assigned by Maine DOT to a node and have no relationship to each other, the direction of travel, or distance along the highway. Furthermore, the distance identified in the HSIS data for a given link often disagreed with the actual length measured on road maps or aerial photographs. These issues were compounded by Maine DOT’s conversion to a new node numbering system in 2007. The interchange maps provided by Maine DOT used the 2007 node numbers and the HSIS data used the pre-2007 node numbers. This issue was overcome through the development of additional SAS algorithms and link distance scaling techniques. Speed-Change Crash Identification. A second issue that was encountered related to the identification of freeway speed-change-related crashes. None of the HSIS databases includes crash attributes that could be used to identify with certainty whether a crash was related to the speed-change lane. This issue was resolved by assuming that all crashes located (by milepost) on the freeway segment between the mileposts that define the speed-change lane are speed-change- lane-related crashes. The location of a speed-change lane was defined to match the “ramp entrance length” and the “ramp exit length” dimensions shown in Figure 11. Crossroad-Ramp-Terminal-Related Crash Identification. A third issue that was encountered related to the identification of crossroad-ramp-terminal-related crashes. The

78 Washington HSIS database includes a variable that indicates whether a crash is related to the intersection’s activity, behavior, or control. This variable was evaluated using various cross- variable tabulations and found to be reasonably accurate in identifying crossroad-ramp-terminal- related crashes. In contrast, the HSIS databases for California and Maine do not include a variable that identifies intersection-related crashes. When developing robust safety prediction models, it is important to have criteria for identifying intersection-related crashes that is uniform from state to state. Vogt (1999) examined this issue in some detail and developed the following criteria for identifying intersection-related crashes: 1. The crash must occur within 250 ft of the intersection center, and 2. The crash must have one or more of the following attributes: a. involve a pedestrian; b. one vehicle involved in the crash is making a left turn, right turn, or U turn; or c. if two or more vehicles are involved, manner of collision is sideswipe, rear end, or angle. The criteria developed by Vogt have a logical dependence on the distance between the crash and the intersection. It is defensible when applied to typical highway intersections; however, it may not be adequate for identifying ramp-terminal-related crashes on ramp segments. In fact, many rear-end crashes on exit ramps are related to the ramp terminal even when they occur more than 250 ft from the terminal because of extensive terminal-related queuing that occurs on some exit ramps. Bauer and Harwood (1998) evaluated 100 reports for rear-end crashes on exit ramps and found that 95 percent of these crashes were related to the operation of the crossroad ramp terminal, regardless of the distance between the crash and the terminal. Based on these findings, the criteria in Table 15 were developed and used to identify crossroad-ramp-terminal-related crashes. It is noted that crash location, in terms of its distance to the subject crossroad ramp terminal, is available in both the California and Maine HSIS databases. Level of Effort Indicators The merging of the road inventory and crash databases was facilitated using SAS data manipulation and merge procedures. A separate SAS processing code was developed for each of the nine databases. Each code reflects the unique nature of each state’s data and its treatment of crash records. The lines of code developed for this activity are identified in Table 12. They illustrate the magnitude of the development effort.

79 TABLE 15. Criteria for defining crossroad-ramp-terminal-related crashes State Leg Criteria Washington Crossroad 1. The crash must occur within 250 ft of the terminal and have one of the following attributes: a. intersection related; b. at intersection; c. at driveway, or d. driveway related. Ramp 1. The crash must have one or more of the following attributes: a. intersection related; or b. at intersection. OR 2. The ramp is an exit ramp and the manner of collision is rear end. California and Maine Crossroad 1. The crash must occur within 250 ft of the terminal and have the following attribute: a. at intersection. OR 2. The crash must occur within 250 ft of the terminal and have one or more of the following attributes: a. involve a pedestrian; b. one vehicle involved in the crash is making a left turn, right turn, or U turn; or c. if two or more vehicles are involved, the manner of collision is sideswipe, rear end, or angle. Ramp 1. The crash must satisfy the same criteria as specified for the crossroad. OR 2. The ramp is an exit ramp and the manner of collision is rear end. DATABASE SUMMARY This part of the chapter summarizes the data assembled for the purpose of calibrating the predictive models needed to evaluate the safety of freeways and interchanges. The discussion herein is organized in terms of the freeway and interchange components that comprise the freeway system. They are: ● freeway segment, ● interchange ramp, ● crossroad ramp terminal, and ● freeway speed-change lane. A separate section is devoted to the discussion associated with each component. The discussion of in this part of the chapter focuses on FI crashes. This focus is intended to minimize differences in reporting threshold (both formal and informal) that may vary among the states represented in the database. FI crash data are more consistently reported throughout the U.S. and thus, they provide a more uniform basis for comparing crash trends among different jurisdictions. Evidence of the variability encountered when using property-damage-only (PDO) crashes for comparison is provided by Zegeer et al. (1998). Their examination of rural freeway crashes in

80 four states found that the percentage of PDO crashes varies from 60 to 77 percent. Slightly smaller variation was found for urban freeways. They speculated that this variation is due to differences in reporting threshold, as opposed to differences in freeway safety. Table 16 lists the final sample size for each of the nine databases that were assembled. The numbers in this table can be compared with the target sample sizes identified in Table 14. In many instances, the final sample size is lower than the target sample size. This reduction reflects decisions made during the database enhancement activity. In some cases, Street View photographs were not available during the enhancement activity such that barrier could not be accurately located. In other instances, road construction activities missed during the database assembly activity were discovered during the enhancement activity. For a variety of reasons, about 18 percent of the target segments or intersections were eliminated during the enhancement activity. TABLE 16. Database sample size Category Roadway Component Sample Size by State Calif. Maine Wash. Total Number of segments or ramp terminals Freeway segment and freeway speed-change lane 533 203 1,144 1,880 Ramp proper (including collector-distributor ramps) 405 209 923 1,537 Crossroad ramp terminal 216 62 291 569 Total: 1,154 474 2,358 3,986 Total segment length, miles Freeway segment and freeway speed-change lane 209 101 200 510 Ramp proper (including collector-distributor ramps) 65 49 114 228 Total: 274 150 314 738 Average segment length, miles Freeway segment and freeway speed-change lane 0.39 0.50 0.17 0.27 Ramp proper (including collector-distributor ramps) 0.16 0.23 0.12 0.15 Freeway Segment This section describes the freeway segments represented in the safety database. Initially, the traffic and geometric characteristics of each segment are presented. Then, the segment crash data are examined. The data presented for freeway segments includes segments with speed- change lanes. This approach is taken to provide a “complete” picture of freeway segment elements; it recognizes the difficulty of accurately isolating speed-change-related crashes. A subsequent section provides a more detailed examination of the crash characteristics of freeway speed-change lanes. Segment Characteristics A total of 1,880 freeway segments are represented in the combined freeway segment database. These segments represent about 209, 101, and 200 miles of freeway in California, Maine, and Washington, respectively. Selected segment characteristics are provided in Table 17.

81 TABLE 17. Summary characteristics for freeway segments State Area Type Through Lanes Total Segments Total Length, mi Seg. Length Range, mi Volume Range, veh/day Minimum Maximum Minimum Maximum Californ ia Rural 4 83 42.1 0.13 1.3 17,000 74,000 6 56 31.9 0.13 1.6 45,300 139,000 8 64 32.8 0.13 1.5 71,700 143,000 Urban 4 86 28.4 0.11 1.1 26,000 97,700 6 111 40.2 0.11 0.8 55,000 194,000 8 80 20.2 0.11 0.8 104,000 270,000 10 53 13.2 0.12 0.5 198,000 308,000 Overall: 533 208.8 0.11 1.6 17,000 308,000 Maine Rural 4 116 66.4 0.12 1.2 11,300 62,000 6 17 7.9 0.12 1.3 42,600 58,000 8 Urban 4 67 25.7 0.12 1.1 11,400 69,500 6 2 0.5 0.18 0.3 56,200 62,000 8 1 0.2 0.19 0.2 83,700 83,700 10 Overall: 203 100.6 0.12 1.3 11,300 83,700 Washin gton Rural 4 179 81.0 0.03 1.6 9,600 55,800 6 312 49.6 0.01 0.5 44,000 110,000 8 Urban 4 266 28.5 0.01 0.4 18,400 98,300 6 272 27.6 0.01 0.3 38,800 121,000 8 115 13.7 0.01 0.4 89,300 197,000 10 Overall: 1,144 200.4 0.01 1.6 9,600 197,000 The number of segments for each combination of area type and through lanes is indicated in column four of Table 17. The numbers shown for California and Washington are relatively uniform among rows and reflect the sampling technique described in a previous section. Sampling was not used for the Maine data because of the limited number of freeway miles available in that state. As a result, the distribution of segment count for the Maine data roughly reflects the mileage of each combination on the Maine freeway system. The barrier and ramp characteristics for freeway segments are listed in Table 18. This table lists the average proportion of the segment length that has barrier. This proportion is separately tabulated for the inside (i.e., median) and outside (i.e., roadside) of the roadway. The proportion of the segment length with barrier is shown to increase with the number of through lanes and tends to be higher in urban areas. These trends likely reflect situations where through

82 lanes were added to the alignment without taking additional right-of-way. In these situations, the clear zone or the median width is reduced and barrier is needed to protect motorists from the resulting hazards. TABLE 18. Barrier and ramp characteristics for freeway segments State Area Type Through Lanes Proportion Barrier Proportion Ramp Number of Ramps Inside 1 Outside 1 Entrance 2 Weave 2 Entrance Exit California Rural 4 0.649 0.118 0.080 0.017 53 48 6 0.997 0.130 0.061 0.000 22 25 8 0.949 0.098 0.067 0.015 25 24 Urban 4 0.734 0.178 0.089 0.013 47 41 6 0.964 0.188 0.112 0.060 87 65 8 0.988 0.327 0.106 0.236 39 38 10 1.000 0.624 0.104 0.118 31 30 Overall: 0.887 0.223 0.091 0.066 304 271 Maine Rural 4 0.650 0.242 0.032 0.000 21 21 6 0.985 0.149 0.071 0.000 4 3 8 Urban 4 0.806 0.370 0.067 0.051 24 20 6 0.727 0.750 0.161 0.493 1 1 8 1.000 0.998 0.000 0.997 0 1 10 Overall: 0.733 0.285 0.048 0.027 50 46 Washington Rural 4 0.563 0.146 0.085 0.007 19 19 6 0.584 0.257 0.096 0.016 40 40 8 Urban 4 0.631 0.334 0.068 0.327 44 42 6 0.550 0.417 0.048 0.256 44 36 8 0.921 0.343 0.136 0.133 20 20 10 Overall: 0.618 0.304 0.080 0.156 167 157 Notes: 1 - Proportion of segment length with rigid barrier. Inside barrier is in the median. Outside barrier is on the roadside. 2 - Proportion of segment length associated with total ramp entrance length or total weave length. The data in Table 18 also indicate that entrance-ramp-related speed-change lanes constitute about 5 to 9 percent of the segment length. For this statistic, the speed-change lane length is computed as the ramp entrance length or the ramp exit length, as shown in Figure 11. Although not shown in the table, ramp exits tended to account for an additional 6 percent of the segment length.

83 The proportion of the total segment length that is located in a weaving section varies widely among the states. Maine has the least mileage in the database with weaving. In contrast, the Washington data has the largest amount of mileage with weaving. Many weaving sections in California were coincident with managed-lane facilities, which were explicitly excluded from the database assembled for this project. Crash Characteristics Crash data were identified for each segment using the most recently available data from the HSIS. Three years of crash data were identified for each segment. The analysis period is 2005, 2006, and 2007 for the California and Washington segments. It is 2004, 2005, and 2006 for the Maine segments. The crash records associated with each segment were categorized in terms of whether they described a multiple-vehicle non-ramp-related crash, single-vehicle crash, ramp-entrance-related crash, or ramp-exit-related crash. The latter two crash types crashes are referred to herein as speed- change-related crashes. Ramp-entrance- and ramp-exit-related crashes do include some crashes that occur on the ramp proper, near the speed-change lane. Crashes of each severity level (i.e., K, A, B, C, or PDO) were included in the database. However, only those associated with injury or fatality are summarized in this subsection. The crash data for the freeway segments are summarized in Table 19. These segments were collectively associated with 8,381 injury or fatal crashes (= 5,492 + 661 + 2,228). The trends in crash rate are provided in the last column of the table. These rates indicate that urban freeway segments experience more crashes than rural freeway segments, for the same volume and length. They also indicate that urban segments with more lanes tend to experience more crashes than urban segments with few lanes. Given that the proportion of barrier was also observed to increase with the number of lanes, it is likely that the increase in urban freeway crash rate with increasing lanes is reflective of the reduced horizontal clearance to barrier and other obstructions. In other words, the addition of lanes to a freeway is not likely causing the observed increase in crashes. Rather, it is the reduced clearance (that often associates with the addition of lanes) that is causing the crashes. These trends were observed in a previous analysis of Texas crash data (Bonneson and Pratt, 2008). Although not shown in the table, the total number of crashes for each state was also computed. This total includes the FI crashes listed in Table 19 plus the PDO crashes that were reported. The total number of crashes for the segments is 26,426. The distribution by state is 17,548, 2,495, and 6,383 for California, Maine, and Washington, respectively. Overall, the FI crashes represent 32 percent of the total crashes. However, this percentage varies among the three states. Specifically, it is 31 percent, 26 percent, and 35 percent for California, Maine, and Washington, respectively. The variation in these percentages is consistent with the variation among states found by Zegeer et al. (1998) and reaffirms the benefit of focusing on FI crashes when comparing crash trends among different jurisdictions. Multiple-vehicle crashes in California and Washington tend to be nearly twice as frequent as single-vehicle crashes. This trend is not maintained in the Maine data. It is likely due to the lower traffic density on the Maine segments.

84 TABLE 19. Crash data summary for freeway segments State Area Type Through Lanes Exposure,1 mvm FI Crashes / 3 years 4 Crash Rate, cr/mvm Multiple- Vehicle 2 Single- Vehicle Ramp Entrance 3 Ramp Exit 3 Total California Rural 4 1,927 108 165 33 20 326 0.17 6 3,271 223 229 22 25 499 0.15 8 4,004 222 269 21 15 527 0.13 Urban 4 1,819 193 128 48 8 377 0.21 6 4,560 641 428 141 54 1,264 0.28 8 4,057 782 274 125 44 1,225 0.30 10 3,873 888 235 112 39 1,274 0.33 Overall: 23,511 3,057 1,728 502 205 5,492 0.23 Maine Rural 4 2,506 154 229 10 8 401 0.16 6 446 25 16 2 0 43 0.10 8 Urban 4 1,050 88 80 14 11 193 0.18 6 31 5 3 3 2 13 0.42 8 17 6 4 0 1 11 0.63 10 Overall: 4,051 278 332 29 22 661 0.16 Washingt on Rural 4 1,954 64 179 4 2 249 0.13 6 3,332 184 189 15 6 394 0.12 8 Urban 4 1,358 175 108 17 9 309 0.23 6 2,259 301 136 15 10 462 0.20 8 2,301 594 115 83 22 814 0.35 10 Overall: 11,203 1,318 727 134 49 2,228 0.20 All states Rural 4 6,387 326 573 47 30 976 0.15 6 7,050 432 434 39 31 936 0.13 8 4,004 222 269 21 15 527 0.13 Urban 4 4,227 456 316 79 28 879 0.21 6 6,850 947 567 159 66 1,739 0.25 8 6,375 1,382 393 208 67 2,050 0.32 10 3,873 888 235 112 39 1,274 0.33 Overall: 38,765 4,653 2,787 665 276 8,381 0.22 All states, excluding speed- change lane- related crashes Rural 4 6,387 326 573 899 0.14 6 7,050 432 434 866 0.12 8 4,004 222 269 491 0.12 Urban 4 4,227 456 316 772 0.18 6 6,850 947 567 1,514 0.22 8 6,375 1,382 393 1,775 0.28 10 3,873 888 235 1,123 0.29 Overall: 38,765 4,653 2,787 7,440 0.19 Notes: 1 - mvm: million vehicle-miles. 2 - Multiple-vehicle crashes do not include speed-change-related crashes. 3 - Referred to herein as speed-change-related crashes. 4 - FI: fatal-and-injury crashes.

85 Interchange Ramp This section describes the ramp segments represented in the safety database. Initially, the traffic and geometric characteristics of each segment are presented. Then, the segment crash data are examined. A ramp segment is defined as a portion of the entire ramp. An entire ramp is defined to extend from the gore point at the freeway speed-change lane to either (1) the gore point at a crossroad speed-change lane or (2) the near edge of the crossroad traveled way at the crossroad ramp terminal. For the California and Maine databases, all of the ramp segments extend for the length of the entire ramp. For the Washington database, the ramp segments are typically only a small portion of the entire ramp. Segment Characteristics A total of 1,537 ramp segments are represented in the combined ramp segment database. These segments represent about 65, 49, and 114 miles of ramps in California, Maine, and Washington, respectively. Selected segment characteristics are provided in Table 20. The ramp configurations identified in this table are shown in Figure 4. The number of segments for each combination of ramp type and configuration is indicated in column four of Table 20. The numbers shown for California are relatively uniform among rows and reflect the sampling technique described in a previous section. Sampling was also applied to the Washington data; however, it was not possible to make the distribution more uniform among ramp configurations given the relatively limited number of viable loop and buttonhook ramps represented in the data. Sampling was not used for the Maine data because all of ramps available in that state were considered for inclusion in the database. Collector-distributor road segments were included in the database for Maine and Washington. These segments could not be accurately located using the HSIS data for California because interchange maps were not available for this state.

86 TABLE 20. Summary characteristics for ramp segments State Ramp Type 1 Ramp Config. Total Segments Total Length, mi Seg. Length Range, mi Volume Range, veh/day Minimum Maximum Minimum Maximum Califor nia C-D road Segment Exit Connector 39 8.4 0.08 0.43 440 21,600 Diagonal 65 12.1 0.12 0.29 400 13,200 Button hook 43 4.8 0.05 0.24 780 15,000 Loop 63 11.0 0.08 0.25 960 12,400 Entrance Connector 40 8.1 0.07 0.65 1,500 18,100 Diagonal 57 9.6 0.10 0.25 520 13,300 Button hook 33 2.7 0.05 0.14 740 8,800 Loop 65 8.6 0.09 0.22 530 11,800 Overall: 405 65.3 0.05 0.65 400 21,600 Maine C-D road Segment 7 1.0 0.09 0.20 3,200 11,100 Exit Connector 39 11.1 0.12 0.76 1,100 9,800 Diagonal 43 10.2 0.15 0.38 400 7,100 Button hook 1 0.1 0.09 0.09 1,800 1,800 Loop 21 3.7 0.10 0.30 1,400 6,400 Entrance Connector 34 7.8 0.11 0.40 750 8,800 Diagonal 40 10.4 0.14 0.42 720 8,500 Button hook 1 0.1 0.08 0.08 2,500 2,500 Loop 23 5.1 0.11 0.38 560 6,600 Overall: 209 49.5 0.08 0.76 400 11,100 Washi ngton C-D road Segment 195 17.1 0.02 0.21 2,200 29,500 Exit Connector 79 12.2 0.02 0.40 1,400 33,000 Diagonal 239 30.6 0.02 0.26 140 10,900 Button hook 13 1.7 0.05 0.41 450 11,000 Loop 34 5.3 0.02 0.42 160 11,400 Entrance Connector 103 12.6 0.02 0.36 1,800 23,800 Diagonal 229 30.1 0.02 0.27 140 9,800 Button hook 2 0.1 0.04 0.06 5,000 5,000 Loop 29 4.2 0.01 0.35 740 15,300 Overall: 923 113.9 0.01 0.42 140 33,000 Note: 1 - C-D road: collector-distributor ramp roadway. The barrier and ramp characteristics for ramp segments are listed in Table 21. This table lists the average proportion of the segment length that has barrier. This proportion is separately tabulated for the inside (i.e., left side) and outside (i.e., right side) of the ramp roadway. The

87 proportion of the segment with barrier for Maine ramps is about twice that of the Washington and California ramps. TABLE 21. Barrier and cross section characteristics for ramp segments State Ramp Type Ramp Config. Proportion Barrier Proportion Curve 2 Average Width, ft Inside 1 Outside 1 Left Shldr. Lane Right Shldr. Californi a C-D road Segment Exit Connector 0.203 0.328 0.524 3.6 12.7 7.0 Diagonal 0.047 0.136 0.441 3.4 12.6 5.8 Button hook 0.163 0.159 0.477 3.0 12.6 5.5 Loop 0.204 0.116 0.651 3.4 13.4 5.8 Entrance Connector 0.070 0.078 0.556 3.6 13.1 6.5 Diagonal 0.041 0.047 0.417 3.2 13.1 6.5 Button hook 0.042 0.078 0.707 3.3 15.5 5.6 Loop 0.118 0.027 0.822 3.4 14.4 6.1 Overall: 0.111 0.121 0.574 3.4 13.4 6.1 Maine C-D road Segment 0.124 0.416 0.126 4.0 12.0 10.0 Exit Connector 0.303 0.476 0.612 5.4 15.5 6.4 Diagonal 0.237 0.229 0.438 5.2 15.3 6.6 Button hook 0.422 0.433 0.731 0.0 19.0 0.0 Loop 0.321 0.186 0.844 4.6 18.3 5.0 Entrance Connector 0.266 0.391 0.516 5.1 16.7 5.9 Diagonal 0.150 0.253 0.343 5.5 15.2 6.6 Button hook 0.543 0.364 0.658 4.0 18.0 4.0 Loop 0.426 0.390 0.792 4.3 17.7 4.6 Overall: 0.310 0.349 0.562 4.2 16.4 5.5 Washingt on C-D road Segment 0.740 0.416 0.353 6.0 13.8 8.3 Exit Connector 0.376 0.567 0.589 5.3 13.8 7.7 Diagonal 0.079 0.221 0.470 5.2 14.3 7.9 Button hook 0.000 0.000 0.380 5.1 13.4 6.4 Loop 0.000 0.000 0.749 5.4 16.2 7.6 Entrance Connector 0.225 0.197 0.609 5.6 14.0 8.7 Diagonal 0.137 0.283 0.458 5.2 14.1 8.1 Button hook 0.000 0.000 0.348 6.4 13.9 8.6 Loop 0.000 0.000 0.855 5.0 15.7 7.4 Overall: 0.173 0.187 0.535 5.5 14.4 7.9 Notes: 1 - Proportion of segment length with rigid barrier. Inside barrier is in the median. Outside barrier is on the roadside. 2 - Proportion of segment length on horizontal curve.

88 The sixth column of Table 21 indicates the average proportion of the ramp segment length that has curvature. This proportion is highest for the loop ramp configurations, which is logical given their design. All of the ramp configurations tend to have curvature for 40 percent or more of their length. The last three columns of Table 21 list the average width of the shoulders and lanes on the ramp segments. It is noteworthy that the average lane width on the Maine ramps is two feet wider than the Washington ramps and three feet wider than the California ramps. It is also noteworthy that the combined cross section of the California ramps is several feet narrower than the ramps in either Maine or Washington. Crash Characteristics Crash data were identified for each ramp segment using the most recently available data from the HSIS. Five years of crash data were identified for each segment. The analysis period is 2003, 2004, 2005, 2006, and 2007 for the California and Washington segments. It is 2002, 2003, 2004, 2005, and 2006 for the Maine segments. The crash records associated with each segment were categorized in terms of whether they described a multiple-vehicle crash or a single-vehicle crash. Crashes of each severity level (i.e., K, A, B, C, or PDO) were included in the database. However, only those associated with injury or fatality are summarized in this subsection. The crash data for the ramp segments are summarized in Table 22. These segments were collectively associated with 1,178 injury or fatal crashes (= 382 + 89 + 707). The trends in crash rate are provided in the last column of the table. These rates suggest that crash risk on a ramp is about twice that on a freeway segment. This trend is likely due to the sharper curves on ramps, relative to freeways and the significant speed change associated with ramp driving. A closer examination of the crash rates in Table 22 indicates that most ramp crashes are single-vehicle crashes, which is logical given the typical lack of adjacent or oncoming lanes and the frequent presence of sharp curves on ramps. The crash rates for the Maine ramps are notably smaller than that found for the California and Washington ramps. This trend may reflect the longer length of the Maine ramps, their wider cross section, or both, relative to the California and Washington ramps. The longer length may provide a more accommodating distance for the speed change that occurs on the ramp. Wider lane and shoulder widths have been found to reduce crashes on street and highway segments (as noted in Chapter 2). Although not shown in the table, the total number of crashes for each state was also computed. This total includes the FI crashes listed in Table 22 plus the PDO crashes that were reported. The total number of crashes for the segments is 3,541. The distribution by state is 1,120, 324, and 2,097 for California, Maine, and Washington, respectively. The FI crashes represent 33 percent of the total crashes. However, this percentage varies among the three states. Specifically, it is 34 percent, 27 percent, and 34 percent for California, Maine, and Washington,

89 respectively. The variation in these percentages is consistent with the variation among the states found in the freeway segment crash data. TABLE 22. Crash data summary for ramp segments State Ramp Type Ramp Config. Exposure,1 mvm FI Crashes / 5 years Crash Rate, cr/mvm Multiple- Vehicle 2 Single- Vehicle Total California C-D road Segment Exit Connector 145.1 3 62 65 0.45 Diagonal 137.9 0 36 36 0.26 Button hook 50.7 1 45 46 0.91 Loop 126.9 3 96 99 0.78 Entrance Connector 120.9 6 40 46 0.38 Diagonal 89.5 3 36 39 0.44 Button hook 22.5 0 5 5 0.22 Loop 62.5 7 39 46 0.74 Overall: 756.0 23 359 382 0.51 Maine C-D road Segment 12.9 0 1 1 0.08 Exit Connector 101.9 0 20 20 0.20 Diagonal 47.0 3 15 18 0.38 Button hook 0.3 0 0 0 0.00 Loop 22.6 0 9 9 0.40 Entrance Connector 65.2 8 17 25 0.38 Diagonal 56.0 4 3 7 0.13 Button hook 0.4 0 0 0 0.00 Loop 32.0 5 4 9 0.28 Overall: 338.2 20 69 89 0.26 Washington C-D road Segment 366.8 114 44 158 0.43 Exit Connector 202.4 17 169 186 0.92 Diagonal 172.3 6 73 79 0.46 Button hook 18.7 2 21 23 1.23 Loop 39.2 2 27 29 0.74 Entrance Connector 231.8 37 102 139 0.60 Diagonal 149.3 12 44 56 0.38 Button hook 0.9 1 5 6 6.57 Loop 32.8 1 30 31 0.95 Overall: 1,214.2 192 515 707 0.58 Notes: 1 - mvm: million vehicle-miles. 2 - Multiple-vehicle crashes do not include speed-change-related crashes.

90 The data in Table 22 are summarized by ramp type and configuration in Table 23 to facilitate some preliminary examination of trend. The crash rates shown indicate that button hook exit ramps tend to have the highest crash rate, which is intuitive given their inherently short length and sharp curvature. In contrast, the diagonal entrance ramp has the lowest crash rate. A pair-wise comparison of exit and entrance ramps by configuration indicates that the exit ramps tend to have a higher crash rate than entrance ramps. TABLE 23. Crash data summary by ramp configuration Ramp Type Ramp Config. Exposure, 1 mvm FI Crashes / 5 years Crash Rate, cr/mvm C-D road Segment 379.7 159 0.42 Exit Connector 449.4 271 0.60 Diagonal 357.2 133 0.37 Button hook 69.7 69 0.99 Loop 188.7 137 0.73 Entrance Connector 417.9 210 0.50 Diagonal 294.8 102 0.35 Button hook 23.8 11 0.46 Loop 127.3 86 0.68 Overall: 2,308.5 1,178 0.51 Note: 1 - mvm: million vehicle-miles. Crossroad Ramp Terminal This section describes the crossroad ramp terminals represented in the safety database. Initially, the traffic and geometric characteristics of each ramp terminal are presented. Then, the ramp terminal crash data are examined. There are many different ramp configurations found at interchanges. The more common ones are identified in Figure 37. Differences among ramp terminals are shown to reflect the number of ramp legs, ramp configuration, number of left-turn movements, and location of crossroad left-turn storage (i.e., internal or external to interchange). Although not shown, control type (i.e., signalized or unsignalized) is also an important factor in characterizing ramp terminal safety and operation.

91 Crossroad Ramp Type: D3ex Type: D3en Ramp Fr ee w ay Fr ee w ay Crossroad Ramp Type: D4 Type: D4 Ramp Fr ee w ay Fr ee w ay Crossroad Ramp Type: A4 Type: A4 RampRamp Ramp Fr ee w ay Fr ee w ay a. Three-leg ramp terminal with diagonal exit ramp or entrance ramp (D3ex and D3en). b. Four-leg ramp terminal with diagonal ramps (D4). c. Ramp terminal at four-quadrant parclo A (A4). Figure 37. Ramp terminal configurations.

92 Crossroad Ramp Type: B4 Type: B4 Ramp RampRamp Fr ee w ay Fr ee w ay Crossroad Ramp Type: A2 Type: A2 Ramp Fr ee w ay Fr ee w ay Crossroad Type: B2 Type: B2 Ramp Ramp Fr ee w ay Fr ee w ay d. Ramp terminal at four-quadrant parclo B (B4). e. Ramp terminal at two-quadrant parclo A (A2). f. Ramp terminal at two-quadrant parclo B (B2). Figure 37. Ramp terminal configurations (continued).

93 Ramp Terminal Characteristics All total, 569 ramp terminals are represented in the combined database. More specifically, the database includes 216, 62, and 291 ramp terminals in California, Maine, and Washington, respectively. Selected ramp terminal characteristics are provided in Table 24. The number of ramp terminals for each combination of control type and configuration is indicated in column 4 of Table 24. When distributed among the 13 combinations shown, there are several combinations with a small number of ramp terminals. Most notable is the small number of “parclo B4” and “diagonal 3-leg” configurations. As noted previously, this sample represents all of the ramp terminals that are fully represented in the HSIS databases. Sample size issues are discussed at the end of this section. Crash Characteristics Crash data were identified for each ramp terminal using the most recently available data from the HSIS. Three years of crash data were identified for each ramp terminal. The analysis period is 2005, 2006, and 2007 for the California and Washington segments. It is 2004, 2005, and 2006 for the Maine segments. Crashes of each severity level (i.e., K, A, B, C, or PDO) were included in the database. Those crashes associated with an injury or fatality are summarized separately from those associated with PDO. The crash data for the ramp terminals are summarized in Table 25. These ramp terminals were collectively associated with 2,177 injury or fatal crashes (= 810 + 171 + 1,196). The trends in crash rate are provided in the last column of the table. Typical FI crash rates for three-leg intersections are 0.14 and 0.20 cr/mev for unsignalized and signalized control, respectively (Bonneson and Pratt, 2008). The rates shown in Table 25 are consistent with these typical values. The overall crash rates for the California ramp terminals (i.e., 0.13 and 0.42 cr/mvm) are notably lower than those for the Maine and Washington ramp terminals. However, the average exposure per ramp terminal for the California sites is 26 mev (= 6123.4 / 216), which is much larger than that for Maine (13 mev) and Washington (19 mev). Thus, the noted trend in crash rate is likely a reflection of the nonlinear relationship between ramp terminal volume and crash frequency.

94 TABLE 24. Summary characteristics for crossroad ramp terminals State Control Type Terminal Configuration Total Terminals Crossroad Volume, veh/day Ramp Volume, veh/day Minimum Maximum Minimum Maximum Californ ia Signal Parclo A2 11 7,700 48,700 2,900 15,100 Parclo A4 41 13,100 68,700 1,900 22,000 Parclo B2 6 7,300 45,700 2,000 11,700 Parclo B4 0 Diagonal 3 leg 4 12,100 30,000 1,000 10,500 Diagonal 4 leg 55 6,800 46,400 3,000 15,500 Unsignalized 1 Parclo A2 8 700 10,500 500 4,400 Parclo A4 16 4,000 17,300 340 10,200 Parclo B2 6 745 23,000 310 7,200 Parclo B4 1 13,600 13,600 4,300 4,300 Diagonal 3 leg 5 11,800 40,900 1,000 6,100 Diagonal 4 leg 63 1,900 18,900 320 4,700 Speed-change 0 Overall: 216 700 68,700 310 22,000 Maine Signal Parclo A2 0 Parclo A4 1 15,200 15,200 4,700 4,700 Parclo B2 2 14,700 15,600 3,200 3,400 Parclo B4 0 Diagonal 3 leg 1 34,000 34,000 10,800 10,800 Diagonal 4 leg 4 13,300 17,800 4,000 6,000 Unsignalized 1 Parclo A2 7 3,400 17,300 1,700 14,700 Parclo A4 0 Parclo B2 9 4,600 12,100 1,200 4,600 Parclo B4 1 18,300 18,300 3,100 3,100 Diagonal 3 leg 7 4,700 11,100 590 2,400 Diagonal 4 leg 24 1,900 12,700 510 4,400 Speed-change 6 8,270 20,530 373 7,300 Overall: 62 1,900 34,000 373 14,700 Washin gton Signal Parclo A2 11 8,100 42,000 1,800 13,100 Parclo A4 8 11,900 42,800 1,200 7,900 Parclo B2 6 10,400 25,300 3,600 8,500 Parclo B4 2 31,700 38,300 5,300 14,700 Diagonal 3 leg 6 2,900 33,400 4,700 21,600 Diagonal 4 leg 87 9,400 43,300 3,500 11,700 Unsignalized 1 Parclo A2 11 3,000 17,200 860 7,000 Parclo A4 2 12,900 22,700 590 5,000 Parclo B2 7 1,200 26,800 380 5,400 Parclo B4 4 8,800 24,800 940 9,900 Diagonal 3 leg 19 1,200 22,400 270 17,700 Diagonal 4 leg 127 370 10,400 180 4,200 Speed-change 1 24,200 24,200 4,700 4,700 Overall: 291 370 43,300 180 21,600 Note: 1 - Unsignalized intersections have an uncontrolled major street and a stop-controlled minor street.

95 TABLE 25. Crash data summary for crossroad ramp terminals State Control Type Terminal Configuration Exposure,1 mev Crashes / 3 years Crash Rate, cr/mev PDO 2 FI 3 Total FI 3 Total Califo rnia Signal Parclo A2 442.1 111 41 152 0.09 0.34 Parclo A4 2,252.2 552 287 839 0.13 0.37 Parclo B2 253.6 47 19 66 0.07 0.26 Parclo B4 Diagonal 3 leg 125.3 16 14 30 0.11 0.24 Diagonal 4 leg 1,900.5 742 298 1,040 0.16 0.55 Unsignalized 1 Parclo A2 57.6 15 8 23 0.14 0.40 Parclo A4 208.4 89 43 132 0.21 0.63 Parclo B2 73.1 16 8 24 0.11 0.33 Parclo B4 19.5 0 1 1 0.05 0.05 Diagonal 3 leg 143.1 36 19 55 0.13 0.38 Diagonal 4 leg 648.0 144 72 216 0.11 0.33 Speed-change Overall: 6,123.4 1,768 810 2,578 0.13 0.42 Maine Signal Parclo A2 Parclo A4 21.8 8 4 12 0.18 0.55 Parclo B2 40.5 19 16 35 0.40 0.86 Parclo B4 Diagonal 3 leg 49.0 41 15 56 0.31 1.14 Diagonal 4 leg 90.2 87 45 132 0.50 1.46 Unsignalized 1 Parclo A2 114.7 56 26 82 0.23 0.71 Parclo A4 Parclo B2 105.6 29 10 39 0.09 0.37 Parclo B4 23.4 12 4 16 0.17 0.68 Diagonal 3 leg 73.3 10 9 19 0.12 0.26 Diagonal 4 leg 203.0 63 28 91 0.14 0.45 Speed-change 107.0 34 14 48 0.13 0.45 Overall: 828.5 359 171 530 0.21 0.64 Washi ngton Signal Parclo A2 320.9 125 82 207 0.26 0.65 Parclo A4 227.8 92 43 135 0.19 0.59 Parclo B2 169.0 78 46 124 0.27 0.73 Parclo B4 98.5 48 24 72 0.24 0.73 Diagonal 3 leg 231.1 73 36 109 0.16 0.47 Diagonal 4 leg 2,883.9 1,310 770 2,080 0.27 0.72 Unsignalized 1 Parclo A2 133.6 14 15 29 0.11 0.22 Parclo A4 45.1 2 3 5 0.07 0.11 Parclo B2 90.5 21 13 34 0.14 0.38 Parclo B4 88.1 22 14 36 0.16 0.41 Diagonal 3 leg 252.2 46 26 72 0.10 0.29 Diagonal 4 leg 870.5 211 121 332 0.14 0.38 Speed-change 31.7 3 3 6 0.09 0.19 Overall: 5,442.8 2,045 1,196 3,241 0.22 0.60 Notes: 1 - mev: million-entering-vehicles per year. 2 - PDO: property-damage-only crashes. 3 - FI: fatal-and-injury crashes.

96 The data in Table 25 are summarized by control type and terminal configuration in Table 26 to facilitate some preliminary examination of trend. The crash rates shown indicate that the signalized “parclo B4” configuration has the highest crash rate. However, these statistics are based on only two ramp terminals and may not be representative of similar terminals at other locations. The signalized “diagonal 4 leg” has the second highest crash rate shown in Table 26. This trend may be related to the fact that this configuration is the only one that has four legs (all other configurations typically have three legs). In contrast, the unsignalized “parclo B2” and “diagonal 3 leg” terminals have lowest crash rate. A pair-wise comparison of signalized and unsignalized terminals by configuration indicates that the signalized terminals tend to have a higher crash rate than the unsignalized terminals. Sample Size Considerations Sample size considerations are complex and involve many factors. However, it is considered desirable to have at least 25 ramp terminals for each combination of control type and configuration to ensure reasonable representation of that type of facility. Also, statistical considerations indicate a desirable minimum of 150 crashes in the database for each combination. Table 27 indicates the total number of ramp terminals for each configuration across all three states represented in the combined database. An examination of columns 3 and 4 indicates that only 3 of the 13 configurations satisfy both criteria. However, 5 of 13 configurations have more than 25 ramp terminals. Several options are available to address these concerns. The option chosen was to aggregate selected terminal configurations with similar turn movement patterns. For example, consultation of Figure 37 indicates that the “parclo A4” has movements that are very similar to a “diagonal 3 leg” where the ramp leg is associated with an exit ramp. Similarly, the “parclo B4” has movements that are very similar to a “diagonal 3 leg” where the ramp leg is associated with an entrance ramp. Also, the “parclo B2” and “parclo A2” have very similar movements. These configurations were combined to examine their impact on the sample size considerations. The results are shown in the last three columns of Table 27. With this option, 4 of the 9 configurations satisfy both minimum criteria. Moreover, 6 of 9 configurations have more than 25 ramp terminals.

97 TABLE 26. Crash data summary by ramp terminal configuration Control Type Terminal Configuration Exposure, mev Crashes / 3 years Crash Rate, cr/mev PDO FI Total FI Total Signal Parclo A2 763.0 236 123 359 0.16 0.47 Parclo A4 2,501.8 652 334 986 0.13 0.39 Parclo B2 463.1 144 81 225 0.17 0.49 Parclo B4 98.5 48 24 72 0.24 0.73 Diagonal 3 leg 405.4 130 65 195 0.16 0.48 Diagonal 4 leg 4,874.6 2,139 1,113 3,252 0.23 0.67 Unsignalized Parclo A2 305.9 85 49 134 0.16 0.44 Parclo A4 253.5 91 46 137 0.18 0.54 Parclo B2 269.2 66 31 97 0.12 0.36 Parclo B4 131.0 34 19 53 0.15 0.40 Diagonal 3 leg 468.6 92 54 146 0.12 0.31 Diagonal 4 leg 1,721.5 418 221 639 0.13 0.37 Speed-change 138.7 37 17 54 0.12 0.39 Overall: 12,394.8 4,172 2,177 6,349 0.18 0.51 TABLE 27. Sample size considerations for crossroad ramp terminals Control Type Terminal Configuration Total Terminals 1 FI Crashes / 3 years 1 Possible Combinations Terminal Configuration Total Terminals1 FI Crashes / 3 years 1 Signal Parclo A2 22 123 Parclo A4 50 334 Parclo A4 and Diagonal 3 leg exit ramps 55 367 Parclo B2 14 81 Parclo B2 and Parclo A2 36 204 Parclo B4 2 24 Parclo B4 and Diagonal 3 leg entrance ramps 8 56 Diagonal 3 leg 11 65 Diagonal 4 leg 146 1,113 Diagonal 4 leg 146 1,113 Unsignalized Parclo A2 26 49 Parclo A4 18 46 Parclo A4 and Diagonal 3 leg exit ramps 33 73 Parclo B2 22 31 Parclo B2 and Parclo A2 48 80 Parclo B4 6 19 Parclo B4 and Diagonal 3 leg entrance ramps 22 46 Diagonal 3 leg 31 54 Diagonal 4 leg 214 221 Diagonal 4 leg 214 221 Speed-change 7 17 Speed-change 7 17 Overall: 569 2,177 569 2,177 Note: 1 - Underlined values identify configurations that meet or exceed desired sample sizes.

98 Freeway Speed-Change Lane This section describes the freeway speed-change lanes represented in the safety database. As described previously, all crashes located (by milepost) on the freeway segment between the mileposts that define the speed-change lane are speed-change-lane-related crashes. The location of a speed-change lane is defined by the mileposts of its taper and gore points. The crash data for speed-change lanes are summarized in Table 28. The data are categorized by state, ramp type, and number of speed-change lanes per freeway segment. Almost all of the segments with speed-change lanes have one entrance speed-change lane and one exit speed-change lane. A few of the longer segments have as many as three entrance speed-change lanes and two exit speed-change lanes. The exposure statistic in Table 28 is based on the length of the speed-change lane and the freeway segment AADT volume. The ramp AADT volume is not factored into the calculation of exposure. The last column of Table 28 lists the crash rate associated with the speed-change lanes. A comparison of these rates with those in Table 19 indicates that the crash rate in a speed-change lane exceeds that of a basic freeway segment (i.e., with speed-change-related crashes excluded). This trend is logical and likely reflects the increased risk of crash associated with frequent lane changes in the speed-change lane. An examination of the crash rates for entrance- and exit- related speed-change lanes does not indicate that there is any correlation between ramp type and crash rate.

99 TABLE 28. Summary characteristics for freeway speed-change lane segments State Ramp Type No. S-C Lanes/Seg Total Segments S-C Lane Length, mi Exposure,1 mvm FI Crashes / 3years Crash Rate, cr/mvm Californ ia Entrance 1 231 20.7 1,352.7 369 0.27 2 54 9.7 574.2 127 0.22 3 3 0.5 19.2 6 0.31 Exit 1 222 9.8 593.4 177 0.30 2 23 2.4 139.8 28 0.20 3 0 Overall: 533 43.1 2,679.3 707 0.26 Maine Entrance 1 52 5.8 129.1 22 0.17 2 3 0.7 25.2 7 0.28 3 0 Exit 1 58 5.3 106.4 18 0.17 2 3 0.5 12.7 3 0.24 3 0 Overall: 116 12.3 273.4 50 0.18 Washin gton Entrance 1 239 17.1 624.6 129 0.21 2 6 0.9 25.5 5 0.20 3 0 Exit 1 197 7.3 285.8 49 0.17 2 0 3 0 Overall: 442 25.2 935.9 183 0.20 All States Entrance 1 522 43.6 2,106.4 520 0.25 2 63 11.3 624.9 139 0.22 3 3 0.5 19.2 6 0.31 Exit 1 477 22.4 985.6 244 0.25 2 26 2.9 152.5 31 0.20 3 0 Overall: 1,091 80.7 3,888.6 940 0.24 Note: 1 - mev: million-entering-vehicles per year.