National Academies Press: OpenBook

Development of a Posted Speed Limit Setting Procedure and Tool (2021)

Chapter: APPENDIX D. RELATIONSHIP AMONG URBAN/SUBURBAN ROADWAY CHARACTERISTICS, OPERATING SPEED, AND CRASHES IN AUSTIN, TEXAS

« Previous: APPENDIX C. SPEED LIMIT SETTING APPROACHES
Page 80
Suggested Citation:"APPENDIX D. RELATIONSHIP AMONG URBAN/SUBURBAN ROADWAY CHARACTERISTICS, OPERATING SPEED, AND CRASHES IN AUSTIN, TEXAS ." National Academies of Sciences, Engineering, and Medicine. 2021. Development of a Posted Speed Limit Setting Procedure and Tool. Washington, DC: The National Academies Press. doi: 10.17226/26200.
×
Page 80
Page 81
Suggested Citation:"APPENDIX D. RELATIONSHIP AMONG URBAN/SUBURBAN ROADWAY CHARACTERISTICS, OPERATING SPEED, AND CRASHES IN AUSTIN, TEXAS ." National Academies of Sciences, Engineering, and Medicine. 2021. Development of a Posted Speed Limit Setting Procedure and Tool. Washington, DC: The National Academies Press. doi: 10.17226/26200.
×
Page 81
Page 82
Suggested Citation:"APPENDIX D. RELATIONSHIP AMONG URBAN/SUBURBAN ROADWAY CHARACTERISTICS, OPERATING SPEED, AND CRASHES IN AUSTIN, TEXAS ." National Academies of Sciences, Engineering, and Medicine. 2021. Development of a Posted Speed Limit Setting Procedure and Tool. Washington, DC: The National Academies Press. doi: 10.17226/26200.
×
Page 82
Page 83
Suggested Citation:"APPENDIX D. RELATIONSHIP AMONG URBAN/SUBURBAN ROADWAY CHARACTERISTICS, OPERATING SPEED, AND CRASHES IN AUSTIN, TEXAS ." National Academies of Sciences, Engineering, and Medicine. 2021. Development of a Posted Speed Limit Setting Procedure and Tool. Washington, DC: The National Academies Press. doi: 10.17226/26200.
×
Page 83
Page 84
Suggested Citation:"APPENDIX D. RELATIONSHIP AMONG URBAN/SUBURBAN ROADWAY CHARACTERISTICS, OPERATING SPEED, AND CRASHES IN AUSTIN, TEXAS ." National Academies of Sciences, Engineering, and Medicine. 2021. Development of a Posted Speed Limit Setting Procedure and Tool. Washington, DC: The National Academies Press. doi: 10.17226/26200.
×
Page 84
Page 85
Suggested Citation:"APPENDIX D. RELATIONSHIP AMONG URBAN/SUBURBAN ROADWAY CHARACTERISTICS, OPERATING SPEED, AND CRASHES IN AUSTIN, TEXAS ." National Academies of Sciences, Engineering, and Medicine. 2021. Development of a Posted Speed Limit Setting Procedure and Tool. Washington, DC: The National Academies Press. doi: 10.17226/26200.
×
Page 85
Page 86
Suggested Citation:"APPENDIX D. RELATIONSHIP AMONG URBAN/SUBURBAN ROADWAY CHARACTERISTICS, OPERATING SPEED, AND CRASHES IN AUSTIN, TEXAS ." National Academies of Sciences, Engineering, and Medicine. 2021. Development of a Posted Speed Limit Setting Procedure and Tool. Washington, DC: The National Academies Press. doi: 10.17226/26200.
×
Page 86
Page 87
Suggested Citation:"APPENDIX D. RELATIONSHIP AMONG URBAN/SUBURBAN ROADWAY CHARACTERISTICS, OPERATING SPEED, AND CRASHES IN AUSTIN, TEXAS ." National Academies of Sciences, Engineering, and Medicine. 2021. Development of a Posted Speed Limit Setting Procedure and Tool. Washington, DC: The National Academies Press. doi: 10.17226/26200.
×
Page 87
Page 88
Suggested Citation:"APPENDIX D. RELATIONSHIP AMONG URBAN/SUBURBAN ROADWAY CHARACTERISTICS, OPERATING SPEED, AND CRASHES IN AUSTIN, TEXAS ." National Academies of Sciences, Engineering, and Medicine. 2021. Development of a Posted Speed Limit Setting Procedure and Tool. Washington, DC: The National Academies Press. doi: 10.17226/26200.
×
Page 88
Page 89
Suggested Citation:"APPENDIX D. RELATIONSHIP AMONG URBAN/SUBURBAN ROADWAY CHARACTERISTICS, OPERATING SPEED, AND CRASHES IN AUSTIN, TEXAS ." National Academies of Sciences, Engineering, and Medicine. 2021. Development of a Posted Speed Limit Setting Procedure and Tool. Washington, DC: The National Academies Press. doi: 10.17226/26200.
×
Page 89
Page 90
Suggested Citation:"APPENDIX D. RELATIONSHIP AMONG URBAN/SUBURBAN ROADWAY CHARACTERISTICS, OPERATING SPEED, AND CRASHES IN AUSTIN, TEXAS ." National Academies of Sciences, Engineering, and Medicine. 2021. Development of a Posted Speed Limit Setting Procedure and Tool. Washington, DC: The National Academies Press. doi: 10.17226/26200.
×
Page 90
Page 91
Suggested Citation:"APPENDIX D. RELATIONSHIP AMONG URBAN/SUBURBAN ROADWAY CHARACTERISTICS, OPERATING SPEED, AND CRASHES IN AUSTIN, TEXAS ." National Academies of Sciences, Engineering, and Medicine. 2021. Development of a Posted Speed Limit Setting Procedure and Tool. Washington, DC: The National Academies Press. doi: 10.17226/26200.
×
Page 91
Page 92
Suggested Citation:"APPENDIX D. RELATIONSHIP AMONG URBAN/SUBURBAN ROADWAY CHARACTERISTICS, OPERATING SPEED, AND CRASHES IN AUSTIN, TEXAS ." National Academies of Sciences, Engineering, and Medicine. 2021. Development of a Posted Speed Limit Setting Procedure and Tool. Washington, DC: The National Academies Press. doi: 10.17226/26200.
×
Page 92
Page 93
Suggested Citation:"APPENDIX D. RELATIONSHIP AMONG URBAN/SUBURBAN ROADWAY CHARACTERISTICS, OPERATING SPEED, AND CRASHES IN AUSTIN, TEXAS ." National Academies of Sciences, Engineering, and Medicine. 2021. Development of a Posted Speed Limit Setting Procedure and Tool. Washington, DC: The National Academies Press. doi: 10.17226/26200.
×
Page 93
Page 94
Suggested Citation:"APPENDIX D. RELATIONSHIP AMONG URBAN/SUBURBAN ROADWAY CHARACTERISTICS, OPERATING SPEED, AND CRASHES IN AUSTIN, TEXAS ." National Academies of Sciences, Engineering, and Medicine. 2021. Development of a Posted Speed Limit Setting Procedure and Tool. Washington, DC: The National Academies Press. doi: 10.17226/26200.
×
Page 94
Page 95
Suggested Citation:"APPENDIX D. RELATIONSHIP AMONG URBAN/SUBURBAN ROADWAY CHARACTERISTICS, OPERATING SPEED, AND CRASHES IN AUSTIN, TEXAS ." National Academies of Sciences, Engineering, and Medicine. 2021. Development of a Posted Speed Limit Setting Procedure and Tool. Washington, DC: The National Academies Press. doi: 10.17226/26200.
×
Page 95
Page 96
Suggested Citation:"APPENDIX D. RELATIONSHIP AMONG URBAN/SUBURBAN ROADWAY CHARACTERISTICS, OPERATING SPEED, AND CRASHES IN AUSTIN, TEXAS ." National Academies of Sciences, Engineering, and Medicine. 2021. Development of a Posted Speed Limit Setting Procedure and Tool. Washington, DC: The National Academies Press. doi: 10.17226/26200.
×
Page 96
Page 97
Suggested Citation:"APPENDIX D. RELATIONSHIP AMONG URBAN/SUBURBAN ROADWAY CHARACTERISTICS, OPERATING SPEED, AND CRASHES IN AUSTIN, TEXAS ." National Academies of Sciences, Engineering, and Medicine. 2021. Development of a Posted Speed Limit Setting Procedure and Tool. Washington, DC: The National Academies Press. doi: 10.17226/26200.
×
Page 97
Page 98
Suggested Citation:"APPENDIX D. RELATIONSHIP AMONG URBAN/SUBURBAN ROADWAY CHARACTERISTICS, OPERATING SPEED, AND CRASHES IN AUSTIN, TEXAS ." National Academies of Sciences, Engineering, and Medicine. 2021. Development of a Posted Speed Limit Setting Procedure and Tool. Washington, DC: The National Academies Press. doi: 10.17226/26200.
×
Page 98
Page 99
Suggested Citation:"APPENDIX D. RELATIONSHIP AMONG URBAN/SUBURBAN ROADWAY CHARACTERISTICS, OPERATING SPEED, AND CRASHES IN AUSTIN, TEXAS ." National Academies of Sciences, Engineering, and Medicine. 2021. Development of a Posted Speed Limit Setting Procedure and Tool. Washington, DC: The National Academies Press. doi: 10.17226/26200.
×
Page 99
Page 100
Suggested Citation:"APPENDIX D. RELATIONSHIP AMONG URBAN/SUBURBAN ROADWAY CHARACTERISTICS, OPERATING SPEED, AND CRASHES IN AUSTIN, TEXAS ." National Academies of Sciences, Engineering, and Medicine. 2021. Development of a Posted Speed Limit Setting Procedure and Tool. Washington, DC: The National Academies Press. doi: 10.17226/26200.
×
Page 100
Page 101
Suggested Citation:"APPENDIX D. RELATIONSHIP AMONG URBAN/SUBURBAN ROADWAY CHARACTERISTICS, OPERATING SPEED, AND CRASHES IN AUSTIN, TEXAS ." National Academies of Sciences, Engineering, and Medicine. 2021. Development of a Posted Speed Limit Setting Procedure and Tool. Washington, DC: The National Academies Press. doi: 10.17226/26200.
×
Page 101
Page 102
Suggested Citation:"APPENDIX D. RELATIONSHIP AMONG URBAN/SUBURBAN ROADWAY CHARACTERISTICS, OPERATING SPEED, AND CRASHES IN AUSTIN, TEXAS ." National Academies of Sciences, Engineering, and Medicine. 2021. Development of a Posted Speed Limit Setting Procedure and Tool. Washington, DC: The National Academies Press. doi: 10.17226/26200.
×
Page 102
Page 103
Suggested Citation:"APPENDIX D. RELATIONSHIP AMONG URBAN/SUBURBAN ROADWAY CHARACTERISTICS, OPERATING SPEED, AND CRASHES IN AUSTIN, TEXAS ." National Academies of Sciences, Engineering, and Medicine. 2021. Development of a Posted Speed Limit Setting Procedure and Tool. Washington, DC: The National Academies Press. doi: 10.17226/26200.
×
Page 103
Page 104
Suggested Citation:"APPENDIX D. RELATIONSHIP AMONG URBAN/SUBURBAN ROADWAY CHARACTERISTICS, OPERATING SPEED, AND CRASHES IN AUSTIN, TEXAS ." National Academies of Sciences, Engineering, and Medicine. 2021. Development of a Posted Speed Limit Setting Procedure and Tool. Washington, DC: The National Academies Press. doi: 10.17226/26200.
×
Page 104
Page 105
Suggested Citation:"APPENDIX D. RELATIONSHIP AMONG URBAN/SUBURBAN ROADWAY CHARACTERISTICS, OPERATING SPEED, AND CRASHES IN AUSTIN, TEXAS ." National Academies of Sciences, Engineering, and Medicine. 2021. Development of a Posted Speed Limit Setting Procedure and Tool. Washington, DC: The National Academies Press. doi: 10.17226/26200.
×
Page 105
Page 106
Suggested Citation:"APPENDIX D. RELATIONSHIP AMONG URBAN/SUBURBAN ROADWAY CHARACTERISTICS, OPERATING SPEED, AND CRASHES IN AUSTIN, TEXAS ." National Academies of Sciences, Engineering, and Medicine. 2021. Development of a Posted Speed Limit Setting Procedure and Tool. Washington, DC: The National Academies Press. doi: 10.17226/26200.
×
Page 106

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.

NCHRP Web-Only Document 291: Development of a Posted Speed Limit Setting Procedure and Tool 68 APPENDIX D. RELATIONSHIP AMONG URBAN/SUBURBAN ROADWAY CHARACTERISTICS, OPERATING SPEED, AND CRASHES IN AUSTIN, TEXAS OVERVIEW The goal of NCHRP Project 17-76 was to identify the relationships among roadway characteristics (including posted speed limit), travel speeds, and crashes on urban/suburban streets. The database was developed as a fusion of several databases, as discussed below. It includes data for operating speed, crashes, and roadway characteristics, including posted speed limit and vehicle volume. DATABASE DEVELOPMEMT Operating Speed Data from On-Road Tubes On-road tube speed data collection is a common method for obtaining short-term vehicular speed and volume data. Typically, the count of vehicles is stored within 5-mph speed bins. An advantage of on-road tube data is that the speed readings are associated with the volume present during that time period. While the volume for the hour is associated with the speed data, a limitation with on-road tube speed binned data is that free-flow speed cannot be determined for the site. This study used data from on-road tube sites (RTSs) from two different sources:  City of Austin traffic count data.  Data collected from selected sites in Austin as part of this research. The City of Austin used on-road tubes to collect volume and speed data in several locations between 2016 and 2017. The data were binned in 5-mph increments and were used to investigate conditions for potential traffic calming solutions, among other purposes. The City of Austin made those counts available online (211). The research team downloaded the binned speed data and traffic volume data. Table 19 lists the description of the data fields for the counts. The research team disregarded a few of the sites (around 2 percent of all sites) because it appeared that the speed data were incomplete for the sites. Sites were not considered for analysis if many of the hours had zero vehicles or if the number of readings for one direction was much lower than the number of readings for the opposing direction. Furthermore, speed readings were only available for one one-way street, so that site was also removed from the database. A review of the data available from the City of Austin demonstrated that most of the sites studied were on two-lane roads classified as local streets. To increase the number of four-lane sites, a local vendor installed road tubes at 26 locations (52 sites). Speed and vehicle volume data were recorded for a 2-week period in July 2018. The dataset contained speed data for 4,128,083 vehicles after filtering out vehicles with an axle value as 0. The data were collected at the individual vehicle level, which provided the following key information:  Speed of individual vehicle (mph).  Length of individual vehicle (inches).  Vehicle class.  Numbers of axles.

NCHRP Web-Only Document 291: Development of a Posted Speed Limit Setting Procedure and Tool 69  Gap (seconds).  Follow (inches). Figure 9 shows the count locations included in the database. These locations are primarily in Travis County, with a few sites in Williamson County. Table 19. Austin on-road binned tube data file format. Field Name Type Description ROW_ID Text Unique record identifier DATA_FILE Text Unique identifier assigned to data file SITE_CODE Text Unique identifier assigned to site DATETIME Time stamp Time stamp YEAR Integer Year LATITUDE Float Latitude of the road tube location LONGITUDE Float Longitude of the road tube location MONTH Integer Month DAY_OF_MONTH Integer Day of the month DAY_OF_WEEK Integer Day of the week TIME Time stamp Hour SPEED_CHANNEL Text Direction COUNT_TOTAL Integer Traffic volume count SPEED_0_14 Integer Count of vehicles within the speed limit bin SPEED_15_19 Integer Count of vehicles within the speed limit bin SPEED_20_24 Integer Count of vehicles within the speed limit bin SPEED_25_29 Integer Count of vehicles within the speed limit bin SPEED_30_34 Integer Count of vehicles within the speed limit bin SPEED_35_39 Integer Count of vehicles within the speed limit bin SPEED_40_44 Integer Count of vehicles within the speed limit bin SPEED_45_49 Integer Count of vehicles within the speed limit bin SPEED_50_54 Integer Count of vehicles within the speed limit bin SPEED_55_59 Integer Count of vehicles within the speed limit bin SPEED_60_64 Integer Count of vehicles within the speed limit bin SPEED_65_69 Integer Count of vehicles within the speed limit bin SPEED_70_200 Integer Count of vehicles within the speed limit bin

NCHRP Web-Only Document 291: Development of a Posted Speed Limit Setting Procedure and Tool 70 Figure 9. Austin road tube traffic counter locations. Table 20 lists the distribution of number of segments and total segment length by functional classification. Urban local roadway is the dominant functional class of the collected data, while urban major collector is the second most dominant functional class.

NCHRP Web-Only Document 291: Development of a Posted Speed Limit Setting Procedure and Tool 71 Table 20. Number of segments by functional classification. Functional Classification Code Number of Segments Length (mi) Urban Other Principal Arterial U3 63 31 Urban Minor Arterial U4 60 44 Urban Major Collector U5 151 88 Urban Minor Collector U6 29 11 Urban Local U7 360 132 All Sites All 663 305 Table 21 and Table 22 list the distribution of number of segments and total segment length by posted speed limit and number of lanes, respectively. The dominant groups for these variables are roadways with a posted speed limit of 30 mph and two-lane roadways. Table 21. Number of segments by posted speed limit. Posted Speed Limit (mph) Total Length (mi) 25 169 52 30 318 138 35 68 36 40 51 37 45 43 28 50 12 13 55 2 2 Grand Total 663 305 Table 22. Number of segments by number of lanes. Number of Lanes Total Length (mi) 2 529 222 4 126 78 6 8 6 Grand Total 663 305 Crash Records Information System TxDOT is responsible for assembling and maintaining the traffic crash database known as the Crash Records Information System (CRIS). CRIS contains multiple tables (crash, unit, and person) that are linked by a common crash designation identification number. The research team used data for 7 years (2011–2017). Average Annual Daily Traffic TxDOT also maintains a database that includes a variety of roadway characteristics as the Roadway Highway Inventory Network Offload (RHiNO). TxDOT specialists typically associate AADT values with corresponding roadway sections, and the data are stored in RHiNO. The research team used the 2016 RHiNO to extract annual average daily volume (ADT_ADJ) for 2013–2016. The 2017 data were not available at the time this database was assembled.

NCHRP Web-Only Document 291: Development of a Posted Speed Limit Setting Procedure and Tool 72 Roadway Geometry and Traffic Control Device Characteristics Data Table 23 lists descriptions of the specific geometric variables considered for this investigation. These variables were primarily chosen based on the findings from the literature review. Table 23. Roadway and traffic control device variables for City of Austin, Texas. Variable Description Site Unique name for each site, consists of a segment number plus the primary direction for traffic (e.g., NB, SB, EB, or WB). Beg_IT_Legs Number of legs for intersection at beginning of segment. Bike_1yes Bike lane presence: 1=yes, 0=no. Curb_1yes Is curb and gutter present on segment: 1=yes or 0=no. Develop Development: Com/Ret/Ind (for commercial, retail, or industrial), Residential, or Rural/Parks (for areas with park-like or rural-like settings). DrvUsigPerMileBoth Driveways and unsignalized intersections per mile in both directions. End_IT_Legs Number of legs for intersection at ending of segment. Horz_1tan Horizontal alignment: 1=straight(tangent), 0=some horizontal curvature. Len_mi Segment length (mi). LnWdG Lane width (ft) for the segment grouped into N=Narrow (7, 8, 9, or 10 ft), T=Typ (11 or 12 ft), W=Wide (13 ft or more). Median Median type: none, TWLTL, raised. MedWidth Typical or average median width for the segment (ft). NumSigInt Number of signalized intersections along segment, including the signals at the beginning or end of the segment. OnStreetPark On-street parking (either marked or unmarked), subdivided by parking width: None, Yes—Nar (6 or 7 ft wide), Yes—Typ (8 or 9 ft wide). PedAuto Typical or average distance between the sidewalk and the automobile lane for the segment, sum of the following (when present): parking width, bike width, bike-auto separation, and sidewalk to road separation (ft). PedCross_1yes Is a midblock marked pedestrian crossing present within the segment: 1=yes or 0=no. PSL Posted speed limit (mph). RoadSurf Distance between the driving surface edges, calculated as number of through lanes multiplied by average lane width plus median width plus parking widths plus bike widths. RU_F_SYSTEM TxDOT functional classification for street: U3=urban other principal arterial, U4=urban minor arterial, U5=urban major collector, U6=urban minor collector, U7=urban local. SchZone_1yes School zone presence: 1=yes, 0=no. Sidewalk_1yes Is a sidewalk present within the segment: 1=yes or 0=no. Vol_Day Volume per day in both directions. Typically, the value is from TxDOT RHiNO’s ADT_ADJ. If ADT_ADJ is not available or when ADT_ADJ = 405 (a known placeholder), the value is the average daily volume from the on-road counter. The research team determined that the most efficient approach to building the database was to investigate existing sources to obtain the needed data. TxDOT maintains several datasets and ArcGIS maps; however, none contain all the desired information. Therefore, the research

NCHRP Web-Only Document 291: Development of a Posted Speed Limit Setting Procedure and Tool 73 team used Google® Earth to gather the necessary roadway and traffic control device data. The team members gathered the geometric data using the measurement tool, acquired the posted speed limit information using the Street View feature, and used the historical Street View feature to confirm the speed limit that existed at the time the traffic count was made. Data Integration Overview To determine the association among operating speed, roadway characteristics, and crash outcomes, this study required the research team to develop a conflated database where the RTS was conflated with RHiNO and CRIS to develop the database for analysis. RTSs have three geo-locations (in latitude and longitude):  Segment begin location.  Segment end location.  Road tube counter location. The spatial conflation of these datasets is challenging because the RTS locations contain network-level information for each direction, and the RHiNO data are non-directional. Additionally, RHiNO has roadway segments that are not included in the RTS database. The data integration work was divided into four processes:  Process 1: Conflate RTS on RHiNO segments.  Process 2: Assign crashes to RTS.  Process 3: Develop a wider list of speed measures at site and hour level.  Process 4: Identify segment and intersection crashes. It is important to note that RHiNO datasets do not cover all roadway networks, for example, local city streets. The RTS dataset included a site that was outside the boundary of Travis County. When the site was selected for study, it appeared to be within the county border; however, the research team learned later that it was just beyond the county border. In order to not lose that site from the study, the researchers considered crashes from Williamson County. Process 1: Conflation of Tube Locations with Roadway Inventory Data The objective of Process 1 was to assign the road tube locations to the RHiNO segments to be able to perform the analysis at the roadway segment level. The following steps were taken to assign the RTS to RHiNO segments:  Step 0: Create a shapefile of RHiNO data (ArcGIS).  Step 1: Create point-level shapefiles for the RTS beginning and end locations (ArcGIS).  Step 2: Use the “Near” function to assign the RTS locations to the nearby RHiNO segments (ArcGIS).  Step 3: Develop an identifier to separate RHiNO segments that contain several RTSs (ArcGIS).  Step 4: Divide RHiNO segments to match with RTS beginning and end locations (ArcGIS).  Step 5: Create segment-level RTS shapefiles with appropriate RHiNO sites (ArcGIS). Process 2: CRIS Crash Assignment The objective of Process 2 was to integrate line and point shapefiles (RTS and CRIS crashes) such that each relevant CRIS crash record was associated with a roadway segment that

NCHRP Web-Only Document 291: Development of a Posted Speed Limit Setting Procedure and Tool 74 had speed data to allow for the speed and crash analysis. The software used during this effort included ArcGIS and R (212). The following steps were taken in Process 2:  Step 0: Identify relevant crash data from CRIS (2011–2017). Create a shapefile of crash data by converting crashes with available geo-locations into shapefiles, as shown in Figure 10 (Crash Dataset 1). Develop a separate spreadsheet of crash data that do not have geo-locations but where the name of the main road matches with road RTS roadway names (Crash Dataset 2; R and ArcGIS).  Step 1: Create a 30-ft buffer around the RTS segments (ArcGIS).  Step 2: Spatially join Crash Dataset 1 with resulting nearest distance (see Figure 11). It is important to note that the beginning and end points of the segments had an additional 30-ft buffer for this analysis. In this step, each crash case was assigned to the closest RTS segments in both directions (ArcGIS).  Step 3: Generate crash-level output csv file. In these data, each row represents a crash case and is assigned to an RTS when applicable (R).  Step 4: Produce RTS-level output csv file. This step combines RTS-level manual geometric data, RHiNO attributes, and segment-level crash information (R). Figure 10. Example of crashes (dots) and road tube locations (squares with labels starting with ST) for a road in Austin.

NCHRP Web-Only Document 291: Development of a Posted Speed Limit Setting Procedure and Tool 75 Figure 11. Example of crash assignment using 30-ft buffer along a sample road segment. Process 3: Calculation of Speed Measures The objective of Process 3 was to construct candidate speed measures that were suitable to demonstrate the association among speed measures, roadway geometry, traffic volume, and crash outcomes. RTS traffic volume and speed measures were collected from two sources—the City of Austin road tube data and the vendor-collected data. The City of Austin road tube data contain the number of vehicles recorded within the bin limits for a given hour, while the vendor- collected data have the speed for the individual vehicle. Each vehicle within a bin was randomly assigned a value according to a uniform distribution so that the speed measures, especially standard deviation, would better represent a typical value. For example, if there were 50 vehicles within the speed bin of 25 to 29, these vehicles would be randomly assigned a value of 25, 26, 27, 28, or 29. The field data for 52 sites in Austin had speed data at an individual level. These data were converted to an hour level of speed measures and traffic volume data for consistency. The research team developed a wider list of speed measures for these two datasets. Then, a list of common speed measures between both datasets was developed. Table 24 provides a list of these speed measures.

NCHRP Web-Only Document 291: Development of a Posted Speed Limit Setting Procedure and Tool 76 Table 24. Initial speed measures developed. Initial Speed Measures Description PSLMinusSped85 Posted speed limit minus 85th percentile speed (mph) Spd85 85th percentile speed (mph) SpdAve Average speed for the site (mph) StdSpd Standard deviation of speeds for the site (mph) PerOverPSL Percent of observations over the speed limit for the site Pace_LV Lower speed value of 10-mph pace for the site Pace_UV Upper speed value of 10-mph pace for the site Pace_Bin Range of the pace Pace_Per Percent of vehicles in 10-mph pace for the site SpdAve_Hr_Ca.Qa Average speed per hour per site for both the City of Austin traffic count data and the data collected as part of this research (mph) Process 4: Identify Segment and Intersection Crashes Segment crashes were identified using the “intersection-related” variable within the CRIS dataset. Crashes coded as “driveway access” or “not intersection” were considered segment crashes, while the remaining levels of “intersection” and “intersection related” were considered intersection crashes. To obtain a shorter variable name, the segment crashes were called NID (for not intersection and driveways), while the intersection crashes were called IRI (for intersection and intersection related). The research team also grouped the crashes by those with fatalities or injuries (i.e., KABC) and those at all severity levels (KABCO). Descriptive Statistics The research team developed two different data structures to perform the analysis:  Site-level data (663 site-level data points).  Site–temporal-level data (15,446 hourly observations, where each hour had at least 30 vehicles). Roadway Characteristics Table 25 provides the descriptive statistics for the variables considered in the analyses.

NCHRP Web-Only Document 291: Development of a Posted Speed Limit Setting Procedure and Tool 77 Table 25. Descriptive statistics of variables for City of Austin, Texas. Variablea Variable Typeb Minimum Maximum Mean Std. Deviation Beg_IT_Legs Numerical 1 5 3.48 0.59 DrvUsigPerMileBoth Numerical 0 174.4 47.33 41.76 End_IT_Legs Numerical 1 5 3.48 0.58 Len_mi Numerical 0.06 2.77 0.46 0.29 MedWidth Numerical 0 50 3.01 6.79 NumSigInt Numerical 0 2 0.68 0.85 PedAuto Numerical 0 37 5.83 5.62 PSL Numerical 25 55 31.42 6.20 RoadSurf Numerical 18 100 41.67 15.07 Vol_Day Numerical 92 44673 6749.17 9404.19 Bike_1yes Dichotomous 0 1 0.22 0.42 Curb_1yes Dichotomous 0 1 0.95 0.22 Horz_1tan Dichotomous 0 1 0.35 0.48 PedCross_1yes Dichotomous 0 1 0.10 0.31 SchZone_1yes Dichotomous 0 1 0.09 0.28 Sidewalk_1yes Dichotomous 0 1 0.64 0.48 Develop Polychotomous ComRetInd (n=130), Residential (n=525), Rural/Parks (n=8) LnWdG Polychotomous N=Narrow (7 to 10 ft wide) (n=208), T=Typical (11 to 12 ft wide) (n=157), or W=Wide (greater than 13 ft wide) (n=298) Median Polychotomous Raised (n=53), TWLTL (n=81), None (n=529) OnStreetPark Polychotomous Yes—Typ=parking lane present with width of 8 or 9 ft (n=82), Yes—Nar=parking lane present with width of 6 or 7 ft (n=117), None (n=464) RU_F_SYSTEM Polychotomous U3 c (n=63), U4 (n=60), U5 (n=151), U6 (n=29), U7 (n=360) a Variable descriptions are in Table 23. b For dichotomous variables, “1” indicates the presence of the feature and “0” indicates its absence. For polychotomous variables, the numbers in parentheses represent frequencies of the corresponding categories. c See Table 20. Speed Distributions The research team first performed exploratory data analysis, examining distributions and the empirical relationship between pairs of variables based on plots. Figure 12 shows the basic relationship between average and 85th percentile speed and posted speed limit using the hourly data. Overall, average speed is similar to the posted speed limit, while the 85th percentile speed exceeds the posted speed limit. Note that this plot reflects binned speed data rather than free-flow speed data. Free-flow speed data include vehicles that are not affected by other vehicles, while binned speed data include all vehicles. Free-flow speed is traditionally used to set posted speed limits and is frequently used when relating operating speed to roadway characteristics (114); however, the emphasis of this study was to consider crashes, which could be influenced by the speed distribution for all vehicles, not just the free-flow vehicles.

NCHRP Web-Only Document 291: Development of a Posted Speed Limit Setting Procedure and Tool 78 (a) Average Speed (mph) (b) 85th Percentile Speed (mph) Figure 12. Operating speed versus posted speed limit for city streets in Austin. Violin plots are similar to box and whisker plots in that they compare distributions of quantitative data across several levels of categorical variables. They show the variabilities between key contributing factors. Unlike the box plot, the violin plot illustrates a kernel density estimation of the underlying distribution. This can be an effective and attractive way to show multiple distributions of data at once; however, the estimation procedure is influenced by the sample size, so violins for relatively small samples might look misleadingly smooth. Additionally, the lower smoothing points of these estimations go beyond zero values, but the actual speed measures are always greater than zero.

NCHRP Web-Only Document 291: Development of a Posted Speed Limit Setting Procedure and Tool 79 Figure 13 contains several violin plots for average speed. It shows that the average speed is greatest on road segments in rural or park-like areas compared to commercial/retail/industrial or roads in residential areas, which had the lowest mean. It is important to note that only eight of the 663 segments were classified as rural or park development in this study, which could explain why the kernel distribution is so heavily concentrated around the mean. Another observation from Figure 13 is that road segments with a raised median and road segments with a TWLTL had similar means and kernel distributions for average speed, while road segments with no median present had a lower mean with a kernel distribution that was more heavily concentrated below the mean. The violin plots also reveal that roads with wide lane width have a lower mean speed average than narrow or typical lane widths. The presence of a sidewalk was also associated with a greater average speed, while the presence of on-street parking was associated with a lower average speed. Figure 13 also shows that road segments classified as either U3 (urban other principal) or U4 (urban minor arterial) had the highest average speed, while U5 (urban major collector), U6 (urban minor collector), and U7 (urban local) streets each had a lower mean than the previous. Figure 13. Violin plots for average speed (SpdAve) for different geometric features. Figure 14 provides the violin plots for the percent of vehicles over the speed limit by several roadway characteristics. It shows that the percent of observations over the speed limit is greatest on road segments in rural or park-like areas compared to commercial/retail/industrial or

NCHRP Web-Only Document 291: Development of a Posted Speed Limit Setting Procedure and Tool 80 roads in residential areas; commercial development had the lowest mean. As noted previously, only eight of the 663 segments were classified as rural or park development in this study, which could explain why the kernel distribution is so heavily concentrated around the mean. Another finding shown in Figure 14 is that road segments with a raised median had the greatest mean for percent over the speed limit and the kernel distribution was more concentrated above the mean in comparison to road segments with a TWLTL and road segments with no median present. The violin plots also reveal that roads with wide lane width have a lower mean value than narrow or typical lane widths. The presence of a sidewalk did not seem to have a significant effect on the percent over the speed limit, although road segments with no sidewalk did have a slightly lower mean. The presence of on-street parking was associated with a lower percent over the speed limit. Road segments classified as either U4 (urban minor arterial) or U5 (urban major collector) had the highest percent of drivers over the posted speed limit. Figure 14. Violin plots for percent of vehicles over the posted speed limit (PerOverPSL) for different geometric features. ROUND 1: DATA ANALYSIS USING NEGATIVE BINOMIAL REGRESSION Overview Researchers applied NB regression models to the site-level data to investigate the relationship between crashes and speed measures while controlling for the effects of other

NCHRP Web-Only Document 291: Development of a Posted Speed Limit Setting Procedure and Tool 81 variables including AADT and roadway geometry variables. The research team examined several NB regression models. The models included the log of segment length as an offset variable. To account for correlations in crash counts from the same road segment in estimation, the generalized estimating equations (GEE) procedure was used to estimate NB regression models. The same crash count (segment-level crash count) was used for both directions of travel at a site. When multiple hours or days of speed data were available at a site, all speed data were used to generate a representative single speed measure for each direction of travel. Variable Relationships with Crashes To examine the relationships among crashes and roadway characteristics, including posted speed limit and volume, the research team selected non-intersection or segment crashes rather than all crashes. All crashes would have included intersection crashes, especially signalized intersection crashes. Focusing on segment crashes was expected to make it easier to identify a relationship between crashes and posted speed limit. The research team conducted evaluations using crashes with injuries (i.e., KABC) and all severity-level crashes (KABCO) in case the evaluations that included only the fatal/injury crashes were limited by sample size. Fatal and Injury Segment Crashes (KABC_NID) Multiple NB regression models were developed in order to identify a model that was physically meaningful as well as statistically significant (containing variables that were significant with p-values about 0.1 or lower). Table 26 provides the significance per variable, while Table 27 provides the estimate along with the p-value for each variable/level combination for the selected model. Key observations include the following:  As expected, greater vehicle volume was associated with a higher number of crashes.  Presence of a TWLTL was associated with more KABC segment crashes compared to no median, and presence of a raised median was associated with fewer segment crashes compared to no median or TWLTL. Care must be taken in interpreting higher KABC segment crashes at TWLTL segments compared to no median segments because road segments with no median present had lower speeds on average compared to road segments with TWLTL (or raised medians), as can be seen from Figure 5, while speeds of road segments with TWLTL and raised medians were similar. This finding suggests that more KABC segment crashes at TWLTL segments compared to no median segments could have been due (at least in part) to higher speeds (as well as other extraneous factors) at TWLTL segments compared to those at segments with no median present.  Wider medians were associated with fewer KABC segment crashes.  More signals (1 versus 0 or 2 versus 1 since the variable only had values of 0, 1, 2) were associated with more KABC crashes, even for segment (midblock) crashes.  The presence of on-street parking was associated with more KABC crashes. When the available space for the on-street parking was narrow (6 to 7 ft), there were more crashes than when the on-street parking was a typical width (8 to 9 ft).  Larger standard deviations (more variability) of speeds were associated with more KABC crashes.  The presence of curb and gutter was associated with fewer KABC crashes.

NCHRP Web-Only Document 291: Development of a Posted Speed Limit Setting Procedure and Tool 82 Table 26. Score statistics for model shown in Table 27. Variable DF Chi-Square Pr > ChiSq NumSigInt 1 15.51 <.0001 LnVol 1 21.53 <.0001 StdSpd_Ca_Qa 1 7.72 0.0055 Median 2 8.41 0.0149 OnStreetPark 2 5.22 0.0735 MedWidth 1 4.25 0.0392 Sidewalk_1yes 1 3.66 0.0556 Curb_1yes 1 2.67 0.1020 Note: Score statistics for Type 3 GEE analysis. Variable descriptions are in Table 23. LnVol=Log(Vol_Day). Table 27. Variables with significant effects on KABC segment (KABC_NID) crashes. Variable Level Estimate Standard Error 95% Confidence Limits Z Pr > |Z| Intercept −4.2184 0.9168 −6.0152 −2.4215 −4.60 <.0001 NumSigInt 0.6834 0.1230 0.4424 0.9244 5.56 <.0001 LnVol 0.5887 0.1057 0.3816 0.7958 5.57 <.0001 StdSpd_Ca_Qa 0.1758 0.0508 0.0762 0.2754 3.46 0.0005 Median Raised −0.0422 0.3352 −0.6991 0.6147 −0.13 0.8999 Median TWLTL 0.5902 0.2794 0.0427 1.1378 2.11 0.0346 Median None 0.0000 0.0000 0.0000 0.0000 . . OnStreetPark Yes—Typ 0.0731 0.2707 −0.4575 0.6037 0.27 0.7872 OnStreetPark Yes—Nar 0.3970 0.1458 0.1112 0.6829 2.72 0.0065 OnStreetPark None 0.0000 0.0000 0.0000 0.0000 . . MedWidth −0.0435 0.0160 −0.0749 −0.0121 −2.71 0.0067 Sidewalk_1yes 0.2597 0.1528 −0.0397 0.5591 1.70 0.0891 Curb_1yes −0.9195 0.3396 −1.5850 −0.2540 −2.71 0.0068 Notes: Analysis of GEE parameter estimates. Empirical standard error estimates. Variable descriptions are in Table 23. LnVol=Log(Vol_Day). . = value is not relevant since this level is the base for the variable. {blank} = value not relevant because the variable is not a multicategory variable. The presence of a sidewalk was associated with more KABC crashes. Caution needs to be taken when interpreting this finding because association does not imply causation. That is, it does not imply that the presence of a sidewalk results in more crashes. The research team currently does not have a theory as to why the presence of a sidewalk would be associated with more vehicle crashes. It is possible that a confounding factor or lurking variable such as the location where a sidewalk is present (recall that the presence of a sidewalk was associated with a greater average speed in Figure 13) may explain this outcome. All Segment Crashes (KABCO_NID) Similar to the effort for segment (midblock) injury crashes, the research team developed several NB regression models to understand the relationship between all segment crashes and roadway characteristics. Table 28 provides the significance per variable, while Table 29 provides the estimate along with the p-value for each variable/level combination for the selected model. The following observations are similar to the findings for midblock injury crashes:  As expected, more vehicle volume was associated with more crashes.

NCHRP Web-Only Document 291: Development of a Posted Speed Limit Setting Procedure and Tool 83  Presence of a TWLTL was associated with more segment crashes compared to no median, and presence of a raised median was associated with fewer segment crashes compared to no median or TWLTL. Care must be taken in interpreting higher segment crashes at TWLTL segments compared to no median segments because road segments with no median present had lower speeds on average compared to road segments with TWLTL (or raised medians), as can be seen from Figure 5, while speeds of road segments with TWLTL and raised medians were similar. This finding suggests that more segment crashes at TWLTL segments compared to no median segments could have been due (at least in part) to higher speeds (as well as other extraneous factors) at TWLTL segments compared to those at segments with no median present.  Wider medians were associated with fewer segment crashes.  More signals were associated with more crashes.  The only speed measure that was significant was standard deviation (of speeds). Larger standard deviations were associated with more crashes.  Presence of a curb and gutter was associated with fewer crashes.  Presence of a sidewalk was associated with more crashes. The research team currently does not have a theory as to why the presence of a sidewalk would be associated with more vehicle crashes. Note again that this is not a causal relationship, as mentioned before. An observation that is different from the midblock injury crash findings is:  Presence of on-street parking was not significant when examining midblock crashes that included all severity levels. Table 28. Score statistics for model shown in Table 29. Source DF Chi-Square Pr > ChiSq NumSigInt 1 16.52 <.0001 LnVol 1 26.23 <.0001 Sidewalk_1yes 1 9.67 0.0019 Median 2 9.92 0.0070 StdSpd_Ca_Qa 1 6.94 0.0084 MedWidth 1 3.50 0.0613 Curb_1yes 1 2.66 0.1028 Note: Score statistics for Type 3 GEE analysis. Variable descriptions are in Table 23. LnVol=Log(Vol_Day).

NCHRP Web-Only Document 291: Development of a Posted Speed Limit Setting Procedure and Tool 84 Table 29. Variables with significant effects on KABCO segment (KABCO_NID) crashes. Variable Level Estimate Standard Error 95% Confidence Limits Z Pr > |Z| Intercept −2.0840 0.7516 −3.5572 −0.6108 −2.77 0.0056 Curb_1yes −0.7279 0.2919 −1.2999 −0.1558 −2.49 0.0126 Median Raised −0.3299 0.2271 −0.7750 0.1151 −1.45 0.1462 Median TWLTL 0.4068 0.1968 0.0210 0.7926 2.07 0.0388 Median None 0.0000 0.0000 0.0000 0.0000 . . MedWidth −0.0285 0.0091 −0.0464 −0.0106 −3.12 0.0018 NumSigInt 0.5533 0.0986 0.3601 0.7465 5.61 <.0001 Sidewalk_1yes 0.3892 0.1331 0.1284 0.6500 2.93 0.0034 StdSpd_Ca_Qa 0.1478 0.0445 0.0607 0.2350 3.33 0.0009 LnVol 0.4601 0.0840 0.2954 0.6248 5.47 <.0001 Note: Variable descriptions are in Table 23. LnVol=Log(Vol_Day). . = value is not relevant since this level is the base for the variable. {blank} = value not relevant because the variable is not a multicategory variable. Variable Relationships with Posted Speed Limit A better understanding of relationships between roadway characteristics and posted speed limit may assist with identifying how the context—sometimes referred to as the look and feel— of a road could help to communicate the anticipated speed and the anticipated posted speed for the facility. Table 30 provides a summary of the variables with statistically significant effects on posted speed limit. These results illustrate how certain roadway characteristics tend to be associated with different posted speed limits. Key observations include the following:  Roads with on-street parking were associated with lower speed limits.  Higher access density (i.e., DrvUsigPerMileBoth) was associated with lower speed limits.  Lower speed limits were associated with the presence of bike lanes and curb/gutter.  Higher speed limits were associated with straight roads.  For this Austin database, roads with raised medians had lower posted speed limits, while roads with TWLTL had higher posted speed limits, compared to roads with no medians.  For this database, roads with rural or park-like development had higher posted speeds limits compared to the other development types (i.e., residential or commercial/retail/industrial). This finding should be used with caution due to the low number of segments with rural/park development (only eight of the 663 segments).

NCHRP Web-Only Document 291: Development of a Posted Speed Limit Setting Procedure and Tool 85 Table 30. Variables with significant effects on posted speed limit. Parameter Level DF Estimate Standard Error Wald 95% Confidence Limits Wald Chi- Square Pr > ChiSq Intercept 1 31.7757 1.4273 28.9782 34.5731 495.64 <.0001 Bike_1yes 1 −1.3754 0.4205 −2.1996 −0.5512 10.70 0.0011 Curb_1yes 1 −1.7321 0.6196 −2.9466 −0.5177 7.81 0.0052 Develop ComRetInd 1 −6.4299 1.3017 −8.9811 −3.8787 24.40 <.0001 Develop Residential 1 −7.5124 1.2292 −9.9217 −5.1031 37.35 <.0001 Develop Rural/Parks 0 0.0000 0.0000 0.0000 0.0000 . . DrvUsigPerMileBoth 1 −0.0132 0.0035 −0.0202 −0.0063 14.00 0.0002 Horz_1tan 1 −0.7000 0.2721 −1.2333 −0.1667 6.62 0.0101 LnWdG N 1 1.3765 0.4203 0.5528 2.2002 10.73 0.0011 LnWdG T 1 1.9571 0.3678 1.2363 2.6780 28.32 <.0001 LnWdG W 0 0.0000 0.0000 0.0000 0.0000 . . Median Raised 1 −2.2826 0.6774 −3.6102 −0.9550 11.36 0.0008 Median TWLTL 1 0.4392 0.5785 −0.6947 1.5731 0.58 0.4478 Median None 0 0.0000 0.0000 0.0000 0.0000 . . NumSigInt 1 0.6765 0.2096 0.2656 1.0874 10.41 0.0013 OnStreetPark Yes—Typ 1 −4.1290 0.5339 −5.1754 −3.0826 59.81 <.0001 OnStreetPark Yes—Nar 1 −2.6834 0.4490 −3.5635 −1.8033 35.71 <.0001 OnStreetPark None 0 0.0000 0.0000 0.0000 0.0000 . . PedAuto 1 0.0659 0.0322 0.0028 0.1290 4.19 0.0407 RoadSurf 1 0.1619 0.0164 0.1298 0.1940 97.87 <.0001 Vol_Day 1 0.0002 0.0000 0.0001 0.0002 46.62 <.0001 Len_mi 1 2.5828 0.5452 1.5143 3.6513 22.45 <.0001 Scale 1 2.9643 0.0814 2.8089 3.1282 NR NR Notes: Analysis of maximum likelihood parameter estimates. Variable descriptions are in Table 23. . = value is not relevant since this level is the base for the variable. {blank} = value not relevant because the variable is not a multicategory variable. NR = value not relevant for the scale parameter. ROUND 2: DATA ANALYSIS USING PATH ANALYSIS APPROACH Overview There are two primary relationships of interest among variables: the relationship between speed-related variables (speed measures) and roadway characteristics, including traffic volume and other roadway geometry and traffic control device variables; and the relationship between crashes and speed-related variables while accounting for other roadway characteristic variables that may confound the relationship between speeds and crashes if not taken into account. In the previous analysis (Round 1 data analysis), the relationship between speed-related variables and roadway characteristic variables and the relationship between crashes and speed-related variables along with roadway characteristic variables were assessed separately. In Round 2 data analysis, the research team analyzed the relationships among variables using a more enhanced statistical modeling approach. For speed, researchers considered several different measures of speed that can quantify various aspects of speed distributions at each segment, including newly developed measures as well as some previous speed measures explored in the Round 1 analysis. Table 31 contains the speed measures, and Table 32 lists the roadway characteristic variables considered in this analysis.

NCHRP Web-Only Document 291: Development of a Posted Speed Limit Setting Procedure and Tool 86 Table 31. Round 2 speed measures considered. Round 2 Speed Measures Description Abs(PSL−Avg) Absolute value of posted speed limit minus average speed (mph) CoefVar Coefficient of variation of speed Pace Percent of vehicles in 10-mph pace for the site (%) PerOvPSL Percent of observations over the speed limit for the site (%) PSL Posted speed limit (mph) PSL−Avg Posted speed limit minus average speed (mph) PSL−S85 Posted speed limit minus 85th percentile speed (mph) S85−Avg 85th percentile speed minus average speed (mph) SpdAve Average speed (mph) StdSpd Standard deviation (mph) Table 32. Roadway characteristic variables considered in path analysis. Path Analysis Variable Original Variable Bike1yes Bike_1yes Curb1yes Curb_1yes DUPMBoth DrvUsigPerMileBoth Horz1tan Horz_1tan MedWidth MedWidth NSigInt NumSigInt PedAuto PedAuto PdCr1yes PedCross_1yes RoadSurf RoadSurf ScZn1yes SchZone_1yes SdWk1yes Sidewalk_1yes LnVol Log(Vol_Day) LnLen Log(Len_mi) Median2 (with 2 categories: Raised=1, NotRaised=0) Median (3 categories: Raised, TWLTL, None) OnStPk2 (with 2 categories: OnStrPrk=1, None=0) OnStreetPark (3 categories: Yes—Typ, Yes—Nar, None) Develop2 (with 2 categories: Resident=1, Other=0) Develop (3 categories: Residential, ComRetInd, Rural/Parks) RU_F_rev (with 2 categories: Local [U7]=1, Not-Loc [U3-U6]=0) RU_F_SYSTEM (5 categories: U3, U4, U5, U6, U7) The focus of this analysis was to assess the effect of speed on crashes while accounting for the effects of other roadway characteristic variables on speed and crashes. Note that while roadway characteristic variables (e.g., traffic volume) may affect both speed and crashes, some variables such as posted speed limit may affect crashes only through speeds and only indirectly affect crashes. A speed variable plays the role of a mediator variable (or intervening variable) between crashes and other variables that affect crashes only indirectly in this case. In addition to assessing the speed-crash relationship, it is also of interest to evaluate indirect effects of roadway characteristics on crashes through a mediator variable (speed) as well as direct effects of roadway characteristics on crashes. To accommodate these general relationships among variables, researchers jointly modeled the relationship between speeds and roadway characteristics and the relationship

NCHRP Web-Only Document 291: Development of a Posted Speed Limit Setting Procedure and Tool 87 between crashes and speeds along with roadway characteristics as well as the speed limit simultaneously based on a coherent structural equation modeling (SEM) framework (213), specifically using path analysis. Path analysis is a special case of SEM where there is no latent variable in the model (i.e., all variables in the model are measured variables). Regardless of its many advantages, path analysis has not been widely used in safety analysis yet. A notable exception is the study by Gargoum and El-Basyouny (10). The path analysis model consists of two submodels in this case: (1) crash model (Model 1) describing the relationship between crashes (outcome variable) and speed (mediator variable) as well as other roadway characteristic variables (independent variables); and (2) speed model (Model 2) describing the relationship between speed and other roadway characteristic variables. For the crash model (Model 1), an NB model with the mean given in Equation (1), which expresses the log mean crash rate as a function of covariates corresponding to a speed variable and other roadway characteristic variables in Table 31 and Table 32, was adopted.  0 1 1expi i m i Ki Km X X         , (1) where iy denotes the observed outcome variable (the number of crashes that occurred on the segment in 7 years) on segment i ( 1, ,i I  );  i iE y  is the expected number of crashes for 7 years; im is the mediator variable (a measure of speed); 1 , ,i KiX X are K covariates; and 0 1, , , ,m K    denote regression coefficients for the outcome variable. For the speed model (Model 2), a normal linear model given in Equation (2) was employed. 0 1 1i i Li L im X X        , (2) where 0 1, , , L   are regression coefficients. Estimation was performed by SEM software Mplus version 8.3 (214). Several different models (with different sets of independent variables) for each mediator variable in Table 31 were explored. Table 32 shows the roadway characteristic variables used in the path analysis. Note that categorical variables with three or more categories needed to be recoded as two-category variables by regrouping because Mplus does not allow multicategory variables. Researchers kept variables in the model if the corresponding p-values were less than 0.2. Note, however, that there may be multiple models that may be adequate for any given dataset. Researchers used a penalized-likelihood criterion, the Bayesian information criterion (BIC), which is a popular tool used for model selection in various applications (see, for example, Kass and Raftery [215]), to select an appropriate model. Although a model with lower BIC is preferred in general, a more physically meaningful model can be selected whenever there is not much difference in BIC values among competing models. Note that BIC can be used for comparing models with different independent variables but not models with different dependent variables. That is, BIC values should not be compared across different outcome variables or mediator variables since they represent different datasets. Results and Discussion The estimated regression coefficients for crashes in Equations (1) and (2), having each of the speed variables in Table 31 as a mediator variable and roadway characteristic variables selected from those in Table 32, are given in Table 33, Table 34, Table 35, and Table 36 The

NCHRP Web-Only Document 291: Development of a Posted Speed Limit Setting Procedure and Tool 88 results for KABC_NID crashes are presented in Table 33 and Table 34, and the results for KABCO_NID (all) crashes are presented in Table 35 and Table 36. Table 33. Estimated regression coefficients for KABC_NID crashes by path analysis (mediator variable: PSL, CoefVar, PerOvPSL, StdSpd, Pace, or SpdAve). Outcome Variable  Independent Variable Mediator Coefficient PSL  CoefVar PerOvPSL StdSpd Pace SpdAve KABC_NID crashes Intercept 0 −3.259 −3.076 −2.144 −3.386 −0.856 −3.359 Mediator Mediator 0.032 2.059 −0.004 0.143 −0.023 −0.001 Curb1yes Cirb1yes −0.827 −1.037 −1.069 −0.916 −0.872 −0.921 Dvelop2 Develop2 −0.362 −0.429 −0.438 −0.399 −0.368 −0.402 Median2 Median2 −0.723 −0.641 −0.632 −0.684 −0.668 −0.680 NSigInt NSigInt 0.602 0.644 0.642 0.638 0.629 0.639 SdWk1yes SdWk1yes 0.263 0.249 0.260 0.246 0.260 0.245 ONSTPK2 ONSTPK2 0.194 0 0 0.173 0.167 0.171 LnVol LnVol 0.488 0.584 0.545 0.542 0.526 0.545 LnLen LnLen 0.720 0.877 0.853 0.828 0.836 0.836 StdSpd StdSpd 0 0 0 NA 0 0.144 Speed Intercept α0 27.534 0.336 66.776 5.218 76.743 8.817 PSL αPSL 0 0 −2.004 0.061 −0.413 0.433 BIKE1YES αBIKE1YES −0.615 0 0 0 1.482 0 Curb1yes αCirb1yes −2.585 −0.029 0 −0.780 4.196 0 Dvelop2 αDevelop2 −1.940 −0.025 0 −0.335 3.905 0 DUPMBoth αDUPMBoth −0.020 0.000 −0.099 0.003 0 −0.022 Horz1tan αHorz1tan  0 0.023 0 0.265 −1.259 −0.941 Median2 αMedian2 −1.601 0 0 0 0 0 NSigInt αNSigInt 0.656 −0.019 6.480 −0.145 0 1.342 ONSTPK2 αONSTPK2 −2.731 0.021 −11.864 0.335 0 −2.167 ROADSURF αROADSURF 0.150 −0.001 0.487 0 0 0.095 PedAuto αPedAuto 0 0 0 −0.025 0 0 VOL/1000 αVOL/1000 0.179 −0.001 0 0 0 0.044 LEN_MI αLEN_MI 4.167 −0.061 36.210 0 0 7.084 Model fit BIC 5968.8  777.5 8471.8 4424.7 6915.4 6210.5 Notes: The coefficient “0” denotes that the corresponding variable was excluded from the model. Cells are highlighted in light gray when the p-value is between 0.05 and 0.1. Cells are highlighted in dark gray when the p-value is less than 0.05. Table 33 and Table 34 show that the mediator variables—StdSpd, Pace, Abs(PSL−Avg), PSL−Avg, PSL−S85, and S85−Avg—had statistically significant effects at α=0.05 on KABC_NID crash frequency. The association was positive for Abs(PSL−Avg), PSL−Avg, PSL−S85, S85−Avg, and StdSpd (i.e., as the values of those mediators increased, crash frequency increased), but negative for Pace (i.e., as the value of Pace increased, crash frequency decreased). Also, the presence of curb and gutter, residential area, and raised medians were associated with lower crash frequency. The number of signalized intersections, traffic volumes, and segment length were found to be positively correlated with crash frequency. As expected, higher PSL was associated with higher Abs(PSL−Avg), PSL−Avg, PSL−S85, S85−Avg, StdSpd,

NCHRP Web-Only Document 291: Development of a Posted Speed Limit Setting Procedure and Tool 89 and AvgSpd, but with lower Pace and PerOvPSL. This finding implies that PSL has indirect effects on crashes through its effect on those speed measures. Table 34. Estimated regression coefficients for KABC_NID crashes by path analysis (mediator variable: Abs(PSL−Avg), PSL−Avg, PSL−S85, or S85−Avg). Outcome Variable Independent Variable Mediator Coefficient Abs(PSL−Avg)  PSL−Avg PSL−S85 S85−Avg KABC_NID crashes Intercept 0 −2.877 −2.400 −3.021 −2.199 −3.171 Mediator Mediator 0.052 0.036 0.025 0.031 0.127 Curb1yes Cirb1yes −0.883 −0.941 −0.858 −0.992 −0.881 Dvelop2 Develop2 −0.378 −0.427 −0.402 −0.438 −0.412 Median2 Median2 −0.615 −0.599 −0.630 −0.603 −0.687 NSigInt NSigInt 0.635 0.632 0.627 0.635 0.639 SdWk1yes SdWk1yes 0.224 0.245 0.247 0.250 0.249 ONSTPK2 ONSTPK2 0.144 0 0 0 0 LnVol LnVol 0.565 0.541 0.528 0.544 0.535 LnLen LnLen 0.875 0.909 0.887 0.887 0.811 StdSpd StdSpd 0 0 0.103 0 0 Speed Intercept α0 −1.346 −8.817 −8.817 −12.625 5.623 PSL αPSL 0.357 0.567 0.567 0.497 0.044 BIKE1YES αBIKE1YES 0.039 0 0 0 0 Curb1yes αCirb1yes −0.633 0 0 0 −0.950 Dvelop2 αDevelop2 −1.291 0 0 0 −0.424 DUPMBoth αDUPMBoth 0.008 0.022 0.022 0.019 0.003 Horz1tan αHorz1tan  0.576 0.941 0.941 0.732 0 Median2 αMedian2 −0.194 0 0 0 0 NSigInt αNSigInt −0.793 −1.342 −1.342 −1.188 −0.127 ONSTPK2 αONSTPK2 0.634 2.167 2.167 1.931 0.420 ROADSURF αROADSURF −0.058 −0.095 −0.095 −0.087 0 PedAuto αPedAuto 0.031 0 0 0 −0.023 VOL/1000 αVOL/1000 −0.078 −0.044 −0.044 −0.050 0 LEN_MI αLEN_MI −3.367 −7.084 −7.084 −6.914 0 Model fit BIC 5907.6 6200.4 6200.4 6172.7 4489.2 Notes: The coefficient “0” denotes that the corresponding variable was excluded from the model. Cells are highlighted in light gray when the p-value is between 0.05 and 0.1. Cells are highlighted in dark gray when the p- value is less than 0.05. Table 35 and Table 36 illustrate that S85−Avg had statistically significant effects at α=0.05, and PSL and PSL−Avg had statistically significant effects at α=0.1 on all crash frequency. The association between those mediator variables and crashes was positive (i.e., as the values of those mediators increased, crash frequency increased). Other mediator variables were, however, statistically insignificant. Also, the presence of curb and gutter and raised medians were associated with lower crash frequency. The number of signalized intersections, presence of a sidewalk, presence of on-street parking, traffic volumes, and segment length were found to be positively correlated with all crash frequency. It can also be observed that higher PSL was associated with higher PSL−Avg and S85−Avg, which implies that PSL has indirect effects on crashes through its effect on those speed measures.

NCHRP Web-Only Document 291: Development of a Posted Speed Limit Setting Procedure and Tool 90 Table 35. Estimated regression coefficients for KABCO_NID crashes by path analysis (mediator variable: PSL, CoefVar, PerOvPSL, StdSpd, Pace, or SpdAve). Outcome Variable Independent Variable Mediator Coefficient PSL  CoefVar PerOvPSL StdSpd Pace AvgSpd KABCO_NID Intercept 0 −1.970 −1.699 −1.183 −2.132 0.322 −2.247 Mediator Mediator 0.027 1.215 −0.001 0.127 −0.022 0.005 Curb1yes Cirb1yes −0.854 −1.034 −1.048 −0.905 −0.849 −0.884 Dvelop2 Develop2 −0.261 −0.318 −0.329 −0.289 −0.255 −0.279 Median2 Median2 −0.706 −0.648 −0.650 −0.678 −0.654 −0.697 NSigInt NSigInt 0.491 0.537 0.537 0.522 0.519 0.516 SdWk1yes SdWk1yes 0.346 0.340 0.343 0.340 0.347 0.343 ONSTPK2 ONSTPK2 0.366 0.324 0.326 0.348 0.347 0.356 LnVol LnVol 0.454 0.530 0.505 0.499 0.475 0.489 LnLen LnLen 0.682 0.793 0.763 0.764 0.762 0.736 StdSpd StdSpd 0 0 0 0 0 0.126 Speed Intercept α0 27.534 0.336 66.776 5.218 75.661 8.817 PSL αPSL NA 0 −2.004 0.061 −0.379 0.433 BIKE1YES αBIKE1YES −0.615 0 0 0 0 0 Curb1yes αCirb1yes −2.585 −0.029 0 −0.780 4.693 0 Dvelop2 αDevelop2 −1.940 −0.025 0 −0.335 3.760 0 DUPMBoth αDUPMBoth −0.020 0.000 −0.099 0.003 0 −0.022 Horz1tan αHorz1tan 0 0.023 0 0.265 −1.333 −0.941 Median2 αMedian2 −1.601 0 0 0 0 0 NSigInt αNSigInt 0.656 −0.019 6.480 −0.145 0 1.342 ONSTPK2 αONSTPK2 −2.731 0.021 −11.864 0.335 0 −2.167 ROADSURF αROADSURF 0.150 −0.001 0.487 0 0 0.095 PedAuto αPedAuto 0 0 0 −0.025 0 0 VOL/1000 αVOL/1000 0.179 −0.001 0 0 0 0.044 LEN_MI αLEN_MI 4.167 −0.061 36.210 0 0 7.084 Model fit BIC 6915.0 1731.3 9426.0 5369.8 7855.2 7155.1 Notes: The coefficient “0” denotes that the corresponding variable was excluded from the model. Cells are highlighted in light gray when the p-value is between 0.05 and 0.1. Cells are highlighted in dark gray when the p-value is less than 0.05.

NCHRP Web-Only Document 291: Development of a Posted Speed Limit Setting Procedure and Tool 91 Table 36. Estimated regression coefficients for KABCO_NID crashes by path analysis (mediator variable: Abs(PSL−Avg), PSL−Avg, PSL−S85, or S85−Avg). Outcome Variable  Independent Variable Mediator Coefficient Abs(PSL−Avg) PSL−Avg PSL−S85 S85−Avg KABCO _NID Intercept 0 −1.433 −1.316 −2.055 −1.187 −2.101 Mediator Mediator 0.022 0.023 0.012 0.016 0.123 Curb1yes Cirb1yes −0.970 −0.988 −0.894 −1.022 −0.885 Dvelop2 Develop2 −0.304 −0.314 −0.284 −0.324 −0.293 Median2 Median2 −0.641 −0.613 −0.649 −0.626 −0.675 NSigInt NSigInt 0.529 0.526 0.517 0.531 0.529 SdWk1yes SdWk1yes 0.334 0.335 0.336 0.339 0.330 ONSTPK2 ONSTPK2 0.328 0.323 0.341 0.325 0.332 LnVol LnVol 0.509 0.507 0.502 0.506 0.503 LnLen LnLen 0.771 0.815 0.800 0.791 0.764 StdSpd StdSpd 0 0 0.111 0 0 Speed Intercept α0 −1.346 −8.817 −8.817 −12.625 5.623 PSL αPSL 0.357 0.567 0.567 0.497 0.044 BIKE1YES αBIKE1YES 0.039 0 0 0 0 Curb1yes αCirb1yes −0.633 0 0 0 −0.950 Dvelop2 αDevelop2 −1.291 0 0 0 −0.424 DUPMBoth αDUPMBoth 0.008 0.022 0.022 0.019 0.003 Horz1tan αHorz1tan 0.576 0.941 0.941 0.732 0 Median2 αMedian2 −0.194 0 0 0 0 NSigInt αNSigInt −0.793 −1.342 −1.342 −1.188 −0.127 ONSTPK2 αONSTPK2 0.634 2.167 2.167 1.931 0.420 ROADSURF αROADSURF −0.058 −0.095 −0.095 −0.087 0 PedAuto αPedAuto 0.031 0 0 0 −0.023 VOL/1000 αVOL/1000 −0.078 −0.044 −0.044 −0.050 0 LEN_MI αLEN_MI −3.367 −7.084 −7.084 −6.914 0 Model fit BIC 6862.8 7157.4 7153.8 7129.3 5437.2 Notes: The coefficient “0” denotes that the corresponding variable was excluded from the model. Cells are highlighted in light gray when the p-value is between 0.05 and 0.1. Cells are highlighted in dark gray when the p- value is less than 0.05. ROUND 3: DATA ANALYSIS USING PATH ANALYSIS APPROACH AND SEGMENTS WITH POSTED SPEED LIMITS OF 45 MPH AND LOWER Overview Researchers refitted the path analysis model for KABC crashes with Abs(PSL−Avg) as a mediator variable after excluding the road segments with posted speed limits of 50 and 55 mph because streets with those speed limits may be considered as non-city streets. There was one segment with 55 mph as the PSL and six segments with 50 mph as the PSL in the original data. Excluding those segments resulted in removal of 14 out of 663 sites, leaving 649 sites corresponding to 25- to 45-mph segments in the data. Initially, a full model with all of the

NCHRP Web-Only Document 291: Development of a Posted Speed Limit Setting Procedure and Tool 92 roadway characteristic variables in Table 23 was fitted to the dataset consisting of 649 sites, and then the model was refitted after removing variables that were insignificant at α=0.1. The crash rate for the Austin data was calculated and graphed with the speed metric of PSL−Avg (difference between posted speed limit and average speed) to provide an appreciation of the potential relationship (see Figure 15). While Figure 15 shows crash rate, crash frequency along with segment length and volume were used in the statistical analyses. Figure 15 includes a simple trendline to help illustrate the relationship. The minimum crash rate appears to be near the point when posted speed limit equals average speed. Another observation is that the crash rate is lower when vehicles are traveling within about 5 mph of the posted speed limit. These observations do come with some cautions. The reason the specific speed limit was posted for each of the 649 sites is not known, such as whether the engineer’s decision for the posted speed limit was influenced by existing crashes or if the posted speed limit represents a default speed limit for the road (in Texas the default speed limit for residential streets is 30 mph). Another notable observation is that for many 25- to 45-mph roads, a 5-mph increase from average speed is very close to a typical 85th percentile speed. When PSL−S85 is compared to crash rate, the low point is below the zero point, which adds challenges to interpreting the findings. Therefore, the Abs(PSL−Avg) may be an easier speed measure (in terms of interpretation) to identify variables that are affecting safety on city streets. Figure 15. Comparison of crash rate to the difference between posted speed limit and average speed.

NCHRP Web-Only Document 291: Development of a Posted Speed Limit Setting Procedure and Tool 93 Results and Discussion The estimated model coefficients for KABC crashes from the Round 3 path analysis are presented in Table 37. The table shows that the mediator variable Abs(PSL−Avg) was statistically significant at α=0.1, having a positive association with KABC crashes (i.e., as the values of Abs(PSL−Avg) increased, KABC crash frequency increased). Also, residential area and the presence of a raised median were associated with lower crash frequency. The number of signalized intersections, traffic volumes, and segment length were found to be positively correlated with crash frequency. As expected, higher PSL was associated with higher Abs(PSL−Avg), which implies that PSL has indirect effects on KABC crashes through its effect on Abs(PSL−Avg). Note that the presence of a bike lane, the presence of a school zone, and RU_F_rev=Local also had indirect effects on KABC crashes since they had statistically significant positive associations with Abs(PSL−Avg). ROADSURF had a statistically significant negative association with Abs(PSL−Avg), which also had an indirect effect on KABC crashes, subsequently. The relationship of the variables with either KABC_NID or Abs(PSL−Avg) is shown in Figure 16. Note that the standard errors of model coefficient estimates are provided in parentheses. Table 37. Estimated regression coefficients for KABC_NID crashes by path analysis with mediator variable Abs(PSL−Avg) based on 649 sites. Outcome Variable Independent Variable Mediator Coefficient Abs(PSL−Avg)  KABC_NID crashes Intercept 0 −3.648 Mediator Mediator 0.044 Dvelop2 Develop2 −0.483 Median2 Median2 −0.618 NSigInt NSigInt 0.542 LnVol LnVol 0.599 LnLen LnLen 0.852 Speed Intercept α0 −0.471 PSL αPSL 0.327 BIKE1YES αBIKE1YES 1.123 DEVELOP2 αDevelop2 −0.870 ROADSURF αROADSURF −0.044 SCZN1YES αPedAuto 1.497 RU_F_rev αRU_F_rev 1.189 LNVOL αLNVOL −0.078 LNLEN αLNLEN −3.367 Model fit BIC 5568.8 Notes: Cells are highlighted in light gray when the p-value is between 0.05 and 0.1. Cells are highlighted in dark gray when the p-value is less than 0.05.

NCHRP Web-Only Document 291: Development of a Posted Speed Limit Setting Procedure and Tool 94 Figure 16. Path analysis findings for segments with posted speed limits of 20 to 45 mph. FINDINGS’ IMPACT ON SLS-TOOL The findings from this effort support the following decision rules for the SLS-Tool:  Inclusion of the following variables: o Number of signals or signal density. o On-street parking.  Addition of the following variable: o Median type (raised medians were associated with fewer KABC crashes compared to no median or TWLTL, and TWLTLs were associated with more KABC crashes compared to no median). The findings from the path analysis support the consideration of the 50th percentile speed in identifying a suggested speed limit.

Next: APPENDIX E. RELATIONSHIP AMONG URBAN/SUBURBAN ROADWAY CHARACTERISTICS, POSTED SPEED LIMIT, AND CRASHES IN WASHTENAW COUNTY, MICHIGAN »
Development of a Posted Speed Limit Setting Procedure and Tool Get This Book
×
MyNAP members save 10% online.
Login or Register to save!
Download Free PDF

Several types of speed limits exist, including statutory speed limit, posted speed limit, school zone speed limit, work zone speed limit, variable speed limit, and advisory speed.

The TRB National Cooperative Highway Research Program's NCHRP Web-Only Document 291: Development of a Posted Speed Limit Setting Procedure and Tool documents the research efforts and findings from an NCHRP Project 17-76 to identify factors that influence a driver’s operating speed and the development of a Speed Limit Setting Procedure and Tool.

The document is supplemental to NCHRP Research Report 966: Posted Speed Limit Setting Procedure and Tool: User Guide.

  1. ×

    Welcome to OpenBook!

    You're looking at OpenBook, NAP.edu's online reading room since 1999. Based on feedback from you, our users, we've made some improvements that make it easier than ever to read thousands of publications on our website.

    Do you want to take a quick tour of the OpenBook's features?

    No Thanks Take a Tour »
  2. ×

    Show this book's table of contents, where you can jump to any chapter by name.

    « Back Next »
  3. ×

    ...or use these buttons to go back to the previous chapter or skip to the next one.

    « Back Next »
  4. ×

    Jump up to the previous page or down to the next one. Also, you can type in a page number and press Enter to go directly to that page in the book.

    « Back Next »
  5. ×

    To search the entire text of this book, type in your search term here and press Enter.

    « Back Next »
  6. ×

    Share a link to this book page on your preferred social network or via email.

    « Back Next »
  7. ×

    View our suggested citation for this chapter.

    « Back Next »
  8. ×

    Ready to take your reading offline? Click here to buy this book in print or download it as a free PDF, if available.

    « Back Next »
Stay Connected!