|
Intersection Safety Implementation Plan Process
Step 4: Analyze Crash and Applicable Roadway DataThe Data Analysis ProcessThe intersection safety implementation plan process is data-driven. The primary source of data for intersection crash analysis is a State's crash data system. The data used in the analysis helps identify candidate intersections where countermeasures can be considered for cost-effective implementation. The most recent 5 years of crash data is recommended for use in the analysis. More years may be used if the data is available in the crash data system and factors that can change exposure (e.g., significant land use changes, traffic volume changes) have not occurred over the crash data period. Three years of data, while acceptable for identifying high-crash locations, is considered too unstable for identifying intersections with lower repetitive crash histories to be considered for systematic deployment of low-cost countermeasures. The five levels in the data analysis process are described below. Separate Intersection Crashes into Sub-GroupsIntersection crashes should be divided into sub-groups based on State or local ownership, urban or rural location, and traffic control type. This results in eight sub-groups for analysis:
The division of crashes into State and local ownership is helpful since the processes for implementing similar improvements on the State verses the local system are significantly different. Crashes are separated by rural and urban areas because similar types of intersection crashes are more severe in rural areas (e.g., Table 6 shows that typical values for fatalities per 100 crashes for angle crashes are 3.53 and 0.88 for State rural and State urban intersections, respectively). This is important if the State's goal is to reduce fatalities and/or incapacitating injuries instead of crashes. Finally, the types of countermeasure will be different at stop-controlled intersections compared to signalized intersections for similar crash patterns. In addition, the severity of similar crashes is greater at stop-controlled intersections than at signalized intersections. Determine Target Crash TypesBased on the countermeasures selected in Step 3, States should identify the crash types and characteristics which the countermeasures are designed to impact. The typical types of crashes and associated countermeasures are shown in Table 19. This list provides the basis for analyzing the crash data. The combinations of crash types and sub-groups represent all of the various cuts of data that can be used in subsequent levels of the data analysis process.
There are a number of special case or supplementary countermeasures that do not appear in Table 19 because they will be deployed only if an intersection warrants them. This determination cannot be made from the data; it requires field evaluation. These special case/supplementary countermeasures include:
Calculate Average Crash Costs and Crash SeveritiesUsing the countermeasures selected in Step 3 and the related information in Table 19, States should calculate the average crash costs and severity of crashes for each crash type/sub-group (i.e., State and local ownership, rural and urban area, and traffic control type) combination. The formula for average crash costs uses the cost data in Table 4 and the number of injury types for each crash type/sub-group combination:
Severity usually is measured in terms of fatalities per 100 crashes and incapacitating injuries per 100 crashes using 5 years of data. Thus, if implementation of a given countermeasure is expected to prevent 200 crashes in a specified subgroup with a severity of 1 fatality per 100 crashes, it can be expected that 2 fatalities can be prevented through the implementation of the countermeasures at the identified number of intersections.1 Table 20 shows an example of a severity and average crash cost report for angle crashes at signalized intersections.
Determine Distribution of Crash DensitiesUsing the same list of crash type/sub-group combinations as in the average crash cost and crash severity analysis, States need to determine the distribution of crash densities across intersections. The first step in this process creates a standard location definition for each intersection. On State roads, this definition typically is a combination of county, route, and milepost. For local roads, county, city, and the two intersecting road names are often used. The second step is to group crashes by intersection, so that the number of crashes per intersection can be obtained. Table 21 shows a sample, partial listing of crashes per intersection. It is noted that there are over 10,000 intersections included in the complete listing of intersections associated with this sample.
The final step is to create a summarized frequency distribution of intersections based on their number of crashes. Table 22 shows an example distribution for State, rural, stop-controlled intersections. This data corresponds with that shown in Table 21. This example shows that almost 5 percent2 of the intersections (i.e., those with 10 or more crashes) account for approximately 30 percent3 of all the crashes at State, rural, stop-controlled intersections over 5 years.
It is to be noted that only those intersections with at least one crash within the overall crash data period are listed. Therefore, there are more actual intersections in existence than those listed from the crash data. Prepare Data Analysis PackageStates should prepare a data analysis package to develop the straw man outline including a set of countermeasures, deployment level, costs, and estimated statewide annual crash reductions (Step 5), to provide relevant intersection crash information during the workshop (Step 6), and to develop the draft intersection safety implementation plan (Step 7). The package should include at least the following information:
An example data analysis package and straw man outline can be found on the FHWA Intersection Safety web page (http://safety.fhwa.dot.gov/intersection/). Data Problems and SolutionsA State may encounter at least four types of problems while analyzing its crash data to apply a systematic approach: data quality, data availability, exposure and rates, and intersections with multiple countermeasures. Approaches to addressing these problems have been found as discussed below. Data QualityInconsistency of the Rural/Urban Designation at the Same Intersection. The rural/urban differential is used to define the probable severity of similar crashes. On the whole, crashes in rural areas are much more severe than similar types of crashes in urban areas. The rural/urban designation for a crash can come from two sources: directly from the police crash report or transferred from the State's roadway data file. If the information comes from the roadway data file, it is consistent for all crashes that occur at a specific intersection. If the information comes from the police crash report, the rural/urban designation may differ among crash reports for the same intersection. This becomes a problem when the data are grouped according to rural/urban designation. If an intersection has 20 crashes total, but the police crash reports show 13 of these in urban areas and 7 in rural areas, the intersection will appear on both the rural intersection reports and the urban intersection reports (i.e., an urban intersection with 13 crashes and a rural intersection with 7 crashes). The preferred solution is to link the crash and roadway data files and use the rural/urban designation from the roadway data file for each crash. If the rural/urban information is not available from the roadway file or cannot be transferred, then the State should identify the level of data inconsistency to determine if an informed estimate of the correct urban/rural designation can be made. This can be accomplished by tabulating the distribution of crashes at high-crash intersections above the threshold level by urban and rural areas to determine if one type of area is predominant. Table 23 provides an example tabulation.
In this case, 8 of the 10 intersections are most likely rural. Two of the 10 intersections (Intersections C and D) are too close to call. The urban/rural designation for crashes at Intersections C and D should be determined by consulting designation information from the roadway data file and correcting the urban/rural data in the crash files for the appropriate crashes. Police-reported speed limit information may be used as an alternate source of crash severity differences, as it often has more report consistency than police-reported rural/urban designation. A speed limit at and above 45 mph provides a breakpoint where the speed limit data for the same intersection is consistent. In addition, similar crashes at intersections where the speed limit is 45 mph or greater have been found to have significantly higher fatality rates (e.g., fatalities per 100 crashes) than similar crashes that occur at intersections where the speed limit is 40 mph or lower. States that do not have accurate and consistent urban/rural crash data elements can use speed limits as an indicator of intersections with higher approach speeds and more severe crashes similar to those experienced in rural areas. Inconsistency of the Traffic Control Device Information at the Same Intersection. Most States do not have a computerized traffic control device inventory. Information on the type of traffic control at the intersection must come from police-reported traffic control device information on the crash report. Almost all intersections are controlled by a Stop sign or a traffic signal. However, the type of traffic control device reported on the police report can vary widely and include warning signs, pavement markings, or no traffic control devices at all. The traffic control device information is critical to the implementation plan. It relates to the probable severity of future crashes and the type of countermeasure to apply to reduce future crashes. The traffic control device consistency problem is similar to the rural/urban designation problem. Like the rural/urban designation problem, when the data is grouped by traffic control device, an intersection could appear on more than one data report. For example, an intersection with 20 crashes and inconsistent and/or incorrect identification of traffic control devices could appear on a stop-controlled report (assuming 10 crashes were identified as stop-controlled), a signalized report (assuming 4 crashes were identified as signalized, and an unknown report (assuming 6 crashes were identified as no, other, or unknown traffic control device). If a State is developing a computerized traffic control device inventory, or at least a traffic signal inventory (i.e., all non-signalized intersections can be assumed to be stop-controlled), the State should complete the development of that inventory prior to the data analysis described here and use that inventory to determine if an intersection is signalized or stop-controlled. If a traffic control device inventory is not near completion or is not readily available, the overall accuracy of the police-reported traffic control device information should determine the course of action. The thresholds for crashes at stop-controlled intersections considered for countermeasure improvement are generally 5 for rural intersections and 10 for urban intersections. If about 70 percent or more of total intersection crash reports have a correctly-identified traffic control device, then those intersections with a number of crashes equaling or exceeding the thresholds noted above (which are the minimum used) should have sufficient numbers of correctly identified traffic control devices to predict the type of traffic control device for all crashes occurring at those intersections.4 A State can proceed toward implementation plan development using the predicted traffic control device values. If the percentage of correctly-identified traffic control devices is not sufficient to predict the type of traffic control devices at intersections with crashes above the threshold levels, then, States should use a secondary source to determine traffic control devices at these intersections, preferably prior to the development of the straw man outline (Step 5). Video logs, photo logs, or field reviews can be used to determine or verify the type of traffic control device. Inconsistency of Lit/Unlit Information at Intersections with Night Crashes. The data from police crash reports on the time period of the crash (e.g., day, dusk, dawn, night) is generally very good. However, in most States the data on night crashes identifying whether the intersection is lit or unlit have a significant amount of variability for crashes at the same intersection. Compounding the problem are intersections that are inadequately lit compared to current standards. To address this problem, States can identify intersections with a high frequency and proportion (e.g., night/total crashes) of night crashes. Table 24 shows an example of the distribution of night crashes by intersection in rural areas.
States should identify the statewide mean proportion of night crashes to total crashes for both rural and urban areas. For example, with a statewide mean of 18 percent of night to total crashes in rural areas, those intersections that have both a high frequency and proportion of night crashes substantially above the mean (i.e., in this case, 25 percent or more night/total ratio) have been identified in Table 24 and should be considered for some type of lighting enhancement. A field review of these intersections is necessary to determine if lighting exists and if so, to what degree. Intersection Information for Crashes at Locally-Owned Intersections. Crash reports for crashes that occur at locally-owned intersections usually have reliable information on the county and municipality in which the crash occurred. However, information on the intersecting streets can be characterized in different ways depending on the reporting officer. For example, Fifth Street could be identified as Fifth St., Fifth Str, 5th Street, or 5th St, among other variations. A standard term of "Fifth St" would address this issue. Unless the information is standardized, crashes that occur at the same intersection may be spread over many intersection identifiers when an analyst tries to group crashes by intersection. This can create significant problems when trying to identify intersections with crash levels above a threshold. To address this issue, States could establish standard nomenclature to consolidate some of the crashes for local intersections. Ideally States should do this at the time data is entered into the crash data system. However if standard nomenclature is not pre-established, it can be selectively applied to those municipalities in which intersection safety initiatives are being considered during development of the intersection safety implementation plan, thus reducing the level of effort. A sample listing of common terms, currently being used by Arizona, includes:
Data AvailabilityInsufficient Information to Determine if an Approach Pavement Has Both an Inordinate Number and Proportion of Wet Pavement Crashes and a Slippery Surface. The data from police crash reports on pavement surface conditions (e.g., wet, dry, icy, snow covered) is usually very reliable. However, the physical attributes of the pavement that may be contributing to the inordinate number and proportion of wet pavement crashes often are not known. To address this problem, States can identify intersections with a high frequency and proportion (e.g., wet pavement/total crashes) of wet pavement crashes. Table 25 shows an example of the distribution of wet pavement crashes by intersection.
States should determine the statewide mean proportion of wet to total crashes for rural and urban intersections where the speed limit is at or above 45mph. For example, if the statewide mean for rural intersections with speed limits of 45mph or above is 16 percent, those intersections that have both a high frequency and proportion of wet pavement crashes substantially above the mean (i.e., in this case, 25 percent or more wet pavement/total ratio) have been identified in Table 25 and should be considered for some type of pavement surface improvement. The State should conduct a skid test of the approach to determine if the pavement has a low coefficient of friction. Then the State should conduct a field review of the intersection and a review of the pavement history to determine if other surface factors such as significant rutting (i.e., greater than 2 inches) exist in the wheel paths which could contribute to hydroplaning. Exposure and RatesExposure for Intersections is Different than Highway Segments. Exposure at intersections is measured in terms of the number of entering vehicles from all of the intersection legs rather than VMT. A few States have extracted volume information from their roadway data file and developed entering vehicle numbers for each of the completely State-owned intersections in the State. This information can be used to establish rates of crashes per million entering vehicles. However, most States only pull the mainline annual average daily traffic (AADT) from the roadway data file and attach it to each specific crash in the crash data file. As a result, an intersection with several crashes over a 5-year period will have different levels of AADT if the AADT is updated over the crash history period. Without complex and time-consuming programming, it can be difficult to consolidate these differences at the same intersection into single values for computing rates. Fortunately, the use of rates is not as critical in the systematic application of cost-effective, low-cost countermeasures compared to the traditional approach. There are a number of approaches to refine the number of intersections that should be considered for systematic improvement considering exposure differences. The two key questions that need to be addressed are:
If the number of through lanes or the functional classification for the mainline route in the roadway data file has been linked to the crash data file, one of these pieces of information can be used to establish different threshold levels, either based on the number of through lanes or the mainline functional classifications. For example, a higher crash threshold may be established for stop-controlled intersections with three through approach lanes as opposed to an intersection with a single through approach lane, since the volumes and exposure on the three through approach lane intersection are much greater. Once threshold levels are established, assuming that the mainline AADT for each crash is listed in the output, the mainline AADT for intersections slightly below, at, and slightly above the crash threshold level can be scanned to determine which intersections to consider for improvement. Any very low AADT intersections slightly below the threshold may be added to the list of intersections being considered for improvement. Any very high AADT intersections either at or slightly above the threshold could be removed from improvement consideration. Intersections with Multiple CountermeasuresDeveloping the Intersection Safety Implementation Plan Focuses on Identifying Separate Countermeasure Deployments at Intersections. As a result, intersections often will appear on more than one list of countermeasures. For example, a rural stop-controlled intersection may be above the crash thresholds for the basic set of sign and marking improvements, new or upgraded lighting, and skid resistance surfaces. Grouping countermeasures together for the same intersection is important since that it can reduce the number of multiple field reviews at one location. States can create a set of matrices for each of the eight possible intersection sub-groups (i.e., State/local, rural/urban, stop-controlled/signalized) to identify intersections with multiple countermeasures. Table 26 is a sample matrix. This table shows the number of crashes above the given threshold for a specific countermeasure by intersection. It is created by combining all of the distributions of crashes by intersection, using only those intersections where the number of crashes exceeds the threshold for that given countermeasure.
1 The expected number of fatalities prevented is the expected number of crashes prevented (200) multiplied by the fatalities per 100 crashes (1), or 200 x (1/100) = 200 x 0.01 = 2. 2 Almost 5 percent is calculated by dividing the number of intersections with 10 or more crashes (513) by the total number of intersections (10,653), or 513/10,653 = 0.482. 3 Approximately 30 percent is calculated by dividing the number of crashes at intersections with 10 or more crashes (8,601) by the total number of crashes (30,232), or 8,601/30,232 = 0.2845. 4 By assessing the distribution of traffic control devices for crashes at the same intersection using the same process as described for the rural/urban designation problem. For example, if there were six crashes at a rural intersection with three crashes indicating a stop-controlled traffic control device, one indicating a signalized traffic control device, and two indicating no traffic control device, it can be reasonably assumed that the intersection is stop-controlled.
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
