Click for next page ( 6


The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 5
5 that were developed in other states. The extent to which account for these effects or, at a minimum, recognize their AMFs are valid when transferred to places beyond the devel- existence. opment domain (e.g., from one state to another) that have Lack of Effectiveness Information. AMFs have not been different roadway, traffic, weather, driver and other relevant developed for many ITS improvements and other opera- characteristics, as well as different accident investigation tional strategies. For example, on many freeways, safety serv- practices, is unknown. ice patrols have become more common as a way of reducing Methodological Issues. Many existing AMFs are derived the impact of incidents and reducing secondary accidents. from before-after analysis of actual countermeasure imple- However, no AMFs exist for this countermeasure. Other ITS mentation. Indeed, such before-after analysis, as opposed to countermeasures of high interest for which no reliable cross-sectional/regression-type analysis, produces the best AMFs exist include dynamic or changeable message signs AMF estimates, but only if conducted properly. Unfortu- (including those related to variable speed limits), real-time nately, many current studies reflect changes in crash experi- warning systems (e.g., severe alignment or adverse weather), ence resulting from improvements at sites that experienced and pedestrian safety treatments (e.g., in-pavement cross- an unusually high number of crashes in the before-treatment walk lighting and countdown signals). period. The selection bias that results from this approach can Combinations of Improvements. Most AMFs are designed yield significantly exaggerated AMF estimates due to the for individual improvements. However, multiple improve- phenomenon of regression to the mean. Other methodolog- ments are typically made when a facility is being rebuilt. States ical problems that are often found include the following: use different formulas for combining individual AMFs when Failure to properly separate out the safety effects of other considering multiple treatments. However, there is very little changes (e.g., traffic volumes, the impacts of other treat- sound research on the multitude of actual combinations of ments implemented at the same time, crash reporting treatments that exist in practice. Thus, it is unknown whether differences between jurisdictions or across time, and current predictions based on combining individual AMFs underlying crash trends across time). accurately capture the true combined effect. Sample sizes that are too small. Large numbers of sites Publication/Citation Issues. A less-cited issue that is preva- with the same combination of applied countermeasures lent in much of the research is related to the quality of the are needed for a valid analysis. For some treatments and material that is available and often used in the development the subsequent type of crash reduction expected, hun- of AMFs. Specific problems include the following: dreds or thousands of locations may be necessary, along Publication bias--the tendency to only publish studies with many years of crash data. (Pedestrian treatments that produced favorable results for the treatment being and crashes are a good example of this problem.) evaluated. Use of comparison groups that are unsuitable for a Selective citing of results--the tendency to ignore the variety of reasons. negative aspects of results such as declining effects over Incorrect interpretation of accuracy of estimates or time or unintended consequences that would lead to in- presentation of results without statements of accuracy. creases in some crash types. In some cases, a sponsoring Variability. The value of an AMF may depend on a vari- organization may not want negative results published ety of factors, such as traffic volumes, crash experience, because it invested significant funds in a countermea- and site characteristics. Thus, research that results in a sin- sure/intervention program. gle AMF value may be of limited applicability. Accident modification functions rather than factors may be more In summary, although several AMFs already exist, there are appropriate. Several of the AMFs presented in this report a number of issues related to the quality of those factors now are indeed functions. being used in many states and local jurisdictions. There is a Crash Migration and Spillover Effects. It is possible that need to improve the quality of existing AMFs and to develop countermeasures implemented in a particular location may additional AMFs where there are currently voids. be followed by migration of crashes to adjacent locations. For example, the conversion of two-way stop control to all- Project Objective and Overview way stop control at an intersection may lead to an increase in crashes at surrounding intersections that continue to The objective of this project was to develop reliable AMFs operate as two-way stop control due to driver confusion. for traffic engineering and ITS improvements. Reliable Likewise, the prohibition of left turns at an intersection may AMFs, at a minimum, must meet the following criteria: lead to an increase in left-turn crashes at upstream and downstream intersections. Existing AMFs rarely account The AMFs are methodologically and statistically valid. for this phenomenon. For AMFs to be useful, they have to Separate values for AMFs are defined for various influencing

OCR for page 5
6 factors such as the highway facility, operating condition, Task 5--Development of interim report and work plan, weather, time of day, percentage of truck traffic, and pre- and existing crash history as appropriate. (Alternatively, a Task 6--Project briefing for oversight committee. method could be developed for adjusting the AMFs for these influencing factors.) Should expert judgment be used in Phase II included the following tasks: developing AMFs, it must be analysis-driven. The applicability of the AMF is known and documented. Task 7--Execution of work plan to collect necessary data For example, some AMFs may denote an impact on and develop/improve AMFs and crashes only at a specific location whereas other AMFs may Task 8--Preparation of final report. affect crashes for an entire stretch of roadway. Some AMFs may apply only to specific accident types or to specific pre- Thus, Phase I of the effort was focused first on the extrac- existing conditions (e.g., high percentages of wet weather tion of information on existing AMFs through a critical review crashes). Some may be applicable only to the state or area of research literature. The documentation included detailed where the AMF is developed, and thus adjustment factors descriptors of the AMFs, the conditions (e.g., roadway types must be developed to allow application of the AMF to and locations) for which each AMF was applicable, and a other regions. judgment of the level of predictive certainty (i.e., the quality) The AMFs reflect improvements or combinations of associated with a given AMF. The remainder of Phase I improvements that are of interest to DOTs. The survey focused on a series of steps to determine high-priority AMF conducted as part of this study requested respondents to needs--which existing AMFs should be improved and which indicate which treatments were considered priorities for new AMFs should be developed within the project budget. AMF development. This prioritization was based on inputs concerning "most The AMFs reflect the impact of the improvement on differ- important safety treatments" from a survey of state DOTs ent crash categories. Crash categories might include total combined with other factors concerning the quality of the crashes, severe-injury crashes, property-damage-only existing AMF, the size of the crash problem affected by the crashes, and specific crash types (such as rear-end and angle). treatment, and the availability of data needed to develop or The AMFs reflect variability. The best estimate of the improve an AMF. All this information was then used to AMFs, along with some technique that reflects their vari- develop a work plan for the AMF development/upgrade effort. ability (such as ranges, confidence intervals, standard After approval by the oversight panel, the plan was exe- deviation, or some other technique) should be presented. cuted in Phase II of the effort. This execution involved four The AMFs reflect the savings in "total harm" provided by basic types of analyses: the treatments. Many treatments affect both crash fre- quency and crash severity, some affect just crash severity, Empirical Bayes (EB) Before-After Evaluation. Original and some decrease crashes at one level of severity and data were acquired to conduct an analysis and determine increase crashes at another level of severity (e.g., traffic sig- the crash-harm effects of a specific high-priority treatment nalization can decrease more-severe angle crashes but that had been implemented within a state or states (usually increase less-severe rear-end crashes). AMFs must capture at multiple sites) in the late 1990s or early 2000s in order to changes in crash severity as well as changes in crash fre- have a sufficient after-treatment period. The analysis made quency in order to measure "harm savings." use of the latest statistical methodologies for before-after The identification and development of AMFs that meet studies and produced AMFs with the highest feasible level most of the above requirements involved a project effort with of confidence. This analysis required not only detailed the following eight tasks divided among two phases. Phase I descriptions of the historic treated sites (e.g., treatment included the following tasks: specifics, locations, and dates of installation) and good before-treatment and after-treatment crash, inventory, Task 1--Review of completed and ongoing studies to doc- and traffic data on the treated sites, but also comparable ument existing AMFs, data on a large reference group of somewhat similar sites. Task 2--Survey of states to determine AMF use and priori- Thus, the analysis required either that the project team had ties and availability of data concerning treatment installation easy access to both current and historical crash, inventory, for use in new AMF development, and traffic data, or that the implementing state was willing Task 3--Follow-up interviews with states having poten- and able to provide the linkable data files necessary. Since tially usable treatments and data, FHWA's Highway Safety Information System (HSIS) Task 4--Determination of whether available data from includes historic crash, inventory, and traffic data from states or other sources could be used in AMF development, nine states, and since most non-HSIS states do not store

OCR for page 5
7 historic roadway inventory data, attempts were made to NCHRP Project 17-27, Parts I and II of the Highway Safety find treatments implemented in these HSIS states. Manual, included documentation of the state of current Reanalysis of Existing/Supplemental Data. An existing safety knowledge and preparation of chapters that in- AMF was improved by applying a more rigorous evaluation cluded AMFs. methodology to existing data from a prior study. The pre- NCHRP Project 17-26, Methodology to Predict the Safety ferred methodology was again the EB before-after approach. Performance of Urban and Suburban Arterials, included Again, this required adequate data for both treatment and the development of predictive tools and an HSM chapter non-treatment sites. In many cases, supplemental data for urban and suburban arterials. (e.g., data for the development of a reference group) were NCHRP Project 17-29, Methodology to Predict the Safety acquired to meet this requirement and enhance the analysis. Performance of Rural Multilane Highways, included the Analysis-Driven Expert Panel. An expert panel was con- development of predictive tools and an HSM chapter for vened to review the existing studies concerning a specific rural multilane highways. AMF and then define a consensus AMF based on the stud- NCHRP Project 17-34, Prepare Parts IV and V of the High- ies reviewed. The expert panel included expert researchers way Safety Manual, included the development of chapters (knowledgeable about the AMFs of interest and the for roadway safety management and safety evaluations. strengths and weaknesses of study methods) and a group of NCHRP Project 17-36, Production of the First Edition of expert state and local AMF users (i.e., safety engineers) with the Highway Safety Manual, was preparation of the first knowledge of the specifics of the AMFs needed and the real- edition of the HSM for publication. world conditions under which those evaluated treatments were probably implemented. At times, limited additional The safety knowledge documented in the HSM is being analyses were conducted. The use of an expert panel stated in terms of AMFs, and the HSM safety prediction tools required that the body of literature be robust enough to be include AMFs as a key component. Because the AMFs devel- subject to assimilation/meta-analysis by team members and oped in NCHRP Project 17-25 are so closely related to the then presented to an expert panel to develop a reliable AMF AMFs developed for and documented in the HSM and with at least a medium-high level of confidence. because the safety predictive tools for urban and suburban Cross-Section Modeling. A cross-sectional model was de- arterials and for rural multilane roads both require AMFs, veloped and used for the derivation of an AMF for a specific this project and NCHRP Projects 17-26 and 17-29 were treatment. Some treatments of interest are not installed or closely coordinated by the different teams, with information changed in a manner that allows for a before-after evalua- being shared on a regular basis. The criteria developed in tion. For example, it is unlikely that changes would be made NCHRP Project 17-25 for assessing AMF quality served as the to roadside slopes without making other changes, such as basis for criteria used in NCHRP Project 17-27. However, addition of a shoulder, at the same time. For these types of there were differences in how the final decisions on what con- treatments, the development of road safety models is still an stituted "acceptable AMFs" were made. There are also differ- alternative to determine safety effectiveness. ences in AMFs chosen for publication here and those AMFs that will be in the HSM. While the AMFs published here It is important to note that other AMF projects were going on are only those judged to have high or medium-high levels at the same time as NCHRP Project 17-25. At approximately the of predictive certainty, the HSM will be more inclusive, same time that NCHRP Project 17-25 was initiated, NCHRP publishing AMFs with lower levels of predictive certainty ac- and a TRB task force also began a series of projects aimed at the companied by a rating and warnings concerning use. The re- planned 2008 publication of the first edition of the Highway search team for NCHRP Project 17-25 conducted a detailed Safety Manual (HSM) (see www.highwaysafetymanual.org/) comparative review of the AMFs developed for this project (3). The HSM will be a repository of (1) current knowledge and the AMFs developed for the HSM under NCHRP Project related to roadway safety treatments, (2) tools for use in pre- 17-27. This review led to some changes in the final procedures dicting the safety effects of different roadway design alternatives used in NCHRP Project 17-27 and to a high level of consis- for various classes of roadways, and (3) tools for identifying sites tency between the AMF-related results of the two projects for needing safety improvements and the best treatments for them. the higher-certainty AMFs. In addition, to develop AMFs NCHRP has funded the following projects in support of the for treatments on urban/suburban arterials (NCHRP Project HSM: 17-26) and on rural multilane highways (NCHRP Project 17-29), researchers for NCHRP Project 17-25 organized two NCHRP Project 17-18(04), Highway Safety Manual, was a analysis-driven expert panels jointly with the two project scoping study that included the development of the initial teams, again ensuring both coordination of the efforts and concept, outline, and prototype procedure chapter. consistency in the results.