U.S. Department of Transportation
Federal Highway Administration
1200 New Jersey Avenue, SE
Washington, DC 20590

Skip to content
Facebook iconYouTube iconTwitter iconFlickr iconLinkedInInstagram

Federal Highway Administration Research and Technology
Coordinating, Developing, and Delivering Highway Transportation Innovations

This report is an archived publication and may contain dated technical, contact, and link information
Publication Number: FHWA-RD-99-207

Prediction of the Expected Safety Performance of Rural Two-Lane Highways


This section presents the conclusions concerning the accident prediction algorithm developed in this report and presents recommendations for possible future enhancements of the algorithm.


The primary conclusion of this report is that an accident prediction algorithm has been developed and that this algorithm appears to be a useful tool for predicting the safety performance of rural two-lane highways. The primary strengths of the algorithm are as follows:

  • The algorithm makes quantitative estimates of accident frequency and of the accident severity and accident type distributions for any two-lane highway section or project.
  • The algorithm has been developed with a modular structure that combines base models and AMFs. The base models serve as scale factors to assure that the magnitude of the predicted accident frequency is appropriate, while the AMFs assure that the predicted accident frequency is sensitive to site-specific geometric and traffic control features.
  • The use of AMFs that are separate from the base models assures that the effects of individual geometric design and traffic control features are not dependent upon inappropriate regression coefficients that are too large, too small, or in the wrong direction. Each AMF has been developed by a panel of experts to represent the best information currently available on the safety effects of that particular geometric design or traffic control feature.
  • The modular structure makes the algorithm easy to update as better information, including new research results, become available.
  • A calibration procedure is provided to allow individual highway agencies to adapt the algorithm to the safety conditions present on their rural two-lane highway system. The calibration procedure allows IHSDM users to adjust the predicted accident frequencies for agency-to-agency and State-to-State differences in factors such as accident reporting thresholds, accident reporting practices, animal populations, driver populations, and climates.
  • A procedure based on the EB method allows users to combine accident predictions obtained from the algorithm with observed site-specific accident history data.

The major weakness of the algorithm is that it incorporates the effects on safety of most, but not all, geometric and traffic control features of interest to highway agencies. The algorithm incorporates only those features whose effects were considered by a panel of experts to be well established in quantitative terms. Geometric and traffic control features that are poorly understood, or not understood at all, have necessarily been omitted. The model generally treats the effects of individual geometric design and traffic control features as independent of one another and ignores potential interactions between them. It is likely that such interactions exist and, ideally, they should be accounted for in the accident prediction algorithm. However, such interactions are poorly understood and none could be quantified by the expert panels that participated in the development of the algorithm. It is the assessment of the expert panels that the base models and AMFs presented in this report represent the current state of knowledge about safety on rural two-lane highways and cannot be improved without further research. The next section of report discusses potential areas to which future research might be directed to improve the model.

FHWA plans to incorporate the accident prediction algorithm for rural two-lane highways presented in this report in software for implementation as part of the IHSDM. A stand-alone version of the software may also be available for use independent of a CAD system.

Future Enhancement of the Accident Prediction Algorithm

It is recommended that future enhancements be made to the accident prediction algorithm as further research is completed and that forthcoming research on rural two-lane highways be structured so that results are obtained in a form that can be directly implemented in the accident prediction algorithm. It is also recommended that a program of additional research be undertaken with the specific goal of filling gaps in the accident prediction algorithm and expanding its scope. Specific areas for future enhancement of the accident prediction algorithm are discussed below.

Base Models

The base models for roadway sections and for three- and four-leg STOP-controlled intersections appear to be well established and there is no immediate need for work to improve them. By contrast, the base model for four-leg signalized intersections is based on a small sample size (only 49 intersections split between two different States). The resulting models were not as satisfactory as desired, and it was difficult to choose among the available candidates. Signalized intersections are relatively rare on rural two-lane highways, so the limitations of the base model for signalized intersections do not overly limit the utility of the algorithm, but it would be desirable to assemble a larger database on signalized intersections on rural two-lane highways for the purpose of developing an improved base model. It would also be desirable to develop a base model for three-leg signalized intersections on rural two-lane highways, which are outside the scope of the current accident prediction algorithm.

A minor drawback of the base model for roadway segments is that it was based on sites with a mix of shoulder types, but the shoulder type itself was not a statistically significant factor in the model. This is understandable given the small effect of shoulder type on safety based on the expert panel’s assessment shown in table 3. However, for consistency, if the roadway segment base model should be updated in the future, it would be desirable to (1) include an effect of shoulder type in the base model if found to be statistically significant; or (2) omit the shoulder type factor from the model and develop the model solely with data for sites with paved shoulders.

Accident Modification Factors

The accident prediction algorithm omits AMFs for several geometric design and traffic control features which the accident prediction algorithm should desirable address. Other AMFs that are included in the model could be improved through further research. These improvement needs are described below:

Roadway Segments

One of the greatest limitations of the roadway segment algorithm is the lack of an AMFs for bridge width. The expert panel on roadway segments strongly desired to include a bridge width factor in the algorithm because narrow bridges are known to be associated with accident concentrations on two-lane highways. However, the panel found that even the best study of the relationship between bridge width and safety had a major flaw that limited its use in the algorithm. The flaw was that the study included only bridges that had experienced one or more accidents during the study period.(41) Omission of sites which have experienced no accidents is a known source of bias in accident research. It is recommended that a well designed study of the relationship between bridge width and accidents on rural two-lane highways be undertaken and that its results, if found to be satisfactory, be incorporated in the accident prediction algorithm.

The roadway segment algorithm lacks an AMF to account for the effect on safety of vertical curve design and stopping sight distance. This effect has never been satisfactorily quantified, and recent research suggests that the safety effect of limited stopping sight distance at a crest vertical curve is relatively small.(42) However, should this effect—even if small—be reliably quantified in the future, it would be desirable to include it in the accident prediction algorithm.

It would be desirable to improve the representation of driveway effects in the accident prediction algorithm for roadway effects. The algorithm currently bases driveway effects on the driveway density (driveways per mi). The AMF for driveway density, based on a regression equation, indicates that there is greater sensitivity of safety to driveway density at lower ADT than at higher ADT. This appears to be the opposite of what might be expected. Furthermore, it would be more desirable to develop a method to quantify the safety effect of each individual driveway, but such a method does not currently exist. Such an approach would also require the user to supply detailed data on individual driveway locations and types (e.g., commercial vs. residential) and driveway traffic volumes which are not necessarily available to all users.

The effects of passing lanes and shoulder width are treated as independent in the accident prediction algorithm, but there may in fact be an interaction between them. In the research that established the safety effects of passing lanes, some of the sites at which passing lanes were installed may also have had full shoulders provided as part of the same project. Other sites may have had a portion of the shoulder converted to the passing lane. If further research quantifies the separate effects of passing lanes and shoulder widths, or the interactions between them, it would be desirable to incorporate these effects in the accident prediction algorithm.

The effect of center two-way left-turn lanes currently included in the accident prediction algorithm is clearly an oversimplification of a much more complex effect. Many evaluations of two-way left-turn lanes have been conducted, but little of this work is specific to rural two-lane highways. Further research to improve the AMF for two-way left-turn lanes would be desirable.

The roadside design AMF incorporated in the model is based on a qualitative roadside hazard rating system (a subjective 1 to 7 rating scale), rather than addressing the explicit effects of specific roadside design features. It would be desirable to provide the capability for individual roadside design features to be evaluated explicitly in situations where detailed data on roadside design features are available. Research is currently underway as part of the NCHRP program which may lead to such a capability.(22)


Two undesirable omissions in the accident prediction algorithm for at-grade intersections are the lack of effects for roadside design and driveways in the vicinity of an intersection. The accident prediction algorithm for roadway segments includes the effect of both roadside design and driveways over the entire length of a study section or project, but the intersection algorithm places no special weight on restrictive roadside design at an intersection or driveways located near an intersection. The omission of roadside design issues at intersections may be a minor limitation, because run-off-the-road accidents are generally understood to be a roadway segment, rather than an intersection, problem. For example, table 2 shows that the proportion of intersection-related single-vehicle run-off-the-road accidents is small. However, poor roadside design in the vicinity of an intersection could increase the severity of multiple-vehicle accidents in which one or more of the involved vehicles leaves the traveled way. The lack of an effect for driveways near intersections is an omission of greater concern, although it is not as serious an omission on rural two-lane highways as it would be on urban or suburban arterials. Driveways near intersections are known to be a safety-related access management concern, but the expert panel found that the safety effects of driveways near intersections have not been well quantified. Research on this issue should be encouraged and should address, at a minimum, the type of driveway (e.g., commercial vs. residential), the distance from the intersection to the driveway, and the presence or absence of access control measures that restrict turning maneuvers (e.g., medians or turn prohibitions).

It would be desirable to improve the AMFs for left- and right-turn lanes at intersections. The expert panel found substantial past research on the effects of left- and right-turn lanes, but no definitive results. The AMFs included in the accident prediction algorithm relied heavily upon expert judgment in interpreting the available results and additional research to improve these AMFs would be desirable. FHWA has such research underway, and this research may lead to an update of the accident prediction algorithm.

A final concern with the accident prediction algorithm for intersections is that it does not address before-and-after evaluations of improvement projects in which an existing STOP-controlled intersection is signalized. It would be desirable to develop a specific AMF for such projects based on well-designed before-and-after studies.

Accident Prediction Algorithms for Other Facility Types

The accident prediction algorithm presented in this report applies only to rural two-lane highways. Rural two-lane highways were selected for the initial development of the IHSDM because existing rural two-lane highways have a wide range of variation in the quality of their design features and, therefore, present substantial opportunities for improvement of safety. Most freeways, by contrast, have been built much more recently and are more consistent in their design features.

It would be desirable to expand the accident prediction algorithm to other facility types, following the general approach developed for the accident prediction algorithm for rural two-lane highways presented in this report. It is recommended that the scope of the accident prediction algorithm be expanded in the following priority order: rural multilane highways, urban and suburban arterials, and, finally, freeways.

Previous | Table of Contents | Next

Federal Highway Administration | 1200 New Jersey Avenue, SE | Washington, DC 20590 | 202-366-4000
Turner-Fairbank Highway Research Center | 6300 Georgetown Pike | McLean, VA | 22101