Background

One of the major tasks of the Federal Highway Administration's (FHWA) Design Concepts Research Division is to develop and implement an Interactive Highway Safety Design Model (IHSDM). The purpose of this model is to provide engineers and planners with a tool to evaluate interactively the safety of alternative highway designs. (See related article.)

The model is being developed at two levels. The level 1 model is for the planning or programming stage of a project. Normally, at this stage, the engineer has only general project information, such as projected traffic volume; type of highway; and number of lanes, intersections, and interchanges. The level 1 model, given limited input data, would provide the estimated number of accidents by severity for each alternative.

The more detailed level 2 model assesses the safety of alternative designs when the plans, specifications, and estimates for a project are being prepared. Thus, the model consists of eight modules or submodelsan accident prediction model, a design consistency review model, a design policy review model, a benefit-cost model, a driver model, a vehicle dynamics model, a traffic model, and a finite element model.

Developing a level 1 accident prediction model requires a large amount of good quality accident and roadway data. Researchers explored the use of the Highway Safety Information System (HSIS) to develop average accident rates for different highway types for use in the level 1 accident prediction model. HSIS is a safety data base that contains accident, roadway inventory, and traffic volume data for a select group of states. This data base is maintained by FHWA in cooperation with the participating states.

This article describes the study to determine if HSIS data could be used to develop the accident prediction model.

Analysis of Data Files

For this analysis, roadways were classified into eight different types; however, the focus of this study was roadway section accidentsthe accidents that occur on the stretch of the roadway between the intersections or interchanges.

Researchers hoped to be able to combine the roadway section accidents and the roadway data of similar roadway types from different states to increase the data size and to determine a stable roadway section accident rate. However, to combine the data of two or more states, several factorssuch as weather conditions, terrain conditions, roadway types, and accident rateshave to be the same for the states. If the accident rates for similar highway types show no statistically significant difference, then the data could be combined.

A preliminary study of the five participating HSIS states showed that only two states had the required accident, geometric, and traffic variables to subdivide the highway into eight roadway types and assign the appropriate accidents to different highway sections. Also, these two states have similar weather and terrain conditions. These states are referred to as state A and state B. Each state has a very different file structure and a lot of manipulation and massaging was required to obtain the accident frequencies along with the appropriate exposure data to calculate the accident rates for roadway sections.

The roadway data was grouped into eight categories based on their geometric configuration, using variable such as access control, rural/urban designation, number of lanes, and presence or absence of median. Great care was taken to use good quality data from both states. To achieve this, unpaved sections and single-lane roads were deleted from the roadway file. These eight categories are:

• Urban freeways.
• Urban, two-lane roads.
• Urban, multilane divided highways.
• Urban, multilane undivided highways.
• Rural freeways.
• Rural, two-lane roads.
• Rural, multilane divided highways.
• Rural, multilane undivided highways.

The accident files for both states were processed to keep only those accidents occurring on the primary road systems (interstates, state routes, and U.S. routes). Three years of accident data were used to develop the final accident rates. These accidents were separated into roadway section accidents, intersection accidents, and interchange accidents. The accidents showing a collision with a train were deleted from the final file. The roadway section accidents were then matched to corresponding roadway records. The results of the above match were roadway orientation files for states A and B with accident variables attached to those sections where the accidents occurred. Thus, the matching process retained all the roadway sections with zero or more accidents per section in the final data set.

Summary of Data

The roadway section accidents, separated into the final eight roadway categories, stratified by severity level for both the states, are shown in tables 1 and 2. These tables also show the roadway mileage for each roadway category and the million vehicle-miles of travel. Table 3 shows the accident rates for four roadway types for states A and B. The accident rates are the number of accidents per million vehicle-miles of travel per year.

Table 1 -- Classification of Roadway Section Accidents By Severity Using HSIS Data for State A (1985-87)

Notes: *Fatal + Incapacitating Injuries, ** Other Injuries, *** Property Damage Only, ****Travel in million vehicle-miles for three-year period.

• Accidents and roadway mileage are only for interstates, state routes, and U.S. routes.
• Unpaved roadway surfaces deleted from roadway file.

In tables 1 and 2, the roadway mileage is relatively small for multilane highways compared to the freeways and two-lane highways. However, in cases where there are adequate miles for multilane highways in one state, the other state has very low miles. Therefore, the data for the two states cannot be compared for that roadway type. Only the data for rural and urban freeways and for rural and urban, two-lane roads could be combined for the two states. Therefore, the remainder of the study and statistical analysis focuses on these four roadway types that are shaded in tables 1 and 2. Table 3 shows accident rates for these four roadway types only.

Table 1 -- Classification of Roadway Section Accidents By Severity Using HSIS Data for State B (1986-88)

Notes: *Fatal + Incapacitating Injuries, ** Other Injuries, *** Property Damage Only, ****Travel in million vehicle-miles for three-year period.

• Accidents and roadway mileage are only for interstates, state routes, and U.S. routes.
• Unpaved roadway surfaces deleted from roadway file.

Figure 1 -- Conversion factors for tables 1 & 2.

To determine the significant difference between roadway section accident rates for states A and B, a statistical test was performed on accident rates for rural and urban freeways and for rural and urban, two-lane roads using binomial probability distribution equations.(1) The results showed that there is no significant difference between accident rates of two states for urban freeways and urban two-lane highways. However, the accident rates are significantly different for rural freeways and rural, two-lane highways. Based on this test, the accident data for urban freeways and two-lane highways could be combined to form one data set.

Table 3 -- Comparison of Roadway Section Accident Rates for States A and B

*Accidents per million vehicle-miles of travel per year.

Results

For state A, the numbers of miles of roadway classified as urban and rural, multilane, undivided highway are very small. Therefore, the accident rates calculated for these types of highways should not be used to estimate the accident potential for future highways of this type.

Similarly, for state B, the numbers of miles of roadway classified as urban, multilane, divided, and undivided; and as rural, multilane, divided, and undivided are also very small. Thus, the accident rates of these highway types should not be used to estimate the accident potential for future highways.

An examination of the accident rates in table 3 for the different highway types for the two states shows agreement for certain highway types and no agreement for other highway types. For example, there are close agreements in the accident rates of the two states for urban freeways and urban, two-lane highways; however, there are statistically significant differences in the accident rates for rural freeways and rural, two-lane highways.

Since the rural roads have more variability in lane widths and traffic volumes, these roads should be further stratified in different volume groups. Then the accident rates should be calculated for each of the volume groups and compared by similar volume groups for both states.

Conclusions

The successful development of accident frequencies for eight roadway types demonstrates that state data from a large data base like HSIS can be used to develop the accident prediction model for different roadway types. However, this requires judicious manipulation of the data and sound engineering judgment. The statistical analysis showed that data for all roadway types cannot be combined from HSIS states. Additional data is required to supplement current HSIS data to combine other roadway types and the remaining three states for the development of IHSDM.

Based on the analysis of data from two states participating in the HSIS, the following conclusions were reached:

• State data from the HSIS can be used to develop average accident rates for different roadway types.
• Based on the variables available in the geometric (roadway) files, it is possible to stratify the roadway into eight categories, and it also allows the separation of the highway network into roadway sections, intersections, and interchanges.
• The files have the capability of linking variables to assign the accidents and volumes to the various highway sections.
• The development of these average accident rates requires judicious manipulation of the data and sound engineering judgment.

Future Research

This study was an initial effort to develop accident rates for various roadway types to be used in the level 1 IHSDM. Additional work is planned in this area to stratify different roadway types by a number of key variables, such as type of terrain, width of clear zone, traffic volume, overall alignment, land-use characteristics, and weather condition of the region. This additional research will lead to the development of stable accident rates for use in the level 1 model.

References

(1) Jay L. Devore. Probability and Statistics for Engineering and the Sciences, Brooks/Cole Publishing Company, Pacific Grove, Calif., 1991, pp. 355-358.

Yusuf M. Mohamedshah is a transportation engineer with Advanced Engineering and Planning Corporation Inc. (AEPCO) that provides automated data processing support to the FHWA. Mr. Mohamedshah works with the FHWA's Design Concepts Research Division to develop IHSDM. Specifically, he provides technical support to the highway design team in obtaining information from the HSIS data base and in the use of computer-based roadway design packages. He has a bachelor's degree in civil engineering from the University of Bombay, India, and a master's degree in civil engineering from Virginia Polytechnic Institute and State University.

Amy R. Kohls was a cooperative education student in the Design Concepts Research Division. Her primary research during her co-op assignment was in accident analysis of state data bases. She received a bachelor's degree in civil engineering from The Catholic University of America. Currently, she is pursuing a master's degree in civil engineering at Virginia Polytechnic Institute and State University.