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-98-133
Date: October 1998

Accident Models for Two-Lane Rural Roads: Segment and Intersections

2. Literature Review

Segment Models

Average Daily Traffic (ADT)

ADT is one of the most significant variables in predicting accidents, yet it is not controllable. Many models have used traffic exposure as a dependent variable although its relationship with accident counts is not fully linear. In general, it is recommended to use ADT as an independent variable for greater accuracy because it interacts with other controllable variables, and it measures the effect of traffic flow intensity (Hauer, 1994).

Lane Width, Shoulder Width, and Shoulder Type

Modeling approaches vary from study to study, and techniques of data collection and analysis likewise vary. Thus the effect of lane width and shoulder width on accident frequency has some variation in different studies. Generally it has been found that accident rates decrease when lane and shoulder widths increase. The report by Zegeer et al. (1986) on the effect of cross-section for two-lane rural roads indicated that a paved shoulder widening of 2 feet per side reduces accidents by 16%, while reports of Miaou et al. (1993) and Zegeer et al. (1986) found reductions of 8% and 6.6%, respectively. The latter two reports take into account horizontal curvature and curve length as explanatory variables, while the former does not explicitly include horizontal alignment. Luyanda et al. (1983) showed that shoulder type, an amalgam that includes width and surface type, is a significant variable but did not define this variable in detail. The synthesis of Jorgensen (1978) reported a negative relationship between accidents and shoulder width for two-lane rural highways on the basis of studies done primarily in the 1950's and 1960's. Variation of shoulder width for Interstate Highways and other freeways exists mostly along the inside shoulder, and older reports indicate that accidents increase as the inside shoulder width increases, contrary to the findings of Miaou et al. (1993). The increase of accidents with inside shoulder width may be due to emergency parking on wider shoulders or to insufficient accident history in the older studies.

Horizontal and Vertical Alignment

Horizontal and vertical alignment can be expressed in alternative ways to capture the effect of individual curves (disaggregate) or a sequence of curves (aggregate). Examples of measures of horizontal curvature are as follows:

where L is the segment length in miles and {i} is the absolute horizontal angle between the i-th and (i+1)-th tangents, in degrees. Here AC is aggregate and {i} is disaggregate. Vertical grade variables can be expressed similarly. Researchers have used both aggregate explanatory variables (Polus, 1980; Kulmala and Roine, 1988) and disaggregate ones (Miaou et al. 1993; Zegeer et al., 1991) in the modeling process, although aggregate variables are not directly helpful to designers who are improving individual curves. Nevertheless, aggregate variables are useful as surrogates in evaluating alignment safety. In most of the referenced reports, the results confirm the common sense opinion that sharper and longer curves result in more accidents, regardless of whether the statistical techniques applied are multiple linear regression or generalized linear models.

Roadside and Terrain Condition

When roadside features such as slopes, guardrails, trees, poles, etc. are considered separately, the portion of accident rates explained by roadside features is weak. The reports by Graham and Harwood (1982) and Zegeer et al. (1986) indicate this drawback. Zegeer et al. (1991) reported that mountainous terrain type has a negative effect on safety. Zegeer et al. (1987), as noted in Chapter 1, packaged the roadside variables in a subjective measure called Roadside Hazard Rating based on visual evaluation of clear zone and sideslope. Roadside Hazard Rating takes numerical values from one to seven. This measure "indicates the accident damage likely to be sustained by errant vehicles on a scale from one (low likelihood of an off-road collision or overturn) to seven (high likelihood of an accident resulting in a fatality or severe injury)." On a segment length with variable hazards, an average or middle value is assigned.


Various attempts have failed to find relationships between accidents and speed, whether the latter is design speed, posted speed, or operating speed. One of the few models where speed is considered comes from Finland (Kulmala and Roine, 1988). A report of Fridstrøm et al. (1995) indicates that a change in posted speed lowered fatal accidents in Denmark.


The influence of driveway accidents was highlighted by two studies (Fee et al., 1970; McGuirk and Staterly, 1976). Driveway density and driveway spacing were found to be significant safety factors. McGuirk and Staterly (1976) developed a linear model for accident rates Y:


where X is driveway spacing. Figure 1, illustrating the relationship of accidents to driveway density, appears in Cirillo (1992), and was taken from the report of Fee et al. (1970).


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