Crash Prediction Models

Crash prediction models have been developed for signalized intersections in the U.S., as discussed previously in Chapter 3. However, no crash prediction models exist yet for U.S. roundabouts and driver behavior. Given the relatively recent introduction of roundabouts to the U.S. and driver unfamiliarity with them, crash prediction models from other countries should be used cautiously. As reported earlier in Section 5.3, crash statistics vary from country to country, both in terms of magnitude and in terms of collision types. Consequently, the application of a crash prediction model from another country may not accurately predict crash frequencies at U.S. locations. Nonetheless, these crash prediction models from other countries can be useful in understanding the relative effects of various geometric features on the number of crashes that might be expected. The user is thus cautioned to use these models only for comparative purposes and for obtaining insights into the refinement of individual geometric elements, not to use them for predicting absolute numbers of crashes under U.S. conditions.

Crash models relating crash frequency to roundabout characteristics are available from the United Kingdom. The sample consisted of 84 four-leg roundabouts of all sizes, small to large and with various number of approach lanes and entry lanes (flared or parallel entries) (1). Approach speeds were also evenly represented between 48 to 64 km/h (30 to 40 mph) and 80 to 113 km/h (50 to 70 mph). Crash data were collected for periods of 4 to 6 years, a total of 1,427 fatal, serious, and slight injuries only. The proportion of crashes with one casualty was 83.7 percent, and those with two casualties was 12.5 percent. The models are based on generalized linear regression of the exponential form, which assumes a Poisson distribution. Their goodness of fit is expressed in terms of scaled deviations that are moderately reliable. No additional variables, other than those listed below, could further improve the models significantly (see also (8)).

The British crash prediction equations (1), for each type of crash are listed in Equations 5-1 through 5-5. Note that these equations are only valid for roundabouts with four legs. However, the use of these models for relative comparisons may still be reasonable.

According to the U.K. crash models, the major physical factors that were statistically significant are entry width, circulatory width, entry path radius, approach curvature, and angle between entries. Some of the effects of these parameters are as follows:

• Entry width: For a total entry flow of 20,000 vehicles per day, widening an entry from one lane to two lanes is expected to cause 30 percent more injury crashes. At 40,000 vehicles per day, widening an entry from two lanes to three lanes will cause a 15 percent rise in injury crashes. Moreover, the models could not take into account the added hazard to bicyclists and pedestrians who will have to travel longer exposed distances. (8)
• Circulatory width: Widening the circulatory roadway has less impact on crashes than entry width. Crashes are expected to rise about 5 percent for a widening of two meters. (8)
• Entry path radius: Entry-circulating collision type increases with entry path radius (for the fastest path), while single vehicle and approach collision types decrease. For a double-lane approach, an optimum entry path radius is 50 to 70 m (165 to 230 ft). (8)
• Approach curvature: Approach curvature is safer when the approach curve is to the right and less so when the curve is to the left. This implies that a design is slightly safer when reverse curves are provided to gradually slow drivers before entry. For a double-lane approach roundabout with entering flow of 50,000 vehicles per day, changing a straight approach to a right-turning curve of 200 m (650 ft) radius reduces crash frequency by 5 percent. (8)
• Angle between entries: As the angle between entries decreases, the frequency of crashes increases. For example, an approach with an angle of 60 degrees to the next leg of the roundabout increases crash frequency by approximately 35 percent over approaches at 90-degree angles. Therefore, the angle between entries should be maximized to improve safety.

An approach suggested in Australia (13) differs from the British approach in that the independent variables are based on measures related to driver behavior. For instance, the collision rate for single vehicle crashes was found to be:

These equations demonstrate a direct relationship between the number of crashes, overall speed magnitudes, and the change in speed between elements. Therefore, this equation can be used to estimate the relative differences in safety benefits between various geometric configurations by estimating vehicle speeds through the various parts of a roundabout.

Congratulations, you have completed the knowledge section of the course.

You may now complete the course by successfully passing the course quiz.