Improving traffic safety

Jere Meredith, P.E.
Assistant Engineering Director
City of Kansas City, Missouri

In the 1980s, the City of Kansas City, Missouri, Public Works Department staff developed a procedure to identify critical locations and establish priorities for improvements. The method developed was described in an article entitled, "A Procedure for Identifying Needed Improvements in a Traffic Network," which was published in the November 1988 issue of the ITE Journal.

In the 1990s, the firm of Bucher, Willis, and Ratliff (BWR) refined this procedure and developed computer programs to identify critical locations, determine the magnitude of deficiencies, and establish priorities. The procedure was tested in Kansas City and other urban areas. Many of the intersection improvements recommended by BWR in their 1993 report have been implemented and some have been programmed for future years.

The above procedure utilized performance measures such as accidents, delays, and levels of service. The methodology proposed involved the determination of standards for performance measures mentioned. The actual operation of each location was then compared to the minimal level of acceptability and a deficiency index was developed. The locations were then ranked in descending order of deficiencies.

The traffic operational deficiency index was defined as the additional cost per year incurred by the community for inefficient traffic operation at an intersection or roadway section. The procedure converts the magnitude of deficiencies of individual performance measures into a dollar value by assigning standard road user costs to these measures. The total cost experienced by motorists at locations operating below acceptable levels are then determined. If solutions to eliminate the deficiencies can be determined, along with the cost of improvements, the deficiency/cost ratio can be used to establish priorities.

More recently, the staff incorporated current data into computer programs developed by BWR and executed runs to identify intersections experiencing deficiencies at present. No solutions or benefit cost ratios were pre-determined. At this time, a decision was made to emphasize safety improvements. Therefore, a list of intersections in descending order of accident deficiencies was prepared, along with a curve showing intersection rank vs. accident deficiency costs. The knee of the curve was used to determine the cut-off point. Intersections falling above the cut-off point were investigated in detail to determine short and long range improvements.

The purpose of this report is to describe recent efforts made to improve safety and the benefits realized through the implementation of some of the recommended short range actions. So far, short range actions have been implemented at a limited number of intersections. Therefore, the report describes the results of one year of "before" and "after" accident data at the first 12 locations where improvements were made.

Procedure
Collision diagrams were generated for each of the high accident locations. These were studied in detail and field inspections were made to correlate accidents with physical and operational features at over 50 intersections. Traffic volume data, intersection and signal drawings, etc., were also reviewed. Short-range actions most frequently recommended included installation of all red intervals, replacing 8" lenses with 12" lenses, enhancement of the display of signal indications by installing additional signal heads, and phasing changes. Long range actions included installation of mast mounted signal indications, signal reconstruction, widening of approaches, roadway realignments, and geometric improvements.

Comparison of "before" and "after" accident data
The 12 locations where short-range actions were implemented, along with a description of the action, installation date, "before" and "after" accident statistics, and the percent of change in the occurrence of accidents realized, is shown in Table 1. A majority of these intersections experienced a large number of right angle accidents. Data showed that accidents were reduced at all locations except two. At Benton Boulevard and 12th Street, an increase was noted and, at Benton Boulevard and 27th Street, no change in the frequency of accidents was observed. The reduction in accidents at locations other than those mentioned above ranged from 9 percent to 61 percent. Overall, a reduction of 33 percent was realized with "before" accidents totaling 245 and "after" accidents totaling 165.

Testing for significance of the reduction in accidents observed was done for two groups. In group 1, all locations shown in Table 1 were included. In group 2, the intersections of Benton Boulevard and 12th Street, Prospect Avenue and Brushcreek Boulevard, and Meyer Boulevard and Troost Avenue were omitted from the comparison. At Benton Boulevard and 12th Street, the length of the clearance interval was increased for the off-peak period only. Therefore, there is no reason to believe that the increase in accidents occurred due to a change in signal timing. At Paseo and Brush Creek Boulevard, the occurrence of accidents may have been influenced by detour plans implemented for construction projects in the area; at Meyer Boulevard and Troost Avenue, complaints were made and the original signal phasing was re-installed. Even with the elimination of the above locations from the comparison, an overall reduction of 33 percent was realized with the "before" accidents totaling 157 and the "after" accidents totaling 104.

Testing for significance
Traffic engineers have known for quite some time that the short-range actions described above result in reductions in accidents at locations experiencing a large number of right angle accidents. The analysis described here further supports this statement. However, for a complete evaluation it must be determined whether the observed percentage reduction in accidents is large enough to be ascribed to the change introduced or whether it is due to chance factors. Such an evaluation can be made by utilizing Poisson or Chi-Square tests.

Accidents are rare events. Therefore, an applicable statistical model is the Poisson distribution. In general, the assumption of the Poison distribution is met in most situations. Several studies have shown that this model fits traffic accident data very well. Therefore, one way to approach the problem and determine the significance of a percentage reduction in accidents is to assume that the observed data is a sample from Poisson distribution. If the before time period accident frequency is used as the expected value of the Poisson distribution, then it is possible to determine the percentage decrease in accidents that is statistically significant.

In order to make a more conservative estimate, an alternate is the Chi-Square test. By using this test, it is possible to determine whether the two samples differ significantly, that is, the reduction in accidents can be attributed to corrective measure implemented.

The number of "before" and "after" accidents described above, along with the percent reductions in overall accidents for groups 1 and 2, show that a statistically significant reduction in accidents has occurred even if the conservative test (Chi-Square) is used in the analysis. Thus, it is concluded that the observed reduction in accidents is large enough to be ascribed to the change introduced and is not a chance occurrence.

To contact Jere Meredith, please call 816-513-2777 or send e-mail to Jere_Meredith@kcmo.org.