TFHRC Home > Safety > Safety Publications > Surrogate Safety Assessment Model and Validation: Final Report > Chapter 3 (Continued)

Chapter 3. Theoretical Validation (Continued)

CASE STUDIES

Various types of intersections have been implemented and evaluated in three simulation systems: VISSIM, TEXAS, and AIMSUN.

The goal of this portion of the validation effort was not to compare the results of the simulation model with traffic at a comparable real-world location. Hence, no calibration effort was necessary or performed in this study. Reasonable driver behavior was verified, and appropriate control measures were used to avoid gridlock during high-volume test cases. As such, for all the intersection designs, default driving behavior models and parameters were applied for each simulation model. The same underlying simulation parameters were used for each comparison case to maintain comparability.

Eleven comparison cases were executed among the three simulation systems as follows:

TEXAS Cases

• Case 1: Signalized, four-leg intersection with permitted left turn versus protected left turn.
• Case 2: Signalized, four-leg intersection with and without left turn bay.
• Case 3: Signalized, four-leg intersection with and without right turn bay.

VISSIM Cases

• Case 4: Signalized, four-leg intersection with leading left turns versus lagging left turns.
• Case 5: Signalized, four-leg intersection versus offset T-intersection;
• Case 6: Diamond interchange with three-phase timing versus four-phase timing.
• Case 7: Single point urban interchange (SPUI) versus diamond interchange.

AIMSUN Cases

• Case 8: Signalized, four-leg intersection with left turns versus signalized intersection with median U-turns.
• Case 9: Signalized, four-leg intersection versus single roundabout.
• Case 10: Signalized, three-leg, T-intersection versus single roundabout with three legs.
• Case 11: Diamond interchange versus double roundabout.

Three sets of traffic volumes (low, medium, and high) were applied for each intersection design, and timing plans were designed to ensure no over-saturation would occur.

Case 1: Conventional Four-Leg Intersection with Permitted Left Turn Versus Protected Left Turn (TEXAS)

There are two basic alternative designs for left turns: protected and permitted. Protected left-turn design allocates an exclusive phase for left turn only, which will make the left-turn maneuvers have fewer conflict events with the opposing through traffic. Permitted left-turn design allows vehicles to make a left turn during the through traffic green phase and provides no specific green phase for left-turn only. This logic applies mostly to traffic conditions with low left-turn volumes. When the left-turn volumes become higher, there have been more conflict events when drivers begin accepting smaller gaps to cross the intersection.

According to the crash prediction models for four-leg signalized intersection, the existence of left-turn phase will result in lower crash frequency. Thus, it is hypothesized that protected left turn should have lower predicted conflict frequency than permitted left turn when other network parameters remain the same. Also, it would be reasonable to expect that severity values of the surrogate measures would be less critical for protected left turns versus permitted left turns.

Intersection Description for Case Study

The intersection used to test the alternative traffic control logic for left turns is four-legged intersections with three through lanes in the main travel directions and two through lanes on the side-street approaches to the intersection, as shown in figure 29. All left-turn bays are 76.25 m (250 ft) long.

Figure 29. Screen Capture. Intersection Geometry for Testing Control Logic.

Table 8 lists the traffic volumes applied for each approach of the intersection. Fixed time traffic control is applied in this test. Figure 30 through figure 35 provide the key timing plan parameters for each testing scenario.

Note: L, TH, and R correspond to vehicles proceeding left, through, or right at the intersection.

Figure 30. Illustration. Timing Plan for Permitted Left Turn in Low Volumes.

Figure 31. Illustration. Timing Plan for Protected Left Turn in Low Volumes.

Figure 32. Illustration. Timing Plan for Permitted Left Turn in Medium Volumes.

Figure 33. Illustration. Timing Plan for Protected Left Turn in Medium Volumes.

Figure 34. Illustration. Timing Plan for Permitted Left Turn in High Volumes.

Figure 35. Illustration. Timing Plan for Protected Left Turn in High Volumes.

Data Analysis and Comparison Results

Ten replications were performed for each design case and the resulting output trajectory data were analyzed by SSAM. F-test and t-tests were applied to compare surrogate measures of safety and the aggregations of those measures.

Table 9 through Table 13 list the values of all surrogate measures of safety and corresponding t test results for different types of aggregations with the low-speed events and crash data excluded (TTC ≤ 0 and MaxS ≥ 16.1 km/h (10 mi/h)).

Note: Shaded cells indicate statistically significant differences between the two alternatives.

This table illustrates a counterintuitive result. The design with the protected left turn has, on average, more total conflicts than the case with the permitted left turn for medium- and high-traffic volumes. The following tables explain this result by breaking the total results into a result for each conflict type.

Note: Shaded cells indicate statistically significant differences between the two alternatives.

This table indicates what is expected to happen from adding a protected left-turn phase; that the total crossing conflicts are reduced for all levels of traffic volume. This indicates that SSAM, in its most basic form, is a valid indicator of safety.

Note: Shaded cells indicate statistically significant differences between the two alternatives.

This table indicates the large increase in rear-end conflicts for high and medium volumes that is generated by adding the protected left-turn phase. This large increase in rear-end events is the primary cause of the total conflicts being counter-indicative. Thus, it should be important to analyze all types of conflicts rather than just examining the total number of events when comparing designs.

Note: Shaded cells indicate statistically significant differences between the two alternatives.

This table indicates a 40-percent increase in lane change events at medium volumes and a 125-percent increase at high volumes. These results are likely due to longer queues in the left-turn bay because there is no permitted portion of the phase.

Note: Shaded cells indicate statistically significant differences between the two alternatives. The tan and blue colors indicate extreme values to the right and left columns respectively.

Correlations with Predicted Crash Frequency

The predicted crash rates (crashes per year) for all scenarios in this test are listed in table 14 with the corresponding surrogate measures of safety (conflicts per hour). Rank orders for each category of data are also listed in the table. The Spearman rank correlation coefficients are calculated for each test.

Note: Rows labeled "M" provide mean values and rows labeled "R" provide the ranking of each alternative. The Rs column provides Spearman rank correlation coefficients indicating agreement with theoretical crash estimates.

Findings and Conclusions

Based on the observation of the safety surrogate test data, the following conclusions can be drawn:

• Total number of conflicts for protected left turn is significantly more than that of permitted left turn.
• Protected left turn has less crossing conflicts but more rear-end and lane-change conflicts than permitted left turn.
• The comparison results for all other safety measures show no distinct safety preference between protected left turn and permitted left turn.

In general, the surrogate measures present mixed results; however, an appropriate conclusion can be drawn with consideration of the differing severities of different conflict types. The addition of a protected left-turn phase tended to increase the total number of conflicts while the protected phase substantially decreased crossing conflicts, as expected. The increase in conflicts came primarily from rear ends and, under higher flows, from lane-changing maneuvers. This result is elucidated by considering that, in adding a protected left-turn phase, the cycle time was increased and the proportion of green time for through phases was decreased. This change in the timing has the effect of increasing the number of vehicle stops for through traffic. The number of vehicle stops is known to correlate with the number of rear-end crashes and thus with higher rear-end conflicts. It is also possible that with a greater proportion of vehicles arriving to a standing or dispersing queue, there is an increased tendency of drivers to change to a lane with a shorter queue, despite all lanes having stopped traffic. Drivers in free-flowing lanes may be relatively content to stay in their lane when all lanes are flowing at the same (nonzero) speed. Thus, the conflict frequency results appear reasonable and have provoked consideration of the effect of timing changes and driver behavior. However, the severity-related surrogate measures indicate that the protected left-turn case has improved average values (increased TTC and PET and decreased DeltaV, especially at high volumes), indicating that the protected left-turn phasing is safer than the permitted left-turn case, as would be expected.

The Spearman-rank correlation coefficients from all tests show a strong positive relationship between the rank orders of the surrogate measures of safety and the rank orders of the predicted crash rates. The relationships between the rank order of the totals of all conflict types and crossing conflict types are stronger than the relationship of rear-end and lane-change crossing conflicts. This, again, would be expected because it has been validated in the field that protected left turns reduce crossing crashes. TEXAS, however, shows a very high rate of rear-end and lane-change events per hour, indicating that the default driver behavior parameters may allow vehicles to perform maneuvers that allow closer proximity than the "rule of thumb" threshold of TTC = 1.5 would preclude in the real world.

Case 2: Left-Turn Bay Versus No Left-Turn Bay (TEXAS)

A left-turn bay on an approach to an intersection provides an independent lane for the storage and movement of the left-turn vehicles. With the left-turn bay, the conflict events between through movement vehicles (primarily traveling in the same direction as the turn vehicles) and left-turn vehicles is hypothesized to be significantly reduced. This has been tested over a range of traffic volume scenarios from light traffic to heavy traffic.

According to the crash prediction models for all conventional intersections, the existence of a left-turn bay will reduce the crash frequency under the same traffic conditions.⁽¹³⁾

Intersection Description

The intersection used to test the left-turns bay versus no left-turn bay is a four-legged intersection with two through lanes with shared right turn for all approaches to the intersection, as shown in figure 36 and figure 37. All left-turn bays are 76.25 m (250 ft) long. Table 15 indicates the traffic volumes arriving to each approach of the intersection. Fixed-time traffic control is applied in this test. The ring-diagrams from figure 38 through figure 43 show the timing plans for each testing scenario.

Figure 36. Screen Capture. Exclusive Left-Turn Lane.

Figure 37. Screen Capture. Shared Use Left-Turn and Through Lane.

Note: L, TH, and R correspond to vehicles proceeding left, through, or right at the intersection.

Figure 38. Illustration. Timing Plan for Intersection with Left-Turn Bay in Low Volumes.

Figure 39. Illustration. Timing Plan for Intersection without Left-Turn Bay in Low Volumes.

Figure 40. Illustration. Timing Plan for Intersection with Left-Turn Bay in Medium Volumes.

Figure 41. Illustration. Timing Plan for Intersection without Left-Turn Bay in Medium Volumes.

Figure 42. Illustration. Timing Plan for Intersection with Left-Turn Bay in High Volumes.

Figure 43. Illustration. Timing Plan for Intersection without Left-Turn Bay in High Volumes.

Data Analysis and Comparison Results

Ten replications were performed for each simulation scenario, and the resulting output trajectory data were analyzed by SSAM. F-test and t-tests were applied to identify statistical significance. Table 16 through table 20 list the values of all surrogate measures of safety and corresponding t 'test results for different types of aggregations with the low speed events and crash data excluded (TTC ≤ 0 and MaxS ≥ 16.1 km/h (10 mi/h)).

Note: NLB indicates no left-turn bay and WLB indicates with left-turn bay. Shaded cells indicate statistically significant differences between the two alternatives. The tan and blue colors indicate extreme values to the right and left columns respectively.

Note: NLB indicates no left-turn bay and WLB indicates with left-turn bay. Shaded cells indicate statistically significant differences between the two alternatives. The tan and blue colors indicate extreme values to the right and left columns respectively.

This table indicates that, for all traffic volumes, the number of severe-crossing conflicts is increased when the left-turn bay is added. This result likely reflects the increase in the number of available left-turn maneuvers due to the bay.

Note: NLB indicates no left-turn bay and WLB indicates with left-turn bay. Shaded cells indicate statistically significant differences between the two alternatives. The tan and blue colors indicate extreme values to the right and left columns respectively.

This table indicates a definite decrease in the number of rear-end conflicts when a left- turn bay is added to the intersection, as expected from field experience.

Note: NLB indicates no left-turn bay and WLB indicates with left-turn bay. Shaded cells indicate statistically significant differences between the two alternatives. The tan and blue colors indicate extreme values to the right and left columns respectively.