• Federal Highway Administration >
• Publications >
• Research Publications >
• 03050 >
• 09.Cfm >
• Surrogate Safety Measures From Traffic Simulation Models

Surrogate Safety Measures From Traffic Simulation Models

PDF files can be viewed with the Acrobat® Reader®

9. Validation of Surrogate Measures

There is a significant amount of additional effort required to validate that the proposed surrogate measures can adequately assess the safety of particular intersection conditions. The proposed surrogate measures are largely not observable by an independent roadside observer with only visually subjective information on vehicle locations and speeds. Past studies on TTC estimation have used categories to judge the value of TTC in bins (i.e., "high," "medium," "low" to correlate with 0-0.5 s, 0.5 s-1.0 s, 1.0 s-1.5 s) (33). Video analysis could be used to improve the estimated speed, acceleration, etc. of vehicles involved in particular conflict events so that better estimates of TTC, PET, etc. could be produced. The issue, however, is not whether the surrogates can be replicated in a field study, but rather whether the surrogates are correlated with observable behaviors that indicate the safety of a traffic facility. This does not mean that the surrogates need to be correlated directly to the actual number of crashes or conflicts at a particular intersection, but rather that the relative differences (or perhaps rank order) of various intersection designs as evaluated by the surrogate safety methodology are correlated with a similar study with real-world conflict measurements.

Three hypotheses for surrogate safety measures from simulation models and a corresponding validation test approach for each are listed in this section. Each validation test includes an estimate of the level of effort (LOE) required for executing the test activities. The hypotheses for the utility of the surrogate measures are:

1. Discriminating between the safety of two design alternatives in a simulation.
2. Correlation of the surrogate measures with real-world traffic conflict studies.
3. Correlation of surrogate measure reductions with predicted reductions in traffic conflicts.

Discrimination Between Intersection Design Alternatives

Hypothesis: Two different intersection designs produce different frequencies of traffic conflict events predicted by a simulation model. This indicates that one intersection design or strategy is more or less safe than another.

Positive Result: Validation that traffic simulation results could be used in evaluating proposed future alternatives for intersection redesign. Conclude that surrogate measure distributions are appropriate discriminators of relative intersection safety performance.

This hypothesis must be satisfied before the other hypotheses can be tested.

Approach

1. Code intersection design A in simulation model.
2. Code intersection design B in simulation model.
3. Simulate intersection designs over range of volume and turning probability scenarios.
   1. Replicate n times per scenario for statistical significance.
4. Collect surrogate measures for each design and compare statistical distributions of various aggregations (distributions of distributions). Test comparisons of:
   1. Total number of conflict events.
   2. Number of events of a particular type.
   3. Number of total events on a particular approach or movement.
   4. Other types of aggregations as appropriate.

Correlation With Traffic Conflicts

Hypothesis: High frequency of traffic conflict events predicted by a simulation model is correlated with high frequency of traffic conflicts as measured in a real-world study by the traffic conflicts technique.

Positive Result: Validation that traffic simulation results could be used to replace or augment traditional data gathering for safety analysis.

Approach

1. Code intersection design(s) in simulation model to match real-world intersection(s) with traffic conflict data.
2. Simulate intersection operations over volume and turning probability scenarios as experienced during the traffic conflicts study.
   1. Replicate n times per scenario for statistical significance during each scenario.
3. Collect surrogate measures from simulation model scenarios and compare how statistical distributions of various aggregations change with how the traffic conflicts data change for several control variables. Test comparisons of:
   1. Total number of conflict events.
   2. Number of events of a particular type.
   3. Number of total events on a particular approach or movement.
   4. Other types of aggregations as appropriate.

Prediction of Reductions in Traffic Conflicts

Hypothesis: Frequency of traffic conflict events predicted by the simulation model for a particular intersection improvement alternative is correlated with the actual change in the frequency of conflict events in the real world as measured in a real-world study.

Positive Result: Validation that the safety improvements predicted by the simulation model are not only relatively comparable (i.e., percentage improvements) across alternatives, but are also comparable in an absolute sense (total number of conflict events of particular types).

1. Code intersection design(s) for "before" condition A in simulation model to match intersection before improvements.
2. Code intersection design(s) for "after" condition B to match intersection after improvements.
3. Simulate intersection operations over volume and turning probability scenarios as experienced during the traffic conflicts study for before and after conditions.
   1. Replicate n times per scenario for statistical significance during each scenario.
4. Collect surrogate measures from simulation model scenarios.
5. Compare how statistical distributions of various aggregations change in the simulation model "before and after" with how the traffic conflicts data changed for the "before and after" conditions. Test comparisons of:
   1. Total number of conflict events.
   2. Number of events of a particular type.
   3. Number of total events on a particular approach or movement.
   4. Other types of aggregations as appropriate.
6. ALTERNATIVE TO (5): Compare predicted conflict reduction of the "after" condition with published collision and/or conflict reduction factors (average percent reductions). Repeat 1 through 4 for several other intersection designs and compare results to published conflict reduction factors.

Alternative Approach

1. Code various types of intersection designs.
2. Simulate intersection operations over volume and turning probability scenarios as experienced during the traffic conflicts studies.
   1. Replicate n times per scenario for statistical significance during each scenario.
3. Collect surrogate measures from simulation model scenarios.
4. Rank surrogate measure results for design scenarios by combining conflict statistical results into indices.
   1. Compare the rank order of the simulation design scenarios with the rank order of the design scenarios according to the potential for conflict reduction ranking.

Previous | Table of Contents | Next