Selecting the Representative Travel Time Reliability Measure Based on Metric (Dis)Agreement Patterns

Available TTR metrics can exhibit discrepancies even for the same travel time data, i.e., one metric indicates a reliable travel time distribution, while another one indicates otherwise. This conflict raises concerns for practitioners regarding which metric to use for policy making. To identify the (dis)agreement patterns between TTR metrics, Lognormal, Weibull, Gamma, and Inverse Gaussian distributions were selected to generate travel time distributions. Similarly, six commonly used TTR metrics (Percent variation, Width of travel time distribution, Skew index, Buffer index, 95th percentile, and Misery index) were used to calculate the TTR for hypothetical distributions. Descriptive analysis, majority voting and k-means clustering approaches were utilized to identify the representative TTR metric. The findings revealed that distribution skewness, regardless of the distribution type, can indicate when the metrics are more likely to agree, i.e., the practitioners can choose the TTR metric arbitrarily for travel time distributions that have a skewness smaller than 1.2 or larger than 1.8, because all metrics unanimously agree. It was also shown that for travel time distributions in the disagreement skewness range, buffer index can be used as the representative TTR metric. The results were also validated with USDOT Next Generation Simulation (NGSIM) vehicle trajectory dataset.