Office of Operations
21st Century Operations Using 21st Century Technologies

Application of Travel Time Data and Statistics to Travel Time Reliability Analyses: Handbook and Support Materials

Chapter 5. Conclusions

Data Sources for Travel Time Reliability

A wide variety of travel time data are available for the development of reliability performance measures. These data are of two general types:

  • Roadway based. Data from freeway detectors have a long history of use in performance measurement as they have been available for over 25 years. Speeds, volumes, and lane occupancies are collected at specific points on roadways. An advantage of these data are that volumes and speeds are paired so that detailed exposure measures can be developed. A large disadvantage is that the speeds that are taken at a point may not be reflective of actual travel time over entire roadway segments. Assumptions must be made to convert the spot speed to travel times over a distance.
  • GPS based. These data are derived from tracking individual devices as users traverse the roadway network and are commonly referred to as “probe vehicle data” because their motion implies that the device is on or in a motorized vehicle. The data may be reported in its raw form—the position of devices in time and space—and referred to as “vehicle trajectory data.” The actual paths (trajectories) of devices through the network can be calculated. These data can also be summarized temporally to an “epoch” (e.g., 5-minute time increments), and spatially to a unidirectional link.

Of these data types, vehicle trajectory data offer the higher resolution and therefore can be used to calculate reliability measures at any temporal and spatial level desired. For example, the travel time variability between individual vehicles on the same path and time can be ascertained. This flexibility comes with a higher cost of processing as vehicles in the raw data must be assigned to roadway segments by time. Vehicle trajectory data can also be used to monitor trip performance (including reliability) directly; trip performance must be created synthetically with other forms of travel time data.

Results of the Case Study

  • Because trajectory data measures the paths of individual vehicles, the travel time distributions display high variability in travel times compared to other sources (vehicle probe and detector) that are temporarily pre-aggregated (5-minute intervals for this study). The variability appears to drop during times of congestion as vehicle movements are hampered by heavy traffic flow and queues. But using pre-aggregated data obscures variability, and the resulting reliability measures are reduced in value. The question then is: “Does it matter for congestion monitoring?”; in other words, “Is the variation between vehicles an important aspect of reliability?” Vehicle-to-vehicle variation is important in the context of operations strategies, many of which are aimed at smoothing traffic flow. Within the broader context of congestion monitoring and performance reporting, pre-aggregated travel time data (at least at 5-minute intervals) is adequate to capture macroscopic congestion conditions.
  • Reliability measures developed from trajectory data are generally higher than those developed from pre-aggregated probe data. This difference is probably due to the inclusion of vehicle-to-vehicle variation described above. However, the difference indicates that the two sources should not be combined for congestion monitoring purposes, as misleading results will ensue.
  • Reliability measures developed from pre-aggregated probe data via different processing methods are reasonably close in values. Two processing methods were tested: the snapshot method and the virtual probe method. While the virtual probe method is theoretically more representative of how vehicles pass through the system, there does not appear to be a reason to choose one method over the other. For relatively short length urban facilities and trips, either method may be used, but no strong reason exists to go through the extra calculation complexity of the virtual probe method. Likewise, no strong reason exists to go through the extra calculation complexity of the virtual probe method for the development of reliability measures.
  • Reliability measures developed from roadway detector data are almost always lower in value than those developed from pre-aggregated probe data, based on data from detectors in Los Angeles and San Francisco.
  • All the reliability measures tested in the study were correlated with each other. The MTTI, TTI80, and PTI were strongly correlated with each other and loosely correlated with the semistandard deviation and the LOTTR metric. Understanding the relationship between TTI, TTI80, and PTI could be useful in planning applications which require that reliability measures be developed from model-developed average conditions.
  • Reliability measures for long-distance trips are more sensitive to the pre-aggregated vehicle probe database processing method than for urban facilities. When considering trips that are made continuously throughout the year, the difference in processing method is generally around 5 percent, although trips exposed to a high number of urban conditions show more deviation. When discrete periods are considered, the deviation is greater. This result leads us to use the virtual probe method to develop reliability measures for long-distance trips rather than the snapshot method.
  • The PM3 system reliability measure generally decreases as PTI increases for freeways, although the correlation is weak. The difference most likely lies in the nature of the measures: System reliability relies on a threshold to determine if a facility is unreliable (binary) whereas the PTI is a continuous variable. With binary variables, one facility may be only slightly over the threshold while another might be way over the threshold. In both cases, the facility is deemed to be unreliable, but the latter case is more severe, a condition captured by the PTI but ignored by the system reliability measure.

    On signalized arterials, the correlation between the PM3 system reliability measure and the PTI is extremely low. In general, the reliability measures indicate that travel is more unreliable on signalized arterials (as they have higher PTI values), but the reliability may be a scaling issue. PTI is determined by assuming a free-flow or ideal travel time. On freeways, this assumption is clear but not for arterials. Many references, including the HCM, use what is essentially the midblock speed as the free-flow benchmark; this PM3 measure assumes that the signal has no influence when in fact its mere presence even under low traffic volumes will introduce delay, depending on the phasing and progression. Many researchers and practitioners feel this delay should be included when measuring congestion. Fortunately, the HCM can be used as a guide. Urban street LOS thresholds are set as fractional multipliers of the (midblock) travel speeds. For this study, the authors of this document used a multiplier of 0.75, which corresponds to LOS B. However, even accounting for this adjustment signalized arterial reliability is still high; the ones used the analysis are all over PTI = 2.0. A possible explanation is that on a signalized section, some portions (TMCs in our case) will be strongly influenced by the signal (e.g., approaches) while others that are some distance away will not. Another option is to apply the relationships from NCHRP Report 387, which estimates free-flow speed (including signal control delay) from midblock speed and estimated signal delay.