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 1. Introduction

Task Purpose

The purpose of this work is to provide a handbook for the application of travel time data to travel time reliability analyses. The handbook covers several topics including:

  • Data sources for reliability
  • Data processing methods for reliability
  • Reliability measure creation and comparison, highlighting differences and similarities in the use of data from different sources

The intent of the handbook is twofold:

  • To provide practitioners with the ability to understand the differences in reliability measures derived from the different data sources
  • To document the steps needed to turn high-resolution travel time data into reliability performance measures

Issues Surrounding Reliability

Defining reliability up to this point has largely been a technical exercise aimed at practitioners and researchers. For example, in the original Future Strategic Highway Research Program, reliability was defined as: “ travel times vary over time(e.g., hour-to-hour, day-to-day).”1 This definition has persisted and formed the basis for developing reliability performance measures and analytical methods. From an analyst’s perspective, reliability is often depicted as a travel time distribution to convey variability, such as is shown in figure 1. Additional measures that describe the size and shape of the travel time distribution such as the semistandard deviation also have been used. Essentially, reliability is just a characteristic of overall congestion rather than a distinct phenomenon—how congestion varies over time.

It is generally acknowledged that the travel time distribution is used to measure reliability, but how is travel time itself defined? Travel time is measured in a variety of ways with a variety of different data (direct, indirect, and purely synthetic) and all of these methods have been used to calculate reliability. There has been almost no resource material describing these various data and methods and the implications they have on the values of reliability measures. The specific issues dealt with in this project are described below.

Example distribution of travel times taken from a segment of freeway.

Source: Transportation Research Board, National Academy of Sciences.

Figure 1. Graph. Example travel time distribution and associated reliability measures.2

Example distribution of travel times taken from a segment of freeway. The distribution is skewed toward higher travel times. Superimposed on the distribution are the performance measures used to describe travel time reliability: planning time, buffer time, misery time, standard deviation, and the upper percentiles of the travel time distribution. The point here is that all of the measures used to describe travel time reliability are developed from the distribution of travel times for a facility or trip.
Trip-Based Travel Time Reliability

Because of the nature of the data that have been available, nearly all reliability reporting is based on the facility perspective. The data measurements used relate to the performance of a facility, not an end-to-end trip as made by travelers. Trip performance can be synthesized from facility-based data using the virtual probe method, but how well this method represents actual vehicle travel times has not been determined.3

A comprehensive mobility measurement program will involve using both trip- and facility-based measures because they both inform analysis about the nature of mobility in a region:

  • Many transportation investments are focused on improving and managing facilities, so facility-based measures are highly useful to planners and engineers. This focus is particularly true for operations and capacity improvements as well as some types of demand management.
  • Other transportation investments—as well as land-use and development policies—are more oriented to the entire trip-making process, so understanding trip performance informs us about our customers’ (i.e., travelers) transportation system experience.
  • Emerging operations strategies—such as active transportation and demand management and integrated facility management—also need to consider the entire trip-making process.

Data from vendors are now becoming available that allow trip-based measures to be developed; these data track the location and time of individual vehicles and are described throughout this handbook. However, trip-based reliability measurement poses its own challenges. Facility measurement describes the nature of congestion to which travelers are exposed. When using these data, analysts are left to decide the origin and destination of a trip. Trip measurement includes factors in addition to congestion exposure—how travelers interact with the entire landscape. Regarding trips, travelers are generally free to change departure times and routes and, in some cases, destinations and modes as well. In this project, the research team fixed the origins and destinations to compare trajectory data to facility-based data.

Over time, however, trip purposes and destinations change, resulting in multiple definitions of what a “trip” is even though it may be measured with the same measure. For example, a work trip could have the same start time and route as well as exclusively use a car every day. Alternately, these factors could vary to different degrees. Measuring a true trip from the traveler’s perspective entails measuring a variety of factors, many of which are beyond the control of transportation agencies. Finally, measuring trip performance can be viewed as how participants (travelers and businesses) adapt to the landscape. This adaptation no doubt includes congestion avoidance (e.g., selecting origins and destinations to minimize congestion exposure) and associated costs, which are not captured in trip-based measures.

Standard Processing Procedures

Standard processing procedures for calculating performance measures from high-resolution data do not exist. Analysts use different methods for performing quality control (QC), imputation/handling of missing data, aggregating data, and computing measures, resulting in different values for performance measures created from the same data.

Because little detailed data collected under rigorous controls exist for comparison, QC procedures for travel time data are primitive. For freeway detectors, where volumes and speed measurements exist, checks can be made against traffic flow parameters but deciding how far astray a value should be before it is considered erroneous is problematic. For vehicle probe data, the situation is even more restrictive. Cross-checking travel time data against disruption data (i.e., weather, incidents, and work zones) would be a way of verifying that low speeds are legitimate, but this project did not deal with data from these other data sources.

1 Transportation Research Board of the National Academies of Sciences, Engineering, and Medicine. 2003. NCHRP Report 510: Interim Planning for a Future Strategic Highway Research Program. Washington, DC: National Academy of Sciences. [Return to note 1]

2 Transportation Research Board of the National Academies of Sciences, Engineering, and Medicine. 2014. Incorporating Travel Time Reliability into the Highway Capacity Manual, Transportation Research Board. Report No. S2-L08-RW-1. Washington, DC: Transportation Research Board. [Return to note 2]

3 In this method, a time-space matrix of facility-based travel times is prepared. The movement of vehicles across the matrix is then simulated. [Return to note 3]