Reliability as a measure of transportation system performance

Abstract

This investigation involved a comparative analysis to assess the quantitative and qualitative characteristics of seven previously proposed methods of measuring travel time reliability. For each method, reliability was calculated using a l2-week sample of traffic flow data for a 23 mile portion of the 1-405 corridor in Seattle, Washington. The methods were then evaluated and ranked at the qualitative level using such criteria as the clarity and simplicity of the measure, its analysis capability, descriptive ability, accuracy of results, and general applicability. It was determined that the Range of Travel Times and the Florida Reliability Method were best suited for measuring reliability in terms of clarity and simplicity. With regards to analysis capability, the Range of Travel Times, the Florida Reliability Method, and the Percent Variation were easy to calculate. The Range of Travel Times, Florida Reliability Method, Percent Variation, and Traffic Characteristic Profiles best reflected changes in traffic flow on the corridor. The Success Measure was the only measure that met accuracy requirements. With regards to general applicability, the Florida Reliability Method was the only method which characterized reliability as an indicator of how well conditions on the corridor met travelers' expectations. These observations formed the basis of the qualitative ranking, in which the Florida Reliability Method received the highest overall rating, followed by the Range of Travel Times, Percent Variation of Average Travel Times, Traffic Characteristic Profiles, and Variability Index. The Reliability Performance Indicator and the Success Measure received the lowest overall ratings. Data collection requirements for reliability measurement were also examined. An analysis of minimum sample size and aggregation intervals suggested that a 6-week and 3- month sample produced similar estimates of corridor travel times; however, it appeared that traffic flow data aggregated to a l5-minute interval did not provide the same results as a 5- minute interval. Based on these observations, it was concluded that a l-week data collection effort was too short, and the optimum data collection period for reliability measurement is a 6-week period using traffic data aggregated to a 5-minute interval.