The impact of airline flight schedules on flight delays: An analysis of block-time, delay propagation, and schedule optimization using stochastic models

Mazhar Arikan, Purdue University

Abstract

This dissertation is motivated by the problem of airline flight delays which has come under increased scrutiny with the report released by the Joint Economic Committee (JEC) of the US Congress. The JEC report estimated the total cost of flight delays to the US economy at as much as $41 billion in 2007 (Schumer and Maloney, 2008). This dissertation studies airline schedules and its impact on delays in the US air transportation system. In the first part of this dissertation, we examine the impact of the scheduled block-time allocated for a flight on on-time arrival performance. We analyze a large database that contains detailed data from the Bureau of Transportation Statistics (BTS) website on each domestic flight flown in the years 2005–2007. Our purpose is to understand how operational, competitive, revenue, and cost factors drive the heterogeneity in airline scheduling decisions, by performing a series of hypothesis tests. By using a structural estimation approach, based on the newsvendor framework, our model disentangles how a specific factor—such as congestion—affects the scheduled block-time decision which impacts flight on-time performance. Our results show that airlines systematically "under-emphasize" flight delays, i.e., implied flight delay costs are less than the implied costs of early arrivals for a large fraction of scheduled flights. The second part analyzes flight schedules to measure the extent of the propagation of operational risk associated with network and schedule decisions of airlines. We attribute the operational disruptions of airlines mainly to two factors: (1) the randomness of intrinsic travel time of a flight, and (2) the propagation of this randomness through the airline network. Using these two factors, we construct a stochastic delay propagation model, and then answer several policy questions regarding airline networks and air-travel infrastructure. This stochastic model provides practical guidelines to airlines about which bottleneck resources they should focus on in order to minimize the total network impact of flight delays. We develop several measures regarding the robustness of airline schedules. Our analysis suggests that the overall on-time performance figures of airlines is highly dependent on the on-time definition used. By eliminating the 15 minutes buffer provided in the current BTS on-time metric, we show that overall on-time performance of airlines drops by around 20–25%. In addition, we observe that on-time performance figures of some airlines fall around 2–3% when passenger on-time arrival probability is considered instead of the flight on-time metric. We also provide an analysis of alleviating the impact of congestion by augmenting arrival/departure capacities at different airports. Our analysis shows that alleviating the impact of congestion at Atlanta Hartsfield-Jackson and Chicago O'Hare airports would provide the largest network benefit to the US air-travel system. In the third part of this dissertation, we address the issue of block-time allocation to flights that minimizes both planned resource and delay costs while ensuring a target scheduled on-time arrival probability for each flight. We develop three different optimization formulations and create efficient frontiers for a major US carrier, which capture the tradeoff in total network arrival delay minutes and total block-time budget. Our results indicate that the allocation of block-times in 2007 was not on the efficient frontier. We show that with our recommended solution, this airline could reduce the total network delay cost by 0.58% and achieve about 0.65% reduction in total network arrival delays without adding additional block-time minutes. This corresponds to about $3.7 Million in annual savings based on Air Transport Association estimates of per minute arrival delay cost. Additionally, our recommended solution leads to more robust schedules, as we obtain more stable on-time arrival probabilities throughout a day (across each departure hour block) and throughout a week (across different days of a week).

Degree

Ph.D.

Advisors

Deshpande, Purdue University.

Subject Area

Management|Economics|Transportation

COinS