Decomposing flight delay distributions: Tail, location, and center effects

Flight delays impose substantial costs on airlines, airports, and passengers. While existing studies predominantly focus on predicting average delays, extreme delays cause disproportionate operational disruptions and passenger dissatisfaction. This study introduces a distributional sensitivity analysis framework to identify factors that specifically drive extreme flight delays. Using 2.08 million U.S. domestic flights from 2024, we employ quantile regression models combined with Global Distributional Importance and Tangent-SHAP to decompose factor contributions across the entire delay distribution. Our analysis reveals three distinct effect patterns: tail effects, exhibited by weather conditions and carrier service history, which disproportionately amplify extreme delays; location effects, exhibited by realized delay variables such as departure delay and late aircraft delay, which shift the entire distribution approximately uniformly; and center effects, exhibited by flight duration and distance, which primarily affect typical delays. The classification is supported by an extensive robustness suite covering estimator noise, quadrature design, train–test partitioning, and correlation-preserving perturbations. We identify 60 min of departure delay as a critical threshold (bootstrap 95% confidence interval 55–65 min), beyond which 67 percent of flights experience severe arrival delays exceeding 60 min. Summer months exhibit 2.04 times higher tail risk than fall. The estimated annual cost of delays exceeding 60 min is approximately 1.9 billion USD, with potential savings of approximately 1.0 billion USD through targeted interventions. These findings provide actionable insights for airlines and airports to prioritize interventions targeting extreme delay prevention rather than average delay reduction.