Multilevel modelling of flight demand under various hierarchical structures

This study compares route-based, airline-based, and time slot-based multilevel models of flight-level passenger demand under identical empirical conditions. Using 81,973 outbound flights from Taoyuan International Airport, the models share common fixed effects and random intercepts and are evaluated using information criteria, variance decomposition, intraclass correlations, and five-fold cross-validation. Route-based and airline-based hierarchies outperform time slot-based alternatives. The route-based hierarchy provides the best overall fit, whereas the airline-based hierarchy captures the strongest residual clustering. Aircraft characteristics, airline service frequency, and departure timing exhibit broadly stable associations across specifications. Airfare is positively associated with demand but represents a route-month-airline market characteristic rather than a causal price elasticity because of aggregation and potential endogeneity. The findings demonstrate the importance of hierarchy selection and identify demand relationships that remain robust across alternative hierarchical representations.