Flight cancellations are costly events for both airlines and passengers, yet are poorly understood. This dissertation expands upon literature that has studied flight cancellations by incorporating more variables and using advanced model specifications. In addition, it is necessary to understand the drivers of flight cancellations to quantify the relationship between flight cancellations and flight delay forecasts, which has been poorly documented in the literature. This dissertation investigates the factors leading to flight cancellations and quantifies the effect of flight cancellations on flight delay forecasts.

First, econometric choice models are applied to a large dataset of historical flight information to determine the preferences and behaviors of airlines with respect to flight cancellations. The binary logit estimation results show that flight characteristics, such as load factor, distance, and flight frequency, are significant for determining the likelihood of flight cancellations, even when accounting for adverse weather effects. Airline-specific logit models indicate large heterogeneity with respect to flight cancellation tendencies across the industry. Inter-flight heterogeneity is explored through the use of mixed logit and latent class models, but lack of significant heterogeneity and long computation times provide evidence that a basic binary model can be sufficient for capturing the flight cancellation behavior of airlines. Cancellation predictions are made at an airport-level, but the distribution of predicted cancellations does not match well with the actual distribution observed in the data.

Second, deterministic queueing methods are used to quantify the effect flight cancellations have on queueing delay forecasts. The cancellation model estimates are used to predict flight cancellations for a sample of all flights for 160 airport-days. The reductions in delay due to cancellations are captured using Monte Carlo simulation and a first-order approximation. The simulation results show that delays are reduced by 22% when considering the effect of cancellations and the first-order approximation results are no more than 4% larger than those from the Monte Carlo simulation.

Finally, a case study was performed based on the current operating environment at San Francisco International Airport, where capacity reductions are expected during the summer of 2014 due to runway construction. Moreover, airlines are proposing schedules with 5% more demand. The increased schedule combined with the capacity decrease leads to an large increase in the queueing delay forecasts. A cancellation model is used to predict the changes in delay that result from cancellations induced by the change in operating conditions. The results from the cancellation model indicate that departure cancellations will increase at an almost one-to-one ratio with the proposed demand increase, thus negating any benefit to airlines from a denser schedule. The feedback of cancellations on queueing delay is further explored with analytical models. As witnessed in the case study, queueing delay can reach a theroetical maximum where any additions to the flight schedule results in higher queueing delays and an associated increase in flight cancellations that compensate for the additional flight and return the demand, and queueing delay, to its original level.