Estimating time of day demand with errors in reported preferred times: An application to airline travel

摘要

An essential element of demand modeling in the airline industry is the representation of time of day demand—the demand for agiven itinerary as a function of its departure or arrival times. It is an important datum that drives successful scheduling and fleetdecisions. There are two key components to this problem: the distribution of the time of day demand and how preferred traveltime influences itinerary choice. This paper focuses on estimating the time of day distribution. Our objective is to estimate it in amanner that is not confounded with air travel supply; is a function of the characteristics of the traveler, the trip, and the market;and accounts for potential measurement errors in self-reported travel time preferences. We employ a stated preference datasetcollected by intercepting people who were booking continental US trips via an internet booking service. Respondents reportedpreferred travel times as well as choices from a hypothetical set of itineraries. We parameterize the time of day distribution as amixture of normal distributions (due to the strong peaking nature of travel time preferences) and allow the mixing function tovary by individual characteristics and trip attributes. We estimate the time of day distribution and the itinerary choice modeljointly in a manner that can account for measurement error in the self-reported travel time preferences. We find that the mixtureof normal distributions fits the time of day distribution well and is behaviorally intuitive, and the strongest covariates of traveltime preferences are party size and time zone change. The methodology employed to treat self-reported travel time preferences aspotentially having error contributes to the broader transportation time of day demand literature, which either assumes that thedesired travel times are known with certainty or that they are unknown. We find that the error in self-reported travel timepreferences is statistically significant and impacts the inferred time of day demand distribution.