An application of survival and frailty analysis to the study of taxi-out time: A case of New York Kennedy Airport Tony Diana * Federal Aviation Administration, 800 Independence Avenue, SW Washington, DC 20591, USA Keywords: Survival analysis Flight delay Airport performance Airport taxing abstract This study uses survival models to evaluate how selected operational factors affect the duration of aircraft taxi-out times at John F. Kennedy Airport, New York. Frailty models help assess whether fixed or random effects are likely to explain differences between two summers, 2006 and 2007. The hourly departure records for summer are censored when operations occurred below the airport’s ceiling and visibility minima, that is, in instrument meteorological conditions. Cox regression models showed that block delay and the percent of airport utilized capacity are most likely to increase the risk of longer taxi- out times in instrument meteorological conditions compared with other factors such as departure delays, arrival delays and the volume of departures. Frailty analysis reveals that taxi-out times are not signifi- cantly affected by either fixed or random effects. Published by Elsevier Ltd. 1. Introduction This study compares taxi-out time between June and August 2006 and same period for 2007 at New York John F. Kennedy International Airport (JFK). Based on the Airline Service Quality Performance data (ASQP), 1 the percent of on-time gate departures declined from 73.1% in the summer 2006 to 64.6% in 2007, while taxi-out time increased from 36.2 to 43.46 min over the same period. Based on ASQP, 31,275 departures were reported by the US carriers in summer 2007, up 12.8% from summer 2006. Delta Air Lines and JetBlue increased their scheduled operations respectively 20.1% and 19.0% in summer 2007 compared with previous summer. On the other hand, departure cancellations went up from 535 in summer 2006 to 1218 in summer 2007. Overall,10.65% of takeoff and landing operations in summer 2006 occurred in instrument meteorological conditions (IMC) compared with 11.81% in summer 2007. Instrument meteorological conditions at JFK are determined by a cloud ceiling lower than 2000 feet and visibility below 4 nautical miles. While there was not a significant difference in the percentage of operations in IMC and despite more cancellations between the two summers, taxi-out time still went up. To understand the reasons for longer taxi-out times between the two time periods, we focus on the relationship between selected operational variables and taxi- out times in semi and parametric models as well as to determine whether the duration of taxi-out time can be attributed to fixed or random effects. Frailty models are extensions of the proportional hazards model, the most common model in survival analysis. Frailty models are used to determine whether differences in taxi- out times between summer 2006 and 2007 could be explained by unobserved individual departure-level factors otherwise unac- counted for by other survival models’ estimates. 2. Survival and frailty analysis Survival analysis, also called reliability analysis in the engi- neering field, represents a tool to compute time-to-event proba- bilities given specific circumstances such as the treatment for a disease in the medical field or component failure in engineering. It has not, however, been widely applied to the analysis of airport efficiency and airline delay. Proportional hazard models assume that observations have hazard functions proportional to one another and they can accommodate a variety of shapes for the common hazard function * Tel.: þ1 410 564 0286. E-mail addresses: Tony.Diana@faa.gov, tonydiana1@verizon.net. 1 The data compiled by the US Bureau of Transportation Statistics are available at http://www.bts.gov as well as at https://aspm.faa.gov/asqp. ASQP statistics are based on the flights reported by the US carriers. Under 14 C.F.R Part 234 (Airline Service Quality Performance Reports), the reporting carriers are those certificated under 49 U.S.C. 41102 that accounted for at least 1% of domestic scheduled- passenger revenues in the 12 months ending March 31. A reportable flight is any nonstop flight, including a mechanically delayed flight, to or from any airport within the contiguous 48 states that accounts for at least 1% of domestic scheduled- passenger enplanements in the previous calendar year, as reported to the Depart- ment pursuant to part 241 of this title. Contents lists available at SciVerse ScienceDirect Journal of Air Transport Management journal homepage: www.elsevier.com/locate/jairtraman 0969-6997/$ e see front matter Published by Elsevier Ltd. http://dx.doi.org/10.1016/j.jairtraman.2012.10.002 Journal of Air Transport Management 26 (2013) 40e43