PASSENGER TRIP DELAYS IN THE U.S. AIRLINE TRANSPORTATION SYSTEM IN 2007

Transcription

1 Submied 2/08 Inernaional Conference on Research in Air Transporaion (ICRAT 2008) PASSENGER TRIP DELAYS IN THE U.S. AIRLINE TRANSPORTATION SYSTEM IN 2007 Guillermo Calderón-Meza PhD candidae Cener for Air Transporaion Sysem and Research/GMU Fairfax, VA, USA Lance Sherry, PhD Cener for Air Transporaion Sysem and Research/GMU Fairfax, VA, USA George Donohue Cener for Air Transporaion Sysem and Research /GMU Fairfax, VA, USA Absrac The value of he air ransporaion sysem is he ransporaion of ligh-weigh, high-value cargo, and passengers. Indusry and governmen merics for he performance of he air ransporaion focus on he performance of he flighs. Previous research has idenified he discrepancy beween fligh performance and passenger rip performance, and has developed algorihms for he esimaion of passenger rip performance from publicly available daa. This paper describes an analysis of passenger rip delays for 5224 roues beween 309 air pors in he U.S. air ransporaion sysem for The average rip delay experienced by passengers was 24.3 minues for a naionwide oal of 247 Million hours. Flighs delayed 15 minues or more conribued 48% of he oal delays, cancelled flighs 43%, divered flighs 3%, and flighs delayed less han 15 minues conribued he remaining 6%. Passenger rip delays for oversold flighs were negligible. Analysis of passenger rip delays for roues and airpors, and he implicaions of hese resuls are also discussed. Keywords- passenger rip delay; fligh delay, airpor delay. I. INTRODUCTION The value proposiion of he air ransporaion sysem is he rapid, safe, and cos effecive ransporaion of high-value, lighweigh cargo, and human passengers. This ransporaion is achieved by combining air ransporaion beween airpor erminals wih ground ransporaion beween origin (e.g. home)/desinaion (e.g. meeing) and he airpor. The air componen of he ransporaion is achieved hrough via single segmen or muliple connecing segmen scheduled airline operaions. To leverage economies of scale, airlines schedule and operae a daily iinerary ha neworks passengers, aircraf, fligh, and cabin crews in connecing segmens hroughou he day. Individual flighs on a segmen may be delayed for several reasons such as: (e.g. mechanical) problems, weahe or raffic congesion. To mainain inegriy of heir neworks in he presence of individually delayed flighs, airlines may choose o delay, diver, or cancel flighs. When flighs are delayed, he passenger rip for his segmen is also delayed for he duraion of he fligh delay. When flighs are cancelled or divered, or passengers are bumped for overbooking, he passenger rip delay includes he duraion of delay accrued waiing for he re-booked fligh. All of hese delays represen passenger rip delays. Previous research by Brau & Barnhar [2005] idenified he discrepancy beween fligh performance and passenger rip performance. Wang [2007] showed ha he 2% of passengers experiencing cancelled flighs accrued delays of approximaely 10 hours each, and ha he oal delays experienced by hese passengers accouned for 4 of he oal passenger rip delays. This research provides he resuls of analysis of he U.S. air ransporaion sysem in The resuls are summarized as follows: 1. Passengers experienced a oal of 247 Million hours of delays. The average delay was 24.3 minues. Flighs delayed 15 minues or more accouned for 48% of he oal delays, cancelled fligh 43%, divered flighs 3%, and flighs delayed less han 15 minues accouned for almos all he remaining 6%. Passenger rip delays for overbooked passengers were less han 1%. 2. For flighs on he 5224 roues beween 309 airpors, 5 of he roues experience an average passenger rip delay less han 15 minues. 9 of he roues experience an average rip delay of less han 30 minues. 3. For flighs inbound and oubound of he 309 airpors, 4 of he airpors experience an average passenger rip delay of less han 15 minues, 9 less han 30 minues. Poorly performing airpors included major hub airpors as well as small commuer airpors. 4. Passenger rip delay exhibied similar performance on roues of differen sage-lenghs The paper is organized as follows: Secion 2 provides a summary of previous research. Secion 3 describes he algorihm and daabase srucure used o compue esimaes of passenger rip delay in Secion 4 describes he resuls of 1 Copyrigh Cener for Air Transporaion Sysems Research (CATSR)/GMU 02/08

2 Submied 2/08 Inernaional Conference on Research in Air Transporaion (ICRAT 2008) he analysis. Secion 5, Conclusions, discusses he implicaions of hese resuls. II. PREVIOUS RESEARCH Researchers have shown ha fligh-based merics, like he merics repored in he Deparmen of Transporaion s Airline Travel Consumer Repors (ATCR) [DOT, 2007] are a poor proxy for passenger experience [Wang, Schaefe Wojik, 2003; Mukherjee, Ball, Subramanian, 2006; Ball, 2006; Brau & Barnhar, 2005]. Brau & Barnhar [2005] used proprieary airline daa o sudy passenger rip imes from a hub of a major U.S. airline. This sudy showed ha ha fligh-based merics are poor surrogaes for passenger delays for hub-and-spoke airlines as hey do no capure he effec of missed connecions, and fligh cancellaions. For example, for a 10 day period in Augus 2000, Brau & Barnhar [2005] cie ha 85.7% of passengers ha are no disruped by missed connecions and cancelled flighs arrive wihin one hour of heir scheduled arrival ime and experience an average delay of 16 minues. This is roughly equivalen o he average fligh delay of 15.4 minues for his period. In conras, he 14.3% of he passengers ha are disruped by missed connecions or cancelled flighs experienced an average delay of 303 minues. Wang [2007], Sherry, Wang & Donohue [2006] developed an algorihm o esimae passenger rip delay for publicly available daa from he Bureau of Transporaion Saisics (hp:// One par of he algorihm joins separae daabases wih secondary daa o derive he parameers o perform he passenger rip delay analysis. The nex par of he algorihm compues an esimae of passenger rip delay for each scheduled fligh. Key among hose parameers used in he algorihm is he Passenger Load Facor for a fligh. This algorihm uses he quarerly average Passenger Load Facor for flighs on a given roue. This resuls in undercouning for peak operaions, and possible overcouning for non-peak operaions. Furher his analysis accouns for fligh delays and cancelled flighs only for roues beween he OEP-35 airpors. The main resuls of his analysis are ha passenger rip delays are disproporionaely generaed by cancelled flighs. Passengers scheduled on cancelled flighs represen 3 percen of oal enplanemens, bu generaed 45 percen of oal passenger rip delay. On average, passengers scheduled on cancelled flighs experienced 607 minues delay, and passengers who missed he connecions experienced 341 minues delay in Figure 1. ER diagram of he local daabase 2 Copyrigh Cener for Air Transporaion Sysems Research (CATSR)/GMU 02/08

3 Submied 2/08 Inernaional Conference on Research in Air Transporaion (ICRAT 2008) The analysis described in his paper improved he algorihm by increasing he pre-processing of daa o eliminae infeasible daa and check for referenial inegriy. Furher improvemens were made o he algorihm o include divered flighs, improve processing hroughpu and auomaing manual seps in he processing. III. DATABASE AND ALGORITHM A. The local daabase A local relaional daabase sores daa impored from public daabases. The daa consis of acual fligh and performance values colleced by compeen insiuions. Being as massive as hey are, he raw daa conain errors. Because of h he daabase includes consrains o improve he qualiy of he inpu daa. The design of he local daabase is illusraed by an ER diagram as shown in Fig. 1; i consiss of six eniies and hireen inegriy and referenial consrains. Since he daa are ime dependen all several eniies idenify he uples using year and monh among oher aribues. Oher aribues ha idenify uples in he eniies are he carrier or airline, and he roue (composed of one origin airpor and one desinaion airpor). The Airpor and Airline eniies make sure ha he oher eniies conain only known airpors and airline codes: all of he oher eniies have foreign keys referring o Airpor and Airline. The On_Time eniy conains he daa abou each individual fligh. In paricula he aribue canceled, if is value is one, indicaes ha he fligh was canceled (a value of one).; oherwise, is value is zero. The aribue div_delay is eiher 0 for no divered flighs or 360 (min) for divered flighs. The aribues avasea and avgpax are only used as emporal variables during he compuaion of Esimaed Passenger Trip Delay, EPTD [Wang, 2007, Sherry, Wang & Donohue, 2006]. The aribue pax_delay (min) is he cumulaed EPTD for all he passengers of he fligh. Clearly, if canceled is 1, div_delay mus be 0, and if div_delay is no 0, hen canceled mus be 0. The aribues carrier and airline are only differen when he acual carrier is a subsidiary of an airline. The T_100 eniy conains he inpu daa concerning performance of pairs of roue and carrier for domesic flighs only. There are no daa for individual flighs. The eniy includes informaion abou he oal number of deparures done for a roue and a carrier in he paricular monh (deparures_performed), he oal number of passengers ranspored (passengers), he oal number of seas including all he flighs (seas), and he disance of he paricular roue (in miles). The eniy Load_facors conains daa derived from T_100. For a paricular roue and airline, he each record conains he average number of unoccupied (available) seas in he flighs (avasea), he average number of passengers per fligh (avgpax), and he average number seas in he plane -he size of he plane- (avgsea). Clearly, he following condiions mus be rue a all imes: avgsea avasea and avgsea avgpax. The eniy PTDI conains he resul of he Passenger Trip Delay Index (PTDI) compuaion. In his case, flighs are idenified by heir roue, carrie and deparure ime: no individual flighs are recorded in his eniy, bu only averages of he flighs ha occur periodically a he given roue, carrie and deparure ime. The eniy also includes daa abou he oal number of enplanemens 1 (enp), he average oal number of seas available (avg_avail), he average load facor of his fligh (avg_ld_facor), he number of scheduled flighs (schfl), he number of canceled flighs (canceled_fl), he number of divered flighs (divered_fl), and he average delay ime in minues and number of passengers delayed for each caegory (canceled, divered, delayed, and on-ime) of fligh. Finally, he eniy also conains (hough redundanly because i can be derived from he oher aribues) he PTDI value in minues. Noice ha he delays can be zero, negaive or posiive real numbers. Negaive numbers indicae ha he passengers were no delayed bu hey arrived early. The number of enplanemens mus be greaer han zero for he PTDI o make sense. The same happens wih he number of scheduled flighs. Clearly, he condiion canceled_fl + divered_fl Schfl mus be rue a all imes. B. Inpu daa The compuaion of he PTDI uses daa from he Bureau of Transporaion Saisics (BTS); paricularly from wo daabases ha are available on-line o download. The firs daabase is he T-100 for he domesic segmen [BTS, 2006b]. This daabase allows he download of a whole year for all he carriers in he domesic (USA) segmen. The fields seleced o download are: yea monh, origin, des, carrie seas, deparures performed, passengers, carrier region, and disance. This experimen uses a single file conaining daa for he year 2007 from January o Ocober 2. The file conains records for 203 differen carriers, 1142 airpors 3, and roues. The process o compue load facors for he flighs and disance informaion for he roues uses hese values. Every record of his file mus comply wih he condiions saes in Table I o ener he local daabase. TABLE I. CONDITIONS FOR EACH RECORD OF THE T_100 DATABASE Field Condiion Year Equal o 2007 Monh In range [1, 10] Origin The value mus be already in he Airpor able Des The value mus be already in he Airpor able Carrier The value mus be already in he Airline able Seas An ineger number ha is greaer han or equal o Passengers Deparures performed A posiive ineger number Passengers A posiive ineger number Carrier region Only he value D (for domesic) is acceped Disance A posiive real number 1 An enplanemen is a ranspored passenger. 2 November and December were no available a he ime of he experimen. 3 These daa include airpors in Puero Rico, and airpors in projec ha are being used already. 3 Copyrigh Cener for Air Transporaion Sysems Research (CATSR)/GMU 02/08

4 Submied 2/08 Inernaional Conference on Research in Air Transporaion (ICRAT 2008) A record ha does no comply wih all he condiions does no ener he local daabase, so ha i is no used during he compuaion of he PTDIs. A oal of records acually enered he local daabase including 932 airpors, 115 carriers, and roues. Noice ha some of he airpors, carriers and roues are no acually referred in he On-Time daabase for he same period of ime. These exra records in T_100 have no effec in he final resuls because he algorihm does no use hem. The values for seas, passengers, and deparures performed are monhly oals. There are no daa for individual flighs; herefore, average values are used in his experimen o approximae he acual values. The local daabase derives and sores he following values concerning load facors per yea monh, roue, and carrier: average number of seas, avgsea = seas / deparures performed average number of passengers, avgpax = passengers / deparures performed average number of available seas, avasea = (seas passengers) / deparures performed Therefore, he average load facor for a yea monh, roue, and carrier is: lf = avgpax / avgsea. The second daabase is he so-called Airline On-Time Performance [BTS, 2006a]. This daabase allows he download of individual monhs of a paricular year for all he airpors and carriers in he USA. The fields seleced o download are: fligh_dae, carrie origin, des, arr delay, crs arr ime, dep delay, crs dep ime, cancelled, divered, fl_num, and ail_num. This experimen uses en separae files for he year 2007, one for each monh from January o Ocober. Table II summarizes he figures for each one of he files. TABLE II. STATISTICS FOR EACH OF THE ON-TIME INPUT FILES Monh Records Carriers Airpors Roues January February March April May June July Augus Sepember Ocober Toal enered Noice ha only 17 of he 20 carriers enered he local daabase. I is because he records wih he hree missing carriers did no comply wih he condiions saed below. To ener he local daabase, each record mus comply wih he condiions saed in Table III. TABLE III. CONDITIONS FOR EACH RECORD OF THE ON-TIME DATABASE Field Condiion Fligh dae Any valid dae for he year 2007 Origin The value mus be already in he Airpor able Des The value mus be already in he Airpor able Carrier The value mus be already in he Airline able Arrival delay Any ineger number (including 0 and negaive ones). Scheduled arrival ime A four digi posiive ineger number. The wo lefmos digis represen he hour in 24 hr forma. The wo righ-mos digis represen he minues. Deparure delay Any ineger number (including 0 and negaive ones). Scheduled deparure A four digi posiive ineger number. The wo lefmos ime digis represen he hour in 24 hr forma. The wo righ-mos digis represen he minues. Cancelled Eiher 0 (no cancelled) or 1 (cancelled) Divered Eiher 0 (no divered) or 360 (6 hrs in minues) Fligh number Any value, bu usually a hree or four digi ineger number. Tail number Any value. Used only o filer invalid records. Each record mus be unique wih respec o fligh dae, origin, desinaion, carrie and fligh number. If here are repeaed records, only one of hem eners he local daabase. When he repeaed records show differences in oher fields, he user decides which one o keep. For insance, one of he records saes ha he fligh was delayed and he ohe ha i was cancelled. The cancelled fligh eners he local daabase in his case. Siuaions like his are no frequen: for he curren inpu daa only 53 records were repeaed. C. The algorihm A a very high level of absracion he algorihm o compue he PTDI is as follows: Impor he T_100 daa ino he local daabase. This implies he compuaion of he load facor-relaed values. Impor he on-ime daa ino he local daabase. This implies he consideraion of he carrier / subsidiaries relaions. This means ha subsidiaries are changed o heir paren carrier every ime hey appear. Compue he EPTD based on he local load facor values and he local on-ime daa. This is done flighby-fligh, one monh a a ime. Fig. 2 illusraes he compuaion process of he EPTD. Compue he PTDI based on he EPTD, he delay, cancellaion, and diversion daa. The following formulas compue he EPTD for each caegory of passengers: 4 Copyrigh Cener for Air Transporaion Sysems Research (CATSR)/GMU 02/08

5 Submied 2/08 Inernaional Conference on Research in Air Transporaion (ICRAT 2008) Compue he passenger delay for on-ime flighs: hose arriving early or up o 15 minues afer he scheduled arrival ime 4. Compue he passenger delay for delayed flighs: hose arriving 15 or more minues afer he scheduled arrival ime. Compue he passenger delay for canceled flighs. Compue he passenger delay for divered flighs. Compue he number of enplanemens. Compue he PTDI-relaed load facors. Eliminae null values (if any) and merge flighs ha depar less han 40 minues afer anoher fligh of he same carrier on he same roue. Compue he PTDI. Fig. 3 illusraes he compuaion process of he PTDI. The following formula compues he PTDI: EPTD EPTD EPTD EPTD on ime delayed cancelled ( f ) = ( f ) * ArrDelay ( f ) = ( f ) * ArrDelay ( f ) = ( f, j) * max(15 * 60, SchArr( j) SchArr( f ) + ArrDelay( j)) divered Figure 2. Algorihm o compue he EPTD j ( f ) = ( f ) * 6 * 60 < ( f ) ( f ) Where (f) is he number of passenger in he fligh f. (f, j) is he number of passenger from fligh f, ha were reloaded on fligh j. ArrDelay <15 (f) is he arrival delay of fligh f (in minues) when i is less han 15 minues (fligh arrives onime). ArrDelay 15 (f) is he arrival delay of fligh f (in minues) when i is delayed (15 minues or more delay). SchArr(f) is he scheduled arrival ime of fligh f. The consan 15*60 represens he maximum wai ime (assumed) he passengers will olerae before changing o anoher airline or ransporaion means, i equals 15 hours (in minues). The consan 6 * 60 is he esimaed delay ime for a divered fligh; i equals 6 hours (in minues). A a high level of absracion, he compuaion of he PTDI consiss of eigh seps: PTDI on ime = * EPTDon ime delayed canceled divered * EPTD * EPTD * EPTD delayed canceled divered Where on-ime is he number of passenger on-ime (less han 15 minues delay), delayed is he number of passengers delayed, canceled is he number of passengers in canceled flighs, and divered is he number of passengers in divered flighs. Noice ha he summaions are performed afer grouping he flighs by roue (r), airline (a), and deparure ime (). Corresponding definiions are valid for he EPTD. Sub or superscrips indicae ha he associaed values correspond o he average EPTD for he caegory (on-ime, delayed, canceled, divered) afer grouping by roue, airline, and deparure ime The convenion is ha flighs arriving wih less hen 15 minues of delay are on-ime. 5 Copyrigh Cener for Air Transporaion Sysems Research (CATSR)/GMU 02/08

6 Submied 2/08 Inernaional Conference on Research in Air Transporaion (ICRAT 2008) IV. RESULTS The following analysis was conduced for 2007 for he monhs January hrough Ocober using daa derived from he BTS daabase for hose monhs and year. The daa included 512.8M passengers on 6.2 million flighs on 5224 roues beween 309 airpors. The passenger rip delay includes an esimae of he oal number of delay hours for on-ime, delayed, cancelled, and divered flighs. Esimaed oal passenger rip was M hours. The average rip delay was minues. Esimaed oal passenger rip delay for passengers on flighs delayed more han 15 minues M hours. The average rip delay for hese passengers was minues. Esimaed oal passenger rip delay for passengers on cancelled flighs was M hours. The average rip delay for hese passengers was minues. Esimaed oal passenger rip delay for passengers on divered flighs was 7.77 M hours. The average rip delay for hese passengers was 360 minues. Esimaed for passenger rip delay for over-booked passengers was negligible. A. Comparison of fligh delay and passenger delay Fig. 4 shows a graphical comparison of fligh delay and passenger rip delay (PTD). The y axis of he char shows percenage of he oal delay hours. The caegories included are delayed, cancelled, and divered flighs. Flighs ha arrived early or wih less han 15 minues of delay are no included in he char: hey are considered on-ime. Because of hese onime flighs, he bars do no add o 10. In oher words, he on-ime flighs can also generae delays, bu hey are low enough o consider hem as negligible. The oal delay measured using fligh delay is 1.63 million hours as indicaed in he char. Noice ha he fligh delay meric does no consider canceled flighs because hose flighs do no incur in delays. On he oher hand, he oal delay measured using PTD is million hours. This amoun is very differen from he 1.63 million of he oher meric. In his case he oal also considers he delays due o canceled flighs, and no only divered and delayed flighs. The PTD meric is more deailed and faihful o he real siuaion: passengers from a canceled fligh experience considerable delays. In fac, he delays for passenger from canceled flighs amoun for abou 43% of he oal delay. Abou 48% of he oal delay is due o delayed flighs, and he res of he delay is disribued among divered and on-ime flighs. B. Comparison of roues Fig. 5 compares he hisograms and cumulaive disribuions of he average PTDI and he maximum PTDI for all he roues wih respec o he delay ranges (15 minues each range). Form he poin of view of he average PTDI, 5 of he roues show on-ime flighs; and 9 of hem show flighs ha are delayed less han 30 minues. In exreme siuaions (maximum PTDI) abou of he roues show on-ime flighs and 5 show flighs delayed 30 minues or less. This disribuion shows a peak no a he 0-15 minue range as he one for he average PTDI, bu a he minues range. Afer he peak, he disribuion descends monoonically slower han i he disribuion of he average PTDI. Percenage Figure 3. Algorihm o compue he PTDI Fligh delay (1.63M hours) Performance meric Divered Delayed Canceled PTD (247.08M hours) Figure 4. Comparison of fligh delay and passenger delay as performance merics 6 Copyrigh Cener for Air Transporaion Sysems Research (CATSR)/GMU 02/08

7 Submied 2/08 Inernaional Conference on Research in Air Transporaion (ICRAT 2008) 35% % # Roues % Roues 15% 5% PTDI (mins) Avg PTDI Max PTDI Cum avg Cum max Figure 5. Disribuion of roues wih respec o he delay range Comparison of roue disance The disribuion of roues is similar for each disance range as shown in Fig. 6. Noice ha he disance ranges are given in % Roues Avg PTDI (mins) < 300nm > 1000nm Figure 6. Percenage of roues grouped by disance per delay range nauical miles (nm). All he ranges show beween 55 and 47 percen of on-ime roues. Beween 29 and 38 percen of he roues show delays of 15 o 30 minues. For he delay of 45 minues he percenages are beween 8 and 12. For he oher disance ranges he behavior is also similar hough wih smaller percenage values. Though he differences are no big (8% a mos), shorer roues end o perform beer: mos of he roues of 500 nm and less are on-ime (delay smaller han 15 minues). Longer roues end o delay more ofen. A significan par of he roues longer hen 500 nm delay 30 minues. The informal comparison of he disribuion of delays across disance ranges shows ha he disribuion has he same shape for all he disance ranges as shown in Fig. 7. In all he cases mos of he flighs are on-ime and hen he number of delayed flighs decreases wih each increase in he delay range. Bu, his char also says ha for shorer roues, is less probable o have long delay han i is for longer roues. For insance, he raio of on-ime o 30 minues delay is abou 17/10 = 1.7 for roues of 300 nm or less, bu i is 31/24.5 = 1.2 for roues of 500 o 1000 nm. C. Comparison of airpors The nex sep afer comparing he roues is he comparison of he airpors. In he case of inbound airpors, Fig. 8 shows ha mos of hem receive flighs on-ime or wih 30 minues delays: 4 of he airpors show on-ime flighs, and 9 show delays of 30 minues or less. Only few airpors show average delays of 45 minues or longer. Table 4 summarizes a ranking of all he inbound airpors in he daabase wih respec o he average delay. TABLE IV. < 300nm > 1000nm Disance Range (nm) < 15 mins < 30 mins < 45 mins < 60 mins < 75 mins < 90 mins < 105 mins < 120 mins < 135 mins # Airpors Figure 7. Percenage of roues grouped by disance range and delay range More Avg PTDI (mins) Avg PTDI Cum avg PTDI Figure 8. Inbound airpor performance BEST AND WORST INBOUND AIRPORTS RANKED ACCORDING TO PTDI Bes Wors Rank Airpor (delay) Rank Airpor (delay) 1 Greenville, MS 202 PHL (23) 2 Hilo, HN 226 IAD (26) 3 Pocaello, ID 239 DFW (31) 22 HNL 241 EWR (31) 31 SJC 245 LGA (33) 35 HOU 248 ORD (33) 39 OAK (10) 255 JFK (37) 40 MDW (10) 268 Meridian Regional (95) 59 LAS (11) 269 Rhinelander-Oneida (171) 61 DAL (11) 75 BWI (12) This ranking is based on he average PTDI for he airpor. Rank ies are possible as shown in he able. Airpors in bold belong o he OEP-35. Noice ha some of he OEP-35 7 Copyrigh Cener for Air Transporaion Sysems Research (CATSR)/GMU 02/08

8 Submied 2/08 Inernaional Conference on Research in Air Transporaion (ICRAT 2008) airpors are ranked among he 75 bes ones wih respec he PTDI values. The oubound airpors behave as he inbound ones wih respec o PTDI (see Fig. 9). Abou 9 he of he airpors show delays of 30 minues or less, and 4 show delays of 15 minues or less. Again, only few airpors show average delays of 45 minues or more. # Airpors Table 5 summarizes he ranking of all he oubound airpors wih respec o he average PTDI. TABLE V Avg PTDI (mins) Avg Cum avg Figure 9. Oubound airpor performance More BEST AND WORST OUTBOUND AIRPORTS RANKED ACCORDING TO PTDI Bes Wors Rank Airpor (delay) Rank Airpor (delay) 1 Brisol/Johnson, TN 194 PHL (23) 2 Pocaello, ID 214 IAD (26) 6 Greenville, MS 229 EWR (29) 25 SJC 238 DFW (31) 28 HNL 239 ORD (32) 36 OAK (9) 248 LGA (34) 38 HOU (9) 249 JFK (35) 42 DAL (11) 265 Rhinelander-Oneida (55) 53 MDW (11) 270 Middle GA Reg (260) 89 BWI (13) 109 LAS (15) This ranking is based on he average PTDI for he airpor. Rank ies are possible as shown in he able. Airpors in bold belong o he OEP-35. Noice ha some of he OEP-35 airpors are ranked among he 89 bes ones wih respec he PTDI values. D. Comparison of airlines Finally, Table 6 summarizes he ranking of he airlines wih respec o he average and maximum PTDI. Noice ha in he case of average PTDI he difference is a mos 27 minues. In he case of he maximum PTDI, he difference is a mos 700 minues. This ranking is based on eiher he average or he maximum PTDI for he airpor as indicaed in he column headings of he able. Rank ies are possible as shown in he able. TABLE VI. AIRLINES RANKED BY PTDI Average PTDI Maximum PTDI Rank Airline (delay) Rank Airline (delay) 1 Hawaiin (5) 1 Alaska (50) 2 Aloha 2 Aloha 3 Souhwes 3 Hawaiin 4 Fronier 4 Fronier 5 Air Tran 5 Souhwes (200) 6 Coninenal 6 USAirways 7 Alaska 7 Air Tran 8 ExpressJe (19) 8 Coninenal (250) 9 Unied (19) 9 JeBlue 10 SkyWes (19) 10 SkyWes 11 USAirways 11 Unied 12 Dela 12 ExpressJe 13 Norhwes 13 Norhwes/Airlink (291) 14 Norhwes/Airlink 14 Mesa 15 Mesa 15 American 16 JeBlue 16 Dela 17 American (32) 17 Norhwes (750) V. CONCLUSIONS Passenger rip delay is a criical performance meric for he airline ransporaion sysem. This meric assesses he performance of he rue end-users of he sysem, and provides a measure of he rue cos of delays. Fuure research is planned o: (1) exend he algorihm o include los luggage and refine he overbooked passenger algorihm, (2) add an algorihm adjus he load facor for peak and non-peak periods, (3) coninue o refine he auomaion of daa rerieval and processing. ACKNOWLEDGMENT The auhors would like o acknowledge he echnical assisance and suggesions from Maria Consiglio, Brian Baxley, and Kur Niezke (NASA-LaRC), Todd Farley (NASA-ARC), Joe Pos, Dan Murhy, Sephanie Chung, Dave Knor Anne Suissa (FAA, ATO-P), John Shorle, Rajesh Ganesan, Melanie Larson, Loni Nah, and Bengi Manley (GMU). This research was funded by NASA NRA NNN and Cener for Air Transporaion - George Mason Universiy Research Foundaion. REFERENCES Ball,M., D. Lovell, A. Mukherjee, and A. Subramanian. (2006) Analisys of Passenger Delays: developing a passenger delay meric, NEXTOR NAS Performance Merics Conference. ASiloma CA. March Brau, and C. Barnhar, (2005) An Analysis of Passenger Delays Using Fligh Operaions and Passenger Booking Da Air Traffic Conrol, Volume 13, Number. 1, 1-27 Bureau of Transporaion and Saisics, (2006a). Airline On-Time Performance. Daa. Available: hp:// ne%20ontime%20performance%20daa&db_shor_name=on-time Bureau of Transporaion and Saisics, (2006b). Form 41 Traffic T-100 Domesic Segmen Daa. Available: hp:// 20Carrier%20Saisics%28Form%2041%20Traffic%29&DB_Shor _Name=Air%20Carries Deparmen of Transporaion, (2006). Air Travel Consumer Repor. Available: hp://airconsumer.os.do.gov/repors/index.hm 8 Copyrigh Cener for Air Transporaion Sysems Research (CATSR)/GMU 02/08

9 Submied 2/08 Inernaional Conference on Research in Air Transporaion (ICRAT 2008) Mukherjee,. M. Ball, B. Subramanian (2006) Models for Esimaing Monhly Delays and Cancellaions in he NAS. NEXTOR NAS Performance Merics Conference, ASiloma CA. March Wang, D.(2007) Mehods for Analysis of Passenger Trip Performance In a Complex Neworked Transporaion Sysem, Disseraion, George Mason Universiy. Available hp://casr.ie.gmu.edu. Wang, D., Sherry, L. & Donohue, G. (2006). Passenger Trip Time Meric for Air Transporaion. The 2nd Inernaional Conference on Research in Air Transporaion. Wang, P., Schaefe L. & Wojcik, L. (2003). Fligh Connecions and Their Impacs on Delay Propagaion. In Proceedings of he 22nd Digial Avionics Sysems Conference. Volume 1, Wang,D. L.Sherry, Ning Xu, and M. Larson. (2008) Saisical Comparison of Passenger Trip Delay and Fligh Delay Merics. In Proceedings Transporaion Review Board 26 h Annual Conference, Washingon D.C. 9 Copyrigh Cener for Air Transporaion Sysems Research (CATSR)/GMU 02/08