The Impact of Travel Time Reliability and Perceived Service Quality on Airport Ground Access Mode Choice

Transcription

1 Joural of Choice Modellig, 4(2), pp The Impact of Travel Time Reliability ad Perceived Service Quality o Airport Groud Access Mode Choice Mei-Lig Tam 1,* William H. K. Lam 2, Hig-Po Lo 3,Ŧ 1,3 Departmet of Maagemet Scieces, City Uiversity of Hog Kog, Hog Kog 2 Departmet of Civil ad Structural Egieerig, The Hog Kog Polytechic Uiversity, Hog Kog Received 11 Jauary 2010, revised versio received 6 April 2011, accepted 7 July 2011 Abstract This study makes two cotributios to existig airport groud access mode choice models. The first is a assessmet of travel time reliability o air passeger airport groud access mode choice decisios. Revealed preferece questios were asked to determie the safety margi allowed for groud access jourey to airports. The larger the safety margi allowaces, the less reliable the passeger perceived the mode to be. Stated preferece questios were also used to determie the impact of travel time reliability o mode choice decisios. The secod cotributio of this research is the icorporatio of air passeger perceived service quality i the calibratio of airport groud access mode choice model. With the use of the survey data, the effects of safety margi allowaces, travel time reliability, ad perceived service quality o groud access mode choices to Hog Kog Iteratioal Airport are quatified by a multiomial logit-type mode choice model. For strategic plaig, the calibrated model ca be used by the airport authority ad various trasport operators for evaluatig the chages i the service attributes o modal split patter i iteratioal airports, hece improvig the access mode services. Keywords: Groud access mode choice, perceived service quality, safety margi, travel time reliability * Correspodig author, T: , F: , susaat@cityu.edu.hk T: , F: , cehklam@polyu.edu.hk Ŧ T: , F: , mshplo@cityu.edu.hk

2 1 Itroductio The demad of air trasportatio icreased overwhelmigly i the past decade owig to the ecoomic growth worldwide. Air passeger traffic is expected to keep risig at four percet per year util at least 2020 (Airports Coucil Iteratioal 2005). The air traffic expasio has lead to demad overload o airport groud access traffic, ad is a subject of growig cocer by airport authorities. I order to facilitate airport groud access plaig, the Bay Area Metropolita Trasportatio Commissio (MTC) has coducted surveys regularly at three iteratioal airports i the Sa Fracisco Bay Area, Sa Fracisco Iteratioal Airport, Oaklad Iteratioal Airport ad Norma Y Mieta Sa Jose Iteratioal Airport. With the use of the MTC survey data, Harvey (1986) ad Pels et al. (2001; 2003) developed discrete choice models to quatify the relative importace of various factors regardig air passeger groud access mode choice decisios. They highlighted that busiess air passegers are more sesitive to travel time while obusiess air passegers are more sesitive to travel cost. Other factors ifluecig airport groud access mode choices icluded trip purpose, party size, residetial status ad the umber of pieces of baggage carried by a air passeger (Clark ad Lam 1990; Harvey 1986; Shapiro et al. 1996). Moteiro ad Hase (1996) calibrated both multiomial logit (MNL) ad ested logit (NL) models, with 1990 MTC survey data, for aalyzig departig air passeger airport-access mode choices. Their model results idicated that groud access mode choice strogly iflueces the airport choice i a multiple airport regio. They coteded that rail has a market for accessig the airports i the Sa Fracisco Bay Area, ad recommeded that the Sa Fracisco Bay Area Rapid Trasit (BART) rail system to Sa Fracisco Iteratioal Airport be exteded. Thus icreasig the utilizatio of the BART system ad stregthe Sa Fracisco Iteratioal Airport as the domiat airport i the Sa Fracisco Bay Area. Loo (2008) also applied MNL modelig to idetify attributes that affect passegers airport choice i multiple airport regios. However, based o the stated preferece (SP) data collected at Hog Kog Iteratioal Airport (HKIA), Loo (2008) foud that the umber of access modes ad access cost were ot sigificat o passegers airport choice. I recet years, the use of more flexible model formats, such as mixed multiomial logit (MMNL) model ad cross-ested logit (CNL) model have icreased. Through the relaxatio of the IIA property, the MMNL model eables the model to be specified i such a way that the error compoets i differet choice situatios from a give idividual are correlated (McFadde ad Trai 2000). With the use of MMNL modelig, Hess ad Polak (2005a,b) modeled the choice of airport by air passegers departig from Sa Fracisco, with attributes of groud access modes icluded i the models as explaatory variables. Their model results idicated that there are sigificat differeces across air passegers i their sesitivity to various factors. I additio, the MMNL models lead to importat gais i modelig accuracy ad explaatory power i the aalysis of air travel behavior. The CNL models allow for joit represetatio of correlatio alog the choice dimesios, without requiremets to use a multi-level estig structure. It allocates a fractio of each alterative to a set of ests with equal logsum parameters across ests (Vovsha 1997). Hess ad Polak (2006) applied CNL modelig to joitly aalyze air passeger choices of airport, airlie ad access mode i the Greater Lodo area. They foud that CNL model offers sigificat improvemets over simple MNL ad NL models, ad ca serve as a valuable tool i airport choice modelig. 50

3 Besides travel time ad travel cost, Nam et al. (2005) poited out that whe makig mode choice decisios, travel time reliability also plays a importat role. This is because lower travel time reliability results i late arrival ad imposes a potetially high travel cost o the travelers. Noetheless, due to o-recurret traffic cogestio, travelers caot predict the exact travel time required. As a result, whe makig a travel mode choice, travelers allow extra time, geerally referred to i the literature as a safety margi, i order to avoid late arrival. Owig to arrival time pressure to meet scheduled departure flight times ad the high persoal pealty for missig flights, allowig a adequate safety margi is particularly importat to departig air passegers. I Tam et al. (2008), the safety margi for airport groud access jourey is defied as the differece betwee air passegers preferred arrival time at the airport passeger termial for check-i ad their expected arrival time. Their results showed that the safety margi allowaces have some impacts o airport groud access mode choice decisios, while busiess departig air passegers place a sigificatly higher value o safety margi for their groud access trips tha o-busiess oes. I additio to the effects of the safety margi allowace, the relative itesity of the relatioship betwee satisfactio level ad airport groud access mode choice behavior should also be cosidered. This is because, for service providers, uderstadig exactly what customers expect ad their satisfactio level is the most importat step i defiig ad deliverig the quality service (Zeithaml et al. 1990). However, the service quality is difficult to measure quatitatively because of its itagibility, iseparability ad heterogeeity. Therefore, McFadde (1986) proposed to iclude latet psychological variables, which measure the service performace ad traveler satisfactio with the service, i choice models. The iclusio of the latet psychological variables ca lead to a more behaviorally realistic represetatio of the choice behavior, ad cosequetly, higher explaatory power. A umber of researchers, icludig Kitamura et al. (1997), Morikawa et al. (2002), Sasaki et al. (1999), ad Tam et al. (2010) cosidered the suggestio made by McFadde (1986) ad itroduced attitude factors i the choice model as explaatory variables. They foud that iteractive, two-way relatioship exist betwee attitudes ad behavior. From the above literature review, two importat observatios ca be made. Firstly, both travel time reliability ad perceived service quality were foud to be importat factors affectig air passegers airport groud access mode choices. Secodly, there is a eed to stregth the uderstadig of air passegers airport groud access mode choice i Asia coutries with a high market share of public trasport for their air passegers do exit. Thus, i this paper, the groud access mode choice behavior at HKIA is studied. As the statistics show that the total umber of air passegers usig HKIA has reached 49.7 millio i the year 2010 (Hog Kog Civil Aviatio Departmet 2011), ad this figure is expected to be cotiuously icreasig. Similar to Tam et al. (2010), a attitudial variable regardig satisfactio level towards the mode choices is first costructed usig Structural Equatio Modelig (SEM), ad the itroduced ito the MNL mode choice model as a explaatory variable. Other explaatory variables to be cosidered i this study iclude the safety margi allowaces, travel time, travel cost, ad some traveler characteristics such as residetial status ad age. It is believed that the itegrated model eables the evaluatio of the relatioship betwee satisfactio level ad airport groud access mode choice behavior, ad a more reliable represetatio of air passeger groud access mode choice behavior ca be idetified. Hece, improvemet ad expasio o the existig HKIA trasportatio system ca 51

4 be grouded for copig with the icreasig demad for airport related traffic. The fidigs from this study ca be trasferred to coutries which have airport related groud trasportatio problems for further cosideratios. 2 Hog Kog Iteratioal Airport HKIA, located i the orth of Latau Islad, is approximately 28 km from the Cetral Busiess District (CBD) that comprises Cetral ad Tsim Sha Tsui (Figure 1). To esure efficiecy of travel to ad from HKIA, the North Latau Highway ad the Airport Express (AE) Lie were built. The North Latau Highway provides a railway ad roadway access to Latau Islad ad the airport from urba Kowloo ad Hog Kog. The Highway carries six traffic laes, three i each directio, o the upper deck ad two railway tracks o the eclosed lower deck, oe i each directio. I additio, two traffic laes are provided o the lower deck for emergecy use, such as i severe weather coditios. The AE Lie covers 35.3 km betwee HKIA ad Cetral (i.e., Hog Kog Statio), with two itermediate stops at Tsim Sha Tsui (i.e., Kowloo Statio) ad Tsig Yi (i.e., Tsig Yi Statio) (Figure 1). Five major groud access mode types to HKIA, the AE, buses, taxis, private cars, ad courtesy vehicles comprisig hotel vehicles ad tour coaches have bee idetified. Both AE ad buses are scheduled public trasport modes ad charge a predetermied fare. The AE, operated by the Mass Trasit Railway (MTR) Corporatio, is a fastdedicated railway, likig the airport with the CBD. Over 20 bus routes operate, with two to three servig each of the major regios/districts i Hog Kog. Each of the bus routes has a specific travel path, ad the buses stop at some pre-determied locatios for passeger boardig ad alightig. Taxis provide a coveiet persoalized poitto-poit trasport service, with fares typically calculated accordig to trip legth by a Figure 1 Locatio of Hog Kog Iteratioal Airport 52

5 taximeter ad accordig to rates established by the govermet. Private car users, i additio to fuel cost, have to pay toll fees for the Latau Lik ad all other tuels used. To attract private car users, the Airport Authority Hog Kog (AAHK) offered 30 miutes free parkig at HKIA, with each additioal hour charged at US$2.5. Courtesy vehicles are available to hotel guests ad the public, ad ca be classified ito two types, scheduled service ad private limousie service. The scheduled courtesy vehicles ru i 30 miutes itervals, while private limousie service is available upo the requests of passegers. Table 1 summarizes the characteristics of differet groud access modes for HKIA. For easier compariso, the figures preseted are based o travelig betwee Cetral ad HKIA. Although HKIA is a critical hub for iteratioal passegers ad cargo, oly a few studies o the groud access mode choices to HKIA exist. Thus, to cater for the growth of airport groud access traffic ad facilitate actios for further developmet of the groud trasportatio system at HKIA, this study bega from the fudametal level of iterview surveys. 3 Data Collectio A two-wave modal split survey was carried out aimig to collect ecessary data for determiig the HKIA groud access modal split patter ad factors affectig such patter. The first wave modal split survey was carried out betwee 30 Jue 2004 ad 2 July 2004, while the secod wave survey commeced o 1 May 2005 ad eded o 3 May These two periods were selected because they are the peak air passeger traffic periods. The two-wave modal split survey allows comparisos betwee differet seasos, high summer ad sprig. The surveys targeted departig air passegers whose groud access trip bega from the Hog Kog territory. Such passegers face greater arrival time pressures, owig to the ecessity to meet scheduled flight departure times ad also the travel time ucertaity for accessig the airport. These air passegers have greater eed for reliable trasportatio to the airport. It was also felt that, more detailed iformatio could be obtaied from departig air passegers, whose groud access trip was already complete. Table 1 Characteristics of Differet Groud Access Modes (for Travelig betwee Cetral ad HKIA) Mode Travel cost (US$ a ) Travel time b (miutes) Headway (miutes) Hours of operatio Airport Express :50 to 01:15 Bus c 05:20 to 00:30 Taxi Private car d Courtesy vehicle (scheduled service) :00 24:00 a US$1.00 = HK$7.80. b Travel time required is estimated by the bus operator, the MTR Corporatio ad the Hog Kog Trasport Departmet. c Headway for a particular bus lie, where headway = 60 / frequecy per hour. d This icludes the toll fee for Latau Lik (US$3.85), the toll fees for cross harbour tuels (ragig from US$2.56 to US$6.41) ad the estimated fuel cost (US$12.82). Parkig cost is excluded. 53

6 As suggested by Metropolita Trasportatio Commissio (2003), the legth of the air jourey is expected to correlate with the umber of pieces of baggage carried by a traveler ad hece with mode choice decisios. Therefore, stratified samplig was used i the surveys to esure both log-haul ad short-haul travelers were sampled. The flights were stratified ad selected proportioal to the legth of the air jourey to the first destiatio. The list of scheduled flights was dowloaded from the HKIA website. The flights were classified ito three major groups, (1) log-haul flights (air jourey time of more tha 6 hours), (2) short-haul flights departig to Mailad Chia, ad (3) short-haul flights departig to other coutries, such as Taiwa, Sigapore ad Japa. The departure time of the selected flights was sufficietly well-spaced i time (i.e., 1 to 1.5 hours apart), so that, for each selected flight, a reasoable umber of departig air passegers could be iterviewed. Departig air passegers waitig to board the aircraft at the boardig gates were ivited for iterview. Seated air passegers were selected to esure passeger comfort durig the 10 miutes take to complete the questioaires. A systematic approach was used to select the respodets. The first perso (from left to right) sittig i each row of seats ext to their respective boardig gates was iterviewed. Sice air passegers should arrive at the boardig gates at least 30 miutes before the scheduled flight departure times, there was sufficiet time to complete the questioaire. From a total 891 eligible respodets, i the first wave modal split survey, there was a 56 percet respose rate. From a total of 963 eligible respodets iterviewed i the secod wave modal split survey, there was a 58 percet respose rate. A total of 994 resposes (475 from the first wave survey ad 519 from the secod wave survey) were useful for aalyzig departig air passeger mode choice behavior for accessig HKIA. 3.1 Questioaire Desig The questioaires used i the two-wave surveys are slightly differet. The questioaire used i the first wave modal split survey icludes three mai parts. I the first part, iformatio about the groud access trip to HKIA, icludig trip origi, travel time ad travel cost of the chose mode ad alteratives available to the respodet, party size ad umber of pieces of baggage carried by the travel party, was requested. Respodets were asked to rate their satisfactio level o five selected service attributes of the groud access modes that they have ever used for accessig HKIA. The satisfactio level is rated usig a 5-poit Likert scale. The five service attributes icluded waitig time, i-vehicle travel time, travel time reliability, travel cost ad walkig distace to ad from public trasport statios ad/or car parks. These five service attributes are critical factors affectig air passeger satisfactio level toward the mode choices. The secod part was SP scearios. The aim of the SP questios was to determie how travel time reliability affects departig air passeger airport groud access mode choices. Four attributes, icludig mode choice, travel cost, travel time, ad travel time reliability, were icluded i the SP scearios. Mode choice has five levels, while each of the other three attributes has three levels, with oe level better tha, oe level same as, ad oe level worse tha the curret situatio for accessig HKIA. I the SP scearios, travel time reliability is represeted as the fractio of trips with uexpected delays of 15 miutes or more. It has bee show that survey respodets uderstood this represetatio of travel time reliability more easily (Browstoe ad Small 2005; Small et al. 2005). Usig the techique of orthogoal desig, two sets of 25 profiles 54

7 are geerated usig SPSS. A radom umber is assiged to each profile, ad these radom umbers are used to pair the profiles for compariso. However, it was foud that some of the pairs cosist of a mode choice better perform i all aspects. I additio, some of the choice pairs are rarely faced by the respodets i the reality. As a result, these pairs are dropped, so as to make the compariso as realistic as possible. A total of 12 choice pairs remaied ad icluded i the fial questioaire. Durig the iterview, respodets were asked to view two pairs of profiles, ad choose oe of the two alteratives each time accordig to his/her preferece. Table 2 illustrates oe of the SP scearios preseted i the survey. I the third part of the questioaire, demographic iformatio, such as trip purpose, geder, age ad residetial status of the air passegers, was requested. Based o the resposes obtaied from the first wave modal split survey, travel time reliability was foud to be a critical factor affectig air passeger mode choices. Therefore, i the secod wave modal split survey, questios relatig to the safety margi allowace were added so as to examie departig air passeger perceived travel time variability of various groud access modes. To determie the magitude of safety margi allowed for airport groud access joureys, i the secod wave modal split survey, the respodets were first asked to state the expected travel time required for usig differet modes, the earliest ad latest time for havig check-i at the airport termial, the their preferred arrival time at the airport termial such that they would feel comfortable ad safe. The differece betwee air passegers preferred arrival time at the airport passeger termial for check-i ad their expected arrival time is the safety margi allowed for the airport groud access jourey. Figure 2 illustrates the time elemets asked i the secod wave modal split survey. The safety margi obtaied from these questios were reasoable, as departig air passegers must have a preferred arrival time at the airport before the check-i couters cease acceptace of passegers. Other questios asked i the secod wave modal split survey were the same as those asked i the first wave. However, i order to keep the questioaire short ad icrease the respose rate, questios relatig to passeger satisfactio o groud access services ad the SP questios were discarded i the secod wave modal split survey. Table 2 Example of Stated Preferece Sceario Icluded i the First Wave Modal Split Survey Alterative A Alterative B Mode Airport Express Bus Total travel cost (US$) Total travel time required (miutes) Frequecy of beig late by at least 15miutes 0 of 10 trips 4 of 10 trips 55

8 Expected Travel Time Safety Margi Departure Time from Origi Expected Arrival Time Preferred Arrival Time Scheduled Flight Departure Time Latest Check-i Time (i.e. Closure of Check-i Couters) Figure 2 Graphical Represetatio of Time Elemets Requested i the Secod Wave Modal Split Survey 3.2 Survey Fidigs It was foud i the two-wave modal split survey that of the total 994 respodets, a majority (77 percet) used a sigle mode to access HKIA, ad the remaiig 23 percet used a combiatio of modes. I order to have a reasoable sample size for each mode type, i this study, the access mode refers to the fial mode used by a respodet, as this mode is directly related to the desig ad improvemet of the existig HKIA groud trasportatio system ad the departure curb. Public trasport (icludig AE ad buses) domiates the HKIA groud access market. Buses have a large proportio, 44 percet, while AE has 25 percet. The questioaire resposes revealed low travel cost to be the primary reaso attractig respodets to usig buses, while high travel time reliability was the major reaso for those usig AE. As HKIA is far from the urba area (i.e., 28 km from the CBD), access by taxis icurs a much higher travel cost. This cosequetly leads to a smaller market share by taxi, with oly 14 percet of the airport groud access market. The use of private cars as the groud access mode to HKIA is limited, oly accouts for seve percet of the airport groud access market, possibly owig to the low car owership rate, five percet, i Hog Kog (Hog Kog Trasport Departmet 2005). Based o the results of the secod wave modal split survey, it was foud that busiess ad log-haul departig air passegers accessig HKIA allowed a larger safety margi tha the air passegers with other agedas. Those trasported by buses also ted to allow a large safety margi as the travel time variatio of buses is comparatively larger tha the other airport groud access modes. More detailed survey results regardig the modal split patter at HKIA ad the safety margi allowed by departig air passegers ca be foud i Tam et al. (2008). I the first wave modal split survey, the respodets were asked to rate their satisfactio level o five selected service attributes icluded waitig time, i-vehicle travel time, travel time reliability, travel cost ad walkig distace to ad from public trasport statios ad/or car parks, usig a five-poit Likert scale. Over 90 percet of the respodets reported that they had ot ecoutered ay traffic cogestio o their jourey to HKIA, hece a relatively high satisfactio o travel time reliability is obtaied. Waitig time is the least satisfactory attribute for bus users, while users of other modes are least satisfied with the travel cost. I geeral, departig air passegers perceived the best service from taxis, ad it obtaied relatively high satisfactio ratigs i compariso with other airport groud access modes. The summary of passegers satisfactio ratigs o the five service attributes is preseted i Tam et al. (2010). Regardig the SP samples, a higher proportio of the respodets chose those 56

9 alteratives with a low probability of uexpected delays. However, if the travel cost of a mode was much higher tha that of a alterative, it is likely that the cheaper travel mode regardless of the higher probability of uexpected delays would be chose. It was clear from the SP resposes that travel time reliability has sigificat impact o departig air passeger mode choice decisios. Usig the revealed preferece () resposes collected i the first wave survey ad SEM approach, a satisfactio latet variable is costructed to capture departig air passeger prefereces o the five airport groud access modes idetified. A MNL mode choice model is the calibrated with the use of both ad SP survey data collected from both waves of the modal split survey. Combiig the ad SP data allows elimiatio of the potetial biases caused by (1) strog correlatios amog variables, such as travel time ad travel cost, ad (2) the habitual preferece towards the choice whe respodig to SP questios. The satisfactio latet variable is icluded i the mode choice model as a explaatory variable. 4 Modelig Framework 4.1 Itegratio of Latet Variable Model ad Discrete Choice Model I this study, the framework applied to icorporate latet variable ad the discrete choice model is similar to the oe demostrated by Morikawa et al. (2002). The framework cosists of two compoets: (1) a discrete choice model ad (2) a latet variable model. Both of these models cosist of structural ad measuremet equatios. ad SP resposes are described by the discrete choice model while the latet variable is described by a multiple idicators ad multiple causes (MIMIC) model. The MIMIC model is a special case of the geeral structural equatio models ad has oly oe latet variable. The latet variable is directly affected by oe or more idepedet observable variables, ad it is idicated by oe or more idicators. The MIMIC model is a alterative to multiple group models where there is more tha oe groupig variable ad a larger sample size is required (Bolle 1989; Boomsma 1982; Muthe 1989). The MIMIC model is applied i this study as, based o the collected survey data, it was foud that the sample size of usig each access mode, particularly those of private cars ad courtesy vehicles, is limited. Thus, to avoid o-covergece ad improper estimatios due to the bias of small sample size, the coceptual framework of the latet variable model is a MIMIC model. I Figure 3, the ellipses represet uobservable variables ad rectagles represet observable variables. Solid arrows represet the structural equatios while dashed arrows represet the measuremet equatios. The latet variable model, which is the same as the oe preseted i Tam et al. (2010), describes the relatioships betwee the latet variable ad its idicators ad causes. The discrete choice model explais departig air passeger mode choices. For the sake of simplicity, a MNL model itegrated with a latet variable model is preseted. I the equatios below, superscripts ad SP deote the correspodig data. The structural equatios are represeted as: U = β X + α η + ε (1) SP SP SP U = β X + ω y + ε (2) η = γx + ζ (3) 57

10 i which U is the utility of alterative. X is a vector of observable explaatory variables, icludig the characteristics of the modes, of the trip ad of the traveler. η is the latet variable regardig the satisfactio of departig air passegers o various airport groud access modes. is the mode choice idicator, it is equal to 1 if alterative y is chose ad 0 otherwise. β, α, ω ad γ are vectors of um at ukow param eters. ε ad ζ are the radom disturbace terms. It is ass ed th ε ad ζ are idepedetly logistic distributed. This yields the familiar logit formula for the choice probability coditioal o parameters. The measuremet equatios are represeted as: 1, if U = max{ U } m y = m 0, otherwise (4) SP SP 1, if U = max{ U } SP m y = m 0, otherwise (5) I = λη + δ (6) Observable Explaatory Variables X Satisfactio Latet Variable η Idicators I Latet Variable Model Utility U Utility SP U Choice y Choice SP y Discrete Choice Model Figure 3 Itegrated Discrete Choice ad Latet Variable Model 58

11 where I is a vector of observable idicators of η. λ is a vector of ukow parameters for η, ad δ is a vector of radom disturbace terms. ζ ad δ are assumed to be mutually u-correlated. I this study, Equatios (1), (3) ad (4) form the choice model. The SP choice model composes Equatios (2) ad (5). It is oted that people usually display time icosistecy i the actual ad stated cotexts (Small et al. 2005), while travel time is a compoet determiig the value of the satisfactio latet variable. Therefore, to reduce the potetial error, the satisfactio latet variable is excluded i the SP choice models. As suggested i Morikawa et al. (2002) ad Small et al. (2005), the SP resposes may be biased towards the travel mode which has bee selected i the actual choice cotext. Therefore, a iertia dummy variable y, which idicated the actual choice, is icluded i Equatio (2) to capture the correlatio betwee the ad SP resposes. Equatios (3) ad (6) costruct the latet variable model. SP I combiig the ad SP data, the variaces of ε ad ε are usually allowed to be differet so as to distiguish differet sources for radom prefereces over the revealed ad stated choices. This ca be accomplished by ormalizig the variace of, which is represeted as: ε μ / σ σ SP (7) where μ is the -SP scale parameter, ad each σ is the stadard deviatio of the correspodig ε. The questios asked i both waves of the modal split surveys were geerally the same, except i the secod wave survey, questios regardig safety margi allowaces for the airport groud access joureys were asked. Hece, it is assumed that the variaces of ε betwee the two waves are the same, ad o scale parameter is required. However, as departig air passegers are expected to react differetly to travel time ad travel cost i a actual ad i a hypothetical situatio, the parameters of the variables varyig by modes are allowed to differ across the ad SP data, whereas the other variables have the same effect across the two data sets (Mark ad Swait 2004; Small et al. 2005). This approach provides a more precise estimatio of departig air passeger airport groud access mode choice behavior. I this study, a sequetial estimatio method is employed to calibrate the ukow parameters of the itegrated model because it is less cumbersome to estimate the model sequetially tha simultaeously (Johasso et al. 2005). LISREL (Joreskog ad Sorbom 1993) was employed for the developmet of the MIMIC model with the use of first wave modal split survey data. Resposes to the attitudial questios of the first wave survey are listed as idicators of the satisfactio latet variable. The structural relatioships betwee departig air passeger satisfactio ad their observable characteristics are therefore determied. Based o the calibrated structural equatios of the MIMIC models, departig air passeger satisfactio o various airport groud access modes is quatified. The satisfactio latet variable is the icluded i the discrete choice models as a explaatory variable. The calibrated structural equatios of the MIMIC model are the used to estimate the value of satisfactio latet variable for the secod wave survey respodets. The satisfactio latet variable is icluded i the MNL model, which calibrated with the use of NLOGIT3.0 59

12 (Greee 2002). The icorporatio of the satisfactio latet variable ca lead to a more robust represetatio of departig air passeger choice processes. 4.2 Exogeous Sample Weights Although stratified samplig was employed i the surveys, it was foud that there is differece betwee the profile of the respodets ad that of the populatio, regardig their trip purposes ad residetial statuses. As show i Table 3, i the two-wave modal split survey, Hog Kog residets, particularly those who traveled for busiess, were over-sampled. I cotrast, visitors who traveled for o-busiess purpose were uder-sampled. Thus, to obtai cosistet estimates i the mode choice models, each idividual observatio is weighted by the fractio of populatio belogig to a segmet over the correspodig fractio of the respodets (Garrow et al. 2005). I this study, the departig air passegers are classified ito four segmets, (1) busiess Hog Kog residets, (2) o-busiess Hog Kog residets, (3) busiess visitors, ad (4) o-busiess visitors. The populatio fractios of departig air passegers belogig to each of these four segmets were obtaied from the AAHK, while the correspodig fractios of the respodets were calculated from the collected survey data. The samplig weights are applied to the process of the model calibratio so as to elimiate the estimatio errors raised from the samplig bias. 5 Calibratio Results The latet variable model preseted i Tam et al. (2010) cosists of five idicators icludig waitig time, i-vehicle travel time, travel time reliability, travel cost ad walkig distace to ad from public trasport statios ad/or car parks. I additio, four observable explaatory variables, icludig oe characteristic of the modes (i.e., travel time) ad three traveler characteristics (i.e., geder, age ad educatio level), were idetified as the causes of the satisfactio latet variable. As this paper emphasizes the results of the itegrated model, the fidigs of the latet variable model are ot discussed, but it ca be foud i Tam et al. (2010). As metioed i Sectio 4, travelers usually react differetly to travel time ad travel cost i a actual ad i a hypothetical situatio, ad their values of time (VOTs) ad values of trasfer varied i accordace with differet circumstaces. Therefore, i the itegrated model, the parameters of the variables varyig by modes are allowed to differ across the ad SP data, whereas the other variables have the same effects across the two data sets. This approach provides a more precise estimatio of departig air passeger airport groud access mode choice behavior. Table 3 Profile of the Populatio ad the Sample Segmet Populatio share a Sample share Busiess Hog Kog residets No-busiess Hog Kog residets Busiess visitors No-busiess visitors a Figures as at 2004 (aual data) from the Airport Authority Hog Kog. 60

13 I additio, to accomplish the differet sources for radom prefereces over revealed ad stated choices, a -SP scale parameter ( μ ) has to be icluded i the itegrated model estimatio. The -SP scale parameter of maximized the model loglikelihood, hece it is selected ad applied for estimatig the model. The results of the itegrated model, which calibrated usig the 994 resposes obtaied from the two-wave modal split survey, are give i Table 4. Most of the parameters are sigificatly differet from zero at the 0.05 level of sigificace ad have the expected sig. It is reassurig that the estimated coefficiets i the itegrated model basically maitai the relative values obtaied i Tam et al. (2008) (also show i the last colum of Table 4). The safety margi, which referred to as the additioal time allowed by the departig air passegers i aticipatio of travel time ucertaity for their groud access to HKIA, is egatively related to the mode utility. This is because departig air passegers ted to select a mode which they perceive to be more reliable, to miimize the chace of ecouterig uexpected delays. Based o the model results, the value of safety margi (VOSM) for departig air passegers is US$0.11 per miute. I additio to safety margi allowace, departig air passegers trade-off travel costs with travel time ad umber of trasfers required for accessig HKIA. The model results showed that busiess departig air passegers are substatially more averse to travel time isecurity tha o-busiess oes. The former are willig to pay for a higher cost to reduce the travel time required for accessig the airport. The VOTs for busiess ad o-busiess passegers are US$0.27 per miute ad US$0.11 per miute respectively. Similar fidigs were revealed by Harvey (1986), Pels et al. (2003) ad Tam et al. (2008). The VOTs for busiess ad o-busiess departig air passegers obtaied from Harvey (1986) were US$0.69 per miute ad US$0.33 per miute respectively. Pels et al. (2003) foud that the VOTs for busiess departig air passegers raged from US$1.97 per miute to US$2.90 per miute ad raged for o-busiess departig air passegers from US$1.58 per miute to US$1.60 per miute. I Tam et al. (2008), the VOTs for busiess departig air passegers were US$0.25 per miute, ad that for o-busiess departig air passegers were US$0.10 per miute. The higher VOTs obtaied by Harvey (1986) ad Pels et al. (2003) may be due to travelers beig trasported by private cars tedig to have a higher VOT tha those travelig by other modes. Similar to the results obtaied i Tam et al. (2008), both models revealed that departig air passegers have a lower safety margi value tha for travel time. This is because departig air passegers, who allowed a large safety margi, arrivig at HKIA earlier tha expected, ca oly sped their uused safety margi withi the airport passeger termial. It was oted that the SP VOTs for busiess ad o-busiess departig air passegers are US$0.25 per miute ad US$0.08 per miute respectively, which is smaller tha the values. The differeces arise because people ted to overstate the actual travel time they experieced i the revealed choice cotext. Hece, they are more resposive to a give actual time savig tha to a hypothetical time savig of the same amout, ad yield a smaller VOT i the SP tha i the circumstaces. By ormalizig the variaces betwee the ad SP data, the value of travel time reliability (VOR), which equals US$1.66 per icidet, is obtaied from the itegrated model. 61

14 Table 4 Parameter Estimates for Multiomial Logit Model I this study I Tam et al. (2008) Coefficiet (t-statistics) Coefficiet (t-statistics) variable Alterative specific costat for AE (1.104) (-0.369) Bus (5.184) (2.451) Taxi (2.111) (-5.265) Private car (3.346) (1.275) Cost ( ) ( ) Time (-2.424) (-5.111) Time(Busiess) (-2.653) (-4.077) Safety margi (-4.779) (-2.642) Trasfer (-2.220) (-2.727) Satisfactio (2.798) SP variable Alterative specific costat for AE (-0.249) Bus (-0.534) Taxi (0.694) Private car (1.724) Cost (-2.555) Time (-0.713) Time(Busiess) (-1.984) Ureliability of travel time (-2.003) Trasfer (-1.539) Pooled variable Party size (specific to AE) (-1.050) (-1.039) Baggage (specific to AE) (-3.164) (-3.199) Party size (specific to bus) (-2.644) (-2.255) Baggage (specific to bus) (-2.352) (-1.784) Log-haul (specific to bus) (-3.369) (-5.917) Age 25 (specific to bus) (0.888) (3.545) HK (specific to AE ad bus) (4.108) (2.955) Age 65 (specific to taxi, private car ad courtesy vehicle) (2.064) (3.101) Other parameter Iertia dummy (6.993) -SP scale parameter ( μ ) (4.710) * * The t-statistics correspodig to the -SP scale parameter is computed with respect to a value of 1; a value of 1 idicates o scale differece i the ad SP choice cotexts. The VOTs obtaied from the SP-based MNL model are compared with those obtaied i the Travel Characteristics Survey (TCS) coducted by the Hog Kog Trasport 62

15 Departmet i I the TCS, SP questios were asked with the aim of ivestigatig the VOT of the populatio regardig their daily travel. The overall VOT deduced from the TCS is US$0.07 per miute (Hog Kog Trasport Departmet 2003), which is lower tha those values obtaied i this study. This result idicates that owig to the higher pealty for arrivig late at HKIA, departig air passegers, particularly those travelig for busiess, are willig to pay a higher cost for accessig the airport tha for their daily travel. The satisfactio latet variable has a positive estimated parameter. This implies that, as expected, the mode utility icreases with passeger satisfactio level. More importatly the parameter estimated for the satisfactio latet variable has a larger magitude compared with other variables. This implies that the satisfactio latet variable domiates other variables, which led support to the otio that satisfactio level is associated with departig air passeger airport groud access mode choice behavior. It was further oted that the iclusio of the satisfactio latet variable 2 sigificatly improves the model calibratio results, with the likelihood ratio test χ = 7.56 give a chi-square statistic 3.84 with oe degree of freedom at the 0.05 level of sigificace. This idicates that the iclusio of the satisfactio latet variable ito the mode choice model provides a better estimatio of the departig air passeger groud access mode choice behavior. It is further revealed that, i the sub-model, the coefficiet of the travel time variable becomes less sigificat, as compared with Tam et al. (2008), probably due to multicolliearity. This is caused by the reaso that the variable of travel time is icluded i both the structural equatios of the MIMIC model ad the MNL mode choice model. I the itegrated model, the trip ad traveler characteristics affectig alterative specific prefereces are assumed to have the same effects across the ad SP data. The results of the itegrated model show that party size egatively affects the usage of AE ad buses. This is because these two modes charged a fixed per-perso cost. Thus, departig air passegers travelig i groups ted ot to select these two modes but use taxis ad private cars, for which the travel cost ca be shared amog the travelig party. The umber of pieces of baggage carried by a departig air passeger also has a egative impact o the utilizatio of AE ad buses. This ca be partly explaied by the fact that these two modes offer less direct services, ad departig air passegers carryig more ad/or larger baggage require greater physical effort to walk to the public trasport statios ad are thus less likely to use these two modes. Comparatively, the umber of pieces of baggage has a smaller impact o the usage of buses tha that of AE. This is because departig air passegers sometimes give preferece to the buses particularly whe the bus stops are close to the departure poit at their origis, thus they avoid trasferrig betwee modes as whe usig AE. Departig air passegers havig log-haul trips are less likely to access HKIA by buses due to the ureliability of travel time perceived by departig air passegers. However, those aged below 25 have a higher tedecy to use buses owig to the comparatively low travel cost. Hog Kog residets are believed to be more familiar with the local trasport system, ad thus have a higher tedecy to be trasported by AE ad buses. The itegrated model further revealed that elderly (i.e., aged above 65) preferred opublic modes (icludig taxis, private cars ad courtesy vehicles) as these modes offer poit-to-poit ad comfortable access services to HKIA. 63

16 The iertia dummy variable, which idicates the impact of the choice o the SP choice, is also estimated precisely. The positive coefficiet of the iertia dummy variable idicated that, whe respodig to the SP scearios, the respodets have a habitual preferece towards their choices. The -SP scale parameter ( μ ) of value idicates the error variace of the SP data to be much higher tha that of the data. 2 The high likelihood ratio idex ( ρ =0.72) represets a good fit for the model. Aroud 80 percet of all the observatios, i which 84 percet (836 out of 994) observatios ad 76 percet (731 out of 956) SP observatios were predicted correctly by the itegrated model. As the itegrated model provides reliable estimates for both the ad SP choice cotexts, as well as havig a high predictio power, it is recommeded to the AAHK ad various trasport operators that it ca be used as a dialogistic tool for evaluatig the impacts of chages of the service attributes o modal split patter at HKIA. The sub-model allows predictio of modal split patter uder actual choice cotexts, whereas the SP sub-model permits to ivestigate the impact of travel time reliability o departig air passeger mode choice decisios. Hece, the AAHK ca improve the termial frotage desig ad various trasport operators ca also ehace their services more effectively. 6 Policy Implicatios I respose to the airport-geerated cogestio problems, govermets have iitiated a icrease i the use of public trasport for accessig airports. For example, it was suggested that icreasig bus frequecies, providig ew bus routes ad developig rail liks to airports, ca reduce car depedecy for groud access trips to the Uited Kigdom airports (Humphreys ad Iso 2003). To propose strategies for coordiatig HKIA groud access problems ad further icreasig the utilizatio of public trasport (i.e., AE ad buses), sesitivity tests are performed with the use of the calibrated model. Hece, suggestig the ecessary operatioal restructurigs. 6.1 Improvemet i Bus Services The survey results idicated that departig air passegers are least satisfied with the travel time reliability of buses, ad hece have to allow a large safety margi for accessig HKIA. Thus, for bus operators, they should focus o measures to improve the travel time reliability, hece reduce the required safety margi. Figure 4 shows the chages of bus market shares i respose to chages i safety margi allowace ad travel cost. It was foud that departig air passegers are more sesitivity to safety margi allowace tha travel cost for joureys made by buses. If the bus operators ca improve travel time reliability of buses ad subsequetly reduce departig air passeger safety margi allowace by 10 percet, a bigger market share would be resulted. The bus operators ca also icrease the bus fares by at most 20 percet i order to maitai its market share. Oe of the ways to improve bus travel time reliability is to make ivestmet i iformatio techology ad provide passegers up-to-date iformatio o departure times ad arrival times of airport buses at major bus stops. Hece, departig air passegers ca adjust their departure times ad reduce the safety margi allowed for 64

17 Figure 4 Chages i Bus Market Share i Respose to Chages i Safety Margi ad Travel Cost the airport groud access joureys, accordigly. The results from the sesitivity test suggested that the bus operators may trasfer the ivestmet cost partially to passegers by icreasig the fare charged. 6.2 Improvemet i Airport Express Services Rather tha attemptig to improve the already satisfactory travel time reliability, it could be advisable for the MTR Corporatio to cosider providig discouts to the AE users as survey respodets respoded that the curret charges of usig AE are too high, with a average travel cost US$12.74 per perso. By assumig (1) the daily umber of departig air passegers at HKIA is 30,000; (2) the groud access modal split of the populatio is the same as that revealed by the surveys; ad (3) the distributio of party size of the populatio is the same as that of the survey samples, Figure 5 shows the reveue chage i respect to differet fare discout provided for departig air passegers travelig by AE i group of 2, 3, ad 4 or more. As show i Figure 5, the MTR Corporatio could maximize its reveue by offerig a 20 percet fare discout to departig air passegers travelig i groups of two or three, ad a 35 percet off the existig fare for those with four or more persos. This discout scheme would icrease the MTR Corporatio reveue by US$5,640 per day. By offerig fare discout to AE users, it is ot oly improvig the market share of AE ad its correspodig reveue. More tha that, it helps to lesse the cogestio problems o the road-way as well as at the departure curb. Hece, improvig the departig air passegers satisfactio level towards the airport groud access jourey. 65

18 6,000 5,000 Chage i Reveue (US$) 4,000 3,000 2,000 1, ,000-2,000 5% 10% 15% 20% 25% 30% 35% 40% Discout % Overall Party size =2 Party size = 3 Party size >=4 Figure 5 Reveue Chage agaist Fare Discout for Grouped AE Users 7 Coclusios I this paper, two cotributios have bee made to existig airport groud access mode choice models. The first is a assessmet of travel time reliability o air passeger airport groud access mode choice decisios. The secod cotributio is the icorporatio of air passeger perceived service quality i the calibratio of the mode choice model. I additio, revealed ad stated preferece data were combied to calibrate the mode choice model so as to allow elimiatio of the potetial biases caused by the strog correlatios amog variables. The habitual preferece towards the choice whe respodig to stated preferece questios ca also be distiguished. The model calibrated i this paper outperforms the oe developed i Tam et al. (2008), with a improvemet o the ability of correctly predictig the modal choices of departig air passegers. With the developed itegrated model, sesitivity aalyses have bee carried out to assess the effects of various proposed strategies to icrease the use of public trasport (i.e., Airport Express (AE) ad buses) for accessig Hog Kog Iteratioal Airport (HKIA). It was foud that if the travel time reliability of buses improved, a sigificat proportio of departig air passegers would switch to these buses for accessig HKIA. Offerig discouts to grouped AE users icreases the railway s utilizatio. Attractig more trips to airports by public trasport, particularly rail mode, is likely to reduce roadway cogestio caused by a excess of o-public modes. The calibrated model ca be served as a diagostic tool for evaluatig the chages i service attributes o airport groud access modal split patter i the short-term horizo, istead of a forecastig tool for log-term plaig purpose owig to the difficulty to estimate the future values of safety margis. I further studies, the relatioship betwee safety margis ad observable factors icludig travel time, trip 66

19 purpose, legth of the air jourey ad air ticket type, has to be calibrated. Thus, safety margi ca be icluded i the choice model as a explaatory variable, ad the model the ca be used for log-term plaig purpose. I the curret empirical settig, oly five service attributes regardig airport groud access modes are listed as idicators of the satisfactio latet variable. I further studies, more service attributes, such as ease of travelig with large baggage, safety ad relaxatio durig the trip, ca be used as the idicators of the latet variable. Hece, to obtai more detailed iformatio o the service quality perceived by departig air passegers regardig the airport gourd access trasportatio service. This would also help to idetify the areas that require service improvemets, ad add competitive advatages to the trasport operators. I total, 994 observatios were obtaied i the modal split surveys ad used to calibrate the model i this paper. Further studies are suggested so that a larger sample size with adequate represetatio of less-used modes, particularly taxis ad private cars, ca be collected. This would eable to quatify separately the effects of differet explaatory variables o taxis ad private cars. I additio, by collectig more samples, separate models ca be calibrated for departig air passegers with various trip purposes ad/or air ticket types so as to allow a more fie-graied idetificatio of distict sub-markets. Ackowledgemets The authors wish to thak Dr. Stephe W.K. Lam, Maager (Lad Trasport ad Commuicatios Divisio) of the Airport Authority Hog Kog, ad his team for their assistace, advice ad resources supplied for this study. Refereces Airports Coucil Iteratioal, ACI passeger ad freight forecasts flag shortage of airport capacity. Bolle, K. A., Structural Equatios with Latet Variables. Wiley, New York. Boomsma, A., The robustess of LISREL agaist small sample sizes i factor aalysis models. I: Joreskog, K. G. ad H., Wold (eds.) Systems uder Idirect Observatio: Causality, Structure, Predictio, Part I. North-Hollad, Amsterdam. Browstoe, D. ad K. A., Small, Valuig time ad reliability: assessig the evidece from road pricig demostratios. Trasportatio Research part A, 39(4) Clark, M. C. ad W. H. K., Lam, Airport groud access cosideratios. Paper preseted at the Airports ito the 21 st Cetury, Hog Kog, 5-7 February Garrow, L. A., F. S., Koppelma ad B. L., Nelso, Efficiet estimatio of ested logit models usig choice-based samples. I: Mahmassai, H. S. (ed.) Trasportatio ad Traffic Theory: Proceedigs of the 16 th Iteratioal Symposium o Trasportatio ad Traffic Theory, Elsevier, Amsterdam. Greee, W. H., LIMDEP Versio 8.0 Referece Guide. Ecoometric Software, Ic., New York. Harvey, G., Study of airport access mode choice. Joural of Trasportatio Egieerig ASCE, 112(5), Hess, S. ad J. W., Polak, 2005a. Mixed Logit Modelig of Airport Choice i Multi- Airport Regios. Joural of Air Trasport Maagemet, 11(2),