The city-airport connection in the low-cost carrier era: Implications for urban transport planning

The city-airport connection in the low-cost carrier era: Implications for urban transport planning José I. Castillo-Manzano * Facultad de Ciencias Económicas y Empresariales, University of Seville, Avda. Ramón y Cajal, 1, 41018 Seville, Spain Keywords: Airport ground access Urban transport planning Passenger behavior Low-cost airlines abstract This article, examines differences between the behavior of passengers of low-cost and network airlines when choosing their transport mode for travel to airports. It is found that a passenger flying with a lowcost carrier is 6% less likely to take a taxi to the airport, but more than 4% more likely to drive a rented car and 2% more likely to use public transport than a user of a network carrier. Ó2010 Elsevier Ltd. All rights reserved. 1. Introduction One of the features of air transport deregulation has been the development of low-cost carriers (LCCs) and their new management model, the main characteristics of which have undermined belief in the industry’s previous structure, procedures, and business models and brought about major changes in the strategies and behavior of the different economic agents on both the supply and the demand sides of the air transport industry. On the supply side, for example, there has been an upsurge in the use and importance of many secondary, often underused (Francis et al., 2004), regional airports compared to the hubs (Reynolds-Feighan, 2001). These airports have seen their bargaining power diminished by the aggressive bargaining methods of LCCs (Barrett, 2004) seeking to achieving minimum airport charges. Broadly speaking, competition between airports to attract LCCs can be said to be on the increase (Pels et al., 2009). LCCs’use of secondary airports also means airline trafficis further scattered across multiple airports serving the same metropolitan area. This is a drawback for airport access planning, the main purpose of which is to reduce the market share of private vehicle use by both passengers and airport staff (Humphreys and Ison, 2005 on policies for changing airport employee travel behavior), as private vehicles are the modes of transportation that most contribute to noise, congestion levels, and air pollution (Graham, 2008) in airport hinterlands. On the demand side, one of the most important changes brought about by the LCCs for our analysis is the increased demand for air transport services by younger and price-sensitive travelers (O’Connell and Williams, 2005). A priori this might be surmised to impact on the market shares of the various modes of transportation that passengers use to get to airports and benefit the less expensive. This would explain the interest that some LCCs in Europe have in either developing or working in cooperation with bus and coach companies (such as Terravision, tightly linked to the Ryanair group, or the Easybus connections to the airports around London where Easyjet operates). This paper looks at differences in the behavior of LCC and network airline passengers when choosing of a mode of transportation to an airport? And if these differences do exist, what are they, and what effects do they have on airport ground access planning? Apart from some theoretical considerations, there have been fewer empirical studies looking at these questions. 1 2. Data The database comprised 20,383 passengers, 6247 of whom were LCC passengers. All of these were interviewed in departure lounges *Tel.: þ34 678542833. E-mail address: [email protected] 1 Echevarne (2008) supposed that logic would dictate that low-cost passengers would generate a greater demand for car parking facilities at regional and secondary airports whilst traditional airline passengers would present a higher taxi usage. De Neufville (2006) similarly surmised that price-conscious travelers on lowcost carriers are the passengers most likely to use public transport, and that the market for airport access is now more closely aligned with the traditional market for urban public transport. Contents lists available at ScienceDirect Journal of Air Transport Management journal homepage: www.elsevier.com/locate/jairtraman 0969-6997/$ esee front matter Ó2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.jairtraman.2010.02.005 Journal of Air Transport Management 16 (2010) 295e298 at seven different Spanish secondary airports, none of which had efficient rail-based public transportation to the city at the time the survey campaign was conducted (Table 1). 2 The size of the sample and the large number of passenger attributes and characteristics provide an effective database for an analysis of the air transport industry in Mediterranean countries (see Castillo-Manzano and López-Valpuesta (in press) for another application of this database). ThemainfeaturesofAENA’s(theSpanish PublicAirportAuthority) 2005e2007 survey that is used to construct the database are listed in Table 1. AENA has rotated annual surveys round some of its airports since 1996. The methodology used has, therefore, finely honed and there has been sufficient funding to obtain samples that closely approximate to a random process for a large population and with a low sampling error. As with similar databases, each observation was weighted according to the number of passengers on the flight so that the sample could be expanded to represent the population. 3. Methodology Airport ground access systems play an important role in airport planning and management (Tsamboulas and Nikoleris, 2008) and are increasingly seen as a major problem worldwide. However, few studies have been done to date on the choice of means of airport access. Traditionally, travelers’choices of modes of transportation have been the object of microeconometric analyses using discrete choice models. Since the focus of earlier studies has been on airport ground access and possible differences in the decisions and behavior of passengers according to whether they are business or tourism/ leisure travelers it makes sense to complement this work with a study of differences in passengers’airport ground access mode choice behavior depending on whether they fly on low cost or network carriers. LCC growth is altering the classic distinction between business and tourist passengers. Firstly, there are generally no specific seats assigned to business passengers in these airplanes making it difficult to distinguish business travelers from other passengers on low-cost airline flights. Despite this, as the market share of LCCs rises, business passengers increasingly choose their services. Without rejecting discrete demand models, the proposed methodology is framed by statistical causal inference and based on an estimation of the effect that a specific measure or fact can have on one or more relevant variables. In contrast with traditional analyses, this methodology allows consistent estimators of the effects of the evaluated measure to be obtained (Rotnitzky and Robins, 1995) by determining and isolating the possible impact of additional contaminating variables. Starting with an N-size random sample (Table 2)wedefined the binary variable Dthat indicates whether the observation corresponds toapassengerflyingwithanLCC(D i ¼1)oratraditionalairline(D i ¼0). Thus, our Nobservations were divided into N 1 and N 0 observations (LCC vs. traditional airline). In our case, N 1 stands for the 6247 passengers who used an LCC, while N 0 represents the remaining 14,136passengers. Thus,theconditionthatstates that“N 0 isatleast the same order of magnitude ofN 1 ”is satisfied (Abadie and Imbens, 2006). We defined the outcome variable Y j as the decision to use a specific mode of transportation for a given city-airport transfer. In our case, we considered five possible modes of transportation: taxi, public bus, hotel bus, private car and rented car. Using the potential-outcome notation of the RCM (Rubin, 1974), the response variable was given as Y ij (1) when idenoted a passenger using an LCC and as Y ij (0) when icorresponded to a passenger flying on a traditional airline. Hence, Y ij was equal to Yij ¼DijYijð1Þþ1DijYijð0Þ(1) The Average Treatment Effect (ATE) of the LCC on the selected sample for each of the analyzed modes of transportation was (Imbens, 2004) a j¼EYijð1ÞYijð0Þ¼1 NX N i¼1Yijð1ÞYijð0Þ(2) We also defined a K-dimensional vector of observed covariates as X. A triad was therefore observed for each individual (D ij ,Y ij ,X ij ). The aim was to guarantee two conditions in the evaluation process. Firstly, the unconfoundedness condition (Rosenbaum and Rubin, 1983), also known as the conditional independence assumption,i.e.DtðYð1Þ;Yð0ÞÞjX. Secondly,werequiredcompliance with the overlap condition, according to which every value of vector Xis associated with a positive probability that allows (Hotz et al., 2005). The condition can be read as follows: 0 <PðD¼1jXÞ<1. To guarantee both conditions, the evaluation process conformed to the following phases (Heckman and Vytlacil, 2005). 3.1. Estimation of propensity score Firstly, we noted whether the observations corresponded to a passenger using an LCC (D i ¼1) or a legacy airline (D i ¼0). We then estimated the propensity score, defined by Rosenbaum and Table 1 Technical data survey. Airport Alicante Bilbao Seville Valencia Santiago Valladolid Zaragoza Airport traffic in 2008 9,578,308 4172,901 4,391,794 5779,336 1,917,434 479,716 594,952 Information gathering Questionnaire Available in 12 languages Available in 6 languages Available in 5 languages Available in 4 languages General Departing passengers >15 years of age. Sampling (before weighting) Sample size 2420 3182 4140 4965 3497 1042 1137 Sampling method Stratified by traffic segments with selections of flights for each route and groups of passenger participants done by systematic sampling. Sampling error a 2% 1.7% 1.5% 1.4% 1.5% 2.7% 2.5% Number of waves 1 Field work Time period Sept. 22e28 May 4e10 June 6e12 July 12e18 June 28eJuly 4 June 15e21 June 15e21 Location Departure lounges. Timetable MoneSun. 6ame10pm shifts extended during peak traffic periods. Year 2006 2007 2006 2006 2006 2005 2006 a Error ¼Kffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðNnÞ=ðNnÞ p ffiffiffiffiffiffiffiffiffiffiffi pq=n p;where: N¼population size; n¼sample size; p¼q¼0.5 complementary probabilities of the answer to an event at the point of greatest indeterminacy; k¼parameter for the level of answer to an event, where k¼2 for a 95.45% confidence level. 2 A subway line connecting Valencia to its airport was opened a few months after the survey was conducted. J.I. Castillo-Manzano / Journal of Air Transport Management 16 (2010) 295e298296 Rubin (1983) as the conditional probability of “participating in the evaluated measure,”given a vector Xof observed covariates. Different binary response models can be used to estimate the propensity score depending on the choice of hypothesis regarding the configuration of the Fdistribution function. In this case, we used the binary response model (e.g. logit or probit) that maximized the log pseudo-likelihood: 3 (X)¼P(D¼1jX)¼F( b X) (3) where b is the vector of parameters associated with X. Here, X comprised the 20 covariates presented in Table 2 along with their descriptive statistics. 3.2. Estimation of average treatment effect In a second phase, we calculated the average treatment effect of the measure being evaluated on the response variable, in our case, the probability of a passenger choosing a specific mode of transportation to get to the airport. The average effect on the selected sample was estimated using: a ¼E½ a ðXÞ (4) In our case, with Ybeing a discrete choice variable (using one of the five possible modes of transportation), we used a multinomial logit model to estimate the average treatment effect. Therefore, according to Hirano and Imbens (2001), the multinomial logit probability formula for a passenger iwhen a person chooses a mode of transportation jfor five category outcomes and frequency weights is pij ¼Prðyi¼jÞ¼(1=1þS5 m¼2eðx r i s m Þ;if j ¼1 eðx 0 i s m Þ=1þS5 m¼2eðx r i s m Þ;if js1(5) where: x’ i s m¼ s m0þ a Diþ s m1b 3 ðxiÞþ s m2b 3 ðxiÞEhb 3 ðxÞiDiþuij (6) Table 2 Covariates and their descriptive statistics. Variable Description Mean Std. Dev. a) Socio-demographic factors and employment status. Base category includes the unemployed. Sex 1 ¼male; 0 ¼female. 0.537 0.498 Age 1 ¼30 and under; 2 ¼31-49; 3 ¼50e64; 4¼65 and above. 1.985 0.825 Non-Spanish 1 ¼non-Spanish passenger; 0 otherwise. 0.300 0.458 Frequent flyer Number of flights taken by passenger in previous twelve months: 1¼0flights; 2 ¼1-3; 3 ¼4-12; and 4 ¼over 12 flights. 2.433 1.003 Homemaker 1 ¼passenger is homemaker; 0 otherwise. 0.032 0.175 Self-employed 1 ¼passenger is non-salaried, generally self-employed; 0 otherwise. 0.168 0.374 Salaried worker 1 ¼passenger is salaried worker; 0 otherwise. 0.584 0.493 Student 1 ¼passenger is student; 0 otherwise. 0.108 0.311 Retired 1 ¼passenger is retired; 0 otherwise. 0.082 0.274 b) Trip category. Base category includes passengers visiting friends and relatives (VFR) on a network carrier. Low-cost carrier 1 ¼passenger is flying with a low-cost carrier (LCC); 0 otherwise. 0.306 0.461 Vacation 1 ¼vacation trip; 0 otherwise. 0.441 0.497 Business 1 ¼business trip; 0 otherwise. 0.310 0.462 Length of stay (LOS) 1 ¼passenger returns after one night or earlier; 2 ¼two nights to a week; 3 ¼seven to 14 days; 4 ¼more than two weeks but less than a month; 5 ¼more than a month. 2.408 0.955 c) Social interaction. Base category includes passengers traveling without children and not accompanied to/from the airport. Group size 1 ¼traveling alone; 2 ¼two people; 3 ¼three or more people. 1.684 0.741 Children 1 ¼traveling with children; 0 otherwise. 0.078 0.267 Seen off 1 ¼someone sees passenger off at airport; 0 otherwise. 0.279 0.448 d) Environment. Base category includes passengers traveling on workdays. Weekend 1¼survey taken on Saturday or Sunday, when taxi rate (flat or regular) is higher; 0 otherwise. 0.265 0.441 Hotel 1 ¼passenger departs from a hotel, boarding house, or other paid accommodation; 0 otherwise. 0.214 0.410 Home 1 ¼passenger departs from own primary or secondary home; 0 otherwise. 0.523 0.499 Friends or family 1 ¼passenger departs from home of friends or relatives; 0 otherwise. 0.142 0.349 Table 3 Probit estimation of the propensity score. Covariate Coefficient Covariate Coefficient Sex 0.031 Business 0.502*** Age 0.039 Length of stay 0.077*** Non-Spanish 0.885*** Group Size 0.063*** Frequent flyer 0.029*** Children 0.110 Homemaker 0.089 Seen off 0.004 Self-employed 0.028 Weekend 0.092** Salaried worker 0.027 Hotel 0.183*** Student 0.205** Home 0.035 Retired 0.018 Friends or family 0.155*** Vacation 0.038 Constant 0.907*** No. of observations (before weighting) 19930 Log pseudo-likelihood 14640197 Pseudo R 2 0.137 Wald Chi 2 without clusters (p-value) 4638753.72 (0.000) Note: *, **, or *** indicate coefficient significance at the 10%, 5%, and 1% levels calculated from standard errors robust to heteroskedasticity and clustered by airport of origin. Table 4 Multinomial estimation of relevant effects. LCC (D i ) Constant b 3 ðx i Þð b 3 ðx i ÞE½b 3 ðxÞÞD i Public Bus 0.227* 1.825*** 0.794 1.154*** Hotel Bus 0.017 3.290*** 4.584*** 2.912*** Rent-a- car 0.169* 1.251 1.966* 0.178 Taxi 0.218*** 0.080 0.795* 0.428 Note: *,**,or*** indicatecoefficientsignificanceat the10%,5%, and1%levels,calculated from standard errors robust to heteroskedasticity and clustered by airport of origin. J.I. Castillo-Manzano / Journal of Air Transport Management 16 (2010) 295e298 297 3.3. The marginal effect of b a As in binary outcome models, only the sign of the coefficient is directly interpreted in multinomial models. Thus, a positive coefficient in the multinomial logit means that, as the regressor increases, alternative jis more likely to be chosen than alternative k. In order to make the reading of the results easier, the odds ratios or relative-risk ratios for the binary variable D i (using an LCC) are also considered. Following Cameron and Trivedi (2009), the relative probability or odds ratio of choosing alternative jrather than alternative 1, also called the base outcome, is given by Prðyi¼jÞ Prðyi¼1Þ¼ex 0 i s j (7) However, multinomial logit coefficients and odds ratios only allow the substitutability relationship between pairs of options to be studied, that is, the relationship between each option and the base category, which in our case is the private car. In order to overcome this focus on pairwise oppositions, we calculated the marginal effects of D i across all considered options. This enabled the direct substitutability relationship, should one exist, to be obtained between the five modes of transportation for LCC passengers (D i ¼1), as opposed to the control group including all passengers flying with legacy airlines. According to Cameron and Trivedi (2009), the marginal effects at the mean (MEMs) for the multinomial logit model are: d pij d Di ¼pij a j a i(8) a i¼Xpil a l; where is a probability-weighted average of a l . 4. Results Table 3 summarizes the results of propensity score estimation (Model 3), in the context of the 20 covariates in Table 2. A probit specification was opted for as it maximized the log pseudolikelihood. A multinomial logit specification (Models 5 and 6) was then used to estimate the average treatment effects. The results are presented in Table 4. Finally, the marginal effect at the mean of an LCC passenger is estimated by applying Model 8 (Table 5). 5. Conclusions The results show that being a LLC passenger reduces the probability of choosing a taxi to go to the airport by 5.85%, but increases the likelihood of a LLC passenger and that of choosing a rented car or a public mode of transportation by about 4% and 2%, respectively. The outcome cannot be explained by factors such as LCC passengers having a lower income level than the network carrier passengers because the scores are corrected using income level proxy variables. It would therefore be more appropriate to speak of passengers who are increasingly price-conscious. There could, however, be a threshold type price effect on mode choice if there is a strong psychological shock in paying more for a 10 km taxi ride than for a 1000 km airplane flight. The average city-airport distance for the seven airports in our sample is 9.86 km, with a standard deviation of only 1.21. Generally-speaking, 10 km is a reasonable city-airport distance measure for Spanish secondary airports. Acknowledgements The author would like to express his gratitude to AENA for its support. The author is also grateful to Kenneth Button, Guido Imbens and the anonymous reviewers for very helpful comments. References Abadie, A., Imbens, G.W., 2006. Large sample properties of matching estimators for average treatment effects. Econometrica 74, 235e267. Barrett, S.D., 2004. How do the demands for airport services differ between fullservice carriers and low-cost carriers? Journal of Air Transport Management 10, 33e39. Cameron, A.C., Trivedi, P.K., 2009. Microeconometrics Using Stata. Stata Publication, College Station. Castillo-Manzano, J.I., López Valpuesta, L. The decline of the traditional travel agent mode. Transportation Research E in press, doi:10.1016/j.tre.2009.12.009. De Neufville, R., 2006. Planning airport access in an era of low-cost airlines. Journal of the American Planning Association 72, 347e356. Echevarne, R., 2008. The impact of attracting low cost carriers to airports. In: Graham, A., Papatheodouru, A., Forsyth, P. (Eds.), Aviation and Tourism. Ashgate Publishing, Hampshire. Francis, G., Humphreys, I., Ison, S., 2004. Airports’perspectives on the growth of low-cost airlines and the remodelling of the airporteairline relationship. Tourism Management 25, 507e514. Graham, A., 2008. Managing Airports: An International Perspective, third ed. Butterworth Heinemann-Elsevier, Oxford. Heckman, J.J., Vytlacil, E., 2005. Structural equations, treatment effects, and econometric policy evaluation. Econometrica 73, 669e738. Hirano, K., Imbens, G.W., 2001. Estimation of causal effects using propensity score weighting: an application to data on right heart catheterization. Health Services & Outcomes Research Methodology 2, 259e278. Hotz, V.J., Imbens, G.W., Mortimer, J.H., 2005. Predicting the efficacy of future training programs using past experiences at other locations. Journal of Econometrics and Statistics 125, 241e270. Humphreys, I., Ison, S., 2005. Changing airport employee travel behaviour: the role of airport surface access strategies. Transport Policy 12, 1e9. Imbens, G.W., 2004. Nonparametric estimation of average treatment effects under exogeneity: a review. Review of Economics and Statistics 86, 4e29. O’Connell, J.F., Williams, G., 2005. Passengers’perceptions oflow cost airlines and full service carriers: a case study involving Ryanair, Aer Lingus, Air Asia and Malaysia Airlines. Journal of Air Transport Management 11, 259e272. Pels, E., Njegovan, N., Behrens, C., 2009. Low-cost airlines and airport competition. Transportation Research E 45, 335e344. Reynolds-Feighan, A., 2001. Traffic distribution in low-cost and full-service carrier networks in the US air transportation market. Journal of Air Transport Management 7, 265e275. Rosenbaum, P.R., Rubin, D.B., 1983. The central role of the propensity score in observational studies for causal effects. Biometrika 70, 41e55. Rotnitzky, A., Robins, J., 1995. Semiparametric regression estimation in the presence of dependent censoring. Biometrika 82, 805e820. Rubin, D.B., 1974. Estimating causal effects of treatments in randomized and non-randomized studies. Journal of Educational Psychology 66, 688e701. Tsamboulas, D.A., Nikoleris, A., 2008. Passengers’willingness to pay for airport ground access time savings. Transportation Research A 42, 1274e1282. Table 5 Marginal effect of b a : Private Car 0.0003 /¼0.03% Public Bus 0.0185 / D 1.85% Hotel Bus 0.0010 /¼0.10% Rent-a-Car 0.0385 / D 3.85% Taxi 0.0585 /V5.85% J.I. Castillo-Manzano / Journal of Air Transport Management 16 (2010) 295e298298