Content uploaded by José Cabral Vieira
Author content
All content in this area was uploaded by José Cabral Vieira on Mar 13, 2014
Content may be subject to copyright.
WORKING PAPER SERIES
Universidade dos Açores
Universidade da Madeira
CEEAplA WP No. 01/2006
Willingness to Pay for Airline Services Attributes:
Microeconometric Evidence from a Stated
Preference Discrete Choice Model
António Gomes de Menezes
José António Cabral Vieira
February 2006
Willingness to Pay for Airline Services Attributes:
Microeconometric Evidence from a Stated
Preference Discrete Choice Model
António Gomes de Menezes
Universidade dos Açores (DEG)
e CEEAplA
José António Cabral Vieira
Universidade dos Açores (DEG)
e CEEAplA
Working Paper n.º 01/2006
Fevereiro de 2006
CEEAplA Working Paper n.º 01/2006
Fevereiro de 2006
RESUMO/ABSTRACT
Willingness to Pay for Airline Services Attributes: Microeconometric
Evidence from a Stated Preference Discrete Choice Model
We analyze consumer preferences for airline services attributes between Ponta
Delgada and Lisbon. For this purpose, we conduct a stated preferences choice
game and estimate a microeconometric model à la McFadden (1974).
Our results are statistically significant and imply willingness to pay measures
economically high for attributes such as punctuality warranties and comfort.
We interpret these results at the light of the theories found in Kahneman (2003).
Willingness to pay for additional daily flights is quite low. This result is important
to how should the policy maker liberalize this sector.
Keywords: Stated Preferences Choice Games, Conditional Logit, Willing-
ness to Pay, Airline Services, Air Transportation Policy.
JEL Codes: C35, C93, R41
António Gomes de Menezes
Departamento de Economia e Gestão
Universidade dos Açores
Rua da Mãe de Deus, 58
9501-801 Ponta Delgada
José António Cabral Vieira
Departamento de Economia e Gestão
Universidade dos Açores
Rua da Mãe de Deus, 58
9501-801 Ponta Delgada
Willingness to Pay for Airline Services Attributes:
Microeconometric Evidence from a Stated
Preference Discrete Choice Model
António Gomes de Menezes and José Cabral Vieira
University of the Azores and CEEAplA
February 2006
Abstract
We analyze consumer preferences for airline services attributes between
Ponta Delgada and Lisbon. For this purpose, we conduct a stated preferences
choice game and estimate a microeconometric model à la McFadden (1974).
Our results are statistically signi…cant and imply willingness to pay measures
economically high for attributes such as punctuality warranties and comfort.
We interpret these results at the light of the theories found in Kahneman (2003).
Willingness to pay for additional daily ‡ights is quite low. This result is im-
portant to how should the policy maker liberalize this sector.
Keywords: Stated Preferences Choice Games, Conditional Logit, Willing-
ness to Pay, Airline Services, Air Transportation Policy
JEL Codes: C35, C93, R41
We thank support from Interreg III-B, MOVIECAM Project. We thank Concepción Román for
her comments and suggestions. Please address all correspondence to António Gomes de Menezes,
Department of Economics and Management, University of the Azores, Rua da Mãe de Deus, 9501-801
Ponta Delgada, Portugal. E-mail: menezesa@notes.uac.pt.
1
1 Introduction
It is usually the case that policy makers impose constraints on how airline companies
operate. Hence, and if one is interested in the welfare implications of such air trans-
portation policies, then one needs to have a sound knowledge of consumer preferences.
Otherwise, one will be left clueless with respect to the welfare properties of a given
air transportation policy package.
The aim of our paper is twofold. First, and on a methodological perspective, we
are interested in assessing the e¤ectiveness of a stated preferences choice game as an
instrument to reveal consumer preferences with respect to airline services attributes.
Second, and on an policy perspective, we set ourselves out to shed light on which
policy changes may induce social welfare increases. We do just that for the Ponta
Delgada - Lisbon corridor: the most important corridor between the Azores and
Mainland Portugal.
We note that our methodology is agnostic with respect to the geographical place
of its implementation. However, we do have good reasons to focus our attention in
the Ponta Delgada - Lisbon corridor: As we argue below, on the one hand, stated
preferences data come especially handy, as there are no revealed preferences data,
and, on the other hand, policy guidance is much needed.
The Azores are a Portuguese archipelago, with an autonomous government, in
the North Atlantic, about two hours by ‡ight west of Lisbon, with roughly the same
latitude (36o) as Lisbon and New York. The Azores have a disperse and exiguous
territory, with nine inhabited islands, within 600 kilometers apart, with a total surface
of 2.333 km2and a population of 241.000 inhabitants. Ponta Delgada is the main
city of the Azores, in the island of São Miguel, the largest and richest island in the
Azores.
Given its geography and population, it should come as no surprise that airline
services are commonly perceived as critical to the economic development and to the
social cohesion of the Azores. Thus, there has been heavy governmental regulation
2
in the airline services sector on, at least, two counts: (i) On equity grounds, inter-
island mobility and equal access to other regions regardless of island of origin are
politically understood as necessary to the social cohesion of the Azores. Hence, inter-
island mobility is and has been treated as a public service obligation (on this, more
below). SATA - the Azorean ‡ag carrier, owned by the Azorean Government - provides
and has provided such service as a monopolist operating under stringent regulations,
regarding fares, ‡ight capacity, ‡ight frequencies, among other services attributes.
(ii) On e¢ ciency grounds, due to an arguably lacking demand, on the one hand, and
high capital and operating costs, on the other hand, airline services are and have been
thought of as a natural monopoly.
Under these arguments, there has never been an open skies policy in the Azores.
Nowadays, the Azorean Government enforces stringent regulation on air transport,
which is allowed in the European Union within the framework of Article 4 of Council
Regulation 2408/92. In fact, until 2004 only one airline at a time ‡ew between a
given Azorean gateway and Mainland Portugal. Since 2005, two airlines - SATA and
TAP (the Portuguese ‡ag carrier, owned by the Portuguese Government) - operate
our route of interest, Ponta Delgada - Lisbon, via a code share agreement, as the
sole and joint concessionaires of air transportation services between the Azores and
Mainland Portugal.
However, both SATA and TAP are obliged to follow a stringent set of regulations
regarding several dimensions of their services, including fares, ‡ight frequencies, ‡ight
capacities, punctuality warranties and so on.1In essence, both SATA and TAP have
to implement twin operations strategies and procedures, with virtually no degrees of
freedom whatsoever. Therefore, there are no revealed preferences data that can shed
light on consumer preferences. But we do need to know consumer preferences if we
1See O¢ cial Journal of the European Union, 2004/C 248/06, 7.10.2004 (http://europa.eu.int/eur-
lex/lex/JOIndex.do?), the European Union policy directive that regulates ‡ights between the Azores
and Mainland Portugal.
3
aim to promote social welfare: the sum of producer surplus with consumer surplus.
Therefore, by learning consumer preferences regarding airline services we may provide
guidance to future changes in airline services that may promote increases in consumer
surplus. Hence, we implement a stated preferences choice game, and then we estimate
a discrete choice model à la McFadden (1974).
We resort to a stated preferences choice game and associated discrete choice model
since with this methodology, and to be brief, airline customers are asked to choose
between competing alternatives that di¤er, in a trade-o¤ sense, in several attributes.
Hence, our choice-based approach is based on a quite realistic task that airline cus-
tomers perform every day. In addition, our willingness to pay measures are consistent
with utility theory (see Merino-Castelló (2003) and Hanley et al. (2001) for extensive
discussions on stated preference discrete choice models and the reasons behind the
growing popularity of such models).
Several authors have successfully applied discrete choice models to transportation
policy issues in a number of ways and settings (see, among others, Ben-Akiya and
Lerman (1985), Wardman (1988), for surveys, and Burris and Pendalya (2002), for an
application). Cao and Mokhtarian (2005a, 2005b) argue that individuals adapt their
travel-related strategies according to a number of objective and subjective in‡uences,
and, hence, one should consider individual experiences and characteristics when fore-
casting the expected outcome of a given policy choice. We follow this reasoning and
control in our experiment for a number of individual characteristics.
The evidence that we provide also sheds light on consumer preferences towards
‡ight frequency. Thus, we can use this evidence as an input in the debate if we are
indeed in the presence of a natural monopoly or not. Hence, our paper contributes to
the literature on the e¢ ciency of the application of public service obligations (PSOs)
in air transport within the EU. As Williams and Pagliari (2004) argue, despite the
widespread application of PSOs across the European Union, with the aim of promot-
ing sustainable air services to remote regions for economic development purposes, as
4
in the Azorean case, there is very little research on how e¢ cient have such policies
been applied. Our paper shows that stated preference discrete choice models are an
e¤ective way to root PSOs on deep, structural consumer preferences parameters.
The paper is organized as follows. Section 2 describes the data. Section 3 presents
our econometric model. Section 4 discusses the results. Section 5 concludes.
2 Data
2.1 The Sated Preferences Choice Game
Our stated preferences choice game was implemented through questionnaires minis-
tered at Ponta Delgada’s Airport, near the boarding gate, after security checkpoint. A
total of 347 questionnaires were asked from April 27th to May 5th of 2005. The num-
ber of questionnaires ensures a number of observations large enough to estimate the
econometric model described below. The interviews were conducted in Portuguese.
Only people who were about to take a ‡ight from Ponta Delgada to Lisbon were
interviewed, to make sure that they were familiar with the questions asked. Moreover,
people who were traveling with tourist packages, namely, packages with a combination
of hotel, air travel, rent a car, and so on, were not considered since these people did
not have a clear idea of the exact cost of the air travel portion of their travel package.
The questionnaires had 3 sections. In the …rst section, a number of questions were
asked about the trip, such as: airline; connection at destination; connecting airline;
fare class (business, economy); departure time; trip cost; trip motive; trip frequency;
who pays for the trip; number of people ‡ying with the interviewee; advance of
purchasing the ticket; mode of purchasing the ticket; and frequent ‡yer program.
In the second section, the individuals were confronted with a stated preferences
choice game. In particular, with the aid of a laptop computer, the individuals were
asked to choose one of two virtual airlines that di¤ered in the following dimensions,
5
based, on the on hand, on the status quo,2and, on the other hand, on what we
observe elsewhere, namely in more deregulated and competitive markets:
Attribute Level
0
1
2
Business Cheap Fare
030% 100%
110% 50%
2 0% 30%
Business Cheap Fare
0Cold sandwiches + drink Not available
1Hot food + drink Cold sandwiches + drink
2A la carta (when buying the ticket) Hot food + drink
0
1
0
1
2
0
1
2
No compensation for delay
Free ticket for the same trip
Reimbursement of the cost of the ticket
Reliability
Frequency
2 flights / day
4 flights / day
6 flights / day
Penaly for
changes in
the ticket
Free Food
Comfort
Small space between seats
Wide space between seats
Price
Definition
P + 20%
P
P - 20%
Figure 1: Stated Preferences Choice Game
Other attributes which we may care about were left out of the game in order to
preserve a good understanding of the trade-o¤s involved (see Sudman and Bradburn
(1982) for practical issues on questionnaire design). As a corollary, travel time was
left out since it is, to a great extent, exogenous to the operator and regulator.
The following picture is a "Print Screen" of WinMint v. 2.1 (in Portuguese), the
software used to randomly generate the game menus.
2The status quo, to be brief, entails: two fares, economy and business; no penalty to change
tickets within a year; cold sandwiches if economy, hot food if business; small space between seats for
both fares; two ‡ights per day; and no compensation for delay.
6
Figure 2: Print Screen of Choice Game
In essence, the stated preferences choice game presented the passengers with a
choice between two virtual airlines, none of which dominated the other in all di-
mensions, as expected. That is, all games considered had trade-o¤s built-in. Each
individual played the game 10 times.
In the third and last section, the individuals were asked about their socioeconomic
status, such as: residence county; number of people living in the household; number
of workers in the household; household income; age; gender; educational attainment;
sector of occupation; type of job; weekly working hours and net monthly individual
income.
2.2 Descriptive Statistics
Table 1 summarizes some of the continuous variables in the data set:
7
Variable Observations Mean S. Deviation Minimum Maximum
Trip cost (€) 347 122,37 37,98 - 250,00
Net household monthly income (€) 347 2.645,08 1.679,55 150,00 12.500,00
Weekly working hours (hours) 347 18,80 13,10 0,00 60,00
Net individual monthly income (€) 347 1.196,04 1.325,54 0,00 10.000,00
Age (years) 347 36,53 13,57 19,00 85,00
Table 1: Descriptive Statistics
Mean reported one way ticket cost is e122. In addition, we note that most in-
terviewees ‡ew with SATA, in a domestic ‡ight with no connection and were males.
Most interviewees, 67%, bought the tickets with one week or less in advance of de-
parture day. The travel agency was the mode of purchasing ticket chosen by 69%
of the individuals. While 50% of the interviewees paid for their tickets, 35% of the
interviewees had their tickets paid for their companies. A slight majority, 51%, of the
interviewees had some sort of frequent ‡yer program. Perhaps not surprisingly, many
interviewees held a university degree, 51%, since being at the boarding gate is not a
random event across the overall Portuguese population.
3 Model
3.1 Benchmark Model
The econometric work carried out in the paper is based on the random utility the-
ory (see McFadden (1974), Greene (2003) or Train (2003)), brie‡y described below.
Consider that the random utility of alternative jfor an individual q,Ujq, is given by:
Ujq =Vjq +"jq (1)
where Vjq is the systematic or representative utility (conditional indirect utility) and
"jq is a random term.
Individual qchooses alternative jif and only if Ujq Uiq ,8i6=j. In such a case,
8
and given (1):
Ujq Uiq ()
Vjq +"jq Viq +"iq ()
"iq "jq Vjq Viq ,8i6=j
As utilities are random variables, we can obtain the probability that individual q
chooses alternative jas:
Pjq =P("iq "jq Vjq Viq ),8i6=j(2)
When the random term "jq follows a Gumbel distribution, then Pjq reads (see
McFadden 1973):
Pjq =eVj q
PN
i=1 eViq
(3)
where Nis the number of alternatives. The expression for Pjq given by (3) is the
essence of the well-known multinomial logit model.
3.2 Microeconometric Model
We estimate a conditional logit model, since we have several observations (games)
per individual, and, hence, we control for individual …xed e¤ects. The estimation was
carried out with STATA Intercooled 8.
As usual in the literature (Bateman et al. (2002), Espíno et al. (2003), Fowkes
and Wardman (1998), Fowkes (2000), and Louviére et al. (2000)), we estimate two
alternative speci…cations of the conditional indirect utility, described below. In Model
1 we do not consider interactions between attributes and the conditional indirect
utility reads:
Vj=CC+PP+F1F1+F2F2+(4)
+LRLR +F r F r +R1R1+R2R2; j = 1;2
9
In Model 2 we consider interactions between attributes and hence we write the
conditional indirect utility as follows:
Vj=CC+ (P+P W )P+(5)
+(F1+F1Ec Ec)F1+
+(F2+F2Ec )F2+LRLR +F r F r +
+(R1+R1WW)R1+ (R2+R2WW)R2; j = 1;2
Table 2 provides a list of variables de…nitions.
Table 2: Variables De…nitions
Variable Meaning
Ctravel cost (euros)
Ppenalty for changes in the ticket
F1binary variable equal to 1 if food level equals 1
F2binary variable equal to 1 if food level equals 2
LR binary variable equal to 1 if comfort (more leg room) is 1
F r daily ‡ight frequency (continuous variable)
R1binary variable equal to 1 if reliability level equals 1
R2binary variable equal to 1 if reliability level equals 2
Ecbinary variable equal to 1 if fare is economy
Wbinary variable equal to 1 if trip motive is work
After estimation of the models above, it is possible to compute the willingness
to pay (WTP) for improvements. For continuous variables the subjective value of
attribute qkj reads:
WTPj
qkj =dI
dqkj
=
@Vj
@qkj
@Vj
@I
=
@Vj
@qkj
@Vj
@cj
=dcj
dqkj
where Istands for income and @Vj
@I =@ Vj
@cj. For binary variables the relevant expres-
sion is as follows:
WTPj
qkj =V1
jV0
j
@Vj
@I
10
where Vi
jis the conditional indirect utility of alternative jwhen the level of the
attribute equals i= 0;1.
4 Results
Table 3 summarizes the results for models 1 and 2. The signs are as expected and the
estimates are statistically signi…cant, with the notable exception of the interaction
terms. Adding the interaction terms seems to matter little, both at a qualitative level
and at a quantitative level.
Table 3: Results for Model 1 and Model 2
11
Variable Model 1 Model 2
Cost (C)0:0251
(18:02)
0:0252
(18:04)
Penalty (P)0:0140
(6:97)
0:0138
(5:79)
Food 1 (F1)0:2505
(3:77) 0:7208
(2:86)
Food 2 (F1)0:4403
(6:24) 0:8944
(3:83)
Leg Room (LR)0:5123
(8:98) 0:5135
(8:99)
Frequency (F r )0:1266
(7:09) 0:1279
(7:15)
Reliability 1 (R1)0:9894
(14:68) 0:9868
(11:46)
Reliability 2 (R2)0:8294
(11:66) 0:8667
(11:46)
Food 1*Economy (F1Ec )0:5005
(1:93)
Food 2*Economy (F2Ec )0:4828
(2:03)
Penalty*Work (PW )0:0009
(0:23)
Reliability 1*Work (R1W)0:0174
(0:13)
Reliability 2*Work (R2W0:0849
(0:70)
Log L()3959 3956
Log L(0) 4207 4207
Number of observations 6940 6940
1%; 5%; 10%
In order to obtain a feel of the economic importance of these results we compute
the willingness to pay measures, presented in Tables 4 and 5.
Table 4: Willingness to Pay Measures for Model 1
12
WTP - Model 1
Event WTP (euros)
Penalty for changes in the ticket 0:57
Food: level 0 to level 1 9:97
Food: level 0 to level 2 17:52
Comfort (more leg room) 20:39
Frequency 5:04
Reliability: level 0 to level 1 39:39
Reliability: level 0 to level 2 33:02
Given that the sample mean cost of a one way ticket is about e122, we …nd that
willingness to pay measures are quite high. In particular, the willingness to pay to
improve reliability from level 0 to 1 is about e39 or 32% of the sample mean of the
reported one way ticket cost. Apparently, comfort is quite valuable: the willingness
to pay to have some more leg room is more than e20.
Willingness to pay measures do not change substantially when we consider inter-
actions between trip attributes (Model 2):
Table 5: Willingness to Pay Measures for Model 2
13
WTP - Model 2
Event WTP (euros)
Penalty for changes in the ticket:
Trip motive: work/business 0:58
Trip motive: other 0:55
Food: level 0 to level 1
Economy class 8:74
Other type of fare 28:59
Food: level 0 to level 2
Economy class 16:33
Other type of fare 35:48
Comfort (more leg room) 20:37
Frequency 5:08
Reliability: level 0 to level 1
Trip motive: work/business 39:83
Trip motive: other 39:14
Reliability: level 0 to level 2
Trip motive: work/business 31:01
Trip motive: other 34:38
We note that the willingness to pay for one additional ‡ight per day is about 5
euros. Hence, the subjective value of increased daily ‡ight frequency is far less, in
an economic sense, than the subjective value of improvement in attributes such as
reliability or comfort.
5 Conclusions
The McFadden Discrete Choice Model is an informative tool about consumer pref-
erences over di¤erent attributes across competing alternatives, including in environ-
14
ments where revealed preferences do not take us far. Obviously, this is the case of
airline services in the Ponta Delgada –Lisbon corridor where there are no data which
can be used in a revealed preferences exercise. Thus, a stated preferences exercise
was conducted to reveal consumer preferences. Being armed with such knowledge
on consumer preferences is a must if one is interested in implementing social welfare
maximizing policies. This is certainly the case in heavily regulated markets, such
as the Azorean case, where air transport is regulated as a public service obligation
within the EU framework for remote regions.
The main results were as expected from utility theory and some willingness to
pay measures are quite high, in an economic sense, such as regarding punctuality
(reliability) and comfort. However, some other willingness to pay measures were
found to be revealingly low. This is the case of willingness to pay for increases in
daily ‡ight frequency: about …ve euros. This result is somewhat puzzling considering
that the Ponta Delgada - Lisbon corridor is the most important corridor servicing
the Azores and that quite often ‡ights are fully booked and waiting lists several day
long. Taken at face value, this anecdotal evidence on waiting lists suggests that
‡ight frequency is a binding constraint and that passengers would be willing to pay
a sizeable amount to have such constraint relaxed. It turns out not to be the case.
Instead, our result suggests that passengers do not perceive ‡ight availability as a
bidding constraint. In addition, this result should be upward biased in the sense that
we did not interview a random sample of the population but people who were actually
‡ying, and, hence, everything else the same, more willing to pay for increased ‡ight
availability. However, it should be noted that this result does not imply that there is
no demand for extra ‡ights. It is logically coherent with a scenario of a highly elastic
demand. It simply suggests that there is no demand for more ‡ights at increased
cost. But there may be demand for more ‡ights at given or lower prices.
We also note that this result may be in‡uenced by the interviewee’s own judgement
about his ability to secure a ‡ight through, say, planning in advance. As Kahneman
15
(2003) argues, individuals, in general, are prone to over estimate their own ability in
a number of settings. It is also quite interesting to note that the willingness to pay
for avoiding penalties for changing tickets is quite low: less than one euro. Pereira
et al (2005) …nd similar results to ours to the Funchal - Lisbon route. Like us, in
their study willingness to pay measures seem lower for attributes arguably perceived
as endogenous from the interviewee’s perspective, in the sense that the interviewee
may believe that he may act in a way to avoid penalties, secure ‡ights and so on.
Airline regulators and operators alike should take heed of these results to root
their policies and operations in deep, structural consumer preferences parameters.
16
References
[1] Bateman et al., 2002. Economic valuation with stated preferences techniques: A
manual, UK Department of Transport. Edward Elgar Publishing Inc.
[2] Ben-Akyia, M., Lerman, S., 1985. Discrete choice analysis: Theory and applica-
tion to travel demand. The MIT Press, Cambridge, MA.
[3] Burris, M., Pendalya, R., 2002. Discrete choice models of traveler participation
in di¤erential time of day pricing programs. Transportation Policy 9 (3), 241-251.
[4] Cao, X., Mokhtarian, P., 2005a. How do individuals adapt their personal travel?
A conceptual exploration of the consideration of travel-related strategies. Trans-
port Policy (12) 3, 199-206.
[5] Cao, X., Mokhtarian, P., 2005b. How do individuals adapt their personal travel?
Objective and subjective in‡uences on de consideration of travel-related strate-
gies for San Francisco Bay Area commuters. Transport Policy (12) 4, 291-302.
[6] Espíno, R., Ortúzar, J., Román, C., 2003. Analyzing the e¤ect of latent variables
on willingness to pay in mode choice models. Proceedings of the ETC 2003,
Strasbourg, France.
[7] Fowkes, A., Wardman, M., 1988. The design of stated preference travel choice
experiments with special reference to interpersonal taste variations. Journal of
Transport Economics and Policy XXII, 27-44.
[8] Fowkes, A., 2000. Recent developments in stated preference techniques in trans-
port research. In J. de D. Ortúzar (ed.) Stated Preference Modelling Techniques,
Perspectives 4, PTRC; London.
[9] Greene, W., 2003. Econometric Analysis. Prenctice Hall, New Jersey.
17
[10] Hanley, N., Mourato, S., Wright, R., 2001. Choice modeling approaches: A
superior alternative for environmental valuation? Journal of Economic Surveys
15 (3).
[11] Kahneman, D., 2003. Maps of bounded rationality: Psychology for behavioral
economics. American Economic Review 93 (5), 1449-1475.
[12] Louviere, J., Hensher, D., Swait, J., 2000. Stated choice methods: Analysis and
applications. Cambridge University Press, Cambridge.
[13] McFadden, D., 1973. Conditional logit analysis of qualitative choice behavior.
Frontiers in Economics, 105-142, Academic Press, New York.
[14] McFadden, D., 1974. The measurement of urban travel demand. Journal of Public
Economics 3, 303-328.
[15] Merino-Castelló, A., 2003. Eliciting consumer preferences using stated preference
discrete choice models: Contingent ranking versus choice experiment. Universi-
dad Pompeu i Fabra, Working Paper no705.
[16] Pereira, P., Budria, S., Almeida, A., 2005. Analyzing mobility in the Atlantic
Area of Madeira, Azores and Canary Islands. University of Madeira, mimeo.
[17] Sudman, S., Bradburn, N., 1982. Asking questions: A practical guide to ques-
tionnaire design. Jossey Bass, San Francisco.
[18] Train, K., 2003. Discrete choice models using simulation. Cambridge University
Press, Cambridge.
[19] Wardman, M., 1988. A comparison of revealed preference and stated preference
models of travel behaviour. Journal of Transport Economics and Policy 22 (1),
71-91.
18
[20] Williams, G., Pagliari, R., 2004. A comparative analysis of the application and
use of public-service obligations in air transport within the EU. Transport Policy
11 (1), 55-66.
19
