Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
a b s t r a c t
Keywords:
Charter airlines
Choice experiment
Mixed logit
In the competitive aviation market as a result of the emergence of low cost carriers, charter airlines have
had to reconsider their approach to service provision. Specically, the reduction in service and comfort
levels offered by the low cost airlines provides charter carriers with an opportunity to differentiate their
product based on the quality of the offering. To consider this strategic option we employ an on-line
choice experiment to examine consumer choices with respect to the bundle of services on offer when
deciding to purchase a ight. With these data we use the Bayesian methods to estimate a mixed logit
specication. Our results reveal that in principle passengers are willing to pay a relatively large amount
for enhanced service quality.
2008 Elsevier Ltd. All rights reserved.
1. Introduction
The competitive landscape of the European aviation industry
has changed signicantly since deregulation in 1992 as a result of
price pressures due to erce competition from the growth of low
cost carriers (LCC). The success of LCCs results from an uncompromising no-frills strategy that minimizes costs per by only
offering a modicum of comfort in terms of legroom, seat dimensions, not installing window blinds, headrests, seat-pockets or seat
reclining mechanisms (Franke, 2003). Furthermore, they charge
passengers to check baggage into the hold, which simplies ground
handling, airport operations and turn-around times, plus food,
drinks and in-ight entertainment provision are supplied on
a buy-on-board basis (Barrett, 2003). The associated ancillary
revenues form a signicant part of the business model; for Ryanair
(2006) 15% of revenues in 2006.
However, because the LCC business model is based on maximizing utilization of aircraft (two return journeys per day); to keep
them in the air for the longest possible time, ight range is limited
to 3 h. Thus, UK based LCC operations only y to European destinations (Francis et al., 2006). In response, existing carriers, especially charter airlines (CAs) have engaged in signicant eet
redeployment outside of the 3-h range. For example, operations to
Spain and the Balearic Islands have been reduced, with destinations
* Corresponding author.
E-mail address: i.m.fraser@kent.ac.uk (I. Fraser).
0969-6997/$ see front matter 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jairtraman.2008.12.005
such as Turkey, northern Africa and the Canary Islands being given
greater emphasis (Bieger and Wittmer, 2006).1 However, the
extensive cost-cutting of LCCs presents an opportunity for CAs to
differentiate away from the commodity travel product on offer
(Dennis, 2007).
We examine how CAs might differentiate their products.
Specically, we consider which attributes of in-ight cabin comfort
and service may have a value to consumers by analyzing consumer
willingness to pay (WTP) for in-ight attributes that are derived
from an Internet delivered choice experiment (CE). With a CE
survey respondents choose between alternative options that
contain a number of attributes of different levels. Our CE was
designed to consider air traveler choice as it relates to on-board
service and comfort levels for CAs for a ight of between 4 and 5 h.
The issue of service provision and ticket price has been previously examined, for example, by Lee and Luengo-Prado (2004) who
considered how legroom impacted the price paid by consumers. For
one airline passengers did appear willing to pay more for additional
legroom, whereas for another they found the opposite. Espino et al.
(2008) and Martin et al. (2008) used a CE undertaken in Spain in
which passengers were presented six attributes (price, penalty for
ticket changes, food (sandwich, hot meal or a la carte), comfort (i.e.,
legroom), frequency and reliability) relating to two virtual airlines
1
Generally CA ownership is heavily concentrated with the largest players
belonging to large vertically integrated tour groups. Vertical integration means that
a large proportion of European CA seats are provided for inclusion in such tour
packages (Bieger and Wittmer, 2006).
222
Uj;s;n x0j;s;n t bj es;j;n
(1)
0
t bj t1 b1j 1; t2 b2j ; ..; tk bkj
(2)
that means that the quantities t2 b2j ; ..; tk bkj are the marginal
rates of substitution (MRS) with the numeraire being the rst
attribute, which will always be the price or cost attribute within the
CE. We assume that the bj is normally distributed, and that the reparameterisation t(.) does change the distribution, but t(bj) bj
preserves the normality assumption. Finally, the error term es,j,n is
assumed to be extreme value distributed, independent of x0 s,j,n and
uncorrelated across individuals or choices.
The re-parameterisation in Eq. (2) means model estimation is
being performed in WTP space. Typically, the ML is estimated in
preference space such that we would rst estimate marginal utilities. The resulting MRS is derived from these estimates. However,
by estimating our model in WTP space, all the MRS is estimated
directly. It has been found that this approach can signicantly
reduce the instability associated with WTP estimates in preference
space. The issue of WTP instability in preference space has been
found to be particularly problematic when the price or cost variable
is not bounded above zero. In existing studies it has been common
practice to x the payment coefcient that is an ad hoc approach to
resolve the instability especially as xing the price coefcient may
2
Other examples of CEs used to assess other aspects of consumer choice in terms
of air travel include Hess and Polak (2005), Rose et al. (2005), Hess et al. (2007) and
Teichert et al. (2008).
3
A detailed description of algorithm, in this case Gibbs with a MetropolisHastings (MH) Step is provided in Balcombe et al.
4
As a referee indicated, another reason for excluding lounges is that they are
likely to correlate with the quality of seat on the plane and such could be a source of
multicollinearity in the resulting analysis.
Units
Status Quo
Levels
Additional Levels
Seat Pitch
Seat Width
In-Flight Meal
In-Flight Entertainment
Inches
Inches
Level
Level
28
17
Hot Meal
Standard
Complementary in-Flight
Drinks
Ticket Prices
Level
None
31, 34
18.5
None, Sandwich
Standard plus Amenity Pack,
Standard plus Amenity Pack
plus Own Screen
Complementary Bar Service
Euros
300
223
includes the AP plus a private screen built into the seats, which
allow personalized viewing and other media. This option represents the current Premium Economy offering on some CAs longhaul eet services. Complimentary bar service is kept as a simple
binary yes/no option. On-board drinks sales can constitute signicant ancillary revenue. The added perceived value of complimentary drinks can be compared to this, to assess its potential to add
value.
Finally, the price attribute was set as a representative gure for
the 4.55.5-h ight duration presented to the respondent.
Departing from the UK, this would include destinations in countries
such as Egypt, Turkey, and as far aeld as Gambia. The actual
destinations used in the CE were Dalaman, Izmir and Antalya
(Turkey); Sharm el Sheikh and Hurghada (Egypt); and Banjul
(Gambia) with departures from Manchester and London Gatwick.
Prices were researched on the basis of a booking in the summer
season made two months in advance.6
Given the selection of attributes and their associated levels,
employing an orthogonal fractional factorial design and taking
account of dominated options 15 proles are derived. In this case
a prole refers to a combination of different levels of attributes,
with each prole taking on exactly one level of each attribute. The
aim was to construct a choice card based on each of the proles
from the orthogonal design. Each choice card is made up of three
proles. To construct the choice cards each of the proles were
taken as the rst value on each card. Then to generate a second
prole, the shift method was used. Then, to generate the nal set
of choice cards with three proles a status quo was added to each
which represented a standard offering for a CA seat. The status quo
option is 28inch seat pitch, 17inch seat width, a hot meal, standard
in-ight entertainment, no bar service and a price of V300. Finally,
a dont know option was added. This option allows the respondents to provide an answer that does not force an inappropriate
choice. Thus, it enables the elicitation of a more realistic view of the
respondents true preferences. In the analysis we present we
treated dont know responses as a zero choice: that is they were
not included for estimation purposes.
Fig. 1 provides an example of the choice cards presented to
respondents. This card includes various explanatory notes because
it was the one used in the introduction to the survey respondents
how the survey would work. It includes information to improve
response quality but without being so long as to detrimentally
affect response rates. It can be seem that we did not label the
status quo option. This removes potential labeling effects, and
potentially avoids anchoring the respondents choice. However, if
any respondents recognize this combination of attributes, or if the
combination already represents a favored combination for the
sample, this may manifest itself in a bias toward its selection.
The survey instrument was delivered via the Internet. It was
made available to the public via the web site of a large travel
company. We adopted this approach to survey delivery because it
provided the most cost effective means by which to generate
a sample of sufcient size.
In terms of survey delivery, potential survey participants were
initially asked if they would be prepared to undertake the survey.
This took the form of link on the web site. They were informed that
the survey was being conducted for academic purposes and that
the information provided would remain condential. If they agreed
to participate they were then automatically directed to the survey
instrument via an appropriate link. To ensure that the data
6
In this CE we have not addressed the issue of how a price can vary depending
upon time of purchase and associated seat availability. To include this aspect of
consumer behavior would require a very different CE to that developed here.
224
Thank you for your time, this survey should not take more than 5-10 minutes.
You will be asked to select which flight you would choose on a 4 - 5 1/2 hour flight. This
could be to a destination such as the Canary Islands, North- or Western Africa or Turkey.
Assume you are now booking the flight 2 months in advance.
You will be faced with a series of choices between 3 airline tickets. They will each include
different levels of seat-pitch, seat width, in-flight meal, in-flight entertainment and bar
service.
The following shows an example choice card. It also gives some detail about the different
attributes. Please review these and then proceed to the survey.
Standard entertainment includes
movies and short programmes
playing on overhead screens.
Seat Pitch is given in inches,
ranging from 28 to 34.
28 Seat Pitch will give an
average person roughly one inch
legroom, if sitting upright.
Increasing Seat Pitch to 31 or 34
will give 3 or 6 inches extra
legroom, respectively.
This Price is
for a Return
flight with all
taxes, fees
etc included.
Seat
Width
17"
none
Seat
Pitch
31"
Seat
Width
18.5"
sandwich
Seat
Pitch
28"
Seat
Width
17"
hot meal
In-Flight Meal
In-Flight Entertainment
standard plus amenity pack
In-Flight Meal
In-Flight Entertainment
private screen plus amenity pack
In-Flight Meal
In-Flight Entertainment
standard
Complimentary
Bar Service
yes
Ticket
Price
300
Complimentary
Bar Service
none
Ticket
Price
400
Complimentary
Bar Service
none
Ticket
Price
300
Don't Know
Extra seat
width is
available on
some options
For each card, please select only one option. If you are unsure which you prefer, tick
the dont know box.
Fig. 1. Front page of the survey.
(3)
where Price is the price of the ticket, Pitch and Width are seat
specication characteristics, Bar is a drinks service, Amenity Pack
and Screen are on-board entertainment options, and Sandwich
and No Meal are the food options. These parameters provide us
with a measure of how much our survey respondents are WTP to
change their in-ight on-board travel services. Finally, the Status
Quo captures an option common to all choice sets that can be
consider the current default with respect to in-ight travel service
provision.
In terms of computational performance the MCMC sampler
worked very well for this particular model specication. The results
we generated by employing a burn-in of 1000 draws, then undertaking a further 10,000 draws. Both the burn-in and the subsequent
draws used in the analysis are constructed by sampling only one in
every 100 draws so as to yield robust statistical results. Both
observation of the sample and our convergence statistics indicated
that our model converged.
The rst thing to note about the results reported in Table 2 is
that the Price coefcient is the sign we would expected a priori.
Second, as this model is estimated in WTP space these estimates
can be directly interpreted as WTP estimates. However, it is
necessary to note that some of the variables have been scaled as
part of the estimation process. The Price coefcient has been
divided by 100 and Pitch and Width by 10. Thus, we can see that on
average our survey respondents are WTP an extra V22 for additional Pitch (leg space) and V29 for extra seat width. With respect
to the other attributes we can see that respondents are not WTP for
an Amenity Pack (AP) because of the negative sign, although for this
parameter the posterior mass is to the positive and negative side of
zero as indicated by the lower quartile estimate being negative and
the upper quartile estimate being positive, which means that we
are not condent that this parameter is not equal to zero. Similarly
the sandwich parameter also has a posterior density both positive
and negative. However, respondents are prepared more for a Bar
service and a screen. Finally, the negative sign on No Meal indicates
that survey participants are WTA the loss of this service but only
with an associated reduction in the price of the ticket, i.e., V31. The
WTP for these changes attribute levels are V31, V37, and V31,
respectively. Although these might appear to be rather large it has
to be remembered that once on-board an airline is a monopoly
provider of all services and as such consumers may well have
become conditioned to expect these kinds of values for these
additional services. It also needs to borne in mind that these
additional amounts in and of themselves are relatively small
compared to the overall price of the ticket. To place these values in
context if an individual was to purchase a ticket that included the
225
highest specication on offer relative to the status quo then the net
WTP is V121.
The only CE study which provides comparable estimates is
Espino et al. (2008) who found that the MNL yielded higher WTP
estimates than the ML. In the case of additional legroom the MNL
yielded WTP of V33 and the ML V15. Our estimate is V22 which is
clearly comparable. However, our estimate for the WTP for
a sandwich is such that we cannot be condent that the parameter
is not zero. We also nd that the WTP for having No Meal is much
higher than Espino et al. report, although the duration of the ight
examined in our study is signicantly longer. It could also be the
case that the estimate we report is for a reduction in the level of
service provided, as opposed to a level higher than the status quo.
In terms of actual pricing by airlines for various on-board
services, Dennis (2007) notes that Aer Lingus offers a full Irish
breakfast for V7 and a hot ham and cheese Panini for V5. Although
the price of the breakfast is signicantly less than the WTP for the
hot meal we estimate there is a reasonably correspondence with
our mean estimate for the price of a sandwich and the Panini. Thus,
the WTP estimates are comparable with other results and data.
The correlation coefcients are reported in Table 3 reveal that
there is no consistent pattern. For example, we can see that having
more leg space and seat width are negatively correlated. Thus, there
is no reason to assume that tall passengers who would prefer more
leg space demand wider seats. We can also see that extra leg space
is positively correlated with having a screen and correlated with No
Meal. Interestingly, there is a positive correlation between seat
width and having bar service, and a negative correlation between
seat width and no meal.
The socio-economic variables included in the analysis are Age,
Income, Gender and Education. All variables have been included as
dummy variables. In the case of Age and Income the data have been
divided at the sample mean; 34 years old and 35,000 per annum.
These variables are included in the ML as
(4)
This model was estimated assuming the same random parameter distributions. In terms of computational details this model
required a burn-in of 2000 draws, then undertaking a further
10,000 draws. Both were constructed by sampling only one in every
250 draws. This conguration was necessary given the higher
dimension of the model compared to the basic ML specication and
the resulting problems of achieving model convergence. Our results
for this model are presented in Table 4.
As can be seen in the table all the socio-economic variables have
an impact on the various WTPs. Starting with Age we can see that
older respondents are WTP more for aspects of seat comfort
whereas younger respondents are WTP more for an on-board bar
and screen. Thus, differences in Age indicate that older travelers are
more concerned with comfort and Younger travelers prefer very
Table 3
WTP correlation coefcients.
Table 2
MXL regression results (excluding socio-economic variables).
Variable
Mean
Standard Deviation
Lower Quartile
Upper Quartile
Price
Pitch
Width
Bar
AP
Screen
Sandwich
No Meal
ASC
1.337
2.187
2.886
0.307
0.080
0.373
0.068
0.316
1.042
0.484
2.872
0.755
0.372
0.341
0.477
0.364
0.389
1.265
0.991
0.236
2.372
0.056
0.312
0.049
0.174
0.054
1.889
1.592
4.121
3.402
0.561
0.153
0.695
0.316
0.580
0.188
Pitch
Width
Bar
AP
Price
0.042
0.018 0.024
0.043
0.005 0.045
Pitch
1
0.744 0.205
0.368
0.607
0.523
Width
1
0.125 0.249 0.460 0.433
Bar
1
0.077 0.032 0.185
AP
1
0.202
0.135
Screen
1
0.290
Sandwich
1
No Meal
ASC
0.019
0.522
0.418
0.077
0.187
0.312
0.412
1
0.026
0.940
0.71
0.233
0.374
0.587
0.505
0.502
1
226
Table 4
Mixed logit results conditioned on socio-economic variables.
Age Variable
Price
Pitch
Width
Bar
AP
Screen
Sandwich
No Meal
ASC
Income Variable
Young
Old
Mean
stdv
Mean
stdv
1.019
2.028
2.773
0.473
0.141
0.408
0.044
0.372
0.991
0.425
3.662
1.656
0.398
0.361
0.596
0.406
0.449
1.451
1.179
2.702
3.738
0.200
0.024
0.188
0.157
0.279
1.559
0.496
3.602
1.709
0.396
0.363
0.594
0.403
0.449
1.434
Below Average
Above Average
Mean
stdv
Mean
stdv
Price
Pitch
Width
Bar
AP
Screen
Sandwich
No Meal
ASC
1.083
1.623
2.500
0.351
0.138
0.322
0.010
0.407
1.206
0.466
3.527
1.441
0.416
0.359
0.603
0.398
0.440
1.456
1.078
3.309
4.126
0.375
0.021
0.307
0.209
0.221
1.258
0.456
3.519
1.717
0.429
0.360
0.607
0.397
0.441
1.450
Gender Variable
Male
Price
Pitch
Width
Bar
AP
Screen
Sandwich
No Meal
ASC
Education Variable
Price
Pitch
Width
Bar
AP
Screen
Sandwich
No Meal
ASC
Female
Mean
stdv
Mean
stdv
1.293
2.997
2.652
0.322
0.116
0.463
0.168
0.272
1.366
0.480
3.575
1.645
0.420
0.359
0.586
0.400
0.446
1.456
0.824
1.513
3.826
0.409
0.058
0.137
0.003
0.406
1.071
0.298
3.590
1.668
0.415
0.364
0.585
0.399
0.448
1.459
School
Higher
Mean
stdv
Mean
stdv
1.065
2.491
2.450
0.447
0.073
0.303
0.096
0.301
1.201
0.452
3.631
1.139
0.392
0.365
0.609
0.406
0.449
1.470
1.152
1.794
5.598
0.079
0.147
0.373
0.092
0.422
1.331
0.489
3.617
1.196
0.386
0.365
0.604
0.410
0.449
1.461