Вы находитесь на странице: 1из 9

Transportation Research Part A 75 (2015) 42–50

Contents lists available at ScienceDirect

Transportation Research Part A

journal homepage: www.elsevier.com/locate/tra

Using DEA models to jointly estimate service quality perception

and profitability – Evidence from international airports
Rico Merkert a,⇑, A. George Assaf b
Institute of Transport and Logistics Studies, The University of Sydney, Sydney, NSW 2006, Australia
Isenberg School of Management, University of Massachusetts, Amherst, MA 01003, USA

a r t i c l e i n f o a b s t r a c t

Article history: With increasing competition between airports and a growing share of non-aeronautical
Received 21 August 2014 revenues, particularly at large international hubs, perceived quality and consumer satisfac-
Received in revised form 6 February 2015 tion have not only become a key focus of management but potentially the most important
Accepted 5 March 2015
tool to achieve or maintain a competitive advantage. However, in the long run stakeholders
aim for profit maximisation and it has been shown for other parts of the aviation supply
chain that quality does significantly impact on profitability. In this paper we aim to
investigate whether perceived airport quality has an impact on airport profit margins.
Service quality
We further apply two-stage Data Envelopment Analysis (DEA) models to estimate a single
Profitability efficiency measure that combines the potentially conflicting indicators of perceived service
Data Envelopment Analysis quality and profitability for the airport context. We also identify determinants of the jointly
Airports estimated single quality/profitability measure.
Ó 2015 Elsevier Ltd. All rights reserved.

1. Introduction

Management theory would argue that profitability will follow once firms offer goods or services at optimal quantities and
more importantly high quality levels. The concepts of customer service, perceived service quality, and customer satisfaction
are of importance in this context. Consistent levels of customer services will lead to perceived high levels of service quality
which should result in stronger firm-consumer relationships and therefore greater customer satisfaction. This, in turn, will
translate into higher willingness to spend and greater repeat business (Sterling and Lambert, 1989). That this is however no
guarantee for improved profitability has been shown in the context of airlines (Merkert and Pearson, 2015).
One of the reasons for increased complexity in the service quality – profitability relationship is that the quality of a
consumer’s overall experience depends on the action of every element within the industry’s value chain (Wright and
Race, 2004), thereby suggesting mutual support and understanding of each industry actor despite often conflicting objectives
(Shaw, 2011).
Airports are a great example of the complexity of such value chains, as they are rapidly moving away from just being
gateways for travellers and cargo, to large shopping and logistics hubs and often even mini cities, also referred to as the
airport metropolis business model (Baker and Freestone, 2010). This means that they may have not only a conflict between
the overall profitability and offered service quality but a number of such conflicts over the multiple market segments that
many large airports cover, including passenger, freight and non-aeronautical activities.

⇑ Corresponding author. Tel.: +61 2 9114 1883; fax: +61 2 9351 0088.
E-mail address: rico.merkert@sydney.edu.au (R. Merkert).

0965-8564/Ó 2015 Elsevier Ltd. All rights reserved.
R. Merkert, A.G. Assaf / Transportation Research Part A 75 (2015) 42–50 43

This paper aims to first establish whether there is a correlation between airport profitability and perceived service quality
levels to then for the first time develop a joint indicator that combines both with the traditional airport objectives of growing
passenger and freight traffic (aeronautical income). We also aim to evaluate the impact of various factors, such as share of
low cost operations and share of non-aeronautical activities on this combined profitability/quality/traffic airport perfor-
mance indicator. The remainder of the paper is a follows: Section 2 sets the background and presents gaps in the literature.
The methodology will be presented in Section 3 which is followed by a discussion of the results in Section 4 and concluding
remarks in Section 5.

2. Setting the scene and review of the literature

Although the profitability of airports is on average significantly higher and more stable than that of airlines (Pearce,
2013), there is no doubt that also airports face profit margin performance pressures particularly from their owners or
administrators. Those pressures have intensified since the deregulation (Bel and Fageda, 2010) and/or privatisation
(Basso, 2008) of large and medium sized airports in many parts of the world. Given the large sunk cost that result from
the capital intensive nature of operating airports (runways, terminals and other large scale facilities), airports have little
choice but to try and benefit from economies of scale (Bazargan and Vasigh, 2003) and if possible economies of density if
the airport is in a geographical position that allows for hub-operations. The geographical position and business model will
also determine the level of competition and hence the efficiency (which should include quality) and profitability of an airport
(Merkert and Mangia, 2014).
In terms of business model consideration the quality/profitability balance will also be determined by the careful calibra-
tion of the airport strategy in regards to balancing passenger and freight operations (for a further discussion on potentially
different objectives and overlaps of the two see, Merkert and Ploix, 2014). But even within the air passenger context, there
are significant differences in business models as airports apply differentiated price and service levels to various market seg-
ments. For instance, as shown by Zuidberg (2014) hub airports will focus more on transfer and international passengers
while regional airports will favour domestic and regional passengers in their pricing strategies.
And then there is the aspect of aeronautical revenues/profits versus non-aeronautical activities. Airports increasingly
recognise the importance of improving the quality of passenger experience as this will not only result in higher cus-
tomer satisfaction but also in higher non-aeronautical revenues (e.g. Graham, 2008). In particular within the context
of at large international hub airports, the focus on commercial (non-aeronautical) revenues increase singifcantly over
the last decade, with Atlanta, Frankfurt International, Paris Charles de Gaulle, Singapore Changi, Hong Kong
International and Seoul Incheon airport all featuring a share of non-aeronautical revenues in total revenues of more than
60% in 2012 (ATRS, 2014). This is not only important because it prompts the airports with the choice of focusing on
improving the aeronautical and/or non-aeronautical consumer experiences but also because it allows airports to optimise
the two revenue/profit streams and even cross-subsidise aeronautical charges with profits from non-aeronautical sources
(for evidence on the latter see, Choo, 2014). One could of course argue that many non-aeronautical activities at airports
are outsourced (typically food and beverages, speciality shops or duty free) but the level of outsourcing differs and inter-
estingly the most profitable non-aeronautical activity is usually not outsourced namely, car parks (see Ison et al., 2014).
As such, car parks are under the direct control of airport management, at both strategic and day-to-day operational
And then there are Low Cost Carrier (LCCs) operations at both the airline and airport level, which again make the quality/
profitability relationship yet again an even more complex one. LCCs in the airline context, with value propositions based on
offering the best (average) price for standardised services and good enough quality, performance, and style (Ryans, 2009),
ordinarily avoid inclusive attributes which has implications on the airport-airline-passenger relationship (Francis et al.,
2004). The LCC has also been applied in the airport context as many now offer low cost terminals with no or very little service
attributes (such as lounges or luxurious shopping or food and beverages facilities).
All this shows that measuring airport performance is complex and that quality and profitability matter at different levels.
Existing airport performance measurement studies are getting increasingly international in scope and comprehensive in the
number of input/output variables and deployed benchmarking techniques (i.e. ATRS, 2014). Merkert et al. (2012) provide a
broad overview of the literature on performance modelling in the airport industry. The two most common methodologies
have been the Data Envelopment Analysis (DEA) and DEA bootstrap methods (e.g. Gillen and Lall, 1997; Graham and
Holvad, 1997; Sarkis, 2000; Pels et al., 2001; Lin and Hong, 2006; Barros and Dieke, 2008; Barros and Weber, 2009; Curi
et al., 2010; Assaf, 2010, 2011; Tsekeris, 2011; Gitto and Mancuso, 2012; Wanke, 2012; Merkert and Williams, 2013) with
few other studies also using the stochastic frontier analysis (SFA) method (Oum et al., 2008; Martín et al., 2009; Assaf and
Gillen, 2012; Scotti et al., 2012). For more details on the characteristics of the above literature, refer to Merkert et al. (2012).
Overall, we found that while the literature is rich, none of the existing studies investigated the efficiency of airports in
combination with perceived service quality.
In parallel there are increasingly sophisticated airport quality rankings published (such as by ACI or Skytrax), but there is
no study allowing service quality and profitability to be integrated into a more comprehensive ‘‘efficiency’’ measure. As the
different airport management objectives may conflict, it is of interest to try and combine original technical efficiency
measures with profit margins and perceived quality service levels.
44 R. Merkert, A.G. Assaf / Transportation Research Part A 75 (2015) 42–50

3. Methodology

This paper aims to provide a more comprehensive performance indicator that is different from the literature in terms of
accounting for service quality. The literature so far has accounted for service quality through some indirect indicators in par-
ticular profitability. However, as mentioned above, the two are different and should not be merged together. In other words,
we need a separate output measure for service quality. To provide further evidence to this, we aim first to evaluate whether
for our analysed 30 airports there is a correlation between perceived airport service quality levels and airport profitability.
For instance, a strong correlation would indicate that there is no need to separate between the two (profitability and quality).
To proxy profitability we use MARGINi which is the operating (EBITDA) margin of the ith airport in our sample. By using a
ratio rather than monetary values we avoid distortions that are commonly associated with currency conversions. For per-
ceived airport service quality we consider two variables SKYTRAXi (ranking score 0–5) and PAX_REVIEWSi (with 10 as the
highest and 0 the lowest possible scores). While SkyTrax reviews are determined by an independent UK based industry body
PAX_REVIEWSi are entirely determined by customer/consumer views/experiences (our data set consists of on average 248
observations per airport).
The correlation results indicate that for our sample1 the SKYTRAXi and MARGINi correlation coefficient is r = 0.22 (p = 0.209).
The PAX_REVIEWSi and MARGINi correlation coefficient is with r = 0.32 only marginally higher. This confirms that (as for airlines
as shown in Merkert and Pearson, 2015) for our analysed airports there is no significant relationship between either indicator of
perceived service quality and profitability.2 Hence, this provides further support to our argument that service level perception
has very little, if any, impact upon airport profitability, at least in the short-to-medium term. In other words not having quality
included in the efficiency analysis as a separate output may distort the efficiency results (i.e. in other words, profitability should
not be used as a proxy for quality). We therefore attempt to develop for the first time an efficiency measure that includes both
quality and profitability as well as other common airport outputs (e.g. passenger numbers (PAX), cargo tonnes lifted (Cargo) and
air traffic movements (ATM)) to then evaluate what airport management could do in order to improve both or optimise its effi-
ciency. Hence, the goal is to develop such a single efficiency measure and to benchmark our airport sample simultaneously on
perceived service quality levels, profitability, and traffic output by applying a number of DEA models.
Similar to Merkert and Hensher (2011), we apply a two-stage DEA efficiency approach to try and compute not only the
individual airport efficiency scores but also determinants of airport efficiency. In the first stage we apply a number of
different non-parametric bootstrapped (and also non-bootstrapped/original) Data Envelopment Analysis (DEA)3 models to
compute the efficiency of the relevant airports. DEA is a linear programming technique (Charnes et al., 1978) for evaluating
the relative performance of individual organisations or decision making units (in our case airports) based on observed data.
The DEA specifications focus on two important choices, firstly input versus output orientation,4 and secondly whether constant
returns to scale (CRS) or variable returns to scale (VRS) are assumed. The mathematical formulation for the input-oriented DEA
model can be expressed as follows:
(  )
 Xn Xn Xn
^hi ¼ min hy 6 ki yi ; hxi P ki xi ; ki ¼ 1 ð1Þ
 i i¼1 i¼1 i¼1

where yi represents an M  1 vector of outputs, xi is an N  1 vector of inputs, hi is a scalar and k is a I  1 vector of constants.
The hi’s represent the technical efficiency scores, and ranges between zero (i.e. least efficient) and one (i.e. most efficient).
The term ni¼1 ki ¼ 1 is added to account for variables returns to scale (VRS).
The output oriented DEA model can be expressed as:
(  )
 Xn Xn Xn
^hi ¼ max hxi P ki xi ; hxi 6 ki yi ; ki ¼ 1 ð2Þ
 i¼1 i¼1 i¼1

where all terms are as identified above but the model here seeks output maximisation rather than input minimisation.
At this point it is worth pointing out that in this paper we refer to technical efficiency when talking about the efficiency of
airports, and are therefore interested in the optimal use of resources in the production process (maximum output from a
given set of inputs or minimising inputs from a given set of outputs) which is similar to the concept of productivity.
While we initially attempted to also estimate allocative and cost efficiency, data availability and a lack of international
comparability of the relevant input prices did not allow yielding robust results from such analysis.
In terms of orientation, there is an on-going debate in the literature on whether input or output oriented models are more
appropriate for the aviation sector. While the output oriented model is beneficial as it assumes that airports have more
influence on the outputs (as modifying runways or terminal is difficult at least in the short to medium run), one could argue
that the input oriented model may still be valid as airport outputs are often heavily impacted by endogenous factors. Output

As we identified in a first step a strong correlation (r = 0.82, p = 0.00) between SKYTRAXi and PAX_REVIEWSi, we decided not to use both of these variables in
the same regression model but use correlation coefficients instead to evaluate whether there is a relationship between perceived quality and profitability.
Note that as this correlation analysis is performed at the sample level, it refers to averages rather than the individual airport level. Airports might therefore
still find themselves in a situation where profitability and service level provision become conflicting objectives in the airport’s strategic management decisions.
Since the bootstrapped algorithm for DEA is now well established in the literature, we refer to Simar and Wilson (1998, 2007) for more details.
For a detailed discussion on the choice of orientation refer to Chambers et al. (1996) and Coelli et al. (2005).
R. Merkert, A.G. Assaf / Transportation Research Part A 75 (2015) 42–50 45

traffic volumes are for example in many cases significantly affected by economic conditions or long-term slot contract inter-
ests. We briefly tested the input oriented model but as its results were inferior and because both of our infrastructure inputs
are rather fixed assets, we decided to focus on the output oriented model. We argue that airports will find it naturally easier
to adjust outputs such as quality compared to the length of their runways or the size of their terminal building floor space
(at least in the short to medium run).
In terms of the returns to scale model specification, we focus on the variable returns to scale (VRS) model, as the CRS
model would require assuming that all airports in our sample would produce at the optimal size (which is in our view
unrealistic given the level of regulation and other constraints).
The DEA scores are estimated using the Coelli et al.’s (2005) optimisation approaches for output (max) orientation. The
control/base DEA model (TEBase), includes no quality indicators. For all other models we use SKYTRAXi and alternatively
PAX_REVIEWSi as proxies for perceived service quality and in addition to MARGINi we use the traditional variables PAXi,
CARGOi and ATMi as output factors of the airport production process. RUNWAY_LENGTHi (as the sum of all the length of all
operational runways at airport i), TERMINAL_SIZEi (as the sum of the floor size of all terminals at airport i in m2) and FTEi
are used as the inputs. By doing so, our inputs represent physical proxies for the capital/labour trade-off in airport opera-
tions. We follow the literature as these models are essentially the traditional models in airport DEA analysis but extended
by profitability and perceived quality indicators. As shown in Table 1, in a further round of DEA analysis (e.g. TESky2) we then
focus on MARGINi and SKYTRAXi (and alternatively PAX_REVIEWSi) as the only two outputs.
This allows the model to optimise the trade-off between the two key management objectives that are of relevance to most
firms, profitability and perceived quality. Another benefit of these models is that they focus more on MARGINi (rather than
pure operational outputs) which covers all areas of airport activity, including non-aeronautical activities. Methodologically
we are interested in evaluating whether such model specifications produce meaningful results. Since DEA results are gener-
ally sensitive to the number of inputs and outputs, all of our models follow Banker et al. (1989) by ensuring that the number
of inputs and outputs do not exceed one third of the number of observed units (airports).
As technical efficiency is about measuring the output to inputs ratio (e.g. see Banker et al., 1984) it should primarily con-
sist of physical rather than financial measures. In that sense it is preferable to use operating margins (a ratio) rather than
revenues or operating profits which are both associated with problems of comparability (currency conversion, etc.).
The first-stage DEA analysis is followed by five second-stage truncated regression models which seek to evaluate of sev-
eral variables on our single (combined profitability/service level/PAX/Cargo/ATM) technical efficiency scores (as mentioned
above we use VRS scores only). For this we use the following truncated regressions (truncated at 0 as a result of the non-
negativity of the efficiency scores):

yi ¼ a þ b1 NONAERO SHAREi þ b2 OWNERi þ b3 LCC SHAREi þ b4 ASIAPi þ v i ð3Þ

where yi represents in our first model TESkyboot (the bootstrapped output oriented SkyTrax VRS scores that are based on the full
set of output variables) and then all other models being variations of the first, including the efficiency scores where use
profitability and quality as outputs only. We focus mainly on the bootstrapped models but use both the original and bias
corrected efficiency scores for the most robust model (TESky2 boot and TE
) as the debate on the value of bootstrapping is still
on-going. We use the smoothing homogenous bootstrap approach with 2000 iterations to overcome the potential problem of
biased results in our second-stage regressions (Simar and Wilson, 2000, 2007, 2008). Initially, we included GATESi in the sec-
ond-stage regressions but later dropped it as it is highly correlated with TERMINAL_SIZEi. We also dropped the share of inter-
national passengers in total passengers as it was insignificant in all models. In terms of the potential impact factors we focus
therefore on the share of non-aeronautical revenues in total revenues (NONAERO_SHAREi), the share of LCC airline operations
(LCC_SHAREi) and as with Assaf and Gillen (2012), we postulate that ownership (OWNERi) will have an impact on the joint
performance indicator. We also evaluate whether airports in the Asia–Pacific region (ASIAPi) perform differently to the rest of
the sample.
As shown in Table 2, the sample is consisting of airports that all have at least some international traffic and most of them
are hubs. There is still significant variation with regards to all of the chosen input and output variables, which is positive as
we aim to reveal differences in performance as well as factors that impact on the combined performance indicator. All air-
ports have a positive margin, which is most useful as one of the conditions for DEA models is the non-negativity of input and
output factors.

4. Sample

As shown in Table 3, our sample consists of 30 international airports, all of medium to large size with hub character. The
airports have been chosen based on which airports had a Skytrax ranking in 2013. We managed to collect the data (other
input/output variables) for all but three (Amman Queen Alia Int. airport, Bahrain Int. airport, Kuwait Int. airport) of those
33 airports that had a ranking in 2013. We believe that our sample has a unique international coverage that goes beyond
the ATRS (2014) study. Table 3 also provides the IATA airport codes for the analysed airports, which is useful as the results
will use those codes only.
We have also provided rankings of the airports based on a range of proxies of the size of the analysed airports and it
becomes immediately apparent that there are stark differences in those rankings depending which proxy is chosen. This
46 R. Merkert, A.G. Assaf / Transportation Research Part A 75 (2015) 42–50

Table 1
DEA models and input/output combinations.

TEBase TESky TEPaxRev TESky2 TEPaxRev2

boot TESky
boot TEPaxRev
boot TESky2
boot TEPaxRev2

Inputs (all end 2012)

Outputs (all 2013)
PAX (000s)   
CARGO (000s tonnes)   
ATM (000s)   
SKYTRAX (1–5)  

Note: ‘‘boot’’ refers to bootstrapped.

Table 2
Descriptive statistics of the DEA efficiency models and regressions. Source: Own analysis based on data from IATA, ICAO, Flightglobal, CAPA, Skytrax, airport
accounts and data provided by the airports.

N Mean St.D. Min Max

Inputs (all end 2012)
RUNWAY_LENGTH (total m) 30 9819.1 3501.8 3600 19,467
TERMINAL_SIZE (000s m2) 30 613.2 423.9 99.3 1707.8
FTE 30 2990.8 3635.6 177 18,939
Outputs (all 2013)
PAX (000s) 30 44676.2 21134.3 11935.5 94431.2
CARGO (000s tonnes) 30 1159.2 992.7 110.8 4122.0
ATM (000s) 30 341.7 161.3 82.9 911.1
MARGIN 30 44.7 14.9 5.8 68.8
SKYTRAX (1–5) 30 3.8 0.6 2.5 5
PAXREVIEWS (1–10) 30 6.9 1.4 3.7 9.2
Second stage regression variables
NONAERO_SHAREi 30 49.1 14.4 22.4 72
OWNERi (1 = private) 30 0.33 0.47 0 1
LCC_SHAREi (in% of total seats) 30 13.9 13.3 0.9 51.3
ASIAPi (1 = yes) 30 0.5 0.5 0 1
Share of int. pax. (%) (dropped) 30 63.3 35.1 4 100
GATESi (dropped) 30 118.9 58.9 17 226

suggests that the ranking and also partial productivity measures will fail to portrait a comprehensive picture of the airports’
overall performance. Atlanta will for example perform well in any passenger related productivity measure but will fail to
impress once cargo is included in the analysis. When it comes to inputs Dubai has for example the largest terminal area
available but not that much total runway length. It should be noted that in terms of inputs Frankfurt airport outlier with
more than 18,000 employees. Another outlier is Dubai International airport as the profit margin we use is an approximation
of data from previous years and interviews with management of the airport company (the airport does not officially report
any financial data).

5. Results

As illustrated in Table 4, the results of first-stage DEA models suggest that our developed single profitability/perceived
service quality /traffic efficiency indicator produces results that are within the expected range (see for example, Assaf,
2010). As the debate on bootstrapping in DEA is still on-going, we present both the original TE and the bootstrapped (bias
corrected) TEboot scores. As we use an output-oriented approach the efficiency scores a bound at the minimum of 1 and the
closer the values are to 1 the more efficient the airport.
Our results show clearly that the original results overestimate airport efficiency in all models. That the TESky scores are on
average better (closer to 1) than the TESky2 and the TEBase scores, supports our argument of quality indicators being of impor-
tance in efficiency analyses but has also been interpret carefully as DEA is sensitive to the number of outputs and inputs
used.5 Comparing the scores of TEBase with TESky or TESky2 (and TEPaxRev with TEPaxRev2 respectively) by using the ranking shown
in Table 5 is hence more meaningful than just using the overall average scores of each model.

For example, whenever one adds another output variable (in an output oriented model), the decision making units get another chance to be relatively more
efficient than their peers and hence the more output variables the higher the likelihood of a higher average efficiency score.
R. Merkert, A.G. Assaf / Transportation Research Part A 75 (2015) 42–50 47

Table 3
Sample with airport codes and proxies of size. Source: Own analysis based on data from IATA, ICAO, Flightglobal, CAPA, ATRS, airport accounts and data provided
by the airports. Note that passenger, cargo, ATM and runway data is from 2013, while Terminal size, gates and FTE relates to a snapshot at the end of 2012.

Airport IATA code Rankings (proxies of the size of the airports)

Pax Cargo ATM RunwayL (#⁄m) Terminal size (m2) Gates FTE
Amsterdam Airport Schiphol AMS 11 10 7 1 13 5 15
Atlanta Hartsfield-Jackson Int. Airport ATL 1 18 1 4 12 3 22
Bangkok Suvarnabhumi Airport BKK 13 13 18 20 14 14 10
Beijing Capital Int. Airport PEK 2 9 2 10 3 13 16
Brussels Airport BRU 26 20 25 13 27 15 25
Copenhagen Katrup Airport CPH 24 22 22 12 25 16 13
Dubai International Airport DXB 5 4 14 17 1 6 9
Düsseldorf International Airport DUS 25 30 26 29 24 21 12
Frankfurt am Main Airport FRA 9 5 4 5 4 9 1
Guangzhou Baiyun Int. Airport CAN 12 12 10 24 20 22 8
Haikou Meilan Int. Airport HAK 30 28 29 30 30 29 27
Haneda Airport (Tokyo Int.) HND 4 15 8 9 15 27 30
Helsinki Vantaa Airport HEL 29 29 30 15 28 28 28
Hong Kong International Airport HKG 8 1 12 21 9 25 19
Houston George Bush Intercont. Airport IAH 17 21 3 2 19 18 23
Jakarta Int. Airport Soekarno-Hatta CGK 7 26 9 25 22 12 14
Johannesburg OR Tambo Int. Airport JNB 27 24 27 19 21 30 21
Osaka Kansai Int. Airport KIX 28 17 28 23 23 24 29
Kuala Lumpur Int. Airport KUL 14 16 17 16 16 11 11
London Heathrow Airport LHR 3 11 6 22 11 8 6
Madrid Barajas Airport MAD 18 23 16 3 6 1 20
Moscow Sheremetyevo Int. Airport SVO 22 25 23 26 17 23 3
Munich Airport MUC 19 27 11 18 18 4 2
Narita (Tokyo) Int. Airport NRT 20 7 24 28 7 10 26
Paris Charles de Gaulle Airport CDG 6 6 5 6 2 2 4
Seoul Incheon Int. Airport ICN 16 3 19 7 10 7 24
Shanghai Hongqiao Int. Airport PVG 15 2 13 8 8 20 5
Shenzhen Bao’an Int. Airport SZX 21 14 21 27 29 26 7
Singapore Changi Airport SIN 10 8 15 11 5 19 17
Zurich Airport ZRH 23 19 20 14 26 17 18

Table 4
Aggregated first-stage DEA results.

Original Bootstrapped
TEBase 1.12935 1.254851
TESky 1.077016 1.163477
TEPaxRev 1.084864 1.172153
TESky2 1.212118 1.344677
TEPaxRev2 1.237875 1.378865

Note: The above efficiency scores are computed under the assumption of variable returns to scale (VRS) and are estimated with
the software package FEAR 2.0 (Wilson, 2010).
Values in bold indicate the best model.

The important issue is that ignoring quality has clearly distorted the ranking of some airports. For example, we see that
the very high quality scores of HKG (3rd in quality) or SIN (2nd in quality) are not reflected in the base efficiency results and
similarly BRU airport which is not even in the top 20 in terms of quality ranks third in the base model (TEBase boot ) clearly sug-
gesting that quality should be considered when assessing the overall performance of airports. This also supports and pro-
vides first evidence to previous arguments by Assaf et al. (2014) who emphasised that ignoring quality from efficiency
estimation will simply fail to give credits to airports for improving their quality standards over time.
From the results presented in Table 5 it can be further seen that the focus on profitability and perceived quality as the
only two outputs in model TESky2 boot has explanatory value. Atlanta ranked, for example, highly for a number of proxies of air-
port size, traditional airport outputs (as shown in Table 3) and hence also in the efficiency scores coming out of the TESky
model. That such original non-bootstrapped TE scores can be misleading can be best seen in the Atlanta case. While it is
ranked number one for TESky (due to the top 17 all having a score of 1 and the alphabetical order), bootstrapping overcomes
that problem as with model TESky Sky2
boot Atlanta is suddenly ‘‘only’’ ranked 17. As model TEboot is then solely focused on profitabil-
ity/quality, Atlanta loses its dominant position in terms of passenger output (numbers and ATM) and disappears from the top
20. The more the DEA focuses on quality as an output, the lower Atlanta’s ranking. Atlanta’s positive profit margin (ranked
16) has therefore not been enough to convince in the overall performance as it is perceived quality (quality in Table 5
equates to the average of Skytrax and PaxReviews for each airport) is inferior to those airports in top 20 quality ranking.
48 R. Merkert, A.G. Assaf / Transportation Research Part A 75 (2015) 42–50

Table 5
Top 20 airport rankings.

Atlanta airport as an example is indicated in bold.

Table 6
Second-stage regression results.

boot TESky
boot TEPaxRev
boot TESky2
boot TEPaxRev2
boot TESky2
⁄⁄⁄ ⁄⁄⁄ ⁄⁄⁄ ⁄⁄⁄ ⁄⁄⁄
Cons 1.501 1.425 1.549 1.689 1.461 1.578⁄⁄⁄
NONAERO_SHAREi 0.003 0.004⁄⁄ 0.003 0.003⁄ 0.001 0.003
OWNERi 0.025 0.042 0.035 0.132⁄ 0.072 0.152⁄
LCC_SHAREi 0.001 0.003⁄ 0.003 0.006⁄⁄ 0.004 0.009⁄
ASIAPi 0.091 0.032 0.048 0.131⁄⁄ 0.099 0.218⁄
8 left c. obs.
⁄ ⁄⁄ ⁄⁄⁄
Note: P < 0.1; P < 0.05 and P < 0.01, these represent significant p values.

The importance of quality as an indicator can also be seen at the lower end of airport performance as JNB is ranked 28th in all
TE models but 23rd in TEbootSky2 as a result of its relatively high quality of services and the focus ofmodel TEbootSky2 on
quality and profitability. Overall, it appears that despite conflicting rankings, airports which do well in both profitability and
quality do well in the combined efficiency scores, regardless of how many outputs we choose (most notably SIN).
The second-stage regression results presented in Table 6 exclude Gates which had to be dropped as a result of being
highly correlated with terminal size. Apart from that multi-colinearity was not an issue. The variable that appears to have
the most significant impact in all models is Asia Pacific, with airports in that region performing much better than their peers
from the rest of the world. In terms of the best model, we believe that TESky2boot produces the most robust results. Again, the
benefit of bootstrapping becomes apparent as the variables in the TESky2 are not only less significant but also faced with
the problem of eight observation being left censored.
What is encouraging is that the TESky2 Base
boot results suggest (in stark contrast to the base results TEboot that do not consider
quality as an output) that all variables have a significant impact on the single and bootstrapped (and hence more reliable)
efficiency measure. However, the direction of the impacts is not always as expected. As anticipated, the higher the share in
non-aeronautical revenues the more efficient (high profitability and high quality) the airport. Also that airports in Asia Pacific
perform well is not unexpected and that private ownership improves airport efficiency in all aspects has been shown
elsewhere. However, that a higher LCC share has a positive impact on our airport single efficiency measure is surprising
but may be explainable by the strong growth stories that these carriers have usually brought to the relevant airports.

6. Conclusion

This study provided for the first time a comprehensive approach to the measurement of efficiency of international air-
ports. We combined and measured profitability/quality/traffic outputs and inputs simultaneously to evaluate whether such
a single indicator can be meaningful established and whether two-stage DEA models can be used to identify factors that have
a significant impact on it. As we have achieved both of these aims, airports can now use our results for their strategic man-
agement decisions when aiming for improved overall performance.
R. Merkert, A.G. Assaf / Transportation Research Part A 75 (2015) 42–50 49

We emphasised the importance of including quality as an output measure by comparing between various DEA models,
estimated using output oriented approaches. Overall we found that excluding quality as an output measure can distort
the true overall efficiency ranking of international airports. Our attempt of combining profitability, quality and traffic as
equally important and desirable outputs in DEA models produces useful results, particularly when we apply a bootstrapping
procedure to the original efficiency scores.
In the second-stage regressions, the model that produces the most significant results is the one that is in addition to being
output oriented and bootstrapped also focusing on quality and profit margin as the only two outputs in the first-stage DEA
model (rather than including traffic output variables too). This is not unexpected as the trade-off for management is usually
between quality and profit, despite the fact that we have shown that there is no significant relationship between the two.
The results of the TESky2
boot model (the one we most believe in) suggest that based on our combined single efficiency measure
airports with the following characteristics perform well on average: located in Asia Pacific, high share of non-aeronautical in
total revenues, high share of LCC airline in total airline seats and private ownership. While the region of the airport cannot be
changed, the latter three attributes are certainly within the power of airport management (or its administrators or owners).
At this point it is worth mentioning some limitations of this research. Firstly, our quality indicators only reflect the pas-
senger side of airport operations (while profit margins also include cargo operations). Future research should aim at also
incorporating quality aspects of both passenger and cargo operations. Secondly, there has been some discussion in the media
on the reliability of Skytrax. As we have also used passenger reviews and because other quality indicators face similar con-
cerns, we feel confident enough to argue that our quality indicators are as reliable as research in this area can get. Thirdly,
some of the airports may not see profit maximisation as a priority or in other words profits as a desired output and in fact,
the two U.S. airports in our sample are not allowed to earn profits (as any U.S. airport would lose public funds as a result).
However, we assume that all airports try to at least break even and hence there is an implicit utility in loss minimisation,
which justifies using profitability as an output in the DEA models. We acknowledge further that our research is a snapshot
which may have distorting effects as many airports are not only fast growing (i.e. in the Middle East and Asia Pacific) but also
constantly adapting their business models and hence their attitude to providing specific levels of service quality. Future
research should therefore, once the data becomes available, evaluate our proposed framework within the longitudinal
context. A further possible extension of our research is to also include measures of negative outputs (e.g., bad service)
and combine these with good outputs. This way we can differentiate between negative and positive service and see how this
impacts efficiency.


The authors gratefully acknowledge the support of a number of airports who have kindly provided sensitive data and par-
ticipated in our research. We also appreciate the insightful comments of two anonymous reviewers. Their contribution has
made for a more robust paper and for that we are grateful.


Assaf, A.G., 2010. Bootstrapped scale efficiency measures of UK airports. J. Air Transport Manage. 16, 42–44.
Assaf, A.G., 2011. Bootstrapped Malmquist indices of Australian airports. The Service Indust. J. 31, 829–846.
Assaf, A.G., Gillen, D., 2012. Measuring the joint impact of governance form and economic regulation on airport efficiency. Eur. J. Operat. Res. 220 (1), 187–
Assaf, A.G., Josiassen, A., Gillen, D., 2014. Measuring firm performance: Bayesian estimates with good and bad outputs. J. Business Res. 67, 1249–1256.
ATRS, 2014. Airport benchmarking report 2014: Global Standards for Airport Excellence. Air Transport Research Society (ATRS), Sauder School of Business,
Vancouver, Canada..
Baker, D.C., Freestone, R., 2010. The airport city: a new business model for airport development. In: Macario, R., Van de Voorde, E. (Eds.), Critical Issues in Air
Transport Economics and Business. Routledge/Taylor and Francis, Oxon, UK, pp. 150–164.
Banker, R.D., Charnes, A., Cooper, W.W., 1984. Some models for estimating technical and scale inefficiencies in data envelopment analysis. Manage. Sci. 30
(9), 1078–1092.
Banker, R.D., Charnes, A., Cooper, W.W., Swarts, J., Thomas, D., 1989. An introduction to data envelopment analysis with some of its models and their uses.
Res. Govern. Nonprofit Account. 5, 125–163.
Barros, C.P., Dieke, P., 2008. Measuring the economic efficiency of airports: a Simar–Wilson methodology analysis. Transport. Res. Part E: Logistics Transport.
Rev. 44, 1039–1051.
Barros, C.P., Weber, W.L., 2009. Productivity growth and biased technological change in UK airports. Transport. Res. Part E: Logistics Transport. Rev. 45, 642–
Basso, L.J., 2008. Airport deregulation: effects on pricing and capacity. Int. J. Indust. Organ. 26 (4), 1015–1031.
Bazargan, M., Vasigh, B., 2003. Size versus efficiency: a case study of US commercial airports. J. Air Transport Manage. 9 (3), 187–193.
Bel, G., Fageda, X., 2010. Privatization, regulation and airport pricing: an empirical analysis for Europe. J. Regulatory Econom. 37 (2), 142–161.
Chambers, R.G., Chung, Y., Färe, R., 1996. Benefit and distance functions. J. Econom. Theory 70, 407–419.
Charnes, A., Cooper, W.W., Rhodes, E.L., 1978. Measuring the efficiency of decision making units. Eur. J. Operation. Res. 2 (6), 429–444.
Choo, Y.Y., 2014. Factors affecting aeronautical charges at major US airports. Transport. Res. Part A: Policy Practice 62, 54–62.
Coelli, T.J., Rao, P.D.S., O’Donnell, C.J., Battese, G.E., 2005. An Introduction to Efficiency and Productivity Analysis. Springer, New York.
Curi, C., Gitto, S., Mancuso, P., 2010. The Italian airport industry in transition: a performance analysis. J. Air Transport Manage. 16, 218–221.
Francis, G., Humphreys, I., Ison, S.G., 2004. Airports’ perspectives on the growth of low-cost airlines and the remodelling of the airport–airline relationship.
Tourism Manage. 25 (4), 507–514.
Gillen, D., Lall, A., 1997. Developing measures of airport productivity and performance: an application of data envelopment analysis. Transport. Res. Part E:
Logistics Transport. Rev. 33, 261–273.
Gitto, S., Mancuso, P., 2012. Bootstrapping the Malmquist indexes for Italian airports. Int. J. Product. Econom. 135, 403–411.
Graham, A., 2008. Managing Airports: An International Perspective, third ed. Butterworth-Heinemann, Oxford.
50 R. Merkert, A.G. Assaf / Transportation Research Part A 75 (2015) 42–50

Graham, A., Holvad, T., 1997. Efficiency variations for European and Australian airports. In: EURO XV/INFORMS joint International Conference, Barcelona.
Ison, S., Merkert, R., Mulley, C., 2014. Policy approaches to public transport at airports – some diverging evidence from the UK and Australia. Transport
Policy 35, 265–274.
Lin, L.C., Hong, C.H., 2006. Operational performance evaluation of international major airports: an application of data envelopment analysis. J. Air Transport
Manage. 12, 342–351.
Martín, J.C., Román, C., Voltes-Dorta, A., 2009. A stochastic frontier analysis to estimate the relative efficiency of Spanish airports. J. Product. Anal. 31, 163–
Merkert, R., Hensher, D.A., 2011. The impact of strategic management and fleet planning on airline efficiency – a random effects Tobit model based on DEA
efficiency scores. Transport. Res. Part A: Policy Practice 45 (7), 686–695.
Merkert, R., Mangia, L., 2014. Efficiency of Italian and Norwegian airports: a matter of management or of the level of competition in remote regions?.
Transport. Res. Part A: Policy Practice 62, 30–38.
Merkert, R., Pearson, J., 2015. A non-parametric efficiency measure incorporating perceived airline service levels and profitability. J. Transport Econom.
Policy 49 (2), 261–275.
Merkert, R., Ploix, B., 2014. The impact of terminal re-organisation on belly-hold freight operation chains at airports. J. Air Transport Manage. 36, 78–84.
Merkert, R., Odeck, J., Bråthen, S., Pagliari, R., 2012. A review of different benchmarking methods in the context of regional airports. Transport Rev. 32 (3),
Merkert, R., Williams, G., 2013. Determinants of European PSO airline efficiency – Evidence from a semi-parametric approach. J. Air Transport Manage. 29,
Oum, T.H., Yan, J., Yu, C., 2008. Ownership forms matter for airport efficiency: a stochastic frontier investigation of worldwide airports. J. Urban Econom. 64,
Pearce, B., 2013. Profitability and the air transport value chain – an analysis of investor returns within the airline industry and its supply chain. IATA
Pels, E., Nijkamp, P., Rietveld, P., 2001. Relative efficiency of European airports. Transport Policy 8, 183–192.
Ryans, A., 2009. Beating Low-Cost Competition: How Premium Brands can Respond to Cut Price Rivals. John Wiley & Sons, Chichester.
Sarkis, J., 2000. An analysis of the operational efficiency of major airports in the United States. J. Operations Manage. 18, 335–351.
Scotti, D., Malighetti, P., Martini, G., Volta, N., 2012. The impact of airport competition on technical efficiency: a stochastic frontier analysis applied to Italian
airport. J. Air Transport Manage. 22, 9–15.
Shaw, S., 2011. Airline Marketing and Management. Ashgate Publishing, Aldershot.
Simar, L., Wilson, P.W., 1998. Sensitivity analysis of efficiency scores: how to bootstrap in nonparametric frontier models. Manage. Sci. 44 (1), 49–61.
Simar, L., Wilson, P.W., 2000. A general methodology for bootstrapping in non parametric frontier models. J. Appl. Stat. 27 (6), 779–802.
Simar, L., Wilson, P.W., 2007. Estimation and inference in two-stage, semi-parametric models of production processes. J. Econom. 136, 31–64.
Simar, L., Wilson, P.W., 2008. Statistical inference in nonparametric frontier models: recent developments and perspectives. In: Fried, H.O., Lovell, C.A.K.,
Schmidt, S.S. (Eds.), The Measurement of Productive Efficiency and Productivity Change. Oxford University Press, New York, pp. 421–522.
Sterling, J.U., Lambert, D.M., 1989. Customer service research: past, present, and future. Int. J. Phys. Distrib. Logistics Manage. 19 (2), 2–23.
Tsekeris, T., 2011. Greek airports: efficiency measurement and analysis of determinants. J. Air Transport Manage. 17, 140–142.
Wanke, P.F., 2012. Efficiency of Brazil’s airports: evidences from bootstrapped DEA and FDH estimates. J. Air Transport Manage. 23, 47–53.
Wilson, P.W., 2010. Package ‘FEAR’ – Frontier Efficiency Analysis with R. Department of Economics, Clemson University, Clemson.
Wright, J.N., Race, P., 2004. The Management of Service Operations, second ed. Thomson Learning, London.
Zuidberg, J., 2014. Key drivers for differentiated airport passenger service charges. J. Transport Econom. Policy 48 (Part 2), 279–295.