Vous êtes sur la page 1sur 13

Tourism Management 32 (2011) 7587

Contents lists available at ScienceDirect

Tourism Management
journal homepage: www.elsevier.com/locate/tourman

Efciency of travel agencies: A case study of Alicante, Spain


Ramon Fuentes*
Faculty of Economics, Department of Applied Economic Research, University of Alicante, Ap. Correos 99, E-03080. Spain

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

Article history: This study analyses the relative efciency of 22 travel agencies of similar characteristics based in Alicante
Received 9 December 2008 (Spain). This analysis is carried out using the Data Envelopment Analysis (DEA) technique and smoothed
Accepted 2 December 2009 bootstrap. Following the analysis, possible lines of action that the agencies can take in order to improve
their efciency in the future are provided. Finally, using the Mann Whitney U Test, the relationship, or
Keywords: lack thereof, between the levels of efciency of these agencies and their ownership type, location and
Travel agencies
level of experience is examined.
Efciency
2009 Elsevier Ltd. All rights reserved.
Data envelopment analysis
DEA
Smoothed bootstrap

1. Introduction proportion of agencies per inhabitant was 0.05% in Alicante,


compared with 0.022% for the whole of Spain (County Council of
The growing amount of competition in the economy over the Alicante, 2009; National Statistics Institute, 2009). A setting with
last few years has stimulated interest in the analysis and assess- such a high level of competitiveness was therefore a very suitable
ment of efciency in all economic sectors. The services sector is no choice for the study of efciency in this sort of agency.
exception, despite the fact that its unique characteristics (such as Then, based on the results of this analysis, it will assess the way in
intangibility or heterogeneity of its outputs) mean that it is difcult which certain characteristics of the agencies may affect their ef-
to assess and quantify its efciency (McLaughlin & Coffey, 1990). ciency parameters. In particular, the study aims to ascertain whether
Travel agencies work in the services sector, and more specically or not the agencys ownership type, location or the length of time it
in the tourism sector. Given the importance and global scope of has been in operation are factors which affect its level of efciency.
todays tourism sector, it seems important to carry out an analysis of By examining these variables, the study will try to ascertain whether
the efciency of agencies whose main, albeit not only, aim is to help or not the fact that an agency forms part of a chain, is located in the
connect supply to demand. It would also be useful for any economic city centre or has many years of experience in the sector are relevant
agents with direct or indirect links to these agencies to be able to factors when it comes to improving its level of efciency.
access information about their level of efciency so that they can
make informed decisions about investment and/or management.
2. Literature review
With this in mind, this project aims to analyse the efciency of
a group of travel agencies based in Alicante (Spain). Alicante was
To date, only seven articles analysing the efciency of travel
selected for the analysis for two main reasons: rstly, because it is
agencies have been published. The rst two of these, written by Bell
an area of Spain where the tourism sector is of great importance, in
and Morey (1994, 1995), did not analyse the efciency of travel
terms of both supply and demand; and secondly, because the
agencies exactly, but that of 31 corporate travel departments. Input-
number of travel agencies in the area has increased at such an
oriented Data Envelopment Analysis (DEA) was used for these
exceptionally high rate over the last few years that the number of
studies, using the level of service provided as a representative
agencies per capita has now more than doubled the national
variable for output, whilst input variables used were the levels of
average. Between 2000 and 2007, the number of travel agencies in
travel expenditure (such as car and airline costs and hotel bills),
Alicante increased by 205.26%, while the total increase for the
labour costs, other general expenditure (space and technology
whole of Spain was just 67.41%. Furthermore, in 2007, the
costs, for example) and, nally, other environmental factors (to
illustrate a companys ability to obtain travel cost discounts).
* Tel./fax: 34 965 90 97 11. Some years later, Anderson, Lewis, and Parker (1999) used the
E-mail address: rfuentes@ua.es same data that was used in the afore-mentioned study to show the

0261-5177/$ see front matter 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.tourman.2009.12.003
76 R. Fuentes / Tourism Management 32 (2011) 7587

differences between input-oriented DEA and stochastic frontier on the calculation of the Malmquist index, breaking it down into
results. However, the constructed variables used were different. four factors. They also used a bootstrapped Tobit model. In order to
The output was the number of company trips made and the inputs calculate productivity, they used sales and prots as outputs, and
used were labour costs, the sum of air, hotel and car expenses and, wages, capital, total costs excluding wages and book value of
nally, other expenses (such as technology or occupancy costs). The premises as inputs. In the second stage of the analysis, the efcient
authors concluded that corporate travel departments are a good Malmquist scores were analysed using a Tobit model where the
investment for companies as they are highly efcient and they explicative variables were: foreign ownership of the company, the
make it possible to control the increase in travel costs. ratio of operational costs to sales, the market share of the agencies
Subsequently, another article was written by Barros and Matias and whether or not the agency belonged to a commercial chain
(2006). It examines the efciency of 25 Portuguese travel agencies which could give it access to economies of scale. The authors
using stochastic cost frontier analysis. This study includes an concluded that the level of capital, market share, control of factor
exhaustive review of previously-published articles which have costs and belonging to a group were the main factors which
applied frontier analysis to assess levels of efciency in the tourism determined efciency in the Portuguese sector.
sector (essentially hotels and travel agencies), showing that DEA is To date, no further analyses of efciency in the travel agency
the most widely-used method. sector have been carried out. However, the fact that these studies are
The variables used by Barros and Matias were selected based on often included in hospitality research literature means that it is
their availability and the fact that they had been used in previous reasonable to include other studies that have analysed efciency in
works. Specically, they used: operational costs at constant prices, this context in this review (other previously-published works with
price of labour, price of capital-stock (proxied by the ratio of excellent reviews of this topic include, for example, Barros & Dieke,
earnings to stock), price of capital-premises (proxied by the ratio of 2007; Sigala, Jones, Lockwood, & Airey, 2005 or Wober, 2006). Table
expenses in premises divided by the value of real assets), dummy 1 provides a list of works published to date that analyse efciency in
variables for Spanish companies operating in the Portuguese the tourism and hospitality sectors.
market (based on the idea that due to their recent entry into this The review of the works included in the above table reveals
market, they may have been still undergoing a convergence and relevant information with respect to the different methods that have
consolidation process), additional dummy variables for companies been used in the past to analyse the efciency of rms operating in
which had carried out mergers and acquisitions and, nally, sales at the tourism and hospitality sectors and the variables that have been
constant prices (the output). These authors concluded that the selected as inputs and outputs of these methods. Therefore, the
main factors determining efciency in the sector were capital, conclusions that may be obtained based on this analysis will be
labour, sales, and mergers and acquisitions activities. useful in determining the type of statistical procedure to follow and
In the same year, Wober (2006) analysed efciency data gath- the most appropriate variables for the analysis. With regard to the
ered in 2003 relating to 80 branch ofces of a tour operator in latter, as well as the information obtained from the bibliographical
Austria, using a DEA model which considered variable returns to review, the opinion of company managers from the sector under
scale, which was input-oriented in the rst instance, and output- analysis will also be taken into account as will the availability of the
oriented later. He also calculated levels of super-efciency in order information required (Barros & Matias, 2006).
to draw up a ranking of the efcient branch ofces. The controllable On the one hand, with respect to the statistical procedure used,
inputs used were: personnel, occupancy, marketing and other the main conclusions that can be drawn from the published litera-
variable and xed costs, the number of employees (weighted by the ture both on tourism in general and on travel agencies in particular
number of working days per year) and their average job experience. are, rstly, that DEA is the method that has been used most often to
The non-controllable inputs used were: number of residents living carry out efciency analyses, although other methods, such as
near each agency, a visibility and competitiveness index (based on stochastic frontier analysis, have been used to a lesser extent. In fact,
the size of the window display and the number of agencies nearby it may be observed that DEA was used in 86.37% of studies analyzing
in the close neighbourhood) and ease of access via car or public the efciency of DMUs and in 38.64% of these cases it was combined
transport. The outputs used were: total number of contracts, total with other methods in order to obtain additional information
turnover and contribution margin for each of the outlets. As a result regarding their behaviour (multiple regression, analysis of variance,
of the analysis, the author was able to suggest ways in which the the Mann Whitney U Test ..). Furthermore, on the whole, the kind
management scores for the agencies could be improved, both in of efciency that has been most analysed is technical efciency, in
terms of the use of inputs and the production of outputs. He also spite of the fact that other forms (such as overall, allocative or scale
provided a ranking of all the ofces studied. All of this information efciency) have also been considered on occasion.
could be used to provide practical solutions for the different Finally, the type of DEA model applied has gradually incorpo-
management targets. rated new characteristics and/or complementary methods which
Only one year later, Koksal and Aksu (2007) used input-oriented have helped authors to gather more information and results.
DEA to assess the efciency of 24 travel agencies in the city of Studies such as those conducted by Barros and Mascarenhas
Antalya (Turkey). They also employed the Mann Whitney U test to (2005), Reynolds and Thompson (2007), Sigala (2003, 2004), Sigala,
analyse the relationship between the ownership type of an agency Airey, Jones, and Lockwood (2004) or Wober and Fesenmaier
and its level of efciency, and concluded that there is no link (2004) used stepwise DEA models and/or combined DEA with
between the two variables. The authors also applied DEA to regression analysis, ANOVA and other statistical techniques in
calculate changes in the level of inputs that inefcient agencies order to enrich their results.
would have to achieve in order to become efcient. The variables On the other hand, with respect to the variables, it is not possible
were obtained through surveys and comprised: the number of staff, to arrive at conclusions which are as highly dened as those for the
the level of annual expenses, the potential level of service that they case of the statistical procedure applied. This is due to the wide
can provide (inputs) and the number of customers served (outputs). diversity of the type of DMUs analysed in the bibliographical review
Finally, Barros and Dieke (2007) analysed changes in the which means that the inputs and outputs employed are also diverse.
productivity of travel agencies based on the study of a signicant However, it is possible to make general conclusions when the
sample group of agencies operating in the Portuguese market variables are grouped into broad categories. So, if we rst focus on
between 2000 and 2004. They applied a quantitative method based the inputs, it may be afrmed that the type of variables most used as
Table 1
Analysis of efciency in the tourism and hospitality sectors.

Author/s Model DMUs/Sector Inputs Outputs Prices


Banker and Morey (1986) Input and output-oriented DEA 60 Fast food restaurants. Labour, supplies and advertising Breakfast, lunch and dinner No prices.
models with discretionary and expenditures, age of the store, sales.
non-discretionary inputs, location and existence of drive-in
technical and scale efciency, window.
CRS*, VRS**.
Hruschka (1986) DEA, CRS. 10 hospitality companies. Number of seats, labour expenses, Total revenues. No prices.
operative resources, other
operational resources.
Baker and Riley (1994) Multiple regression analysis. 20 hotels based in each of Room rate, full-time employees, Sales per full-time employee Room rate.
Germany, France and UK. number or rooms. (fte), f&b****(fte), room
occupancy, value added (fte).
Morey and Dittman (1995) Allocative input-oriented DEA. 54 owner-managed hotels Properties physical characteristics Total room revenue and Average daily
in the USA during 1993. (number of rooms), environmental average level of customer rate of
factors (unionised employees, satisfaction on two competitive sets.
average occupancy and daily rate dimensions: physical
for competitive sets) and facilities and services
controllable factors (room-division provided.
wages, prots and meals .).

R. Fuentes / Tourism Management 32 (2011) 7587


Johns, Howcroft, and Drake (1997) Input-oriented DEA. 15 hotels (same chain). Rooms, nights available, total Rooms sold, total covers served, No prices.
labour hours, total f&b costs, total total beverage revenue.
utilities costs.
Morey and Dittman (1997) DEA, VRS, prot maximization- Hotels. Exogenous factors: near airport, Expected annualised prots No prices.
oriented and regression unionised employees, cost of (total room sales f&b
analysis. labour, competition, occupancy sales miscellaneous sales 
rates, competitors average rates, annual operating costs 
cost of acquiring the site. amortised building costs).
Endogenous: brand, number
and mix of rooms, f&b
capacities, other facilities,
levels of service, marketing
expenditures and supplies,
administration and repairs
expenditures.
Donthu and Yoo (1998) DEA, window analysis and 24 restaurants in the USA Store size, store manager Sales and customer satisfaction. No prices.
regression analysis. (chain) experience, store location and
promotion expenses.
Anderson, Fish, Xia, and Michello Stochastic cost frontier analysis. 48 hotel companies in the Full-time equivalent employees, Rooms, gaming, food and F&b, gaming &
(1999) USA. number of rooms, total gaming beverage revenues and other rooms. Proxies for
expenses, total f&b expenses and revenues. wages.
other expenses.
Anderson, Fok and Scott (2000) Input-oriented DEA (overall, 48 hotels in the USA in Full-time equivalent employees, Total revenues in 1994. Proxies for wages
allocative, technical, pure 1994. number of rooms, total gaming and room price.
technical and scale efciency), expenses, total f&b expenses and
CRS, VRS. other expenses.
Tarim, Dener and Tarim (2000) Input & output-oriented DEA, 21 4/5-star hotels (Antalya) Investment costs, number of Customer loyalty index, No prices.
CRS (modied restrictions). employees, administrative occupancy rate and net prot.
expenses.
Wober (2000) Input-oriented DEA, CRS. 61 hotels in Austria in 1997. Controllable: Total payroll and Total accommodation revenue, No prices.
related costs, material-types food and beverage revenue and
expenses and energy, cleaning, average bed occupancy.
maintenance, communication,
marketing and administration
costs. Uncontrollable: number
of beds, seats and
opening days.
(continued on next page)

77
Table 1 (continued)

78
Author/s Model DMUs/Sector Inputs Outputs Prices
Brown and Ragsdale (2002) Output-oriented DEA, CRS and 46 hotel brands in the USA. Complaints, service quality, Guests satisfaction and overall Average room
cluster analysis. condition and cleanliness value of the hotel chain. price.
of rooms, hotel properties
and guest rooms in
the USA.
Hwang and Chang (2003) Output-oriented DEA, CRS. 45 hotels from 1994 to Number of full-time employees, Room revenues, food and No prices.
Malmquist index, super- 1998 in Taiwan. number of guest rooms, total beverages revenue and other
efciency and ANOVA. area of meal department and revenues.
operating expenses.
Morey and Dittman (2003) Input-oriented DEA, allocative 54 owner-managed hotels Resources consumed, occupancy Revenues, prots, service- Salaries and
efciency. of a chain in the USA in rate and, unionised labour, satisfaction index, physical- competitors
1993. number or rooms. facilities-satisfaction index. average daily rate.
Sigala (2003) Output-oriented DEA model, Internet marketing Variables related to customer Visits to websites, requests, No prices.
ANOVA, Scheffe and Pearson strategies for 60 Greek relations and the virtual reservations, quality of
chi-square, t-tests. hotels. information, communication, customer service and average
distribution and transaction daily rate.
spaces.
Barros (2004) Stochastic Cobb-Douglas cost 42 pousadas in Portugal Labour, capital and food and Sales and nights occupied Labour, capital
frontier model. from 1999 to 2001 a dummy variable and food

R. Fuentes / Tourism Management 32 (2011) 7587


(historical vs. regional
pousadas). Managers
opinion is recommended.
Barros and Alves (2004) Output-oriented DEA and 42 Enatur hotels from 1991 Full-time workers, cost of Sales, number of guests and No prices
Malmquist index (divided into to 2001 labour, book value or property, nights spent in the hotel
4 factors). operating costs and
external costs
Chiang, Tsai, and Wang (2004) Input-oriented DEA model, VRS, 25 four or ve-star hotels in Hotel rooms, food and beverage Yielding index, food and No prices
CRS, overall, pure technical and Taipei (three types). capacity, number of employees beverage revenue and other
scale efciency scores. and total costs. revenues.
Fuchs (2004) Two staged output-oriented 21 Tyrolean destination Carrying capacity, advertising Tourist satisfaction and price Salaries and
DEA, VRS. units (tourism summer costs, energy and recycling costs, level for different hotel prices for hotel
resorts) salaries and typical resource categories multiplied by categories.
aspects such as, overnight stays (sales).
accommodation or
restaurants.
Hu and Cai (2004) Input-oriented DEA, CRS and 242 hotels in California Full-time equivalent managers and Total revenues and number of Average daily rate
regression analysis. (USA). workers, service quality (such as rooms sold. and wage of
average daily rate or hotel managers and staff.
category), physical properties
(number of rooms), employee
skills (wage rates of managers
and staff).
Reynolds (2004) Output-oriented DEA with 38 same-brand restaurants Front-of-house hours worked Lunch and dinner sales, tips for Average wage.
uncontrollable inputs, CRS. in the USA in 2001. during lunch and dinner. lunch and dinner as
Uncontrollable: number of a percentage of lunch and
competitors and seating dinner sales.
capacity.
Sigala (2004) Four stepped input and output- 93 three-star hotels in the Rooms, rooms section total Non-food and beverage total Average rate room
oriented DEA, CRS and ANOVA UK. expenses, front ofce wages, revenues, average rate room,
administration non-payroll roomnights, non-roomnights
expenses, other rooms revenues (other outputs
division payroll, other rooms correlated with DEA results:
division non-payroll, total hotel prot, rooms division
demand variability (other inputs revenues, non-rooms division
correlated with DEA results: revenues.)
length of stay, number of
reservations.)
Sigala et al. (2004) Four stepwise input and 93 three-star hotels in the Number of rooms, total Non food and beverages total No prices.
output-oriented DEA, CRS, UK. payroll and other expenses revenue, roomnights, non-
ANOVA, cluster analysis and with respect to rooms, room revenues and average
t-tests. front ofce payroll, room rate.
administration costs
and demand variability.
Wober and Fesenmaier (2004) Input and output-oriented DEA, 48 state tourism advertising Total state ofce domestic Annual business and leisure No prices.
super-efciency. programmes in the USA. advertising budget, total state trips, domestic & international
ofce international advertising visitors expenditures, lodging
budget, other public budget sales and arts & leisure rms
sources and market size revenues.
(uncontrollable).
Barros (2005a) Output-oriented Malmquist 42 public hotel chains from Full-time employees, cost Sales and number of customers. No prices.
index (divided into four factors) 1999 to 2001. of labour, book value of
and Tobit regression property and operating and
external costs.
Barros (2005b) Output-oriented DEA, CRS, VRS. 43 pousada hotels in 2001. Full-time workers, cost of labour, Sales, number of guests and No prices
area of the hotel, book value of the nights spent.
property, rooms, operational and
external costs.
Barros and Mascarenhas (2005) Technical and allocative 43 Portuguese public chain Employees, physical capital and Sales, number of guests and Labour, physical
output-oriented DEA, VRS, CRS. hotels in 2001. rooms. nights spent. capital and rooms.

R. Fuentes / Tourism Management 32 (2011) 7587


Donthu, Hershberger, and Base DEA model (neither inputs 26 fast food outlets. Advertising and promotion Sales and customer satisfaction. No prices.
Osmonbekov (2005) nor outputs to be more expenses, manager experience
important). and number of employees.
Sigala et al. (2005) Stepwise input and output- 93 three-star hotels in the Rooms, rooms section Non-food and beverage total Average rate room.
oriented DEA, CRS, ANOVA and UK in 1999. total wages, room division total revenues, average rate room,
correlation analysis. non-wages expenses, front ofce roomnights, non-roomnights
wages, administration revenues (other outputs
non-payroll expenses, correlated with DEA results:
other rooms division payroll, other hotel prot, rooms division
rooms division non-payroll, total revenues, non-rooms division
demand variability (other inputs revenues.).
correlated with DEA results:
length of stay, number of
reservations,.).
Sigala and Mylonakis (2005) Three stage input-oriented 93 three-star hotels in the Rooms, f&b capacity, information Room and f&b divisions No prices.
DEA, CRS, ANOVA, Pearson UK. technology systems, total revenues.
correlations and post-Scheffee payroll, total material
analysis. and other expenses,
management arrangement and
design, market segments
served, distributions
channels used, repeat customers.
Barros (2006) Stochastic cost frontier analysis. 15 Portuguese hotels Operational costs, total salary Sales and market share. No prices.
during 19982002. expenditure, employees, earnings,
book value of premises, trend.
Bosetti, Casinelli, and Lanza (2006) Output-based Malmquist 20 Italian tourist regions. Tourism development Total presences of tourists, No prices.
productivity change index and (beds in hotels, camp sites, homogeneity of tourism ows
DEA. registered holiday houses during the year, percentage of
and other structures per 100 protected areas and index of
inhabitants), public efciency in waste treatment.
management, advertising
and environmental
protection expenditures and
market size (uncontrollable).
(continued on next page)

79
80 R. Fuentes / Tourism Management 32 (2011) 7587

inputs refer to aspects such as labour, operational costs, the

room rates and f&b.


investment made and environmental characteristics. With respect
Average wage,
to outputs, in general, the types of variable contemplated largely

Server wage.
refer to total sales or sales by section and the number of clients.
No prices.

No prices.

No prices.
There are a few cases where other types of variable have been used,
Prices

such as prots (6.82%) or a customer satisfaction index (13.64%). In


any event, as we have previously mentioned, the variables that are
nally selected will also depend on their availability and the opinion
of professionals working in the sector.
and beverages department and
Revenues from room and food

number of available beds and

a percentage of staff and tips.


Room and food and beverage

Beds sold divided by total

3. Methodology
Sales and quality index.

Sales per day and as


revenues and room

DEA will be used to analyse the efciency of the travel agencies


other revenues.

examined in this study. DEA is a technique based on obtaining an


occupancy.

efciency frontier using a set of observations. Taking into account


beds sold.
Outputs

the results obtained for all other units, the method focuses on
solving a linear programme that is used to obtain the maximum level
of relative efciency of each decision making unit (DMU) analysed.
The DEA efciency measurement is based on the ratio between
the sum of the outputs and inputs of each unit. Furthermore, the
CRS* stands for Constant Returns to Scale, VRS** for Variable Returns to Scale, DRS*** for Decreasing Returns to Scale and f&b**** for Food and Beverage.
server hours per week and server

specic weights used to calculate this ratio are selected by DEA so


location, years open and kind of
Controllable: Number of shifts,
Available beds, employees and

that the Pareto efciency measure of each DMU is calculated


number of rooms, marketing

Uncontrollable: seats, area,


Total server staff, number

(Charnes, Cooper, Lewin, & Seiford, 1997a).


rooms, area of food and
employees in room and

of seats, location index,


department, number of

beverages department.

Therefore, in terms of the input-oriented evaluation process,


average spending per
Total room expenses,

customer, number of
Number of full-time

(excluding salaries).
food and beverages

operating expenses

a DMU is considered to be efcient when it uses the minimum input


empirically observable from any DMU given its output vector
competitors.

(Charnes, Cooper, & Rhodes,1981). In other words, a failure of a DMU


expenses.

facilities.

to generate maximal output levels with minimal input consumption


Inputs

count.

implies a lower efciency level (Cooper, Seiford, & Zhu, 2004).


Given the specic characteristics of the units that are analysed
in this study (which are explained in the following section), this
result can be obtained by solving the following type of input-
establishments of a Spanish
21 four and ve-star hotels

oriented primal linear programme (Charnes, Cooper, Lewin, &


60 full-service restaurants
49 international hotels in
49-unit Asia-Pacic hotel

Seiford, 1997b) which is an adapted version of the original


proposed by Charnes, Cooper, and Rhodes (1978):
chain (19992000).

in the USA in 2001.


Taiwan in 2001.

fast-food chain.

!
54 restaurant

I
X R
X
DMUs/Sector

Min4o;l;Si;Sr 4o  3 Si  Sr
(Antalya)

i1 r1
S:A
_
X n
lj $Xij Si 4o $Xio ci : 1.I (1)
j1
oriented DEA, window analysis,

Xn
technical, scale, pure technical

uncontrollable variables), CRS,


Input-oriented DEA, CRS, VRS,

regression (overall, allocative,

Input-oriented DEA and Tobit

Three step input and output-


Triangular input and output-

lj $Yrj  Sr Yro cr : 1.R


CRS, DRS***, VRS, regression

j1
Bootstrapping and Tobit

lj ; Si ; Sr  0
Output-oriented DEA

regression analysis.
oriented DEA, VRS.

(controllable and

where:
4o : parameter which measures the efciency of the unit ana-
regression.
efciency).

lysed (the subindex o refers to the assessed DMU, o:1.n).


analysis.
Model

Yrj: r-th output of the j-th DMU, r:1.R 2, j:1.n.


Xij: i-th output of the j-th DMU, i:1.I 3, j:1.n.
lj: weights obtained as a solution to the programme which
express the weight of each DMU in the peer group of the DMUo.
n: total number of units analysed (n 22).
Gimenez-Garca, Martnez-Parra,
Wang, Hung, and Shang (2006)

Si : slack variables for inputs (they express the additional


amount of inputs that should be reduced after decreasing all the
Reynolds and Thompson
Keh, Chu, and Xu (2006)

inputs by 1  4o in order to obtain the optimal level of efciency


for the unit analysed).
and Buffa (2007)
Table 1 (continued)

Sr: slack variables for outputs (they express the amount of


Davutyan (2007)

outputs that should be increased in order to achieve its maximum


Source: Author.

level of efciency a DMU is efcient when 4o 1 and the values


(2007)
Author/s

of Si and Sr are 0 -) (Cooper, Seiford, & Tone, 2007, chaps. 1 & 4).
3: a small positive real number (usually, in empirical calcula-
tions, 106) (Norman & Stoker, 1991).
R. Fuentes / Tourism Management 32 (2011) 7587 81

Unlike parametric methods, which are used to obtain the to variations in the sample values as a result, means that it is useful to
hyperplane most suited to the set of observations, DEA is an apply a method which can analyse the sensitivity of efciency results
alternative method of extracting information from observations to changes in sample values (Simar & Wilson, 1998). Furthermore, as
because it aims to optimise the level of efciency of each unit DEA does not incorporate any type of randomness in the process, it
analysed in order to create an efcient frontier based on the Pareto cannot offer any information about uncertainty of efciency esti-
criteria. Furthermore, in addition to the fact that it is not a para- mates for each unit (Lothgren & Tambour, 1999). The bootstrap is
metric method, it is also not a stochastic method, as it does not a statistical procedure which can be used to eliminate these two
assume that the non-calculated efciency follows some kind of drawbacks of DEA. It was introduced by Efron (1979) and is based on
probabilistic distribution (Charnes et al., 1997b). the idea of simulating the data-generating process (DGP) with the
Amongst other advantages, DEA can be used to analyse activities aim of obtaining a new estimate of each simulated sample. In this
in sectors which use a number of different inputs in their produc- way, the estimates obtained mimic the distribution of the real
tive process and generate several outputs. The method does not population estimator (Simar & Wilson, 1998). In fact, the use of the
require any information about which variables are most important term bootstrap is derived from the phrase to pull oneself up by
in the assessment, and it can even use non-discretionary input ones bootstrap, which aims to reect the fact that the method is
variables (Banker & Morey, 1986; Wober & Fesenmaier, 2004). It is based on an initial sample which, using statistical methods, can be
also suitable for situations where the prices of factors and products used to carry out statistical inference (Efron & Tibshirani, 1998, chap.
are unknown or difcult to calculate. For example, in the case 1). More specically, it is possible to calculate condence intervals
where the DMUs to be analysed belong to an economic sector, such for estimated efciency parameters which can be used to nd out if
as the public sector, in which there are frequently outputs that do the levels of efciency of DMUs obtained using DEA are statistically
not have a market price (Charnes et al., 1997a). signicant (Tortosa-Ausina, Grifell-Tatje, Armero, & Conesa, 2008).
Furthermore, the method provides specic information about In this work, the method described by Simar and Wilson (1998)
each DMU (such as peer groups, slack variables, efciency parameters (smoothed bootstrap) will be used. This method provides better
or input and output weights, for example) which can be used to set estimates than those obtained when resampling directly from the
out guidelines regarding how to improve the efciency of inefcient original data sample, because this procedure (naive bootstrap)
units (Cooper et al., 2007, chaps. 1 & 4). In particular, amongst other provides a poor estimate of the DGP. Furthermore, the method will
things, this information can be used to nd out how many inputs (or also incorporate the reection method described by Silverman
outputs) need to be reduced (or increased), or even which efcient (1986), which avoids estimate problems caused by the fact that,
units can be used as a management model for the inefcient ones. when input-oriented, efciency parameters reach a maximum of
However, the method is not free from drawbacks. One of the one. In order to implement this method, the bandwidth of the
main disadvantages is, for example, the need for homogeneity of kernel density estimator (h) must be selected. In this study, h is
the units analysed. Basically, this means that the DMUs must use calculated using the estimation algorithm proposed by Sheater and
the same types of inputs in order to obtain the same types of Jones (1991), which is available at www.stat.unc.edu/faculty/
outputs, and the circumstances which make up the area of action of marron.html. Both the smoothed bootstrap and the h value calcu-
the units also need to be similar (which does not imply that the lations were carried out using MATLAB 7.6.
units of measurement of each variable have to be identical, but For the DEA approach, the smoothed bootstrap algorithm
merely that the DMUs use the same inputs to produce the same follows the steps below (Simar & Wilson, 1998):
outputs) (Cooper et al., 2007, chaps. 1 & 4). Furthermore, DEA is
a deterministic method. It therefore assumes that if any input or a) Compute the efciency parameters for each DMU by using
_
output allocation is higher or lower than the frontier value, this is DEA: f 4 jg j:1.n
due solely to inefcient behaviour, and does not allow for the b) Generate a random sample of size n (f*lb. f*nb) using kernel
possibility of inefciency for random reasons (Coelli, Rao, ODon- density estimation and the reection method, where f*jb is the
nell, & Battese, 2005). In addition, the reliability of the results also efciency parameter of unit j generated with smoothed
depends on the relationship between the number of variables bootstrap.
included and the units analysed. As such, Cooper et al. (2007, chaps. c) Compute a new data set (X*jb, Yj) where
1 & 4) recommend that the number of organisations analysed _ 
should be at least the max {(I  R), 3  (I R)} (where R and I are *
Xjb 4 j =4*jb Xj for j : 1.n (2)
the number of outputs and inputs, respectively), otherwise the
efciency discrimination among DMUs would be questionable,
where X*jb is the new input vector of unit j and Yj is the original
given the models inadequate number of degrees of freedom. The
output vector of unit j.
results provided by DEA are also inuenced by the type and
number of variables included in the model. This means that if these _ _
d) Calculate 4 *ob , the bootstrap estimate of 4 o for o:1.n by
variables were modied, the efciency parameters would also
solving the following linear programming model for unit o:
change. As such, the incorporation of a process which would tell the
user about the impact that the different inputs and outputs have on
efciency parameters could give that user a better overall under-
standing of how to explain the results (Wober & Fesenmaier, 2004). Min_4* ;lj_4*
ob ob
Taking into account all of the factors mentioned above, it is _
SA_
reasonable to assume that the DEA technique is suitable to be used to
n
X
study the travel agency sector, because of the possibilities it offers in * _*
lj  Xjb 4 ob  Xo
terms of minimising drawbacks, and the advantages it offers over
j1
other methods used to quantify efciency. Furthermore, as outlined
n
X
in the section above, the method has already been used for this
lj  Yj Yo
purpose and for other studies with similar characteristics to this one.
j1
However, the fact that non-parametrical DEA estimators are
based on a nite sample of observations, and that they are sensitive lj  0; j : 1.n (3)
82 R. Fuentes / Tourism Management 32 (2011) 7587

where the inputs (X) and outputs (Y) are expressed in vector address of each agency surveyed, its economic data, and details of
notation and X*jb is the inputs vector obtained in step c. its customers.
The set of variables considered in the survey questions was
e) Repeat steps bd B times in order to obtain a set of estimates selected based on those deemed relevant for the analysis, both
_
4 *ob ; b 1.B for o:1.n. B 2000 to ensure that you have through a literature review of the topic and based on opinions from
an adequate estimate of condence intervals (Simar & Wilson, experts in the sector who were interviewed beforehand. Before
2000). carrying out the survey, personal interviews were held with the
managers of travel agencies who agreed to take part (eleven in
In addition to DEA and smoothed bootstrap, another method total). During these interviews, the managers were asked their
will be used (the Mann Whitney U Test) in order to analyse the opinion about the inputs and outputs that would form part of the
relationship between the efciency of travel agencies and their productive process of their agencies, and about any other variables
location, ownership type and amount of experience in the sector. that could inuence their level of efciency.
This method is based on the idea that the possible link existing In order to adhere to the need for homogeneity of DMUs used in
between two variables may be observed when their values are DEA, efforts were made to ensure that the characteristics of the
organised in increasing order because they can offer information agencies studied were as similar as possible. Specically, the fact
about the relationship between their populations. For example, that the agencies carried out the same type of retail activity, selling
a sample where most of the values of one of the variables are tourism products for Spain and abroad, was taken into account.
greater than most of the values of the other would indicate that The surveys were carried out in person in each of the travel
there would be no evidence of random mixing (Gibbons & Chak- agencies. In order to increase the response rate, they were sent to
raborti, 1992, chap. 7). those who did not want to, or could not, respond initially by email or
The use of this method has been deemed appropriate given that, fax. As the staff in the agencies were all native Spanish speakers, the
like DEA, it is not a parametric method (unlike other tests which questionnaires were all written in this language. In the end, 30
can be used for the same purposes, such as the Students t-test) and agencies answered the survey, out of a total of 127. Of these, 8 had to be
because there is no reason to assume that there is any type of excluded following the initial selection process because the infor-
underlying probabilistic distribution of levels of efciency nor of mation provided was limited and, in some cases, incoherent, despite
the variables whose links are being analysed. In these cases, this the fact that the agencies were contacted via telephone in order to ask
test is an effective method for testing hypotheses (Sheskin, 2000). them to complete and/or clarify the answers given more carefully.
Furthermore, the Mann Whitney U Test is more powerful than As agencies were reluctant to provide information about their
other non-parametric alternatives such as the Sign Test (Conover, economic activities (e.g. prot levels), no more responses were
1999, chap. 5), and the fact that it has been used before in similar received, despite repeated attempts and utmost condentiality. As
studies published prior to this one (Koksal & Aksu, 2007) offers a result, the 22 agencies analysed represent just 17.32% of all
a further guarantee of its suitability. agencies surveyed.
The test involves the calculation of a statistic, usually called U, as Based on the results obtained through the questionnaires,
follows: information was obtained about certain aspects which can be used
to determine and summarise the economic activity of travel
nn 1 agencies during the year 2007.
U R (4)
2 To do this, the variables chosen as activity inputs were: the
number of employees (NE), annual expenditure (AE) and the
where R is the sum of the ranks in sample 1 and n is the sample size
potential service (PS) which the agency would be capable of
for sample 1.
providing (the maximum number of customers that the agency
Based on the result of this calculation, the existence of a rela-
would have been able to help that year based on its service capacity
tionship between both samples is rejected with a level of signi-
during that period, as a proxy variable for the level of investment as
cance a when the value of U is less than its percentile a/2 or when it
proposed by Koksal and Aksu (2007)). The outputs selected were:
is higher than its percentile 1  a/2. It is accepted in all other cases
the number of customers (NC) and the average spend per customer
(Conover, 1999, chap. 5).
(AS). A summary of the most important statistical characteristics of
However, this test is not free from limitations. More specically,
these variables can be found in Table 2.
when a set of data is transformed into a range, part of the infor-
These variables were selected not just because the data were
mation is lost. However, this is not a problem if the initial data are
readily available, but also because similar variables have been used
only signicant as an ordinal comparison with the other values,
in previous studies which have focused on the same type of activity
which is the case with DEA estimates when the relationship with
(Barros & Matias, 2006; Koksal & Aksu, 2007; Wober, 2006, for
the variables mentioned is analysed (Conover, 1999, chap. 5).
example). Furthermore, the choice was also made based on the
Another disadvantage of non-parametric tests is that they are
opinions of the sector representatives mentioned above.
designed to test statistical hypotheses only and not to estimate
Finally, in order for the results to be reliable, the number of
parameters, so the MannWhitney U-test can give a p-value but
organisations analysed (n 22) should be at least max{(I  R),
cannot give an estimate of the relationship between two variables.
3  (I R)}. Given that in this study a total number of 5 variables
However, this point is not a problem when it comes to obtaining
have been examined (3 inputs and 2 outputs) and 22 agencies have
results about the existence, or lack thereof, of a relationship
been considered, the relationship required to guarantee the reli-
between the efciency of the agencies and the variables studied.
ability of the results is easily met because 22 > max{(I  R),
3  (I R)} 15, where I is the number of inputs used and R is the
4. Data analysed number of outputs (Cooper et al., 2007, chaps. 1 & 4).

In order to perform the efciency analysis, a survey was con- 5. Results


ducted in travel agencies located within the Alicante city area
between March and April 2008. The questionnaire was divided into The results obtained using the data available for the agencies
three different sections which aimed to nd out the name and studied were analysed using an input-oriented DEA model
R. Fuentes / Tourism Management 32 (2011) 7587 83

Table 2
Input and output characteristics. year 2007.

Number of Employees Potential service Annual expenditure Number of customers Average spend per
(NE) (PS) (AE) (NC) customer (AS)
Average 2.9545 3258.0909 7312.5455 1837.2273 1088.6364
Standard deviation 1.3644 3584.8441 1298.7608 2161.0604 536.3877
Maximum 7 15 850 10 600 9500 2200
Minimum 1 280 5750 140 200

Source: Author.

programmed in MATLAB 7.6, assuming that those agencies aim to should be decreased by the highest number of units is the number
operate incurring the minimum possible cost in order to achieve of employees and the output that needs to be increased most by
maximum prots, given their usual level of income. The input- agencies is the number of customers.
oriented model was chosen based on the views expressed by These results would be consistent with the existence of
sector representatives who explained that, given how difcult it increased competition for capturing clients and an oversized
was to increase the number of customers due to the rapid increase workforce. Both of these factors, in turn, would be inuenced by the
in competitiveness, they had focused their efforts on reducing afore-mentioned rapid proliferation of travel agencies in the city of
costs. Alicante in recent years (see Introduction) as well as the highly rigid
_
These results are shown in Table 3 where 4 j represents the labour market in Spain. This rigidity inuences decisions relating to
efciency parameter of each DMU and SNE, SAE, SPS, SNC and SAS the redeployment of labour and could have led to an incorrect
are the slack variables for each of the model variables: number of management of human resources by the agencies analysed (Alonso
employees (NE), annual expenditure (AE), potential service (PS), & Galdon, 2007).
number of customers (NC) and average spend per customer (AS), Finally, the peer group column in Table 3 shows the DMUs that
respectively. the non-efcient agencies should focus on in order to nd ways to
As Table 3 shows, only 7 of the 22 agencies studied (31.82%) improve their management processes and increase their levels of
_
work efciently, achieving efciency ratios 4 j equal to one and efciency. Agency number 9 is therefore the one that should be
slack variables equal to zero. The average efciency ratio for all used as a model referred to by other non-efcient agencies for
units was 0.8133, which represents quite a high average level of management tips. Of all the efcient DMUs, number 9 was the one
efciency (81.33%). However, it may also imply the existence of used most as a reference for inefcient units (9 times), followed by
room for improving the efciency levels obtained by applying numbers 3 and 8 (7 times), then 21 (6 times) and 22 (5 times) and,
changes to the management methods used. More specically, the nally, numbers 16 and 19 (3 times).
value obtained for each agency reects the proportion of resources In addition to this, Table 4 shows the input and output values
that should be used by each DMU. For example, agency number 2 which the agencies should aim to achieve in order to be efcient. It
should only use 90.91% of the inputs that it uses. Furthermore, the therefore shows those agencies how they can improve their levels
non-zero slack variables reect the additional amount of each input of efciency. In short, as the results of this table are a direct
(output) which each inefcient DMU should reduce (increase) in consequence of the results obtained in Table 3, the management
order to become efcient. The table shows that the input that strategies that may be obtained would follow the same lines as

Table 3
Input-oriented DEA model: efciency, peer groups and slack variables.
_
DMU 4j Peer group (DMUs) Slack variables

SNE SAE SPS SNC SAS


1 0.5575 8, 22 1.3393 2094.3510 0.0000 0.0000 0.0000
2 0.9091 3, 22 1.0970 698.6220 0.0000 238.9474 0.0000
3 1.0000 3 0.0000 0.0000 0.0000 0.0000 0.0000
4 0.6898 8, 22 1.2361 2086.7310 0.0000 0.0000 0.0000
5 0.2745 3, 21 0.0000 0.0000 2783.2640 0.0000 10.3935
6 0.6965 9, 21 0.0000 0.0000 1032.8960 54.1619 0.0000
7 0.8323 8, 9 1.8548 0.0000 0.0000 112.7208 0.0000
8 1.0000 8 0.0000 0.0000 0.0000 0.0000 0.0000
9 1.0000 9 0.0000 0.0000 0.0000 0.0000 0.0000
10 0.6079 9, 16, 21 0.0000 0.0000 1211.9830 0.0000 0.0000
11 0.6061 3, 8, 9 0.3006 0.0000 0.0000 0.0000 0.0000
12 0.6650 3, 8, 9, 16 0.0000 0.0000 0.0000 0.0000 0.0000
13 0.8415 9, 19, 21 0.0000 0.0000 94.2106 0.0000 0.0000
14 0.7447 3, 9, 19 0.3430 0.0000 0.0000 0.0000 0.0000
15 0.6718 3, 8, 22 0.0108 0.0000 0.0000 0.0000 0.0000
16 1.0000 16 0.0000 0.0000 0.0000 0.0000 0.0000
17 0.9748 9, 16,21 0.0000 0.0000 2322.1320 0.0000 0.0000
18 0.8575 3, 9, 19, 21 0.0000 0.0000 0.0000 0.0000 0.0000
19 1.0000 19 0.0000 0.0000 0.0000 0.0000 0.0000
20 0.9639 8, 22 1.0944 2093.9350 0.0000 0.0000 0.0000
21 1.0000 21 0.0000 0.0000 0.0000 0.0000 0.0000
22 1.0000 22 0.0000 0.0000 0.0000 0.0000 0.0000

SNE, SAE, SPS, SNC and SAS are the slack variables for each of the model variables: number of employees (NE), annual expenditure (AE), potential service (PS), number of
customers (NC) and average spend per customer (AS), respectively.
Source: Author.
84 R. Fuentes / Tourism Management 32 (2011) 7587

Table 4 Table 5
Estimated input and output efciency values. Bootstrap of efciency values.
^ ^ ^ ^ ^ _
DMU NE AE PS NC AS DMU 4j 4
~j Bias Lower (95%) Upper (95%)
1 0.3333 1446.0261 156.1111 200.0000 200.0000 1 0.5575 0.5260 0.0315 0.5010 0.5470
2 3.4484 5074.1052 4680.0000 738.9474 500.0000 2 0.9091 0.8621 0.0470 0.8049 0.8959
3 7.0000 10 300.0000 9500.0000 1500.0000 1500.0000 3 1.0000 0.9085 0.0915 0.8138 0.9968
4 0.8333 2810.9541 206.9444 500.0000 500.0000 4 0.6898 0.6503 0.0395 0.5939 0.6749
5 0.8236 2178.4622 1568.1696 300.0000 310.3935 5 0.2745 0.2430 0.0315 0.2151 0.2661
6 1.3930 4213.6986 708.3018 1054.1619 1000.0000 6 0.6965 0.6526 0.0439 0.6182 0.6842
7 2.3066 6855.5455 457.7577 1512.7208 1400.0000 7 0.8323 0.7911 0.0412 0.7402 0.8212
8 2.0000 6103.0000 350.0000 1200.0000 1200.0000 8 1.0000 0.9138 0.0862 0.7404 0.9923
9 3.0000 6870.0000 1200.0000 2000.0000 2000.0000 9 1.0000 0.9155 0.0845 0.8462 0.9960
10 1.8237 5392.0251 1219.5954 1300.0000 1300.0000 10 0.6079 0.5709 0.0370 0.5418 0.5908
11 2.1238 3824.4658 2121.3360 800.0000 800.0000 11 0.6061 0.5539 0.0522 0.5098 0.5973
12 1.9950 5227.0038 1330.0264 1000.0000 1000.0000 12 0.6650 0.6231 0.0419 0.5902 0.6562
13 1.6831 5230.2239 2430.4350 1200.0000 1200.0000 13 0.8415 0.7976 0.0439 0.7559 0.8309
14 1.8910 4281.8364 1117.0008 1200.0000 1200.0000 14 0.7447 0.7105 0.0342 0.6616 0.7396
15 2.0047 5441.5905 1007.7020 1000.0000 1000.0000 15 0.6718 0.6253 0.0465 0.5886 0.6656
16 1.0000 6355.0000 1000.0000 950.0000 950.0000 16 1.0000 0.8943 0.1057 0.7615 0.9985
17 2.9245 10 333.2383 3526.8708 2200.0000 2200.0000 17 0.9748 0.9297 0.0451 0.8729 0.9629
18 2.5726 202.5651 1715.0740 1600.0000 1600.0000 18 0.8575 0.8289 0.0286 0.7945 0.8495
19 3.0000 8135.0000 6000.0000 2000.0000 2000.0000 19 1.0000 0.9193 0.0807 0.8380 0.9980
20 0.8333 4017.1206 481.9445 500.0000 500.0000 20 0.9639 0.9331 0.0308 0.8941 0.9601
21 1.0000 7143.0000 4000.0000 1000.0000 1000.0000 21 1.0000 0.9242 0.0758 0.8104 0.9894
22 1.0000 6670.0000 1000.0000 600.0000 600.0000 22 1.0000 0.9224 0.0776 0.7858 0.9948

Number of employees (NE), annual expenditure (AE), potential service (PS), number Source: Author.
of customers (NC) and average spend per customer (AS).
Source: Author.
As we can see, the bias-corrected measure 4
~ j may be derived
from the following expression:
those previously described. The fundamental difference resides in
the fact that here the gures show the nal target values. _
4
~ j 4 j  bias (7)
The input values would be corrected using the efciency
parameter for each agency and the slack variable values for each Furthermore, lower and upper refer to the highest and lowest
one as follows (Charnes et al., 1997b): limits respectively of the condence intervals obtained for the
corrected values 4 ~ j , according to the steps of the previously
mentioned smoothed bootstrap algorithm (see Methodology).
b 4 X S 
X ci:1.I cj:1.n (5) The results contained in this table show that the conclusions
ij o ij i
_
reached based on the DEA estimates 4 j should be analysed with
where X b is the estimated value of the level of input i for DMUj, _
ij care given that the bias-corrected values 4 j are, sometimes, quite
4o is the input-oriented efciency parameter, Xij is the real value of _
different from the estimates e:g: 416 1 while 4 ~ 16 0:8943. In
input i for agency j and Si is the slack variable value for that input fact, it may be observed that the efciency values originally
and agency. _
obtained through 4 j are higher than the results of the estimates of
For outputs, given that the model is input-oriented, the proce- the bias-corrected measures 4 ~ j which would lead to all bias being
dure would be slightly different (Charnes et al., 1997b): _
positive and therefore the original DEA estimates 4j would be
upward biased. Evidently, this would indicate that the average
b Y Sr
Y cr:1.R cj:1.n (6)
rj rj efciency levels reached in function with the latter (81.33%) are
higher than those obtained on average from the corrected values
where Y b rj is the estimated value of the level of output r for DMUj, 4~ j (75.89%).
Yrj is the real value of output r for agency j and Sr is the slack In particular, these gures show that the reduction in efciency
variable value for that output and agency. is more pronounced in the units classied as efcient through DEA
So, for example, the gures for agency 2 show the result of rst (3, 8, 9, 16, 19, 21 and 22) and two agencies that DEA did not nd
carrying out, a general reduction of all inputs to 90.91%, as shown in them to be efcient are the ones with the best results following the
Table 3, and then a specic reduction of 1.0970 in the number of bootstrap (17 and 20).
employees and of 698.6220 in its level of average expenditure Furthermore, the condence intervals are also useful for
Xb b
12 0:9091*X12  1:0970 3:4484 and X 22 0:9091*X22  reaching conclusions about the levels of efciency of the DMUs. So,
698:6220 5074:1052. Furthermore, in terms of its output level, when the levels of two different units do not overlap, it is possible
its number of customers would have to increase by 238.9474 to state that their levels of efciency are signicantly different. If
Yb
22 Y22 238:9474 738:9474. the opposite is true, no conclusion of this sort can be reached. As
In order to eliminate the limitations of DEA outlined above, i.e. such, Table 5 shows that, in many cases, there is not enough
the sensitivity of estimators when there are variations in values in empirical evidence to prove that two agencies are not equally
the original sample group and the inability to carry out processes of efcient. This shows that it is important to be cautious when
statistical inference, the study went on to analyse the data using the comparing efciency scores based on DEA estimates.
smoothed bootstrap method described above. The results of this Finally, the Mann Whitney U Test can be used to nd out
analysis are shown in Table 5. whether the levels of efciency of the different agencies may be
The columns in Table 5 show the values of the DEA estimates inuenced by the fact that they belong (or not) to a chain of
_
4 j , the bias-corrected measure 4 ~ j , the bootstrap bias estimate agencies (Koksal & Aksu, 2007), their physical location in the city, or
(Bias) and the 95% condence intervals for the bias-corrected ef- the experience acquired since they began operation (Wober, 2006).
ciency estimates (Lower and Upper), respectively, for each DMU. Not only have these three variables been found, in previous studies,
R. Fuentes / Tourism Management 32 (2011) 7587 85

to have a potential inuence on the levels of efciency of agencies, Table 7


but this was conrmed by the managers interviewed in this study. Comparison of the efciency of travel agencies based on bias-corrected efciency
values in relation to agency ownership type, location and experience.
The data relating to these variables were obtained from a public
list of travel agencies in Alicante, published by the regional Ownership type N Mid-range Sum of ranges
government (Regional Government of Valencia, 2008). They were Part of a Group 10 12.20 122.00
analysed using the statistical package programme SPSS 14.0. The Privately-Owned 12 10.92 131.00
variable relating to the experience acquired by the agency was Total 22
selected as a result of opinions provided by the agencies surveyed, Mann Whitney-U 53.00
which believed that the number of years of experience that Asymptotic signicance (bilateral) 0.644

a company had in the sector, and not the experience of its


Level of experience N Mid-range Sum of ranges
employees, would have the most inuence over that companys
Low 12 12.75 153.00
efciency, given that the employees, after a brief adaptation period,
High 10 10.00 100.00
would learn about the work system that the company itself had
developed over time. Furthermore, the variable relating to the Total 22
Mann Whitney-U 45.00
location of an agency refers to whether or not that agency is located Asymptotic signicance (bilateral) 0.323
in the city centre. As it was not possible to obtain as much infor-
mation as in other works (Wober, 2006), this variable was used as Location N Mid-range Sum of ranges
a proxy variable. Centre 14 13.79 193.00
The possible relationship between the variables mentioned and Outskirts 8 7.50 60.00
the levels of efciency of the DMUs was analysed taking into Total 22
account the results of both the DEA and the smoothed bootstrap. In Mann Whitney-U 24.00
order to analyse the results of the latter, the Mann Whitney U test Asymptotic signicance (bilateral) 0.029
was used taking into account the bias-corrected efciency values Source: Author.
4~ j . Tables 6 and 7 show the results obtained through DEA and 4 ~j,
respectively.
As Table 6 shows, the results of the Mann Whitney U Test deal with an agent in person when planning a holiday through
suggest, with a level of signicance of 0.05, that an agencys a travel agency.
ownership type and level of experience do not affect its level of At the same time, the lack of inuence of the rst two variables
efciency, but that its location does have an inuence. In this would imply, rstly that belonging to a large chain is not a relevant
respect, the efciency of a DMU may be positively inuenced by factor that affects the efciency of agencies and secondly, that after
a change of location to the city centre. The greater afuence of an initial level of experience has been obtained, any further expe-
potential customers and/or the higher purchasing power of resi- rience acquired does not alter efciency values.
dents in these areas may be the variables which ultimately deter- In fact, the same conclusions can be obtained when the test is
mine this result. In fact, both factors could directly inuence the carried out on the bias-corrected efciency values (Table 7). As
capacity of the DMUs to increase all of their output levels in general, a result, it is possible to state that the relationship detected using
and more specically the output level which the DEA indicated as DEA estimates can be maintained given the variations in the sample
that which should be increased by a greater number of agencies carried out using smoothed bootstrap. This shows that, in this
(number of customers). Furthermore, this could highlight the fact particular situation, the conclusions obtained when using boot-
that, despite the development of the Internet and the option to strapping as an approach to statistical inference are not different
make reservations online, the average customer may still prefer to from those obtained through initial DEA estimates, in spite of the
fact that the results of the smoothed bootstrap were quite different
from the values of the DEA estimates.
Table 6
Comparison of the efciency of travel agencies based on DEA results in relation to
agency ownership type, location and experience. 6. Conclusions

Ownership type N Mid-range Sum of ranges


This study has assessed the level of efciency of the economic
Part of a Group 10 11.90 119.00 activity of travel agencies in the city of Alicante in the year 2007. It has
Privately-Owned 12 11.17 134.00
also examined the relationship between certain variables (agency
Total 22 ownership type, location and experience) and the level of efciency
Mann Whitney-U 56.00
of those agencies. Data Envelopment Analysis, smoothed bootstrap
Asymptotic signicance (bilateral) 0.789
and the Mann Whitney U Test were used for these purposes.
Level of experience N Mid-range Sum of ranges In the rst part of the study, relating to DEA efciency levels, it
Low 12 13.25 159.00
was found that 7 of the 22 agencies assessed are efcient, repre-
High 10 9.40 94.00 senting 31.82% of the sample total. Having said that, the average
level of efciency is quite high (81.33%). However, some agencies
Total 22
Mann Whitney-U 39.00 have quite low levels of inefciency (i.e. DMUs 1, 5, 10 and 11).
Asymptotic signicance (bilateral) 0.159 Nevertheless, it is possible to increase these levels by reducing the
amount of resources used and increasing the outputs produced by
Location N Mid-range Sum of ranges inefcient agencies until they reach the levels shown in Table 4. At
Centre 14 13.79 193.00 the same time, a change required in the management strategy of
Outskirts 8 7.50 60.00 each inefcient DMU in order to achieve these targets should be
Total 22 based on the strategies used in units which belong to its peer group,
Mann Whitney-U 24.00 as this would reect the efcient agencies that DEA has used as
Asymptotic signicance (bilateral) 0.026 a reference in order to evaluate their level of efciency. In any case,
Source: Author. and given that it may be difcult for the inefcient agencies to
86 R. Fuentes / Tourism Management 32 (2011) 7587

obtain this information, the target values shown in Table 4 are the Nevertheless, this study has attempted to incorporate new
levels of the inputs and outputs which should have been used to features not included in a number of other very good studies
identify operative responses to management priorities in order to carried out on this sort of establishment (e.g. Barros & Dieke, 2007;
implement strategies to make them efcient. Koksal & Aksu, 2007 and Wober, 2006).
However, smoothed bootstrap was also used to analyse the In particular, with regard to the variables used, the experience
sensitivity of DEA estimates to sampling variation and to obtain variable has been approached from the point of view of the expe-
measures of uncertainty relating to these estimates in order to rience acquired by an agency over the course of the period during
conrm or refute the initial values. which it has been operating in the sector. Furthermore, it has also
The bootstrap results are quite different from the initial DEA been possible to incorporate a variable relating to the average
estimates. They also show that condence intervals signicantly spend per customer, which could be used to gain a better insight
change the conclusions reached earlier about efciency levels, as into the output level of each unit by combining it with the number
frequent overlaps can make it impossible to rank DMUs in the same of customers served by the agency during the period studied.
way as the DEA results obtained at the beginning. In fact, DMU17 and In terms of the analysis method used, the target values for each
DMU20 have the highest scores based on their high bias-corrected variable were calculated using efciency variables and slack vari-
value, despite the fact that the DEA results did not lead to them being ables, as opposed to using specic weights of each efcient DMU in
_
classied as efcient (even though it gave them high 4 j values). the peer groups of each inefcient agency. This makes it possible to
In the second part of the study, which analysed the relationship see where additional efforts need to be made in some resources
between three different variables and the efciency of the travel even when the level of efciency of an agency has been corrected.
agencies, the results show that the location of an agency is the only In addition to this, the smoothed bootstrap technique was also used
one of these factors that affects their levels of efciency. This shows to study the stability of the results after sampling variation and to
that setting up an agency in areas close to the city centre has carry out statistical inference.
a major inuence on the efciency of that agency. As such, a higher Finally, the analysis could be continued through the calculation
concentration of the agencies around the city centre would imply of allocative and economic efciency levels or even, if dynamic data
an increase in their level of efciency in spite of the fact that it could could be obtained, through the calculation of changes in the levels
result in more elevated initial costs for them. of productivity. Clearly, this information could add to the results
In addition to this, no link was found between the levels of ef- because, for example, Malmquist indices or quasi-Malmquist
ciency of agencies and their ownership type, nor with their level of indices could be used in order to nd out how the productivity
experience. These results suggest that neither belonging to a group levels of agencies develop over time, and the way in which inputs
nor having a lot of experience affect the efciency of an agency. and outputs inuence those levels. It would also be useful to
In the rst case, the result shows that the advantages of include output-oriented analyses which could show the maximum
belonging to a group are not statistically signicant. This means production levels that DMUs could generate (with the amount of
that agencies which form part of large chains may have access to resources they use) if they demonstrate efcient behaviour.
management conditions which do not, in the end, have any inu-
ence on their efciency. From this point of view, it would seem References
reasonable for units which form part of groups to look at the
advantages that they should, in theory, enjoy because they belong Alonso, C., & Galdon, E. (2007). La proteccion al empleo en Espana: evolucion y con-
to a larger organisation, to nd out whether they are really mate- secuencias. Informacion Comercial Espanola, Revista de Economa, 837, 157177.
Anderson, R. I., Fish, M., Xia, Y., & Michello, F. (1999). Measuring efciency in the hotel
rialising. In any case, including information about the prots that industry: a stochastic frontier approach. Hospitality Management, 18, 4557.
each agency makes may lead to a change in this conclusion. Anderson, R. I., Fok, R., & Scott, J. (2000). Hotel industry efciency: an advanced
Unfortunately, it was not possible to obtain this type of data. linear programming examination. American Business Review, 18(1), 4048.
Anderson, R. I., Lewis, D., & Parker, M. E. (February, 1999). Another look at the
In the second case (level of experience), the result obtained is
efciency of corporate travel management departments. Journal of Travel
similar to the rst. This means that having more years of experience Research, 37, 267272.
in the sector does not have a statistically signicant link with the Baker, M., & Riley, M. (1994). News perspectives on productivity in hotels: some
advances and new directions. International Journal of Hospitality Management,
level of efciency of agencies. This seems to suggest that, once an
13(4), 297311.
agency has overcome the logical rst stage of learning and become Banker, R. D., & Morey, R. C. (1986). Efciency analysis for exogenously xed inputs
established, increasing the amount of time that it has been oper- and outputs. Operations Research, 34(4), 513521.
ating in the sector does not necessarily mean that it becomes more Barros, C. P. (2004). A stochastic cost frontier in the Portuguese hotel industry.
Tourism Economics, 10(2), 177192.
efcient. Barros, C. P. (2005a). Evaluating the efciency of a small hotel chain with Malmquist
With regard to the limitations of this study, the main drawback productivity index. International Journal of Tourism Research, 7, 173184.
lies in the fact that not all of the information requested was Barros, C. P. (2005b). Measuring efciency in the hotel sector. Annals of Tourism
Research, 32(2), 465477.
obtained. It was impossible to increase the number of units analysed Barros, C. P. (2006). Analysing the rate of technical change in the Portuguese hotel
as higher response rates were not obtained. As a result, access to an industry. Tourism Economics, 12(3), 325346.
ofcial or public database which included the economic information Barros, C. P., & Alves, F. P. (2004). Productivity in the tourism industry. International
Advances in Economic Research, 10(3), 215225.
required to carry out a study of this type may have made the task Barros, C. P., & Dieke, P. U. C. (2007). Analyzing the total productivity change in
easier and improved results. In any case, the representativeness of travel agencies. Tourism Analysis, 12, 2737.
the sample falls within the limits established in other similar studies Barros, C. P., & Mascarenhas, M. J. (2005). Technical and allocative efciency in
a chain of small hotels. Hospitality Management, 24, 415436.
(such as Hu & Cai, 2004 or Koksal & Aksu, 2007). In addition, another Barros, C. P., & Matias, A. (2006). Assessing the efciency of travel agencies with
limitation of this study lies in the fact that there are a number of a stochastic cost frontier: a Portuguese case study. International Journal of
non-controllable factors which affect the efciency of DMUs and Tourism Research, 8, 367379.
Bell, R. A., & Morey, R. C. (1994). The search for appropriate benchmarking partners:
which have not been taken into account here (such as the number of
a macro approach and application to corporate travel management. Omega,
inhabitants in the area surrounding the agency, or their level of International Journal of Management Science, 22(5), 477490.
income). Access to information relating to these variables would Bell, R. A., & Morey, R. C. (1995). Increasing the efciency of corporate travel manage-
make it possible to analyse their overall inuence using DEA models ment through macro benchmarking. Journal of Travel Research, 33(3), 1120.
Bosetti, V., Cassinelli, M., & Lanza, A. (2006). Benchmarking in tourism destination,
which include non-controllable variables, or other complementary keeping in mind the sustainable paradigm. Nota di Lavoro 12.2006, Fondazione
statistical methods (e.g. Tobit regression). Eni Enrico Mattei.
R. Fuentes / Tourism Management 32 (2011) 7587 87

Brown, R. J., & Ragsdale, C. T. (2002). The competitive market efciency of hotel McLaughlin, C. P., & Coffey, S. (1990). Measuring productivity on services. Interna-
brands: an application of data envelopment analysis. Journal of Hospitality & tional Journal of Service Industry Management, 1(1), 4664.
Tourism Research, 26(4), 332360. Morey, R. C., & Dittman, D. A. (October, 1995). Evaluating a hotel GMs performance.
Charnes, A., Cooper, W. W., Lewin, A. Y., & Seiford, L. M. (1997a). Introduction. In A case study in benchmarking. Cornell Hotel and Restaurant Administration
A. Charnes, W. W. Cooper, A. Y. Lewin, & L. M. Seiford (Eds.), Data envelopment Quarterly 3035.
analysis: Theory, methodology and applications (2nd ed.). (pp. 322) Kluwer Morey, R. C., & Dittman, D. A. (1997). An aid in selecting the brand, size and other
Academic Publishers. strategic choices for a hotel. The Council on Hotel, Restaurant and Institutional
Charnes, A., Cooper, W. W., Lewin, A. Y., & Seiford, L. M. (1997b). Basic DEA models. Education, 21(1), 7199.
In A. Charnes, W. W. Cooper, A. Y. Lewin, & L. M. Seiford (Eds.), Data envelopment Morey, R. C., & Dittman, D. A. (2003). Evaluating a hotel GMs performance. Updated and
analysis: Theory, methodology and applications (2nd ed.). (pp. 2348) Kluwer extension. Cornell Hotel and Restaurant Administration Quarterly, 44(5/6), 5368.
Academic Publishers. National Statistics Institute. (2009). Annual services survey. 2007, Available at:
Charnes, A., Cooper, W. W., & Rhodes, E. (1978). Measuring the efciency of decision http://www.ine.es/jaxi/menu.do?typepcaxis&path%2Ft37%2Fe01%2Fprovi%
making units. European Journal of Operational Research, 2, 429444. 2F&lepcaxis&L0 (Accessed 14.04.09).
Charnes, A., Cooper, W. W., & Rhodes, E. (1981). Evaluating program and managerial Norman, M., & Stoker, B. (1991). DEA. The assessment of performance. John Wiley and
efciency: an application of data envelopment analysis to program follow Sons. (Appendix A).
through. Management Science, 27(6), 668697. Regional Government of Valencia. (2008), Public list of travel agencies in Alicante,
Chiang, W., Tsai, M., & Wang, L. S. (2004). A DEA evaluation of Taipei hotels. Annals Available at: http://xml.turisme.gva.es:7777/establecimientos/aavv_alacant.
of Tourism Research, 31(3), 712715. xsql (Accessed 14.04.09).
Coelli, T. J., Rao, D. S. P., ODonnell, C. J., & Battese, G. E. (2005). An introduction to Reynolds, D. (2004). An exploratory investigation of multiunit restaurant produc-
efciency and productivity analysis (2nd ed.). New York: NY Springer. tivity assessment using data envelopment analysis. Journal of Travel & Tourism
Conover, W. J. (1999). Practical nonparametric statistics (3rd ed.). New York: John Marketing, 16(2/3), 1926.
Wiley & Sons. Reynolds, D., & Thompson, G. M. (2007). Multiunit restaurant productivity assessment
Cooper, W. W., Seiford, L. M., & Tone, K. (2007). Data envelopment analysis. A using three-phase data envelopment analysis. Hospitality Management, 26, 2032.
comprehensive text with models, applications, references and dea-solver software Sheater, S. J., & Jones, M. C. (1991). A reliable data-based bandwidth selection
(2nd ed.). Springer. method for kernel density estimation. Journal of the Royal Statistical Society,
Cooper, W. W., Seiford, L. M., & Zhu, J. (2004). Data envelopment analysis: history, Series B, 53(3), 683690.
models and interpretations. In W. W. Cooper, L. M. Sieford, & J. Zhu (Eds.), Hand- Sheskin, D. (2000). Handbook of parametric and nonparametric statistical procedures
book on data envelopment analysis (pp. 140). Kluwer Academic Publishers Group. (2nd ed.). Chapman & Hall/CRC. (Test 12).
County Council of Alicante. (2009). Turistic data of Alicante, Available at: http://www. Sigala, M. (2003). Developing and benchmarking internet marketing strategies in the
dip-alicante.es/documentacion/8turismo.asp?codigo03014, (Accessed 14.04.09). hotel sector in Greece. Journal of Hospitality & Tourism Research, 27(4), 375401.
Davutyan, N. (2007). Measuring the quality of hospitality at Antalya. International Sigala, M. (2004). Using data envelopment analysis for measuring and bench-
Journal of Tourism Research, 9, 5157. marking productivity in the hotel sector. Journal of Travel & Tourism Marketing,
Donthu, N., & Yoo, B. (1998). Retail productivity assessment using data envelopment 16(2/3), 3960.
analysis. Journal of Retailing, 74(1), 89105. Sigala, M., Airey, D., Jones, P., & Lockwood, A. (November, 2004). ICT paradox lost? A
Donthu, N., Hershberger, E. K., & Osmonbekov, T. (2005). Benchmarking marketing stepwise DEA methodology to evaluate technology investments in tourism
productivity using data envelopment analysis. Journal of Business Research, 58, settings. Journal of Travel Research, 43, 180192.
14741482. Sigala, M., Jones, P., Lockwood, A., & Airey, D. (2005). Productivity in hotels:
Efron, B. (1979). Bootstrap methods: another look at the jacknife. The Annals of a stepwise data envelopment analysis of hotels rooms division processes. The
Statistics, 7(1), 126. Service Industrial Journal, 25(1), 6181.
Efron, B., & Tibshirani, R. J. (1998). An introduction to the bootstrap (2nd ed.). Florida: Sigala, M., & Mylonakis, J. (2005). Developing a data envelopment analysis model
CRS Press LLC. for measuring and isolating the impact of contextual factors on hotel produc-
Fuchs, M. (2004). Strategy development in tourism destinations: a DEA approach. tivity. International Journal of Business Performance Management, 7(2), 174190.
The Poznan University of Economics Review, 4(1), 5273. Silverman, B. W. (1986). Density estimation for statistics and data analysis. London:
Gibbons, J. D., & Chakraborti, S. (1992). Nonparametric statistical inference (3rd ed.). Chapman and Hall.
New York: Marcel Dekker, INC. Simar, L., & Wilson, P. W. (1998). Sensitivity analysis of efciency scoresh: how to
Gimenez-Garca, V., Martnez-Parra, J. L., & Buffa, F. P. (2007). Improving resource bootstrap in nonparametric frontier models. Management Science, 44(1), 4961.
utilization in multi-unit networked organizations: the case of a Spanish Simar, L., & Wilson, P. W. (2000). Statistical inference in nonparmetric frontier
restaurant chain. Tourism Management, 28, 262270. models: the state of the art. Journal of Productivity Analysis, 13(1), 4978.
Hruschka, H. (1986). Ansatze der Efzienmessung von Betrieben. Journal fur Tarim, S., Dener, H. I., & Tarim, S. A. (2000). Efciency measurement in the hotel
Betriebswirtschaft, 36(2), 7685. industry: output factor constrained DEA application. Anatolia: An International
Hu, B. A., & Cai, L. A. (2004). Hotel labor productivity assessment: a data envelop- Journal of Tourism and Hospitality Research, 11(2), 111123.
ment analysis. Journal of Travel & Tourism Marketing, 16(2/3), 2738. Tortosa-Ausina, E., Grifell-Tatje, E., Armero, C., & Conesa, D. (2008). Sensitivity anal-
Hwang, S., & Chang, T. (2003). Using data envelopment analysis to measure hotel ysis of efciency and Malmquist productivity indices: an application to spanish
managerial efciency change in Taiwan. Tourism Management, 24, 357369. savings Banks. European Journal of Operational Research, 184, 10621084.
Johns, N., Howcroft, B., & Drake, L. (1997). The use of data envelopment analysis to Wang, F., Hung, W., & Shang, J. (2006). Measuring the cost efciency of international
monitor hotel productivity. Progress in Tourism and Hospitality Research, 3,119127. tourist hotels in Taiwan. Tourism Economics, 12(1), 6585.
Keh, H. T., Chu, S., & Xu, J. (2006). Efciency, effectiveness and productivity of Wober, K. W. (2000). Benchmarking hotel operations on the Internet: a data envel-
marketing in services. European Journal of Operational Research, 170, 265276. opment analysis approach. Information Technology and Tourism, 3(3/4), 195212.
Koksal, C. D., & Aksu, A. A. (2007). Efciency evaluation of A-group travel agencies Wober, K. W. (2006). Data envelopment analysis. Journal of Travel & Tourism
with data envelopment analysis (DEA): a case study in the Antalya region, Marketing, 21(4), 91108.
Turkey. Tourism Management, 28, 830834. Wober, K. W., & Fesenmaier, D. R. (2004). A multi-criteria approach to destination
Lothgren, M., & Tambour, M. (1999). Bootstrapping the data envelopment analysis benchmarking: a case study of state tourism advertising programs in the United
mamquist productivity index. Applied Economics, 31, 417425. States. Journal of Travel & Tourism Marketing, 16(2/3), 118.

Vous aimerez peut-être aussi