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

Tourism Management 32 (2011) 75–87

Contents lists available at ScienceDirect

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

Efficiency of travel agencies: A case study of Alicante, Spain Ramo´n 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: Received 9 December 2008 Accepted 2 December 2009

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

Keywords: Travel agencies Efficiency Data envelopment analysis DEA Smoothed bootstrap

1. Introduction The growing amount of competition in the economy over the last few years has stimulated interest in the analysis and assessment of efficiency in all economic sectors. The services sector is no exception, despite the fact that its unique characteristics (such as intangibility or heterogeneity of its outputs) mean that it is difficult to assess and quantify its efficiency (McLaughlin & Coffey, 1990). Travel agencies work in the services sector, and more specifically in the tourism sector. Given the importance and global scope of today’s tourism sector, it seems important to carry out an analysis of the efficiency of agencies whose main, albeit not only, aim is to help connect supply to demand. It would also be useful for any economic agents with direct or indirect links to these agencies to be able to access information about their level of efficiency so that they can make informed decisions about investment and/or management. With this in mind, this project aims to analyse the efficiency of a group of travel agencies based in Alicante (Spain). Alicante was selected for the analysis for two main reasons: firstly, because it is an area of Spain where the tourism sector is of great importance, in terms of both supply and demand; and secondly, because the number of travel agencies in the area has increased at such an exceptionally high rate over the last few years that the number of agencies per capita has now more than doubled the national average. Between 2000 and 2007, the number of travel agencies in Alicante increased by 205.26%, while the total increase for the whole of Spain was just 67.41%. Furthermore, in 2007, the

* Tel./fax: þ34 965 90 97 11. E-mail address: [email protected] 0261-5177/$ – see front matter Ó 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.tourman.2009.12.003

proportion of agencies per inhabitant was 0.05% in Alicante, compared with 0.022% for the whole of Spain (County Council of Alicante, 2009; National Statistics Institute, 2009). A setting with such a high level of competitiveness was therefore a very suitable choice for the study of efficiency in this sort of agency. Then, based on the results of this analysis, it will assess the way in which certain characteristics of the agencies may affect their efficiency parameters. In particular, the study aims to ascertain whether or not the agency’s ownership type, location or the length of time it has been in operation are factors which affect its level of efficiency. By examining these variables, the study will try to ascertain whether or not the fact that an agency forms part of a chain, is located in the city centre or has many years of experience in the sector are relevant factors when it comes to improving its level of efficiency. 2. Literature review To date, only seven articles analysing the efficiency of travel agencies have been published. The first two of these, written by Bell and Morey (1994, 1995), did not analyse the efficiency of travel agencies exactly, but that of 31 corporate travel departments. Inputoriented Data Envelopment Analysis (DEA) was used for these studies, using the level of service provided as a representative variable for output, whilst input variables used were the levels of travel expenditure (such as car and airline costs and hotel bills), labour costs, other general expenditure (space and technology costs, for example) and, finally, other environmental factors (to illustrate a company’s ability to obtain travel cost discounts). Some years later, Anderson, Lewis, and Parker (1999) used the same data that was used in the afore-mentioned study to show the

76

R. Fuentes / Tourism Management 32 (2011) 75–87

differences between input-oriented DEA and stochastic frontier results. However, the constructed variables used were different. The output was the number of company trips made and the inputs used were labour costs, the sum of air, hotel and car expenses and, finally, other expenses (such as technology or occupancy costs). The authors concluded that corporate travel departments are a good investment for companies as they are highly efficient and they make it possible to control the increase in travel costs. Subsequently, another article was written by Barros and Matias (2006). It examines the efficiency of 25 Portuguese travel agencies using stochastic cost frontier analysis. This study includes an exhaustive review of previously-published articles which have applied frontier analysis to assess levels of efficiency in the tourism sector (essentially hotels and travel agencies), showing that DEA is the most widely-used method. The variables used by Barros and Matias were selected based on their availability and the fact that they had been used in previous works. Specifically, they used: operational costs at constant prices, price of labour, price of capital-stock (proxied by the ratio of earnings to stock), price of capital-premises (proxied by the ratio of expenses in premises divided by the value of real assets), dummy variables for Spanish companies operating in the Portuguese market (based on the idea that due to their recent entry into this market, they may have been still undergoing a convergence and consolidation process), additional dummy variables for companies which had carried out mergers and acquisitions and, finally, sales at constant prices (the output). These authors concluded that the main factors determining efficiency in the sector were capital, labour, sales, and mergers and acquisitions activities. In the same year, Wo¨ber (2006) analysed efficiency data gathered in 2003 relating to 80 branch offices of a tour operator in Austria, using a DEA model which considered variable returns to scale, which was input-oriented in the first instance, and outputoriented later. He also calculated levels of super-efficiency in order to draw up a ranking of the efficient branch offices. The controllable inputs used were: personnel, occupancy, marketing and other variable and fixed costs, the number of employees (weighted by the number of working days per year) and their average job experience. The non-controllable inputs used were: number of residents living near each agency, a visibility and competitiveness index (based on the size of the window display and the number of agencies nearby in the close neighbourhood) and ease of access via car or public transport. The outputs used were: total number of contracts, total turnover and contribution margin for each of the outlets. As a result of the analysis, the author was able to suggest ways in which the management scores for the agencies could be improved, both in terms of the use of inputs and the production of outputs. He also provided a ranking of all the offices studied. All of this information could be used to provide practical solutions for the different management targets. Only one year later, Ko¨ksal and Aksu (2007) used input-oriented DEA to assess the efficiency of 24 travel agencies in the city of Antalya (Turkey). They also employed the Mann Whitney U test to analyse the relationship between the ownership type of an agency and its level of efficiency, and concluded that there is no link between the two variables. The authors also applied DEA to calculate changes in the level of inputs that inefficient agencies would have to achieve in order to become efficient. The variables were obtained through surveys and comprised: the number of staff, the level of annual expenses, the potential level of service that they can provide (inputs) and the number of customers served (outputs). Finally, Barros and Dieke (2007) analysed changes in the productivity of travel agencies based on the study of a significant sample group of agencies operating in the Portuguese market between 2000 and 2004. They applied a quantitative method based

on the calculation of the Malmquist index, breaking it down into four factors. They also used a bootstrapped Tobit model. In order to calculate productivity, they used sales and profits as outputs, and wages, capital, total costs excluding wages and book value of premises as inputs. In the second stage of the analysis, the efficient Malmquist scores were analysed using a Tobit model where the explicative variables were: foreign ownership of the company, the ratio of operational costs to sales, the market share of the agencies and whether or not the agency belonged to a commercial chain which could give it access to economies of scale. The authors concluded that the level of capital, market share, control of factor costs and belonging to a group were the main factors which determined efficiency in the Portuguese sector. To date, no further analyses of efficiency in the travel agency sector have been carried out. However, the fact that these studies are often included in hospitality research literature means that it is reasonable to include other studies that have analysed efficiency in this context in this review (other previously-published works with excellent reviews of this topic include, for example, Barros & Dieke, 2007; Sigala, Jones, Lockwood, & Airey, 2005 or Wo¨ber, 2006). Table 1 provides a list of works published to date that analyse efficiency in the tourism and hospitality sectors. The review of the works included in the above table reveals relevant information with respect to the different methods that have been used in the past to analyse the efficiency of firms operating in the tourism and hospitality sectors and the variables that have been selected as inputs and outputs of these methods. Therefore, the conclusions that may be obtained based on this analysis will be useful in determining the type of statistical procedure to follow and the most appropriate variables for the analysis. With regard to the latter, as well as the information obtained from the bibliographical review, the opinion of company managers from the sector under analysis will also be taken into account as will the availability of the information required (Barros & Matias, 2006). On the one hand, with respect to the statistical procedure used, the main conclusions that can be drawn from the published literature both on tourism in general and on travel agencies in particular are, firstly, that DEA is the method that has been used most often to carry out efficiency analyses, although other methods, such as stochastic frontier analysis, have been used to a lesser extent. In fact, it may be observed that DEA was used in 86.37% of studies analyzing the efficiency of DMUs and in 38.64% of these cases it was combined with other methods in order to obtain additional information regarding their behaviour (multiple regression, analysis of variance, the Mann Whitney U Test ..). Furthermore, on the whole, the kind of efficiency that has been most analysed is technical efficiency, in spite of the fact that other forms (such as overall, allocative or scale efficiency) have also been considered on occasion. Finally, the type of DEA model applied has gradually incorporated new characteristics and/or complementary methods which have helped authors to gather more information and results. Studies such as those conducted by Barros and Mascarenhas (2005), Reynolds and Thompson (2007), Sigala (2003, 2004), Sigala, Airey, Jones, and Lockwood (2004) or Wo¨ber and Fesenmaier (2004) used stepwise DEA models and/or combined DEA with regression analysis, ANOVA and other statistical techniques in order to enrich their results. On the other hand, with respect to the variables, it is not possible to arrive at conclusions which are as highly defined as those for the case of the statistical procedure applied. This is due to the wide diversity of the type of DMUs analysed in the bibliographical review which means that the inputs and outputs employed are also diverse. However, it is possible to make general conclusions when the variables are grouped into broad categories. So, if we first focus on the inputs, it may be affirmed that the type of variables most used as

Table 1 Analysis of efficiency in the tourism and hospitality sectors. Author/s

Model

DMUs/Sector

Inputs

Outputs

Prices

Banker and Morey (1986)

60 Fast food restaurants.

Labour, supplies and advertising expenditures, age of the store, location and existence of drive-in window.

Breakfast, lunch and dinner sales.

No prices.

Hruschka (1986)

Input and output-oriented DEA models with discretionary and non-discretionary inputs, technical and scale efficiency, CRS*, VRS**. DEA, CRS.

10 hospitality companies.

Total revenues.

No prices.

Baker and Riley (1994)

Multiple regression analysis.

20 hotels based in each of Germany, France and UK.

Number of seats, labour expenses, operative resources, other operational resources. Room rate, full-time employees, number or rooms.

Room rate.

Morey and Dittman (1995)

Allocative input-oriented DEA.

54 owner-managed hotels in the USA during 1993.

Sales per full-time employee (fte), f&b****(fte), room occupancy, value added (fte). Total room revenue and average level of customer satisfaction on two dimensions: physical facilities and services provided.

Johns, Howcroft, and Drake (1997)

Input-oriented DEA.

15 hotels (same chain).

Rooms sold, total covers served, total beverage revenue.

No prices.

Morey and Dittman (1997)

DEA, VRS, profit maximizationoriented and regression analysis.

Hotels.

Expected annualised profits (total room sales þ f&b sales þ miscellaneous sales  annual operating costs  amortised building costs).

No prices.

Donthu and Yoo (1998)

DEA, window analysis and regression analysis.

24 restaurants in the USA (chain)

Sales and customer satisfaction.

No prices.

Anderson, Fish, Xia, and Michello (1999)

Stochastic cost frontier analysis.

48 hotel companies in the USA.

Rooms, gaming, food and beverage revenues and other revenues.

F&b, gaming & rooms. Proxies for wages.

Anderson, Fok and Scott (2000)

Input-oriented DEA (overall, allocative, technical, pure technical and scale efficiency), CRS, VRS. Input & output-oriented DEA, CRS (modified restrictions).

48 hotels in the USA in 1994.

Total revenues in 1994.

Proxies for wages and room price.

Customer loyalty index, occupancy rate and net profit.

No prices.

Input-oriented DEA, CRS.

61 hotels in Austria in 1997.

Total accommodation revenue, food and beverage revenue and average bed occupancy.

No prices.

Wo¨ber (2000)

21 4/5-star hotels (Antalya)

Average daily rate of competitive sets.

77

(continued on next page)

R. Fuentes / Tourism Management 32 (2011) 75–87

Tarim, Dener and Tarim (2000)

Properties’ physical characteristics (number of rooms), environmental factors (unionised employees, average occupancy and daily rate for competitive sets) and controllable factors (room-division wages, profits and meals .). Rooms, nights available, total labour hours, total f&b costs, total utilities costs. Exogenous factors: near airport, unionised employees, cost of labour, competition, occupancy rates, competitor’s average rates, cost of acquiring the site. Endogenous: brand, number and mix of rooms, f&b capacities, other facilities, levels of service, marketing expenditures and supplies, administration and repairs expenditures. Store size, store manager experience, store location and promotion expenses. Full-time equivalent employees, number of rooms, total gaming expenses, total f&b expenses and other expenses. Full-time equivalent employees, number of rooms, total gaming expenses, total f&b expenses and other expenses. Investment costs, number of employees, administrative expenses. Controllable: Total payroll and related costs, material-types expenses and energy, cleaning, maintenance, communication, marketing and administration costs. Uncontrollable: number of beds, seats and opening days.

78

Table 1 (continued) Model

DMUs/Sector

Inputs

Outputs

Prices

Output-oriented DEA, CRS and cluster analysis.

46 hotel brands in the USA.

Guest’s satisfaction and overall value of the hotel chain.

Average room price.

Hwang and Chang (2003)

Output-oriented DEA, CRS. Malmquist index, superefficiency and ANOVA.

45 hotels from 1994 to 1998 in Taiwan.

Room revenues, food and beverages revenue and other revenues.

No prices.

Morey and Dittman (2003)

Input-oriented DEA, allocative efficiency.

Sigala (2003)

Output-oriented DEA model, ANOVA, Scheffe´ and Pearson chi-square, t-tests.

54 owner-managed hotels of a chain in the USA in 1993. Internet marketing strategies for 60 Greek hotels.

Revenues, profits, servicesatisfaction index, physicalfacilities-satisfaction index. Visits to websites, requests, reservations, quality of customer service and average daily rate.

Salaries and competitors’ average daily rate. No prices.

Barros (2004)

Stochastic Cobb-Douglas cost frontier model.

42 pousadas in Portugal from 1999 to 2001

Sales and nights occupied

Labour, capital and food

Barros and Alves (2004)

Output-oriented DEA and Malmquist index (divided into 4 factors).

42 Enatur hotels from 1991 to 2001

Sales, number of guests and nights spent in the hotel

No prices

Chiang, Tsai, and Wang (2004)

Input-oriented DEA model, VRS, CRS, overall, pure technical and scale efficiency scores. Two staged output-oriented DEA, VRS.

25 four or five-star hotels in Taipei (three types).

Yielding index, food and beverage revenue and other revenues. Tourist satisfaction and price level for different hotel categories multiplied by overnight stays (sales).

No prices

Hu and Cai (2004)

Input-oriented DEA, CRS and regression analysis.

242 hotels in California (USA).

Total revenues and number of rooms sold.

Average daily rate and wage of managers and staff.

Reynolds (2004)

Output-oriented DEA with uncontrollable inputs, CRS.

38 same-brand restaurants in the USA in 2001.

Lunch and dinner sales, tips for lunch and dinner as a percentage of lunch and dinner sales.

Average wage.

Sigala (2004)

Four stepped input and outputoriented DEA, CRS and ANOVA

93 three-star hotels in the UK.

Complaints, service quality, condition and cleanliness of rooms, hotel properties and guest rooms in the USA. Number of full-time employees, number of guest rooms, total area of meal department and operating expenses. Resources consumed, occupancy rate and, unionised labour, number or rooms. Variables related to customer relations and the virtual information, communication, distribution and transaction spaces. Labour, capital and food and a dummy variable (historical vs. regional pousadas). Manager’s opinion is recommended. Full-time workers, cost of labour, book value or property, operating costs and external costs Hotel rooms, food and beverage capacity, number of employees and total costs. Carrying capacity, advertising costs, energy and recycling costs, salaries and typical resource aspects such as, accommodation or restaurants. Full-time equivalent managers and workers, service quality (such as average daily rate or hotel category), physical properties (number of rooms), employee skills (wage rates of managers and staff). Front-of-house hours worked during lunch and dinner. Uncontrollable: number of competitors and seating capacity. Rooms, rooms section total expenses, front office wages, administration non-payroll expenses, other rooms division payroll, other rooms division non-payroll, total demand variability (other inputs correlated with DEA results: length of stay, number of reservations.)

Non-food and beverage total revenues, average rate room, roomnights, non-roomnights revenues (other outputs correlated with DEA results: hotel profit, rooms division revenues, non-rooms division revenues.)

Average rate room

Fuchs (2004)

21 Tyrolean destination units (tourism summer resorts)

Salaries and prices for hotel categories.

R. Fuentes / Tourism Management 32 (2011) 75–87

Author/s Brown and Ragsdale (2002)

Four stepwise input and output-oriented DEA, CRS, ANOVA, cluster analysis and t-tests.

93 three-star hotels in the UK.

Wo¨ber and Fesenmaier (2004)

Input and output-oriented DEA, super-efficiency.

48 state tourism advertising programmes in the USA.

Barros (2005a)

Output-oriented Malmquist index (divided into four factors) and Tobit regression

42 public hotel chains from 1999 to 2001.

Barros (2005b)

Output-oriented DEA, CRS, VRS.

43 pousada hotels in 2001.

Barros and Mascarenhas (2005)

Technical and allocative output-oriented DEA, VRS, CRS. Base DEA model (neither inputs nor outputs to be more important). Stepwise input and outputoriented DEA, CRS, ANOVA and correlation analysis.

43 Portuguese public chain hotels in 2001. 26 fast food outlets.

Sigala and Mylonakis (2005)

Three stage input-oriented DEA, CRS, ANOVA, Pearson correlations and post-Scheffee analysis.

93 three-star hotels in the UK.

Barros (2006)

Stochastic cost frontier analysis.

15 Portuguese hotels during 1998–2002.

Bosetti, Casinelli, and Lanza (2006)

Output-based Malmquist productivity change index and DEA.

20 Italian tourist regions.

Donthu, Hershberger, and Osmonbekov (2005) Sigala et al. (2005)

93 three-star hotels in the UK in 1999.

Number of rooms, total payroll and other expenses with respect to rooms, front office payroll, administration costs and demand variability. Total state office domestic advertising budget, total state office international advertising budget, other public budget sources and market size (uncontrollable). Full-time employees, cost of labour, book value of property and operating and external costs. Full-time workers, cost of labour, area of the hotel, book value of the property, rooms, operational and external costs. Employees, physical capital and rooms. Advertising and promotion expenses, manager experience and number of employees. Rooms, rooms section total wages, room division total non-wages expenses, front office wages, administration non-payroll expenses, other rooms division payroll, other rooms division non-payroll, total demand variability (other inputs correlated with DEA results: length of stay, number of reservations,.). Rooms, f&b capacity, information technology systems, total payroll, total material and other expenses, management arrangement and design, market segments served, distributions channels used, repeat customers. Operational costs, total salary expenditure, employees, earnings, book value of premises, trend. Tourism development (beds in hotels, camp sites, registered holiday houses and other structures per 100 inhabitants), public management, advertising and environmental protection expenditures and market size (uncontrollable).

Non food and beverages total revenue, roomnights, nonroom revenues and average room rate.

No prices.

Annual business and leisure trips, domestic & international visitors’ expenditures, lodging sales and arts & leisure firms’ revenues.

No prices.

Sales and number of customers.

No prices.

Sales, number of guests and nights spent.

No prices

Sales, number of guests and nights spent. Sales and customer satisfaction.

Labour, physical capital and rooms. No prices.

Non-food and beverage total revenues, average rate room, roomnights, non-roomnights revenues (other outputs correlated with DEA results: hotel profit, rooms division revenues, non-rooms division revenues.).

Average rate room.

Room and f&b divisions revenues.

No prices.

Sales and market share.

No prices.

Total presences of tourists, homogeneity of tourism flows during the year, percentage of protected areas and index of efficiency in waste treatment.

No prices.

R. Fuentes / Tourism Management 32 (2011) 75–87

Sigala et al. (2004)

(continued on next page)

79

Output-oriented DEA (controllable and uncontrollable variables), CRS, regression analysis. Reynolds and Thompson (2007)

60 full-service restaurants in the USA in 2001.

Three step input and outputoriented DEA, VRS. Gime´nez-Garcı´a, Martı´nez-Parra, and Buffa (2007)

54 restaurant establishments of a Spanish fast-food chain.

Input-oriented DEA and Tobit regression. Davutyan (2007)

21 four and five-star hotels (Antalya)

Number of full-time employees in room and food and beverages department, number of rooms, area of food and beverages department. Available beds, employees and operating expenses (excluding salaries). Total server staff, number of seats, location index, average spending per customer, number of competitors. Controllable: Number of shifts, server hours per week and server count. Uncontrollable: seats, area, location, years open and kind of facilities. 49 international hotels in Taiwan in 2001.

3. Methodology

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. Source: Author.

No prices.

Beds sold divided by total number of available beds and beds sold. Sales and quality index.

Server wage.

Average wage, room rates and f&b. Revenues from room and food and beverages department and other revenues.

Total room expenses, number of rooms, marketing expenses.

Wang, Hung, and Shang (2006)

inputs refer to aspects such as labour, operational costs, the investment made and environmental characteristics. With respect to outputs, in general, the types of variable contemplated largely refer to total sales or sales by section and the number of clients. There are a few cases where other types of variable have been used, such as profits (6.82%) or a customer satisfaction index (13.64%). In any event, as we have previously mentioned, the variables that are finally selected will also depend on their availability and the opinion of professionals working in the sector.

Sales per day and as a percentage of staff and tips.

No prices. Room and food and beverage revenues and room occupancy.

Inputs

Triangular input and outputoriented DEA, window analysis, CRS, DRS***, VRS, regression analysis. Input-oriented DEA, CRS, VRS, Bootstrapping and Tobit regression (overall, allocative, technical, scale, pure technical efficiency).

49-unit Asia-Pacific hotel chain (1999–2000).

Model

Keh, Chu, and Xu (2006)

DMUs/Sector

No prices.

Prices Outputs

R. Fuentes / Tourism Management 32 (2011) 75–87

Author/s

Table 1 (continued)

80

DEA will be used to analyse the efficiency of the travel agencies examined in this study. DEA is a technique based on obtaining an efficiency frontier using a set of observations. Taking into account 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 efficiency of each decision making unit (DMU) analysed. The DEA efficiency measurement is based on the ratio between the sum of the outputs and inputs of each unit. Furthermore, the specific weights used to calculate this ratio are selected by DEA so that the Pareto efficiency measure of each DMU is calculated (Charnes, Cooper, Lewin, & Seiford, 1997a). Therefore, in terms of the input-oriented evaluation process, a DMU is considered to be efficient when it uses the minimum input empirically observable from any DMU given its output vector (Charnes, Cooper, & Rhodes,1981). In other words, a failure of a DMU to generate maximal output levels with minimal input consumption implies a lower efficiency level (Cooper, Seiford, & Zhu, 2004). Given the specific 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 inputoriented primal linear programme (Charnes, Cooper, Lewin, & Seiford, 1997b) which is an adapted version of the original proposed by Charnes, Cooper, and Rhodes (1978):

Min4o;l;Si;Srþ 4o  3 S:A _ n X j¼1 n X

I X i¼1

Si  þ

R X

! Srþ

r¼1

lj $Xij þ Si ¼ 4o $Xio ci : 1.I

(1)

lj $Yrj  Srþ ¼ Yro cr : 1.R

j¼1

lj ; Si ; Srþ  0 where: 4o : parameter which measures the efficiency of the unit analysed (the subindex o refers to the assessed DMU, o:1.n). Yrj: r-th output of the j-th DMU, r:1.R ¼ 2, j:1.n. Xij: i-th output of the j-th DMU, i:1.I ¼ 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 ¼ 22). Si : slack variables for inputs (they express the additional amount of inputs that should be reduced after decreasing all the inputs by ð1  4o Þ in order to obtain the optimal level of efficiency for the unit analysed). Srþ: slack variables for outputs (they express the amount of outputs that should be increased in order to achieve its maximum level of efficiency – a DMU is efficient when 4o ¼ 1 and the values of Si and Srþ are 0 -) (Cooper, Seiford, & Tone, 2007, chaps. 1 & 4). 3: a small positive real number (usually, in empirical calculations, 106) (Norman & Stoker, 1991).

R. Fuentes / Tourism Management 32 (2011) 75–87

Unlike parametric methods, which are used to obtain the hyperplane most suited to the set of observations, DEA is an alternative method of extracting information from observations because it aims to optimise the level of efficiency of each unit analysed in order to create an efficient frontier based on the Pareto criteria. Furthermore, in addition to the fact that it is not a parametric method, it is also not a stochastic method, as it does not assume that the non-calculated efficiency follows some kind of probabilistic distribution (Charnes et al., 1997b). Amongst other advantages, DEA can be used to analyse activities in sectors which use a number of different inputs in their productive process and generate several outputs. The method does not require any information about which variables are most important in the assessment, and it can even use non-discretionary input variables (Banker & Morey, 1986; Wo¨ber & Fesenmaier, 2004). It is also suitable for situations where the prices of factors and products are unknown or difficult to calculate. For example, in the case where the DMUs to be analysed belong to an economic sector, such as the public sector, in which there are frequently outputs that do not have a market price (Charnes et al., 1997a). Furthermore, the method provides specific information about each DMU (such as peer groups, slack variables, efficiency parameters or input and output weights, for example) which can be used to set out guidelines regarding how to improve the efficiency of inefficient units (Cooper et al., 2007, chaps. 1 & 4). In particular, amongst other things, this information can be used to find out how many inputs (or outputs) need to be reduced (or increased), or even which efficient units can be used as a management model for the inefficient ones. However, the method is not free from drawbacks. One of the main disadvantages is, for example, the need for homogeneity of the units analysed. Basically, this means that the DMUs must use the same types of inputs in order to obtain the same types of outputs, and the circumstances which make up the area of action of the units also need to be similar (which does not imply that the units of measurement of each variable have to be identical, but merely that the DMUs use the same inputs to produce the same outputs) (Cooper et al., 2007, chaps. 1 & 4). Furthermore, DEA is a deterministic method. It therefore assumes that if any input or output allocation is higher or lower than the frontier value, this is due solely to inefficient behaviour, and does not allow for the possibility of inefficiency for random reasons (Coelli, Rao, O’Donnell, & Battese, 2005). In addition, the reliability of the results also depends on the relationship between the number of variables included and the units analysed. As such, Cooper et al. (2007, chaps. 1 & 4) recommend that the number of organisations analysed should be at least the max {(I  R), 3  (I þ R)} (where R and I are the number of outputs and inputs, respectively), otherwise the efficiency discrimination among DMUs would be questionable, given the model’s inadequate number of degrees of freedom. The results provided by DEA are also influenced by the type and number of variables included in the model. This means that if these variables were modified, the efficiency parameters would also change. As such, the incorporation of a process which would tell the user about the impact that the different inputs and outputs have on efficiency parameters could give that user a better overall understanding of how to explain the results (Wo¨ber & Fesenmaier, 2004). Taking into account all of the factors mentioned above, it is reasonable to assume that the DEA technique is suitable to be used to study the travel agency sector, because of the possibilities it offers in terms of minimising drawbacks, and the advantages it offers over other methods used to quantify efficiency. Furthermore, as outlined in the section above, the method has already been used for this purpose and for other studies with similar characteristics to this one. However, the fact that non-parametrical DEA estimators are based on a finite sample of observations, and that they are sensitive

81

to variations in the sample values as a result, means that it is useful to apply a method which can analyse the sensitivity of efficiency results to changes in sample values (Simar & Wilson, 1998). Furthermore, as DEA does not incorporate any type of randomness in the process, it cannot offer any information about uncertainty of efficiency estimates for each unit (Lo¨thgren & Tambour, 1999). The bootstrap is a statistical procedure which can be used to eliminate these two drawbacks of DEA. It was introduced by Efron (1979) and is based on the idea of simulating the data-generating process (DGP) with the aim of obtaining a new estimate of each simulated sample. In this way, the estimates obtained mimic the distribution of the real population estimator (Simar & Wilson, 1998). In fact, the use of the term ‘‘bootstrap’’ is derived from the phrase ‘‘to pull oneself up by one’s bootstrap’’, which aims to reflect the fact that the method is based on an initial sample which, using statistical methods, can be used to carry out statistical inference (Efron & Tibshirani, 1998, chap. 1). More specifically, it is possible to calculate confidence intervals for estimated efficiency parameters which can be used to find out if the levels of efficiency of DMUs obtained using DEA are statistically significant (Tortosa-Ausina, Grifell-Tatje´, Armero, & Conesa, 2008). In this work, the method described by Simar and Wilson (1998) (smoothed bootstrap) will be used. This method provides better estimates than those obtained when resampling directly from the original data sample, because this procedure (naive bootstrap) provides a poor estimate of the DGP. Furthermore, the method will also incorporate the reflection method described by Silverman (1986), which avoids estimate problems caused by the fact that, when input-oriented, efficiency parameters reach a maximum of one. In order to implement this method, the bandwidth of the kernel density estimator (h) must be selected. In this study, h is calculated using the estimation algorithm proposed by Sheater and Jones (1991), which is available at www.stat.unc.edu/faculty/ marron.html. Both the smoothed bootstrap and the h value calculations were carried out using MATLAB 7.6. For the DEA approach, the smoothed bootstrap algorithm follows the steps below (Simar & Wilson, 1998): a) Compute the efficiency parameters for each DMU by using _ DEA: f 4 jg j:1.n b) Generate a random sample of size n (f*lb. f*nb) using kernel density estimation and the reflection method, where f*jb is the efficiency parameter of unit j generated with smoothed bootstrap. c) Compute a new data set (X*jb, Yj) where * Xjb ¼



_

4 j =4*jb Xj for j : 1.n

(2)

where X*jb is the new input vector of unit j and Yj is the original output vector of unit j. _

_

d) Calculate 4 *ob , the bootstrap estimate of 4 o for o:1.n by solving the following linear programming model for unit o:

Min_4* ;lj_4* ob ob _ SA_ n X

_* ob

* lj  Xjb ¼ 4

 Xo

j¼1 n X

lj  Yj ¼ Yo

j¼1

lj  0; j : 1.n

(3)

82

R. Fuentes / Tourism Management 32 (2011) 75–87

where the inputs (X) and outputs (Y) are expressed in vector notation and X*jb is the inputs vector obtained in step c. e) Repeat steps b–d B times in order to obtain a set of estimates _ ð 4 *ob ; b ¼ 1.BÞ for o:1.n. B ¼ 2000 to ensure that you have an adequate estimate of confidence intervals (Simar & Wilson, 2000). In addition to DEA and smoothed bootstrap, another method will be used (the Mann Whitney U Test) in order to analyse the relationship between the efficiency of travel agencies and their location, ownership type and amount of experience in the sector. This method is based on the idea that the possible link existing between two variables may be observed when their values are organised in increasing order because they can offer information about the relationship between their populations. For example, a sample where most of the values of one of the variables are greater than most of the values of the other would indicate that there would be no evidence of random mixing (Gibbons & Chakraborti, 1992, chap. 7). The use of this method has been deemed appropriate given that, like DEA, it is not a parametric method (unlike other tests which can be used for the same purposes, such as the Student’s t-test) and because there is no reason to assume that there is any type of underlying probabilistic distribution of levels of efficiency nor of the variables whose links are being analysed. In these cases, this test is an effective method for testing hypotheses (Sheskin, 2000). Furthermore, the Mann Whitney U Test is more powerful than other non-parametric alternatives such as the Sign Test (Conover, 1999, chap. 5), and the fact that it has been used before in similar studies published prior to this one (Ko¨ksal & Aksu, 2007) offers a further guarantee of its suitability. The test involves the calculation of a statistic, usually called U, as follows:

U ¼ R

nðn þ 1Þ 2

(4)

where R is the sum of the ranks in sample 1 and n is the sample size for sample 1. Based on the result of this calculation, the existence of a relationship between both samples is rejected with a level of significance a when the value of U is less than its percentile a/2 or when it is higher than its percentile 1  a/2. It is accepted in all other cases (Conover, 1999, chap. 5). However, this test is not free from limitations. More specifically, when a set of data is transformed into a range, part of the information is lost. However, this is not a problem if the initial data are only significant as an ordinal comparison with the other values, which is the case with DEA estimates when the relationship with the variables mentioned is analysed (Conover, 1999, chap. 5). Another disadvantage of non-parametric tests is that they are designed to test statistical hypotheses only and not to estimate parameters, so the Mann–Whitney U-test can give a p-value but cannot give an estimate of the relationship between two variables. However, this point is not a problem when it comes to obtaining results about the existence, or lack thereof, of a relationship between the efficiency of the agencies and the variables studied. 4. Data analysed In order to perform the efficiency analysis, a survey was conducted in travel agencies located within the Alicante city area between March and April 2008. The questionnaire was divided into three different sections which aimed to find out the name and

address of each agency surveyed, its economic data, and details of its customers. The set of variables considered in the survey questions was selected based on those deemed relevant for the analysis, both through a literature review of the topic and based on opinions from experts in the sector who were interviewed beforehand. Before carrying out the survey, personal interviews were held with the managers of travel agencies who agreed to take part (eleven in total). During these interviews, the managers were asked their opinion about the inputs and outputs that would form part of the productive process of their agencies, and about any other variables that could influence their level of efficiency. In order to adhere to the need for homogeneity of DMUs used in DEA, efforts were made to ensure that the characteristics of the agencies studied were as similar as possible. Specifically, the fact that the agencies carried out the same type of retail activity, selling tourism products for Spain and abroad, was taken into account. The surveys were carried out in person in each of the travel agencies. In order to increase the response rate, they were sent to those who did not want to, or could not, respond initially by email or fax. As the staff in the agencies were all native Spanish speakers, the questionnaires were all written in this language. In the end, 30 agencies answered the survey, out of a total of 127. Of these, 8 had to be excluded following the initial selection process because the information provided was limited and, in some cases, incoherent, despite the fact that the agencies were contacted via telephone in order to ask them to complete and/or clarify the answers given more carefully. As agencies were reluctant to provide information about their economic activities (e.g. profit levels), no more responses were received, despite repeated attempts and utmost confidentiality. As a result, the 22 agencies analysed represent just 17.32% of all agencies surveyed. Based on the results obtained through the questionnaires, information was obtained about certain aspects which can be used to determine and summarise the economic activity of travel agencies during the year 2007. To do this, the variables chosen as activity inputs were: the number of employees (NE), annual expenditure (AE) and the potential service (PS) which the agency would be capable of providing (the maximum number of customers that the agency would have been able to help that year based on its service capacity during that period, as a proxy variable for the level of investment as proposed by Ko¨ksal and Aksu (2007)). The outputs selected were: the number of customers (NC) and the average spend per customer (AS). A summary of the most important statistical characteristics of these variables can be found in Table 2. These variables were selected not just because the data were readily available, but also because similar variables have been used in previous studies which have focused on the same type of activity (Barros & Matias, 2006; Ko¨ksal & Aksu, 2007; Wo¨ber, 2006, for example). Furthermore, the choice was also made based on the opinions of the sector representatives mentioned above. Finally, in order for the results to be reliable, the number of organisations analysed (n ¼ 22) should be at least max{(I  R), 3  (I þ R)}. Given that in this study a total number of 5 variables have been examined (3 inputs and 2 outputs) and 22 agencies have been considered, the relationship required to guarantee the reliability of the results is easily met because 22 > max{(I  R), 3  (I þ R)} ¼ 15, where I is the number of inputs used and R is the number of outputs (Cooper et al., 2007, chaps. 1 & 4). 5. Results The results obtained using the data available for the agencies studied were analysed using an input-oriented DEA model

R. Fuentes / Tourism Management 32 (2011) 75–87

83

Table 2 Input and output characteristics. year 2007.

Average Standard deviation Maximum Minimum

Number of Employees (NE)

Potential service (PS)

Annual expenditure (AE)

Number of customers (NC)

Average spend per customer (AS)

2.9545 1.3644 7 1

3258.0909 3584.8441 15 850 280

7312.5455 1298.7608 10 600 5750

1837.2273 2161.0604 9500 140

1088.6364 536.3877 2200 200

Source: Author.

programmed in MATLAB 7.6, assuming that those agencies aim to operate incurring the minimum possible cost in order to achieve maximum profits, given their usual level of income. The inputoriented model was chosen based on the views expressed by sector representatives who explained that, given how difficult it was to increase the number of customers due to the rapid increase in competitiveness, they had focused their efforts on reducing costs. _ These results are shown in Table 3 where 4 j represents the efficiency parameter of each DMU and 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. As Table 3 shows, only 7 of the 22 agencies studied (31.82%) _ work efficiently, achieving efficiency ratios ð 4 j Þ equal to one and slack variables equal to zero. The average efficiency ratio for all units was 0.8133, which represents quite a high average level of efficiency (81.33%). However, it may also imply the existence of room for improving the efficiency levels obtained by applying changes to the management methods used. More specifically, the value obtained for each agency reflects the proportion of resources that should be used by each DMU. For example, agency number 2 should only use 90.91% of the inputs that it uses. Furthermore, the non-zero slack variables reflect the additional amount of each input (output) which each inefficient DMU should reduce (increase) in order to become efficient. The table shows that the input that

should be decreased by the highest number of units is the number of employees and the output that needs to be increased most by agencies is the number of customers. These results would be consistent with the existence of increased competition for capturing clients and an oversized workforce. Both of these factors, in turn, would be influenced by the afore-mentioned rapid proliferation of travel agencies in the city of Alicante in recent years (see Introduction) as well as the highly rigid labour market in Spain. This rigidity influences decisions relating to the redeployment of labour and could have led to an incorrect management of human resources by the agencies analysed (Alonso & Galdo´n, 2007). Finally, the ‘‘peer group’’ column in Table 3 shows the DMUs that the non-efficient agencies should focus on in order to find ways to improve their management processes and increase their levels of efficiency. Agency number 9 is therefore the one that should be used as a model referred to by other non-efficient agencies for management tips. Of all the efficient DMUs, number 9 was the one used most as a reference for inefficient units (9 times), followed by numbers 3 and 8 (7 times), then 21 (6 times) and 22 (5 times) and, finally, numbers 16 and 19 (3 times). In addition to this, Table 4 shows the input and output values which the agencies should aim to achieve in order to be efficient. It therefore shows those agencies how they can improve their levels of efficiency. In short, as the results of this table are a direct consequence of the results obtained in Table 3, the management strategies that may be obtained would follow the same lines as

Table 3 Input-oriented DEA model: efficiency, peer groups and slack variables. DMU

_ 4j

Peer group (DMUs)

Slack variables SNE

SAE

SPS

SNC

SAS

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22

0.5575 0.9091 1.0000 0.6898 0.2745 0.6965 0.8323 1.0000 1.0000 0.6079 0.6061 0.6650 0.8415 0.7447 0.6718 1.0000 0.9748 0.8575 1.0000 0.9639 1.0000 1.0000

8, 22 3, 22 3 8, 22 3, 21 9, 21 8, 9 8 9 9, 16, 21 3, 8, 9 3, 8, 9, 16 9, 19, 21 3, 9, 19 3, 8, 22 16 9, 16,21 3, 9, 19, 21 19 8, 22 21 22

1.3393 1.0970 0.0000 1.2361 0.0000 0.0000 1.8548 0.0000 0.0000 0.0000 0.3006 0.0000 0.0000 0.3430 0.0108 0.0000 0.0000 0.0000 0.0000 1.0944 0.0000 0.0000

2094.3510 698.6220 0.0000 2086.7310 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 2093.9350 0.0000 0.0000

0.0000 0.0000 0.0000 0.0000 2783.2640 1032.8960 0.0000 0.0000 0.0000 1211.9830 0.0000 0.0000 94.2106 0.0000 0.0000 0.0000 2322.1320 0.0000 0.0000 0.0000 0.0000 0.0000

0.0000 238.9474 0.0000 0.0000 0.0000 54.1619 112.7208 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000

0.0000 0.0000 0.0000 0.0000 10.3935 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 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) 75–87

Table 4 Estimated input and output efficiency values. ^

^

^

Table 5 Bootstrap of efficiency values. ^

^

DMU

NE

AE

PS

NC

AS

DMU

_ 4j

~j 4

Bias

Lower (95%)

Upper (95%)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22

0.3333 3.4484 7.0000 0.8333 0.8236 1.3930 2.3066 2.0000 3.0000 1.8237 2.1238 1.9950 1.6831 1.8910 2.0047 1.0000 2.9245 2.5726 3.0000 0.8333 1.0000 1.0000

1446.0261 5074.1052 10 300.0000 2810.9541 2178.4622 4213.6986 6855.5455 6103.0000 6870.0000 5392.0251 3824.4658 5227.0038 5230.2239 4281.8364 5441.5905 6355.0000 10 333.2383 202.5651 8135.0000 4017.1206 7143.0000 6670.0000

156.1111 4680.0000 9500.0000 206.9444 1568.1696 708.3018 457.7577 350.0000 1200.0000 1219.5954 2121.3360 1330.0264 2430.4350 1117.0008 1007.7020 1000.0000 3526.8708 1715.0740 6000.0000 481.9445 4000.0000 1000.0000

200.0000 738.9474 1500.0000 500.0000 300.0000 1054.1619 1512.7208 1200.0000 2000.0000 1300.0000 800.0000 1000.0000 1200.0000 1200.0000 1000.0000 950.0000 2200.0000 1600.0000 2000.0000 500.0000 1000.0000 600.0000

200.0000 500.0000 1500.0000 500.0000 310.3935 1000.0000 1400.0000 1200.0000 2000.0000 1300.0000 800.0000 1000.0000 1200.0000 1200.0000 1000.0000 950.0000 2200.0000 1600.0000 2000.0000 500.0000 1000.0000 600.0000

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22

0.5575 0.9091 1.0000 0.6898 0.2745 0.6965 0.8323 1.0000 1.0000 0.6079 0.6061 0.6650 0.8415 0.7447 0.6718 1.0000 0.9748 0.8575 1.0000 0.9639 1.0000 1.0000

0.5260 0.8621 0.9085 0.6503 0.2430 0.6526 0.7911 0.9138 0.9155 0.5709 0.5539 0.6231 0.7976 0.7105 0.6253 0.8943 0.9297 0.8289 0.9193 0.9331 0.9242 0.9224

0.0315 0.0470 0.0915 0.0395 0.0315 0.0439 0.0412 0.0862 0.0845 0.0370 0.0522 0.0419 0.0439 0.0342 0.0465 0.1057 0.0451 0.0286 0.0807 0.0308 0.0758 0.0776

0.5010 0.8049 0.8138 0.5939 0.2151 0.6182 0.7402 0.7404 0.8462 0.5418 0.5098 0.5902 0.7559 0.6616 0.5886 0.7615 0.8729 0.7945 0.8380 0.8941 0.8104 0.7858

0.5470 0.8959 0.9968 0.6749 0.2661 0.6842 0.8212 0.9923 0.9960 0.5908 0.5973 0.6562 0.8309 0.7396 0.6656 0.9985 0.9629 0.8495 0.9980 0.9601 0.9894 0.9948

Number of employees (NE), annual expenditure (AE), potential service (PS), number of customers (NC) and average spend per customer (AS). Source: Author.

those previously described. The fundamental difference resides in the fact that here the figures show the final target values. The input values would be corrected using the efficiency parameter for each agency and the slack variable values for each one as follows (Charnes et al., 1997b):

b ¼ 4 X S  X ij ij i o

ci:1.I cj:1.n

(5)

b is the estimated value of the level of input i for DMUj, where X ij 4o is the input-oriented efficiency parameter, Xij is the real value of input i for agency j and Si is the slack variable value for that input and agency. For outputs, given that the model is input-oriented, the procedure would be slightly different (Charnes et al., 1997b):

b ¼ Y þ Sr þ Y rj rj

cr:1.R cj:1.n

(6)

b rj is the estimated value of the level of output r for DMUj, where Y Yrj is the real value of output r for agency j and Srþ is the slack variable value for that output and agency. So, for example, the figures for agency 2 show the result of first carrying out, a general reduction of all inputs to 90.91%, as shown in Table 3, and then a specific reduction of 1.0970 in the number of employees and of 698.6220 in its level of average expenditure b b ðX 12 ¼ 0:9091*X12  1:0970 ¼ 3:4484 and X 22 ¼ 0:9091*X22  698:6220 ¼ 5074:1052Þ. Furthermore, in terms of its output level, its number of customers would have to increase by 238.9474 b ðY 22 ¼ Y22 þ 238:9474 ¼ 738:9474Þ. In order to eliminate the limitations of DEA outlined above, i.e. the sensitivity of estimators when there are variations in values in the original sample group and the inability to carry out processes of statistical inference, the study went on to analyse the data using the smoothed bootstrap method described above. The results of this analysis are shown in Table 5. The columns in Table 5 show the values of the DEA estimates _ ~ j Þ, the bootstrap bias estimate ð 4 j Þ, the bias-corrected measure ð4 (Bias) and the 95% confidence intervals for the bias-corrected efficiency estimates (Lower and Upper), respectively, for each DMU.

Source: Author.

~ j Þ may be derived As we can see, the bias-corrected measure ð4 from the following expression: _

~ j ¼ 4 j  bias 4

(7)

Furthermore, lower and upper refer to the highest and lowest limits respectively of the confidence intervals obtained for the ~ j Þ, according to the steps of the previously corrected values ð4 mentioned smoothed bootstrap algorithm (see Methodology). The results contained in this table show that the conclusions _ reached based on the DEA estimates ð 4 j Þ should be analysed with _ care given that the bias-corrected values ð 4 j Þ are, sometimes, quite _ ~ 16 ¼ 0:8943Þ. In different from the estimates ðe:g: 416 ¼ 1 while 4 fact, it may be observed that the efficiency values originally _ obtained through ð 4 j Þ are higher than the results of the estimates of ~ j Þ which would lead to all bias being the bias-corrected measures ð4 _ positive and therefore the original DEA estimates ð 4j Þ would be upward biased. Evidently, this would indicate that the average efficiency levels reached in function with the latter (81.33%) are higher than those obtained on average from the corrected values ~ j Þ (75.89%). ð4 In particular, these figures show that the reduction in efficiency is more pronounced in the units classified as efficient through DEA (3, 8, 9, 16, 19, 21 and 22) and two agencies that DEA did not find them to be efficient are the ones with the best results following the bootstrap (17 and 20). Furthermore, the confidence intervals are also useful for reaching conclusions about the levels of efficiency of the DMUs. So, when the levels of two different units do not overlap, it is possible to state that their levels of efficiency are significantly different. If the opposite is true, no conclusion of this sort can be reached. As such, Table 5 shows that, in many cases, there is not enough empirical evidence to prove that two agencies are not equally efficient. This shows that it is important to be cautious when comparing efficiency scores based on DEA estimates. Finally, the Mann Whitney U Test can be used to find out whether the levels of efficiency of the different agencies may be influenced by the fact that they belong (or not) to a chain of agencies (Ko¨ksal & Aksu, 2007), their physical location in the city, or the experience acquired since they began operation (Wo¨ber, 2006). Not only have these three variables been found, in previous studies,

R. Fuentes / Tourism Management 32 (2011) 75–87

to have a potential influence on the levels of efficiency of agencies, but this was confirmed by the managers interviewed in this study. The data relating to these variables were obtained from a public list of travel agencies in Alicante, published by the regional government (Regional Government of Valencia, 2008). They were analysed using the statistical package programme SPSS 14.0. The variable relating to the experience acquired by the agency was selected as a result of opinions provided by the agencies surveyed, which believed that the number of years of experience that a company had in the sector, and not the experience of its employees, would have the most influence over that company’s efficiency, given that the employees, after a brief adaptation period, would learn about the work system that the company itself had developed over time. Furthermore, the variable relating to the location of an agency refers to whether or not that agency is located in the city centre. As it was not possible to obtain as much information as in other works (Wo¨ber, 2006), this variable was used as a proxy variable. The possible relationship between the variables mentioned and the levels of efficiency of the DMUs was analysed taking into account the results of both the DEA and the smoothed bootstrap. In order to analyse the results of the latter, the Mann Whitney U test was used taking into account the bias-corrected efficiency values ~j, ~ j Þ. Tables 6 and 7 show the results obtained through DEA and 4 ð4 respectively. As Table 6 shows, the results of the Mann Whitney U Test suggest, with a level of significance of 0.05, that an agency’s ownership type and level of experience do not affect its level of efficiency, but that its location does have an influence. In this respect, the efficiency of a DMU may be positively influenced by a change of location to the city centre. The greater affluence of potential customers and/or the higher purchasing power of residents in these areas may be the variables which ultimately determine this result. In fact, both factors could directly influence the capacity of the DMUs to increase all of their output levels in general, and more specifically the output level which the DEA indicated as that which should be increased by a greater number of agencies (number of customers). Furthermore, this could highlight the fact that, despite the development of the Internet and the option to make reservations online, the average customer may still prefer to

Table 6 Comparison of the efficiency of travel agencies based on DEA results in relation to agency ownership type, location and experience. Ownership type

N

Mid-range

Sum of ranges

Part of a Group Privately-Owned

10 12

11.90 11.17

119.00 134.00

Total 22 Mann Whitney-U Asymptotic significance (bilateral)

56.00 0.789

Level of experience

N

Mid-range

Sum of ranges

Low High

12 10

13.25 9.40

159.00 94.00

Total 22 Mann Whitney-U Asymptotic significance (bilateral)

39.00 0.159

Location

N

Mid-range

Sum of ranges

Centre Outskirts

14 8

13.79 7.50

193.00 60.00

Total 22 Mann Whitney-U Asymptotic significance (bilateral) Source: Author.

24.00 0.026

85

Table 7 Comparison of the efficiency of travel agencies based on bias-corrected efficiency values in relation to agency ownership type, location and experience. Ownership type

N

Mid-range

Sum of ranges

Part of a Group Privately-Owned

10 12

12.20 10.92

122.00 131.00

Total 22 Mann Whitney-U Asymptotic significance (bilateral)

53.00 0.644

Level of experience

N

Mid-range

Sum of ranges

Low High

12 10

12.75 10.00

153.00 100.00

Total 22 Mann Whitney-U Asymptotic significance (bilateral)

45.00 0.323

Location

N

Mid-range

Sum of ranges

Centre Outskirts

14 8

13.79 7.50

193.00 60.00

Total 22 Mann Whitney-U Asymptotic significance (bilateral)

24.00 0.029

Source: Author.

deal with an agent in person when planning a holiday through a travel agency. At the same time, the lack of influence of the first two variables would imply, firstly that belonging to a large chain is not a relevant factor that affects the efficiency of agencies and secondly, that after an initial level of experience has been obtained, any further experience acquired does not alter efficiency values. In fact, the same conclusions can be obtained when the test is carried out on the bias-corrected efficiency values (Table 7). As a result, it is possible to state that the relationship detected using DEA estimates can be maintained given the variations in the sample carried out using smoothed bootstrap. This shows that, in this particular situation, the conclusions obtained when using bootstrapping as an approach to statistical inference are not different 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. 6. Conclusions This study has assessed the level of efficiency of the economic activity of travel agencies in the city of Alicante in the year 2007. It has also examined the relationship between certain variables (agency ownership type, location and experience) and the level of efficiency of those agencies. Data Envelopment Analysis, smoothed bootstrap and the Mann Whitney U Test were used for these purposes. In the first part of the study, relating to DEA efficiency levels, it was found that 7 of the 22 agencies assessed are efficient, representing 31.82% of the sample total. Having said that, the average level of efficiency is quite high (81.33%). However, some agencies have quite low levels of inefficiency (i.e. DMUs 1, 5, 10 and 11). Nevertheless, it is possible to increase these levels by reducing the amount of resources used and increasing the outputs produced by inefficient agencies until they reach the levels shown in Table 4. At the same time, a change required in the management strategy of each inefficient DMU in order to achieve these targets should be based on the strategies used in units which belong to its peer group, as this would reflect the efficient agencies that DEA has used as a reference in order to evaluate their level of efficiency. In any case, and given that it may be difficult for the inefficient agencies to

86

R. Fuentes / Tourism Management 32 (2011) 75–87

obtain this information, the target values shown in Table 4 are the levels of the inputs and outputs which should have been used to identify operative responses to management priorities in order to implement strategies to make them efficient. However, smoothed bootstrap was also used to analyse the sensitivity of DEA estimates to sampling variation and to obtain measures of uncertainty relating to these estimates in order to confirm or refute the initial values. The bootstrap results are quite different from the initial DEA estimates. They also show that confidence intervals significantly change the conclusions reached earlier about efficiency levels, as frequent overlaps can make it impossible to rank DMUs in the same way as the DEA results obtained at the beginning. In fact, DMU17 and DMU20 have the highest scores based on their high bias-corrected value, despite the fact that the DEA results did not lead to them being _ classified as efficient (even though it gave them high 4 j values). In the second part of the study, which analysed the relationship between three different variables and the efficiency of the travel agencies, the results show that the location of an agency is the only one of these factors that affects their levels of efficiency. This shows that setting up an agency in areas close to the city centre has a major influence on the efficiency of that agency. As such, a higher concentration of the agencies around the city centre would imply an increase in their level of efficiency in spite of the fact that it could result in more elevated initial costs for them. In addition to this, no link was found between the levels of efficiency of agencies and their ownership type, nor with their level of experience. These results suggest that neither belonging to a group nor having a lot of experience affect the efficiency of an agency. In the first case, the result shows that the advantages of belonging to a group are not statistically significant. This means that agencies which form part of large chains may have access to management conditions which do not, in the end, have any influence on their efficiency. From this point of view, it would seem reasonable for units which form part of groups to look at the advantages that they should, in theory, enjoy because they belong to a larger organisation, to find out whether they are really materialising. In any case, including information about the profits that each agency makes may lead to a change in this conclusion. Unfortunately, it was not possible to obtain this type of data. In the second case (level of experience), the result obtained is similar to the first. This means that having more years of experience in the sector does not have a statistically significant link with the level of efficiency of agencies. This seems to suggest that, once an agency has overcome the logical first stage of learning and become established, increasing the amount of time that it has been operating in the sector does not necessarily mean that it becomes more efficient. With regard to the limitations of this study, the main drawback lies in the fact that not all of the information requested was obtained. It was impossible to increase the number of units analysed as higher response rates were not obtained. As a result, access to an official or public database which included the economic information required to carry out a study of this type may have made the task easier and improved results. In any case, the representativeness of the sample falls within the limits established in other similar studies (such as Hu & Cai, 2004 or Ko¨ksal & Aksu, 2007). In addition, another limitation of this study lies in the fact that there are a number of non-controllable factors which affect the efficiency of DMUs and which have not been taken into account here (such as the number of inhabitants in the area surrounding the agency, or their level of income). Access to information relating to these variables would make it possible to analyse their overall influence using DEA models which include non-controllable variables, or other complementary statistical methods (e.g. Tobit regression).

Nevertheless, this study has attempted to incorporate new features not included in a number of other very good studies carried out on this sort of establishment (e.g. Barros & Dieke, 2007; Ko¨ksal & Aksu, 2007 and Wo¨ber, 2006). In particular, with regard to the variables used, the experience variable has been approached from the point of view of the experience acquired by an agency over the course of the period during which it has been operating in the sector. Furthermore, it has also been possible to incorporate a variable relating to the average spend per customer, which could be used to gain a better insight into the output level of each unit by combining it with the number of customers served by the agency during the period studied. In terms of the analysis method used, the target values for each variable were calculated using efficiency variables and slack variables, as opposed to using specific weights of each efficient DMU in the peer groups of each inefficient agency. This makes it possible to see where additional efforts need to be made in some resources even when the level of efficiency of an agency has been corrected. In addition to this, the smoothed bootstrap technique was also used to study the stability of the results after sampling variation and to carry out statistical inference. Finally, the analysis could be continued through the calculation of allocative and economic efficiency levels or even, if dynamic data could be obtained, through the calculation of changes in the levels of productivity. Clearly, this information could add to the results because, for example, Malmquist indices or quasi-Malmquist indices could be used in order to find out how the productivity levels of agencies develop over time, and the way in which inputs and outputs influence those levels. It would also be useful to include output-oriented analyses which could show the maximum production levels that DMUs could generate (with the amount of resources they use) if they demonstrate efficient behaviour. References ˜a: evolucio´n y conAlonso, C., & Galdo´n, E. (2007). La proteccio´n al empleo en Espan ´n Comercial Espan ˜ola, Revista de Economı´a, 837, 157–177. secuencias. Informacio Anderson, R. I., Fish, M., Xia, Y., & Michello, F. (1999). Measuring efficiency in the hotel industry: a stochastic frontier approach. Hospitality Management, 18, 45–57. Anderson, R. I., Fok, R., & Scott, J. (2000). Hotel industry efficiency: an advanced linear programming examination. American Business Review, 18(1), 40–48. Anderson, R. I., Lewis, D., & Parker, M. E. (February, 1999). Another look at the efficiency of corporate travel management departments. Journal of Travel Research, 37, 267–272. Baker, M., & Riley, M. (1994). News perspectives on productivity in hotels: some advances and new directions. International Journal of Hospitality Management, 13(4), 297–311. Banker, R. D., & Morey, R. C. (1986). Efficiency analysis for exogenously fixed inputs and outputs. Operations Research, 34(4), 513–521. Barros, C. P. (2004). A stochastic cost frontier in the Portuguese hotel industry. Tourism Economics, 10(2), 177–192. Barros, C. P. (2005a). Evaluating the efficiency of a small hotel chain with Malmquist productivity index. International Journal of Tourism Research, 7, 173–184. Barros, C. P. (2005b). Measuring efficiency in the hotel sector. Annals of Tourism Research, 32(2), 465–477. Barros, C. P. (2006). Analysing the rate of technical change in the Portuguese hotel industry. Tourism Economics, 12(3), 325–346. Barros, C. P., & Alves, F. P. (2004). Productivity in the tourism industry. International Advances in Economic Research, 10(3), 215–225. Barros, C. P., & Dieke, P. U. C. (2007). Analyzing the total productivity change in travel agencies. Tourism Analysis, 12, 27–37. Barros, C. P., & Mascarenhas, M. J. (2005). Technical and allocative efficiency in a chain of small hotels. Hospitality Management, 24, 415–436. Barros, C. P., & Matias, A. (2006). Assessing the efficiency of travel agencies with a stochastic cost frontier: a Portuguese case study. International Journal of Tourism Research, 8, 367–379. Bell, R. A., & Morey, R. C. (1994). The search for appropriate benchmarking partners: a macro approach and application to corporate travel management. Omega, International Journal of Management Science, 22(5), 477–490. Bell, R. A., & Morey, R. C. (1995). Increasing the efficiency of corporate travel management through macro benchmarking. Journal of Travel Research, 33(3), 11–20. Bosetti, V., Cassinelli, M., & Lanza, A. (2006). Benchmarking in tourism destination, keeping in mind the sustainable paradigm. Nota di Lavoro 12.2006, Fondazione Eni Enrico Mattei.

R. Fuentes / Tourism Management 32 (2011) 75–87 Brown, R. J., & Ragsdale, C. T. (2002). The competitive market efficiency of hotel brands: an application of data envelopment analysis. Journal of Hospitality & Tourism Research, 26(4), 332–360. Charnes, A., Cooper, W. W., Lewin, A. Y., & Seiford, L. M. (1997a). Introduction. In A. Charnes, W. W. Cooper, A. Y. Lewin, & L. M. Seiford (Eds.), Data envelopment analysis: Theory, methodology and applications (2nd ed.). (pp. 3–22) Kluwer Academic Publishers. Charnes, A., Cooper, W. W., Lewin, A. Y., & Seiford, L. M. (1997b). Basic DEA models. In A. Charnes, W. W. Cooper, A. Y. Lewin, & L. M. Seiford (Eds.), Data envelopment analysis: Theory, methodology and applications (2nd ed.). (pp. 23–48) Kluwer Academic Publishers. Charnes, A., Cooper, W. W., & Rhodes, E. (1978). Measuring the efficiency of decision making units. European Journal of Operational Research, 2, 429–444. Charnes, A., Cooper, W. W., & Rhodes, E. (1981). Evaluating program and managerial efficiency: an application of data envelopment analysis to program follow through. Management Science, 27(6), 668–697. Chiang, W., Tsai, M., & Wang, L. S. (2004). A DEA evaluation of Taipei hotels. Annals of Tourism Research, 31(3), 712–715. Coelli, T. J., Rao, D. S. P., O’Donnell, C. J., & Battese, G. E. (2005). An introduction to efficiency and productivity analysis (2nd ed.). New York: NY Springer. Conover, W. J. (1999). Practical nonparametric statistics (3rd ed.). New York: John Wiley & Sons. Cooper, W. W., Seiford, L. M., & Tone, K. (2007). Data envelopment analysis. A comprehensive text with models, applications, references and dea-solver software (2nd ed.). Springer. Cooper, W. W., Seiford, L. M., & Zhu, J. (2004). Data envelopment analysis: history, models and interpretations. In W. W. Cooper, L. M. Sieford, & J. Zhu (Eds.), Handbook on data envelopment analysis (pp. 1–40). Kluwer Academic Publishers Group. County Council of Alicante. (2009). Turistic data of Alicante, Available at: http://www. dip-alicante.es/documentacion/8turismo.asp?codigo¼03014, (Accessed 14.04.09). Davutyan, N. (2007). Measuring the quality of hospitality at Antalya. International Journal of Tourism Research, 9, 51–57. Donthu, N., & Yoo, B. (1998). Retail productivity assessment using data envelopment analysis. Journal of Retailing, 74(1), 89–105. Donthu, N., Hershberger, E. K., & Osmonbekov, T. (2005). Benchmarking marketing productivity using data envelopment analysis. Journal of Business Research, 58, 1474–1482. Efron, B. (1979). Bootstrap methods: another look at the jacknife. The Annals of Statistics, 7(1), 1–26. Efron, B., & Tibshirani, R. J. (1998). An introduction to the bootstrap (2nd ed.). Florida: CRS Press LLC. Fuchs, M. (2004). Strategy development in tourism destinations: a DEA approach. The Poznan University of Economics Review, 4(1), 52–73. Gibbons, J. D., & Chakraborti, S. (1992). Nonparametric statistical inference (3rd ed.). New York: Marcel Dekker, INC. Gime´nez-Garcı´a, V., Martı´nez-Parra, J. L., & Buffa, F. P. (2007). Improving resource utilization in multi-unit networked organizations: the case of a Spanish restaurant chain. Tourism Management, 28, 262–270. ¨r Hruschka, H. (1986). Ansa¨tze der Effizienmessung von Betrieben. Journal fu Betriebswirtschaft, 36(2), 76–85. Hu, B. A., & Cai, L. A. (2004). Hotel labor productivity assessment: a data envelopment analysis. Journal of Travel & Tourism Marketing, 16(2/3), 27–38. Hwang, S., & Chang, T. (2003). Using data envelopment analysis to measure hotel managerial efficiency change in Taiwan. Tourism Management, 24, 357–369. Johns, N., Howcroft, B., & Drake, L. (1997). The use of data envelopment analysis to monitor hotel productivity. Progress in Tourism and Hospitality Research, 3,119–127. Keh, H. T., Chu, S., & Xu, J. (2006). Efficiency, effectiveness and productivity of marketing in services. European Journal of Operational Research, 170, 265–276. Ko¨ksal, C. D., & Aksu, A. A. (2007). Efficiency evaluation of A-group travel agencies with data envelopment analysis (DEA): a case study in the Antalya region, Turkey. Tourism Management, 28, 830–834. Lo¨thgren, M., & Tambour, M. (1999). Bootstrapping the data envelopment analysis mamquist productivity index. Applied Economics, 31, 417–425.

87

McLaughlin, C. P., & Coffey, S. (1990). Measuring productivity on services. International Journal of Service Industry Management, 1(1), 46–64. Morey, R. C., & Dittman, D. A. (October, 1995). Evaluating a hotel GM’s performance. A case study in benchmarking. Cornell Hotel and Restaurant Administration Quarterly 30–35. Morey, R. C., & Dittman, D. A. (1997). An aid in selecting the brand, size and other strategic choices for a hotel. The Council on Hotel, Restaurant and Institutional Education, 21(1), 71–99. Morey, R. C., & Dittman, D. A. (2003). Evaluating a hotel GM’s performance. Updated and extension. Cornell Hotel and Restaurant Administration Quarterly, 44(5/6), 53–68. National Statistics Institute. (2009). Annual services survey. 2007, Available at: http://www.ine.es/jaxi/menu.do?type¼pcaxis&path¼%2Ft37%2Fe01%2Fprovi% 2F&file¼pcaxis&L¼0 (Accessed 14.04.09). Norman, M., & Stoker, B. (1991). DEA. The assessment of performance. John Wiley and Sons. (Appendix A). Regional Government of Valencia. (2008), Public list of travel agencies in Alicante, Available at: http://xml.turisme.gva.es:7777/establecimientos/aavv_alacant. xsql (Accessed 14.04.09). Reynolds, D. (2004). An exploratory investigation of multiunit restaurant productivity assessment using data envelopment analysis. Journal of Travel & Tourism Marketing, 16(2/3), 19–26. Reynolds, D., & Thompson, G. M. (2007). Multiunit restaurant productivity assessment using three-phase data envelopment analysis. Hospitality Management, 26, 20–32. Sheater, S. J., & Jones, M. C. (1991). A reliable data-based bandwidth selection method for kernel density estimation. Journal of the Royal Statistical Society, Series B, 53(3), 683–690. Sheskin, D. (2000). Handbook of parametric and nonparametric statistical procedures (2nd ed.). Chapman & Hall/CRC. (Test 12). Sigala, M. (2003). Developing and benchmarking internet marketing strategies in the hotel sector in Greece. Journal of Hospitality & Tourism Research, 27(4), 375–401. Sigala, M. (2004). Using data envelopment analysis for measuring and benchmarking productivity in the hotel sector. Journal of Travel & Tourism Marketing, 16(2/3), 39–60. Sigala, M., Airey, D., Jones, P., & Lockwood, A. (November, 2004). ICT paradox lost? A stepwise DEA methodology to evaluate technology investments in tourism settings. Journal of Travel Research, 43, 180–192. Sigala, M., Jones, P., Lockwood, A., & Airey, D. (2005). Productivity in hotels: a stepwise data envelopment analysis of hotels’ rooms division processes. The Service Industrial Journal, 25(1), 61–81. Sigala, M., & Mylonakis, J. (2005). Developing a data envelopment analysis model for measuring and isolating the impact of contextual factors on hotel productivity. International Journal of Business Performance Management, 7(2), 174–190. Silverman, B. W. (1986). Density estimation for statistics and data analysis. London: Chapman and Hall. Simar, L., & Wilson, P. W. (1998). Sensitivity analysis of efficiency scoresh: how to bootstrap in nonparametric frontier models. Management Science, 44(1), 49–61. Simar, L., & Wilson, P. W. (2000). Statistical inference in nonparmetric frontier models: the state of the art. Journal of Productivity Analysis, 13(1), 49–78. Tarim, S., Dener, H. I., & Tarim, S. A. (2000). Efficiency measurement in the hotel industry: output factor constrained DEA application. Anatolia: An International Journal of Tourism and Hospitality Research, 11(2), 111–123. Tortosa-Ausina, E., Grifell-Tatje´, E., Armero, C., & Conesa, D. (2008). Sensitivity analysis of efficiency and Malmquist productivity indices: an application to spanish savings Banks. European Journal of Operational Research, 184, 1062–1084. Wang, F., Hung, W., & Shang, J. (2006). Measuring the cost efficiency of international tourist hotels in Taiwan. Tourism Economics, 12(1), 65–85. Wo¨ber, K. W. (2000). Benchmarking hotel operations on the Internet: a data envelopment analysis approach. Information Technology and Tourism, 3(3/4), 195–212. Wo¨ber, K. W. (2006). Data envelopment analysis. Journal of Travel & Tourism Marketing, 21(4), 91–108. Wo¨ber, K. W., & Fesenmaier, D. R. (2004). A multi-criteria approach to destination benchmarking: a case study of state tourism advertising programs in the United States. Journal of Travel & Tourism Marketing, 16(2/3), 1–18.